blob: f7fdfff81453b575aa16b582c1c34de3e0e64a20 [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 Tarreau0dc9e6d2022-08-01 11:55:57 +020052#include <import/ebmbtree.h>
Willy Tarreau48fbcae2020-06-03 18:09:46 +020053
Willy Tarreau4c7e4b72020-05-27 12:58:42 +020054#include <haproxy/api.h>
Christopher Faulet6b0a0fb2022-04-04 11:29:28 +020055#include <haproxy/applet.h>
Willy Tarreauc13ed532020-06-02 10:22:45 +020056#include <haproxy/chunk.h>
Willy Tarreau7c18b542020-06-11 09:23:02 +020057#include <haproxy/dgram.h>
Willy Tarreauf268ee82020-06-04 17:05:57 +020058#include <haproxy/global.h>
Willy Tarreau86416052020-06-04 09:20:54 +020059#include <haproxy/hlua.h>
Willy Tarreau213e9902020-06-04 14:58:24 +020060#include <haproxy/listener.h>
Willy Tarreau7a00efb2020-06-02 17:02:59 +020061#include <haproxy/namespace.h>
Christopher Faulet9553de72021-02-26 09:12:50 +010062#include <haproxy/net_helper.h>
Willy Tarreau5fc93282020-09-16 18:25:03 +020063#include <haproxy/protocol.h>
Emeric Brunc9437992021-02-12 19:42:55 +010064#include <haproxy/resolvers.h>
Willy Tarreau5edca2f2022-05-27 09:25:10 +020065#include <haproxy/sc_strm.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 Tarreaucb086c62022-05-27 09:47:12 +020069#include <haproxy/stconn.h>
Willy Tarreaucea0e1b2020-06-04 17:25:40 +020070#include <haproxy/task.h>
Willy Tarreau48fbcae2020-06-03 18:09:46 +020071#include <haproxy/tools.h>
Willy Tarreau08093cc2022-11-24 08:09:12 +010072#include <haproxy/xxhash.h>
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +010073
Thierry Fournier93127942016-01-20 18:49:45 +010074/* This macro returns false if the test __x is false. Many
75 * of the following parsing function must be abort the processing
76 * if it returns 0, so this macro is useful for writing light code.
77 */
78#define RET0_UNLESS(__x) do { if (!(__x)) return 0; } while (0)
79
Erwan Le Goas9c766372022-09-14 17:40:24 +020080/* Define the number of line of hash_word */
Erwan Le Goas5eef1582022-09-29 10:25:31 +020081#define NB_L_HASH_WORD 15
Erwan Le Goas9c766372022-09-14 17:40:24 +020082
Willy Tarreau08093cc2022-11-24 08:09:12 +010083/* return the hash of a string and length for a given key. All keys are valid. */
84#define HA_ANON(key, str, len) (XXH32(str, len, key) & 0xFFFFFF)
85
Willy Tarreau56adcf22012-12-23 18:00:29 +010086/* enough to store NB_ITOA_STR integers of :
Willy Tarreau72d759c2007-10-25 12:14:10 +020087 * 2^64-1 = 18446744073709551615 or
88 * -2^63 = -9223372036854775808
Willy Tarreaue7239b52009-03-29 13:41:58 +020089 *
90 * The HTML version needs room for adding the 25 characters
91 * '<span class="rls"></span>' around digits at positions 3N+1 in order
92 * to add spacing at up to 6 positions : 18 446 744 073 709 551 615
Willy Tarreau72d759c2007-10-25 12:14:10 +020093 */
Christopher Faulet99bca652017-11-14 16:47:26 +010094THREAD_LOCAL char itoa_str[NB_ITOA_STR][171];
95THREAD_LOCAL int itoa_idx = 0; /* index of next itoa_str to use */
Willy Tarreaubaaee002006-06-26 02:48:02 +020096
Willy Tarreau588297f2014-06-16 15:16:40 +020097/* sometimes we'll need to quote strings (eg: in stats), and we don't expect
98 * to quote strings larger than a max configuration line.
99 */
Christopher Faulet99bca652017-11-14 16:47:26 +0100100THREAD_LOCAL char quoted_str[NB_QSTR][QSTR_SIZE + 1];
101THREAD_LOCAL int quoted_idx = 0;
Willy Tarreau588297f2014-06-16 15:16:40 +0200102
Willy Tarreau06e69b52021-03-02 14:01:35 +0100103/* thread-local PRNG state. It's modified to start from a different sequence
104 * on all threads upon startup. It must not be used or anything beyond getting
105 * statistical values as it's 100% predictable.
106 */
107THREAD_LOCAL unsigned int statistical_prng_state = 2463534242U;
108
Willy Tarreau5b3cd952022-07-18 13:58:17 +0200109/* set to true if this is a static build */
110int build_is_static = 0;
111
Erwan Le Goas9c766372022-09-14 17:40:24 +0200112/* A global static table to store hashed words */
113static THREAD_LOCAL char hash_word[NB_L_HASH_WORD][20];
114static THREAD_LOCAL int index_hash = 0;
115
Willy Tarreaubaaee002006-06-26 02:48:02 +0200116/*
William Lallemande7340ec2012-01-24 11:15:39 +0100117 * unsigned long long ASCII representation
118 *
119 * return the last char '\0' or NULL if no enough
120 * space in dst
121 */
122char *ulltoa(unsigned long long n, char *dst, size_t size)
123{
124 int i = 0;
125 char *res;
126
127 switch(n) {
128 case 1ULL ... 9ULL:
129 i = 0;
130 break;
131
132 case 10ULL ... 99ULL:
133 i = 1;
134 break;
135
136 case 100ULL ... 999ULL:
137 i = 2;
138 break;
139
140 case 1000ULL ... 9999ULL:
141 i = 3;
142 break;
143
144 case 10000ULL ... 99999ULL:
145 i = 4;
146 break;
147
148 case 100000ULL ... 999999ULL:
149 i = 5;
150 break;
151
152 case 1000000ULL ... 9999999ULL:
153 i = 6;
154 break;
155
156 case 10000000ULL ... 99999999ULL:
157 i = 7;
158 break;
159
160 case 100000000ULL ... 999999999ULL:
161 i = 8;
162 break;
163
164 case 1000000000ULL ... 9999999999ULL:
165 i = 9;
166 break;
167
168 case 10000000000ULL ... 99999999999ULL:
169 i = 10;
170 break;
171
172 case 100000000000ULL ... 999999999999ULL:
173 i = 11;
174 break;
175
176 case 1000000000000ULL ... 9999999999999ULL:
177 i = 12;
178 break;
179
180 case 10000000000000ULL ... 99999999999999ULL:
181 i = 13;
182 break;
183
184 case 100000000000000ULL ... 999999999999999ULL:
185 i = 14;
186 break;
187
188 case 1000000000000000ULL ... 9999999999999999ULL:
189 i = 15;
190 break;
191
192 case 10000000000000000ULL ... 99999999999999999ULL:
193 i = 16;
194 break;
195
196 case 100000000000000000ULL ... 999999999999999999ULL:
197 i = 17;
198 break;
199
200 case 1000000000000000000ULL ... 9999999999999999999ULL:
201 i = 18;
202 break;
203
204 case 10000000000000000000ULL ... ULLONG_MAX:
205 i = 19;
206 break;
207 }
208 if (i + 2 > size) // (i + 1) + '\0'
209 return NULL; // too long
210 res = dst + i + 1;
211 *res = '\0';
212 for (; i >= 0; i--) {
213 dst[i] = n % 10ULL + '0';
214 n /= 10ULL;
215 }
216 return res;
217}
218
219/*
220 * unsigned long ASCII representation
221 *
222 * return the last char '\0' or NULL if no enough
223 * space in dst
224 */
225char *ultoa_o(unsigned long n, char *dst, size_t size)
226{
227 int i = 0;
228 char *res;
229
230 switch (n) {
231 case 0U ... 9UL:
232 i = 0;
233 break;
234
235 case 10U ... 99UL:
236 i = 1;
237 break;
238
239 case 100U ... 999UL:
240 i = 2;
241 break;
242
243 case 1000U ... 9999UL:
244 i = 3;
245 break;
246
247 case 10000U ... 99999UL:
248 i = 4;
249 break;
250
251 case 100000U ... 999999UL:
252 i = 5;
253 break;
254
255 case 1000000U ... 9999999UL:
256 i = 6;
257 break;
258
259 case 10000000U ... 99999999UL:
260 i = 7;
261 break;
262
263 case 100000000U ... 999999999UL:
264 i = 8;
265 break;
266#if __WORDSIZE == 32
267
268 case 1000000000ULL ... ULONG_MAX:
269 i = 9;
270 break;
271
272#elif __WORDSIZE == 64
273
274 case 1000000000ULL ... 9999999999UL:
275 i = 9;
276 break;
277
278 case 10000000000ULL ... 99999999999UL:
279 i = 10;
280 break;
281
282 case 100000000000ULL ... 999999999999UL:
283 i = 11;
284 break;
285
286 case 1000000000000ULL ... 9999999999999UL:
287 i = 12;
288 break;
289
290 case 10000000000000ULL ... 99999999999999UL:
291 i = 13;
292 break;
293
294 case 100000000000000ULL ... 999999999999999UL:
295 i = 14;
296 break;
297
298 case 1000000000000000ULL ... 9999999999999999UL:
299 i = 15;
300 break;
301
302 case 10000000000000000ULL ... 99999999999999999UL:
303 i = 16;
304 break;
305
306 case 100000000000000000ULL ... 999999999999999999UL:
307 i = 17;
308 break;
309
310 case 1000000000000000000ULL ... 9999999999999999999UL:
311 i = 18;
312 break;
313
314 case 10000000000000000000ULL ... ULONG_MAX:
315 i = 19;
316 break;
317
318#endif
319 }
320 if (i + 2 > size) // (i + 1) + '\0'
321 return NULL; // too long
322 res = dst + i + 1;
323 *res = '\0';
324 for (; i >= 0; i--) {
325 dst[i] = n % 10U + '0';
326 n /= 10U;
327 }
328 return res;
329}
330
331/*
332 * signed long ASCII representation
333 *
334 * return the last char '\0' or NULL if no enough
335 * space in dst
336 */
337char *ltoa_o(long int n, char *dst, size_t size)
338{
339 char *pos = dst;
340
341 if (n < 0) {
342 if (size < 3)
343 return NULL; // min size is '-' + digit + '\0' but another test in ultoa
344 *pos = '-';
345 pos++;
346 dst = ultoa_o(-n, pos, size - 1);
347 } else {
348 dst = ultoa_o(n, dst, size);
349 }
350 return dst;
351}
352
353/*
354 * signed long long ASCII representation
355 *
356 * return the last char '\0' or NULL if no enough
357 * space in dst
358 */
359char *lltoa(long long n, char *dst, size_t size)
360{
361 char *pos = dst;
362
363 if (n < 0) {
364 if (size < 3)
365 return NULL; // min size is '-' + digit + '\0' but another test in ulltoa
366 *pos = '-';
367 pos++;
368 dst = ulltoa(-n, pos, size - 1);
369 } else {
370 dst = ulltoa(n, dst, size);
371 }
372 return dst;
373}
374
375/*
376 * write a ascii representation of a unsigned into dst,
377 * return a pointer to the last character
378 * Pad the ascii representation with '0', using size.
379 */
380char *utoa_pad(unsigned int n, char *dst, size_t size)
381{
382 int i = 0;
383 char *ret;
384
385 switch(n) {
386 case 0U ... 9U:
387 i = 0;
388 break;
389
390 case 10U ... 99U:
391 i = 1;
392 break;
393
394 case 100U ... 999U:
395 i = 2;
396 break;
397
398 case 1000U ... 9999U:
399 i = 3;
400 break;
401
402 case 10000U ... 99999U:
403 i = 4;
404 break;
405
406 case 100000U ... 999999U:
407 i = 5;
408 break;
409
410 case 1000000U ... 9999999U:
411 i = 6;
412 break;
413
414 case 10000000U ... 99999999U:
415 i = 7;
416 break;
417
418 case 100000000U ... 999999999U:
419 i = 8;
420 break;
421
422 case 1000000000U ... 4294967295U:
423 i = 9;
424 break;
425 }
426 if (i + 2 > size) // (i + 1) + '\0'
427 return NULL; // too long
Aurelien DARRAGONe3177af2022-11-22 11:42:07 +0100428 i = size - 2; // padding - '\0'
William Lallemande7340ec2012-01-24 11:15:39 +0100429
430 ret = dst + i + 1;
431 *ret = '\0';
432 for (; i >= 0; i--) {
433 dst[i] = n % 10U + '0';
434 n /= 10U;
435 }
436 return ret;
437}
438
439/*
Willy Tarreaubaaee002006-06-26 02:48:02 +0200440 * copies at most <size-1> chars from <src> to <dst>. Last char is always
441 * set to 0, unless <size> is 0. The number of chars copied is returned
442 * (excluding the terminating zero).
443 * This code has been optimized for size and speed : on x86, it's 45 bytes
444 * long, uses only registers, and consumes only 4 cycles per char.
445 */
446int strlcpy2(char *dst, const char *src, int size)
447{
448 char *orig = dst;
449 if (size) {
450 while (--size && (*dst = *src)) {
451 src++; dst++;
452 }
453 *dst = 0;
454 }
455 return dst - orig;
456}
457
458/*
Willy Tarreau72d759c2007-10-25 12:14:10 +0200459 * This function simply returns a locally allocated string containing
Willy Tarreaubaaee002006-06-26 02:48:02 +0200460 * the ascii representation for number 'n' in decimal.
461 */
Emeric Brun3a7fce52010-01-04 14:54:38 +0100462char *ultoa_r(unsigned long n, char *buffer, int size)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200463{
464 char *pos;
465
Willy Tarreau72d759c2007-10-25 12:14:10 +0200466 pos = buffer + size - 1;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200467 *pos-- = '\0';
468
469 do {
470 *pos-- = '0' + n % 10;
471 n /= 10;
Willy Tarreau72d759c2007-10-25 12:14:10 +0200472 } while (n && pos >= buffer);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200473 return pos + 1;
474}
475
Willy Tarreau91092e52007-10-25 16:58:42 +0200476/*
Willy Tarreaue7239b52009-03-29 13:41:58 +0200477 * This function simply returns a locally allocated string containing
Thierry FOURNIER763a5d82015-07-06 23:09:52 +0200478 * the ascii representation for number 'n' in decimal.
479 */
480char *lltoa_r(long long int in, char *buffer, int size)
481{
482 char *pos;
483 int neg = 0;
484 unsigned long long int n;
485
486 pos = buffer + size - 1;
487 *pos-- = '\0';
488
489 if (in < 0) {
490 neg = 1;
491 n = -in;
492 }
493 else
494 n = in;
495
496 do {
497 *pos-- = '0' + n % 10;
498 n /= 10;
499 } while (n && pos >= buffer);
500 if (neg && pos > buffer)
501 *pos-- = '-';
502 return pos + 1;
503}
504
505/*
506 * This function simply returns a locally allocated string containing
Thierry FOURNIER1480bd82015-06-06 19:14:59 +0200507 * the ascii representation for signed number 'n' in decimal.
508 */
509char *sltoa_r(long n, char *buffer, int size)
510{
511 char *pos;
512
513 if (n >= 0)
514 return ultoa_r(n, buffer, size);
515
516 pos = ultoa_r(-n, buffer + 1, size - 1) - 1;
517 *pos = '-';
518 return pos;
519}
520
521/*
522 * This function simply returns a locally allocated string containing
Willy Tarreaue7239b52009-03-29 13:41:58 +0200523 * the ascii representation for number 'n' in decimal, formatted for
524 * HTML output with tags to create visual grouping by 3 digits. The
525 * output needs to support at least 171 characters.
526 */
527const char *ulltoh_r(unsigned long long n, char *buffer, int size)
528{
529 char *start;
530 int digit = 0;
531
532 start = buffer + size;
533 *--start = '\0';
534
535 do {
536 if (digit == 3 && start >= buffer + 7)
537 memcpy(start -= 7, "</span>", 7);
538
539 if (start >= buffer + 1) {
540 *--start = '0' + n % 10;
541 n /= 10;
542 }
543
544 if (digit == 3 && start >= buffer + 18)
545 memcpy(start -= 18, "<span class=\"rls\">", 18);
546
547 if (digit++ == 3)
548 digit = 1;
549 } while (n && start > buffer);
550 return start;
551}
552
553/*
Willy Tarreau91092e52007-10-25 16:58:42 +0200554 * This function simply returns a locally allocated string containing the ascii
555 * representation for number 'n' in decimal, unless n is 0 in which case it
556 * returns the alternate string (or an empty string if the alternate string is
557 * NULL). It use is intended for limits reported in reports, where it's
558 * desirable not to display anything if there is no limit. Warning! it shares
559 * the same vector as ultoa_r().
560 */
561const char *limit_r(unsigned long n, char *buffer, int size, const char *alt)
562{
563 return (n) ? ultoa_r(n, buffer, size) : (alt ? alt : "");
564}
565
Willy Tarreau56d1d8d2021-05-08 10:28:53 +0200566/* Trims the first "%f" float in a string to its minimum number of digits after
567 * the decimal point by trimming trailing zeroes, even dropping the decimal
568 * point if not needed. The string is in <buffer> of length <len>, and the
569 * number is expected to start at or after position <num_start> (the first
570 * point appearing there is considered). A NUL character is always placed at
571 * the end if some trimming occurs. The new buffer length is returned.
572 */
573size_t flt_trim(char *buffer, size_t num_start, size_t len)
574{
575 char *end = buffer + len;
576 char *p = buffer + num_start;
577 char *trim;
578
579 do {
580 if (p >= end)
581 return len;
582 trim = p++;
583 } while (*trim != '.');
584
585 /* For now <trim> is on the decimal point. Let's look for any other
586 * meaningful digit after it.
587 */
588 while (p < end) {
589 if (*p++ != '0')
590 trim = p;
591 }
592
593 if (trim < end)
594 *trim = 0;
595
596 return trim - buffer;
597}
598
Willy Tarreauae03d262021-05-08 07:35:00 +0200599/*
600 * This function simply returns a locally allocated string containing
601 * the ascii representation for number 'n' in decimal with useless trailing
602 * zeroes trimmed.
603 */
604char *ftoa_r(double n, char *buffer, int size)
605{
606 flt_trim(buffer, 0, snprintf(buffer, size, "%f", n));
607 return buffer;
608}
609
Willy Tarreau588297f2014-06-16 15:16:40 +0200610/* returns a locally allocated string containing the quoted encoding of the
611 * input string. The output may be truncated to QSTR_SIZE chars, but it is
612 * guaranteed that the string will always be properly terminated. Quotes are
613 * encoded by doubling them as is commonly done in CSV files. QSTR_SIZE must
614 * always be at least 4 chars.
615 */
616const char *qstr(const char *str)
617{
618 char *ret = quoted_str[quoted_idx];
619 char *p, *end;
620
621 if (++quoted_idx >= NB_QSTR)
622 quoted_idx = 0;
623
624 p = ret;
625 end = ret + QSTR_SIZE;
626
627 *p++ = '"';
628
629 /* always keep 3 chars to support passing "" and the ending " */
630 while (*str && p < end - 3) {
631 if (*str == '"') {
632 *p++ = '"';
633 *p++ = '"';
634 }
635 else
636 *p++ = *str;
637 str++;
638 }
639 *p++ = '"';
640 return ret;
641}
642
Robert Tsai81ae1952007-12-05 10:47:29 +0100643/*
Willy Tarreaubaaee002006-06-26 02:48:02 +0200644 * Returns non-zero if character <s> is a hex digit (0-9, a-f, A-F), else zero.
645 *
646 * It looks like this one would be a good candidate for inlining, but this is
647 * not interesting because it around 35 bytes long and often called multiple
648 * times within the same function.
649 */
650int ishex(char s)
651{
652 s -= '0';
653 if ((unsigned char)s <= 9)
654 return 1;
655 s -= 'A' - '0';
656 if ((unsigned char)s <= 5)
657 return 1;
658 s -= 'a' - 'A';
659 if ((unsigned char)s <= 5)
660 return 1;
661 return 0;
662}
663
Willy Tarreau3ca1a882015-01-15 18:43:49 +0100664/* rounds <i> down to the closest value having max 2 digits */
665unsigned int round_2dig(unsigned int i)
666{
667 unsigned int mul = 1;
668
669 while (i >= 100) {
670 i /= 10;
671 mul *= 10;
672 }
673 return i * mul;
674}
675
Willy Tarreau2e74c3f2007-12-02 18:45:09 +0100676/*
677 * Checks <name> for invalid characters. Valid chars are [A-Za-z0-9_:.-]. If an
678 * invalid character is found, a pointer to it is returned. If everything is
679 * fine, NULL is returned.
680 */
681const char *invalid_char(const char *name)
682{
683 if (!*name)
684 return name;
685
686 while (*name) {
Willy Tarreau90807112020-02-25 08:16:33 +0100687 if (!isalnum((unsigned char)*name) && *name != '.' && *name != ':' &&
Willy Tarreau2e74c3f2007-12-02 18:45:09 +0100688 *name != '_' && *name != '-')
689 return name;
690 name++;
691 }
692 return NULL;
693}
Willy Tarreaubaaee002006-06-26 02:48:02 +0200694
695/*
Frédéric Lécailleb82f7422017-04-13 18:24:23 +0200696 * Checks <name> for invalid characters. Valid chars are [_.-] and those
697 * accepted by <f> function.
Krzysztof Piotr Oledzkiefe3b6f2008-05-23 23:49:32 +0200698 * If an invalid character is found, a pointer to it is returned.
699 * If everything is fine, NULL is returned.
700 */
Frédéric Lécailleb82f7422017-04-13 18:24:23 +0200701static inline const char *__invalid_char(const char *name, int (*f)(int)) {
Krzysztof Piotr Oledzkiefe3b6f2008-05-23 23:49:32 +0200702
703 if (!*name)
704 return name;
705
706 while (*name) {
Willy Tarreau90807112020-02-25 08:16:33 +0100707 if (!f((unsigned char)*name) && *name != '.' &&
Krzysztof Piotr Oledzkiefe3b6f2008-05-23 23:49:32 +0200708 *name != '_' && *name != '-')
709 return name;
710
711 name++;
712 }
713
714 return NULL;
715}
716
717/*
Frédéric Lécailleb82f7422017-04-13 18:24:23 +0200718 * Checks <name> for invalid characters. Valid chars are [A-Za-z0-9_.-].
719 * If an invalid character is found, a pointer to it is returned.
720 * If everything is fine, NULL is returned.
721 */
722const char *invalid_domainchar(const char *name) {
723 return __invalid_char(name, isalnum);
724}
725
726/*
727 * Checks <name> for invalid characters. Valid chars are [A-Za-z_.-].
728 * If an invalid character is found, a pointer to it is returned.
729 * If everything is fine, NULL is returned.
730 */
731const char *invalid_prefix_char(const char *name) {
Thierry Fournierf7b7c3e2018-03-26 11:54:39 +0200732 return __invalid_char(name, isalnum);
Frédéric Lécailleb82f7422017-04-13 18:24:23 +0200733}
734
735/*
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100736 * converts <str> to a struct sockaddr_storage* provided by the caller. The
Willy Tarreau24709282013-03-10 21:32:12 +0100737 * caller must have zeroed <sa> first, and may have set sa->ss_family to force
738 * parse a specific address format. If the ss_family is 0 or AF_UNSPEC, then
739 * the function tries to guess the address family from the syntax. If the
740 * family is forced and the format doesn't match, an error is returned. The
Willy Tarreaufab5a432011-03-04 15:31:53 +0100741 * string is assumed to contain only an address, no port. The address can be a
742 * dotted IPv4 address, an IPv6 address, a host name, or empty or "*" to
743 * indicate INADDR_ANY. NULL is returned if the host part cannot be resolved.
744 * The return address will only have the address family and the address set,
745 * all other fields remain zero. The string is not supposed to be modified.
Thierry FOURNIER58639a02014-11-25 12:02:25 +0100746 * The IPv6 '::' address is IN6ADDR_ANY. If <resolve> is non-zero, the hostname
747 * is resolved, otherwise only IP addresses are resolved, and anything else
Willy Tarreauecde7df2016-11-02 22:37:03 +0100748 * returns NULL. If the address contains a port, this one is preserved.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200749 */
Thierry FOURNIER58639a02014-11-25 12:02:25 +0100750struct sockaddr_storage *str2ip2(const char *str, struct sockaddr_storage *sa, int resolve)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200751{
Willy Tarreaufab5a432011-03-04 15:31:53 +0100752 struct hostent *he;
mildisff5d5102015-10-26 18:50:08 +0100753 /* max IPv6 length, including brackets and terminating NULL */
754 char tmpip[48];
Willy Tarreauecde7df2016-11-02 22:37:03 +0100755 int port = get_host_port(sa);
mildisff5d5102015-10-26 18:50:08 +0100756
757 /* check IPv6 with square brackets */
758 if (str[0] == '[') {
759 size_t iplength = strlen(str);
760
761 if (iplength < 4) {
762 /* minimal size is 4 when using brackets "[::]" */
763 goto fail;
764 }
765 else if (iplength >= sizeof(tmpip)) {
766 /* IPv6 literal can not be larger than tmpip */
767 goto fail;
768 }
769 else {
770 if (str[iplength - 1] != ']') {
771 /* if address started with bracket, it should end with bracket */
772 goto fail;
773 }
774 else {
775 memcpy(tmpip, str + 1, iplength - 2);
776 tmpip[iplength - 2] = '\0';
777 str = tmpip;
778 }
779 }
780 }
Willy Tarreaufab5a432011-03-04 15:31:53 +0100781
Willy Tarreaufab5a432011-03-04 15:31:53 +0100782 /* Any IPv6 address */
783 if (str[0] == ':' && str[1] == ':' && !str[2]) {
Willy Tarreau24709282013-03-10 21:32:12 +0100784 if (!sa->ss_family || sa->ss_family == AF_UNSPEC)
785 sa->ss_family = AF_INET6;
786 else if (sa->ss_family != AF_INET6)
787 goto fail;
Willy Tarreauecde7df2016-11-02 22:37:03 +0100788 set_host_port(sa, port);
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100789 return sa;
Willy Tarreaufab5a432011-03-04 15:31:53 +0100790 }
791
Willy Tarreau24709282013-03-10 21:32:12 +0100792 /* Any address for the family, defaults to IPv4 */
Willy Tarreaufab5a432011-03-04 15:31:53 +0100793 if (!str[0] || (str[0] == '*' && !str[1])) {
Willy Tarreau24709282013-03-10 21:32:12 +0100794 if (!sa->ss_family || sa->ss_family == AF_UNSPEC)
795 sa->ss_family = AF_INET;
Willy Tarreauecde7df2016-11-02 22:37:03 +0100796 set_host_port(sa, port);
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100797 return sa;
Willy Tarreaufab5a432011-03-04 15:31:53 +0100798 }
799
800 /* check for IPv6 first */
Willy Tarreau24709282013-03-10 21:32:12 +0100801 if ((!sa->ss_family || sa->ss_family == AF_UNSPEC || sa->ss_family == AF_INET6) &&
802 inet_pton(AF_INET6, str, &((struct sockaddr_in6 *)sa)->sin6_addr)) {
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100803 sa->ss_family = AF_INET6;
Willy Tarreauecde7df2016-11-02 22:37:03 +0100804 set_host_port(sa, port);
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100805 return sa;
Willy Tarreaufab5a432011-03-04 15:31:53 +0100806 }
807
808 /* then check for IPv4 */
Willy Tarreau24709282013-03-10 21:32:12 +0100809 if ((!sa->ss_family || sa->ss_family == AF_UNSPEC || sa->ss_family == AF_INET) &&
810 inet_pton(AF_INET, str, &((struct sockaddr_in *)sa)->sin_addr)) {
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100811 sa->ss_family = AF_INET;
Willy Tarreauecde7df2016-11-02 22:37:03 +0100812 set_host_port(sa, port);
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100813 return sa;
Willy Tarreaufab5a432011-03-04 15:31:53 +0100814 }
815
Thierry FOURNIER58639a02014-11-25 12:02:25 +0100816 if (!resolve)
817 return NULL;
818
Emeric Brund30e9a12020-12-23 18:49:16 +0100819 if (!resolv_hostname_validation(str, NULL))
Baptiste Assmanna68ca962015-04-14 01:15:08 +0200820 return NULL;
821
David du Colombierd5f43282011-03-17 10:40:16 +0100822#ifdef USE_GETADDRINFO
Nenad Merdanovic88afe032014-04-14 15:56:58 +0200823 if (global.tune.options & GTUNE_USE_GAI) {
David du Colombierd5f43282011-03-17 10:40:16 +0100824 struct addrinfo hints, *result;
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100825 int success = 0;
David du Colombierd5f43282011-03-17 10:40:16 +0100826
827 memset(&result, 0, sizeof(result));
828 memset(&hints, 0, sizeof(hints));
Willy Tarreau24709282013-03-10 21:32:12 +0100829 hints.ai_family = sa->ss_family ? sa->ss_family : AF_UNSPEC;
David du Colombierd5f43282011-03-17 10:40:16 +0100830 hints.ai_socktype = SOCK_DGRAM;
Dmitry Sivachenkoeab7f392015-10-02 01:01:58 +0200831 hints.ai_flags = 0;
David du Colombierd5f43282011-03-17 10:40:16 +0100832 hints.ai_protocol = 0;
833
834 if (getaddrinfo(str, NULL, &hints, &result) == 0) {
Willy Tarreau24709282013-03-10 21:32:12 +0100835 if (!sa->ss_family || sa->ss_family == AF_UNSPEC)
836 sa->ss_family = result->ai_family;
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100837 else if (sa->ss_family != result->ai_family) {
838 freeaddrinfo(result);
Willy Tarreau24709282013-03-10 21:32:12 +0100839 goto fail;
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100840 }
Willy Tarreau24709282013-03-10 21:32:12 +0100841
David du Colombierd5f43282011-03-17 10:40:16 +0100842 switch (result->ai_family) {
843 case AF_INET:
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100844 memcpy((struct sockaddr_in *)sa, result->ai_addr, result->ai_addrlen);
Willy Tarreauecde7df2016-11-02 22:37:03 +0100845 set_host_port(sa, port);
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100846 success = 1;
847 break;
David du Colombierd5f43282011-03-17 10:40:16 +0100848 case AF_INET6:
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100849 memcpy((struct sockaddr_in6 *)sa, result->ai_addr, result->ai_addrlen);
Willy Tarreauecde7df2016-11-02 22:37:03 +0100850 set_host_port(sa, port);
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100851 success = 1;
852 break;
David du Colombierd5f43282011-03-17 10:40:16 +0100853 }
854 }
855
Sean Carey58ea0392013-02-15 23:39:18 +0100856 if (result)
857 freeaddrinfo(result);
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100858
859 if (success)
860 return sa;
Willy Tarreaufab5a432011-03-04 15:31:53 +0100861 }
David du Colombierd5f43282011-03-17 10:40:16 +0100862#endif
Nenad Merdanovic88afe032014-04-14 15:56:58 +0200863 /* try to resolve an IPv4/IPv6 hostname */
864 he = gethostbyname(str);
865 if (he) {
866 if (!sa->ss_family || sa->ss_family == AF_UNSPEC)
867 sa->ss_family = he->h_addrtype;
868 else if (sa->ss_family != he->h_addrtype)
869 goto fail;
870
871 switch (sa->ss_family) {
872 case AF_INET:
873 ((struct sockaddr_in *)sa)->sin_addr = *(struct in_addr *) *(he->h_addr_list);
Willy Tarreauecde7df2016-11-02 22:37:03 +0100874 set_host_port(sa, port);
Nenad Merdanovic88afe032014-04-14 15:56:58 +0200875 return sa;
876 case AF_INET6:
877 ((struct sockaddr_in6 *)sa)->sin6_addr = *(struct in6_addr *) *(he->h_addr_list);
Willy Tarreauecde7df2016-11-02 22:37:03 +0100878 set_host_port(sa, port);
Nenad Merdanovic88afe032014-04-14 15:56:58 +0200879 return sa;
880 }
881 }
882
David du Colombierd5f43282011-03-17 10:40:16 +0100883 /* unsupported address family */
Willy Tarreau24709282013-03-10 21:32:12 +0100884 fail:
Willy Tarreaufab5a432011-03-04 15:31:53 +0100885 return NULL;
886}
887
888/*
Willy Tarreaud4448bc2013-02-20 15:55:15 +0100889 * Converts <str> to a locally allocated struct sockaddr_storage *, and a port
890 * range or offset consisting in two integers that the caller will have to
891 * check to find the relevant input format. The following format are supported :
892 *
893 * String format | address | port | low | high
894 * addr | <addr> | 0 | 0 | 0
895 * addr: | <addr> | 0 | 0 | 0
896 * addr:port | <addr> | <port> | <port> | <port>
897 * addr:pl-ph | <addr> | <pl> | <pl> | <ph>
898 * addr:+port | <addr> | <port> | 0 | <port>
899 * addr:-port | <addr> |-<port> | <port> | 0
900 *
901 * The detection of a port range or increment by the caller is made by
902 * comparing <low> and <high>. If both are equal, then port 0 means no port
903 * was specified. The caller may pass NULL for <low> and <high> if it is not
904 * interested in retrieving port ranges.
905 *
906 * Note that <addr> above may also be :
907 * - empty ("") => family will be AF_INET and address will be INADDR_ANY
908 * - "*" => family will be AF_INET and address will be INADDR_ANY
909 * - "::" => family will be AF_INET6 and address will be IN6ADDR_ANY
910 * - a host name => family and address will depend on host name resolving.
911 *
Willy Tarreau24709282013-03-10 21:32:12 +0100912 * A prefix may be passed in before the address above to force the family :
913 * - "ipv4@" => force address to resolve as IPv4 and fail if not possible.
914 * - "ipv6@" => force address to resolve as IPv6 and fail if not possible.
915 * - "unix@" => force address to be a path to a UNIX socket even if the
916 * path does not start with a '/'
Willy Tarreauccfccef2014-05-10 01:49:15 +0200917 * - 'abns@' -> force address to belong to the abstract namespace (Linux
918 * only). These sockets are just like Unix sockets but without
919 * the need for an underlying file system. The address is a
920 * string. Technically it's like a Unix socket with a zero in
921 * the first byte of the address.
Willy Tarreau40aa0702013-03-10 23:51:38 +0100922 * - "fd@" => an integer must follow, and is a file descriptor number.
Willy Tarreau24709282013-03-10 21:32:12 +0100923 *
mildisff5d5102015-10-26 18:50:08 +0100924 * IPv6 addresses can be declared with or without square brackets. When using
925 * square brackets for IPv6 addresses, the port separator (colon) is optional.
926 * If not using square brackets, and in order to avoid any ambiguity with
927 * IPv6 addresses, the last colon ':' is mandatory even when no port is specified.
928 * NULL is returned if the address cannot be parsed. The <low> and <high> ports
929 * are always initialized if non-null, even for non-IP families.
Willy Tarreaud393a622013-03-04 18:22:00 +0100930 *
931 * If <pfx> is non-null, it is used as a string prefix before any path-based
932 * address (typically the path to a unix socket).
Willy Tarreau40aa0702013-03-10 23:51:38 +0100933 *
Willy Tarreau72b8c1f2015-09-08 15:50:19 +0200934 * if <fqdn> is non-null, it will be filled with :
935 * - a pointer to the FQDN of the server name to resolve if there's one, and
936 * that the caller will have to free(),
937 * - NULL if there was an explicit address that doesn't require resolution.
938 *
Willy Tarreaucd3a55912020-09-04 15:30:46 +0200939 * Hostnames are only resolved if <opts> has PA_O_RESOLVE. Otherwise <fqdn> is
940 * still honored so it is possible for the caller to know whether a resolution
941 * failed by clearing this flag and checking if <fqdn> was filled, indicating
942 * the need for a resolution.
Thierry FOURNIER7fe3be72015-09-26 20:03:36 +0200943 *
Willy Tarreau40aa0702013-03-10 23:51:38 +0100944 * When a file descriptor is passed, its value is put into the s_addr part of
Willy Tarreaua5b325f2020-09-04 16:44:20 +0200945 * the address when cast to sockaddr_in and the address family is
946 * AF_CUST_EXISTING_FD.
Willy Tarreaua93e5c72020-09-15 14:01:16 +0200947 *
Willy Tarreau5fc93282020-09-16 18:25:03 +0200948 * The matching protocol will be set into <proto> if non-null.
949 *
Willy Tarreaua93e5c72020-09-15 14:01:16 +0200950 * Any known file descriptor is also assigned to <fd> if non-null, otherwise it
951 * is forced to -1.
Willy Tarreaufab5a432011-03-04 15:31:53 +0100952 */
Willy Tarreau5fc93282020-09-16 18:25:03 +0200953struct sockaddr_storage *str2sa_range(const char *str, int *port, int *low, int *high, int *fd,
954 struct protocol **proto, char **err,
955 const char *pfx, char **fqdn, unsigned int opts)
Willy Tarreaufab5a432011-03-04 15:31:53 +0100956{
Christopher Faulet1bc04c72017-10-29 20:14:08 +0100957 static THREAD_LOCAL struct sockaddr_storage ss;
David du Colombier6f5ccb12011-03-10 22:26:24 +0100958 struct sockaddr_storage *ret = NULL;
Willy Tarreau5fc93282020-09-16 18:25:03 +0200959 struct protocol *new_proto = NULL;
Willy Tarreau24709282013-03-10 21:32:12 +0100960 char *back, *str2;
Willy Tarreaud4448bc2013-02-20 15:55:15 +0100961 char *port1, *port2;
962 int portl, porth, porta;
Willy Tarreauccfccef2014-05-10 01:49:15 +0200963 int abstract = 0;
Willy Tarreaua93e5c72020-09-15 14:01:16 +0200964 int new_fd = -1;
Willy Tarreaue3b45182021-10-27 17:28:55 +0200965 enum proto_type proto_type;
966 int ctrl_type;
Willy Tarreaud4448bc2013-02-20 15:55:15 +0100967
968 portl = porth = porta = 0;
Willy Tarreau72b8c1f2015-09-08 15:50:19 +0200969 if (fqdn)
970 *fqdn = NULL;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200971
Willy Tarreaudad36a32013-03-11 01:20:04 +0100972 str2 = back = env_expand(strdup(str));
Willy Tarreaudf350f12013-03-01 20:22:54 +0100973 if (str2 == NULL) {
974 memprintf(err, "out of memory in '%s'\n", __FUNCTION__);
Willy Tarreaud5191e72010-02-09 20:50:45 +0100975 goto out;
Willy Tarreaudf350f12013-03-01 20:22:54 +0100976 }
Willy Tarreaubaaee002006-06-26 02:48:02 +0200977
Willy Tarreau9f69f462015-09-08 16:01:25 +0200978 if (!*str2) {
979 memprintf(err, "'%s' resolves to an empty address (environment variable missing?)\n", str);
980 goto out;
981 }
982
Willy Tarreau24709282013-03-10 21:32:12 +0100983 memset(&ss, 0, sizeof(ss));
984
Willy Tarreaue835bd82020-09-16 11:35:47 +0200985 /* prepare the default socket types */
Willy Tarreauf23b1bc2021-03-23 18:36:37 +0100986 if ((opts & (PA_O_STREAM|PA_O_DGRAM)) == PA_O_DGRAM ||
Willy Tarreaue3b45182021-10-27 17:28:55 +0200987 ((opts & (PA_O_STREAM|PA_O_DGRAM)) == (PA_O_DGRAM|PA_O_STREAM) && (opts & PA_O_DEFAULT_DGRAM))) {
988 proto_type = PROTO_TYPE_DGRAM;
989 ctrl_type = SOCK_DGRAM;
990 } else {
991 proto_type = PROTO_TYPE_STREAM;
992 ctrl_type = SOCK_STREAM;
993 }
Willy Tarreaue835bd82020-09-16 11:35:47 +0200994
995 if (strncmp(str2, "stream+", 7) == 0) {
996 str2 += 7;
Willy Tarreaue3b45182021-10-27 17:28:55 +0200997 proto_type = PROTO_TYPE_STREAM;
998 ctrl_type = SOCK_STREAM;
Willy Tarreaue835bd82020-09-16 11:35:47 +0200999 }
1000 else if (strncmp(str2, "dgram+", 6) == 0) {
1001 str2 += 6;
Willy Tarreaue3b45182021-10-27 17:28:55 +02001002 proto_type = PROTO_TYPE_DGRAM;
1003 ctrl_type = SOCK_DGRAM;
Willy Tarreaue835bd82020-09-16 11:35:47 +02001004 }
Willy Tarreau40725a42023-01-16 13:55:27 +01001005 else if (strncmp(str2, "quic+", 5) == 0) {
1006 str2 += 5;
1007 proto_type = PROTO_TYPE_DGRAM;
1008 ctrl_type = SOCK_STREAM;
1009 }
Willy Tarreaue835bd82020-09-16 11:35:47 +02001010
Willy Tarreau24709282013-03-10 21:32:12 +01001011 if (strncmp(str2, "unix@", 5) == 0) {
1012 str2 += 5;
Willy Tarreauccfccef2014-05-10 01:49:15 +02001013 abstract = 0;
Willy Tarreau24709282013-03-10 21:32:12 +01001014 ss.ss_family = AF_UNIX;
1015 }
Emeric Brunce325c42021-04-02 17:05:09 +02001016 else if (strncmp(str2, "uxdg@", 5) == 0) {
1017 str2 += 5;
1018 abstract = 0;
1019 ss.ss_family = AF_UNIX;
Willy Tarreaue3b45182021-10-27 17:28:55 +02001020 proto_type = PROTO_TYPE_DGRAM;
1021 ctrl_type = SOCK_DGRAM;
Emeric Brunce325c42021-04-02 17:05:09 +02001022 }
1023 else if (strncmp(str2, "uxst@", 5) == 0) {
1024 str2 += 5;
1025 abstract = 0;
1026 ss.ss_family = AF_UNIX;
Willy Tarreaue3b45182021-10-27 17:28:55 +02001027 proto_type = PROTO_TYPE_STREAM;
1028 ctrl_type = SOCK_STREAM;
Emeric Brunce325c42021-04-02 17:05:09 +02001029 }
Willy Tarreauccfccef2014-05-10 01:49:15 +02001030 else if (strncmp(str2, "abns@", 5) == 0) {
1031 str2 += 5;
1032 abstract = 1;
1033 ss.ss_family = AF_UNIX;
1034 }
Emeric Brunce325c42021-04-02 17:05:09 +02001035 else if (strncmp(str2, "ip@", 3) == 0) {
1036 str2 += 3;
1037 ss.ss_family = AF_UNSPEC;
1038 }
Willy Tarreau24709282013-03-10 21:32:12 +01001039 else if (strncmp(str2, "ipv4@", 5) == 0) {
1040 str2 += 5;
1041 ss.ss_family = AF_INET;
1042 }
1043 else if (strncmp(str2, "ipv6@", 5) == 0) {
1044 str2 += 5;
1045 ss.ss_family = AF_INET6;
1046 }
Emeric Brunce325c42021-04-02 17:05:09 +02001047 else if (strncmp(str2, "tcp4@", 5) == 0) {
1048 str2 += 5;
1049 ss.ss_family = AF_INET;
Willy Tarreaue3b45182021-10-27 17:28:55 +02001050 proto_type = PROTO_TYPE_STREAM;
1051 ctrl_type = SOCK_STREAM;
Emeric Brunce325c42021-04-02 17:05:09 +02001052 }
Emeric Brun3835c0d2020-07-07 09:46:09 +02001053 else if (strncmp(str2, "udp4@", 5) == 0) {
1054 str2 += 5;
1055 ss.ss_family = AF_INET;
Willy Tarreaue3b45182021-10-27 17:28:55 +02001056 proto_type = PROTO_TYPE_DGRAM;
1057 ctrl_type = SOCK_DGRAM;
Emeric Brun3835c0d2020-07-07 09:46:09 +02001058 }
Emeric Brunce325c42021-04-02 17:05:09 +02001059 else if (strncmp(str2, "tcp6@", 5) == 0) {
1060 str2 += 5;
1061 ss.ss_family = AF_INET6;
Willy Tarreaue3b45182021-10-27 17:28:55 +02001062 proto_type = PROTO_TYPE_STREAM;
1063 ctrl_type = SOCK_STREAM;
Emeric Brunce325c42021-04-02 17:05:09 +02001064 }
Emeric Brun3835c0d2020-07-07 09:46:09 +02001065 else if (strncmp(str2, "udp6@", 5) == 0) {
1066 str2 += 5;
1067 ss.ss_family = AF_INET6;
Willy Tarreaue3b45182021-10-27 17:28:55 +02001068 proto_type = PROTO_TYPE_DGRAM;
1069 ctrl_type = SOCK_DGRAM;
Emeric Brun3835c0d2020-07-07 09:46:09 +02001070 }
Emeric Brunce325c42021-04-02 17:05:09 +02001071 else if (strncmp(str2, "tcp@", 4) == 0) {
1072 str2 += 4;
1073 ss.ss_family = AF_UNSPEC;
Willy Tarreaue3b45182021-10-27 17:28:55 +02001074 proto_type = PROTO_TYPE_STREAM;
1075 ctrl_type = SOCK_STREAM;
Emeric Brunce325c42021-04-02 17:05:09 +02001076 }
Emeric Brun3835c0d2020-07-07 09:46:09 +02001077 else if (strncmp(str2, "udp@", 4) == 0) {
1078 str2 += 4;
1079 ss.ss_family = AF_UNSPEC;
Willy Tarreaue3b45182021-10-27 17:28:55 +02001080 proto_type = PROTO_TYPE_DGRAM;
1081 ctrl_type = SOCK_DGRAM;
Emeric Brun3835c0d2020-07-07 09:46:09 +02001082 }
Frédéric Lécaille10caf652020-11-23 11:36:57 +01001083 else if (strncmp(str2, "quic4@", 6) == 0) {
1084 str2 += 6;
1085 ss.ss_family = AF_INET;
Willy Tarreaue3b45182021-10-27 17:28:55 +02001086 proto_type = PROTO_TYPE_DGRAM;
Frédéric Lécaille10caf652020-11-23 11:36:57 +01001087 ctrl_type = SOCK_STREAM;
1088 }
1089 else if (strncmp(str2, "quic6@", 6) == 0) {
1090 str2 += 6;
1091 ss.ss_family = AF_INET6;
Willy Tarreaue3b45182021-10-27 17:28:55 +02001092 proto_type = PROTO_TYPE_DGRAM;
Frédéric Lécaille10caf652020-11-23 11:36:57 +01001093 ctrl_type = SOCK_STREAM;
1094 }
Willy Tarreau5a7beed2020-09-04 16:54:05 +02001095 else if (strncmp(str2, "fd@", 3) == 0) {
1096 str2 += 3;
1097 ss.ss_family = AF_CUST_EXISTING_FD;
1098 }
1099 else if (strncmp(str2, "sockpair@", 9) == 0) {
1100 str2 += 9;
1101 ss.ss_family = AF_CUST_SOCKPAIR;
1102 }
Willy Tarreau24709282013-03-10 21:32:12 +01001103 else if (*str2 == '/') {
1104 ss.ss_family = AF_UNIX;
1105 }
1106 else
1107 ss.ss_family = AF_UNSPEC;
1108
Willy Tarreau5a7beed2020-09-04 16:54:05 +02001109 if (ss.ss_family == AF_CUST_SOCKPAIR) {
Willy Tarreaua215be22020-09-16 10:14:16 +02001110 struct sockaddr_storage ss2;
1111 socklen_t addr_len;
William Lallemand2fe7dd02018-09-11 16:51:29 +02001112 char *endptr;
1113
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001114 new_fd = strtol(str2, &endptr, 10);
1115 if (!*str2 || new_fd < 0 || *endptr) {
William Lallemand2fe7dd02018-09-11 16:51:29 +02001116 memprintf(err, "file descriptor '%s' is not a valid integer in '%s'\n", str2, str);
1117 goto out;
1118 }
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001119
Willy Tarreaua215be22020-09-16 10:14:16 +02001120 /* just verify that it's a socket */
1121 addr_len = sizeof(ss2);
1122 if (getsockname(new_fd, (struct sockaddr *)&ss2, &addr_len) == -1) {
1123 memprintf(err, "cannot use file descriptor '%d' : %s.\n", new_fd, strerror(errno));
1124 goto out;
1125 }
1126
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001127 ((struct sockaddr_in *)&ss)->sin_addr.s_addr = new_fd;
1128 ((struct sockaddr_in *)&ss)->sin_port = 0;
William Lallemand2fe7dd02018-09-11 16:51:29 +02001129 }
Willy Tarreau5a7beed2020-09-04 16:54:05 +02001130 else if (ss.ss_family == AF_CUST_EXISTING_FD) {
Willy Tarreau40aa0702013-03-10 23:51:38 +01001131 char *endptr;
1132
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001133 new_fd = strtol(str2, &endptr, 10);
1134 if (!*str2 || new_fd < 0 || *endptr) {
Willy Tarreaudad36a32013-03-11 01:20:04 +01001135 memprintf(err, "file descriptor '%s' is not a valid integer in '%s'\n", str2, str);
Willy Tarreau40aa0702013-03-10 23:51:38 +01001136 goto out;
1137 }
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001138
Willy Tarreau6edc7222020-09-15 17:41:56 +02001139 if (opts & PA_O_SOCKET_FD) {
1140 socklen_t addr_len;
1141 int type;
1142
1143 addr_len = sizeof(ss);
1144 if (getsockname(new_fd, (struct sockaddr *)&ss, &addr_len) == -1) {
1145 memprintf(err, "cannot use file descriptor '%d' : %s.\n", new_fd, strerror(errno));
1146 goto out;
1147 }
1148
1149 addr_len = sizeof(type);
1150 if (getsockopt(new_fd, SOL_SOCKET, SO_TYPE, &type, &addr_len) != 0 ||
Willy Tarreaue3b45182021-10-27 17:28:55 +02001151 (type == SOCK_STREAM) != (proto_type == PROTO_TYPE_STREAM)) {
Willy Tarreau6edc7222020-09-15 17:41:56 +02001152 memprintf(err, "socket on file descriptor '%d' is of the wrong type.\n", new_fd);
1153 goto out;
1154 }
1155
1156 porta = portl = porth = get_host_port(&ss);
1157 } else if (opts & PA_O_RAW_FD) {
1158 ((struct sockaddr_in *)&ss)->sin_addr.s_addr = new_fd;
1159 ((struct sockaddr_in *)&ss)->sin_port = 0;
1160 } else {
1161 memprintf(err, "a file descriptor is not acceptable here in '%s'\n", str);
1162 goto out;
1163 }
Willy Tarreau40aa0702013-03-10 23:51:38 +01001164 }
1165 else if (ss.ss_family == AF_UNIX) {
Willy Tarreau8daa9202019-06-16 18:16:33 +02001166 struct sockaddr_un *un = (struct sockaddr_un *)&ss;
Willy Tarreau15586382013-03-04 19:48:14 +01001167 int prefix_path_len;
1168 int max_path_len;
Willy Tarreau94ef3f32014-04-14 14:49:00 +02001169 int adr_len;
Willy Tarreau15586382013-03-04 19:48:14 +01001170
1171 /* complete unix socket path name during startup or soft-restart is
1172 * <unix_bind_prefix><path>.<pid>.<bak|tmp>
1173 */
Willy Tarreauccfccef2014-05-10 01:49:15 +02001174 prefix_path_len = (pfx && !abstract) ? strlen(pfx) : 0;
Willy Tarreau8daa9202019-06-16 18:16:33 +02001175 max_path_len = (sizeof(un->sun_path) - 1) -
Willy Tarreau327ea5a2020-02-11 06:43:37 +01001176 (abstract ? 0 : prefix_path_len + 1 + 5 + 1 + 3);
Willy Tarreau15586382013-03-04 19:48:14 +01001177
Willy Tarreau94ef3f32014-04-14 14:49:00 +02001178 adr_len = strlen(str2);
1179 if (adr_len > max_path_len) {
Willy Tarreau15586382013-03-04 19:48:14 +01001180 memprintf(err, "socket path '%s' too long (max %d)\n", str, max_path_len);
1181 goto out;
1182 }
1183
Willy Tarreauccfccef2014-05-10 01:49:15 +02001184 /* when abstract==1, we skip the first zero and copy all bytes except the trailing zero */
Willy Tarreau8daa9202019-06-16 18:16:33 +02001185 memset(un->sun_path, 0, sizeof(un->sun_path));
Willy Tarreau94ef3f32014-04-14 14:49:00 +02001186 if (prefix_path_len)
Willy Tarreau8daa9202019-06-16 18:16:33 +02001187 memcpy(un->sun_path, pfx, prefix_path_len);
1188 memcpy(un->sun_path + prefix_path_len + abstract, str2, adr_len + 1 - abstract);
Willy Tarreau15586382013-03-04 19:48:14 +01001189 }
Willy Tarreau24709282013-03-10 21:32:12 +01001190 else { /* IPv4 and IPv6 */
mildisff5d5102015-10-26 18:50:08 +01001191 char *end = str2 + strlen(str2);
1192 char *chr;
Willy Tarreau72b8c1f2015-09-08 15:50:19 +02001193
mildisff5d5102015-10-26 18:50:08 +01001194 /* search for : or ] whatever comes first */
1195 for (chr = end-1; chr > str2; chr--) {
1196 if (*chr == ']' || *chr == ':')
1197 break;
1198 }
1199
1200 if (*chr == ':') {
1201 /* Found a colon before a closing-bracket, must be a port separator.
1202 * This guarantee backward compatibility.
1203 */
Willy Tarreau7f96a842020-09-15 11:12:44 +02001204 if (!(opts & PA_O_PORT_OK)) {
1205 memprintf(err, "port specification not permitted here in '%s'", str);
1206 goto out;
1207 }
mildisff5d5102015-10-26 18:50:08 +01001208 *chr++ = '\0';
1209 port1 = chr;
1210 }
1211 else {
1212 /* Either no colon and no closing-bracket
1213 * or directly ending with a closing-bracket.
1214 * However, no port.
1215 */
Willy Tarreau7f96a842020-09-15 11:12:44 +02001216 if (opts & PA_O_PORT_MAND) {
1217 memprintf(err, "missing port specification in '%s'", str);
1218 goto out;
1219 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001220 port1 = "";
mildisff5d5102015-10-26 18:50:08 +01001221 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02001222
Willy Tarreau90807112020-02-25 08:16:33 +01001223 if (isdigit((unsigned char)*port1)) { /* single port or range */
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001224 port2 = strchr(port1, '-');
Willy Tarreau7f96a842020-09-15 11:12:44 +02001225 if (port2) {
1226 if (!(opts & PA_O_PORT_RANGE)) {
1227 memprintf(err, "port range not permitted here in '%s'", str);
1228 goto out;
1229 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001230 *port2++ = '\0';
Willy Tarreau7f96a842020-09-15 11:12:44 +02001231 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001232 else
1233 port2 = port1;
1234 portl = atoi(port1);
1235 porth = atoi(port2);
Willy Tarreau7f96a842020-09-15 11:12:44 +02001236
1237 if (portl < !!(opts & PA_O_PORT_MAND) || portl > 65535) {
1238 memprintf(err, "invalid port '%s'", port1);
1239 goto out;
1240 }
1241
1242 if (porth < !!(opts & PA_O_PORT_MAND) || porth > 65535) {
1243 memprintf(err, "invalid port '%s'", port2);
1244 goto out;
1245 }
1246
1247 if (portl > porth) {
1248 memprintf(err, "invalid port range '%d-%d'", portl, porth);
1249 goto out;
1250 }
1251
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001252 porta = portl;
1253 }
1254 else if (*port1 == '-') { /* negative offset */
Willy Tarreau7f96a842020-09-15 11:12:44 +02001255 if (!(opts & PA_O_PORT_OFS)) {
1256 memprintf(err, "port offset not permitted here in '%s'", str);
1257 goto out;
1258 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001259 portl = atoi(port1 + 1);
1260 porta = -portl;
1261 }
1262 else if (*port1 == '+') { /* positive offset */
Willy Tarreau7f96a842020-09-15 11:12:44 +02001263 if (!(opts & PA_O_PORT_OFS)) {
1264 memprintf(err, "port offset not permitted here in '%s'", str);
1265 goto out;
1266 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001267 porth = atoi(port1 + 1);
1268 porta = porth;
1269 }
1270 else if (*port1) { /* other any unexpected char */
Willy Tarreaudad36a32013-03-11 01:20:04 +01001271 memprintf(err, "invalid character '%c' in port number '%s' in '%s'\n", *port1, port1, str);
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001272 goto out;
1273 }
Willy Tarreau7f96a842020-09-15 11:12:44 +02001274 else if (opts & PA_O_PORT_MAND) {
1275 memprintf(err, "missing port specification in '%s'", str);
1276 goto out;
1277 }
Willy Tarreauceccdd72016-11-02 22:27:10 +01001278
1279 /* first try to parse the IP without resolving. If it fails, it
1280 * tells us we need to keep a copy of the FQDN to resolve later
1281 * and to enable DNS. In this case we can proceed if <fqdn> is
Willy Tarreaucd3a55912020-09-04 15:30:46 +02001282 * set or if PA_O_RESOLVE is set, otherwise it's an error.
Willy Tarreauceccdd72016-11-02 22:27:10 +01001283 */
1284 if (str2ip2(str2, &ss, 0) == NULL) {
Willy Tarreaucd3a55912020-09-04 15:30:46 +02001285 if ((!(opts & PA_O_RESOLVE) && !fqdn) ||
1286 ((opts & PA_O_RESOLVE) && str2ip2(str2, &ss, 1) == NULL)) {
Willy Tarreauceccdd72016-11-02 22:27:10 +01001287 memprintf(err, "invalid address: '%s' in '%s'\n", str2, str);
1288 goto out;
1289 }
Willy Tarreau72b8c1f2015-09-08 15:50:19 +02001290
Willy Tarreauceccdd72016-11-02 22:27:10 +01001291 if (fqdn) {
1292 if (str2 != back)
1293 memmove(back, str2, strlen(str2) + 1);
1294 *fqdn = back;
1295 back = NULL;
1296 }
Willy Tarreau72b8c1f2015-09-08 15:50:19 +02001297 }
Willy Tarreauceccdd72016-11-02 22:27:10 +01001298 set_host_port(&ss, porta);
Willy Tarreaue4c58c82013-03-06 15:28:17 +01001299 }
Willy Tarreaufab5a432011-03-04 15:31:53 +01001300
Willy Tarreaue835bd82020-09-16 11:35:47 +02001301 if (ctrl_type == SOCK_STREAM && !(opts & PA_O_STREAM)) {
Willy Tarreau8d31ab02022-05-09 16:18:26 +02001302 memprintf(err, "stream-type address not acceptable in '%s'\n", str);
Willy Tarreaue835bd82020-09-16 11:35:47 +02001303 goto out;
1304 }
1305 else if (ctrl_type == SOCK_DGRAM && !(opts & PA_O_DGRAM)) {
Willy Tarreau8d31ab02022-05-09 16:18:26 +02001306 memprintf(err, "dgram-type address not acceptable in '%s'\n", str);
Willy Tarreaue835bd82020-09-16 11:35:47 +02001307 goto out;
1308 }
1309
Willy Tarreau65ec4e32020-09-16 19:17:08 +02001310 if (proto || (opts & PA_O_CONNECT)) {
Willy Tarreau5fc93282020-09-16 18:25:03 +02001311 /* Note: if the caller asks for a proto, we must find one,
Emeric Brun26754902021-04-07 14:26:44 +02001312 * except if we inherit from a raw FD (family == AF_CUST_EXISTING_FD)
1313 * orif we return with an fqdn that will resolve later,
Willy Tarreau5fc93282020-09-16 18:25:03 +02001314 * in which case the address is not known yet (this is only
1315 * for servers actually).
1316 */
Willy Tarreaub2ffc992020-09-16 21:37:31 +02001317 new_proto = protocol_lookup(ss.ss_family,
Willy Tarreaue3b45182021-10-27 17:28:55 +02001318 proto_type,
Willy Tarreauaf9609b2020-09-16 22:04:46 +02001319 ctrl_type == SOCK_DGRAM);
Willy Tarreaub2ffc992020-09-16 21:37:31 +02001320
Emeric Brun26754902021-04-07 14:26:44 +02001321 if (!new_proto && (!fqdn || !*fqdn) && (ss.ss_family != AF_CUST_EXISTING_FD)) {
Willy Tarreau3d7b4682022-05-20 17:32:35 +02001322 memprintf(err, "unsupported %s protocol for %s family %d address '%s'%s",
Willy Tarreau2b049b82022-05-20 17:28:30 +02001323 (ctrl_type == SOCK_DGRAM) ? "datagram" : "stream",
1324 (proto_type == PROTO_TYPE_DGRAM) ? "datagram" : "stream",
1325 ss.ss_family,
Willy Tarreau3d7b4682022-05-20 17:32:35 +02001326 str,
Willy Tarreau3d7b4682022-05-20 17:32:35 +02001327#ifndef USE_QUIC
Tim Duesterhus147eeb22022-05-22 12:40:58 +02001328 (ctrl_type == SOCK_STREAM && proto_type == PROTO_TYPE_DGRAM)
1329 ? "; QUIC is not compiled in if this is what you were looking for."
1330 : ""
Willy Tarreau3d7b4682022-05-20 17:32:35 +02001331#else
1332 ""
1333#endif
Tim Duesterhus147eeb22022-05-22 12:40:58 +02001334 );
Willy Tarreau5fc93282020-09-16 18:25:03 +02001335 goto out;
1336 }
Willy Tarreau65ec4e32020-09-16 19:17:08 +02001337
1338 if ((opts & PA_O_CONNECT) && new_proto && !new_proto->connect) {
1339 memprintf(err, "connect() not supported for this protocol family %d used by address '%s'", ss.ss_family, str);
1340 goto out;
1341 }
Willy Tarreau5fc93282020-09-16 18:25:03 +02001342 }
1343
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001344 ret = &ss;
Willy Tarreaud5191e72010-02-09 20:50:45 +01001345 out:
Willy Tarreau48ef4c92017-01-06 18:32:38 +01001346 if (port)
1347 *port = porta;
Willy Tarreaud4448bc2013-02-20 15:55:15 +01001348 if (low)
1349 *low = portl;
1350 if (high)
1351 *high = porth;
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001352 if (fd)
1353 *fd = new_fd;
Willy Tarreau5fc93282020-09-16 18:25:03 +02001354 if (proto)
1355 *proto = new_proto;
Willy Tarreau24709282013-03-10 21:32:12 +01001356 free(back);
Willy Tarreaud5191e72010-02-09 20:50:45 +01001357 return ret;
Willy Tarreauc6f4ce82009-06-10 11:09:37 +02001358}
1359
Thayne McCombs92149f92020-11-20 01:28:26 -07001360/* converts <addr> and <port> into a string representation of the address and port. This is sort
1361 * of an inverse of str2sa_range, with some restrictions. The supported families are AF_INET,
1362 * AF_INET6, AF_UNIX, and AF_CUST_SOCKPAIR. If the family is unsopported NULL is returned.
1363 * If map_ports is true, then the sign of the port is included in the output, to indicate it is
1364 * relative to the incoming port. AF_INET and AF_INET6 will be in the form "<addr>:<port>".
1365 * AF_UNIX will either be just the path (if using a pathname) or "abns@<path>" if it is abstract.
1366 * AF_CUST_SOCKPAIR will be of the form "sockpair@<fd>".
1367 *
1368 * The returned char* is allocated, and it is the responsibility of the caller to free it.
1369 */
1370char * sa2str(const struct sockaddr_storage *addr, int port, int map_ports)
1371{
1372 char buffer[INET6_ADDRSTRLEN];
1373 char *out = NULL;
1374 const void *ptr;
1375 const char *path;
1376
1377 switch (addr->ss_family) {
1378 case AF_INET:
1379 ptr = &((struct sockaddr_in *)addr)->sin_addr;
1380 break;
1381 case AF_INET6:
1382 ptr = &((struct sockaddr_in6 *)addr)->sin6_addr;
1383 break;
1384 case AF_UNIX:
1385 path = ((struct sockaddr_un *)addr)->sun_path;
1386 if (path[0] == '\0') {
1387 const int max_length = sizeof(struct sockaddr_un) - offsetof(struct sockaddr_un, sun_path) - 1;
1388 return memprintf(&out, "abns@%.*s", max_length, path+1);
1389 } else {
1390 return strdup(path);
1391 }
1392 case AF_CUST_SOCKPAIR:
1393 return memprintf(&out, "sockpair@%d", ((struct sockaddr_in *)addr)->sin_addr.s_addr);
1394 default:
1395 return NULL;
1396 }
Tim Duesterhus22535a52022-05-23 09:30:49 +02001397 if (inet_ntop(addr->ss_family, ptr, buffer, sizeof(buffer)) == NULL) {
1398 BUG_ON(errno == ENOSPC);
1399 return NULL;
1400 }
Thayne McCombs92149f92020-11-20 01:28:26 -07001401 if (map_ports)
1402 return memprintf(&out, "%s:%+d", buffer, port);
1403 else
1404 return memprintf(&out, "%s:%d", buffer, port);
1405}
1406
1407
Willy Tarreau2937c0d2010-01-26 17:36:17 +01001408/* converts <str> to a struct in_addr containing a network mask. It can be
1409 * passed in dotted form (255.255.255.0) or in CIDR form (24). It returns 1
Jarno Huuskonen577d5ac2017-05-21 17:32:21 +03001410 * if the conversion succeeds otherwise zero.
Willy Tarreau2937c0d2010-01-26 17:36:17 +01001411 */
1412int str2mask(const char *str, struct in_addr *mask)
1413{
1414 if (strchr(str, '.') != NULL) { /* dotted notation */
1415 if (!inet_pton(AF_INET, str, mask))
1416 return 0;
1417 }
1418 else { /* mask length */
1419 char *err;
1420 unsigned long len = strtol(str, &err, 10);
1421
1422 if (!*str || (err && *err) || (unsigned)len > 32)
1423 return 0;
Tim Duesterhus8575f722018-01-25 16:24:48 +01001424
1425 len2mask4(len, mask);
Willy Tarreau2937c0d2010-01-26 17:36:17 +01001426 }
1427 return 1;
1428}
1429
Tim Duesterhus47185172018-01-25 16:24:49 +01001430/* converts <str> to a struct in6_addr containing a network mask. It can be
Tim Duesterhus5e642862018-02-20 17:02:18 +01001431 * passed in quadruplet form (ffff:ffff::) or in CIDR form (64). It returns 1
Tim Duesterhus47185172018-01-25 16:24:49 +01001432 * if the conversion succeeds otherwise zero.
1433 */
1434int str2mask6(const char *str, struct in6_addr *mask)
1435{
1436 if (strchr(str, ':') != NULL) { /* quadruplet notation */
1437 if (!inet_pton(AF_INET6, str, mask))
1438 return 0;
1439 }
1440 else { /* mask length */
1441 char *err;
1442 unsigned long len = strtol(str, &err, 10);
1443
1444 if (!*str || (err && *err) || (unsigned)len > 128)
1445 return 0;
1446
1447 len2mask6(len, mask);
1448 }
1449 return 1;
1450}
1451
Thierry FOURNIERb0504632013-12-14 15:39:02 +01001452/* convert <cidr> to struct in_addr <mask>. It returns 1 if the conversion
1453 * succeeds otherwise zero.
1454 */
1455int cidr2dotted(int cidr, struct in_addr *mask) {
1456
1457 if (cidr < 0 || cidr > 32)
1458 return 0;
1459
1460 mask->s_addr = cidr ? htonl(~0UL << (32 - cidr)) : 0;
1461 return 1;
1462}
1463
Thierry Fournier70473a52016-02-17 17:12:14 +01001464/* Convert mask from bit length form to in_addr form.
1465 * This function never fails.
1466 */
1467void len2mask4(int len, struct in_addr *addr)
1468{
1469 if (len >= 32) {
1470 addr->s_addr = 0xffffffff;
1471 return;
1472 }
1473 if (len <= 0) {
1474 addr->s_addr = 0x00000000;
1475 return;
1476 }
1477 addr->s_addr = 0xffffffff << (32 - len);
1478 addr->s_addr = htonl(addr->s_addr);
1479}
1480
1481/* Convert mask from bit length form to in6_addr form.
1482 * This function never fails.
1483 */
1484void len2mask6(int len, struct in6_addr *addr)
1485{
1486 len2mask4(len, (struct in_addr *)&addr->s6_addr[0]); /* msb */
1487 len -= 32;
1488 len2mask4(len, (struct in_addr *)&addr->s6_addr[4]);
1489 len -= 32;
1490 len2mask4(len, (struct in_addr *)&addr->s6_addr[8]);
1491 len -= 32;
1492 len2mask4(len, (struct in_addr *)&addr->s6_addr[12]); /* lsb */
1493}
1494
Willy Tarreauc6f4ce82009-06-10 11:09:37 +02001495/*
Willy Tarreaud077a8e2007-05-08 18:28:09 +02001496 * converts <str> to two struct in_addr* which must be pre-allocated.
Willy Tarreaubaaee002006-06-26 02:48:02 +02001497 * The format is "addr[/mask]", where "addr" cannot be empty, and mask
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05001498 * is optional and either in the dotted or CIDR notation.
Willy Tarreaubaaee002006-06-26 02:48:02 +02001499 * Note: "addr" can also be a hostname. Returns 1 if OK, 0 if error.
1500 */
Thierry FOURNIERfc7ac7b2014-02-11 15:23:04 +01001501int str2net(const char *str, int resolve, struct in_addr *addr, struct in_addr *mask)
Willy Tarreaubaaee002006-06-26 02:48:02 +02001502{
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001503 __label__ out_free, out_err;
1504 char *c, *s;
1505 int ret_val;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001506
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001507 s = strdup(str);
1508 if (!s)
1509 return 0;
1510
Willy Tarreaubaaee002006-06-26 02:48:02 +02001511 memset(mask, 0, sizeof(*mask));
1512 memset(addr, 0, sizeof(*addr));
Willy Tarreaubaaee002006-06-26 02:48:02 +02001513
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001514 if ((c = strrchr(s, '/')) != NULL) {
Willy Tarreaubaaee002006-06-26 02:48:02 +02001515 *c++ = '\0';
1516 /* c points to the mask */
Willy Tarreau2937c0d2010-01-26 17:36:17 +01001517 if (!str2mask(c, mask))
1518 goto out_err;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001519 }
1520 else {
Willy Tarreauebd61602006-12-30 11:54:15 +01001521 mask->s_addr = ~0U;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001522 }
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001523 if (!inet_pton(AF_INET, s, addr)) {
Willy Tarreaubaaee002006-06-26 02:48:02 +02001524 struct hostent *he;
1525
Thierry FOURNIERfc7ac7b2014-02-11 15:23:04 +01001526 if (!resolve)
1527 goto out_err;
1528
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001529 if ((he = gethostbyname(s)) == NULL) {
1530 goto out_err;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001531 }
1532 else
1533 *addr = *(struct in_addr *) *(he->h_addr_list);
1534 }
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001535
1536 ret_val = 1;
1537 out_free:
1538 free(s);
1539 return ret_val;
1540 out_err:
1541 ret_val = 0;
1542 goto out_free;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001543}
1544
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001545
1546/*
Willy Tarreau6d20e282012-04-27 22:49:47 +02001547 * converts <str> to two struct in6_addr* which must be pre-allocated.
1548 * The format is "addr[/mask]", where "addr" cannot be empty, and mask
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05001549 * is an optional number of bits (128 being the default).
Willy Tarreau6d20e282012-04-27 22:49:47 +02001550 * Returns 1 if OK, 0 if error.
1551 */
1552int str62net(const char *str, struct in6_addr *addr, unsigned char *mask)
1553{
1554 char *c, *s;
1555 int ret_val = 0;
1556 char *err;
1557 unsigned long len = 128;
1558
1559 s = strdup(str);
1560 if (!s)
1561 return 0;
1562
1563 memset(mask, 0, sizeof(*mask));
1564 memset(addr, 0, sizeof(*addr));
1565
1566 if ((c = strrchr(s, '/')) != NULL) {
1567 *c++ = '\0'; /* c points to the mask */
1568 if (!*c)
1569 goto out_free;
1570
1571 len = strtoul(c, &err, 10);
1572 if ((err && *err) || (unsigned)len > 128)
1573 goto out_free;
1574 }
1575 *mask = len; /* OK we have a valid mask in <len> */
1576
1577 if (!inet_pton(AF_INET6, s, addr))
1578 goto out_free;
1579
1580 ret_val = 1;
1581 out_free:
1582 free(s);
1583 return ret_val;
1584}
1585
1586
1587/*
Willy Tarreau12e10272021-03-25 11:34:40 +01001588 * Parse IPv4 address found in url. Return the number of bytes parsed. It
1589 * expects exactly 4 numbers between 0 and 255 delimited by dots, and returns
1590 * zero in case of mismatch.
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001591 */
David du Colombier6f5ccb12011-03-10 22:26:24 +01001592int url2ipv4(const char *addr, struct in_addr *dst)
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001593{
1594 int saw_digit, octets, ch;
1595 u_char tmp[4], *tp;
1596 const char *cp = addr;
1597
1598 saw_digit = 0;
1599 octets = 0;
1600 *(tp = tmp) = 0;
1601
1602 while (*addr) {
Willy Tarreau12e10272021-03-25 11:34:40 +01001603 unsigned char digit = (ch = *addr) - '0';
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001604 if (digit > 9 && ch != '.')
1605 break;
Willy Tarreau12e10272021-03-25 11:34:40 +01001606 addr++;
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001607 if (digit <= 9) {
1608 u_int new = *tp * 10 + digit;
1609 if (new > 255)
1610 return 0;
1611 *tp = new;
1612 if (!saw_digit) {
1613 if (++octets > 4)
1614 return 0;
1615 saw_digit = 1;
1616 }
1617 } else if (ch == '.' && saw_digit) {
1618 if (octets == 4)
1619 return 0;
1620 *++tp = 0;
1621 saw_digit = 0;
1622 } else
1623 return 0;
1624 }
1625
1626 if (octets < 4)
1627 return 0;
1628
1629 memcpy(&dst->s_addr, tmp, 4);
Willy Tarreau12e10272021-03-25 11:34:40 +01001630 return addr - cp;
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001631}
1632
1633/*
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001634 * Resolve destination server from URL. Convert <str> to a sockaddr_storage.
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05001635 * <out> contain the code of the detected scheme, the start and length of
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001636 * the hostname. Actually only http and https are supported. <out> can be NULL.
1637 * This function returns the consumed length. It is useful if you parse complete
1638 * url like http://host:port/path, because the consumed length corresponds to
1639 * the first character of the path. If the conversion fails, it returns -1.
1640 *
1641 * This function tries to resolve the DNS name if haproxy is in starting mode.
1642 * So, this function may be used during the configuration parsing.
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001643 */
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001644int url2sa(const char *url, int ulen, struct sockaddr_storage *addr, struct split_url *out)
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001645{
1646 const char *curr = url, *cp = url;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001647 const char *end;
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001648 int ret, url_code = 0;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001649 unsigned long long int http_code = 0;
1650 int default_port;
1651 struct hostent *he;
1652 char *p;
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001653
1654 /* Firstly, try to find :// pattern */
1655 while (curr < url+ulen && url_code != 0x3a2f2f) {
1656 url_code = ((url_code & 0xffff) << 8);
1657 url_code += (unsigned char)*curr++;
1658 }
1659
1660 /* Secondly, if :// pattern is found, verify parsed stuff
1661 * before pattern is matching our http pattern.
1662 * If so parse ip address and port in uri.
1663 *
1664 * WARNING: Current code doesn't support dynamic async dns resolver.
1665 */
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001666 if (url_code != 0x3a2f2f)
1667 return -1;
1668
1669 /* Copy scheme, and utrn to lower case. */
1670 while (cp < curr - 3)
1671 http_code = (http_code << 8) + *cp++;
1672 http_code |= 0x2020202020202020ULL; /* Turn everything to lower case */
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001673
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001674 /* HTTP or HTTPS url matching */
1675 if (http_code == 0x2020202068747470ULL) {
1676 default_port = 80;
1677 if (out)
1678 out->scheme = SCH_HTTP;
1679 }
1680 else if (http_code == 0x2020206874747073ULL) {
1681 default_port = 443;
1682 if (out)
1683 out->scheme = SCH_HTTPS;
1684 }
1685 else
1686 return -1;
1687
1688 /* If the next char is '[', the host address is IPv6. */
1689 if (*curr == '[') {
1690 curr++;
1691
1692 /* Check trash size */
1693 if (trash.size < ulen)
1694 return -1;
1695
1696 /* Look for ']' and copy the address in a trash buffer. */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001697 p = trash.area;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001698 for (end = curr;
1699 end < url + ulen && *end != ']';
1700 end++, p++)
1701 *p = *end;
1702 if (*end != ']')
1703 return -1;
1704 *p = '\0';
1705
1706 /* Update out. */
1707 if (out) {
1708 out->host = curr;
1709 out->host_len = end - curr;
1710 }
1711
1712 /* Try IPv6 decoding. */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001713 if (!inet_pton(AF_INET6, trash.area, &((struct sockaddr_in6 *)addr)->sin6_addr))
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001714 return -1;
1715 end++;
1716
1717 /* Decode port. */
William Lallemand3d7a9182022-03-25 17:37:51 +01001718 if (end < url + ulen && *end == ':') {
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001719 end++;
1720 default_port = read_uint(&end, url + ulen);
1721 }
1722 ((struct sockaddr_in6 *)addr)->sin6_port = htons(default_port);
1723 ((struct sockaddr_in6 *)addr)->sin6_family = AF_INET6;
1724 return end - url;
1725 }
1726 else {
William Lallemand8a913742022-02-18 16:13:12 +01001727 /* we need to copy the string into the trash because url2ipv4
1728 * needs a \0 at the end of the string */
1729 if (trash.size < ulen)
1730 return -1;
1731
1732 memcpy(trash.area, curr, ulen - (curr - url));
1733 trash.area[ulen - (curr - url)] = '\0';
1734
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001735 /* We are looking for IP address. If you want to parse and
1736 * resolve hostname found in url, you can use str2sa_range(), but
1737 * be warned this can slow down global daemon performances
1738 * while handling lagging dns responses.
1739 */
William Lallemand8a913742022-02-18 16:13:12 +01001740 ret = url2ipv4(trash.area, &((struct sockaddr_in *)addr)->sin_addr);
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001741 if (ret) {
1742 /* Update out. */
1743 if (out) {
1744 out->host = curr;
1745 out->host_len = ret;
1746 }
1747
William Lallemandb938b772022-03-24 21:59:03 +01001748 curr += ret;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001749
1750 /* Decode port. */
William Lallemand3d7a9182022-03-25 17:37:51 +01001751 if (curr < url + ulen && *curr == ':') {
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001752 curr++;
1753 default_port = read_uint(&curr, url + ulen);
1754 }
1755 ((struct sockaddr_in *)addr)->sin_port = htons(default_port);
1756
1757 /* Set family. */
1758 ((struct sockaddr_in *)addr)->sin_family = AF_INET;
1759 return curr - url;
1760 }
1761 else if (global.mode & MODE_STARTING) {
1762 /* The IPv4 and IPv6 decoding fails, maybe the url contain name. Try to execute
1763 * synchronous DNS request only if HAProxy is in the start state.
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001764 */
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001765
1766 /* look for : or / or end */
1767 for (end = curr;
1768 end < url + ulen && *end != '/' && *end != ':';
1769 end++);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001770 memcpy(trash.area, curr, end - curr);
1771 trash.area[end - curr] = '\0';
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001772
1773 /* try to resolve an IPv4/IPv6 hostname */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001774 he = gethostbyname(trash.area);
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001775 if (!he)
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001776 return -1;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001777
1778 /* Update out. */
1779 if (out) {
1780 out->host = curr;
1781 out->host_len = end - curr;
1782 }
1783
1784 /* Decode port. */
William Lallemand3d7a9182022-03-25 17:37:51 +01001785 if (end < url + ulen && *end == ':') {
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001786 end++;
1787 default_port = read_uint(&end, url + ulen);
1788 }
1789
1790 /* Copy IP address, set port and family. */
1791 switch (he->h_addrtype) {
1792 case AF_INET:
1793 ((struct sockaddr_in *)addr)->sin_addr = *(struct in_addr *) *(he->h_addr_list);
1794 ((struct sockaddr_in *)addr)->sin_port = htons(default_port);
1795 ((struct sockaddr_in *)addr)->sin_family = AF_INET;
1796 return end - url;
1797
1798 case AF_INET6:
1799 ((struct sockaddr_in6 *)addr)->sin6_addr = *(struct in6_addr *) *(he->h_addr_list);
1800 ((struct sockaddr_in6 *)addr)->sin6_port = htons(default_port);
1801 ((struct sockaddr_in6 *)addr)->sin6_family = AF_INET6;
1802 return end - url;
1803 }
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001804 }
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001805 }
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001806 return -1;
1807}
1808
Willy Tarreau631f01c2011-09-05 00:36:48 +02001809/* Tries to convert a sockaddr_storage address to text form. Upon success, the
1810 * address family is returned so that it's easy for the caller to adapt to the
1811 * output format. Zero is returned if the address family is not supported. -1
1812 * is returned upon error, with errno set. AF_INET, AF_INET6 and AF_UNIX are
1813 * supported.
1814 */
Willy Tarreaud5ec4bf2019-04-25 17:48:16 +02001815int addr_to_str(const struct sockaddr_storage *addr, char *str, int size)
Willy Tarreau631f01c2011-09-05 00:36:48 +02001816{
1817
Willy Tarreaud5ec4bf2019-04-25 17:48:16 +02001818 const void *ptr;
Willy Tarreau631f01c2011-09-05 00:36:48 +02001819
1820 if (size < 5)
1821 return 0;
1822 *str = '\0';
1823
1824 switch (addr->ss_family) {
1825 case AF_INET:
1826 ptr = &((struct sockaddr_in *)addr)->sin_addr;
1827 break;
1828 case AF_INET6:
1829 ptr = &((struct sockaddr_in6 *)addr)->sin6_addr;
1830 break;
1831 case AF_UNIX:
1832 memcpy(str, "unix", 5);
1833 return addr->ss_family;
1834 default:
1835 return 0;
1836 }
1837
1838 if (inet_ntop(addr->ss_family, ptr, str, size))
1839 return addr->ss_family;
1840
1841 /* failed */
1842 return -1;
1843}
1844
Simon Horman75ab8bd2014-06-16 09:39:41 +09001845/* Tries to convert a sockaddr_storage port to text form. Upon success, the
1846 * address family is returned so that it's easy for the caller to adapt to the
1847 * output format. Zero is returned if the address family is not supported. -1
1848 * is returned upon error, with errno set. AF_INET, AF_INET6 and AF_UNIX are
1849 * supported.
1850 */
Willy Tarreaud5ec4bf2019-04-25 17:48:16 +02001851int port_to_str(const struct sockaddr_storage *addr, char *str, int size)
Simon Horman75ab8bd2014-06-16 09:39:41 +09001852{
1853
1854 uint16_t port;
1855
1856
Willy Tarreaud7dad1b2017-01-06 16:46:22 +01001857 if (size < 6)
Simon Horman75ab8bd2014-06-16 09:39:41 +09001858 return 0;
1859 *str = '\0';
1860
1861 switch (addr->ss_family) {
1862 case AF_INET:
1863 port = ((struct sockaddr_in *)addr)->sin_port;
1864 break;
1865 case AF_INET6:
1866 port = ((struct sockaddr_in6 *)addr)->sin6_port;
1867 break;
1868 case AF_UNIX:
1869 memcpy(str, "unix", 5);
1870 return addr->ss_family;
1871 default:
1872 return 0;
1873 }
1874
1875 snprintf(str, size, "%u", ntohs(port));
1876 return addr->ss_family;
1877}
1878
Willy Tarreau16e01562016-08-09 16:46:18 +02001879/* check if the given address is local to the system or not. It will return
1880 * -1 when it's not possible to know, 0 when the address is not local, 1 when
1881 * it is. We don't want to iterate over all interfaces for this (and it is not
1882 * portable). So instead we try to bind in UDP to this address on a free non
1883 * privileged port and to connect to the same address, port 0 (connect doesn't
1884 * care). If it succeeds, we own the address. Note that non-inet addresses are
1885 * considered local since they're most likely AF_UNIX.
1886 */
1887int addr_is_local(const struct netns_entry *ns,
1888 const struct sockaddr_storage *orig)
1889{
1890 struct sockaddr_storage addr;
1891 int result;
1892 int fd;
1893
1894 if (!is_inet_addr(orig))
1895 return 1;
1896
1897 memcpy(&addr, orig, sizeof(addr));
1898 set_host_port(&addr, 0);
1899
1900 fd = my_socketat(ns, addr.ss_family, SOCK_DGRAM, IPPROTO_UDP);
1901 if (fd < 0)
1902 return -1;
1903
1904 result = -1;
1905 if (bind(fd, (struct sockaddr *)&addr, get_addr_len(&addr)) == 0) {
1906 if (connect(fd, (struct sockaddr *)&addr, get_addr_len(&addr)) == -1)
1907 result = 0; // fail, non-local address
1908 else
1909 result = 1; // success, local address
1910 }
1911 else {
1912 if (errno == EADDRNOTAVAIL)
1913 result = 0; // definitely not local :-)
1914 }
1915 close(fd);
1916
1917 return result;
1918}
1919
Willy Tarreaubaaee002006-06-26 02:48:02 +02001920/* will try to encode the string <string> replacing all characters tagged in
1921 * <map> with the hexadecimal representation of their ASCII-code (2 digits)
1922 * prefixed by <escape>, and will store the result between <start> (included)
1923 * and <stop> (excluded), and will always terminate the string with a '\0'
1924 * before <stop>. The position of the '\0' is returned if the conversion
1925 * completes. If bytes are missing between <start> and <stop>, then the
1926 * conversion will be incomplete and truncated. If <stop> <= <start>, the '\0'
1927 * cannot even be stored so we return <start> without writing the 0.
1928 * The input string must also be zero-terminated.
1929 */
1930const char hextab[16] = "0123456789ABCDEF";
1931char *encode_string(char *start, char *stop,
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001932 const char escape, const long *map,
Willy Tarreaubaaee002006-06-26 02:48:02 +02001933 const char *string)
1934{
1935 if (start < stop) {
1936 stop--; /* reserve one byte for the final '\0' */
1937 while (start < stop && *string != '\0') {
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001938 if (!ha_bit_test((unsigned char)(*string), map))
Willy Tarreaubaaee002006-06-26 02:48:02 +02001939 *start++ = *string;
1940 else {
1941 if (start + 3 >= stop)
1942 break;
1943 *start++ = escape;
1944 *start++ = hextab[(*string >> 4) & 15];
1945 *start++ = hextab[*string & 15];
1946 }
1947 string++;
1948 }
1949 *start = '\0';
1950 }
1951 return start;
1952}
1953
Thierry FOURNIERe059ec92014-03-17 12:01:13 +01001954/*
1955 * Same behavior as encode_string() above, except that it encodes chunk
1956 * <chunk> instead of a string.
1957 */
1958char *encode_chunk(char *start, char *stop,
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001959 const char escape, const long *map,
Willy Tarreau83061a82018-07-13 11:56:34 +02001960 const struct buffer *chunk)
Thierry FOURNIERe059ec92014-03-17 12:01:13 +01001961{
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001962 char *str = chunk->area;
1963 char *end = chunk->area + chunk->data;
Thierry FOURNIERe059ec92014-03-17 12:01:13 +01001964
1965 if (start < stop) {
1966 stop--; /* reserve one byte for the final '\0' */
1967 while (start < stop && str < end) {
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001968 if (!ha_bit_test((unsigned char)(*str), map))
Thierry FOURNIERe059ec92014-03-17 12:01:13 +01001969 *start++ = *str;
1970 else {
1971 if (start + 3 >= stop)
1972 break;
1973 *start++ = escape;
1974 *start++ = hextab[(*str >> 4) & 15];
1975 *start++ = hextab[*str & 15];
1976 }
1977 str++;
1978 }
1979 *start = '\0';
1980 }
1981 return start;
1982}
1983
Dragan Dosen0edd1092016-02-12 13:23:02 +01001984/*
1985 * Tries to prefix characters tagged in the <map> with the <escape>
Aurelien DARRAGONc5bff8e2022-09-20 14:33:18 +02001986 * character. The input <string> is processed until string_stop
1987 * is reached or NULL-byte is encountered. The result will
Dragan Dosen1a5d0602016-07-22 16:00:31 +02001988 * be stored between <start> (included) and <stop> (excluded). This
1989 * function will always try to terminate the resulting string with a '\0'
1990 * before <stop>, and will return its position if the conversion
1991 * completes.
1992 */
1993char *escape_string(char *start, char *stop,
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001994 const char escape, const long *map,
Aurelien DARRAGONc5bff8e2022-09-20 14:33:18 +02001995 const char *string, const char *string_stop)
Dragan Dosen1a5d0602016-07-22 16:00:31 +02001996{
1997 if (start < stop) {
1998 stop--; /* reserve one byte for the final '\0' */
Aurelien DARRAGONc5bff8e2022-09-20 14:33:18 +02001999 while (start < stop && string < string_stop && *string != '\0') {
Willy Tarreau1bfd6022019-06-07 11:10:07 +02002000 if (!ha_bit_test((unsigned char)(*string), map))
Dragan Dosen1a5d0602016-07-22 16:00:31 +02002001 *start++ = *string;
2002 else {
2003 if (start + 2 >= stop)
2004 break;
2005 *start++ = escape;
2006 *start++ = *string;
2007 }
2008 string++;
2009 }
2010 *start = '\0';
2011 }
2012 return start;
2013}
2014
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002015/* Check a string for using it in a CSV output format. If the string contains
2016 * one of the following four char <">, <,>, CR or LF, the string is
2017 * encapsulated between <"> and the <"> are escaped by a <""> sequence.
2018 * <str> is the input string to be escaped. The function assumes that
2019 * the input string is null-terminated.
2020 *
2021 * If <quote> is 0, the result is returned escaped but without double quote.
Willy Tarreau898529b2016-01-06 18:07:04 +01002022 * It is useful if the escaped string is used between double quotes in the
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002023 * format.
2024 *
Willy Tarreau898529b2016-01-06 18:07:04 +01002025 * printf("..., \"%s\", ...\r\n", csv_enc(str, 0, &trash));
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002026 *
Willy Tarreaub631c292016-01-08 10:04:08 +01002027 * If <quote> is 1, the converter puts the quotes only if any reserved character
2028 * is present. If <quote> is 2, the converter always puts the quotes.
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002029 *
Willy Tarreau83061a82018-07-13 11:56:34 +02002030 * <output> is a struct buffer used for storing the output string.
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002031 *
Willy Tarreau898529b2016-01-06 18:07:04 +01002032 * The function returns the converted string on its output. If an error
2033 * occurs, the function returns an empty string. This type of output is useful
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002034 * for using the function directly as printf() argument.
2035 *
2036 * If the output buffer is too short to contain the input string, the result
2037 * is truncated.
Willy Tarreau898529b2016-01-06 18:07:04 +01002038 *
Willy Tarreaub631c292016-01-08 10:04:08 +01002039 * This function appends the encoding to the existing output chunk, and it
2040 * guarantees that it starts immediately at the first available character of
2041 * the chunk. Please use csv_enc() instead if you want to replace the output
2042 * chunk.
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002043 */
Willy Tarreau83061a82018-07-13 11:56:34 +02002044const char *csv_enc_append(const char *str, int quote, struct buffer *output)
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002045{
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002046 char *end = output->area + output->size;
2047 char *out = output->area + output->data;
Willy Tarreau898529b2016-01-06 18:07:04 +01002048 char *ptr = out;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002049
Willy Tarreaub631c292016-01-08 10:04:08 +01002050 if (quote == 1) {
2051 /* automatic quoting: first verify if we'll have to quote the string */
2052 if (!strpbrk(str, "\n\r,\""))
2053 quote = 0;
2054 }
2055
2056 if (quote)
2057 *ptr++ = '"';
2058
Willy Tarreau898529b2016-01-06 18:07:04 +01002059 while (*str && ptr < end - 2) { /* -2 for reserving space for <"> and \0. */
2060 *ptr = *str;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002061 if (*str == '"') {
Willy Tarreau898529b2016-01-06 18:07:04 +01002062 ptr++;
2063 if (ptr >= end - 2) {
2064 ptr--;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002065 break;
2066 }
Willy Tarreau898529b2016-01-06 18:07:04 +01002067 *ptr = '"';
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002068 }
Willy Tarreau898529b2016-01-06 18:07:04 +01002069 ptr++;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002070 str++;
2071 }
2072
Willy Tarreaub631c292016-01-08 10:04:08 +01002073 if (quote)
2074 *ptr++ = '"';
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002075
Willy Tarreau898529b2016-01-06 18:07:04 +01002076 *ptr = '\0';
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002077 output->data = ptr - output->area;
Willy Tarreau898529b2016-01-06 18:07:04 +01002078 return out;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002079}
2080
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002081/* Decode an URL-encoded string in-place. The resulting string might
2082 * be shorter. If some forbidden characters are found, the conversion is
Thierry FOURNIER5068d962013-10-04 16:27:27 +02002083 * aborted, the string is truncated before the issue and a negative value is
2084 * returned, otherwise the operation returns the length of the decoded string.
Willy Tarreau62ba9ba2020-04-23 17:54:47 +02002085 * If the 'in_form' argument is non-nul the string is assumed to be part of
2086 * an "application/x-www-form-urlencoded" encoded string, and the '+' will be
2087 * turned to a space. If it's zero, this will only be done after a question
2088 * mark ('?').
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002089 */
Willy Tarreau62ba9ba2020-04-23 17:54:47 +02002090int url_decode(char *string, int in_form)
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002091{
2092 char *in, *out;
Thierry FOURNIER5068d962013-10-04 16:27:27 +02002093 int ret = -1;
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002094
2095 in = string;
2096 out = string;
2097 while (*in) {
2098 switch (*in) {
2099 case '+' :
Willy Tarreau62ba9ba2020-04-23 17:54:47 +02002100 *out++ = in_form ? ' ' : *in;
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002101 break;
2102 case '%' :
2103 if (!ishex(in[1]) || !ishex(in[2]))
2104 goto end;
2105 *out++ = (hex2i(in[1]) << 4) + hex2i(in[2]);
2106 in += 2;
2107 break;
Willy Tarreau62ba9ba2020-04-23 17:54:47 +02002108 case '?':
2109 in_form = 1;
Willy Tarreau7de8de02022-11-14 07:20:09 +01002110 __fallthrough;
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002111 default:
2112 *out++ = *in;
2113 break;
2114 }
2115 in++;
2116 }
Thierry FOURNIER5068d962013-10-04 16:27:27 +02002117 ret = out - string; /* success */
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002118 end:
2119 *out = 0;
2120 return ret;
2121}
Willy Tarreaubaaee002006-06-26 02:48:02 +02002122
Willy Tarreau6911fa42007-03-04 18:06:08 +01002123unsigned int str2ui(const char *s)
2124{
2125 return __str2ui(s);
2126}
2127
2128unsigned int str2uic(const char *s)
2129{
2130 return __str2uic(s);
2131}
2132
2133unsigned int strl2ui(const char *s, int len)
2134{
2135 return __strl2ui(s, len);
2136}
2137
2138unsigned int strl2uic(const char *s, int len)
2139{
2140 return __strl2uic(s, len);
2141}
2142
Willy Tarreau4ec83cd2010-10-15 23:19:55 +02002143unsigned int read_uint(const char **s, const char *end)
2144{
2145 return __read_uint(s, end);
2146}
2147
Thierry FOURNIER763a5d82015-07-06 23:09:52 +02002148/* This function reads an unsigned integer from the string pointed to by <s> and
2149 * returns it. The <s> pointer is adjusted to point to the first unread char. The
2150 * function automatically stops at <end>. If the number overflows, the 2^64-1
2151 * value is returned.
2152 */
2153unsigned long long int read_uint64(const char **s, const char *end)
2154{
2155 const char *ptr = *s;
2156 unsigned long long int i = 0, tmp;
2157 unsigned int j;
2158
2159 while (ptr < end) {
2160
2161 /* read next char */
2162 j = *ptr - '0';
2163 if (j > 9)
2164 goto read_uint64_end;
2165
2166 /* add char to the number and check overflow. */
2167 tmp = i * 10;
2168 if (tmp / 10 != i) {
2169 i = ULLONG_MAX;
2170 goto read_uint64_eat;
2171 }
2172 if (ULLONG_MAX - tmp < j) {
2173 i = ULLONG_MAX;
2174 goto read_uint64_eat;
2175 }
2176 i = tmp + j;
2177 ptr++;
2178 }
2179read_uint64_eat:
2180 /* eat each numeric char */
2181 while (ptr < end) {
2182 if ((unsigned int)(*ptr - '0') > 9)
2183 break;
2184 ptr++;
2185 }
2186read_uint64_end:
2187 *s = ptr;
2188 return i;
2189}
2190
2191/* This function reads an integer from the string pointed to by <s> and returns
2192 * it. The <s> pointer is adjusted to point to the first unread char. The function
2193 * automatically stops at <end>. Il the number is bigger than 2^63-2, the 2^63-1
2194 * value is returned. If the number is lowest than -2^63-1, the -2^63 value is
2195 * returned.
2196 */
2197long long int read_int64(const char **s, const char *end)
2198{
2199 unsigned long long int i = 0;
2200 int neg = 0;
2201
2202 /* Look for minus char. */
2203 if (**s == '-') {
2204 neg = 1;
2205 (*s)++;
2206 }
2207 else if (**s == '+')
2208 (*s)++;
2209
2210 /* convert as positive number. */
2211 i = read_uint64(s, end);
2212
2213 if (neg) {
2214 if (i > 0x8000000000000000ULL)
2215 return LLONG_MIN;
2216 return -i;
2217 }
2218 if (i > 0x7fffffffffffffffULL)
2219 return LLONG_MAX;
2220 return i;
2221}
2222
Willy Tarreau6911fa42007-03-04 18:06:08 +01002223/* This one is 7 times faster than strtol() on athlon with checks.
2224 * It returns the value of the number composed of all valid digits read,
2225 * and can process negative numbers too.
2226 */
2227int strl2ic(const char *s, int len)
2228{
2229 int i = 0;
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002230 int j, k;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002231
2232 if (len > 0) {
2233 if (*s != '-') {
2234 /* positive number */
2235 while (len-- > 0) {
2236 j = (*s++) - '0';
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002237 k = i * 10;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002238 if (j > 9)
2239 break;
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002240 i = k + j;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002241 }
2242 } else {
2243 /* negative number */
2244 s++;
2245 while (--len > 0) {
2246 j = (*s++) - '0';
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002247 k = i * 10;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002248 if (j > 9)
2249 break;
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002250 i = k - j;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002251 }
2252 }
2253 }
2254 return i;
2255}
2256
2257
2258/* This function reads exactly <len> chars from <s> and converts them to a
2259 * signed integer which it stores into <ret>. It accurately detects any error
2260 * (truncated string, invalid chars, overflows). It is meant to be used in
2261 * applications designed for hostile environments. It returns zero when the
2262 * number has successfully been converted, non-zero otherwise. When an error
2263 * is returned, the <ret> value is left untouched. It is yet 5 to 40 times
2264 * faster than strtol().
2265 */
2266int strl2irc(const char *s, int len, int *ret)
2267{
2268 int i = 0;
2269 int j;
2270
2271 if (!len)
2272 return 1;
2273
2274 if (*s != '-') {
2275 /* positive number */
2276 while (len-- > 0) {
2277 j = (*s++) - '0';
2278 if (j > 9) return 1; /* invalid char */
2279 if (i > INT_MAX / 10) return 1; /* check for multiply overflow */
2280 i = i * 10;
2281 if (i + j < i) return 1; /* check for addition overflow */
2282 i = i + j;
2283 }
2284 } else {
2285 /* negative number */
2286 s++;
2287 while (--len > 0) {
2288 j = (*s++) - '0';
2289 if (j > 9) return 1; /* invalid char */
2290 if (i < INT_MIN / 10) return 1; /* check for multiply overflow */
2291 i = i * 10;
2292 if (i - j > i) return 1; /* check for subtract overflow */
2293 i = i - j;
2294 }
2295 }
2296 *ret = i;
2297 return 0;
2298}
2299
2300
2301/* This function reads exactly <len> chars from <s> and converts them to a
2302 * signed integer which it stores into <ret>. It accurately detects any error
2303 * (truncated string, invalid chars, overflows). It is meant to be used in
2304 * applications designed for hostile environments. It returns zero when the
2305 * number has successfully been converted, non-zero otherwise. When an error
2306 * is returned, the <ret> value is left untouched. It is about 3 times slower
William Dauchy060ffc82021-02-06 20:47:51 +01002307 * than strl2irc().
Willy Tarreau6911fa42007-03-04 18:06:08 +01002308 */
Willy Tarreau6911fa42007-03-04 18:06:08 +01002309
2310int strl2llrc(const char *s, int len, long long *ret)
2311{
2312 long long i = 0;
2313 int j;
2314
2315 if (!len)
2316 return 1;
2317
2318 if (*s != '-') {
2319 /* positive number */
2320 while (len-- > 0) {
2321 j = (*s++) - '0';
2322 if (j > 9) return 1; /* invalid char */
2323 if (i > LLONG_MAX / 10LL) return 1; /* check for multiply overflow */
2324 i = i * 10LL;
2325 if (i + j < i) return 1; /* check for addition overflow */
2326 i = i + j;
2327 }
2328 } else {
2329 /* negative number */
2330 s++;
2331 while (--len > 0) {
2332 j = (*s++) - '0';
2333 if (j > 9) return 1; /* invalid char */
2334 if (i < LLONG_MIN / 10LL) return 1; /* check for multiply overflow */
2335 i = i * 10LL;
2336 if (i - j > i) return 1; /* check for subtract overflow */
2337 i = i - j;
2338 }
2339 }
2340 *ret = i;
2341 return 0;
2342}
2343
Thierry FOURNIER511e9472014-01-23 17:40:34 +01002344/* This function is used with pat_parse_dotted_ver(). It converts a string
2345 * composed by two number separated by a dot. Each part must contain in 16 bits
2346 * because internally they will be represented as a 32-bit quantity stored in
2347 * a 64-bit integer. It returns zero when the number has successfully been
2348 * converted, non-zero otherwise. When an error is returned, the <ret> value
2349 * is left untouched.
2350 *
2351 * "1.3" -> 0x0000000000010003
2352 * "65535.65535" -> 0x00000000ffffffff
2353 */
2354int strl2llrc_dotted(const char *text, int len, long long *ret)
2355{
2356 const char *end = &text[len];
2357 const char *p;
2358 long long major, minor;
2359
2360 /* Look for dot. */
2361 for (p = text; p < end; p++)
2362 if (*p == '.')
2363 break;
2364
2365 /* Convert major. */
2366 if (strl2llrc(text, p - text, &major) != 0)
2367 return 1;
2368
2369 /* Check major. */
2370 if (major >= 65536)
2371 return 1;
2372
2373 /* Convert minor. */
2374 minor = 0;
2375 if (p < end)
2376 if (strl2llrc(p + 1, end - (p + 1), &minor) != 0)
2377 return 1;
2378
2379 /* Check minor. */
2380 if (minor >= 65536)
2381 return 1;
2382
2383 /* Compose value. */
2384 *ret = (major << 16) | (minor & 0xffff);
2385 return 0;
2386}
2387
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002388/* This function parses a time value optionally followed by a unit suffix among
2389 * "d", "h", "m", "s", "ms" or "us". It converts the value into the unit
2390 * expected by the caller. The computation does its best to avoid overflows.
2391 * The value is returned in <ret> if everything is fine, and a NULL is returned
2392 * by the function. In case of error, a pointer to the error is returned and
2393 * <ret> is left untouched. Values are automatically rounded up when needed.
Willy Tarreau9faebe32019-06-07 19:00:37 +02002394 * Values resulting in values larger than or equal to 2^31 after conversion are
2395 * reported as an overflow as value PARSE_TIME_OVER. Non-null values resulting
2396 * in an underflow are reported as an underflow as value PARSE_TIME_UNDER.
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002397 */
2398const char *parse_time_err(const char *text, unsigned *ret, unsigned unit_flags)
2399{
Willy Tarreau9faebe32019-06-07 19:00:37 +02002400 unsigned long long imult, idiv;
2401 unsigned long long omult, odiv;
2402 unsigned long long value, result;
Christopher Fauletc20ad0d2020-12-11 09:23:07 +01002403 const char *str = text;
2404
2405 if (!isdigit((unsigned char)*text))
2406 return text;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002407
2408 omult = odiv = 1;
2409
2410 switch (unit_flags & TIME_UNIT_MASK) {
2411 case TIME_UNIT_US: omult = 1000000; break;
2412 case TIME_UNIT_MS: omult = 1000; break;
2413 case TIME_UNIT_S: break;
2414 case TIME_UNIT_MIN: odiv = 60; break;
2415 case TIME_UNIT_HOUR: odiv = 3600; break;
2416 case TIME_UNIT_DAY: odiv = 86400; break;
2417 default: break;
2418 }
2419
2420 value = 0;
2421
2422 while (1) {
2423 unsigned int j;
2424
2425 j = *text - '0';
2426 if (j > 9)
2427 break;
2428 text++;
2429 value *= 10;
2430 value += j;
2431 }
2432
2433 imult = idiv = 1;
2434 switch (*text) {
2435 case '\0': /* no unit = default unit */
2436 imult = omult = idiv = odiv = 1;
Christopher Fauletc20ad0d2020-12-11 09:23:07 +01002437 goto end;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002438 case 's': /* second = unscaled unit */
2439 break;
2440 case 'u': /* microsecond : "us" */
2441 if (text[1] == 's') {
2442 idiv = 1000000;
2443 text++;
Thayne McCombsa6838052021-04-02 14:12:43 -06002444 break;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002445 }
Christopher Fauletc20ad0d2020-12-11 09:23:07 +01002446 return text;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002447 case 'm': /* millisecond : "ms" or minute: "m" */
2448 if (text[1] == 's') {
2449 idiv = 1000;
2450 text++;
2451 } else
2452 imult = 60;
2453 break;
2454 case 'h': /* hour : "h" */
2455 imult = 3600;
2456 break;
2457 case 'd': /* day : "d" */
2458 imult = 86400;
2459 break;
2460 default:
2461 return text;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002462 }
Christopher Fauletc20ad0d2020-12-11 09:23:07 +01002463 if (*(++text) != '\0') {
2464 ha_warning("unexpected character '%c' after the timer value '%s', only "
2465 "(us=microseconds,ms=milliseconds,s=seconds,m=minutes,h=hours,d=days) are supported."
2466 " This will be reported as an error in next versions.\n", *text, str);
2467 }
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002468
Christopher Fauletc20ad0d2020-12-11 09:23:07 +01002469 end:
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002470 if (omult % idiv == 0) { omult /= idiv; idiv = 1; }
2471 if (idiv % omult == 0) { idiv /= omult; omult = 1; }
2472 if (imult % odiv == 0) { imult /= odiv; odiv = 1; }
2473 if (odiv % imult == 0) { odiv /= imult; imult = 1; }
2474
Willy Tarreau9faebe32019-06-07 19:00:37 +02002475 result = (value * (imult * omult) + (idiv * odiv - 1)) / (idiv * odiv);
2476 if (result >= 0x80000000)
2477 return PARSE_TIME_OVER;
2478 if (!result && value)
2479 return PARSE_TIME_UNDER;
2480 *ret = result;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002481 return NULL;
2482}
Willy Tarreau6911fa42007-03-04 18:06:08 +01002483
Emeric Brun39132b22010-01-04 14:57:24 +01002484/* this function converts the string starting at <text> to an unsigned int
2485 * stored in <ret>. If an error is detected, the pointer to the unexpected
Joseph Herlant32b83272018-11-15 11:58:28 -08002486 * character is returned. If the conversion is successful, NULL is returned.
Emeric Brun39132b22010-01-04 14:57:24 +01002487 */
2488const char *parse_size_err(const char *text, unsigned *ret) {
2489 unsigned value = 0;
2490
Christopher Faulet82635a02020-12-11 09:30:45 +01002491 if (!isdigit((unsigned char)*text))
2492 return text;
2493
Emeric Brun39132b22010-01-04 14:57:24 +01002494 while (1) {
2495 unsigned int j;
2496
2497 j = *text - '0';
2498 if (j > 9)
2499 break;
2500 if (value > ~0U / 10)
2501 return text;
2502 value *= 10;
2503 if (value > (value + j))
2504 return text;
2505 value += j;
2506 text++;
2507 }
2508
2509 switch (*text) {
2510 case '\0':
2511 break;
2512 case 'K':
2513 case 'k':
2514 if (value > ~0U >> 10)
2515 return text;
2516 value = value << 10;
2517 break;
2518 case 'M':
2519 case 'm':
2520 if (value > ~0U >> 20)
2521 return text;
2522 value = value << 20;
2523 break;
2524 case 'G':
2525 case 'g':
2526 if (value > ~0U >> 30)
2527 return text;
2528 value = value << 30;
2529 break;
2530 default:
2531 return text;
2532 }
2533
Godbach58048a22015-01-28 17:36:16 +08002534 if (*text != '\0' && *++text != '\0')
2535 return text;
2536
Emeric Brun39132b22010-01-04 14:57:24 +01002537 *ret = value;
2538 return NULL;
2539}
2540
Willy Tarreau126d4062013-12-03 17:50:47 +01002541/*
2542 * Parse binary string written in hexadecimal (source) and store the decoded
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05002543 * result into binstr and set binstrlen to the length of binstr. Memory for
Willy Tarreau126d4062013-12-03 17:50:47 +01002544 * binstr is allocated by the function. In case of error, returns 0 with an
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05002545 * error message in err. In success case, it returns the consumed length.
Willy Tarreau126d4062013-12-03 17:50:47 +01002546 */
2547int parse_binary(const char *source, char **binstr, int *binstrlen, char **err)
2548{
2549 int len;
2550 const char *p = source;
2551 int i,j;
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002552 int alloc;
Willy Tarreau126d4062013-12-03 17:50:47 +01002553
2554 len = strlen(source);
2555 if (len % 2) {
2556 memprintf(err, "an even number of hex digit is expected");
2557 return 0;
2558 }
2559
2560 len = len >> 1;
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002561
Willy Tarreau126d4062013-12-03 17:50:47 +01002562 if (!*binstr) {
Tim Duesterhuse52b6e52020-09-12 20:26:43 +02002563 *binstr = calloc(len, sizeof(**binstr));
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002564 if (!*binstr) {
2565 memprintf(err, "out of memory while loading string pattern");
2566 return 0;
2567 }
2568 alloc = 1;
Willy Tarreau126d4062013-12-03 17:50:47 +01002569 }
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002570 else {
2571 if (*binstrlen < len) {
Joseph Herlant76dbe782018-11-15 12:01:22 -08002572 memprintf(err, "no space available in the buffer. expect %d, provides %d",
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002573 len, *binstrlen);
2574 return 0;
2575 }
2576 alloc = 0;
2577 }
2578 *binstrlen = len;
Willy Tarreau126d4062013-12-03 17:50:47 +01002579
2580 i = j = 0;
2581 while (j < len) {
2582 if (!ishex(p[i++]))
2583 goto bad_input;
2584 if (!ishex(p[i++]))
2585 goto bad_input;
2586 (*binstr)[j++] = (hex2i(p[i-2]) << 4) + hex2i(p[i-1]);
2587 }
Thierry FOURNIERee330af2014-01-21 11:36:14 +01002588 return len << 1;
Willy Tarreau126d4062013-12-03 17:50:47 +01002589
2590bad_input:
2591 memprintf(err, "an hex digit is expected (found '%c')", p[i-1]);
Willy Tarreau61cfdf42021-02-20 10:46:51 +01002592 if (alloc)
2593 ha_free(binstr);
Willy Tarreau126d4062013-12-03 17:50:47 +01002594 return 0;
2595}
2596
Willy Tarreau946ba592009-05-10 15:41:18 +02002597/* copies at most <n> characters from <src> and always terminates with '\0' */
2598char *my_strndup(const char *src, int n)
2599{
2600 int len = 0;
2601 char *ret;
2602
2603 while (len < n && src[len])
2604 len++;
2605
Vincent Bernat3c2f2f22016-04-03 13:48:42 +02002606 ret = malloc(len + 1);
Willy Tarreau946ba592009-05-10 15:41:18 +02002607 if (!ret)
2608 return ret;
2609 memcpy(ret, src, len);
2610 ret[len] = '\0';
2611 return ret;
2612}
2613
Baptiste Assmannbb77c8e2013-10-06 23:24:13 +02002614/*
2615 * search needle in haystack
2616 * returns the pointer if found, returns NULL otherwise
2617 */
2618const void *my_memmem(const void *haystack, size_t haystacklen, const void *needle, size_t needlelen)
2619{
2620 const void *c = NULL;
2621 unsigned char f;
2622
2623 if ((haystack == NULL) || (needle == NULL) || (haystacklen < needlelen))
2624 return NULL;
2625
2626 f = *(char *)needle;
2627 c = haystack;
2628 while ((c = memchr(c, f, haystacklen - (c - haystack))) != NULL) {
2629 if ((haystacklen - (c - haystack)) < needlelen)
2630 return NULL;
2631
2632 if (memcmp(c, needle, needlelen) == 0)
2633 return c;
2634 ++c;
2635 }
2636 return NULL;
2637}
2638
Ilya Shipitsinc02a23f2020-05-06 00:53:22 +05002639/* get length of the initial segment consisting entirely of bytes in <accept> */
Christopher Faulet5eb96cb2020-04-15 10:23:01 +02002640size_t my_memspn(const void *str, size_t len, const void *accept, size_t acceptlen)
2641{
2642 size_t ret = 0;
2643
2644 while (ret < len && memchr(accept, *((int *)str), acceptlen)) {
2645 str++;
2646 ret++;
2647 }
2648 return ret;
2649}
2650
Ilya Shipitsinc02a23f2020-05-06 00:53:22 +05002651/* get length of the initial segment consisting entirely of bytes not in <rejcet> */
Christopher Faulet5eb96cb2020-04-15 10:23:01 +02002652size_t my_memcspn(const void *str, size_t len, const void *reject, size_t rejectlen)
2653{
2654 size_t ret = 0;
2655
2656 while (ret < len) {
2657 if(memchr(reject, *((int *)str), rejectlen))
2658 return ret;
2659 str++;
2660 ret++;
2661 }
2662 return ret;
2663}
2664
Willy Tarreau482b00d2009-10-04 22:48:42 +02002665/* This function returns the first unused key greater than or equal to <key> in
2666 * ID tree <root>. Zero is returned if no place is found.
2667 */
2668unsigned int get_next_id(struct eb_root *root, unsigned int key)
2669{
2670 struct eb32_node *used;
2671
2672 do {
2673 used = eb32_lookup_ge(root, key);
2674 if (!used || used->key > key)
2675 return key; /* key is available */
2676 key++;
2677 } while (key);
2678 return key;
2679}
2680
Willy Tarreau9c1e15d2017-11-15 18:51:29 +01002681/* dump the full tree to <file> in DOT format for debugging purposes. Will
2682 * optionally highlight node <subj> if found, depending on operation <op> :
2683 * 0 : nothing
2684 * >0 : insertion, node/leaf are surrounded in red
2685 * <0 : removal, node/leaf are dashed with no background
2686 * Will optionally add "desc" as a label on the graph if set and non-null.
2687 */
2688void 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 +01002689{
2690 struct eb32sc_node *node;
2691 unsigned long scope = -1;
2692
Willy Tarreau9c1e15d2017-11-15 18:51:29 +01002693 fprintf(file, "digraph ebtree {\n");
2694
2695 if (desc && *desc) {
2696 fprintf(file,
2697 " fontname=\"fixed\";\n"
2698 " fontsize=8;\n"
2699 " label=\"%s\";\n", desc);
2700 }
2701
Willy Tarreaued3cda02017-11-15 15:04:05 +01002702 fprintf(file,
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002703 " node [fontname=\"fixed\" fontsize=8 shape=\"box\" style=\"filled\" color=\"black\" fillcolor=\"white\"];\n"
2704 " edge [fontname=\"fixed\" fontsize=8 style=\"solid\" color=\"magenta\" dir=\"forward\"];\n"
Willy Tarreaued3cda02017-11-15 15:04:05 +01002705 " \"%lx_n\" [label=\"root\\n%lx\"]\n", (long)eb_root_to_node(root), (long)root
2706 );
2707
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002708 fprintf(file, " \"%lx_n\" -> \"%lx_%c\" [taillabel=\"L\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002709 (long)eb_root_to_node(root),
2710 (long)eb_root_to_node(eb_clrtag(root->b[0])),
Willy Tarreaued3cda02017-11-15 15:04:05 +01002711 eb_gettag(root->b[0]) == EB_LEAF ? 'l' : 'n');
2712
2713 node = eb32sc_first(root, scope);
2714 while (node) {
2715 if (node->node.node_p) {
2716 /* node part is used */
Willy Tarreau9c1e15d2017-11-15 18:51:29 +01002717 fprintf(file, " \"%lx_n\" [label=\"%lx\\nkey=%u\\nscope=%lx\\nbit=%d\" fillcolor=\"lightskyblue1\" %s];\n",
2718 (long)node, (long)node, node->key, node->node_s, node->node.bit,
2719 (node == subj) ? (op < 0 ? "color=\"red\" style=\"dashed\"" : op > 0 ? "color=\"red\"" : "") : "");
Willy Tarreaued3cda02017-11-15 15:04:05 +01002720
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002721 fprintf(file, " \"%lx_n\" -> \"%lx_n\" [taillabel=\"%c\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002722 (long)node,
2723 (long)eb_root_to_node(eb_clrtag(node->node.node_p)),
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002724 eb_gettag(node->node.node_p) ? 'R' : 'L');
Willy Tarreaued3cda02017-11-15 15:04:05 +01002725
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002726 fprintf(file, " \"%lx_n\" -> \"%lx_%c\" [taillabel=\"L\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002727 (long)node,
2728 (long)eb_root_to_node(eb_clrtag(node->node.branches.b[0])),
Willy Tarreaued3cda02017-11-15 15:04:05 +01002729 eb_gettag(node->node.branches.b[0]) == EB_LEAF ? 'l' : 'n');
2730
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002731 fprintf(file, " \"%lx_n\" -> \"%lx_%c\" [taillabel=\"R\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002732 (long)node,
2733 (long)eb_root_to_node(eb_clrtag(node->node.branches.b[1])),
Willy Tarreaued3cda02017-11-15 15:04:05 +01002734 eb_gettag(node->node.branches.b[1]) == EB_LEAF ? 'l' : 'n');
2735 }
2736
Willy Tarreau9c1e15d2017-11-15 18:51:29 +01002737 fprintf(file, " \"%lx_l\" [label=\"%lx\\nkey=%u\\nscope=%lx\\npfx=%u\" fillcolor=\"yellow\" %s];\n",
2738 (long)node, (long)node, node->key, node->leaf_s, node->node.pfx,
2739 (node == subj) ? (op < 0 ? "color=\"red\" style=\"dashed\"" : op > 0 ? "color=\"red\"" : "") : "");
Willy Tarreaued3cda02017-11-15 15:04:05 +01002740
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002741 fprintf(file, " \"%lx_l\" -> \"%lx_n\" [taillabel=\"%c\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002742 (long)node,
2743 (long)eb_root_to_node(eb_clrtag(node->node.leaf_p)),
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002744 eb_gettag(node->node.leaf_p) ? 'R' : 'L');
Willy Tarreaued3cda02017-11-15 15:04:05 +01002745 node = eb32sc_next(node, scope);
2746 }
2747 fprintf(file, "}\n");
2748}
2749
Willy Tarreau0dc9e6d2022-08-01 11:55:57 +02002750/* dump the full tree to <file> in DOT format for debugging purposes. Will
2751 * optionally highlight node <subj> if found, depending on operation <op> :
2752 * 0 : nothing
2753 * >0 : insertion, node/leaf are surrounded in red
2754 * <0 : removal, node/leaf are dashed with no background
2755 * Will optionally add "desc" as a label on the graph if set and non-null. The
2756 * key is printed as a u32 hex value. A full-sized hex dump would be better but
2757 * is left to be implemented.
2758 */
2759void ebmb_to_file(FILE *file, struct eb_root *root, const struct ebmb_node *subj, int op, const char *desc)
2760{
2761 struct ebmb_node *node;
2762
2763 fprintf(file, "digraph ebtree {\n");
2764
2765 if (desc && *desc) {
2766 fprintf(file,
2767 " fontname=\"fixed\";\n"
2768 " fontsize=8;\n"
2769 " label=\"%s\";\n", desc);
2770 }
2771
2772 fprintf(file,
2773 " node [fontname=\"fixed\" fontsize=8 shape=\"box\" style=\"filled\" color=\"black\" fillcolor=\"white\"];\n"
2774 " edge [fontname=\"fixed\" fontsize=8 style=\"solid\" color=\"magenta\" dir=\"forward\"];\n"
2775 " \"%lx_n\" [label=\"root\\n%lx\"]\n", (long)eb_root_to_node(root), (long)root
2776 );
2777
2778 fprintf(file, " \"%lx_n\" -> \"%lx_%c\" [taillabel=\"L\"];\n",
2779 (long)eb_root_to_node(root),
2780 (long)eb_root_to_node(eb_clrtag(root->b[0])),
2781 eb_gettag(root->b[0]) == EB_LEAF ? 'l' : 'n');
2782
2783 node = ebmb_first(root);
2784 while (node) {
2785 if (node->node.node_p) {
2786 /* node part is used */
2787 fprintf(file, " \"%lx_n\" [label=\"%lx\\nkey=%#x\\nbit=%d\" fillcolor=\"lightskyblue1\" %s];\n",
2788 (long)node, (long)node, read_u32(node->key), node->node.bit,
2789 (node == subj) ? (op < 0 ? "color=\"red\" style=\"dashed\"" : op > 0 ? "color=\"red\"" : "") : "");
2790
2791 fprintf(file, " \"%lx_n\" -> \"%lx_n\" [taillabel=\"%c\"];\n",
2792 (long)node,
2793 (long)eb_root_to_node(eb_clrtag(node->node.node_p)),
2794 eb_gettag(node->node.node_p) ? 'R' : 'L');
2795
2796 fprintf(file, " \"%lx_n\" -> \"%lx_%c\" [taillabel=\"L\"];\n",
2797 (long)node,
2798 (long)eb_root_to_node(eb_clrtag(node->node.branches.b[0])),
2799 eb_gettag(node->node.branches.b[0]) == EB_LEAF ? 'l' : 'n');
2800
2801 fprintf(file, " \"%lx_n\" -> \"%lx_%c\" [taillabel=\"R\"];\n",
2802 (long)node,
2803 (long)eb_root_to_node(eb_clrtag(node->node.branches.b[1])),
2804 eb_gettag(node->node.branches.b[1]) == EB_LEAF ? 'l' : 'n');
2805 }
2806
2807 fprintf(file, " \"%lx_l\" [label=\"%lx\\nkey=%#x\\npfx=%u\" fillcolor=\"yellow\" %s];\n",
2808 (long)node, (long)node, read_u32(node->key), node->node.pfx,
2809 (node == subj) ? (op < 0 ? "color=\"red\" style=\"dashed\"" : op > 0 ? "color=\"red\"" : "") : "");
2810
2811 fprintf(file, " \"%lx_l\" -> \"%lx_n\" [taillabel=\"%c\"];\n",
2812 (long)node,
2813 (long)eb_root_to_node(eb_clrtag(node->node.leaf_p)),
2814 eb_gettag(node->node.leaf_p) ? 'R' : 'L');
2815 node = ebmb_next(node);
2816 }
2817 fprintf(file, "}\n");
2818}
2819
Willy Tarreau348238b2010-01-18 15:05:57 +01002820/* This function compares a sample word possibly followed by blanks to another
2821 * clean word. The compare is case-insensitive. 1 is returned if both are equal,
2822 * otherwise zero. This intends to be used when checking HTTP headers for some
2823 * values. Note that it validates a word followed only by blanks but does not
2824 * validate a word followed by blanks then other chars.
2825 */
2826int word_match(const char *sample, int slen, const char *word, int wlen)
2827{
2828 if (slen < wlen)
2829 return 0;
2830
2831 while (wlen) {
2832 char c = *sample ^ *word;
2833 if (c && c != ('A' ^ 'a'))
2834 return 0;
2835 sample++;
2836 word++;
2837 slen--;
2838 wlen--;
2839 }
2840
2841 while (slen) {
2842 if (*sample != ' ' && *sample != '\t')
2843 return 0;
2844 sample++;
2845 slen--;
2846 }
2847 return 1;
2848}
Willy Tarreau482b00d2009-10-04 22:48:42 +02002849
Willy Tarreaud54bbdc2009-09-07 11:00:31 +02002850/* Converts any text-formatted IPv4 address to a host-order IPv4 address. It
2851 * is particularly fast because it avoids expensive operations such as
2852 * multiplies, which are optimized away at the end. It requires a properly
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05002853 * formatted address though (3 points).
Willy Tarreaud54bbdc2009-09-07 11:00:31 +02002854 */
2855unsigned int inetaddr_host(const char *text)
2856{
2857 const unsigned int ascii_zero = ('0' << 24) | ('0' << 16) | ('0' << 8) | '0';
2858 register unsigned int dig100, dig10, dig1;
2859 int s;
2860 const char *p, *d;
2861
2862 dig1 = dig10 = dig100 = ascii_zero;
2863 s = 24;
2864
2865 p = text;
2866 while (1) {
2867 if (((unsigned)(*p - '0')) <= 9) {
2868 p++;
2869 continue;
2870 }
2871
2872 /* here, we have a complete byte between <text> and <p> (exclusive) */
2873 if (p == text)
2874 goto end;
2875
2876 d = p - 1;
2877 dig1 |= (unsigned int)(*d << s);
2878 if (d == text)
2879 goto end;
2880
2881 d--;
2882 dig10 |= (unsigned int)(*d << s);
2883 if (d == text)
2884 goto end;
2885
2886 d--;
2887 dig100 |= (unsigned int)(*d << s);
2888 end:
2889 if (!s || *p != '.')
2890 break;
2891
2892 s -= 8;
2893 text = ++p;
2894 }
2895
2896 dig100 -= ascii_zero;
2897 dig10 -= ascii_zero;
2898 dig1 -= ascii_zero;
2899 return ((dig100 * 10) + dig10) * 10 + dig1;
2900}
2901
2902/*
2903 * Idem except the first unparsed character has to be passed in <stop>.
2904 */
2905unsigned int inetaddr_host_lim(const char *text, const char *stop)
2906{
2907 const unsigned int ascii_zero = ('0' << 24) | ('0' << 16) | ('0' << 8) | '0';
2908 register unsigned int dig100, dig10, dig1;
2909 int s;
2910 const char *p, *d;
2911
2912 dig1 = dig10 = dig100 = ascii_zero;
2913 s = 24;
2914
2915 p = text;
2916 while (1) {
2917 if (((unsigned)(*p - '0')) <= 9 && p < stop) {
2918 p++;
2919 continue;
2920 }
2921
2922 /* here, we have a complete byte between <text> and <p> (exclusive) */
2923 if (p == text)
2924 goto end;
2925
2926 d = p - 1;
2927 dig1 |= (unsigned int)(*d << s);
2928 if (d == text)
2929 goto end;
2930
2931 d--;
2932 dig10 |= (unsigned int)(*d << s);
2933 if (d == text)
2934 goto end;
2935
2936 d--;
2937 dig100 |= (unsigned int)(*d << s);
2938 end:
2939 if (!s || p == stop || *p != '.')
2940 break;
2941
2942 s -= 8;
2943 text = ++p;
2944 }
2945
2946 dig100 -= ascii_zero;
2947 dig10 -= ascii_zero;
2948 dig1 -= ascii_zero;
2949 return ((dig100 * 10) + dig10) * 10 + dig1;
2950}
2951
2952/*
2953 * Idem except the pointer to first unparsed byte is returned into <ret> which
2954 * must not be NULL.
2955 */
Willy Tarreau74172752010-10-15 23:21:42 +02002956unsigned int inetaddr_host_lim_ret(char *text, char *stop, char **ret)
Willy Tarreaud54bbdc2009-09-07 11:00:31 +02002957{
2958 const unsigned int ascii_zero = ('0' << 24) | ('0' << 16) | ('0' << 8) | '0';
2959 register unsigned int dig100, dig10, dig1;
2960 int s;
Willy Tarreau74172752010-10-15 23:21:42 +02002961 char *p, *d;
Willy Tarreaud54bbdc2009-09-07 11:00:31 +02002962
2963 dig1 = dig10 = dig100 = ascii_zero;
2964 s = 24;
2965
2966 p = text;
2967 while (1) {
2968 if (((unsigned)(*p - '0')) <= 9 && p < stop) {
2969 p++;
2970 continue;
2971 }
2972
2973 /* here, we have a complete byte between <text> and <p> (exclusive) */
2974 if (p == text)
2975 goto end;
2976
2977 d = p - 1;
2978 dig1 |= (unsigned int)(*d << s);
2979 if (d == text)
2980 goto end;
2981
2982 d--;
2983 dig10 |= (unsigned int)(*d << s);
2984 if (d == text)
2985 goto end;
2986
2987 d--;
2988 dig100 |= (unsigned int)(*d << s);
2989 end:
2990 if (!s || p == stop || *p != '.')
2991 break;
2992
2993 s -= 8;
2994 text = ++p;
2995 }
2996
2997 *ret = p;
2998 dig100 -= ascii_zero;
2999 dig10 -= ascii_zero;
3000 dig1 -= ascii_zero;
3001 return ((dig100 * 10) + dig10) * 10 + dig1;
3002}
3003
Willy Tarreauf0b38bf2010-06-06 13:22:23 +02003004/* Convert a fixed-length string to an IP address. Returns 0 in case of error,
3005 * or the number of chars read in case of success. Maybe this could be replaced
3006 * by one of the functions above. Also, apparently this function does not support
3007 * hosts above 255 and requires exactly 4 octets.
Willy Tarreau075415a2013-12-12 11:29:39 +01003008 * The destination is only modified on success.
Willy Tarreauf0b38bf2010-06-06 13:22:23 +02003009 */
3010int buf2ip(const char *buf, size_t len, struct in_addr *dst)
3011{
3012 const char *addr;
3013 int saw_digit, octets, ch;
3014 u_char tmp[4], *tp;
3015 const char *cp = buf;
3016
3017 saw_digit = 0;
3018 octets = 0;
3019 *(tp = tmp) = 0;
3020
3021 for (addr = buf; addr - buf < len; addr++) {
3022 unsigned char digit = (ch = *addr) - '0';
3023
3024 if (digit > 9 && ch != '.')
3025 break;
3026
3027 if (digit <= 9) {
3028 u_int new = *tp * 10 + digit;
3029
3030 if (new > 255)
3031 return 0;
3032
3033 *tp = new;
3034
3035 if (!saw_digit) {
3036 if (++octets > 4)
3037 return 0;
3038 saw_digit = 1;
3039 }
3040 } else if (ch == '.' && saw_digit) {
3041 if (octets == 4)
3042 return 0;
3043
3044 *++tp = 0;
3045 saw_digit = 0;
3046 } else
3047 return 0;
3048 }
3049
3050 if (octets < 4)
3051 return 0;
3052
3053 memcpy(&dst->s_addr, tmp, 4);
3054 return addr - cp;
3055}
3056
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01003057/* This function converts the string in <buf> of the len <len> to
3058 * struct in6_addr <dst> which must be allocated by the caller.
3059 * This function returns 1 in success case, otherwise zero.
Willy Tarreau075415a2013-12-12 11:29:39 +01003060 * The destination is only modified on success.
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01003061 */
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01003062int buf2ip6(const char *buf, size_t len, struct in6_addr *dst)
3063{
Thierry FOURNIERcd659912013-12-11 12:33:54 +01003064 char null_term_ip6[INET6_ADDRSTRLEN + 1];
Willy Tarreau075415a2013-12-12 11:29:39 +01003065 struct in6_addr out;
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01003066
Thierry FOURNIERcd659912013-12-11 12:33:54 +01003067 if (len > INET6_ADDRSTRLEN)
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01003068 return 0;
3069
3070 memcpy(null_term_ip6, buf, len);
3071 null_term_ip6[len] = '\0';
3072
Willy Tarreau075415a2013-12-12 11:29:39 +01003073 if (!inet_pton(AF_INET6, null_term_ip6, &out))
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01003074 return 0;
3075
Willy Tarreau075415a2013-12-12 11:29:39 +01003076 *dst = out;
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01003077 return 1;
3078}
3079
Willy Tarreauacf95772010-06-14 19:09:21 +02003080/* To be used to quote config arg positions. Returns the short string at <ptr>
3081 * surrounded by simple quotes if <ptr> is valid and non-empty, or "end of line"
3082 * if ptr is NULL or empty. The string is locally allocated.
3083 */
3084const char *quote_arg(const char *ptr)
3085{
Christopher Faulet1bc04c72017-10-29 20:14:08 +01003086 static THREAD_LOCAL char val[32];
Willy Tarreauacf95772010-06-14 19:09:21 +02003087 int i;
3088
3089 if (!ptr || !*ptr)
3090 return "end of line";
3091 val[0] = '\'';
Willy Tarreaude2dd6b2013-01-24 02:14:42 +01003092 for (i = 1; i < sizeof(val) - 2 && *ptr; i++)
Willy Tarreauacf95772010-06-14 19:09:21 +02003093 val[i] = *ptr++;
3094 val[i++] = '\'';
3095 val[i] = '\0';
3096 return val;
3097}
3098
Willy Tarreau5b180202010-07-18 10:40:48 +02003099/* returns an operator among STD_OP_* for string <str> or < 0 if unknown */
3100int get_std_op(const char *str)
3101{
3102 int ret = -1;
3103
3104 if (*str == 'e' && str[1] == 'q')
3105 ret = STD_OP_EQ;
3106 else if (*str == 'n' && str[1] == 'e')
3107 ret = STD_OP_NE;
3108 else if (*str == 'l') {
3109 if (str[1] == 'e') ret = STD_OP_LE;
3110 else if (str[1] == 't') ret = STD_OP_LT;
3111 }
3112 else if (*str == 'g') {
3113 if (str[1] == 'e') ret = STD_OP_GE;
3114 else if (str[1] == 't') ret = STD_OP_GT;
3115 }
3116
3117 if (ret == -1 || str[2] != '\0')
3118 return -1;
3119 return ret;
3120}
3121
Willy Tarreau4c14eaa2010-11-24 14:01:45 +01003122/* hash a 32-bit integer to another 32-bit integer */
3123unsigned int full_hash(unsigned int a)
3124{
3125 return __full_hash(a);
3126}
3127
Willy Tarreauf3241112019-02-26 09:56:22 +01003128/* Return the bit position in mask <m> of the nth bit set of rank <r>, between
3129 * 0 and LONGBITS-1 included, starting from the left. For example ranks 0,1,2,3
3130 * for mask 0x55 will be 6, 4, 2 and 0 respectively. This algorithm is based on
3131 * a popcount variant and is described here :
3132 * https://graphics.stanford.edu/~seander/bithacks.html
3133 */
3134unsigned int mask_find_rank_bit(unsigned int r, unsigned long m)
3135{
3136 unsigned long a, b, c, d;
3137 unsigned int s;
3138 unsigned int t;
3139
3140 a = m - ((m >> 1) & ~0UL/3);
3141 b = (a & ~0UL/5) + ((a >> 2) & ~0UL/5);
3142 c = (b + (b >> 4)) & ~0UL/0x11;
3143 d = (c + (c >> 8)) & ~0UL/0x101;
3144
3145 r++; // make r be 1..64
3146
3147 t = 0;
3148 s = LONGBITS;
3149 if (s > 32) {
Willy Tarreau9b6be3b2019-03-18 16:31:18 +01003150 unsigned long d2 = (d >> 16) >> 16;
3151 t = d2 + (d2 >> 16);
Willy Tarreauf3241112019-02-26 09:56:22 +01003152 s -= ((t - r) & 256) >> 3; r -= (t & ((t - r) >> 8));
3153 }
3154
3155 t = (d >> (s - 16)) & 0xff;
3156 s -= ((t - r) & 256) >> 4; r -= (t & ((t - r) >> 8));
3157 t = (c >> (s - 8)) & 0xf;
3158 s -= ((t - r) & 256) >> 5; r -= (t & ((t - r) >> 8));
3159 t = (b >> (s - 4)) & 0x7;
3160 s -= ((t - r) & 256) >> 6; r -= (t & ((t - r) >> 8));
3161 t = (a >> (s - 2)) & 0x3;
3162 s -= ((t - r) & 256) >> 7; r -= (t & ((t - r) >> 8));
3163 t = (m >> (s - 1)) & 0x1;
3164 s -= ((t - r) & 256) >> 8;
3165
3166 return s - 1;
3167}
3168
3169/* Same as mask_find_rank_bit() above but makes use of pre-computed bitmaps
3170 * based on <m>, in <a..d>. These ones must be updated whenever <m> changes
3171 * using mask_prep_rank_map() below.
3172 */
3173unsigned int mask_find_rank_bit_fast(unsigned int r, unsigned long m,
3174 unsigned long a, unsigned long b,
3175 unsigned long c, unsigned long d)
3176{
3177 unsigned int s;
3178 unsigned int t;
3179
3180 r++; // make r be 1..64
3181
3182 t = 0;
3183 s = LONGBITS;
3184 if (s > 32) {
Willy Tarreau9b6be3b2019-03-18 16:31:18 +01003185 unsigned long d2 = (d >> 16) >> 16;
3186 t = d2 + (d2 >> 16);
Willy Tarreauf3241112019-02-26 09:56:22 +01003187 s -= ((t - r) & 256) >> 3; r -= (t & ((t - r) >> 8));
3188 }
3189
3190 t = (d >> (s - 16)) & 0xff;
3191 s -= ((t - r) & 256) >> 4; r -= (t & ((t - r) >> 8));
3192 t = (c >> (s - 8)) & 0xf;
3193 s -= ((t - r) & 256) >> 5; r -= (t & ((t - r) >> 8));
3194 t = (b >> (s - 4)) & 0x7;
3195 s -= ((t - r) & 256) >> 6; r -= (t & ((t - r) >> 8));
3196 t = (a >> (s - 2)) & 0x3;
3197 s -= ((t - r) & 256) >> 7; r -= (t & ((t - r) >> 8));
3198 t = (m >> (s - 1)) & 0x1;
3199 s -= ((t - r) & 256) >> 8;
3200
3201 return s - 1;
3202}
3203
3204/* Prepare the bitmaps used by the fast implementation of the find_rank_bit()
3205 * above.
3206 */
3207void mask_prep_rank_map(unsigned long m,
3208 unsigned long *a, unsigned long *b,
3209 unsigned long *c, unsigned long *d)
3210{
3211 *a = m - ((m >> 1) & ~0UL/3);
3212 *b = (*a & ~0UL/5) + ((*a >> 2) & ~0UL/5);
3213 *c = (*b + (*b >> 4)) & ~0UL/0x11;
3214 *d = (*c + (*c >> 8)) & ~0UL/0x101;
3215}
3216
Willy Tarreauc7a8a3c2022-06-21 20:19:54 +02003217/* Returns the position of one bit set in <v>, starting at position <bit>, and
3218 * searching in other halves if not found. This is intended to be used to
3219 * report the position of one bit set among several based on a counter or a
3220 * random generator while preserving a relatively good distribution so that
3221 * values made of holes in the middle do not see one of the bits around the
3222 * hole being returned much more often than the other one. It can be seen as a
3223 * disturbed ffsl() where the initial search starts at bit <bit>. The look up
3224 * is performed in O(logN) time for N bit words, yielding a bit among 64 in
3225 * about 16 cycles. Its usage differs from the rank find function in that the
3226 * bit passed doesn't need to be limited to the value's popcount, making the
3227 * function easier to use for random picking, and twice as fast. Passing value
3228 * 0 for <v> makes no sense and -1 is returned in this case.
3229 */
3230int one_among_mask(unsigned long v, int bit)
3231{
3232 /* note, these masks may be produced by ~0UL/((1UL<<scale)+1) but
3233 * that's more expensive.
3234 */
3235 static const unsigned long halves[] = {
3236 (unsigned long)0x5555555555555555ULL,
3237 (unsigned long)0x3333333333333333ULL,
3238 (unsigned long)0x0F0F0F0F0F0F0F0FULL,
3239 (unsigned long)0x00FF00FF00FF00FFULL,
3240 (unsigned long)0x0000FFFF0000FFFFULL,
3241 (unsigned long)0x00000000FFFFFFFFULL
3242 };
3243 unsigned long halfword = ~0UL;
3244 int scope = 0;
3245 int mirror;
3246 int scale;
3247
3248 if (!v)
3249 return -1;
3250
3251 /* we check if the exact bit is set or if it's present in a mirror
3252 * position based on the current scale we're checking, in which case
3253 * it's returned with its current (or mirrored) value. Otherwise we'll
3254 * make sure there's at least one bit in the half we're in, and will
3255 * scale down to a smaller scope and try again, until we find the
3256 * closest bit.
3257 */
3258 for (scale = (sizeof(long) > 4) ? 5 : 4; scale >= 0; scale--) {
3259 halfword >>= (1UL << scale);
3260 scope |= (1UL << scale);
3261 mirror = bit ^ (1UL << scale);
3262 if (v & ((1UL << bit) | (1UL << mirror)))
3263 return (v & (1UL << bit)) ? bit : mirror;
3264
3265 if (!((v >> (bit & scope)) & halves[scale] & halfword))
3266 bit = mirror;
3267 }
3268 return bit;
3269}
3270
David du Colombier4f92d322011-03-24 11:09:31 +01003271/* Return non-zero if IPv4 address is part of the network,
Willy Tarreaueec1d382016-07-13 11:59:39 +02003272 * otherwise zero. Note that <addr> may not necessarily be aligned
3273 * while the two other ones must.
David du Colombier4f92d322011-03-24 11:09:31 +01003274 */
Willy Tarreaueec1d382016-07-13 11:59:39 +02003275int in_net_ipv4(const void *addr, const struct in_addr *mask, const struct in_addr *net)
David du Colombier4f92d322011-03-24 11:09:31 +01003276{
Willy Tarreaueec1d382016-07-13 11:59:39 +02003277 struct in_addr addr_copy;
3278
3279 memcpy(&addr_copy, addr, sizeof(addr_copy));
3280 return((addr_copy.s_addr & mask->s_addr) == (net->s_addr & mask->s_addr));
David du Colombier4f92d322011-03-24 11:09:31 +01003281}
3282
3283/* Return non-zero if IPv6 address is part of the network,
Willy Tarreaueec1d382016-07-13 11:59:39 +02003284 * otherwise zero. Note that <addr> may not necessarily be aligned
3285 * while the two other ones must.
David du Colombier4f92d322011-03-24 11:09:31 +01003286 */
Willy Tarreaueec1d382016-07-13 11:59:39 +02003287int in_net_ipv6(const void *addr, const struct in6_addr *mask, const struct in6_addr *net)
David du Colombier4f92d322011-03-24 11:09:31 +01003288{
3289 int i;
Willy Tarreaueec1d382016-07-13 11:59:39 +02003290 struct in6_addr addr_copy;
David du Colombier4f92d322011-03-24 11:09:31 +01003291
Willy Tarreaueec1d382016-07-13 11:59:39 +02003292 memcpy(&addr_copy, addr, sizeof(addr_copy));
David du Colombier4f92d322011-03-24 11:09:31 +01003293 for (i = 0; i < sizeof(struct in6_addr) / sizeof(int); i++)
Willy Tarreaueec1d382016-07-13 11:59:39 +02003294 if (((((int *)&addr_copy)[i] & ((int *)mask)[i])) !=
David du Colombier4f92d322011-03-24 11:09:31 +01003295 (((int *)net)[i] & ((int *)mask)[i]))
3296 return 0;
3297 return 1;
3298}
3299
3300/* RFC 4291 prefix */
3301const char rfc4291_pfx[] = { 0x00, 0x00, 0x00, 0x00,
3302 0x00, 0x00, 0x00, 0x00,
3303 0x00, 0x00, 0xFF, 0xFF };
3304
Joseph Herlant32b83272018-11-15 11:58:28 -08003305/* Map IPv4 address on IPv6 address, as specified in RFC 3513.
Thierry FOURNIER4a04dc32013-11-28 16:33:15 +01003306 * Input and output may overlap.
3307 */
David du Colombier4f92d322011-03-24 11:09:31 +01003308void v4tov6(struct in6_addr *sin6_addr, struct in_addr *sin_addr)
3309{
Thierry FOURNIER4a04dc32013-11-28 16:33:15 +01003310 struct in_addr tmp_addr;
3311
3312 tmp_addr.s_addr = sin_addr->s_addr;
David du Colombier4f92d322011-03-24 11:09:31 +01003313 memcpy(sin6_addr->s6_addr, rfc4291_pfx, sizeof(rfc4291_pfx));
Thierry FOURNIER4a04dc32013-11-28 16:33:15 +01003314 memcpy(sin6_addr->s6_addr+12, &tmp_addr.s_addr, 4);
David du Colombier4f92d322011-03-24 11:09:31 +01003315}
3316
Joseph Herlant32b83272018-11-15 11:58:28 -08003317/* Map IPv6 address on IPv4 address, as specified in RFC 3513.
David du Colombier4f92d322011-03-24 11:09:31 +01003318 * Return true if conversion is possible and false otherwise.
3319 */
3320int v6tov4(struct in_addr *sin_addr, struct in6_addr *sin6_addr)
3321{
3322 if (memcmp(sin6_addr->s6_addr, rfc4291_pfx, sizeof(rfc4291_pfx)) == 0) {
3323 memcpy(&(sin_addr->s_addr), &(sin6_addr->s6_addr[12]),
3324 sizeof(struct in_addr));
3325 return 1;
3326 }
3327
3328 return 0;
3329}
3330
Amaury Denoyelle21e611d2022-12-01 17:46:45 +01003331/* compare two struct sockaddr_storage, including port if <check_port> is true,
3332 * and return:
Baptiste Assmann08b24cf2016-01-23 23:39:12 +01003333 * 0 (true) if the addr is the same in both
3334 * 1 (false) if the addr is not the same in both
3335 * -1 (unable) if one of the addr is not AF_INET*
3336 */
Amaury Denoyelle21e611d2022-12-01 17:46:45 +01003337int ipcmp(struct sockaddr_storage *ss1, struct sockaddr_storage *ss2, int check_port)
Baptiste Assmann08b24cf2016-01-23 23:39:12 +01003338{
3339 if ((ss1->ss_family != AF_INET) && (ss1->ss_family != AF_INET6))
3340 return -1;
3341
3342 if ((ss2->ss_family != AF_INET) && (ss2->ss_family != AF_INET6))
3343 return -1;
3344
3345 if (ss1->ss_family != ss2->ss_family)
3346 return 1;
3347
3348 switch (ss1->ss_family) {
3349 case AF_INET:
Amaury Denoyelle21e611d2022-12-01 17:46:45 +01003350 return (memcmp(&((struct sockaddr_in *)ss1)->sin_addr,
Baptiste Assmann08b24cf2016-01-23 23:39:12 +01003351 &((struct sockaddr_in *)ss2)->sin_addr,
Amaury Denoyelle21e611d2022-12-01 17:46:45 +01003352 sizeof(struct in_addr)) != 0) ||
3353 (check_port && get_net_port(ss1) != get_net_port(ss2));
Baptiste Assmann08b24cf2016-01-23 23:39:12 +01003354 case AF_INET6:
Amaury Denoyelle21e611d2022-12-01 17:46:45 +01003355 return (memcmp(&((struct sockaddr_in6 *)ss1)->sin6_addr,
Baptiste Assmann08b24cf2016-01-23 23:39:12 +01003356 &((struct sockaddr_in6 *)ss2)->sin6_addr,
Amaury Denoyelle21e611d2022-12-01 17:46:45 +01003357 sizeof(struct in6_addr)) != 0) ||
3358 (check_port && get_net_port(ss1) != get_net_port(ss2));
Baptiste Assmann08b24cf2016-01-23 23:39:12 +01003359 }
3360
3361 return 1;
3362}
3363
Christopher Faulet9553de72021-02-26 09:12:50 +01003364/* compare a struct sockaddr_storage to a struct net_addr and return :
3365 * 0 (true) if <addr> is matching <net>
3366 * 1 (false) if <addr> is not matching <net>
3367 * -1 (unable) if <addr> or <net> is not AF_INET*
3368 */
3369int ipcmp2net(const struct sockaddr_storage *addr, const struct net_addr *net)
3370{
3371 if ((addr->ss_family != AF_INET) && (addr->ss_family != AF_INET6))
3372 return -1;
3373
3374 if ((net->family != AF_INET) && (net->family != AF_INET6))
3375 return -1;
3376
3377 if (addr->ss_family != net->family)
3378 return 1;
3379
3380 if (addr->ss_family == AF_INET &&
3381 (((struct sockaddr_in *)addr)->sin_addr.s_addr & net->addr.v4.mask.s_addr) == net->addr.v4.ip.s_addr)
3382 return 0;
3383 else {
3384 const struct in6_addr *addr6 = &(((const struct sockaddr_in6*)addr)->sin6_addr);
3385 const struct in6_addr *nip6 = &net->addr.v6.ip;
3386 const struct in6_addr *nmask6 = &net->addr.v6.mask;
3387
3388 if ((read_u32(&addr6->s6_addr[0]) & read_u32(&nmask6->s6_addr[0])) == read_u32(&nip6->s6_addr[0]) &&
3389 (read_u32(&addr6->s6_addr[4]) & read_u32(&nmask6->s6_addr[4])) == read_u32(&nip6->s6_addr[4]) &&
3390 (read_u32(&addr6->s6_addr[8]) & read_u32(&nmask6->s6_addr[8])) == read_u32(&nip6->s6_addr[8]) &&
3391 (read_u32(&addr6->s6_addr[12]) & read_u32(&nmask6->s6_addr[12])) == read_u32(&nip6->s6_addr[12]))
3392 return 0;
3393 }
3394
3395 return 1;
3396}
3397
Baptiste Assmann08396c82016-01-31 00:27:17 +01003398/* copy IP address from <source> into <dest>
Willy Tarreaudc3a9e82016-11-04 18:47:01 +01003399 * The caller must allocate and clear <dest> before calling.
3400 * The source must be in either AF_INET or AF_INET6 family, or the destination
3401 * address will be undefined. If the destination address used to hold a port,
3402 * it is preserved, so that this function can be used to switch to another
3403 * address family with no risk. Returns a pointer to the destination.
Baptiste Assmann08396c82016-01-31 00:27:17 +01003404 */
3405struct sockaddr_storage *ipcpy(struct sockaddr_storage *source, struct sockaddr_storage *dest)
3406{
Willy Tarreaudc3a9e82016-11-04 18:47:01 +01003407 int prev_port;
3408
3409 prev_port = get_net_port(dest);
3410 memset(dest, 0, sizeof(*dest));
Baptiste Assmann08396c82016-01-31 00:27:17 +01003411 dest->ss_family = source->ss_family;
3412
3413 /* copy new addr and apply it */
3414 switch (source->ss_family) {
3415 case AF_INET:
3416 ((struct sockaddr_in *)dest)->sin_addr.s_addr = ((struct sockaddr_in *)source)->sin_addr.s_addr;
Willy Tarreaudc3a9e82016-11-04 18:47:01 +01003417 ((struct sockaddr_in *)dest)->sin_port = prev_port;
Baptiste Assmann08396c82016-01-31 00:27:17 +01003418 break;
3419 case AF_INET6:
3420 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 +01003421 ((struct sockaddr_in6 *)dest)->sin6_port = prev_port;
Baptiste Assmann08396c82016-01-31 00:27:17 +01003422 break;
3423 }
3424
3425 return dest;
3426}
3427
William Lallemand421f5b52012-02-06 18:15:57 +01003428char *human_time(int t, short hz_div) {
3429 static char rv[sizeof("24855d23h")+1]; // longest of "23h59m" and "59m59s"
3430 char *p = rv;
Willy Tarreau761b3d52014-04-14 14:53:06 +02003431 char *end = rv + sizeof(rv);
William Lallemand421f5b52012-02-06 18:15:57 +01003432 int cnt=2; // print two numbers
3433
3434 if (unlikely(t < 0 || hz_div <= 0)) {
Willy Tarreau761b3d52014-04-14 14:53:06 +02003435 snprintf(p, end - p, "?");
William Lallemand421f5b52012-02-06 18:15:57 +01003436 return rv;
3437 }
3438
3439 if (unlikely(hz_div > 1))
3440 t /= hz_div;
3441
3442 if (t >= DAY) {
Willy Tarreau761b3d52014-04-14 14:53:06 +02003443 p += snprintf(p, end - p, "%dd", t / DAY);
William Lallemand421f5b52012-02-06 18:15:57 +01003444 cnt--;
3445 }
3446
3447 if (cnt && t % DAY / HOUR) {
Willy Tarreau761b3d52014-04-14 14:53:06 +02003448 p += snprintf(p, end - p, "%dh", t % DAY / HOUR);
William Lallemand421f5b52012-02-06 18:15:57 +01003449 cnt--;
3450 }
3451
3452 if (cnt && t % HOUR / MINUTE) {
Willy Tarreau761b3d52014-04-14 14:53:06 +02003453 p += snprintf(p, end - p, "%dm", t % HOUR / MINUTE);
William Lallemand421f5b52012-02-06 18:15:57 +01003454 cnt--;
3455 }
3456
3457 if ((cnt && t % MINUTE) || !t) // also display '0s'
Willy Tarreau761b3d52014-04-14 14:53:06 +02003458 p += snprintf(p, end - p, "%ds", t % MINUTE / SEC);
William Lallemand421f5b52012-02-06 18:15:57 +01003459
3460 return rv;
3461}
3462
3463const char *monthname[12] = {
3464 "Jan", "Feb", "Mar", "Apr", "May", "Jun",
3465 "Jul", "Aug", "Sep", "Oct", "Nov", "Dec"
3466};
3467
3468/* date2str_log: write a date in the format :
3469 * sprintf(str, "%02d/%s/%04d:%02d:%02d:%02d.%03d",
3470 * tm.tm_mday, monthname[tm.tm_mon], tm.tm_year+1900,
3471 * tm.tm_hour, tm.tm_min, tm.tm_sec, (int)date.tv_usec/1000);
3472 *
3473 * without using sprintf. return a pointer to the last char written (\0) or
3474 * NULL if there isn't enough space.
3475 */
Willy Tarreauf16cb412018-09-04 19:08:48 +02003476char *date2str_log(char *dst, const struct tm *tm, const struct timeval *date, size_t size)
William Lallemand421f5b52012-02-06 18:15:57 +01003477{
3478
3479 if (size < 25) /* the size is fixed: 24 chars + \0 */
3480 return NULL;
3481
3482 dst = utoa_pad((unsigned int)tm->tm_mday, dst, 3); // day
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003483 if (!dst)
3484 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003485 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003486
William Lallemand421f5b52012-02-06 18:15:57 +01003487 memcpy(dst, monthname[tm->tm_mon], 3); // month
3488 dst += 3;
3489 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003490
William Lallemand421f5b52012-02-06 18:15:57 +01003491 dst = utoa_pad((unsigned int)tm->tm_year+1900, dst, 5); // year
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003492 if (!dst)
3493 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003494 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003495
William Lallemand421f5b52012-02-06 18:15:57 +01003496 dst = utoa_pad((unsigned int)tm->tm_hour, dst, 3); // hour
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003497 if (!dst)
3498 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003499 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003500
William Lallemand421f5b52012-02-06 18:15:57 +01003501 dst = utoa_pad((unsigned int)tm->tm_min, dst, 3); // minutes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003502 if (!dst)
3503 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003504 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003505
William Lallemand421f5b52012-02-06 18:15:57 +01003506 dst = utoa_pad((unsigned int)tm->tm_sec, dst, 3); // secondes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003507 if (!dst)
3508 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003509 *dst++ = '.';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003510
Willy Tarreau7d9421d2020-02-29 09:08:02 +01003511 dst = utoa_pad((unsigned int)(date->tv_usec/1000)%1000, dst, 4); // milliseconds
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003512 if (!dst)
3513 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003514 *dst = '\0';
3515
3516 return dst;
3517}
3518
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003519/* Base year used to compute leap years */
3520#define TM_YEAR_BASE 1900
3521
3522/* Return the difference in seconds between two times (leap seconds are ignored).
3523 * Retrieved from glibc 2.18 source code.
3524 */
3525static int my_tm_diff(const struct tm *a, const struct tm *b)
3526{
3527 /* Compute intervening leap days correctly even if year is negative.
3528 * Take care to avoid int overflow in leap day calculations,
3529 * but it's OK to assume that A and B are close to each other.
3530 */
3531 int a4 = (a->tm_year >> 2) + (TM_YEAR_BASE >> 2) - ! (a->tm_year & 3);
3532 int b4 = (b->tm_year >> 2) + (TM_YEAR_BASE >> 2) - ! (b->tm_year & 3);
3533 int a100 = a4 / 25 - (a4 % 25 < 0);
3534 int b100 = b4 / 25 - (b4 % 25 < 0);
3535 int a400 = a100 >> 2;
3536 int b400 = b100 >> 2;
3537 int intervening_leap_days = (a4 - b4) - (a100 - b100) + (a400 - b400);
3538 int years = a->tm_year - b->tm_year;
3539 int days = (365 * years + intervening_leap_days
3540 + (a->tm_yday - b->tm_yday));
3541 return (60 * (60 * (24 * days + (a->tm_hour - b->tm_hour))
3542 + (a->tm_min - b->tm_min))
3543 + (a->tm_sec - b->tm_sec));
3544}
3545
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003546/* Return the GMT offset for a specific local time.
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003547 * Both t and tm must represent the same time.
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003548 * The string returned has the same format as returned by strftime(... "%z", tm).
3549 * Offsets are kept in an internal cache for better performances.
3550 */
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003551const char *get_gmt_offset(time_t t, struct tm *tm)
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003552{
3553 /* Cache offsets from GMT (depending on whether DST is active or not) */
Christopher Faulet1bc04c72017-10-29 20:14:08 +01003554 static THREAD_LOCAL char gmt_offsets[2][5+1] = { "", "" };
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003555
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003556 char *gmt_offset;
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003557 struct tm tm_gmt;
3558 int diff;
3559 int isdst = tm->tm_isdst;
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003560
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003561 /* Pretend DST not active if its status is unknown */
3562 if (isdst < 0)
3563 isdst = 0;
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003564
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003565 /* Fetch the offset and initialize it if needed */
3566 gmt_offset = gmt_offsets[isdst & 0x01];
3567 if (unlikely(!*gmt_offset)) {
3568 get_gmtime(t, &tm_gmt);
3569 diff = my_tm_diff(tm, &tm_gmt);
3570 if (diff < 0) {
3571 diff = -diff;
3572 *gmt_offset = '-';
3573 } else {
3574 *gmt_offset = '+';
3575 }
Willy Tarreaue112c8a2019-10-29 10:16:11 +01003576 diff %= 86400U;
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003577 diff /= 60; /* Convert to minutes */
3578 snprintf(gmt_offset+1, 4+1, "%02d%02d", diff/60, diff%60);
3579 }
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003580
Willy Tarreaue112c8a2019-10-29 10:16:11 +01003581 return gmt_offset;
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003582}
3583
William Lallemand421f5b52012-02-06 18:15:57 +01003584/* gmt2str_log: write a date in the format :
3585 * "%02d/%s/%04d:%02d:%02d:%02d +0000" without using snprintf
3586 * return a pointer to the last char written (\0) or
3587 * NULL if there isn't enough space.
3588 */
3589char *gmt2str_log(char *dst, struct tm *tm, size_t size)
3590{
Yuxans Yao4e25b012012-10-19 10:36:09 +08003591 if (size < 27) /* the size is fixed: 26 chars + \0 */
William Lallemand421f5b52012-02-06 18:15:57 +01003592 return NULL;
3593
3594 dst = utoa_pad((unsigned int)tm->tm_mday, dst, 3); // day
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003595 if (!dst)
3596 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003597 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003598
William Lallemand421f5b52012-02-06 18:15:57 +01003599 memcpy(dst, monthname[tm->tm_mon], 3); // month
3600 dst += 3;
3601 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003602
William Lallemand421f5b52012-02-06 18:15:57 +01003603 dst = utoa_pad((unsigned int)tm->tm_year+1900, dst, 5); // year
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003604 if (!dst)
3605 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003606 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003607
William Lallemand421f5b52012-02-06 18:15:57 +01003608 dst = utoa_pad((unsigned int)tm->tm_hour, dst, 3); // hour
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003609 if (!dst)
3610 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003611 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003612
William Lallemand421f5b52012-02-06 18:15:57 +01003613 dst = utoa_pad((unsigned int)tm->tm_min, dst, 3); // minutes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003614 if (!dst)
3615 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003616 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003617
William Lallemand421f5b52012-02-06 18:15:57 +01003618 dst = utoa_pad((unsigned int)tm->tm_sec, dst, 3); // secondes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003619 if (!dst)
3620 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003621 *dst++ = ' ';
3622 *dst++ = '+';
3623 *dst++ = '0';
3624 *dst++ = '0';
3625 *dst++ = '0';
3626 *dst++ = '0';
3627 *dst = '\0';
3628
3629 return dst;
3630}
3631
Yuxans Yao4e25b012012-10-19 10:36:09 +08003632/* localdate2str_log: write a date in the format :
3633 * "%02d/%s/%04d:%02d:%02d:%02d +0000(local timezone)" without using snprintf
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003634 * Both t and tm must represent the same time.
3635 * return a pointer to the last char written (\0) or
3636 * NULL if there isn't enough space.
Yuxans Yao4e25b012012-10-19 10:36:09 +08003637 */
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003638char *localdate2str_log(char *dst, time_t t, struct tm *tm, size_t size)
Yuxans Yao4e25b012012-10-19 10:36:09 +08003639{
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003640 const char *gmt_offset;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003641 if (size < 27) /* the size is fixed: 26 chars + \0 */
3642 return NULL;
3643
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003644 gmt_offset = get_gmt_offset(t, tm);
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003645
Yuxans Yao4e25b012012-10-19 10:36:09 +08003646 dst = utoa_pad((unsigned int)tm->tm_mday, dst, 3); // day
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003647 if (!dst)
3648 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003649 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003650
Yuxans Yao4e25b012012-10-19 10:36:09 +08003651 memcpy(dst, monthname[tm->tm_mon], 3); // month
3652 dst += 3;
3653 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003654
Yuxans Yao4e25b012012-10-19 10:36:09 +08003655 dst = utoa_pad((unsigned int)tm->tm_year+1900, dst, 5); // year
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003656 if (!dst)
3657 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003658 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003659
Yuxans Yao4e25b012012-10-19 10:36:09 +08003660 dst = utoa_pad((unsigned int)tm->tm_hour, dst, 3); // hour
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003661 if (!dst)
3662 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003663 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003664
Yuxans Yao4e25b012012-10-19 10:36:09 +08003665 dst = utoa_pad((unsigned int)tm->tm_min, dst, 3); // minutes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003666 if (!dst)
3667 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003668 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003669
Yuxans Yao4e25b012012-10-19 10:36:09 +08003670 dst = utoa_pad((unsigned int)tm->tm_sec, dst, 3); // secondes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003671 if (!dst)
3672 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003673 *dst++ = ' ';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003674
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003675 memcpy(dst, gmt_offset, 5); // Offset from local time to GMT
Yuxans Yao4e25b012012-10-19 10:36:09 +08003676 dst += 5;
3677 *dst = '\0';
3678
3679 return dst;
3680}
3681
Willy Tarreaucb1949b2017-07-19 19:05:29 +02003682/* Returns the number of seconds since 01/01/1970 0:0:0 GMT for GMT date <tm>.
3683 * It is meant as a portable replacement for timegm() for use with valid inputs.
3684 * Returns undefined results for invalid dates (eg: months out of range 0..11).
3685 */
3686time_t my_timegm(const struct tm *tm)
3687{
3688 /* Each month has 28, 29, 30 or 31 days, or 28+N. The date in the year
3689 * is thus (current month - 1)*28 + cumulated_N[month] to count the
3690 * sum of the extra N days for elapsed months. The sum of all these N
3691 * days doesn't exceed 30 for a complete year (366-12*28) so it fits
3692 * in a 5-bit word. This means that with 60 bits we can represent a
3693 * matrix of all these values at once, which is fast and efficient to
3694 * access. The extra February day for leap years is not counted here.
3695 *
3696 * Jan : none = 0 (0)
3697 * Feb : Jan = 3 (3)
3698 * Mar : Jan..Feb = 3 (3 + 0)
3699 * Apr : Jan..Mar = 6 (3 + 0 + 3)
3700 * May : Jan..Apr = 8 (3 + 0 + 3 + 2)
3701 * Jun : Jan..May = 11 (3 + 0 + 3 + 2 + 3)
3702 * Jul : Jan..Jun = 13 (3 + 0 + 3 + 2 + 3 + 2)
3703 * Aug : Jan..Jul = 16 (3 + 0 + 3 + 2 + 3 + 2 + 3)
3704 * Sep : Jan..Aug = 19 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3)
3705 * Oct : Jan..Sep = 21 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3 + 2)
3706 * Nov : Jan..Oct = 24 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3 + 2 + 3)
3707 * Dec : Jan..Nov = 26 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3 + 2 + 3 + 2)
3708 */
3709 uint64_t extra =
3710 ( 0ULL << 0*5) + ( 3ULL << 1*5) + ( 3ULL << 2*5) + /* Jan, Feb, Mar, */
3711 ( 6ULL << 3*5) + ( 8ULL << 4*5) + (11ULL << 5*5) + /* Apr, May, Jun, */
3712 (13ULL << 6*5) + (16ULL << 7*5) + (19ULL << 8*5) + /* Jul, Aug, Sep, */
3713 (21ULL << 9*5) + (24ULL << 10*5) + (26ULL << 11*5); /* Oct, Nov, Dec, */
3714
3715 unsigned int y = tm->tm_year + 1900;
3716 unsigned int m = tm->tm_mon;
3717 unsigned long days = 0;
3718
3719 /* days since 1/1/1970 for full years */
3720 days += days_since_zero(y) - days_since_zero(1970);
3721
3722 /* days for full months in the current year */
3723 days += 28 * m + ((extra >> (m * 5)) & 0x1f);
3724
3725 /* count + 1 after March for leap years. A leap year is a year multiple
3726 * of 4, unless it's multiple of 100 without being multiple of 400. 2000
3727 * is leap, 1900 isn't, 1904 is.
3728 */
3729 if ((m > 1) && !(y & 3) && ((y % 100) || !(y % 400)))
3730 days++;
3731
3732 days += tm->tm_mday - 1;
3733 return days * 86400ULL + tm->tm_hour * 3600 + tm->tm_min * 60 + tm->tm_sec;
3734}
3735
Thierry Fournier93127942016-01-20 18:49:45 +01003736/* This function check a char. It returns true and updates
3737 * <date> and <len> pointer to the new position if the
3738 * character is found.
3739 */
3740static inline int parse_expect_char(const char **date, int *len, char c)
3741{
3742 if (*len < 1 || **date != c)
3743 return 0;
3744 (*len)--;
3745 (*date)++;
3746 return 1;
3747}
3748
3749/* This function expects a string <str> of len <l>. It return true and updates.
3750 * <date> and <len> if the string matches, otherwise, it returns false.
3751 */
3752static inline int parse_strcmp(const char **date, int *len, char *str, int l)
3753{
3754 if (*len < l || strncmp(*date, str, l) != 0)
3755 return 0;
3756 (*len) -= l;
3757 (*date) += l;
3758 return 1;
3759}
3760
3761/* This macro converts 3 chars name in integer. */
3762#define STR2I3(__a, __b, __c) ((__a) * 65536 + (__b) * 256 + (__c))
3763
3764/* day-name = %x4D.6F.6E ; "Mon", case-sensitive
3765 * / %x54.75.65 ; "Tue", case-sensitive
3766 * / %x57.65.64 ; "Wed", case-sensitive
3767 * / %x54.68.75 ; "Thu", case-sensitive
3768 * / %x46.72.69 ; "Fri", case-sensitive
3769 * / %x53.61.74 ; "Sat", case-sensitive
3770 * / %x53.75.6E ; "Sun", case-sensitive
3771 *
3772 * This array must be alphabetically sorted
3773 */
3774static inline int parse_http_dayname(const char **date, int *len, struct tm *tm)
3775{
3776 if (*len < 3)
3777 return 0;
3778 switch (STR2I3((*date)[0], (*date)[1], (*date)[2])) {
3779 case STR2I3('M','o','n'): tm->tm_wday = 1; break;
3780 case STR2I3('T','u','e'): tm->tm_wday = 2; break;
3781 case STR2I3('W','e','d'): tm->tm_wday = 3; break;
3782 case STR2I3('T','h','u'): tm->tm_wday = 4; break;
3783 case STR2I3('F','r','i'): tm->tm_wday = 5; break;
3784 case STR2I3('S','a','t'): tm->tm_wday = 6; break;
3785 case STR2I3('S','u','n'): tm->tm_wday = 7; break;
3786 default: return 0;
3787 }
3788 *len -= 3;
3789 *date += 3;
3790 return 1;
3791}
3792
3793/* month = %x4A.61.6E ; "Jan", case-sensitive
3794 * / %x46.65.62 ; "Feb", case-sensitive
3795 * / %x4D.61.72 ; "Mar", case-sensitive
3796 * / %x41.70.72 ; "Apr", case-sensitive
3797 * / %x4D.61.79 ; "May", case-sensitive
3798 * / %x4A.75.6E ; "Jun", case-sensitive
3799 * / %x4A.75.6C ; "Jul", case-sensitive
3800 * / %x41.75.67 ; "Aug", case-sensitive
3801 * / %x53.65.70 ; "Sep", case-sensitive
3802 * / %x4F.63.74 ; "Oct", case-sensitive
3803 * / %x4E.6F.76 ; "Nov", case-sensitive
3804 * / %x44.65.63 ; "Dec", case-sensitive
3805 *
3806 * This array must be alphabetically sorted
3807 */
3808static inline int parse_http_monthname(const char **date, int *len, struct tm *tm)
3809{
3810 if (*len < 3)
3811 return 0;
3812 switch (STR2I3((*date)[0], (*date)[1], (*date)[2])) {
3813 case STR2I3('J','a','n'): tm->tm_mon = 0; break;
3814 case STR2I3('F','e','b'): tm->tm_mon = 1; break;
3815 case STR2I3('M','a','r'): tm->tm_mon = 2; break;
3816 case STR2I3('A','p','r'): tm->tm_mon = 3; break;
3817 case STR2I3('M','a','y'): tm->tm_mon = 4; break;
3818 case STR2I3('J','u','n'): tm->tm_mon = 5; break;
3819 case STR2I3('J','u','l'): tm->tm_mon = 6; break;
3820 case STR2I3('A','u','g'): tm->tm_mon = 7; break;
3821 case STR2I3('S','e','p'): tm->tm_mon = 8; break;
3822 case STR2I3('O','c','t'): tm->tm_mon = 9; break;
3823 case STR2I3('N','o','v'): tm->tm_mon = 10; break;
3824 case STR2I3('D','e','c'): tm->tm_mon = 11; break;
3825 default: return 0;
3826 }
3827 *len -= 3;
3828 *date += 3;
3829 return 1;
3830}
3831
3832/* day-name-l = %x4D.6F.6E.64.61.79 ; "Monday", case-sensitive
3833 * / %x54.75.65.73.64.61.79 ; "Tuesday", case-sensitive
3834 * / %x57.65.64.6E.65.73.64.61.79 ; "Wednesday", case-sensitive
3835 * / %x54.68.75.72.73.64.61.79 ; "Thursday", case-sensitive
3836 * / %x46.72.69.64.61.79 ; "Friday", case-sensitive
3837 * / %x53.61.74.75.72.64.61.79 ; "Saturday", case-sensitive
3838 * / %x53.75.6E.64.61.79 ; "Sunday", case-sensitive
3839 *
3840 * This array must be alphabetically sorted
3841 */
3842static inline int parse_http_ldayname(const char **date, int *len, struct tm *tm)
3843{
3844 if (*len < 6) /* Minimum length. */
3845 return 0;
3846 switch (STR2I3((*date)[0], (*date)[1], (*date)[2])) {
3847 case STR2I3('M','o','n'):
3848 RET0_UNLESS(parse_strcmp(date, len, "Monday", 6));
3849 tm->tm_wday = 1;
3850 return 1;
3851 case STR2I3('T','u','e'):
3852 RET0_UNLESS(parse_strcmp(date, len, "Tuesday", 7));
3853 tm->tm_wday = 2;
3854 return 1;
3855 case STR2I3('W','e','d'):
3856 RET0_UNLESS(parse_strcmp(date, len, "Wednesday", 9));
3857 tm->tm_wday = 3;
3858 return 1;
3859 case STR2I3('T','h','u'):
3860 RET0_UNLESS(parse_strcmp(date, len, "Thursday", 8));
3861 tm->tm_wday = 4;
3862 return 1;
3863 case STR2I3('F','r','i'):
3864 RET0_UNLESS(parse_strcmp(date, len, "Friday", 6));
3865 tm->tm_wday = 5;
3866 return 1;
3867 case STR2I3('S','a','t'):
3868 RET0_UNLESS(parse_strcmp(date, len, "Saturday", 8));
3869 tm->tm_wday = 6;
3870 return 1;
3871 case STR2I3('S','u','n'):
3872 RET0_UNLESS(parse_strcmp(date, len, "Sunday", 6));
3873 tm->tm_wday = 7;
3874 return 1;
3875 }
3876 return 0;
3877}
3878
3879/* This function parses exactly 1 digit and returns the numeric value in "digit". */
3880static inline int parse_digit(const char **date, int *len, int *digit)
3881{
3882 if (*len < 1 || **date < '0' || **date > '9')
3883 return 0;
3884 *digit = (**date - '0');
3885 (*date)++;
3886 (*len)--;
3887 return 1;
3888}
3889
3890/* This function parses exactly 2 digits and returns the numeric value in "digit". */
3891static inline int parse_2digit(const char **date, int *len, int *digit)
3892{
3893 int value;
3894
3895 RET0_UNLESS(parse_digit(date, len, &value));
3896 (*digit) = value * 10;
3897 RET0_UNLESS(parse_digit(date, len, &value));
3898 (*digit) += value;
3899
3900 return 1;
3901}
3902
3903/* This function parses exactly 4 digits and returns the numeric value in "digit". */
3904static inline int parse_4digit(const char **date, int *len, int *digit)
3905{
3906 int value;
3907
3908 RET0_UNLESS(parse_digit(date, len, &value));
3909 (*digit) = value * 1000;
3910
3911 RET0_UNLESS(parse_digit(date, len, &value));
3912 (*digit) += value * 100;
3913
3914 RET0_UNLESS(parse_digit(date, len, &value));
3915 (*digit) += value * 10;
3916
3917 RET0_UNLESS(parse_digit(date, len, &value));
3918 (*digit) += value;
3919
3920 return 1;
3921}
3922
3923/* time-of-day = hour ":" minute ":" second
3924 * ; 00:00:00 - 23:59:60 (leap second)
3925 *
3926 * hour = 2DIGIT
3927 * minute = 2DIGIT
3928 * second = 2DIGIT
3929 */
3930static inline int parse_http_time(const char **date, int *len, struct tm *tm)
3931{
3932 RET0_UNLESS(parse_2digit(date, len, &tm->tm_hour)); /* hour 2DIGIT */
3933 RET0_UNLESS(parse_expect_char(date, len, ':')); /* expect ":" */
3934 RET0_UNLESS(parse_2digit(date, len, &tm->tm_min)); /* min 2DIGIT */
3935 RET0_UNLESS(parse_expect_char(date, len, ':')); /* expect ":" */
3936 RET0_UNLESS(parse_2digit(date, len, &tm->tm_sec)); /* sec 2DIGIT */
3937 return 1;
3938}
3939
3940/* From RFC7231
3941 * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
3942 *
3943 * IMF-fixdate = day-name "," SP date1 SP time-of-day SP GMT
3944 * ; fixed length/zone/capitalization subset of the format
3945 * ; see Section 3.3 of [RFC5322]
3946 *
3947 *
3948 * date1 = day SP month SP year
3949 * ; e.g., 02 Jun 1982
3950 *
3951 * day = 2DIGIT
3952 * year = 4DIGIT
3953 *
3954 * GMT = %x47.4D.54 ; "GMT", case-sensitive
3955 *
3956 * time-of-day = hour ":" minute ":" second
3957 * ; 00:00:00 - 23:59:60 (leap second)
3958 *
3959 * hour = 2DIGIT
3960 * minute = 2DIGIT
3961 * second = 2DIGIT
3962 *
3963 * DIGIT = decimal 0-9
3964 */
3965int parse_imf_date(const char *date, int len, struct tm *tm)
3966{
David Carlier327298c2016-11-20 10:42:38 +00003967 /* tm_gmtoff, if present, ought to be zero'ed */
3968 memset(tm, 0, sizeof(*tm));
3969
Thierry Fournier93127942016-01-20 18:49:45 +01003970 RET0_UNLESS(parse_http_dayname(&date, &len, tm)); /* day-name */
3971 RET0_UNLESS(parse_expect_char(&date, &len, ',')); /* expect "," */
3972 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3973 RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_mday)); /* day 2DIGIT */
3974 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3975 RET0_UNLESS(parse_http_monthname(&date, &len, tm)); /* Month */
3976 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3977 RET0_UNLESS(parse_4digit(&date, &len, &tm->tm_year)); /* year = 4DIGIT */
3978 tm->tm_year -= 1900;
3979 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3980 RET0_UNLESS(parse_http_time(&date, &len, tm)); /* Parse time. */
3981 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3982 RET0_UNLESS(parse_strcmp(&date, &len, "GMT", 3)); /* GMT = %x47.4D.54 ; "GMT", case-sensitive */
3983 tm->tm_isdst = -1;
Thierry Fournier93127942016-01-20 18:49:45 +01003984 return 1;
3985}
3986
3987/* From RFC7231
3988 * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
3989 *
3990 * rfc850-date = day-name-l "," SP date2 SP time-of-day SP GMT
3991 * date2 = day "-" month "-" 2DIGIT
3992 * ; e.g., 02-Jun-82
3993 *
3994 * day = 2DIGIT
3995 */
3996int parse_rfc850_date(const char *date, int len, struct tm *tm)
3997{
3998 int year;
3999
David Carlier327298c2016-11-20 10:42:38 +00004000 /* tm_gmtoff, if present, ought to be zero'ed */
4001 memset(tm, 0, sizeof(*tm));
4002
Thierry Fournier93127942016-01-20 18:49:45 +01004003 RET0_UNLESS(parse_http_ldayname(&date, &len, tm)); /* Read the day name */
4004 RET0_UNLESS(parse_expect_char(&date, &len, ',')); /* expect "," */
4005 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
4006 RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_mday)); /* day 2DIGIT */
4007 RET0_UNLESS(parse_expect_char(&date, &len, '-')); /* expect "-" */
4008 RET0_UNLESS(parse_http_monthname(&date, &len, tm)); /* Month */
4009 RET0_UNLESS(parse_expect_char(&date, &len, '-')); /* expect "-" */
4010
4011 /* year = 2DIGIT
4012 *
4013 * Recipients of a timestamp value in rfc850-(*date) format, which uses a
4014 * two-digit year, MUST interpret a timestamp that appears to be more
4015 * than 50 years in the future as representing the most recent year in
4016 * the past that had the same last two digits.
4017 */
4018 RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_year));
4019
4020 /* expect SP */
4021 if (!parse_expect_char(&date, &len, ' ')) {
4022 /* Maybe we have the date with 4 digits. */
4023 RET0_UNLESS(parse_2digit(&date, &len, &year));
4024 tm->tm_year = (tm->tm_year * 100 + year) - 1900;
4025 /* expect SP */
4026 RET0_UNLESS(parse_expect_char(&date, &len, ' '));
4027 } else {
4028 /* I fix 60 as pivot: >60: +1900, <60: +2000. Note that the
4029 * tm_year is the number of year since 1900, so for +1900, we
4030 * do nothing, and for +2000, we add 100.
4031 */
4032 if (tm->tm_year <= 60)
4033 tm->tm_year += 100;
4034 }
4035
4036 RET0_UNLESS(parse_http_time(&date, &len, tm)); /* Parse time. */
4037 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
4038 RET0_UNLESS(parse_strcmp(&date, &len, "GMT", 3)); /* GMT = %x47.4D.54 ; "GMT", case-sensitive */
4039 tm->tm_isdst = -1;
Thierry Fournier93127942016-01-20 18:49:45 +01004040
4041 return 1;
4042}
4043
4044/* From RFC7231
4045 * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
4046 *
4047 * asctime-date = day-name SP date3 SP time-of-day SP year
4048 * date3 = month SP ( 2DIGIT / ( SP 1DIGIT ))
4049 * ; e.g., Jun 2
4050 *
4051 * HTTP-date is case sensitive. A sender MUST NOT generate additional
4052 * whitespace in an HTTP-date beyond that specifically included as SP in
4053 * the grammar.
4054 */
4055int parse_asctime_date(const char *date, int len, struct tm *tm)
4056{
David Carlier327298c2016-11-20 10:42:38 +00004057 /* tm_gmtoff, if present, ought to be zero'ed */
4058 memset(tm, 0, sizeof(*tm));
4059
Thierry Fournier93127942016-01-20 18:49:45 +01004060 RET0_UNLESS(parse_http_dayname(&date, &len, tm)); /* day-name */
4061 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
4062 RET0_UNLESS(parse_http_monthname(&date, &len, tm)); /* expect month */
4063 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
4064
4065 /* expect SP and 1DIGIT or 2DIGIT */
4066 if (parse_expect_char(&date, &len, ' '))
4067 RET0_UNLESS(parse_digit(&date, &len, &tm->tm_mday));
4068 else
4069 RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_mday));
4070
4071 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
4072 RET0_UNLESS(parse_http_time(&date, &len, tm)); /* Parse time. */
4073 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
4074 RET0_UNLESS(parse_4digit(&date, &len, &tm->tm_year)); /* year = 4DIGIT */
4075 tm->tm_year -= 1900;
4076 tm->tm_isdst = -1;
Thierry Fournier93127942016-01-20 18:49:45 +01004077 return 1;
4078}
4079
4080/* From RFC7231
4081 * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
4082 *
4083 * HTTP-date = IMF-fixdate / obs-date
4084 * obs-date = rfc850-date / asctime-date
4085 *
4086 * parses an HTTP date in the RFC format and is accepted
4087 * alternatives. <date> is the strinf containing the date,
4088 * len is the len of the string. <tm> is filled with the
4089 * parsed time. We must considers this time as GMT.
4090 */
4091int parse_http_date(const char *date, int len, struct tm *tm)
4092{
4093 if (parse_imf_date(date, len, tm))
4094 return 1;
4095
4096 if (parse_rfc850_date(date, len, tm))
4097 return 1;
4098
4099 if (parse_asctime_date(date, len, tm))
4100 return 1;
4101
4102 return 0;
4103}
4104
Willy Tarreau4deeb102021-01-29 10:47:52 +01004105/* print the time <ns> in a short form (exactly 7 chars) at the end of buffer
4106 * <out>. "-" is printed if the value is zero, "inf" if larger than 1000 years.
4107 * It returns the new buffer length, or 0 if it doesn't fit. The value will be
4108 * surrounded by <pfx> and <sfx> respectively if not NULL.
4109 */
4110int print_time_short(struct buffer *out, const char *pfx, uint64_t ns, const char *sfx)
4111{
4112 double val = ns; // 52 bits of mantissa keep ns accuracy over 52 days
4113 const char *unit;
4114
4115 if (!pfx)
4116 pfx = "";
4117 if (!sfx)
4118 sfx = "";
4119
4120 do {
4121 unit = " - "; if (val <= 0.0) break;
4122 unit = "ns"; if (val < 1000.0) break;
4123 unit = "us"; val /= 1000.0; if (val < 1000.0) break;
4124 unit = "ms"; val /= 1000.0; if (val < 1000.0) break;
4125 unit = "s "; val /= 1000.0; if (val < 60.0) break;
4126 unit = "m "; val /= 60.0; if (val < 60.0) break;
4127 unit = "h "; val /= 60.0; if (val < 24.0) break;
4128 unit = "d "; val /= 24.0; if (val < 365.0) break;
4129 unit = "yr"; val /= 365.0; if (val < 1000.0) break;
4130 unit = " inf "; val = 0.0; break;
4131 } while (0);
4132
4133 if (val <= 0.0)
4134 return chunk_appendf(out, "%s%7s%s", pfx, unit, sfx);
4135 else if (val < 10.0)
4136 return chunk_appendf(out, "%s%1.3f%s%s", pfx, val, unit, sfx);
4137 else if (val < 100.0)
4138 return chunk_appendf(out, "%s%2.2f%s%s", pfx, val, unit, sfx);
4139 else
4140 return chunk_appendf(out, "%s%3.1f%s%s", pfx, val, unit, sfx);
4141}
4142
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004143/* Dynamically allocates a string of the proper length to hold the formatted
4144 * output. NULL is returned on error. The caller is responsible for freeing the
4145 * memory area using free(). The resulting string is returned in <out> if the
4146 * pointer is not NULL. A previous version of <out> might be used to build the
4147 * new string, and it will be freed before returning if it is not NULL, which
4148 * makes it possible to build complex strings from iterative calls without
4149 * having to care about freeing intermediate values, as in the example below :
4150 *
4151 * memprintf(&err, "invalid argument: '%s'", arg);
4152 * ...
4153 * memprintf(&err, "parser said : <%s>\n", *err);
4154 * ...
4155 * free(*err);
4156 *
4157 * This means that <err> must be initialized to NULL before first invocation.
4158 * The return value also holds the allocated string, which eases error checking
4159 * and immediate consumption. If the output pointer is not used, NULL must be
Willy Tarreaueb6cead2012-09-20 19:43:14 +02004160 * passed instead and it will be ignored. The returned message will then also
4161 * be NULL so that the caller does not have to bother with freeing anything.
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004162 *
4163 * It is also convenient to use it without any free except the last one :
4164 * err = NULL;
4165 * if (!fct1(err)) report(*err);
4166 * if (!fct2(err)) report(*err);
4167 * if (!fct3(err)) report(*err);
4168 * free(*err);
Christopher Faulet93a518f2017-10-24 11:25:33 +02004169 *
4170 * memprintf relies on memvprintf. This last version can be called from any
4171 * function with variadic arguments.
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004172 */
Christopher Faulet93a518f2017-10-24 11:25:33 +02004173char *memvprintf(char **out, const char *format, va_list orig_args)
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004174{
4175 va_list args;
4176 char *ret = NULL;
4177 int allocated = 0;
4178 int needed = 0;
4179
Willy Tarreaueb6cead2012-09-20 19:43:14 +02004180 if (!out)
4181 return NULL;
4182
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004183 do {
Willy Tarreaue0609f52019-03-29 19:13:23 +01004184 char buf1;
4185
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004186 /* vsnprintf() will return the required length even when the
4187 * target buffer is NULL. We do this in a loop just in case
4188 * intermediate evaluations get wrong.
4189 */
Christopher Faulet93a518f2017-10-24 11:25:33 +02004190 va_copy(args, orig_args);
Willy Tarreaue0609f52019-03-29 19:13:23 +01004191 needed = vsnprintf(ret ? ret : &buf1, allocated, format, args);
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004192 va_end(args);
Willy Tarreau1b2fed62013-04-01 22:48:54 +02004193 if (needed < allocated) {
4194 /* Note: on Solaris 8, the first iteration always
4195 * returns -1 if allocated is zero, so we force a
4196 * retry.
4197 */
4198 if (!allocated)
4199 needed = 0;
4200 else
4201 break;
4202 }
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004203
Willy Tarreau1b2fed62013-04-01 22:48:54 +02004204 allocated = needed + 1;
Hubert Verstraete831962e2016-06-28 22:44:26 +02004205 ret = my_realloc2(ret, allocated);
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004206 } while (ret);
4207
4208 if (needed < 0) {
4209 /* an error was encountered */
Willy Tarreau61cfdf42021-02-20 10:46:51 +01004210 ha_free(&ret);
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004211 }
4212
4213 if (out) {
4214 free(*out);
4215 *out = ret;
4216 }
4217
4218 return ret;
4219}
William Lallemand421f5b52012-02-06 18:15:57 +01004220
Christopher Faulet93a518f2017-10-24 11:25:33 +02004221char *memprintf(char **out, const char *format, ...)
4222{
4223 va_list args;
4224 char *ret = NULL;
4225
4226 va_start(args, format);
4227 ret = memvprintf(out, format, args);
4228 va_end(args);
4229
4230 return ret;
4231}
4232
Willy Tarreau21c705b2012-09-14 11:40:36 +02004233/* Used to add <level> spaces before each line of <out>, unless there is only one line.
4234 * The input argument is automatically freed and reassigned. The result will have to be
Willy Tarreau70eec382012-10-10 08:56:47 +02004235 * freed by the caller. It also supports being passed a NULL which results in the same
4236 * output.
Willy Tarreau21c705b2012-09-14 11:40:36 +02004237 * Example of use :
4238 * parse(cmd, &err); (callee: memprintf(&err, ...))
4239 * fprintf(stderr, "Parser said: %s\n", indent_error(&err));
4240 * free(err);
4241 */
4242char *indent_msg(char **out, int level)
4243{
4244 char *ret, *in, *p;
4245 int needed = 0;
4246 int lf = 0;
4247 int lastlf = 0;
4248 int len;
4249
Willy Tarreau70eec382012-10-10 08:56:47 +02004250 if (!out || !*out)
4251 return NULL;
4252
Willy Tarreau21c705b2012-09-14 11:40:36 +02004253 in = *out - 1;
4254 while ((in = strchr(in + 1, '\n')) != NULL) {
4255 lastlf = in - *out;
4256 lf++;
4257 }
4258
4259 if (!lf) /* single line, no LF, return it as-is */
4260 return *out;
4261
4262 len = strlen(*out);
4263
4264 if (lf == 1 && lastlf == len - 1) {
4265 /* single line, LF at end, strip it and return as-is */
4266 (*out)[lastlf] = 0;
4267 return *out;
4268 }
4269
4270 /* OK now we have at least one LF, we need to process the whole string
4271 * as a multi-line string. What we'll do :
4272 * - prefix with an LF if there is none
4273 * - add <level> spaces before each line
4274 * This means at most ( 1 + level + (len-lf) + lf*<1+level) ) =
4275 * 1 + level + len + lf * level = 1 + level * (lf + 1) + len.
4276 */
4277
4278 needed = 1 + level * (lf + 1) + len + 1;
4279 p = ret = malloc(needed);
4280 in = *out;
4281
4282 /* skip initial LFs */
4283 while (*in == '\n')
4284 in++;
4285
4286 /* copy each line, prefixed with LF and <level> spaces, and without the trailing LF */
4287 while (*in) {
4288 *p++ = '\n';
4289 memset(p, ' ', level);
4290 p += level;
4291 do {
4292 *p++ = *in++;
4293 } while (*in && *in != '\n');
4294 if (*in)
4295 in++;
4296 }
4297 *p = 0;
4298
4299 free(*out);
4300 *out = ret;
4301
4302 return ret;
4303}
4304
Willy Tarreaua2c99112019-08-21 13:17:37 +02004305/* makes a copy of message <in> into <out>, with each line prefixed with <pfx>
4306 * and end of lines replaced with <eol> if not 0. The first line to indent has
4307 * to be indicated in <first> (starts at zero), so that it is possible to skip
4308 * indenting the first line if it has to be appended after an existing message.
4309 * Empty strings are never indented, and NULL strings are considered empty both
4310 * for <in> and <pfx>. It returns non-zero if an EOL was appended as the last
4311 * character, non-zero otherwise.
4312 */
4313int append_prefixed_str(struct buffer *out, const char *in, const char *pfx, char eol, int first)
4314{
4315 int bol, lf;
4316 int pfxlen = pfx ? strlen(pfx) : 0;
4317
4318 if (!in)
4319 return 0;
4320
4321 bol = 1;
4322 lf = 0;
4323 while (*in) {
4324 if (bol && pfxlen) {
4325 if (first > 0)
4326 first--;
4327 else
4328 b_putblk(out, pfx, pfxlen);
4329 bol = 0;
4330 }
4331
4332 lf = (*in == '\n');
4333 bol |= lf;
4334 b_putchr(out, (lf && eol) ? eol : *in);
4335 in++;
4336 }
4337 return lf;
4338}
4339
Willy Tarreau9d22e562019-03-29 18:49:09 +01004340/* removes environment variable <name> from the environment as found in
4341 * environ. This is only provided as an alternative for systems without
4342 * unsetenv() (old Solaris and AIX versions). THIS IS NOT THREAD SAFE.
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05004343 * The principle is to scan environ for each occurrence of variable name
Willy Tarreau9d22e562019-03-29 18:49:09 +01004344 * <name> and to replace the matching pointers with the last pointer of
4345 * the array (since variables are not ordered).
4346 * It always returns 0 (success).
4347 */
4348int my_unsetenv(const char *name)
4349{
4350 extern char **environ;
4351 char **p = environ;
4352 int vars;
4353 int next;
4354 int len;
4355
4356 len = strlen(name);
4357 for (vars = 0; p[vars]; vars++)
4358 ;
4359 next = 0;
4360 while (next < vars) {
4361 if (strncmp(p[next], name, len) != 0 || p[next][len] != '=') {
4362 next++;
4363 continue;
4364 }
4365 if (next < vars - 1)
4366 p[next] = p[vars - 1];
4367 p[--vars] = NULL;
4368 }
4369 return 0;
4370}
4371
Willy Tarreaudad36a32013-03-11 01:20:04 +01004372/* Convert occurrences of environment variables in the input string to their
4373 * corresponding value. A variable is identified as a series of alphanumeric
4374 * characters or underscores following a '$' sign. The <in> string must be
4375 * free()able. NULL returns NULL. The resulting string might be reallocated if
4376 * some expansion is made. Variable names may also be enclosed into braces if
4377 * needed (eg: to concatenate alphanum characters).
4378 */
4379char *env_expand(char *in)
4380{
4381 char *txt_beg;
4382 char *out;
4383 char *txt_end;
4384 char *var_beg;
4385 char *var_end;
4386 char *value;
4387 char *next;
4388 int out_len;
4389 int val_len;
4390
4391 if (!in)
4392 return in;
4393
4394 value = out = NULL;
4395 out_len = 0;
4396
4397 txt_beg = in;
4398 do {
4399 /* look for next '$' sign in <in> */
4400 for (txt_end = txt_beg; *txt_end && *txt_end != '$'; txt_end++);
4401
4402 if (!*txt_end && !out) /* end and no expansion performed */
4403 return in;
4404
4405 val_len = 0;
4406 next = txt_end;
4407 if (*txt_end == '$') {
4408 char save;
4409
4410 var_beg = txt_end + 1;
4411 if (*var_beg == '{')
4412 var_beg++;
4413
4414 var_end = var_beg;
Willy Tarreau90807112020-02-25 08:16:33 +01004415 while (isalnum((unsigned char)*var_end) || *var_end == '_') {
Willy Tarreaudad36a32013-03-11 01:20:04 +01004416 var_end++;
4417 }
4418
4419 next = var_end;
4420 if (*var_end == '}' && (var_beg > txt_end + 1))
4421 next++;
4422
4423 /* get value of the variable name at this location */
4424 save = *var_end;
4425 *var_end = '\0';
4426 value = getenv(var_beg);
4427 *var_end = save;
4428 val_len = value ? strlen(value) : 0;
4429 }
4430
Hubert Verstraete831962e2016-06-28 22:44:26 +02004431 out = my_realloc2(out, out_len + (txt_end - txt_beg) + val_len + 1);
Willy Tarreaudad36a32013-03-11 01:20:04 +01004432 if (txt_end > txt_beg) {
4433 memcpy(out + out_len, txt_beg, txt_end - txt_beg);
4434 out_len += txt_end - txt_beg;
4435 }
4436 if (val_len) {
4437 memcpy(out + out_len, value, val_len);
4438 out_len += val_len;
4439 }
4440 out[out_len] = 0;
4441 txt_beg = next;
4442 } while (*txt_beg);
4443
4444 /* here we know that <out> was allocated and that we don't need <in> anymore */
4445 free(in);
4446 return out;
4447}
4448
de Lafond Guillaume88c278f2013-04-15 19:27:10 +02004449
4450/* same as strstr() but case-insensitive and with limit length */
4451const char *strnistr(const char *str1, int len_str1, const char *str2, int len_str2)
4452{
4453 char *pptr, *sptr, *start;
Willy Tarreauc8746532014-05-28 23:05:07 +02004454 unsigned int slen, plen;
4455 unsigned int tmp1, tmp2;
de Lafond Guillaume88c278f2013-04-15 19:27:10 +02004456
4457 if (str1 == NULL || len_str1 == 0) // search pattern into an empty string => search is not found
4458 return NULL;
4459
4460 if (str2 == NULL || len_str2 == 0) // pattern is empty => every str1 match
4461 return str1;
4462
4463 if (len_str1 < len_str2) // pattern is longer than string => search is not found
4464 return NULL;
4465
4466 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 +02004467 while (toupper((unsigned char)*start) != toupper((unsigned char)*str2)) {
de Lafond Guillaume88c278f2013-04-15 19:27:10 +02004468 start++;
4469 slen--;
4470 tmp1++;
4471
4472 if (tmp1 >= len_str1)
4473 return NULL;
4474
4475 /* if pattern longer than string */
4476 if (slen < plen)
4477 return NULL;
4478 }
4479
4480 sptr = start;
4481 pptr = (char *)str2;
4482
4483 tmp2 = 0;
Willy Tarreauf278eec2020-07-05 21:46:32 +02004484 while (toupper((unsigned char)*sptr) == toupper((unsigned char)*pptr)) {
de Lafond Guillaume88c278f2013-04-15 19:27:10 +02004485 sptr++;
4486 pptr++;
4487 tmp2++;
4488
4489 if (*pptr == '\0' || tmp2 == len_str2) /* end of pattern found */
4490 return start;
4491 if (*sptr == '\0' || tmp2 == len_str1) /* end of string found and the pattern is not fully found */
4492 return NULL;
4493 }
4494 }
4495 return NULL;
4496}
4497
Willy Tarreau3ff476e2022-03-30 10:02:56 +02004498/* Returns true if s1 < s2 < s3 otherwise zero. Both s1 and s3 may be NULL and
4499 * in this case only non-null strings are compared. This allows to pass initial
4500 * values in iterators and in sort functions.
4501 */
4502int strordered(const char *s1, const char *s2, const char *s3)
4503{
4504 return (!s1 || strcmp(s1, s2) < 0) && (!s3 || strcmp(s2, s3) < 0);
4505}
4506
Thierry FOURNIER317e1c42014-08-12 10:20:47 +02004507/* This function read the next valid utf8 char.
4508 * <s> is the byte srray to be decode, <len> is its length.
4509 * The function returns decoded char encoded like this:
4510 * The 4 msb are the return code (UTF8_CODE_*), the 4 lsb
4511 * are the length read. The decoded character is stored in <c>.
4512 */
4513unsigned char utf8_next(const char *s, int len, unsigned int *c)
4514{
4515 const unsigned char *p = (unsigned char *)s;
4516 int dec;
4517 unsigned char code = UTF8_CODE_OK;
4518
4519 if (len < 1)
4520 return UTF8_CODE_OK;
4521
4522 /* Check the type of UTF8 sequence
4523 *
4524 * 0... .... 0x00 <= x <= 0x7f : 1 byte: ascii char
4525 * 10.. .... 0x80 <= x <= 0xbf : invalid sequence
4526 * 110. .... 0xc0 <= x <= 0xdf : 2 bytes
4527 * 1110 .... 0xe0 <= x <= 0xef : 3 bytes
4528 * 1111 0... 0xf0 <= x <= 0xf7 : 4 bytes
4529 * 1111 10.. 0xf8 <= x <= 0xfb : 5 bytes
4530 * 1111 110. 0xfc <= x <= 0xfd : 6 bytes
4531 * 1111 111. 0xfe <= x <= 0xff : invalid sequence
4532 */
4533 switch (*p) {
4534 case 0x00 ... 0x7f:
4535 *c = *p;
4536 return UTF8_CODE_OK | 1;
4537
4538 case 0x80 ... 0xbf:
4539 *c = *p;
4540 return UTF8_CODE_BADSEQ | 1;
4541
4542 case 0xc0 ... 0xdf:
4543 if (len < 2) {
4544 *c = *p;
4545 return UTF8_CODE_BADSEQ | 1;
4546 }
4547 *c = *p & 0x1f;
4548 dec = 1;
4549 break;
4550
4551 case 0xe0 ... 0xef:
4552 if (len < 3) {
4553 *c = *p;
4554 return UTF8_CODE_BADSEQ | 1;
4555 }
4556 *c = *p & 0x0f;
4557 dec = 2;
4558 break;
4559
4560 case 0xf0 ... 0xf7:
4561 if (len < 4) {
4562 *c = *p;
4563 return UTF8_CODE_BADSEQ | 1;
4564 }
4565 *c = *p & 0x07;
4566 dec = 3;
4567 break;
4568
4569 case 0xf8 ... 0xfb:
4570 if (len < 5) {
4571 *c = *p;
4572 return UTF8_CODE_BADSEQ | 1;
4573 }
4574 *c = *p & 0x03;
4575 dec = 4;
4576 break;
4577
4578 case 0xfc ... 0xfd:
4579 if (len < 6) {
4580 *c = *p;
4581 return UTF8_CODE_BADSEQ | 1;
4582 }
4583 *c = *p & 0x01;
4584 dec = 5;
4585 break;
4586
4587 case 0xfe ... 0xff:
4588 default:
4589 *c = *p;
4590 return UTF8_CODE_BADSEQ | 1;
4591 }
4592
4593 p++;
4594
4595 while (dec > 0) {
4596
4597 /* need 0x10 for the 2 first bits */
4598 if ( ( *p & 0xc0 ) != 0x80 )
4599 return UTF8_CODE_BADSEQ | ((p-(unsigned char *)s)&0xffff);
4600
4601 /* add data at char */
4602 *c = ( *c << 6 ) | ( *p & 0x3f );
4603
4604 dec--;
4605 p++;
4606 }
4607
4608 /* Check ovelong encoding.
4609 * 1 byte : 5 + 6 : 11 : 0x80 ... 0x7ff
4610 * 2 bytes : 4 + 6 + 6 : 16 : 0x800 ... 0xffff
4611 * 3 bytes : 3 + 6 + 6 + 6 : 21 : 0x10000 ... 0x1fffff
4612 */
Thierry FOURNIER9e7ec082015-03-12 19:32:38 +01004613 if (( *c <= 0x7f && (p-(unsigned char *)s) > 1) ||
Thierry FOURNIER317e1c42014-08-12 10:20:47 +02004614 (*c >= 0x80 && *c <= 0x7ff && (p-(unsigned char *)s) > 2) ||
4615 (*c >= 0x800 && *c <= 0xffff && (p-(unsigned char *)s) > 3) ||
4616 (*c >= 0x10000 && *c <= 0x1fffff && (p-(unsigned char *)s) > 4))
4617 code |= UTF8_CODE_OVERLONG;
4618
4619 /* Check invalid UTF8 range. */
4620 if ((*c >= 0xd800 && *c <= 0xdfff) ||
4621 (*c >= 0xfffe && *c <= 0xffff))
4622 code |= UTF8_CODE_INVRANGE;
4623
4624 return code | ((p-(unsigned char *)s)&0x0f);
4625}
4626
Maxime de Roucydc887852016-05-13 23:52:54 +02004627/* append a copy of string <str> (in a wordlist) at the end of the list <li>
4628 * On failure : return 0 and <err> filled with an error message.
4629 * The caller is responsible for freeing the <err> and <str> copy
4630 * memory area using free()
4631 */
4632int list_append_word(struct list *li, const char *str, char **err)
4633{
4634 struct wordlist *wl;
4635
4636 wl = calloc(1, sizeof(*wl));
4637 if (!wl) {
4638 memprintf(err, "out of memory");
4639 goto fail_wl;
4640 }
4641
4642 wl->s = strdup(str);
4643 if (!wl->s) {
4644 memprintf(err, "out of memory");
4645 goto fail_wl_s;
4646 }
4647
Willy Tarreau2b718102021-04-21 07:32:39 +02004648 LIST_APPEND(li, &wl->list);
Maxime de Roucydc887852016-05-13 23:52:54 +02004649
4650 return 1;
4651
4652fail_wl_s:
4653 free(wl->s);
4654fail_wl:
4655 free(wl);
4656 return 0;
4657}
4658
Willy Tarreau37101052019-05-20 16:48:20 +02004659/* indicates if a memory location may safely be read or not. The trick consists
4660 * in performing a harmless syscall using this location as an input and letting
4661 * the operating system report whether it's OK or not. For this we have the
4662 * stat() syscall, which will return EFAULT when the memory location supposed
4663 * to contain the file name is not readable. If it is readable it will then
4664 * either return 0 if the area contains an existing file name, or -1 with
4665 * another code. This must not be abused, and some audit systems might detect
4666 * this as abnormal activity. It's used only for unsafe dumps.
4667 */
4668int may_access(const void *ptr)
4669{
4670 struct stat buf;
4671
4672 if (stat(ptr, &buf) == 0)
4673 return 1;
4674 if (errno == EFAULT)
4675 return 0;
4676 return 1;
4677}
4678
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004679/* print a string of text buffer to <out>. The format is :
4680 * Non-printable chars \t, \n, \r and \e are * encoded in C format.
4681 * Other non-printable chars are encoded "\xHH". Space, '\', and '=' are also escaped.
4682 * Print stopped if null char or <bsize> is reached, or if no more place in the chunk.
4683 */
Willy Tarreau83061a82018-07-13 11:56:34 +02004684int dump_text(struct buffer *out, const char *buf, int bsize)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004685{
4686 unsigned char c;
Tim Duesterhus18795d42021-08-29 00:58:22 +02004687 size_t ptr = 0;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004688
Tim Duesterhus18795d42021-08-29 00:58:22 +02004689 while (ptr < bsize && buf[ptr]) {
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004690 c = buf[ptr];
Willy Tarreau90807112020-02-25 08:16:33 +01004691 if (isprint((unsigned char)c) && isascii((unsigned char)c) && c != '\\' && c != ' ' && c != '=') {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004692 if (out->data > out->size - 1)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004693 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004694 out->area[out->data++] = c;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004695 }
4696 else if (c == '\t' || c == '\n' || c == '\r' || c == '\e' || c == '\\' || c == ' ' || c == '=') {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004697 if (out->data > out->size - 2)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004698 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004699 out->area[out->data++] = '\\';
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004700 switch (c) {
4701 case ' ': c = ' '; break;
4702 case '\t': c = 't'; break;
4703 case '\n': c = 'n'; break;
4704 case '\r': c = 'r'; break;
4705 case '\e': c = 'e'; break;
4706 case '\\': c = '\\'; break;
4707 case '=': c = '='; break;
4708 }
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004709 out->area[out->data++] = c;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004710 }
4711 else {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004712 if (out->data > out->size - 4)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004713 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004714 out->area[out->data++] = '\\';
4715 out->area[out->data++] = 'x';
4716 out->area[out->data++] = hextab[(c >> 4) & 0xF];
4717 out->area[out->data++] = hextab[c & 0xF];
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004718 }
4719 ptr++;
4720 }
4721
4722 return ptr;
4723}
4724
4725/* print a buffer in hexa.
4726 * Print stopped if <bsize> is reached, or if no more place in the chunk.
4727 */
Willy Tarreau83061a82018-07-13 11:56:34 +02004728int dump_binary(struct buffer *out, const char *buf, int bsize)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004729{
4730 unsigned char c;
4731 int ptr = 0;
4732
4733 while (ptr < bsize) {
4734 c = buf[ptr];
4735
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004736 if (out->data > out->size - 2)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004737 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004738 out->area[out->data++] = hextab[(c >> 4) & 0xF];
4739 out->area[out->data++] = hextab[c & 0xF];
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004740
4741 ptr++;
4742 }
4743 return ptr;
4744}
4745
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004746/* Appends into buffer <out> a hex dump of memory area <buf> for <len> bytes,
4747 * prepending each line with prefix <pfx>. The output is *not* initialized.
4748 * The output will not wrap pas the buffer's end so it is more optimal if the
4749 * caller makes sure the buffer is aligned first. A trailing zero will always
4750 * be appended (and not counted) if there is room for it. The caller must make
Willy Tarreau37101052019-05-20 16:48:20 +02004751 * sure that the area is dumpable first. If <unsafe> is non-null, the memory
4752 * locations are checked first for being readable.
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004753 */
Willy Tarreau37101052019-05-20 16:48:20 +02004754void dump_hex(struct buffer *out, const char *pfx, const void *buf, int len, int unsafe)
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004755{
4756 const unsigned char *d = buf;
4757 int i, j, start;
4758
4759 d = (const unsigned char *)(((unsigned long)buf) & -16);
4760 start = ((unsigned long)buf) & 15;
4761
4762 for (i = 0; i < start + len; i += 16) {
4763 chunk_appendf(out, (sizeof(void *) == 4) ? "%s%8p: " : "%s%16p: ", pfx, d + i);
4764
Willy Tarreau37101052019-05-20 16:48:20 +02004765 // 0: unchecked, 1: checked safe, 2: danger
4766 unsafe = !!unsafe;
4767 if (unsafe && !may_access(d + i))
4768 unsafe = 2;
4769
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004770 for (j = 0; j < 16; j++) {
Willy Tarreau37101052019-05-20 16:48:20 +02004771 if ((i + j < start) || (i + j >= start + len))
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004772 chunk_strcat(out, "'' ");
Willy Tarreau37101052019-05-20 16:48:20 +02004773 else if (unsafe > 1)
4774 chunk_strcat(out, "** ");
4775 else
4776 chunk_appendf(out, "%02x ", d[i + j]);
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004777
4778 if (j == 7)
4779 chunk_strcat(out, "- ");
4780 }
4781 chunk_strcat(out, " ");
4782 for (j = 0; j < 16; j++) {
Willy Tarreau37101052019-05-20 16:48:20 +02004783 if ((i + j < start) || (i + j >= start + len))
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004784 chunk_strcat(out, "'");
Willy Tarreau37101052019-05-20 16:48:20 +02004785 else if (unsafe > 1)
4786 chunk_strcat(out, "*");
Willy Tarreau90807112020-02-25 08:16:33 +01004787 else if (isprint((unsigned char)d[i + j]))
Willy Tarreau37101052019-05-20 16:48:20 +02004788 chunk_appendf(out, "%c", d[i + j]);
4789 else
4790 chunk_strcat(out, ".");
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004791 }
4792 chunk_strcat(out, "\n");
4793 }
4794}
4795
Willy Tarreau762fb3e2020-03-03 15:57:10 +01004796/* dumps <pfx> followed by <n> bytes from <addr> in hex form into buffer <buf>
4797 * enclosed in brackets after the address itself, formatted on 14 chars
4798 * including the "0x" prefix. This is meant to be used as a prefix for code
4799 * areas. For example:
4800 * "0x7f10b6557690 [48 c7 c0 0f 00 00 00 0f]"
4801 * It relies on may_access() to know if the bytes are dumpable, otherwise "--"
4802 * is emitted. A NULL <pfx> will be considered empty.
4803 */
4804void dump_addr_and_bytes(struct buffer *buf, const char *pfx, const void *addr, int n)
4805{
4806 int ok = 0;
4807 int i;
4808
4809 chunk_appendf(buf, "%s%#14lx [", pfx ? pfx : "", (long)addr);
4810
4811 for (i = 0; i < n; i++) {
4812 if (i == 0 || (((long)(addr + i) ^ (long)(addr)) & 4096))
4813 ok = may_access(addr + i);
4814 if (ok)
4815 chunk_appendf(buf, "%02x%s", ((uint8_t*)addr)[i], (i<n-1) ? " " : "]");
4816 else
4817 chunk_appendf(buf, "--%s", (i<n-1) ? " " : "]");
4818 }
4819}
4820
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004821/* print a line of text buffer (limited to 70 bytes) to <out>. The format is :
4822 * <2 spaces> <offset=5 digits> <space or plus> <space> <70 chars max> <\n>
4823 * which is 60 chars per line. Non-printable chars \t, \n, \r and \e are
4824 * encoded in C format. Other non-printable chars are encoded "\xHH". Original
4825 * lines are respected within the limit of 70 output chars. Lines that are
4826 * continuation of a previous truncated line begin with "+" instead of " "
4827 * after the offset. The new pointer is returned.
4828 */
Willy Tarreau83061a82018-07-13 11:56:34 +02004829int dump_text_line(struct buffer *out, const char *buf, int bsize, int len,
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004830 int *line, int ptr)
4831{
4832 int end;
4833 unsigned char c;
4834
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004835 end = out->data + 80;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004836 if (end > out->size)
4837 return ptr;
4838
4839 chunk_appendf(out, " %05d%c ", ptr, (ptr == *line) ? ' ' : '+');
4840
4841 while (ptr < len && ptr < bsize) {
4842 c = buf[ptr];
Willy Tarreau90807112020-02-25 08:16:33 +01004843 if (isprint((unsigned char)c) && isascii((unsigned char)c) && c != '\\') {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004844 if (out->data > end - 2)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004845 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004846 out->area[out->data++] = c;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004847 } else if (c == '\t' || c == '\n' || c == '\r' || c == '\e' || c == '\\') {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004848 if (out->data > end - 3)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004849 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004850 out->area[out->data++] = '\\';
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004851 switch (c) {
4852 case '\t': c = 't'; break;
4853 case '\n': c = 'n'; break;
4854 case '\r': c = 'r'; break;
4855 case '\e': c = 'e'; break;
4856 case '\\': c = '\\'; break;
4857 }
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004858 out->area[out->data++] = c;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004859 } else {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004860 if (out->data > end - 5)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004861 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004862 out->area[out->data++] = '\\';
4863 out->area[out->data++] = 'x';
4864 out->area[out->data++] = hextab[(c >> 4) & 0xF];
4865 out->area[out->data++] = hextab[c & 0xF];
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004866 }
4867 if (buf[ptr++] == '\n') {
4868 /* we had a line break, let's return now */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004869 out->area[out->data++] = '\n';
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004870 *line = ptr;
4871 return ptr;
4872 }
4873 }
4874 /* we have an incomplete line, we return it as-is */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004875 out->area[out->data++] = '\n';
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004876 return ptr;
4877}
4878
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004879/* displays a <len> long memory block at <buf>, assuming first byte of <buf>
Willy Tarreaued936c52017-04-27 18:03:20 +02004880 * has address <baseaddr>. String <pfx> may be placed as a prefix in front of
4881 * each line. It may be NULL if unused. The output is emitted to file <out>.
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004882 */
Willy Tarreaued936c52017-04-27 18:03:20 +02004883void debug_hexdump(FILE *out, const char *pfx, const char *buf,
4884 unsigned int baseaddr, int len)
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004885{
Willy Tarreau73459792017-04-11 07:58:08 +02004886 unsigned int i;
4887 int b, j;
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004888
4889 for (i = 0; i < (len + (baseaddr & 15)); i += 16) {
4890 b = i - (baseaddr & 15);
Willy Tarreaued936c52017-04-27 18:03:20 +02004891 fprintf(out, "%s%08x: ", pfx ? pfx : "", i + (baseaddr & ~15));
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004892 for (j = 0; j < 8; j++) {
4893 if (b + j >= 0 && b + j < len)
4894 fprintf(out, "%02x ", (unsigned char)buf[b + j]);
4895 else
4896 fprintf(out, " ");
4897 }
4898
4899 if (b + j >= 0 && b + j < len)
4900 fputc('-', out);
4901 else
4902 fputc(' ', out);
4903
4904 for (j = 8; j < 16; j++) {
4905 if (b + j >= 0 && b + j < len)
4906 fprintf(out, " %02x", (unsigned char)buf[b + j]);
4907 else
4908 fprintf(out, " ");
4909 }
4910
4911 fprintf(out, " ");
4912 for (j = 0; j < 16; j++) {
4913 if (b + j >= 0 && b + j < len) {
4914 if (isprint((unsigned char)buf[b + j]))
4915 fputc((unsigned char)buf[b + j], out);
4916 else
4917 fputc('.', out);
4918 }
4919 else
4920 fputc(' ', out);
4921 }
4922 fputc('\n', out);
4923 }
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004924}
4925
Willy Tarreaubb869862020-04-16 10:52:41 +02004926/* Tries to report the executable path name on platforms supporting this. If
4927 * not found or not possible, returns NULL.
4928 */
4929const char *get_exec_path()
4930{
4931 const char *ret = NULL;
4932
David Carlier43a56852022-03-04 15:50:48 +00004933#if defined(__linux__) && defined(__GLIBC__) && (__GLIBC__ > 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 16))
Willy Tarreaubb869862020-04-16 10:52:41 +02004934 long execfn = getauxval(AT_EXECFN);
4935
4936 if (execfn && execfn != ENOENT)
4937 ret = (const char *)execfn;
devnexen@gmail.comc4e52322021-08-17 12:55:49 +01004938#elif defined(__FreeBSD__)
4939 Elf_Auxinfo *auxv;
4940 for (auxv = __elf_aux_vector; auxv->a_type != AT_NULL; ++auxv) {
4941 if (auxv->a_type == AT_EXECPATH) {
4942 ret = (const char *)auxv->a_un.a_ptr;
4943 break;
4944 }
4945 }
David Carlierbd2cced2021-08-17 08:44:25 +01004946#elif defined(__NetBSD__)
4947 AuxInfo *auxv;
4948 for (auxv = _dlauxinfo(); auxv->a_type != AT_NULL; ++auxv) {
4949 if (auxv->a_type == AT_SUN_EXECNAME) {
4950 ret = (const char *)auxv->a_v;
4951 break;
4952 }
4953 }
David Carlier7198c702022-05-14 17:15:49 +01004954#elif defined(__sun)
4955 ret = getexecname();
Willy Tarreaubb869862020-04-16 10:52:41 +02004956#endif
4957 return ret;
4958}
4959
Baruch Siache1651b22020-07-24 07:52:20 +03004960#if (defined(__ELF__) && !defined(__linux__)) || defined(USE_DL)
Willy Tarreau9133e482020-03-04 10:19:36 +01004961/* calls dladdr() or dladdr1() on <addr> and <dli>. If dladdr1 is available,
4962 * also returns the symbol size in <size>, otherwise returns 0 there.
4963 */
4964static int dladdr_and_size(const void *addr, Dl_info *dli, size_t *size)
4965{
4966 int ret;
Willy Tarreau7b2108c2021-08-30 10:15:35 +02004967#if defined(__GLIBC__) && (__GLIBC__ > 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 3)) // most detailed one
Willy Tarreauf3d5c4b2022-01-28 09:42:29 +01004968 const ElfW(Sym) *sym __attribute__((may_alias));
Willy Tarreau9133e482020-03-04 10:19:36 +01004969
4970 ret = dladdr1(addr, dli, (void **)&sym, RTLD_DL_SYMENT);
4971 if (ret)
4972 *size = sym ? sym->st_size : 0;
4973#else
David Carlierae5c42f2021-12-31 08:15:29 +00004974#if defined(__sun)
4975 ret = dladdr((void *)addr, dli);
4976#else
Willy Tarreau9133e482020-03-04 10:19:36 +01004977 ret = dladdr(addr, dli);
David Carlierae5c42f2021-12-31 08:15:29 +00004978#endif
Willy Tarreau9133e482020-03-04 10:19:36 +01004979 *size = 0;
4980#endif
4981 return ret;
4982}
Willy Tarreau64192392021-05-05 09:06:21 +02004983
Willy Tarreau5b3cd952022-07-18 13:58:17 +02004984/* Sets build_is_static to true if we detect a static build. Some older glibcs
4985 * tend to crash inside dlsym() in static builds, but tests show that at least
4986 * dladdr() still works (and will fail to resolve anything of course). Thus we
4987 * try to determine if we're on a static build to avoid calling dlsym() in this
4988 * case.
Willy Tarreau288dc1d2022-07-16 13:49:34 +02004989 */
Willy Tarreau5b3cd952022-07-18 13:58:17 +02004990void check_if_static_build()
Willy Tarreau288dc1d2022-07-16 13:49:34 +02004991{
Willy Tarreau5b3cd952022-07-18 13:58:17 +02004992 Dl_info dli = { };
4993 size_t size = 0;
4994
4995 /* Now let's try to be smarter */
4996 if (!dladdr_and_size(&main, &dli, &size))
4997 build_is_static = 1;
4998 else
4999 build_is_static = 0;
Willy Tarreau288dc1d2022-07-16 13:49:34 +02005000}
5001
Willy Tarreau5b3cd952022-07-18 13:58:17 +02005002INITCALL0(STG_PREPARE, check_if_static_build);
Willy Tarreau288dc1d2022-07-16 13:49:34 +02005003
Willy Tarreau64192392021-05-05 09:06:21 +02005004/* Tries to retrieve the address of the first occurrence symbol <name>.
5005 * Note that NULL in return is not always an error as a symbol may have that
5006 * address in special situations.
5007 */
5008void *get_sym_curr_addr(const char *name)
5009{
5010 void *ptr = NULL;
5011
5012#ifdef RTLD_DEFAULT
Willy Tarreau5b3cd952022-07-18 13:58:17 +02005013 if (!build_is_static)
Willy Tarreau288dc1d2022-07-16 13:49:34 +02005014 ptr = dlsym(RTLD_DEFAULT, name);
Willy Tarreau64192392021-05-05 09:06:21 +02005015#endif
5016 return ptr;
5017}
5018
5019
5020/* Tries to retrieve the address of the next occurrence of symbol <name>
5021 * Note that NULL in return is not always an error as a symbol may have that
5022 * address in special situations.
5023 */
5024void *get_sym_next_addr(const char *name)
5025{
5026 void *ptr = NULL;
5027
5028#ifdef RTLD_NEXT
Willy Tarreau5b3cd952022-07-18 13:58:17 +02005029 if (!build_is_static)
Willy Tarreau288dc1d2022-07-16 13:49:34 +02005030 ptr = dlsym(RTLD_NEXT, name);
Willy Tarreau9133e482020-03-04 10:19:36 +01005031#endif
Willy Tarreau64192392021-05-05 09:06:21 +02005032 return ptr;
5033}
5034
5035#else /* elf & linux & dl */
5036
5037/* no possible resolving on other platforms at the moment */
5038void *get_sym_curr_addr(const char *name)
5039{
5040 return NULL;
5041}
5042
5043void *get_sym_next_addr(const char *name)
5044{
5045 return NULL;
5046}
5047
5048#endif /* elf & linux & dl */
Willy Tarreau9133e482020-03-04 10:19:36 +01005049
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01005050/* Tries to append to buffer <buf> some indications about the symbol at address
5051 * <addr> using the following form:
5052 * lib:+0xoffset (unresolvable address from lib's base)
5053 * main+0xoffset (unresolvable address from main (+/-))
5054 * lib:main+0xoffset (unresolvable lib address from main (+/-))
5055 * name (resolved exact exec address)
5056 * lib:name (resolved exact lib address)
5057 * name+0xoffset/0xsize (resolved address within exec symbol)
5058 * lib:name+0xoffset/0xsize (resolved address within lib symbol)
5059 *
5060 * The file name (lib or executable) is limited to what lies between the last
5061 * '/' and the first following '.'. An optional prefix <pfx> is prepended before
5062 * 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 +03005063 * that contains the "main" symbol, or when __ELF__ && USE_DL are not set.
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01005064 *
5065 * The symbol's base address is returned, or NULL when unresolved, in order to
5066 * allow the caller to match it against known ones.
5067 */
Willy Tarreau45fd1032021-01-20 14:37:59 +01005068const void *resolve_sym_name(struct buffer *buf, const char *pfx, const void *addr)
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01005069{
5070 const struct {
5071 const void *func;
5072 const char *name;
5073 } fcts[] = {
5074 { .func = process_stream, .name = "process_stream" },
5075 { .func = task_run_applet, .name = "task_run_applet" },
Willy Tarreau462b9892022-05-18 18:06:53 +02005076 { .func = sc_conn_io_cb, .name = "sc_conn_io_cb" },
Willy Tarreau586f71b2020-12-11 15:54:36 +01005077 { .func = sock_conn_iocb, .name = "sock_conn_iocb" },
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01005078 { .func = dgram_fd_handler, .name = "dgram_fd_handler" },
5079 { .func = listener_accept, .name = "listener_accept" },
Willy Tarreaud597ec22021-01-29 14:29:06 +01005080 { .func = manage_global_listener_queue, .name = "manage_global_listener_queue" },
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01005081 { .func = poller_pipe_io_handler, .name = "poller_pipe_io_handler" },
5082 { .func = mworker_accept_wrapper, .name = "mworker_accept_wrapper" },
Willy Tarreau02922e12021-01-29 12:27:57 +01005083 { .func = session_expire_embryonic, .name = "session_expire_embryonic" },
Willy Tarreaufb5401f2021-01-29 12:25:23 +01005084#ifdef USE_THREAD
5085 { .func = accept_queue_process, .name = "accept_queue_process" },
5086#endif
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01005087#ifdef USE_LUA
5088 { .func = hlua_process_task, .name = "hlua_process_task" },
5089#endif
Ilya Shipitsinbdec3ba2020-11-14 01:56:34 +05005090#ifdef SSL_MODE_ASYNC
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01005091 { .func = ssl_async_fd_free, .name = "ssl_async_fd_free" },
5092 { .func = ssl_async_fd_handler, .name = "ssl_async_fd_handler" },
5093#endif
5094 };
5095
Baruch Siache1651b22020-07-24 07:52:20 +03005096#if (defined(__ELF__) && !defined(__linux__)) || defined(USE_DL)
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01005097 Dl_info dli, dli_main;
Willy Tarreau9133e482020-03-04 10:19:36 +01005098 size_t size;
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01005099 const char *fname, *p;
5100#endif
5101 int i;
5102
5103 if (pfx)
5104 chunk_appendf(buf, "%s", pfx);
5105
5106 for (i = 0; i < sizeof(fcts) / sizeof(fcts[0]); i++) {
5107 if (addr == fcts[i].func) {
5108 chunk_appendf(buf, "%s", fcts[i].name);
5109 return addr;
5110 }
5111 }
5112
Baruch Siache1651b22020-07-24 07:52:20 +03005113#if (defined(__ELF__) && !defined(__linux__)) || defined(USE_DL)
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01005114 /* Now let's try to be smarter */
Willy Tarreau9133e482020-03-04 10:19:36 +01005115 if (!dladdr_and_size(addr, &dli, &size))
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01005116 goto unknown;
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01005117
5118 /* 1. prefix the library name if it's not the same object as the one
5119 * that contains the main function. The name is picked between last '/'
5120 * and first following '.'.
5121 */
5122 if (!dladdr(main, &dli_main))
5123 dli_main.dli_fbase = NULL;
5124
5125 if (dli_main.dli_fbase != dli.dli_fbase) {
5126 fname = dli.dli_fname;
5127 p = strrchr(fname, '/');
5128 if (p++)
5129 fname = p;
5130 p = strchr(fname, '.');
5131 if (!p)
5132 p = fname + strlen(fname);
5133
5134 chunk_appendf(buf, "%.*s:", (int)(long)(p - fname), fname);
5135 }
5136
5137 /* 2. symbol name */
5138 if (dli.dli_sname) {
5139 /* known, dump it and return symbol's address (exact or relative) */
5140 chunk_appendf(buf, "%s", dli.dli_sname);
5141 if (addr != dli.dli_saddr) {
5142 chunk_appendf(buf, "+%#lx", (long)(addr - dli.dli_saddr));
Willy Tarreau9133e482020-03-04 10:19:36 +01005143 if (size)
5144 chunk_appendf(buf, "/%#lx", (long)size);
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01005145 }
5146 return dli.dli_saddr;
5147 }
5148 else if (dli_main.dli_fbase != dli.dli_fbase) {
5149 /* unresolved symbol from a known library, report relative offset */
5150 chunk_appendf(buf, "+%#lx", (long)(addr - dli.dli_fbase));
5151 return NULL;
5152 }
Baruch Siache1651b22020-07-24 07:52:20 +03005153#endif /* __ELF__ && !__linux__ || USE_DL */
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01005154 unknown:
5155 /* unresolved symbol from the main file, report relative offset to main */
5156 if ((void*)addr < (void*)main)
5157 chunk_appendf(buf, "main-%#lx", (long)((void*)main - addr));
5158 else
5159 chunk_appendf(buf, "main+%#lx", (long)(addr - (void*)main));
5160 return NULL;
Willy Tarreau0ebb5112016-12-05 00:10:57 +01005161}
5162
Willy Tarreau6ab7b212021-12-28 09:57:10 +01005163/* On systems where this is supported, let's provide a possibility to enumerate
5164 * the list of object files. The output is appended to a buffer initialized by
5165 * the caller, with one name per line. A trailing zero is always emitted if data
5166 * are written. Only real objects are dumped (executable and .so libs). The
5167 * function returns non-zero if it dumps anything. These functions do not make
5168 * use of the trash so that it is possible for the caller to call them with the
5169 * trash on input. The output format may be platform-specific but at least one
5170 * version must emit raw object file names when argument is zero.
5171 */
5172#if defined(HA_HAVE_DUMP_LIBS)
5173# if defined(HA_HAVE_DL_ITERATE_PHDR)
5174/* the private <data> we pass below is a dump context initialized like this */
5175struct dl_dump_ctx {
5176 struct buffer *buf;
5177 int with_addr;
5178};
5179
5180static int dl_dump_libs_cb(struct dl_phdr_info *info, size_t size, void *data)
5181{
5182 struct dl_dump_ctx *ctx = data;
5183 const char *fname;
5184 size_t p1, p2, beg, end;
5185 int idx;
5186
5187 if (!info || !info->dlpi_name)
5188 goto leave;
5189
5190 if (!*info->dlpi_name)
5191 fname = get_exec_path();
5192 else if (strchr(info->dlpi_name, '/'))
5193 fname = info->dlpi_name;
5194 else
5195 /* else it's a VDSO or similar and we're not interested */
5196 goto leave;
5197
5198 if (!ctx->with_addr)
5199 goto dump_name;
5200
5201 /* virtual addresses are relative to the load address and are per
5202 * pseudo-header, so we have to scan them all to find the furthest
5203 * one from the beginning. In this case we only dump entries if
5204 * they have at least one section.
5205 */
5206 beg = ~0; end = 0;
5207 for (idx = 0; idx < info->dlpi_phnum; idx++) {
5208 if (!info->dlpi_phdr[idx].p_memsz)
5209 continue;
5210 p1 = info->dlpi_phdr[idx].p_vaddr;
5211 if (p1 < beg)
5212 beg = p1;
5213 p2 = p1 + info->dlpi_phdr[idx].p_memsz - 1;
5214 if (p2 > end)
5215 end = p2;
5216 }
5217
5218 if (!idx)
5219 goto leave;
5220
5221 chunk_appendf(ctx->buf, "0x%012llx-0x%012llx (0x%07llx) ",
5222 (ullong)info->dlpi_addr + beg,
5223 (ullong)info->dlpi_addr + end,
5224 (ullong)(end - beg + 1));
5225 dump_name:
5226 chunk_appendf(ctx->buf, "%s\n", fname);
5227 leave:
5228 return 0;
5229}
5230
5231/* dumps lib names and optionally address ranges */
5232int dump_libs(struct buffer *output, int with_addr)
5233{
5234 struct dl_dump_ctx ctx = { .buf = output, .with_addr = with_addr };
5235 size_t old_data = output->data;
5236
5237 dl_iterate_phdr(dl_dump_libs_cb, &ctx);
5238 return output->data != old_data;
5239}
5240# else // no DL_ITERATE_PHDR
5241# error "No dump_libs() function for this platform"
5242# endif
5243#else // no HA_HAVE_DUMP_LIBS
5244
5245/* unsupported platform: do not dump anything */
5246int dump_libs(struct buffer *output, int with_addr)
5247{
5248 return 0;
5249}
5250
5251#endif // HA_HAVE_DUMP_LIBS
5252
Frédéric Lécaille3b717162019-02-25 15:04:22 +01005253/*
5254 * Allocate an array of unsigned int with <nums> as address from <str> string
Ilya Shipitsin46a030c2020-07-05 16:36:08 +05005255 * made of integer separated by dot characters.
Frédéric Lécaille3b717162019-02-25 15:04:22 +01005256 *
5257 * First, initializes the value with <sz> as address to 0 and initializes the
5258 * array with <nums> as address to NULL. Then allocates the array with <nums> as
5259 * address updating <sz> pointed value to the size of this array.
5260 *
5261 * Returns 1 if succeeded, 0 if not.
5262 */
5263int parse_dotted_uints(const char *str, unsigned int **nums, size_t *sz)
5264{
5265 unsigned int *n;
5266 const char *s, *end;
5267
5268 s = str;
5269 *sz = 0;
5270 end = str + strlen(str);
5271 *nums = n = NULL;
5272
5273 while (1) {
5274 unsigned int r;
5275
5276 if (s >= end)
5277 break;
5278
5279 r = read_uint(&s, end);
5280 /* Expected characters after having read an uint: '\0' or '.',
5281 * if '.', must not be terminal.
5282 */
Christopher Faulet4b524122021-02-11 10:42:41 +01005283 if (*s != '\0'&& (*s++ != '.' || s == end)) {
5284 free(n);
Frédéric Lécaille3b717162019-02-25 15:04:22 +01005285 return 0;
Christopher Faulet4b524122021-02-11 10:42:41 +01005286 }
Frédéric Lécaille3b717162019-02-25 15:04:22 +01005287
Frédéric Lécaille12a71842019-02-26 18:19:48 +01005288 n = my_realloc2(n, (*sz + 1) * sizeof *n);
Frédéric Lécaille3b717162019-02-25 15:04:22 +01005289 if (!n)
5290 return 0;
5291
5292 n[(*sz)++] = r;
5293 }
5294 *nums = n;
5295
5296 return 1;
5297}
5298
Willy Tarreau4d589e72019-08-23 19:02:26 +02005299
5300/* returns the number of bytes needed to encode <v> as a varint. An inline
5301 * version exists for use with constants (__varint_bytes()).
5302 */
5303int varint_bytes(uint64_t v)
5304{
5305 int len = 1;
5306
5307 if (v >= 240) {
5308 v = (v - 240) >> 4;
5309 while (1) {
5310 len++;
5311 if (v < 128)
5312 break;
5313 v = (v - 128) >> 7;
5314 }
5315 }
5316 return len;
5317}
5318
Willy Tarreau52bf8392020-03-08 00:42:37 +01005319
5320/* Random number generator state, see below */
Willy Tarreau1544c142020-03-12 00:31:18 +01005321static uint64_t ha_random_state[2] ALIGNED(2*sizeof(uint64_t));
Willy Tarreau52bf8392020-03-08 00:42:37 +01005322
5323/* This is a thread-safe implementation of xoroshiro128** described below:
5324 * http://prng.di.unimi.it/
5325 * It features a 2^128 long sequence, returns 64 high-quality bits on each call,
5326 * supports fast jumps and passes all common quality tests. It is thread-safe,
5327 * uses a double-cas on 64-bit architectures supporting it, and falls back to a
5328 * local lock on other ones.
5329 */
5330uint64_t ha_random64()
5331{
Willy Tarreau1544c142020-03-12 00:31:18 +01005332 uint64_t old[2] ALIGNED(2*sizeof(uint64_t));
5333 uint64_t new[2] ALIGNED(2*sizeof(uint64_t));
Willy Tarreau52bf8392020-03-08 00:42:37 +01005334
5335#if defined(USE_THREAD) && (!defined(HA_CAS_IS_8B) || !defined(HA_HAVE_CAS_DW))
5336 static HA_SPINLOCK_T rand_lock;
5337
5338 HA_SPIN_LOCK(OTHER_LOCK, &rand_lock);
5339#endif
5340
5341 old[0] = ha_random_state[0];
5342 old[1] = ha_random_state[1];
5343
5344#if defined(USE_THREAD) && defined(HA_CAS_IS_8B) && defined(HA_HAVE_CAS_DW)
5345 do {
5346#endif
Willy Tarreau52bf8392020-03-08 00:42:37 +01005347 new[1] = old[0] ^ old[1];
5348 new[0] = rotl64(old[0], 24) ^ new[1] ^ (new[1] << 16); // a, b
5349 new[1] = rotl64(new[1], 37); // c
5350
5351#if defined(USE_THREAD) && defined(HA_CAS_IS_8B) && defined(HA_HAVE_CAS_DW)
5352 } while (unlikely(!_HA_ATOMIC_DWCAS(ha_random_state, old, new)));
5353#else
5354 ha_random_state[0] = new[0];
5355 ha_random_state[1] = new[1];
5356#if defined(USE_THREAD)
5357 HA_SPIN_UNLOCK(OTHER_LOCK, &rand_lock);
5358#endif
5359#endif
Willy Tarreaub2475a12021-05-09 10:26:14 +02005360 return rotl64(old[0] * 5, 7) * 9;
Willy Tarreau52bf8392020-03-08 00:42:37 +01005361}
5362
5363/* seeds the random state using up to <len> bytes from <seed>, starting with
5364 * the first non-zero byte.
5365 */
5366void ha_random_seed(const unsigned char *seed, size_t len)
5367{
5368 size_t pos;
5369
5370 /* the seed must not be all zeroes, so we pre-fill it with alternating
5371 * bits and overwrite part of them with the block starting at the first
5372 * non-zero byte from the seed.
5373 */
5374 memset(ha_random_state, 0x55, sizeof(ha_random_state));
5375
5376 for (pos = 0; pos < len; pos++)
5377 if (seed[pos] != 0)
5378 break;
5379
5380 if (pos == len)
5381 return;
5382
5383 seed += pos;
5384 len -= pos;
5385
5386 if (len > sizeof(ha_random_state))
5387 len = sizeof(ha_random_state);
5388
5389 memcpy(ha_random_state, seed, len);
5390}
5391
5392/* This causes a jump to (dist * 2^96) places in the pseudo-random sequence,
5393 * and is equivalent to calling ha_random64() as many times. It is used to
5394 * provide non-overlapping sequences of 2^96 numbers (~7*10^28) to up to 2^32
5395 * different generators (i.e. different processes after a fork). The <dist>
5396 * argument is the distance to jump to and is used in a loop so it rather not
5397 * be too large if the processing time is a concern.
5398 *
5399 * BEWARE: this function is NOT thread-safe and must not be called during
5400 * concurrent accesses to ha_random64().
5401 */
5402void ha_random_jump96(uint32_t dist)
5403{
5404 while (dist--) {
5405 uint64_t s0 = 0;
5406 uint64_t s1 = 0;
5407 int b;
5408
5409 for (b = 0; b < 64; b++) {
5410 if ((0xd2a98b26625eee7bULL >> b) & 1) {
5411 s0 ^= ha_random_state[0];
5412 s1 ^= ha_random_state[1];
5413 }
5414 ha_random64();
5415 }
5416
5417 for (b = 0; b < 64; b++) {
5418 if ((0xdddf9b1090aa7ac1ULL >> b) & 1) {
5419 s0 ^= ha_random_state[0];
5420 s1 ^= ha_random_state[1];
5421 }
5422 ha_random64();
5423 }
5424 ha_random_state[0] = s0;
5425 ha_random_state[1] = s1;
5426 }
5427}
5428
Willy Tarreauee3bcdd2020-03-08 17:48:17 +01005429/* Generates an RFC4122 UUID into chunk <output> which must be at least 37
5430 * bytes large.
5431 */
5432void ha_generate_uuid(struct buffer *output)
5433{
5434 uint32_t rnd[4];
5435 uint64_t last;
5436
5437 last = ha_random64();
5438 rnd[0] = last;
5439 rnd[1] = last >> 32;
5440
5441 last = ha_random64();
5442 rnd[2] = last;
5443 rnd[3] = last >> 32;
5444
5445 chunk_printf(output, "%8.8x-%4.4x-%4.4x-%4.4x-%12.12llx",
5446 rnd[0],
5447 rnd[1] & 0xFFFF,
5448 ((rnd[1] >> 16u) & 0xFFF) | 0x4000, // highest 4 bits indicate the uuid version
5449 (rnd[2] & 0x3FFF) | 0x8000, // the highest 2 bits indicate the UUID variant (10),
5450 (long long)((rnd[2] >> 14u) | ((uint64_t) rnd[3] << 18u)) & 0xFFFFFFFFFFFFull);
5451}
5452
5453
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005454/* only used by parse_line() below. It supports writing in place provided that
5455 * <in> is updated to the next location before calling it. In that case, the
5456 * char at <in> may be overwritten.
5457 */
5458#define EMIT_CHAR(x) \
5459 do { \
5460 char __c = (char)(x); \
5461 if ((opts & PARSE_OPT_INPLACE) && out+outpos > in) \
5462 err |= PARSE_ERR_OVERLAP; \
5463 if (outpos >= outmax) \
5464 err |= PARSE_ERR_TOOLARGE; \
5465 if (!err) \
5466 out[outpos] = __c; \
5467 outpos++; \
5468 } while (0)
5469
Ilya Shipitsin46a030c2020-07-05 16:36:08 +05005470/* Parse <in>, copy it into <out> split into isolated words whose pointers
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005471 * are put in <args>. If more than <outlen> bytes have to be emitted, the
5472 * extraneous ones are not emitted but <outlen> is updated so that the caller
5473 * knows how much to realloc. Similarly, <args> are not updated beyond <nbargs>
5474 * but the returned <nbargs> indicates how many were found. All trailing args
Willy Tarreau61dd44b2020-06-25 07:35:42 +02005475 * up to <nbargs> point to the trailing zero, and as long as <nbargs> is > 0,
5476 * it is guaranteed that at least one arg will point to the zero. It is safe
5477 * to call it with a NULL <args> if <nbargs> is 0.
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005478 *
5479 * <out> may overlap with <in> provided that it never goes further, in which
5480 * case the parser will accept to perform in-place parsing and unquoting/
5481 * unescaping but only if environment variables do not lead to expansion that
5482 * causes overlapping, otherwise the input string being destroyed, the error
5483 * will not be recoverable. Note that even during out-of-place <in> will
5484 * experience temporary modifications in-place for variable resolution and must
5485 * be writable, and will also receive zeroes to delimit words when using
5486 * in-place copy. Parsing options <opts> taken from PARSE_OPT_*. Return value
5487 * is zero on success otherwise a bitwise-or of PARSE_ERR_*. Upon error, the
5488 * starting point of the first invalid character sequence or unmatched
5489 * quote/brace is reported in <errptr> if not NULL. When using in-place parsing
5490 * error reporting might be difficult since zeroes will have been inserted into
5491 * the string. One solution for the caller may consist in replacing all args
5492 * delimiters with spaces in this case.
5493 */
Maximilian Mader29c6cd72021-06-06 00:50:21 +02005494uint32_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 +02005495{
5496 char *quote = NULL;
5497 char *brace = NULL;
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005498 char *word_expand = NULL;
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005499 unsigned char hex1, hex2;
5500 size_t outmax = *outlen;
Willy Tarreau61dd44b2020-06-25 07:35:42 +02005501 int argsmax = *nbargs - 1;
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005502 size_t outpos = 0;
5503 int squote = 0;
5504 int dquote = 0;
5505 int arg = 0;
5506 uint32_t err = 0;
5507
5508 *nbargs = 0;
5509 *outlen = 0;
5510
Willy Tarreau61dd44b2020-06-25 07:35:42 +02005511 /* argsmax may be -1 here, protecting args[] from any write */
5512 if (arg < argsmax)
5513 args[arg] = out;
5514
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005515 while (1) {
5516 if (*in >= '-' && *in != '\\') {
5517 /* speedup: directly send all regular chars starting
5518 * with '-', '.', '/', alnum etc...
5519 */
5520 EMIT_CHAR(*in++);
5521 continue;
5522 }
5523 else if (*in == '\0' || *in == '\n' || *in == '\r') {
5524 /* end of line */
5525 break;
5526 }
5527 else if (*in == '#' && (opts & PARSE_OPT_SHARP) && !squote && !dquote) {
5528 /* comment */
5529 break;
5530 }
5531 else if (*in == '"' && !squote && (opts & PARSE_OPT_DQUOTE)) { /* double quote outside single quotes */
5532 if (dquote) {
5533 dquote = 0;
5534 quote = NULL;
5535 }
5536 else {
5537 dquote = 1;
5538 quote = in;
5539 }
5540 in++;
5541 continue;
5542 }
5543 else if (*in == '\'' && !dquote && (opts & PARSE_OPT_SQUOTE)) { /* single quote outside double quotes */
5544 if (squote) {
5545 squote = 0;
5546 quote = NULL;
5547 }
5548 else {
5549 squote = 1;
5550 quote = in;
5551 }
5552 in++;
5553 continue;
5554 }
5555 else if (*in == '\\' && !squote && (opts & PARSE_OPT_BKSLASH)) {
5556 /* first, we'll replace \\, \<space>, \#, \r, \n, \t, \xXX with their
5557 * C equivalent value but only when they have a special meaning and within
5558 * double quotes for some of them. Other combinations left unchanged (eg: \1).
5559 */
5560 char tosend = *in;
5561
5562 switch (in[1]) {
5563 case ' ':
5564 case '\\':
5565 tosend = in[1];
5566 in++;
5567 break;
5568
5569 case 't':
5570 tosend = '\t';
5571 in++;
5572 break;
5573
5574 case 'n':
5575 tosend = '\n';
5576 in++;
5577 break;
5578
5579 case 'r':
5580 tosend = '\r';
5581 in++;
5582 break;
5583
5584 case '#':
5585 /* escaping of "#" only if comments are supported */
5586 if (opts & PARSE_OPT_SHARP)
5587 in++;
5588 tosend = *in;
5589 break;
5590
5591 case '\'':
5592 /* escaping of "'" only outside single quotes and only if single quotes are supported */
5593 if (opts & PARSE_OPT_SQUOTE && !squote)
5594 in++;
5595 tosend = *in;
5596 break;
5597
5598 case '"':
5599 /* escaping of '"' only outside single quotes and only if double quotes are supported */
5600 if (opts & PARSE_OPT_DQUOTE && !squote)
5601 in++;
5602 tosend = *in;
5603 break;
5604
5605 case '$':
5606 /* escaping of '$' only inside double quotes and only if env supported */
5607 if (opts & PARSE_OPT_ENV && dquote)
5608 in++;
5609 tosend = *in;
5610 break;
5611
5612 case 'x':
5613 if (!ishex(in[2]) || !ishex(in[3])) {
5614 /* invalid or incomplete hex sequence */
5615 err |= PARSE_ERR_HEX;
5616 if (errptr)
5617 *errptr = in;
5618 goto leave;
5619 }
Willy Tarreauf278eec2020-07-05 21:46:32 +02005620 hex1 = toupper((unsigned char)in[2]) - '0';
5621 hex2 = toupper((unsigned char)in[3]) - '0';
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005622 if (hex1 > 9) hex1 -= 'A' - '9' - 1;
5623 if (hex2 > 9) hex2 -= 'A' - '9' - 1;
5624 tosend = (hex1 << 4) + hex2;
5625 in += 3;
5626 break;
5627
5628 default:
5629 /* other combinations are not escape sequences */
5630 break;
5631 }
5632
5633 in++;
5634 EMIT_CHAR(tosend);
5635 }
5636 else if (isspace((unsigned char)*in) && !squote && !dquote) {
5637 /* a non-escaped space is an argument separator */
5638 while (isspace((unsigned char)*in))
5639 in++;
5640 EMIT_CHAR(0);
5641 arg++;
5642 if (arg < argsmax)
5643 args[arg] = out + outpos;
5644 else
5645 err |= PARSE_ERR_TOOMANY;
5646 }
5647 else if (*in == '$' && (opts & PARSE_OPT_ENV) && (dquote || !(opts & PARSE_OPT_DQUOTE))) {
5648 /* environment variables are evaluated anywhere, or only
5649 * inside double quotes if they are supported.
5650 */
5651 char *var_name;
5652 char save_char;
Willy Tarreaua46f1af2021-05-06 10:25:11 +02005653 const char *value;
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005654
5655 in++;
5656
5657 if (*in == '{')
5658 brace = in++;
5659
Willy Tarreaua46f1af2021-05-06 10:25:11 +02005660 if (!isalpha((unsigned char)*in) && *in != '_' && *in != '.') {
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005661 /* unacceptable character in variable name */
5662 err |= PARSE_ERR_VARNAME;
5663 if (errptr)
5664 *errptr = in;
5665 goto leave;
5666 }
5667
5668 var_name = in;
Willy Tarreaua46f1af2021-05-06 10:25:11 +02005669 if (*in == '.')
5670 in++;
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005671 while (isalnum((unsigned char)*in) || *in == '_')
5672 in++;
5673
5674 save_char = *in;
5675 *in = '\0';
Willy Tarreaua46f1af2021-05-06 10:25:11 +02005676 if (unlikely(*var_name == '.')) {
5677 /* internal pseudo-variables */
5678 if (strcmp(var_name, ".LINE") == 0)
5679 value = ultoa(global.cfg_curr_line);
5680 else if (strcmp(var_name, ".FILE") == 0)
5681 value = global.cfg_curr_file;
5682 else if (strcmp(var_name, ".SECTION") == 0)
5683 value = global.cfg_curr_section;
5684 else {
5685 /* unsupported internal variable name */
5686 err |= PARSE_ERR_VARNAME;
5687 if (errptr)
5688 *errptr = var_name;
5689 goto leave;
5690 }
5691 } else {
5692 value = getenv(var_name);
5693 }
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005694 *in = save_char;
5695
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005696 /* support for '[*]' sequence to force word expansion,
5697 * only available inside braces */
5698 if (*in == '[' && brace && (opts & PARSE_OPT_WORD_EXPAND)) {
5699 word_expand = in++;
5700
5701 if (*in++ != '*' || *in++ != ']') {
5702 err |= PARSE_ERR_WRONG_EXPAND;
5703 if (errptr)
5704 *errptr = word_expand;
5705 goto leave;
5706 }
5707 }
5708
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005709 if (brace) {
Willy Tarreauec347b12021-11-18 17:42:50 +01005710 if (*in == '-') {
5711 /* default value starts just after the '-' */
5712 if (!value)
5713 value = in + 1;
5714
5715 while (*in && *in != '}')
5716 in++;
5717 if (!*in)
5718 goto no_brace;
5719 *in = 0; // terminate the default value
5720 }
5721 else if (*in != '}') {
5722 no_brace:
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005723 /* unmatched brace */
5724 err |= PARSE_ERR_BRACE;
5725 if (errptr)
5726 *errptr = brace;
5727 goto leave;
5728 }
Willy Tarreauec347b12021-11-18 17:42:50 +01005729
5730 /* brace found, skip it */
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005731 in++;
5732 brace = NULL;
5733 }
5734
5735 if (value) {
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005736 while (*value) {
5737 /* expand as individual parameters on a space character */
Willy Tarreaufe2cc412020-10-01 18:04:40 +02005738 if (word_expand && isspace((unsigned char)*value)) {
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005739 EMIT_CHAR(0);
5740 ++arg;
5741 if (arg < argsmax)
5742 args[arg] = out + outpos;
5743 else
5744 err |= PARSE_ERR_TOOMANY;
5745
5746 /* skip consecutive spaces */
Willy Tarreaufe2cc412020-10-01 18:04:40 +02005747 while (isspace((unsigned char)*++value))
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005748 ;
5749 } else {
5750 EMIT_CHAR(*value++);
5751 }
5752 }
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005753 }
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005754 word_expand = NULL;
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005755 }
5756 else {
5757 /* any other regular char */
5758 EMIT_CHAR(*in++);
5759 }
5760 }
5761
5762 /* end of output string */
5763 EMIT_CHAR(0);
Erwan Le Goas8a6767d2022-09-23 15:06:34 +02005764
Willy Tarreau94ab1392022-10-03 08:27:55 +02005765 /* Don't add an empty arg after trailing spaces. Note that args[arg]
5766 * may contain some distances relative to NULL if <out> was NULL, or
5767 * pointers beyond the end of <out> in case <outlen> is too short, thus
5768 * we must not dereference it.
Erwan Le Goas8a6767d2022-09-23 15:06:34 +02005769 */
Willy Tarreau94ab1392022-10-03 08:27:55 +02005770 if (arg < argsmax && args[arg] != out + outpos - 1)
Erwan Le Goas8a6767d2022-09-23 15:06:34 +02005771 arg++;
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005772
5773 if (quote) {
5774 /* unmatched quote */
5775 err |= PARSE_ERR_QUOTE;
5776 if (errptr)
5777 *errptr = quote;
5778 goto leave;
5779 }
5780 leave:
5781 *nbargs = arg;
5782 *outlen = outpos;
5783
Willy Tarreau61dd44b2020-06-25 07:35:42 +02005784 /* empty all trailing args by making them point to the trailing zero,
5785 * at least the last one in any case.
5786 */
5787 if (arg > argsmax)
5788 arg = argsmax;
5789
5790 while (arg >= 0 && arg <= argsmax)
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005791 args[arg++] = out + outpos - 1;
5792
5793 return err;
5794}
5795#undef EMIT_CHAR
5796
Willy Tarreauc54e5ad2020-06-25 09:15:40 +02005797/* This is used to sanitize an input line that's about to be used for error reporting.
5798 * It will adjust <line> to print approximately <width> chars around <pos>, trying to
5799 * preserve the beginning, with leading or trailing "..." when the line is truncated.
5800 * If non-printable chars are present in the output. It returns the new offset <pos>
5801 * in the modified line. Non-printable characters are replaced with '?'. <width> must
5802 * be at least 6 to support two "..." otherwise the result is undefined. The line
5803 * itself must have at least 7 chars allocated for the same reason.
5804 */
5805size_t sanitize_for_printing(char *line, size_t pos, size_t width)
5806{
5807 size_t shift = 0;
5808 char *out = line;
5809 char *in = line;
5810 char *end = line + width;
5811
5812 if (pos >= width) {
5813 /* if we have to shift, we'll be out of context, so let's
5814 * try to put <pos> at the center of width.
5815 */
5816 shift = pos - width / 2;
5817 in += shift + 3;
5818 end = out + width - 3;
5819 out[0] = out[1] = out[2] = '.';
5820 out += 3;
5821 }
5822
5823 while (out < end && *in) {
5824 if (isspace((unsigned char)*in))
5825 *out++ = ' ';
5826 else if (isprint((unsigned char)*in))
5827 *out++ = *in;
5828 else
5829 *out++ = '?';
5830 in++;
5831 }
5832
5833 if (end < line + width) {
5834 out[0] = out[1] = out[2] = '.';
5835 out += 3;
5836 }
5837
5838 *out++ = 0;
5839 return pos - shift;
5840}
Willy Tarreau06e69b52021-03-02 14:01:35 +01005841
Willy Tarreaue33c4b32021-03-12 18:59:31 +01005842/* Update array <fp> with the fingerprint of word <word> by counting the
Willy Tarreauba2c4452021-03-12 09:01:52 +01005843 * transitions between characters. <fp> is a 1024-entries array indexed as
5844 * 32*from+to. Positions for 'from' and 'to' are:
Willy Tarreau9294e882021-03-15 09:34:27 +01005845 * 1..26=letter, 27=digit, 28=other/begin/end.
5846 * Row "from=0" is used to mark the character's presence. Others unused.
Willy Tarreauba2c4452021-03-12 09:01:52 +01005847 */
Willy Tarreaue33c4b32021-03-12 18:59:31 +01005848void update_word_fingerprint(uint8_t *fp, const char *word)
Willy Tarreauba2c4452021-03-12 09:01:52 +01005849{
5850 const char *p;
5851 int from, to;
5852 int c;
5853
Willy Tarreauba2c4452021-03-12 09:01:52 +01005854 from = 28; // begin
5855 for (p = word; *p; p++) {
5856 c = tolower(*p);
5857 switch(c) {
Willy Tarreau9294e882021-03-15 09:34:27 +01005858 case 'a'...'z': to = c - 'a' + 1; break;
5859 case 'A'...'Z': to = tolower(c) - 'a' + 1; break;
5860 case '0'...'9': to = 27; break;
5861 default: to = 28; break;
Willy Tarreauba2c4452021-03-12 09:01:52 +01005862 }
Willy Tarreau9294e882021-03-15 09:34:27 +01005863 fp[to] = 1;
Willy Tarreauba2c4452021-03-12 09:01:52 +01005864 fp[32 * from + to]++;
5865 from = to;
5866 }
5867 to = 28; // end
5868 fp[32 * from + to]++;
5869}
5870
Erwan Le Goas9c766372022-09-14 17:40:24 +02005871/* This function hashes a word, scramble is the anonymizing key, returns
5872 * the hashed word when the key (scramble) != 0, else returns the word.
5873 * This function can be called NB_L_HASH_WORD times in a row, don't call
5874 * it if you called it more than NB_L_HASH_WORD.
5875 */
5876const char *hash_anon(uint32_t scramble, const char *string2hash, const char *prefix, const char *suffix)
5877{
5878 index_hash++;
Erwan Le Goasd2605cf2022-09-21 16:24:23 +02005879 if (index_hash == NB_L_HASH_WORD)
Erwan Le Goas9c766372022-09-14 17:40:24 +02005880 index_hash = 0;
5881
5882 /* don't hash empty strings */
5883 if (!string2hash[0] || (string2hash[0] == ' ' && string2hash[1] == 0))
5884 return string2hash;
5885
5886 if (scramble != 0) {
5887 snprintf(hash_word[index_hash], sizeof(hash_word[index_hash]), "%s%06x%s",
5888 prefix, HA_ANON(scramble, string2hash, strlen(string2hash)), suffix);
5889 return hash_word[index_hash];
5890 }
5891 else
5892 return string2hash;
5893}
5894
5895/* This function hashes or not an ip address ipstring, scramble is the anonymizing
5896 * key, returns the hashed ip with his port or ipstring when there is nothing to hash.
Erwan Le Goas5eef1582022-09-29 10:25:31 +02005897 * Put hasport equal 0 to point out ipstring has no port, else put an other int.
5898 * Without port, return a simple hash or ipstring.
Erwan Le Goas9c766372022-09-14 17:40:24 +02005899 */
Erwan Le Goas5eef1582022-09-29 10:25:31 +02005900const char *hash_ipanon(uint32_t scramble, char *ipstring, int hasport)
Erwan Le Goas9c766372022-09-14 17:40:24 +02005901{
5902 char *errmsg = NULL;
5903 struct sockaddr_storage *sa;
Erwan Le Goas5eef1582022-09-29 10:25:31 +02005904 struct sockaddr_storage ss;
Erwan Le Goas9c766372022-09-14 17:40:24 +02005905 char addr[46];
5906 int port;
5907
5908 index_hash++;
Erwan Le Goasd2605cf2022-09-21 16:24:23 +02005909 if (index_hash == NB_L_HASH_WORD) {
Erwan Le Goas9c766372022-09-14 17:40:24 +02005910 index_hash = 0;
5911 }
5912
Erwan Le Goas5eef1582022-09-29 10:25:31 +02005913 if (scramble == 0) {
Erwan Le Goas9c766372022-09-14 17:40:24 +02005914 return ipstring;
5915 }
Christopher Faulet015bbc22022-09-29 11:53:07 +02005916 if (strcmp(ipstring, "localhost") == 0 ||
5917 strcmp(ipstring, "stdout") == 0 ||
5918 strcmp(ipstring, "stderr") == 0 ||
5919 strncmp(ipstring, "fd@", 3) == 0 ||
5920 strncmp(ipstring, "sockpair@", 9) == 0) {
Erwan Le Goas5eef1582022-09-29 10:25:31 +02005921 return ipstring;
5922 }
Erwan Le Goas9c766372022-09-14 17:40:24 +02005923 else {
Erwan Le Goas5eef1582022-09-29 10:25:31 +02005924 if (hasport == 0) {
5925 memset(&ss, 0, sizeof(ss));
5926 if (str2ip2(ipstring, &ss, 1) == NULL) {
5927 return HA_ANON_STR(scramble, ipstring);
5928 }
5929 sa = &ss;
Erwan Le Goas9c766372022-09-14 17:40:24 +02005930 }
5931 else {
Erwan Le Goas5eef1582022-09-29 10:25:31 +02005932 sa = str2sa_range(ipstring, NULL, NULL, NULL, NULL, NULL, &errmsg, NULL, NULL,
Christopher Faulet7e50e4b2022-09-29 11:46:34 +02005933 PA_O_PORT_OK | PA_O_STREAM | PA_O_DGRAM | PA_O_XPRT | PA_O_CONNECT |
5934 PA_O_PORT_RANGE | PA_O_PORT_OFS | PA_O_RESOLVE);
Erwan Le Goas5eef1582022-09-29 10:25:31 +02005935 if (sa == NULL) {
5936 return HA_ANON_STR(scramble, ipstring);
5937 }
5938 }
5939 addr_to_str(sa, addr, sizeof(addr));
5940 port = get_host_port(sa);
Erwan Le Goas9c766372022-09-14 17:40:24 +02005941
Erwan Le Goas5eef1582022-09-29 10:25:31 +02005942 switch(sa->ss_family) {
5943 case AF_INET:
5944 if (strncmp(addr, "127", 3) == 0 || strncmp(addr, "255", 3) == 0 || strncmp(addr, "0", 1) == 0) {
5945 return ipstring;
5946 }
5947 else {
5948 if (port != 0) {
5949 snprintf(hash_word[index_hash], sizeof(hash_word[index_hash]), "IPV4(%06x):%d", HA_ANON(scramble, addr, strlen(addr)), port);
5950 return hash_word[index_hash];
Erwan Le Goas9c766372022-09-14 17:40:24 +02005951 }
5952 else {
Erwan Le Goas5eef1582022-09-29 10:25:31 +02005953 snprintf(hash_word[index_hash], sizeof(hash_word[index_hash]), "IPV4(%06x)", HA_ANON(scramble, addr, strlen(addr)));
5954 return hash_word[index_hash];
Erwan Le Goas9c766372022-09-14 17:40:24 +02005955 }
Erwan Le Goas5eef1582022-09-29 10:25:31 +02005956 }
5957 break;
Erwan Le Goas9c766372022-09-14 17:40:24 +02005958
Erwan Le Goas5eef1582022-09-29 10:25:31 +02005959 case AF_INET6:
5960 if (strcmp(addr, "::1") == 0) {
5961 return ipstring;
5962 }
5963 else {
5964 if (port != 0) {
5965 snprintf(hash_word[index_hash], sizeof(hash_word[index_hash]), "IPV6(%06x):%d", HA_ANON(scramble, addr, strlen(addr)), port);
5966 return hash_word[index_hash];
Erwan Le Goas9c766372022-09-14 17:40:24 +02005967 }
5968 else {
Erwan Le Goas5eef1582022-09-29 10:25:31 +02005969 snprintf(hash_word[index_hash], sizeof(hash_word[index_hash]), "IPV6(%06x)", HA_ANON(scramble, addr, strlen(addr)));
5970 return hash_word[index_hash];
Erwan Le Goas9c766372022-09-14 17:40:24 +02005971 }
Erwan Le Goas5eef1582022-09-29 10:25:31 +02005972 }
5973 break;
Erwan Le Goas9c766372022-09-14 17:40:24 +02005974
Erwan Le Goas5eef1582022-09-29 10:25:31 +02005975 case AF_UNIX:
5976 return HA_ANON_STR(scramble, ipstring);
5977 break;
5978
5979 default:
5980 return ipstring;
5981 break;
5982 };
Erwan Le Goas9c766372022-09-14 17:40:24 +02005983 }
5984 return ipstring;
5985}
5986
Willy Tarreaue33c4b32021-03-12 18:59:31 +01005987/* Initialize array <fp> with the fingerprint of word <word> by counting the
5988 * transitions between characters. <fp> is a 1024-entries array indexed as
5989 * 32*from+to. Positions for 'from' and 'to' are:
5990 * 0..25=letter, 26=digit, 27=other, 28=begin, 29=end, others unused.
5991 */
5992void make_word_fingerprint(uint8_t *fp, const char *word)
5993{
5994 memset(fp, 0, 1024);
5995 update_word_fingerprint(fp, word);
5996}
5997
Willy Tarreauba2c4452021-03-12 09:01:52 +01005998/* Return the distance between two word fingerprints created by function
5999 * make_word_fingerprint(). It's a positive integer calculated as the sum of
Willy Tarreau714c4c12021-03-15 09:44:53 +01006000 * the differences between each location.
Willy Tarreauba2c4452021-03-12 09:01:52 +01006001 */
6002int word_fingerprint_distance(const uint8_t *fp1, const uint8_t *fp2)
6003{
6004 int i, k, dist = 0;
6005
6006 for (i = 0; i < 1024; i++) {
6007 k = (int)fp1[i] - (int)fp2[i];
Willy Tarreau714c4c12021-03-15 09:44:53 +01006008 dist += abs(k);
Willy Tarreauba2c4452021-03-12 09:01:52 +01006009 }
6010 return dist;
6011}
6012
William Lallemand3aeb3f92021-08-21 23:59:56 +02006013/*
6014 * This function compares the loaded openssl version with a string <version>
6015 * This function use the same return code as compare_current_version:
6016 *
6017 * -1 : the version in argument is older than the current openssl version
6018 * 0 : the version in argument is the same as the current openssl version
6019 * 1 : the version in argument is newer than the current openssl version
6020 *
6021 * Or some errors:
6022 * -2 : openssl is not available on this process
6023 * -3 : the version in argument is not parsable
6024 */
6025int openssl_compare_current_version(const char *version)
6026{
6027#ifdef USE_OPENSSL
6028 int numversion;
6029
6030 numversion = openssl_version_parser(version);
6031 if (numversion == 0)
6032 return -3;
6033
6034 if (numversion < OPENSSL_VERSION_NUMBER)
6035 return -1;
6036 else if (numversion > OPENSSL_VERSION_NUMBER)
6037 return 1;
6038 else
6039 return 0;
6040#else
6041 return -2;
6042#endif
6043}
6044
Remi Tricot-Le Bretonb01179a2021-10-11 15:34:12 +02006045/*
6046 * This function compares the loaded openssl name with a string <name>
6047 * This function returns 0 if the OpenSSL name starts like the passed parameter,
6048 * 1 otherwise.
6049 */
6050int openssl_compare_current_name(const char *name)
6051{
6052#ifdef USE_OPENSSL
6053 int name_len = 0;
6054 const char *openssl_version = OpenSSL_version(OPENSSL_VERSION);
6055
6056 if (name) {
6057 name_len = strlen(name);
6058 if (strlen(name) <= strlen(openssl_version))
6059 return strncmp(openssl_version, name, name_len);
6060 }
6061#endif
6062 return 1;
6063}
6064
Willy Tarreau40dde2d2022-06-19 16:41:59 +02006065#if defined(RTLD_DEFAULT) || defined(RTLD_NEXT)
6066/* redefine dlopen() so that we can detect unexpected replacement of some
6067 * critical symbols, typically init/alloc/free functions coming from alternate
6068 * libraries. When called, a tainted flag is set (TAINTED_SHARED_LIBS).
6069 */
6070void *dlopen(const char *filename, int flags)
6071{
6072 static void *(*_dlopen)(const char *filename, int flags);
Willy Tarreau177aed52022-06-19 16:49:51 +02006073 struct {
6074 const char *name;
6075 void *curr, *next;
6076 } check_syms[] = {
6077 { .name = "malloc", },
6078 { .name = "free", },
6079 { .name = "SSL_library_init", },
6080 { .name = "X509_free", },
6081 /* insert only above, 0 must be the last one */
6082 { 0 },
6083 };
6084 const char *trace;
6085 void *addr;
Willy Tarreau40dde2d2022-06-19 16:41:59 +02006086 void *ret;
Willy Tarreau177aed52022-06-19 16:49:51 +02006087 int sym = 0;
Willy Tarreau40dde2d2022-06-19 16:41:59 +02006088
6089 if (!_dlopen) {
6090 _dlopen = get_sym_next_addr("dlopen");
6091 if (!_dlopen || _dlopen == dlopen) {
6092 _dlopen = NULL;
6093 return NULL;
6094 }
6095 }
6096
Willy Tarreau177aed52022-06-19 16:49:51 +02006097 /* save a few pointers to critical symbols. We keep a copy of both the
6098 * current and the next value, because we might already have replaced
6099 * some of them (e.g. malloc/free with DEBUG_MEM_STATS), and we're only
6100 * interested in verifying that a loaded library doesn't come with a
6101 * completely different definition that would be incompatible.
6102 */
6103 for (sym = 0; check_syms[sym].name; sym++) {
6104 check_syms[sym].curr = get_sym_curr_addr(check_syms[sym].name);
6105 check_syms[sym].next = get_sym_next_addr(check_syms[sym].name);
6106 }
Willy Tarreau40dde2d2022-06-19 16:41:59 +02006107
6108 /* now open the requested lib */
6109 ret = _dlopen(filename, flags);
6110 if (!ret)
6111 return ret;
6112
6113 mark_tainted(TAINTED_SHARED_LIBS);
6114
Willy Tarreau177aed52022-06-19 16:49:51 +02006115 /* and check that critical symbols didn't change */
6116 for (sym = 0; check_syms[sym].name; sym++) {
6117 if (!check_syms[sym].curr && !check_syms[sym].next)
6118 continue;
6119
6120 addr = dlsym(ret, check_syms[sym].name);
6121 if (!addr || addr == check_syms[sym].curr || addr == check_syms[sym].next)
6122 continue;
6123
6124 /* OK it's clear that this symbol was redefined */
6125 mark_tainted(TAINTED_REDEFINITION);
6126
6127 trace = hlua_show_current_location("\n ");
6128 ha_warning("dlopen(): shared library '%s' brings a different definition of symbol '%s'. The process cannot be trusted anymore!%s%s\n",
6129 filename, check_syms[sym].name,
6130 trace ? " Suspected call location: \n " : "",
6131 trace ? trace : "");
6132 }
6133
Willy Tarreau40dde2d2022-06-19 16:41:59 +02006134 return ret;
6135}
6136#endif
6137
Willy Tarreau06e69b52021-03-02 14:01:35 +01006138static int init_tools_per_thread()
6139{
6140 /* Let's make each thread start from a different position */
6141 statistical_prng_state += tid * MAX_THREADS;
6142 if (!statistical_prng_state)
6143 statistical_prng_state++;
6144 return 1;
6145}
6146REGISTER_PER_THREAD_INIT(init_tools_per_thread);
Willy Tarreauc54e5ad2020-06-25 09:15:40 +02006147
Willy Tarreaubaaee002006-06-26 02:48:02 +02006148/*
6149 * Local variables:
6150 * c-indent-level: 8
6151 * c-basic-offset: 8
6152 * End:
6153 */