blob: 02475a769d20b70da3510b19d527c5b83c0fae1c [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>
Willy Tarreau8f6da642023-03-10 12:04:02 +010064#include <haproxy/quic_sock.h>
Emeric Brunc9437992021-02-12 19:42:55 +010065#include <haproxy/resolvers.h>
Willy Tarreau5edca2f2022-05-27 09:25:10 +020066#include <haproxy/sc_strm.h>
Willy Tarreau586f71b2020-12-11 15:54:36 +010067#include <haproxy/sock.h>
Willy Tarreau209108d2020-06-04 20:30:20 +020068#include <haproxy/ssl_sock.h>
William Lallemand3aeb3f92021-08-21 23:59:56 +020069#include <haproxy/ssl_utils.h>
Willy Tarreaucb086c62022-05-27 09:47:12 +020070#include <haproxy/stconn.h>
Willy Tarreaucea0e1b2020-06-04 17:25:40 +020071#include <haproxy/task.h>
Willy Tarreau48fbcae2020-06-03 18:09:46 +020072#include <haproxy/tools.h>
Willy Tarreau08093cc2022-11-24 08:09:12 +010073#include <haproxy/xxhash.h>
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +010074
Thierry Fournier93127942016-01-20 18:49:45 +010075/* This macro returns false if the test __x is false. Many
76 * of the following parsing function must be abort the processing
77 * if it returns 0, so this macro is useful for writing light code.
78 */
79#define RET0_UNLESS(__x) do { if (!(__x)) return 0; } while (0)
80
Erwan Le Goas9c766372022-09-14 17:40:24 +020081/* Define the number of line of hash_word */
Erwan Le Goas5eef1582022-09-29 10:25:31 +020082#define NB_L_HASH_WORD 15
Erwan Le Goas9c766372022-09-14 17:40:24 +020083
Willy Tarreau08093cc2022-11-24 08:09:12 +010084/* return the hash of a string and length for a given key. All keys are valid. */
85#define HA_ANON(key, str, len) (XXH32(str, len, key) & 0xFFFFFF)
86
Willy Tarreau56adcf22012-12-23 18:00:29 +010087/* enough to store NB_ITOA_STR integers of :
Willy Tarreau72d759c2007-10-25 12:14:10 +020088 * 2^64-1 = 18446744073709551615 or
89 * -2^63 = -9223372036854775808
Willy Tarreaue7239b52009-03-29 13:41:58 +020090 *
91 * The HTML version needs room for adding the 25 characters
92 * '<span class="rls"></span>' around digits at positions 3N+1 in order
93 * to add spacing at up to 6 positions : 18 446 744 073 709 551 615
Willy Tarreau72d759c2007-10-25 12:14:10 +020094 */
Christopher Faulet99bca652017-11-14 16:47:26 +010095THREAD_LOCAL char itoa_str[NB_ITOA_STR][171];
96THREAD_LOCAL int itoa_idx = 0; /* index of next itoa_str to use */
Willy Tarreaubaaee002006-06-26 02:48:02 +020097
Willy Tarreau588297f2014-06-16 15:16:40 +020098/* sometimes we'll need to quote strings (eg: in stats), and we don't expect
99 * to quote strings larger than a max configuration line.
100 */
Christopher Faulet99bca652017-11-14 16:47:26 +0100101THREAD_LOCAL char quoted_str[NB_QSTR][QSTR_SIZE + 1];
102THREAD_LOCAL int quoted_idx = 0;
Willy Tarreau588297f2014-06-16 15:16:40 +0200103
Willy Tarreau06e69b52021-03-02 14:01:35 +0100104/* thread-local PRNG state. It's modified to start from a different sequence
105 * on all threads upon startup. It must not be used or anything beyond getting
106 * statistical values as it's 100% predictable.
107 */
108THREAD_LOCAL unsigned int statistical_prng_state = 2463534242U;
109
Willy Tarreau5b3cd952022-07-18 13:58:17 +0200110/* set to true if this is a static build */
111int build_is_static = 0;
112
Erwan Le Goas9c766372022-09-14 17:40:24 +0200113/* A global static table to store hashed words */
114static THREAD_LOCAL char hash_word[NB_L_HASH_WORD][20];
115static THREAD_LOCAL int index_hash = 0;
116
Willy Tarreaubaaee002006-06-26 02:48:02 +0200117/*
William Lallemande7340ec2012-01-24 11:15:39 +0100118 * unsigned long long ASCII representation
119 *
120 * return the last char '\0' or NULL if no enough
121 * space in dst
122 */
123char *ulltoa(unsigned long long n, char *dst, size_t size)
124{
125 int i = 0;
126 char *res;
127
128 switch(n) {
129 case 1ULL ... 9ULL:
130 i = 0;
131 break;
132
133 case 10ULL ... 99ULL:
134 i = 1;
135 break;
136
137 case 100ULL ... 999ULL:
138 i = 2;
139 break;
140
141 case 1000ULL ... 9999ULL:
142 i = 3;
143 break;
144
145 case 10000ULL ... 99999ULL:
146 i = 4;
147 break;
148
149 case 100000ULL ... 999999ULL:
150 i = 5;
151 break;
152
153 case 1000000ULL ... 9999999ULL:
154 i = 6;
155 break;
156
157 case 10000000ULL ... 99999999ULL:
158 i = 7;
159 break;
160
161 case 100000000ULL ... 999999999ULL:
162 i = 8;
163 break;
164
165 case 1000000000ULL ... 9999999999ULL:
166 i = 9;
167 break;
168
169 case 10000000000ULL ... 99999999999ULL:
170 i = 10;
171 break;
172
173 case 100000000000ULL ... 999999999999ULL:
174 i = 11;
175 break;
176
177 case 1000000000000ULL ... 9999999999999ULL:
178 i = 12;
179 break;
180
181 case 10000000000000ULL ... 99999999999999ULL:
182 i = 13;
183 break;
184
185 case 100000000000000ULL ... 999999999999999ULL:
186 i = 14;
187 break;
188
189 case 1000000000000000ULL ... 9999999999999999ULL:
190 i = 15;
191 break;
192
193 case 10000000000000000ULL ... 99999999999999999ULL:
194 i = 16;
195 break;
196
197 case 100000000000000000ULL ... 999999999999999999ULL:
198 i = 17;
199 break;
200
201 case 1000000000000000000ULL ... 9999999999999999999ULL:
202 i = 18;
203 break;
204
205 case 10000000000000000000ULL ... ULLONG_MAX:
206 i = 19;
207 break;
208 }
209 if (i + 2 > size) // (i + 1) + '\0'
210 return NULL; // too long
211 res = dst + i + 1;
212 *res = '\0';
213 for (; i >= 0; i--) {
214 dst[i] = n % 10ULL + '0';
215 n /= 10ULL;
216 }
217 return res;
218}
219
220/*
221 * unsigned long ASCII representation
222 *
223 * return the last char '\0' or NULL if no enough
224 * space in dst
225 */
226char *ultoa_o(unsigned long n, char *dst, size_t size)
227{
228 int i = 0;
229 char *res;
230
231 switch (n) {
232 case 0U ... 9UL:
233 i = 0;
234 break;
235
236 case 10U ... 99UL:
237 i = 1;
238 break;
239
240 case 100U ... 999UL:
241 i = 2;
242 break;
243
244 case 1000U ... 9999UL:
245 i = 3;
246 break;
247
248 case 10000U ... 99999UL:
249 i = 4;
250 break;
251
252 case 100000U ... 999999UL:
253 i = 5;
254 break;
255
256 case 1000000U ... 9999999UL:
257 i = 6;
258 break;
259
260 case 10000000U ... 99999999UL:
261 i = 7;
262 break;
263
264 case 100000000U ... 999999999UL:
265 i = 8;
266 break;
267#if __WORDSIZE == 32
268
269 case 1000000000ULL ... ULONG_MAX:
270 i = 9;
271 break;
272
273#elif __WORDSIZE == 64
274
275 case 1000000000ULL ... 9999999999UL:
276 i = 9;
277 break;
278
279 case 10000000000ULL ... 99999999999UL:
280 i = 10;
281 break;
282
283 case 100000000000ULL ... 999999999999UL:
284 i = 11;
285 break;
286
287 case 1000000000000ULL ... 9999999999999UL:
288 i = 12;
289 break;
290
291 case 10000000000000ULL ... 99999999999999UL:
292 i = 13;
293 break;
294
295 case 100000000000000ULL ... 999999999999999UL:
296 i = 14;
297 break;
298
299 case 1000000000000000ULL ... 9999999999999999UL:
300 i = 15;
301 break;
302
303 case 10000000000000000ULL ... 99999999999999999UL:
304 i = 16;
305 break;
306
307 case 100000000000000000ULL ... 999999999999999999UL:
308 i = 17;
309 break;
310
311 case 1000000000000000000ULL ... 9999999999999999999UL:
312 i = 18;
313 break;
314
315 case 10000000000000000000ULL ... ULONG_MAX:
316 i = 19;
317 break;
318
319#endif
320 }
321 if (i + 2 > size) // (i + 1) + '\0'
322 return NULL; // too long
323 res = dst + i + 1;
324 *res = '\0';
325 for (; i >= 0; i--) {
326 dst[i] = n % 10U + '0';
327 n /= 10U;
328 }
329 return res;
330}
331
332/*
333 * signed long ASCII representation
334 *
335 * return the last char '\0' or NULL if no enough
336 * space in dst
337 */
338char *ltoa_o(long int n, char *dst, size_t size)
339{
340 char *pos = dst;
341
342 if (n < 0) {
343 if (size < 3)
344 return NULL; // min size is '-' + digit + '\0' but another test in ultoa
345 *pos = '-';
346 pos++;
347 dst = ultoa_o(-n, pos, size - 1);
348 } else {
349 dst = ultoa_o(n, dst, size);
350 }
351 return dst;
352}
353
354/*
355 * signed long long ASCII representation
356 *
357 * return the last char '\0' or NULL if no enough
358 * space in dst
359 */
360char *lltoa(long long n, char *dst, size_t size)
361{
362 char *pos = dst;
363
364 if (n < 0) {
365 if (size < 3)
366 return NULL; // min size is '-' + digit + '\0' but another test in ulltoa
367 *pos = '-';
368 pos++;
369 dst = ulltoa(-n, pos, size - 1);
370 } else {
371 dst = ulltoa(n, dst, size);
372 }
373 return dst;
374}
375
376/*
377 * write a ascii representation of a unsigned into dst,
378 * return a pointer to the last character
379 * Pad the ascii representation with '0', using size.
380 */
381char *utoa_pad(unsigned int n, char *dst, size_t size)
382{
383 int i = 0;
384 char *ret;
385
386 switch(n) {
387 case 0U ... 9U:
388 i = 0;
389 break;
390
391 case 10U ... 99U:
392 i = 1;
393 break;
394
395 case 100U ... 999U:
396 i = 2;
397 break;
398
399 case 1000U ... 9999U:
400 i = 3;
401 break;
402
403 case 10000U ... 99999U:
404 i = 4;
405 break;
406
407 case 100000U ... 999999U:
408 i = 5;
409 break;
410
411 case 1000000U ... 9999999U:
412 i = 6;
413 break;
414
415 case 10000000U ... 99999999U:
416 i = 7;
417 break;
418
419 case 100000000U ... 999999999U:
420 i = 8;
421 break;
422
423 case 1000000000U ... 4294967295U:
424 i = 9;
425 break;
426 }
427 if (i + 2 > size) // (i + 1) + '\0'
428 return NULL; // too long
Aurelien DARRAGONe3177af2022-11-22 11:42:07 +0100429 i = size - 2; // padding - '\0'
William Lallemande7340ec2012-01-24 11:15:39 +0100430
431 ret = dst + i + 1;
432 *ret = '\0';
433 for (; i >= 0; i--) {
434 dst[i] = n % 10U + '0';
435 n /= 10U;
436 }
437 return ret;
438}
439
440/*
Willy Tarreaubaaee002006-06-26 02:48:02 +0200441 * copies at most <size-1> chars from <src> to <dst>. Last char is always
442 * set to 0, unless <size> is 0. The number of chars copied is returned
443 * (excluding the terminating zero).
444 * This code has been optimized for size and speed : on x86, it's 45 bytes
445 * long, uses only registers, and consumes only 4 cycles per char.
446 */
447int strlcpy2(char *dst, const char *src, int size)
448{
449 char *orig = dst;
450 if (size) {
451 while (--size && (*dst = *src)) {
452 src++; dst++;
453 }
454 *dst = 0;
455 }
456 return dst - orig;
457}
458
459/*
Willy Tarreau72d759c2007-10-25 12:14:10 +0200460 * This function simply returns a locally allocated string containing
Willy Tarreaubaaee002006-06-26 02:48:02 +0200461 * the ascii representation for number 'n' in decimal.
462 */
Emeric Brun3a7fce52010-01-04 14:54:38 +0100463char *ultoa_r(unsigned long n, char *buffer, int size)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200464{
465 char *pos;
466
Willy Tarreau72d759c2007-10-25 12:14:10 +0200467 pos = buffer + size - 1;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200468 *pos-- = '\0';
469
470 do {
471 *pos-- = '0' + n % 10;
472 n /= 10;
Willy Tarreau72d759c2007-10-25 12:14:10 +0200473 } while (n && pos >= buffer);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200474 return pos + 1;
475}
476
Willy Tarreau91092e52007-10-25 16:58:42 +0200477/*
Willy Tarreaue7239b52009-03-29 13:41:58 +0200478 * This function simply returns a locally allocated string containing
Thierry FOURNIER763a5d82015-07-06 23:09:52 +0200479 * the ascii representation for number 'n' in decimal.
480 */
481char *lltoa_r(long long int in, char *buffer, int size)
482{
483 char *pos;
484 int neg = 0;
485 unsigned long long int n;
486
487 pos = buffer + size - 1;
488 *pos-- = '\0';
489
490 if (in < 0) {
491 neg = 1;
492 n = -in;
493 }
494 else
495 n = in;
496
497 do {
498 *pos-- = '0' + n % 10;
499 n /= 10;
500 } while (n && pos >= buffer);
501 if (neg && pos > buffer)
502 *pos-- = '-';
503 return pos + 1;
504}
505
506/*
507 * This function simply returns a locally allocated string containing
Thierry FOURNIER1480bd82015-06-06 19:14:59 +0200508 * the ascii representation for signed number 'n' in decimal.
509 */
510char *sltoa_r(long n, char *buffer, int size)
511{
512 char *pos;
513
514 if (n >= 0)
515 return ultoa_r(n, buffer, size);
516
517 pos = ultoa_r(-n, buffer + 1, size - 1) - 1;
518 *pos = '-';
519 return pos;
520}
521
522/*
523 * This function simply returns a locally allocated string containing
Willy Tarreaue7239b52009-03-29 13:41:58 +0200524 * the ascii representation for number 'n' in decimal, formatted for
525 * HTML output with tags to create visual grouping by 3 digits. The
526 * output needs to support at least 171 characters.
527 */
528const char *ulltoh_r(unsigned long long n, char *buffer, int size)
529{
530 char *start;
531 int digit = 0;
532
533 start = buffer + size;
534 *--start = '\0';
535
536 do {
537 if (digit == 3 && start >= buffer + 7)
538 memcpy(start -= 7, "</span>", 7);
539
540 if (start >= buffer + 1) {
541 *--start = '0' + n % 10;
542 n /= 10;
543 }
544
545 if (digit == 3 && start >= buffer + 18)
546 memcpy(start -= 18, "<span class=\"rls\">", 18);
547
548 if (digit++ == 3)
549 digit = 1;
550 } while (n && start > buffer);
551 return start;
552}
553
554/*
Willy Tarreau91092e52007-10-25 16:58:42 +0200555 * This function simply returns a locally allocated string containing the ascii
556 * representation for number 'n' in decimal, unless n is 0 in which case it
557 * returns the alternate string (or an empty string if the alternate string is
558 * NULL). It use is intended for limits reported in reports, where it's
559 * desirable not to display anything if there is no limit. Warning! it shares
560 * the same vector as ultoa_r().
561 */
562const char *limit_r(unsigned long n, char *buffer, int size, const char *alt)
563{
564 return (n) ? ultoa_r(n, buffer, size) : (alt ? alt : "");
565}
566
Willy Tarreau56d1d8d2021-05-08 10:28:53 +0200567/* Trims the first "%f" float in a string to its minimum number of digits after
568 * the decimal point by trimming trailing zeroes, even dropping the decimal
569 * point if not needed. The string is in <buffer> of length <len>, and the
570 * number is expected to start at or after position <num_start> (the first
571 * point appearing there is considered). A NUL character is always placed at
572 * the end if some trimming occurs. The new buffer length is returned.
573 */
574size_t flt_trim(char *buffer, size_t num_start, size_t len)
575{
576 char *end = buffer + len;
577 char *p = buffer + num_start;
578 char *trim;
579
580 do {
581 if (p >= end)
582 return len;
583 trim = p++;
584 } while (*trim != '.');
585
586 /* For now <trim> is on the decimal point. Let's look for any other
587 * meaningful digit after it.
588 */
589 while (p < end) {
590 if (*p++ != '0')
591 trim = p;
592 }
593
594 if (trim < end)
595 *trim = 0;
596
597 return trim - buffer;
598}
599
Willy Tarreauae03d262021-05-08 07:35:00 +0200600/*
601 * This function simply returns a locally allocated string containing
602 * the ascii representation for number 'n' in decimal with useless trailing
603 * zeroes trimmed.
604 */
605char *ftoa_r(double n, char *buffer, int size)
606{
607 flt_trim(buffer, 0, snprintf(buffer, size, "%f", n));
608 return buffer;
609}
610
Willy Tarreau588297f2014-06-16 15:16:40 +0200611/* returns a locally allocated string containing the quoted encoding of the
612 * input string. The output may be truncated to QSTR_SIZE chars, but it is
613 * guaranteed that the string will always be properly terminated. Quotes are
614 * encoded by doubling them as is commonly done in CSV files. QSTR_SIZE must
615 * always be at least 4 chars.
616 */
617const char *qstr(const char *str)
618{
619 char *ret = quoted_str[quoted_idx];
620 char *p, *end;
621
622 if (++quoted_idx >= NB_QSTR)
623 quoted_idx = 0;
624
625 p = ret;
626 end = ret + QSTR_SIZE;
627
628 *p++ = '"';
629
630 /* always keep 3 chars to support passing "" and the ending " */
631 while (*str && p < end - 3) {
632 if (*str == '"') {
633 *p++ = '"';
634 *p++ = '"';
635 }
636 else
637 *p++ = *str;
638 str++;
639 }
640 *p++ = '"';
641 return ret;
642}
643
Robert Tsai81ae1952007-12-05 10:47:29 +0100644/*
Willy Tarreaubaaee002006-06-26 02:48:02 +0200645 * Returns non-zero if character <s> is a hex digit (0-9, a-f, A-F), else zero.
646 *
647 * It looks like this one would be a good candidate for inlining, but this is
648 * not interesting because it around 35 bytes long and often called multiple
649 * times within the same function.
650 */
651int ishex(char s)
652{
653 s -= '0';
654 if ((unsigned char)s <= 9)
655 return 1;
656 s -= 'A' - '0';
657 if ((unsigned char)s <= 5)
658 return 1;
659 s -= 'a' - 'A';
660 if ((unsigned char)s <= 5)
661 return 1;
662 return 0;
663}
664
Willy Tarreau3ca1a882015-01-15 18:43:49 +0100665/* rounds <i> down to the closest value having max 2 digits */
666unsigned int round_2dig(unsigned int i)
667{
668 unsigned int mul = 1;
669
670 while (i >= 100) {
671 i /= 10;
672 mul *= 10;
673 }
674 return i * mul;
675}
676
Willy Tarreau2e74c3f2007-12-02 18:45:09 +0100677/*
678 * Checks <name> for invalid characters. Valid chars are [A-Za-z0-9_:.-]. If an
679 * invalid character is found, a pointer to it is returned. If everything is
680 * fine, NULL is returned.
681 */
682const char *invalid_char(const char *name)
683{
684 if (!*name)
685 return name;
686
687 while (*name) {
Willy Tarreau90807112020-02-25 08:16:33 +0100688 if (!isalnum((unsigned char)*name) && *name != '.' && *name != ':' &&
Willy Tarreau2e74c3f2007-12-02 18:45:09 +0100689 *name != '_' && *name != '-')
690 return name;
691 name++;
692 }
693 return NULL;
694}
Willy Tarreaubaaee002006-06-26 02:48:02 +0200695
696/*
Frédéric Lécailleb82f7422017-04-13 18:24:23 +0200697 * Checks <name> for invalid characters. Valid chars are [_.-] and those
698 * accepted by <f> function.
Krzysztof Piotr Oledzkiefe3b6f2008-05-23 23:49:32 +0200699 * If an invalid character is found, a pointer to it is returned.
700 * If everything is fine, NULL is returned.
701 */
Frédéric Lécailleb82f7422017-04-13 18:24:23 +0200702static inline const char *__invalid_char(const char *name, int (*f)(int)) {
Krzysztof Piotr Oledzkiefe3b6f2008-05-23 23:49:32 +0200703
704 if (!*name)
705 return name;
706
707 while (*name) {
Willy Tarreau90807112020-02-25 08:16:33 +0100708 if (!f((unsigned char)*name) && *name != '.' &&
Krzysztof Piotr Oledzkiefe3b6f2008-05-23 23:49:32 +0200709 *name != '_' && *name != '-')
710 return name;
711
712 name++;
713 }
714
715 return NULL;
716}
717
718/*
Frédéric Lécailleb82f7422017-04-13 18:24:23 +0200719 * Checks <name> for invalid characters. Valid chars are [A-Za-z0-9_.-].
720 * If an invalid character is found, a pointer to it is returned.
721 * If everything is fine, NULL is returned.
722 */
723const char *invalid_domainchar(const char *name) {
724 return __invalid_char(name, isalnum);
725}
726
727/*
728 * Checks <name> for invalid characters. Valid chars are [A-Za-z_.-].
729 * If an invalid character is found, a pointer to it is returned.
730 * If everything is fine, NULL is returned.
731 */
732const char *invalid_prefix_char(const char *name) {
Thierry Fournierf7b7c3e2018-03-26 11:54:39 +0200733 return __invalid_char(name, isalnum);
Frédéric Lécailleb82f7422017-04-13 18:24:23 +0200734}
735
736/*
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100737 * converts <str> to a struct sockaddr_storage* provided by the caller. The
Willy Tarreau24709282013-03-10 21:32:12 +0100738 * caller must have zeroed <sa> first, and may have set sa->ss_family to force
739 * parse a specific address format. If the ss_family is 0 or AF_UNSPEC, then
740 * the function tries to guess the address family from the syntax. If the
741 * family is forced and the format doesn't match, an error is returned. The
Willy Tarreaufab5a432011-03-04 15:31:53 +0100742 * string is assumed to contain only an address, no port. The address can be a
743 * dotted IPv4 address, an IPv6 address, a host name, or empty or "*" to
744 * indicate INADDR_ANY. NULL is returned if the host part cannot be resolved.
745 * The return address will only have the address family and the address set,
746 * all other fields remain zero. The string is not supposed to be modified.
Thierry FOURNIER58639a02014-11-25 12:02:25 +0100747 * The IPv6 '::' address is IN6ADDR_ANY. If <resolve> is non-zero, the hostname
748 * is resolved, otherwise only IP addresses are resolved, and anything else
Willy Tarreauecde7df2016-11-02 22:37:03 +0100749 * returns NULL. If the address contains a port, this one is preserved.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200750 */
Thierry FOURNIER58639a02014-11-25 12:02:25 +0100751struct sockaddr_storage *str2ip2(const char *str, struct sockaddr_storage *sa, int resolve)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200752{
Willy Tarreaufab5a432011-03-04 15:31:53 +0100753 struct hostent *he;
mildisff5d5102015-10-26 18:50:08 +0100754 /* max IPv6 length, including brackets and terminating NULL */
755 char tmpip[48];
Willy Tarreauecde7df2016-11-02 22:37:03 +0100756 int port = get_host_port(sa);
mildisff5d5102015-10-26 18:50:08 +0100757
758 /* check IPv6 with square brackets */
759 if (str[0] == '[') {
760 size_t iplength = strlen(str);
761
762 if (iplength < 4) {
763 /* minimal size is 4 when using brackets "[::]" */
764 goto fail;
765 }
766 else if (iplength >= sizeof(tmpip)) {
767 /* IPv6 literal can not be larger than tmpip */
768 goto fail;
769 }
770 else {
771 if (str[iplength - 1] != ']') {
772 /* if address started with bracket, it should end with bracket */
773 goto fail;
774 }
775 else {
776 memcpy(tmpip, str + 1, iplength - 2);
777 tmpip[iplength - 2] = '\0';
778 str = tmpip;
779 }
780 }
781 }
Willy Tarreaufab5a432011-03-04 15:31:53 +0100782
Willy Tarreaufab5a432011-03-04 15:31:53 +0100783 /* Any IPv6 address */
784 if (str[0] == ':' && str[1] == ':' && !str[2]) {
Willy Tarreau24709282013-03-10 21:32:12 +0100785 if (!sa->ss_family || sa->ss_family == AF_UNSPEC)
786 sa->ss_family = AF_INET6;
787 else if (sa->ss_family != AF_INET6)
788 goto fail;
Willy Tarreauecde7df2016-11-02 22:37:03 +0100789 set_host_port(sa, port);
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100790 return sa;
Willy Tarreaufab5a432011-03-04 15:31:53 +0100791 }
792
Willy Tarreau24709282013-03-10 21:32:12 +0100793 /* Any address for the family, defaults to IPv4 */
Willy Tarreaufab5a432011-03-04 15:31:53 +0100794 if (!str[0] || (str[0] == '*' && !str[1])) {
Willy Tarreau24709282013-03-10 21:32:12 +0100795 if (!sa->ss_family || sa->ss_family == AF_UNSPEC)
796 sa->ss_family = AF_INET;
Willy Tarreauecde7df2016-11-02 22:37:03 +0100797 set_host_port(sa, port);
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100798 return sa;
Willy Tarreaufab5a432011-03-04 15:31:53 +0100799 }
800
801 /* check for IPv6 first */
Willy Tarreau24709282013-03-10 21:32:12 +0100802 if ((!sa->ss_family || sa->ss_family == AF_UNSPEC || sa->ss_family == AF_INET6) &&
803 inet_pton(AF_INET6, str, &((struct sockaddr_in6 *)sa)->sin6_addr)) {
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100804 sa->ss_family = AF_INET6;
Willy Tarreauecde7df2016-11-02 22:37:03 +0100805 set_host_port(sa, port);
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100806 return sa;
Willy Tarreaufab5a432011-03-04 15:31:53 +0100807 }
808
809 /* then check for IPv4 */
Willy Tarreau24709282013-03-10 21:32:12 +0100810 if ((!sa->ss_family || sa->ss_family == AF_UNSPEC || sa->ss_family == AF_INET) &&
811 inet_pton(AF_INET, str, &((struct sockaddr_in *)sa)->sin_addr)) {
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100812 sa->ss_family = AF_INET;
Willy Tarreauecde7df2016-11-02 22:37:03 +0100813 set_host_port(sa, port);
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100814 return sa;
Willy Tarreaufab5a432011-03-04 15:31:53 +0100815 }
816
Thierry FOURNIER58639a02014-11-25 12:02:25 +0100817 if (!resolve)
818 return NULL;
819
Emeric Brund30e9a12020-12-23 18:49:16 +0100820 if (!resolv_hostname_validation(str, NULL))
Baptiste Assmanna68ca962015-04-14 01:15:08 +0200821 return NULL;
822
David du Colombierd5f43282011-03-17 10:40:16 +0100823#ifdef USE_GETADDRINFO
Nenad Merdanovic88afe032014-04-14 15:56:58 +0200824 if (global.tune.options & GTUNE_USE_GAI) {
David du Colombierd5f43282011-03-17 10:40:16 +0100825 struct addrinfo hints, *result;
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100826 int success = 0;
David du Colombierd5f43282011-03-17 10:40:16 +0100827
828 memset(&result, 0, sizeof(result));
829 memset(&hints, 0, sizeof(hints));
Willy Tarreau24709282013-03-10 21:32:12 +0100830 hints.ai_family = sa->ss_family ? sa->ss_family : AF_UNSPEC;
David du Colombierd5f43282011-03-17 10:40:16 +0100831 hints.ai_socktype = SOCK_DGRAM;
Dmitry Sivachenkoeab7f392015-10-02 01:01:58 +0200832 hints.ai_flags = 0;
David du Colombierd5f43282011-03-17 10:40:16 +0100833 hints.ai_protocol = 0;
834
835 if (getaddrinfo(str, NULL, &hints, &result) == 0) {
Willy Tarreau24709282013-03-10 21:32:12 +0100836 if (!sa->ss_family || sa->ss_family == AF_UNSPEC)
837 sa->ss_family = result->ai_family;
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100838 else if (sa->ss_family != result->ai_family) {
839 freeaddrinfo(result);
Willy Tarreau24709282013-03-10 21:32:12 +0100840 goto fail;
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100841 }
Willy Tarreau24709282013-03-10 21:32:12 +0100842
David du Colombierd5f43282011-03-17 10:40:16 +0100843 switch (result->ai_family) {
844 case AF_INET:
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100845 memcpy((struct sockaddr_in *)sa, result->ai_addr, result->ai_addrlen);
Willy Tarreauecde7df2016-11-02 22:37:03 +0100846 set_host_port(sa, port);
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100847 success = 1;
848 break;
David du Colombierd5f43282011-03-17 10:40:16 +0100849 case AF_INET6:
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100850 memcpy((struct sockaddr_in6 *)sa, result->ai_addr, result->ai_addrlen);
Willy Tarreauecde7df2016-11-02 22:37:03 +0100851 set_host_port(sa, port);
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100852 success = 1;
853 break;
David du Colombierd5f43282011-03-17 10:40:16 +0100854 }
855 }
856
Sean Carey58ea0392013-02-15 23:39:18 +0100857 if (result)
858 freeaddrinfo(result);
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100859
860 if (success)
861 return sa;
Willy Tarreaufab5a432011-03-04 15:31:53 +0100862 }
David du Colombierd5f43282011-03-17 10:40:16 +0100863#endif
Nenad Merdanovic88afe032014-04-14 15:56:58 +0200864 /* try to resolve an IPv4/IPv6 hostname */
865 he = gethostbyname(str);
866 if (he) {
867 if (!sa->ss_family || sa->ss_family == AF_UNSPEC)
868 sa->ss_family = he->h_addrtype;
869 else if (sa->ss_family != he->h_addrtype)
870 goto fail;
871
872 switch (sa->ss_family) {
873 case AF_INET:
874 ((struct sockaddr_in *)sa)->sin_addr = *(struct in_addr *) *(he->h_addr_list);
Willy Tarreauecde7df2016-11-02 22:37:03 +0100875 set_host_port(sa, port);
Nenad Merdanovic88afe032014-04-14 15:56:58 +0200876 return sa;
877 case AF_INET6:
878 ((struct sockaddr_in6 *)sa)->sin6_addr = *(struct in6_addr *) *(he->h_addr_list);
Willy Tarreauecde7df2016-11-02 22:37:03 +0100879 set_host_port(sa, port);
Nenad Merdanovic88afe032014-04-14 15:56:58 +0200880 return sa;
881 }
882 }
883
David du Colombierd5f43282011-03-17 10:40:16 +0100884 /* unsupported address family */
Willy Tarreau24709282013-03-10 21:32:12 +0100885 fail:
Willy Tarreaufab5a432011-03-04 15:31:53 +0100886 return NULL;
887}
888
889/*
Willy Tarreaud4448bc2013-02-20 15:55:15 +0100890 * Converts <str> to a locally allocated struct sockaddr_storage *, and a port
891 * range or offset consisting in two integers that the caller will have to
892 * check to find the relevant input format. The following format are supported :
893 *
894 * String format | address | port | low | high
895 * addr | <addr> | 0 | 0 | 0
896 * addr: | <addr> | 0 | 0 | 0
897 * addr:port | <addr> | <port> | <port> | <port>
898 * addr:pl-ph | <addr> | <pl> | <pl> | <ph>
899 * addr:+port | <addr> | <port> | 0 | <port>
900 * addr:-port | <addr> |-<port> | <port> | 0
901 *
902 * The detection of a port range or increment by the caller is made by
903 * comparing <low> and <high>. If both are equal, then port 0 means no port
904 * was specified. The caller may pass NULL for <low> and <high> if it is not
905 * interested in retrieving port ranges.
906 *
907 * Note that <addr> above may also be :
908 * - empty ("") => family will be AF_INET and address will be INADDR_ANY
909 * - "*" => family will be AF_INET and address will be INADDR_ANY
910 * - "::" => family will be AF_INET6 and address will be IN6ADDR_ANY
911 * - a host name => family and address will depend on host name resolving.
912 *
Willy Tarreau24709282013-03-10 21:32:12 +0100913 * A prefix may be passed in before the address above to force the family :
914 * - "ipv4@" => force address to resolve as IPv4 and fail if not possible.
915 * - "ipv6@" => force address to resolve as IPv6 and fail if not possible.
916 * - "unix@" => force address to be a path to a UNIX socket even if the
917 * path does not start with a '/'
Willy Tarreauccfccef2014-05-10 01:49:15 +0200918 * - 'abns@' -> force address to belong to the abstract namespace (Linux
919 * only). These sockets are just like Unix sockets but without
920 * the need for an underlying file system. The address is a
921 * string. Technically it's like a Unix socket with a zero in
922 * the first byte of the address.
Willy Tarreau40aa0702013-03-10 23:51:38 +0100923 * - "fd@" => an integer must follow, and is a file descriptor number.
Willy Tarreau24709282013-03-10 21:32:12 +0100924 *
mildisff5d5102015-10-26 18:50:08 +0100925 * IPv6 addresses can be declared with or without square brackets. When using
926 * square brackets for IPv6 addresses, the port separator (colon) is optional.
927 * If not using square brackets, and in order to avoid any ambiguity with
928 * IPv6 addresses, the last colon ':' is mandatory even when no port is specified.
929 * NULL is returned if the address cannot be parsed. The <low> and <high> ports
930 * are always initialized if non-null, even for non-IP families.
Willy Tarreaud393a622013-03-04 18:22:00 +0100931 *
932 * If <pfx> is non-null, it is used as a string prefix before any path-based
933 * address (typically the path to a unix socket).
Willy Tarreau40aa0702013-03-10 23:51:38 +0100934 *
Willy Tarreau72b8c1f2015-09-08 15:50:19 +0200935 * if <fqdn> is non-null, it will be filled with :
936 * - a pointer to the FQDN of the server name to resolve if there's one, and
937 * that the caller will have to free(),
938 * - NULL if there was an explicit address that doesn't require resolution.
939 *
Willy Tarreaucd3a55912020-09-04 15:30:46 +0200940 * Hostnames are only resolved if <opts> has PA_O_RESOLVE. Otherwise <fqdn> is
941 * still honored so it is possible for the caller to know whether a resolution
942 * failed by clearing this flag and checking if <fqdn> was filled, indicating
943 * the need for a resolution.
Thierry FOURNIER7fe3be72015-09-26 20:03:36 +0200944 *
Willy Tarreau40aa0702013-03-10 23:51:38 +0100945 * When a file descriptor is passed, its value is put into the s_addr part of
Willy Tarreaua5b325f2020-09-04 16:44:20 +0200946 * the address when cast to sockaddr_in and the address family is
947 * AF_CUST_EXISTING_FD.
Willy Tarreaua93e5c72020-09-15 14:01:16 +0200948 *
Willy Tarreau5fc93282020-09-16 18:25:03 +0200949 * The matching protocol will be set into <proto> if non-null.
950 *
Willy Tarreaua93e5c72020-09-15 14:01:16 +0200951 * Any known file descriptor is also assigned to <fd> if non-null, otherwise it
952 * is forced to -1.
Willy Tarreaufab5a432011-03-04 15:31:53 +0100953 */
Willy Tarreau5fc93282020-09-16 18:25:03 +0200954struct sockaddr_storage *str2sa_range(const char *str, int *port, int *low, int *high, int *fd,
955 struct protocol **proto, char **err,
956 const char *pfx, char **fqdn, unsigned int opts)
Willy Tarreaufab5a432011-03-04 15:31:53 +0100957{
Christopher Faulet1bc04c72017-10-29 20:14:08 +0100958 static THREAD_LOCAL struct sockaddr_storage ss;
David du Colombier6f5ccb12011-03-10 22:26:24 +0100959 struct sockaddr_storage *ret = NULL;
Willy Tarreau5fc93282020-09-16 18:25:03 +0200960 struct protocol *new_proto = NULL;
Willy Tarreau24709282013-03-10 21:32:12 +0100961 char *back, *str2;
Willy Tarreaud4448bc2013-02-20 15:55:15 +0100962 char *port1, *port2;
963 int portl, porth, porta;
Willy Tarreauccfccef2014-05-10 01:49:15 +0200964 int abstract = 0;
Willy Tarreaua93e5c72020-09-15 14:01:16 +0200965 int new_fd = -1;
Willy Tarreaue3b45182021-10-27 17:28:55 +0200966 enum proto_type proto_type;
967 int ctrl_type;
Willy Tarreaud4448bc2013-02-20 15:55:15 +0100968
969 portl = porth = porta = 0;
Willy Tarreau72b8c1f2015-09-08 15:50:19 +0200970 if (fqdn)
971 *fqdn = NULL;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200972
Willy Tarreaudad36a32013-03-11 01:20:04 +0100973 str2 = back = env_expand(strdup(str));
Willy Tarreaudf350f12013-03-01 20:22:54 +0100974 if (str2 == NULL) {
Thierry Fournier3e83b092023-05-23 17:58:03 +0200975 memprintf(err, "out of memory in '%s'", __FUNCTION__);
Willy Tarreaud5191e72010-02-09 20:50:45 +0100976 goto out;
Willy Tarreaudf350f12013-03-01 20:22:54 +0100977 }
Willy Tarreaubaaee002006-06-26 02:48:02 +0200978
Willy Tarreau9f69f462015-09-08 16:01:25 +0200979 if (!*str2) {
Thierry Fournier3e83b092023-05-23 17:58:03 +0200980 memprintf(err, "'%s' resolves to an empty address (environment variable missing?)", str);
Willy Tarreau9f69f462015-09-08 16:01:25 +0200981 goto out;
982 }
983
Willy Tarreau24709282013-03-10 21:32:12 +0100984 memset(&ss, 0, sizeof(ss));
985
Willy Tarreaue835bd82020-09-16 11:35:47 +0200986 /* prepare the default socket types */
Willy Tarreauf23b1bc2021-03-23 18:36:37 +0100987 if ((opts & (PA_O_STREAM|PA_O_DGRAM)) == PA_O_DGRAM ||
Willy Tarreaue3b45182021-10-27 17:28:55 +0200988 ((opts & (PA_O_STREAM|PA_O_DGRAM)) == (PA_O_DGRAM|PA_O_STREAM) && (opts & PA_O_DEFAULT_DGRAM))) {
989 proto_type = PROTO_TYPE_DGRAM;
990 ctrl_type = SOCK_DGRAM;
991 } else {
992 proto_type = PROTO_TYPE_STREAM;
993 ctrl_type = SOCK_STREAM;
994 }
Willy Tarreaue835bd82020-09-16 11:35:47 +0200995
996 if (strncmp(str2, "stream+", 7) == 0) {
997 str2 += 7;
Willy Tarreaue3b45182021-10-27 17:28:55 +0200998 proto_type = PROTO_TYPE_STREAM;
999 ctrl_type = SOCK_STREAM;
Willy Tarreaue835bd82020-09-16 11:35:47 +02001000 }
1001 else if (strncmp(str2, "dgram+", 6) == 0) {
1002 str2 += 6;
Willy Tarreaue3b45182021-10-27 17:28:55 +02001003 proto_type = PROTO_TYPE_DGRAM;
1004 ctrl_type = SOCK_DGRAM;
Willy Tarreaue835bd82020-09-16 11:35:47 +02001005 }
Willy Tarreau40725a42023-01-16 13:55:27 +01001006 else if (strncmp(str2, "quic+", 5) == 0) {
1007 str2 += 5;
1008 proto_type = PROTO_TYPE_DGRAM;
1009 ctrl_type = SOCK_STREAM;
1010 }
Willy Tarreaue835bd82020-09-16 11:35:47 +02001011
Willy Tarreau24709282013-03-10 21:32:12 +01001012 if (strncmp(str2, "unix@", 5) == 0) {
1013 str2 += 5;
Willy Tarreauccfccef2014-05-10 01:49:15 +02001014 abstract = 0;
Willy Tarreau24709282013-03-10 21:32:12 +01001015 ss.ss_family = AF_UNIX;
1016 }
Emeric Brunce325c42021-04-02 17:05:09 +02001017 else if (strncmp(str2, "uxdg@", 5) == 0) {
1018 str2 += 5;
1019 abstract = 0;
1020 ss.ss_family = AF_UNIX;
Willy Tarreaue3b45182021-10-27 17:28:55 +02001021 proto_type = PROTO_TYPE_DGRAM;
1022 ctrl_type = SOCK_DGRAM;
Emeric Brunce325c42021-04-02 17:05:09 +02001023 }
1024 else if (strncmp(str2, "uxst@", 5) == 0) {
1025 str2 += 5;
1026 abstract = 0;
1027 ss.ss_family = AF_UNIX;
Willy Tarreaue3b45182021-10-27 17:28:55 +02001028 proto_type = PROTO_TYPE_STREAM;
1029 ctrl_type = SOCK_STREAM;
Emeric Brunce325c42021-04-02 17:05:09 +02001030 }
Willy Tarreauccfccef2014-05-10 01:49:15 +02001031 else if (strncmp(str2, "abns@", 5) == 0) {
1032 str2 += 5;
1033 abstract = 1;
1034 ss.ss_family = AF_UNIX;
1035 }
Emeric Brunce325c42021-04-02 17:05:09 +02001036 else if (strncmp(str2, "ip@", 3) == 0) {
1037 str2 += 3;
1038 ss.ss_family = AF_UNSPEC;
1039 }
Willy Tarreau24709282013-03-10 21:32:12 +01001040 else if (strncmp(str2, "ipv4@", 5) == 0) {
1041 str2 += 5;
1042 ss.ss_family = AF_INET;
1043 }
1044 else if (strncmp(str2, "ipv6@", 5) == 0) {
1045 str2 += 5;
1046 ss.ss_family = AF_INET6;
1047 }
Emeric Brunce325c42021-04-02 17:05:09 +02001048 else if (strncmp(str2, "tcp4@", 5) == 0) {
1049 str2 += 5;
1050 ss.ss_family = AF_INET;
Willy Tarreaue3b45182021-10-27 17:28:55 +02001051 proto_type = PROTO_TYPE_STREAM;
1052 ctrl_type = SOCK_STREAM;
Emeric Brunce325c42021-04-02 17:05:09 +02001053 }
Emeric Brun3835c0d2020-07-07 09:46:09 +02001054 else if (strncmp(str2, "udp4@", 5) == 0) {
1055 str2 += 5;
1056 ss.ss_family = AF_INET;
Willy Tarreaue3b45182021-10-27 17:28:55 +02001057 proto_type = PROTO_TYPE_DGRAM;
1058 ctrl_type = SOCK_DGRAM;
Emeric Brun3835c0d2020-07-07 09:46:09 +02001059 }
Emeric Brunce325c42021-04-02 17:05:09 +02001060 else if (strncmp(str2, "tcp6@", 5) == 0) {
1061 str2 += 5;
1062 ss.ss_family = AF_INET6;
Willy Tarreaue3b45182021-10-27 17:28:55 +02001063 proto_type = PROTO_TYPE_STREAM;
1064 ctrl_type = SOCK_STREAM;
Emeric Brunce325c42021-04-02 17:05:09 +02001065 }
Emeric Brun3835c0d2020-07-07 09:46:09 +02001066 else if (strncmp(str2, "udp6@", 5) == 0) {
1067 str2 += 5;
1068 ss.ss_family = AF_INET6;
Willy Tarreaue3b45182021-10-27 17:28:55 +02001069 proto_type = PROTO_TYPE_DGRAM;
1070 ctrl_type = SOCK_DGRAM;
Emeric Brun3835c0d2020-07-07 09:46:09 +02001071 }
Emeric Brunce325c42021-04-02 17:05:09 +02001072 else if (strncmp(str2, "tcp@", 4) == 0) {
1073 str2 += 4;
1074 ss.ss_family = AF_UNSPEC;
Willy Tarreaue3b45182021-10-27 17:28:55 +02001075 proto_type = PROTO_TYPE_STREAM;
1076 ctrl_type = SOCK_STREAM;
Emeric Brunce325c42021-04-02 17:05:09 +02001077 }
Emeric Brun3835c0d2020-07-07 09:46:09 +02001078 else if (strncmp(str2, "udp@", 4) == 0) {
1079 str2 += 4;
1080 ss.ss_family = AF_UNSPEC;
Willy Tarreaue3b45182021-10-27 17:28:55 +02001081 proto_type = PROTO_TYPE_DGRAM;
1082 ctrl_type = SOCK_DGRAM;
Emeric Brun3835c0d2020-07-07 09:46:09 +02001083 }
Frédéric Lécaille10caf652020-11-23 11:36:57 +01001084 else if (strncmp(str2, "quic4@", 6) == 0) {
1085 str2 += 6;
1086 ss.ss_family = AF_INET;
Willy Tarreaue3b45182021-10-27 17:28:55 +02001087 proto_type = PROTO_TYPE_DGRAM;
Frédéric Lécaille10caf652020-11-23 11:36:57 +01001088 ctrl_type = SOCK_STREAM;
1089 }
1090 else if (strncmp(str2, "quic6@", 6) == 0) {
1091 str2 += 6;
1092 ss.ss_family = AF_INET6;
Willy Tarreaue3b45182021-10-27 17:28:55 +02001093 proto_type = PROTO_TYPE_DGRAM;
Frédéric Lécaille10caf652020-11-23 11:36:57 +01001094 ctrl_type = SOCK_STREAM;
1095 }
Willy Tarreau5a7beed2020-09-04 16:54:05 +02001096 else if (strncmp(str2, "fd@", 3) == 0) {
1097 str2 += 3;
1098 ss.ss_family = AF_CUST_EXISTING_FD;
1099 }
1100 else if (strncmp(str2, "sockpair@", 9) == 0) {
1101 str2 += 9;
1102 ss.ss_family = AF_CUST_SOCKPAIR;
1103 }
Willy Tarreau24709282013-03-10 21:32:12 +01001104 else if (*str2 == '/') {
1105 ss.ss_family = AF_UNIX;
1106 }
1107 else
1108 ss.ss_family = AF_UNSPEC;
1109
Willy Tarreau5a7beed2020-09-04 16:54:05 +02001110 if (ss.ss_family == AF_CUST_SOCKPAIR) {
Willy Tarreaua215be22020-09-16 10:14:16 +02001111 struct sockaddr_storage ss2;
1112 socklen_t addr_len;
William Lallemand2fe7dd02018-09-11 16:51:29 +02001113 char *endptr;
1114
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001115 new_fd = strtol(str2, &endptr, 10);
1116 if (!*str2 || new_fd < 0 || *endptr) {
Thierry Fournier3e83b092023-05-23 17:58:03 +02001117 memprintf(err, "file descriptor '%s' is not a valid integer in '%s'", str2, str);
William Lallemand2fe7dd02018-09-11 16:51:29 +02001118 goto out;
1119 }
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001120
Willy Tarreaua215be22020-09-16 10:14:16 +02001121 /* just verify that it's a socket */
1122 addr_len = sizeof(ss2);
1123 if (getsockname(new_fd, (struct sockaddr *)&ss2, &addr_len) == -1) {
Thierry Fournier3e83b092023-05-23 17:58:03 +02001124 memprintf(err, "cannot use file descriptor '%d' : %s.", new_fd, strerror(errno));
Willy Tarreaua215be22020-09-16 10:14:16 +02001125 goto out;
1126 }
1127
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001128 ((struct sockaddr_in *)&ss)->sin_addr.s_addr = new_fd;
1129 ((struct sockaddr_in *)&ss)->sin_port = 0;
William Lallemand2fe7dd02018-09-11 16:51:29 +02001130 }
Willy Tarreau5a7beed2020-09-04 16:54:05 +02001131 else if (ss.ss_family == AF_CUST_EXISTING_FD) {
Willy Tarreau40aa0702013-03-10 23:51:38 +01001132 char *endptr;
1133
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001134 new_fd = strtol(str2, &endptr, 10);
1135 if (!*str2 || new_fd < 0 || *endptr) {
Thierry Fournier3e83b092023-05-23 17:58:03 +02001136 memprintf(err, "file descriptor '%s' is not a valid integer in '%s'", str2, str);
Willy Tarreau40aa0702013-03-10 23:51:38 +01001137 goto out;
1138 }
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001139
Willy Tarreau6edc7222020-09-15 17:41:56 +02001140 if (opts & PA_O_SOCKET_FD) {
1141 socklen_t addr_len;
1142 int type;
1143
1144 addr_len = sizeof(ss);
1145 if (getsockname(new_fd, (struct sockaddr *)&ss, &addr_len) == -1) {
Thierry Fournier3e83b092023-05-23 17:58:03 +02001146 memprintf(err, "cannot use file descriptor '%d' : %s.", new_fd, strerror(errno));
Willy Tarreau6edc7222020-09-15 17:41:56 +02001147 goto out;
1148 }
1149
1150 addr_len = sizeof(type);
1151 if (getsockopt(new_fd, SOL_SOCKET, SO_TYPE, &type, &addr_len) != 0 ||
Willy Tarreaue3b45182021-10-27 17:28:55 +02001152 (type == SOCK_STREAM) != (proto_type == PROTO_TYPE_STREAM)) {
Thierry Fournier3e83b092023-05-23 17:58:03 +02001153 memprintf(err, "socket on file descriptor '%d' is of the wrong type.", new_fd);
Willy Tarreau6edc7222020-09-15 17:41:56 +02001154 goto out;
1155 }
1156
1157 porta = portl = porth = get_host_port(&ss);
1158 } else if (opts & PA_O_RAW_FD) {
1159 ((struct sockaddr_in *)&ss)->sin_addr.s_addr = new_fd;
1160 ((struct sockaddr_in *)&ss)->sin_port = 0;
1161 } else {
Thierry Fournier3e83b092023-05-23 17:58:03 +02001162 memprintf(err, "a file descriptor is not acceptable here in '%s'", str);
Willy Tarreau6edc7222020-09-15 17:41:56 +02001163 goto out;
1164 }
Willy Tarreau40aa0702013-03-10 23:51:38 +01001165 }
1166 else if (ss.ss_family == AF_UNIX) {
Willy Tarreau8daa9202019-06-16 18:16:33 +02001167 struct sockaddr_un *un = (struct sockaddr_un *)&ss;
Willy Tarreau15586382013-03-04 19:48:14 +01001168 int prefix_path_len;
1169 int max_path_len;
Willy Tarreau94ef3f32014-04-14 14:49:00 +02001170 int adr_len;
Willy Tarreau15586382013-03-04 19:48:14 +01001171
1172 /* complete unix socket path name during startup or soft-restart is
1173 * <unix_bind_prefix><path>.<pid>.<bak|tmp>
1174 */
Willy Tarreauccfccef2014-05-10 01:49:15 +02001175 prefix_path_len = (pfx && !abstract) ? strlen(pfx) : 0;
Willy Tarreau8daa9202019-06-16 18:16:33 +02001176 max_path_len = (sizeof(un->sun_path) - 1) -
Willy Tarreau327ea5a2020-02-11 06:43:37 +01001177 (abstract ? 0 : prefix_path_len + 1 + 5 + 1 + 3);
Willy Tarreau15586382013-03-04 19:48:14 +01001178
Willy Tarreau94ef3f32014-04-14 14:49:00 +02001179 adr_len = strlen(str2);
1180 if (adr_len > max_path_len) {
Thierry Fournier3e83b092023-05-23 17:58:03 +02001181 memprintf(err, "socket path '%s' too long (max %d)", str, max_path_len);
Willy Tarreau15586382013-03-04 19:48:14 +01001182 goto out;
1183 }
1184
Willy Tarreauccfccef2014-05-10 01:49:15 +02001185 /* when abstract==1, we skip the first zero and copy all bytes except the trailing zero */
Willy Tarreau8daa9202019-06-16 18:16:33 +02001186 memset(un->sun_path, 0, sizeof(un->sun_path));
Willy Tarreau94ef3f32014-04-14 14:49:00 +02001187 if (prefix_path_len)
Willy Tarreau8daa9202019-06-16 18:16:33 +02001188 memcpy(un->sun_path, pfx, prefix_path_len);
1189 memcpy(un->sun_path + prefix_path_len + abstract, str2, adr_len + 1 - abstract);
Willy Tarreau15586382013-03-04 19:48:14 +01001190 }
Willy Tarreau24709282013-03-10 21:32:12 +01001191 else { /* IPv4 and IPv6 */
mildisff5d5102015-10-26 18:50:08 +01001192 char *end = str2 + strlen(str2);
1193 char *chr;
Willy Tarreau72b8c1f2015-09-08 15:50:19 +02001194
mildisff5d5102015-10-26 18:50:08 +01001195 /* search for : or ] whatever comes first */
1196 for (chr = end-1; chr > str2; chr--) {
1197 if (*chr == ']' || *chr == ':')
1198 break;
1199 }
1200
1201 if (*chr == ':') {
1202 /* Found a colon before a closing-bracket, must be a port separator.
1203 * This guarantee backward compatibility.
1204 */
Willy Tarreau7f96a842020-09-15 11:12:44 +02001205 if (!(opts & PA_O_PORT_OK)) {
1206 memprintf(err, "port specification not permitted here in '%s'", str);
1207 goto out;
1208 }
mildisff5d5102015-10-26 18:50:08 +01001209 *chr++ = '\0';
1210 port1 = chr;
1211 }
1212 else {
1213 /* Either no colon and no closing-bracket
1214 * or directly ending with a closing-bracket.
1215 * However, no port.
1216 */
Willy Tarreau7f96a842020-09-15 11:12:44 +02001217 if (opts & PA_O_PORT_MAND) {
1218 memprintf(err, "missing port specification in '%s'", str);
1219 goto out;
1220 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001221 port1 = "";
mildisff5d5102015-10-26 18:50:08 +01001222 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02001223
Willy Tarreau90807112020-02-25 08:16:33 +01001224 if (isdigit((unsigned char)*port1)) { /* single port or range */
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001225 port2 = strchr(port1, '-');
Willy Tarreau7f96a842020-09-15 11:12:44 +02001226 if (port2) {
1227 if (!(opts & PA_O_PORT_RANGE)) {
1228 memprintf(err, "port range not permitted here in '%s'", str);
1229 goto out;
1230 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001231 *port2++ = '\0';
Willy Tarreau7f96a842020-09-15 11:12:44 +02001232 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001233 else
1234 port2 = port1;
1235 portl = atoi(port1);
1236 porth = atoi(port2);
Willy Tarreau7f96a842020-09-15 11:12:44 +02001237
1238 if (portl < !!(opts & PA_O_PORT_MAND) || portl > 65535) {
1239 memprintf(err, "invalid port '%s'", port1);
1240 goto out;
1241 }
1242
1243 if (porth < !!(opts & PA_O_PORT_MAND) || porth > 65535) {
1244 memprintf(err, "invalid port '%s'", port2);
1245 goto out;
1246 }
1247
1248 if (portl > porth) {
1249 memprintf(err, "invalid port range '%d-%d'", portl, porth);
1250 goto out;
1251 }
1252
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001253 porta = portl;
1254 }
1255 else if (*port1 == '-') { /* negative offset */
Willy Tarreau7f96a842020-09-15 11:12:44 +02001256 if (!(opts & PA_O_PORT_OFS)) {
1257 memprintf(err, "port offset not permitted here in '%s'", str);
1258 goto out;
1259 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001260 portl = atoi(port1 + 1);
1261 porta = -portl;
1262 }
1263 else if (*port1 == '+') { /* positive offset */
Willy Tarreau7f96a842020-09-15 11:12:44 +02001264 if (!(opts & PA_O_PORT_OFS)) {
1265 memprintf(err, "port offset not permitted here in '%s'", str);
1266 goto out;
1267 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001268 porth = atoi(port1 + 1);
1269 porta = porth;
1270 }
1271 else if (*port1) { /* other any unexpected char */
Thierry Fournier3e83b092023-05-23 17:58:03 +02001272 memprintf(err, "invalid character '%c' in port number '%s' in '%s'", *port1, port1, str);
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001273 goto out;
1274 }
Willy Tarreau7f96a842020-09-15 11:12:44 +02001275 else if (opts & PA_O_PORT_MAND) {
1276 memprintf(err, "missing port specification in '%s'", str);
1277 goto out;
1278 }
Willy Tarreauceccdd72016-11-02 22:27:10 +01001279
1280 /* first try to parse the IP without resolving. If it fails, it
1281 * tells us we need to keep a copy of the FQDN to resolve later
1282 * and to enable DNS. In this case we can proceed if <fqdn> is
Willy Tarreaucd3a55912020-09-04 15:30:46 +02001283 * set or if PA_O_RESOLVE is set, otherwise it's an error.
Willy Tarreauceccdd72016-11-02 22:27:10 +01001284 */
1285 if (str2ip2(str2, &ss, 0) == NULL) {
Willy Tarreaucd3a55912020-09-04 15:30:46 +02001286 if ((!(opts & PA_O_RESOLVE) && !fqdn) ||
1287 ((opts & PA_O_RESOLVE) && str2ip2(str2, &ss, 1) == NULL)) {
Thierry Fournier3e83b092023-05-23 17:58:03 +02001288 memprintf(err, "invalid address: '%s' in '%s'", str2, str);
Willy Tarreauceccdd72016-11-02 22:27:10 +01001289 goto out;
1290 }
Willy Tarreau72b8c1f2015-09-08 15:50:19 +02001291
Willy Tarreauceccdd72016-11-02 22:27:10 +01001292 if (fqdn) {
1293 if (str2 != back)
1294 memmove(back, str2, strlen(str2) + 1);
1295 *fqdn = back;
1296 back = NULL;
1297 }
Willy Tarreau72b8c1f2015-09-08 15:50:19 +02001298 }
Willy Tarreauceccdd72016-11-02 22:27:10 +01001299 set_host_port(&ss, porta);
Willy Tarreaue4c58c82013-03-06 15:28:17 +01001300 }
Willy Tarreaufab5a432011-03-04 15:31:53 +01001301
Willy Tarreaue835bd82020-09-16 11:35:47 +02001302 if (ctrl_type == SOCK_STREAM && !(opts & PA_O_STREAM)) {
Thierry Fournier3e83b092023-05-23 17:58:03 +02001303 memprintf(err, "stream-type address not acceptable in '%s'", str);
Willy Tarreaue835bd82020-09-16 11:35:47 +02001304 goto out;
1305 }
1306 else if (ctrl_type == SOCK_DGRAM && !(opts & PA_O_DGRAM)) {
Thierry Fournier3e83b092023-05-23 17:58:03 +02001307 memprintf(err, "dgram-type address not acceptable in '%s'", str);
Willy Tarreaue835bd82020-09-16 11:35:47 +02001308 goto out;
1309 }
1310
Willy Tarreau65ec4e32020-09-16 19:17:08 +02001311 if (proto || (opts & PA_O_CONNECT)) {
Willy Tarreau5fc93282020-09-16 18:25:03 +02001312 /* Note: if the caller asks for a proto, we must find one,
Emeric Brun26754902021-04-07 14:26:44 +02001313 * except if we inherit from a raw FD (family == AF_CUST_EXISTING_FD)
1314 * orif we return with an fqdn that will resolve later,
Willy Tarreau5fc93282020-09-16 18:25:03 +02001315 * in which case the address is not known yet (this is only
1316 * for servers actually).
1317 */
Willy Tarreaub2ffc992020-09-16 21:37:31 +02001318 new_proto = protocol_lookup(ss.ss_family,
Willy Tarreaue3b45182021-10-27 17:28:55 +02001319 proto_type,
Willy Tarreauaf9609b2020-09-16 22:04:46 +02001320 ctrl_type == SOCK_DGRAM);
Willy Tarreaub2ffc992020-09-16 21:37:31 +02001321
Emeric Brun26754902021-04-07 14:26:44 +02001322 if (!new_proto && (!fqdn || !*fqdn) && (ss.ss_family != AF_CUST_EXISTING_FD)) {
Willy Tarreau3d7b4682022-05-20 17:32:35 +02001323 memprintf(err, "unsupported %s protocol for %s family %d address '%s'%s",
Willy Tarreau2b049b82022-05-20 17:28:30 +02001324 (ctrl_type == SOCK_DGRAM) ? "datagram" : "stream",
1325 (proto_type == PROTO_TYPE_DGRAM) ? "datagram" : "stream",
1326 ss.ss_family,
Willy Tarreau3d7b4682022-05-20 17:32:35 +02001327 str,
Willy Tarreau3d7b4682022-05-20 17:32:35 +02001328#ifndef USE_QUIC
Tim Duesterhus147eeb22022-05-22 12:40:58 +02001329 (ctrl_type == SOCK_STREAM && proto_type == PROTO_TYPE_DGRAM)
1330 ? "; QUIC is not compiled in if this is what you were looking for."
1331 : ""
Willy Tarreau3d7b4682022-05-20 17:32:35 +02001332#else
1333 ""
1334#endif
Tim Duesterhus147eeb22022-05-22 12:40:58 +02001335 );
Willy Tarreau5fc93282020-09-16 18:25:03 +02001336 goto out;
1337 }
Willy Tarreau65ec4e32020-09-16 19:17:08 +02001338
1339 if ((opts & PA_O_CONNECT) && new_proto && !new_proto->connect) {
1340 memprintf(err, "connect() not supported for this protocol family %d used by address '%s'", ss.ss_family, str);
1341 goto out;
1342 }
Willy Tarreau5fc93282020-09-16 18:25:03 +02001343 }
1344
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001345 ret = &ss;
Willy Tarreaud5191e72010-02-09 20:50:45 +01001346 out:
Willy Tarreau48ef4c92017-01-06 18:32:38 +01001347 if (port)
1348 *port = porta;
Willy Tarreaud4448bc2013-02-20 15:55:15 +01001349 if (low)
1350 *low = portl;
1351 if (high)
1352 *high = porth;
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001353 if (fd)
1354 *fd = new_fd;
Willy Tarreau5fc93282020-09-16 18:25:03 +02001355 if (proto)
1356 *proto = new_proto;
Willy Tarreau24709282013-03-10 21:32:12 +01001357 free(back);
Willy Tarreaud5191e72010-02-09 20:50:45 +01001358 return ret;
Willy Tarreauc6f4ce82009-06-10 11:09:37 +02001359}
1360
Thayne McCombs92149f92020-11-20 01:28:26 -07001361/* converts <addr> and <port> into a string representation of the address and port. This is sort
1362 * of an inverse of str2sa_range, with some restrictions. The supported families are AF_INET,
1363 * AF_INET6, AF_UNIX, and AF_CUST_SOCKPAIR. If the family is unsopported NULL is returned.
1364 * If map_ports is true, then the sign of the port is included in the output, to indicate it is
1365 * relative to the incoming port. AF_INET and AF_INET6 will be in the form "<addr>:<port>".
1366 * AF_UNIX will either be just the path (if using a pathname) or "abns@<path>" if it is abstract.
1367 * AF_CUST_SOCKPAIR will be of the form "sockpair@<fd>".
1368 *
1369 * The returned char* is allocated, and it is the responsibility of the caller to free it.
1370 */
1371char * sa2str(const struct sockaddr_storage *addr, int port, int map_ports)
1372{
1373 char buffer[INET6_ADDRSTRLEN];
1374 char *out = NULL;
1375 const void *ptr;
1376 const char *path;
1377
1378 switch (addr->ss_family) {
1379 case AF_INET:
1380 ptr = &((struct sockaddr_in *)addr)->sin_addr;
1381 break;
1382 case AF_INET6:
1383 ptr = &((struct sockaddr_in6 *)addr)->sin6_addr;
1384 break;
1385 case AF_UNIX:
1386 path = ((struct sockaddr_un *)addr)->sun_path;
1387 if (path[0] == '\0') {
1388 const int max_length = sizeof(struct sockaddr_un) - offsetof(struct sockaddr_un, sun_path) - 1;
1389 return memprintf(&out, "abns@%.*s", max_length, path+1);
1390 } else {
1391 return strdup(path);
1392 }
1393 case AF_CUST_SOCKPAIR:
1394 return memprintf(&out, "sockpair@%d", ((struct sockaddr_in *)addr)->sin_addr.s_addr);
1395 default:
1396 return NULL;
1397 }
Tim Duesterhus22535a52022-05-23 09:30:49 +02001398 if (inet_ntop(addr->ss_family, ptr, buffer, sizeof(buffer)) == NULL) {
1399 BUG_ON(errno == ENOSPC);
1400 return NULL;
1401 }
Thayne McCombs92149f92020-11-20 01:28:26 -07001402 if (map_ports)
1403 return memprintf(&out, "%s:%+d", buffer, port);
1404 else
1405 return memprintf(&out, "%s:%d", buffer, port);
1406}
1407
1408
Willy Tarreau2937c0d2010-01-26 17:36:17 +01001409/* converts <str> to a struct in_addr containing a network mask. It can be
1410 * passed in dotted form (255.255.255.0) or in CIDR form (24). It returns 1
Jarno Huuskonen577d5ac2017-05-21 17:32:21 +03001411 * if the conversion succeeds otherwise zero.
Willy Tarreau2937c0d2010-01-26 17:36:17 +01001412 */
1413int str2mask(const char *str, struct in_addr *mask)
1414{
1415 if (strchr(str, '.') != NULL) { /* dotted notation */
1416 if (!inet_pton(AF_INET, str, mask))
1417 return 0;
1418 }
1419 else { /* mask length */
1420 char *err;
1421 unsigned long len = strtol(str, &err, 10);
1422
1423 if (!*str || (err && *err) || (unsigned)len > 32)
1424 return 0;
Tim Duesterhus8575f722018-01-25 16:24:48 +01001425
1426 len2mask4(len, mask);
Willy Tarreau2937c0d2010-01-26 17:36:17 +01001427 }
1428 return 1;
1429}
1430
Tim Duesterhus47185172018-01-25 16:24:49 +01001431/* converts <str> to a struct in6_addr containing a network mask. It can be
Tim Duesterhus5e642862018-02-20 17:02:18 +01001432 * passed in quadruplet form (ffff:ffff::) or in CIDR form (64). It returns 1
Tim Duesterhus47185172018-01-25 16:24:49 +01001433 * if the conversion succeeds otherwise zero.
1434 */
1435int str2mask6(const char *str, struct in6_addr *mask)
1436{
1437 if (strchr(str, ':') != NULL) { /* quadruplet notation */
1438 if (!inet_pton(AF_INET6, str, mask))
1439 return 0;
1440 }
1441 else { /* mask length */
1442 char *err;
1443 unsigned long len = strtol(str, &err, 10);
1444
1445 if (!*str || (err && *err) || (unsigned)len > 128)
1446 return 0;
1447
1448 len2mask6(len, mask);
1449 }
1450 return 1;
1451}
1452
Thierry FOURNIERb0504632013-12-14 15:39:02 +01001453/* convert <cidr> to struct in_addr <mask>. It returns 1 if the conversion
1454 * succeeds otherwise zero.
1455 */
1456int cidr2dotted(int cidr, struct in_addr *mask) {
1457
1458 if (cidr < 0 || cidr > 32)
1459 return 0;
1460
1461 mask->s_addr = cidr ? htonl(~0UL << (32 - cidr)) : 0;
1462 return 1;
1463}
1464
Thierry Fournier70473a52016-02-17 17:12:14 +01001465/* Convert mask from bit length form to in_addr form.
1466 * This function never fails.
1467 */
1468void len2mask4(int len, struct in_addr *addr)
1469{
1470 if (len >= 32) {
1471 addr->s_addr = 0xffffffff;
1472 return;
1473 }
1474 if (len <= 0) {
1475 addr->s_addr = 0x00000000;
1476 return;
1477 }
1478 addr->s_addr = 0xffffffff << (32 - len);
1479 addr->s_addr = htonl(addr->s_addr);
1480}
1481
1482/* Convert mask from bit length form to in6_addr form.
1483 * This function never fails.
1484 */
1485void len2mask6(int len, struct in6_addr *addr)
1486{
1487 len2mask4(len, (struct in_addr *)&addr->s6_addr[0]); /* msb */
1488 len -= 32;
1489 len2mask4(len, (struct in_addr *)&addr->s6_addr[4]);
1490 len -= 32;
1491 len2mask4(len, (struct in_addr *)&addr->s6_addr[8]);
1492 len -= 32;
1493 len2mask4(len, (struct in_addr *)&addr->s6_addr[12]); /* lsb */
1494}
1495
Willy Tarreauc6f4ce82009-06-10 11:09:37 +02001496/*
Willy Tarreaud077a8e2007-05-08 18:28:09 +02001497 * converts <str> to two struct in_addr* which must be pre-allocated.
Willy Tarreaubaaee002006-06-26 02:48:02 +02001498 * The format is "addr[/mask]", where "addr" cannot be empty, and mask
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05001499 * is optional and either in the dotted or CIDR notation.
Willy Tarreaubaaee002006-06-26 02:48:02 +02001500 * Note: "addr" can also be a hostname. Returns 1 if OK, 0 if error.
1501 */
Thierry FOURNIERfc7ac7b2014-02-11 15:23:04 +01001502int str2net(const char *str, int resolve, struct in_addr *addr, struct in_addr *mask)
Willy Tarreaubaaee002006-06-26 02:48:02 +02001503{
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001504 __label__ out_free, out_err;
1505 char *c, *s;
1506 int ret_val;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001507
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001508 s = strdup(str);
1509 if (!s)
1510 return 0;
1511
Willy Tarreaubaaee002006-06-26 02:48:02 +02001512 memset(mask, 0, sizeof(*mask));
1513 memset(addr, 0, sizeof(*addr));
Willy Tarreaubaaee002006-06-26 02:48:02 +02001514
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001515 if ((c = strrchr(s, '/')) != NULL) {
Willy Tarreaubaaee002006-06-26 02:48:02 +02001516 *c++ = '\0';
1517 /* c points to the mask */
Willy Tarreau2937c0d2010-01-26 17:36:17 +01001518 if (!str2mask(c, mask))
1519 goto out_err;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001520 }
1521 else {
Willy Tarreauebd61602006-12-30 11:54:15 +01001522 mask->s_addr = ~0U;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001523 }
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001524 if (!inet_pton(AF_INET, s, addr)) {
Willy Tarreaubaaee002006-06-26 02:48:02 +02001525 struct hostent *he;
1526
Thierry FOURNIERfc7ac7b2014-02-11 15:23:04 +01001527 if (!resolve)
1528 goto out_err;
1529
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001530 if ((he = gethostbyname(s)) == NULL) {
1531 goto out_err;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001532 }
1533 else
1534 *addr = *(struct in_addr *) *(he->h_addr_list);
1535 }
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001536
1537 ret_val = 1;
1538 out_free:
1539 free(s);
1540 return ret_val;
1541 out_err:
1542 ret_val = 0;
1543 goto out_free;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001544}
1545
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001546
1547/*
Willy Tarreau6d20e282012-04-27 22:49:47 +02001548 * converts <str> to two struct in6_addr* which must be pre-allocated.
1549 * The format is "addr[/mask]", where "addr" cannot be empty, and mask
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05001550 * is an optional number of bits (128 being the default).
Willy Tarreau6d20e282012-04-27 22:49:47 +02001551 * Returns 1 if OK, 0 if error.
1552 */
1553int str62net(const char *str, struct in6_addr *addr, unsigned char *mask)
1554{
1555 char *c, *s;
1556 int ret_val = 0;
1557 char *err;
1558 unsigned long len = 128;
1559
1560 s = strdup(str);
1561 if (!s)
1562 return 0;
1563
1564 memset(mask, 0, sizeof(*mask));
1565 memset(addr, 0, sizeof(*addr));
1566
1567 if ((c = strrchr(s, '/')) != NULL) {
1568 *c++ = '\0'; /* c points to the mask */
1569 if (!*c)
1570 goto out_free;
1571
1572 len = strtoul(c, &err, 10);
1573 if ((err && *err) || (unsigned)len > 128)
1574 goto out_free;
1575 }
1576 *mask = len; /* OK we have a valid mask in <len> */
1577
1578 if (!inet_pton(AF_INET6, s, addr))
1579 goto out_free;
1580
1581 ret_val = 1;
1582 out_free:
1583 free(s);
1584 return ret_val;
1585}
1586
1587
1588/*
Willy Tarreau12e10272021-03-25 11:34:40 +01001589 * Parse IPv4 address found in url. Return the number of bytes parsed. It
1590 * expects exactly 4 numbers between 0 and 255 delimited by dots, and returns
1591 * zero in case of mismatch.
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001592 */
David du Colombier6f5ccb12011-03-10 22:26:24 +01001593int url2ipv4(const char *addr, struct in_addr *dst)
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001594{
1595 int saw_digit, octets, ch;
1596 u_char tmp[4], *tp;
1597 const char *cp = addr;
1598
1599 saw_digit = 0;
1600 octets = 0;
1601 *(tp = tmp) = 0;
1602
1603 while (*addr) {
Willy Tarreau12e10272021-03-25 11:34:40 +01001604 unsigned char digit = (ch = *addr) - '0';
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001605 if (digit > 9 && ch != '.')
1606 break;
Willy Tarreau12e10272021-03-25 11:34:40 +01001607 addr++;
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001608 if (digit <= 9) {
1609 u_int new = *tp * 10 + digit;
1610 if (new > 255)
1611 return 0;
1612 *tp = new;
1613 if (!saw_digit) {
1614 if (++octets > 4)
1615 return 0;
1616 saw_digit = 1;
1617 }
1618 } else if (ch == '.' && saw_digit) {
1619 if (octets == 4)
1620 return 0;
1621 *++tp = 0;
1622 saw_digit = 0;
1623 } else
1624 return 0;
1625 }
1626
1627 if (octets < 4)
1628 return 0;
1629
1630 memcpy(&dst->s_addr, tmp, 4);
Willy Tarreau12e10272021-03-25 11:34:40 +01001631 return addr - cp;
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001632}
1633
1634/*
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001635 * Resolve destination server from URL. Convert <str> to a sockaddr_storage.
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05001636 * <out> contain the code of the detected scheme, the start and length of
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001637 * the hostname. Actually only http and https are supported. <out> can be NULL.
1638 * This function returns the consumed length. It is useful if you parse complete
1639 * url like http://host:port/path, because the consumed length corresponds to
1640 * the first character of the path. If the conversion fails, it returns -1.
1641 *
1642 * This function tries to resolve the DNS name if haproxy is in starting mode.
1643 * So, this function may be used during the configuration parsing.
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001644 */
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001645int url2sa(const char *url, int ulen, struct sockaddr_storage *addr, struct split_url *out)
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001646{
1647 const char *curr = url, *cp = url;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001648 const char *end;
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001649 int ret, url_code = 0;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001650 unsigned long long int http_code = 0;
1651 int default_port;
1652 struct hostent *he;
1653 char *p;
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001654
1655 /* Firstly, try to find :// pattern */
1656 while (curr < url+ulen && url_code != 0x3a2f2f) {
1657 url_code = ((url_code & 0xffff) << 8);
1658 url_code += (unsigned char)*curr++;
1659 }
1660
1661 /* Secondly, if :// pattern is found, verify parsed stuff
1662 * before pattern is matching our http pattern.
1663 * If so parse ip address and port in uri.
1664 *
1665 * WARNING: Current code doesn't support dynamic async dns resolver.
1666 */
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001667 if (url_code != 0x3a2f2f)
1668 return -1;
1669
1670 /* Copy scheme, and utrn to lower case. */
1671 while (cp < curr - 3)
1672 http_code = (http_code << 8) + *cp++;
1673 http_code |= 0x2020202020202020ULL; /* Turn everything to lower case */
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001674
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001675 /* HTTP or HTTPS url matching */
1676 if (http_code == 0x2020202068747470ULL) {
1677 default_port = 80;
1678 if (out)
1679 out->scheme = SCH_HTTP;
1680 }
1681 else if (http_code == 0x2020206874747073ULL) {
1682 default_port = 443;
1683 if (out)
1684 out->scheme = SCH_HTTPS;
1685 }
1686 else
1687 return -1;
1688
1689 /* If the next char is '[', the host address is IPv6. */
1690 if (*curr == '[') {
1691 curr++;
1692
1693 /* Check trash size */
1694 if (trash.size < ulen)
1695 return -1;
1696
1697 /* Look for ']' and copy the address in a trash buffer. */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001698 p = trash.area;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001699 for (end = curr;
1700 end < url + ulen && *end != ']';
1701 end++, p++)
1702 *p = *end;
1703 if (*end != ']')
1704 return -1;
1705 *p = '\0';
1706
1707 /* Update out. */
1708 if (out) {
1709 out->host = curr;
1710 out->host_len = end - curr;
1711 }
1712
1713 /* Try IPv6 decoding. */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001714 if (!inet_pton(AF_INET6, trash.area, &((struct sockaddr_in6 *)addr)->sin6_addr))
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001715 return -1;
1716 end++;
1717
1718 /* Decode port. */
William Lallemand3d7a9182022-03-25 17:37:51 +01001719 if (end < url + ulen && *end == ':') {
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001720 end++;
1721 default_port = read_uint(&end, url + ulen);
1722 }
1723 ((struct sockaddr_in6 *)addr)->sin6_port = htons(default_port);
1724 ((struct sockaddr_in6 *)addr)->sin6_family = AF_INET6;
1725 return end - url;
1726 }
1727 else {
William Lallemand8a913742022-02-18 16:13:12 +01001728 /* we need to copy the string into the trash because url2ipv4
1729 * needs a \0 at the end of the string */
1730 if (trash.size < ulen)
1731 return -1;
1732
1733 memcpy(trash.area, curr, ulen - (curr - url));
1734 trash.area[ulen - (curr - url)] = '\0';
1735
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001736 /* We are looking for IP address. If you want to parse and
1737 * resolve hostname found in url, you can use str2sa_range(), but
1738 * be warned this can slow down global daemon performances
1739 * while handling lagging dns responses.
1740 */
William Lallemand8a913742022-02-18 16:13:12 +01001741 ret = url2ipv4(trash.area, &((struct sockaddr_in *)addr)->sin_addr);
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001742 if (ret) {
1743 /* Update out. */
1744 if (out) {
1745 out->host = curr;
1746 out->host_len = ret;
1747 }
1748
William Lallemandb938b772022-03-24 21:59:03 +01001749 curr += ret;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001750
1751 /* Decode port. */
William Lallemand3d7a9182022-03-25 17:37:51 +01001752 if (curr < url + ulen && *curr == ':') {
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001753 curr++;
1754 default_port = read_uint(&curr, url + ulen);
1755 }
1756 ((struct sockaddr_in *)addr)->sin_port = htons(default_port);
1757
1758 /* Set family. */
1759 ((struct sockaddr_in *)addr)->sin_family = AF_INET;
1760 return curr - url;
1761 }
1762 else if (global.mode & MODE_STARTING) {
1763 /* The IPv4 and IPv6 decoding fails, maybe the url contain name. Try to execute
1764 * synchronous DNS request only if HAProxy is in the start state.
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001765 */
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001766
1767 /* look for : or / or end */
1768 for (end = curr;
1769 end < url + ulen && *end != '/' && *end != ':';
1770 end++);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001771 memcpy(trash.area, curr, end - curr);
1772 trash.area[end - curr] = '\0';
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001773
1774 /* try to resolve an IPv4/IPv6 hostname */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001775 he = gethostbyname(trash.area);
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001776 if (!he)
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001777 return -1;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001778
1779 /* Update out. */
1780 if (out) {
1781 out->host = curr;
1782 out->host_len = end - curr;
1783 }
1784
1785 /* Decode port. */
William Lallemand3d7a9182022-03-25 17:37:51 +01001786 if (end < url + ulen && *end == ':') {
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001787 end++;
1788 default_port = read_uint(&end, url + ulen);
1789 }
1790
1791 /* Copy IP address, set port and family. */
1792 switch (he->h_addrtype) {
1793 case AF_INET:
1794 ((struct sockaddr_in *)addr)->sin_addr = *(struct in_addr *) *(he->h_addr_list);
1795 ((struct sockaddr_in *)addr)->sin_port = htons(default_port);
1796 ((struct sockaddr_in *)addr)->sin_family = AF_INET;
1797 return end - url;
1798
1799 case AF_INET6:
1800 ((struct sockaddr_in6 *)addr)->sin6_addr = *(struct in6_addr *) *(he->h_addr_list);
1801 ((struct sockaddr_in6 *)addr)->sin6_port = htons(default_port);
1802 ((struct sockaddr_in6 *)addr)->sin6_family = AF_INET6;
1803 return end - url;
1804 }
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001805 }
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001806 }
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001807 return -1;
1808}
1809
Willy Tarreau631f01c2011-09-05 00:36:48 +02001810/* Tries to convert a sockaddr_storage address to text form. Upon success, the
1811 * address family is returned so that it's easy for the caller to adapt to the
1812 * output format. Zero is returned if the address family is not supported. -1
1813 * is returned upon error, with errno set. AF_INET, AF_INET6 and AF_UNIX are
1814 * supported.
1815 */
Willy Tarreaud5ec4bf2019-04-25 17:48:16 +02001816int addr_to_str(const struct sockaddr_storage *addr, char *str, int size)
Willy Tarreau631f01c2011-09-05 00:36:48 +02001817{
1818
Willy Tarreaud5ec4bf2019-04-25 17:48:16 +02001819 const void *ptr;
Willy Tarreau631f01c2011-09-05 00:36:48 +02001820
1821 if (size < 5)
1822 return 0;
1823 *str = '\0';
1824
1825 switch (addr->ss_family) {
1826 case AF_INET:
1827 ptr = &((struct sockaddr_in *)addr)->sin_addr;
1828 break;
1829 case AF_INET6:
1830 ptr = &((struct sockaddr_in6 *)addr)->sin6_addr;
1831 break;
1832 case AF_UNIX:
1833 memcpy(str, "unix", 5);
1834 return addr->ss_family;
1835 default:
1836 return 0;
1837 }
1838
1839 if (inet_ntop(addr->ss_family, ptr, str, size))
1840 return addr->ss_family;
1841
1842 /* failed */
1843 return -1;
1844}
1845
Simon Horman75ab8bd2014-06-16 09:39:41 +09001846/* Tries to convert a sockaddr_storage port to text form. Upon success, the
1847 * address family is returned so that it's easy for the caller to adapt to the
1848 * output format. Zero is returned if the address family is not supported. -1
1849 * is returned upon error, with errno set. AF_INET, AF_INET6 and AF_UNIX are
1850 * supported.
1851 */
Willy Tarreaud5ec4bf2019-04-25 17:48:16 +02001852int port_to_str(const struct sockaddr_storage *addr, char *str, int size)
Simon Horman75ab8bd2014-06-16 09:39:41 +09001853{
1854
1855 uint16_t port;
1856
1857
Willy Tarreaud7dad1b2017-01-06 16:46:22 +01001858 if (size < 6)
Simon Horman75ab8bd2014-06-16 09:39:41 +09001859 return 0;
1860 *str = '\0';
1861
1862 switch (addr->ss_family) {
1863 case AF_INET:
1864 port = ((struct sockaddr_in *)addr)->sin_port;
1865 break;
1866 case AF_INET6:
1867 port = ((struct sockaddr_in6 *)addr)->sin6_port;
1868 break;
1869 case AF_UNIX:
1870 memcpy(str, "unix", 5);
1871 return addr->ss_family;
1872 default:
1873 return 0;
1874 }
1875
1876 snprintf(str, size, "%u", ntohs(port));
1877 return addr->ss_family;
1878}
1879
Willy Tarreau16e01562016-08-09 16:46:18 +02001880/* check if the given address is local to the system or not. It will return
1881 * -1 when it's not possible to know, 0 when the address is not local, 1 when
1882 * it is. We don't want to iterate over all interfaces for this (and it is not
1883 * portable). So instead we try to bind in UDP to this address on a free non
1884 * privileged port and to connect to the same address, port 0 (connect doesn't
1885 * care). If it succeeds, we own the address. Note that non-inet addresses are
1886 * considered local since they're most likely AF_UNIX.
1887 */
1888int addr_is_local(const struct netns_entry *ns,
1889 const struct sockaddr_storage *orig)
1890{
1891 struct sockaddr_storage addr;
1892 int result;
1893 int fd;
1894
1895 if (!is_inet_addr(orig))
1896 return 1;
1897
1898 memcpy(&addr, orig, sizeof(addr));
1899 set_host_port(&addr, 0);
1900
1901 fd = my_socketat(ns, addr.ss_family, SOCK_DGRAM, IPPROTO_UDP);
1902 if (fd < 0)
1903 return -1;
1904
1905 result = -1;
1906 if (bind(fd, (struct sockaddr *)&addr, get_addr_len(&addr)) == 0) {
1907 if (connect(fd, (struct sockaddr *)&addr, get_addr_len(&addr)) == -1)
1908 result = 0; // fail, non-local address
1909 else
1910 result = 1; // success, local address
1911 }
1912 else {
1913 if (errno == EADDRNOTAVAIL)
1914 result = 0; // definitely not local :-)
1915 }
1916 close(fd);
1917
1918 return result;
1919}
1920
Willy Tarreaubaaee002006-06-26 02:48:02 +02001921/* will try to encode the string <string> replacing all characters tagged in
1922 * <map> with the hexadecimal representation of their ASCII-code (2 digits)
1923 * prefixed by <escape>, and will store the result between <start> (included)
1924 * and <stop> (excluded), and will always terminate the string with a '\0'
1925 * before <stop>. The position of the '\0' is returned if the conversion
1926 * completes. If bytes are missing between <start> and <stop>, then the
1927 * conversion will be incomplete and truncated. If <stop> <= <start>, the '\0'
1928 * cannot even be stored so we return <start> without writing the 0.
1929 * The input string must also be zero-terminated.
1930 */
1931const char hextab[16] = "0123456789ABCDEF";
1932char *encode_string(char *start, char *stop,
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001933 const char escape, const long *map,
Willy Tarreaubaaee002006-06-26 02:48:02 +02001934 const char *string)
1935{
1936 if (start < stop) {
1937 stop--; /* reserve one byte for the final '\0' */
1938 while (start < stop && *string != '\0') {
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001939 if (!ha_bit_test((unsigned char)(*string), map))
Willy Tarreaubaaee002006-06-26 02:48:02 +02001940 *start++ = *string;
1941 else {
1942 if (start + 3 >= stop)
1943 break;
1944 *start++ = escape;
1945 *start++ = hextab[(*string >> 4) & 15];
1946 *start++ = hextab[*string & 15];
1947 }
1948 string++;
1949 }
1950 *start = '\0';
1951 }
1952 return start;
1953}
1954
Thierry FOURNIERe059ec92014-03-17 12:01:13 +01001955/*
1956 * Same behavior as encode_string() above, except that it encodes chunk
1957 * <chunk> instead of a string.
1958 */
1959char *encode_chunk(char *start, char *stop,
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001960 const char escape, const long *map,
Willy Tarreau83061a82018-07-13 11:56:34 +02001961 const struct buffer *chunk)
Thierry FOURNIERe059ec92014-03-17 12:01:13 +01001962{
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001963 char *str = chunk->area;
1964 char *end = chunk->area + chunk->data;
Thierry FOURNIERe059ec92014-03-17 12:01:13 +01001965
1966 if (start < stop) {
1967 stop--; /* reserve one byte for the final '\0' */
1968 while (start < stop && str < end) {
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001969 if (!ha_bit_test((unsigned char)(*str), map))
Thierry FOURNIERe059ec92014-03-17 12:01:13 +01001970 *start++ = *str;
1971 else {
1972 if (start + 3 >= stop)
1973 break;
1974 *start++ = escape;
1975 *start++ = hextab[(*str >> 4) & 15];
1976 *start++ = hextab[*str & 15];
1977 }
1978 str++;
1979 }
1980 *start = '\0';
1981 }
1982 return start;
1983}
1984
Dragan Dosen0edd1092016-02-12 13:23:02 +01001985/*
1986 * Tries to prefix characters tagged in the <map> with the <escape>
Aurelien DARRAGONc5bff8e2022-09-20 14:33:18 +02001987 * character. The input <string> is processed until string_stop
1988 * is reached or NULL-byte is encountered. The result will
Dragan Dosen1a5d0602016-07-22 16:00:31 +02001989 * be stored between <start> (included) and <stop> (excluded). This
1990 * function will always try to terminate the resulting string with a '\0'
1991 * before <stop>, and will return its position if the conversion
1992 * completes.
1993 */
1994char *escape_string(char *start, char *stop,
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001995 const char escape, const long *map,
Aurelien DARRAGONc5bff8e2022-09-20 14:33:18 +02001996 const char *string, const char *string_stop)
Dragan Dosen1a5d0602016-07-22 16:00:31 +02001997{
1998 if (start < stop) {
1999 stop--; /* reserve one byte for the final '\0' */
Aurelien DARRAGONc5bff8e2022-09-20 14:33:18 +02002000 while (start < stop && string < string_stop && *string != '\0') {
Willy Tarreau1bfd6022019-06-07 11:10:07 +02002001 if (!ha_bit_test((unsigned char)(*string), map))
Dragan Dosen1a5d0602016-07-22 16:00:31 +02002002 *start++ = *string;
2003 else {
2004 if (start + 2 >= stop)
2005 break;
2006 *start++ = escape;
2007 *start++ = *string;
2008 }
2009 string++;
2010 }
2011 *start = '\0';
2012 }
2013 return start;
2014}
2015
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002016/* Check a string for using it in a CSV output format. If the string contains
2017 * one of the following four char <">, <,>, CR or LF, the string is
2018 * encapsulated between <"> and the <"> are escaped by a <""> sequence.
2019 * <str> is the input string to be escaped. The function assumes that
2020 * the input string is null-terminated.
2021 *
2022 * If <quote> is 0, the result is returned escaped but without double quote.
Willy Tarreau898529b2016-01-06 18:07:04 +01002023 * It is useful if the escaped string is used between double quotes in the
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002024 * format.
2025 *
Emeric Brun36e54272024-01-23 15:44:32 +01002026 * printf("..., \"%s\", ...\r\n", csv_enc(str, 0, 0, &trash));
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002027 *
Willy Tarreaub631c292016-01-08 10:04:08 +01002028 * If <quote> is 1, the converter puts the quotes only if any reserved character
2029 * is present. If <quote> is 2, the converter always puts the quotes.
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002030 *
Emeric Brun36e54272024-01-23 15:44:32 +01002031 * If <oneline> is not 0, CRs are skipped and LFs are replaced by spaces.
2032 * This re-format multi-lines strings to only one line. The purpose is to
2033 * allow a line by line parsing but also to keep the output compliant with
2034 * the CLI witch uses LF to defines the end of the response.
2035 *
2036 * If <oneline> is 2, In addition to previous action, the trailing spaces are
2037 * removed.
2038 *
Willy Tarreau83061a82018-07-13 11:56:34 +02002039 * <output> is a struct buffer used for storing the output string.
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002040 *
Willy Tarreau898529b2016-01-06 18:07:04 +01002041 * The function returns the converted string on its output. If an error
2042 * occurs, the function returns an empty string. This type of output is useful
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002043 * for using the function directly as printf() argument.
2044 *
2045 * If the output buffer is too short to contain the input string, the result
2046 * is truncated.
Willy Tarreau898529b2016-01-06 18:07:04 +01002047 *
Willy Tarreaub631c292016-01-08 10:04:08 +01002048 * This function appends the encoding to the existing output chunk, and it
2049 * guarantees that it starts immediately at the first available character of
2050 * the chunk. Please use csv_enc() instead if you want to replace the output
2051 * chunk.
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002052 */
Emeric Brun36e54272024-01-23 15:44:32 +01002053const char *csv_enc_append(const char *str, int quote, int oneline, struct buffer *output)
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002054{
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002055 char *end = output->area + output->size;
2056 char *out = output->area + output->data;
Willy Tarreau898529b2016-01-06 18:07:04 +01002057 char *ptr = out;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002058
Willy Tarreaub631c292016-01-08 10:04:08 +01002059 if (quote == 1) {
2060 /* automatic quoting: first verify if we'll have to quote the string */
2061 if (!strpbrk(str, "\n\r,\""))
2062 quote = 0;
2063 }
2064
2065 if (quote)
2066 *ptr++ = '"';
2067
Willy Tarreau898529b2016-01-06 18:07:04 +01002068 while (*str && ptr < end - 2) { /* -2 for reserving space for <"> and \0. */
Emeric Brun36e54272024-01-23 15:44:32 +01002069 if (oneline) {
2070 if (*str == '\n' ) {
2071 /* replace LF by a space */
2072 *ptr++ = ' ';
2073 str++;
2074 continue;
2075 }
2076 else if (*str == '\r' ) {
2077 /* skip CR */
2078 str++;
2079 continue;
2080 }
2081 }
Willy Tarreau898529b2016-01-06 18:07:04 +01002082 *ptr = *str;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002083 if (*str == '"') {
Willy Tarreau898529b2016-01-06 18:07:04 +01002084 ptr++;
2085 if (ptr >= end - 2) {
2086 ptr--;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002087 break;
2088 }
Willy Tarreau898529b2016-01-06 18:07:04 +01002089 *ptr = '"';
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002090 }
Willy Tarreau898529b2016-01-06 18:07:04 +01002091 ptr++;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002092 str++;
2093 }
2094
Emeric Brun36e54272024-01-23 15:44:32 +01002095 if (oneline == 2) {
2096 /* remove trailing spaces */
2097 while (ptr > out && *(ptr - 1) == ' ')
2098 ptr--;
2099 }
2100
Willy Tarreaub631c292016-01-08 10:04:08 +01002101 if (quote)
2102 *ptr++ = '"';
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002103
Willy Tarreau898529b2016-01-06 18:07:04 +01002104 *ptr = '\0';
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002105 output->data = ptr - output->area;
Willy Tarreau898529b2016-01-06 18:07:04 +01002106 return out;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002107}
2108
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002109/* Decode an URL-encoded string in-place. The resulting string might
2110 * be shorter. If some forbidden characters are found, the conversion is
Thierry FOURNIER5068d962013-10-04 16:27:27 +02002111 * aborted, the string is truncated before the issue and a negative value is
2112 * returned, otherwise the operation returns the length of the decoded string.
Willy Tarreau62ba9ba2020-04-23 17:54:47 +02002113 * If the 'in_form' argument is non-nul the string is assumed to be part of
2114 * an "application/x-www-form-urlencoded" encoded string, and the '+' will be
2115 * turned to a space. If it's zero, this will only be done after a question
2116 * mark ('?').
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002117 */
Willy Tarreau62ba9ba2020-04-23 17:54:47 +02002118int url_decode(char *string, int in_form)
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002119{
2120 char *in, *out;
Thierry FOURNIER5068d962013-10-04 16:27:27 +02002121 int ret = -1;
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002122
2123 in = string;
2124 out = string;
2125 while (*in) {
2126 switch (*in) {
2127 case '+' :
Willy Tarreau62ba9ba2020-04-23 17:54:47 +02002128 *out++ = in_form ? ' ' : *in;
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002129 break;
2130 case '%' :
2131 if (!ishex(in[1]) || !ishex(in[2]))
2132 goto end;
2133 *out++ = (hex2i(in[1]) << 4) + hex2i(in[2]);
2134 in += 2;
2135 break;
Willy Tarreau62ba9ba2020-04-23 17:54:47 +02002136 case '?':
2137 in_form = 1;
Willy Tarreau7de8de02022-11-14 07:20:09 +01002138 __fallthrough;
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002139 default:
2140 *out++ = *in;
2141 break;
2142 }
2143 in++;
2144 }
Thierry FOURNIER5068d962013-10-04 16:27:27 +02002145 ret = out - string; /* success */
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002146 end:
2147 *out = 0;
2148 return ret;
2149}
Willy Tarreaubaaee002006-06-26 02:48:02 +02002150
Willy Tarreau6911fa42007-03-04 18:06:08 +01002151unsigned int str2ui(const char *s)
2152{
2153 return __str2ui(s);
2154}
2155
2156unsigned int str2uic(const char *s)
2157{
2158 return __str2uic(s);
2159}
2160
2161unsigned int strl2ui(const char *s, int len)
2162{
2163 return __strl2ui(s, len);
2164}
2165
2166unsigned int strl2uic(const char *s, int len)
2167{
2168 return __strl2uic(s, len);
2169}
2170
Willy Tarreau4ec83cd2010-10-15 23:19:55 +02002171unsigned int read_uint(const char **s, const char *end)
2172{
2173 return __read_uint(s, end);
2174}
2175
Thierry FOURNIER763a5d82015-07-06 23:09:52 +02002176/* This function reads an unsigned integer from the string pointed to by <s> and
2177 * returns it. The <s> pointer is adjusted to point to the first unread char. The
2178 * function automatically stops at <end>. If the number overflows, the 2^64-1
2179 * value is returned.
2180 */
2181unsigned long long int read_uint64(const char **s, const char *end)
2182{
2183 const char *ptr = *s;
2184 unsigned long long int i = 0, tmp;
2185 unsigned int j;
2186
2187 while (ptr < end) {
2188
2189 /* read next char */
2190 j = *ptr - '0';
2191 if (j > 9)
2192 goto read_uint64_end;
2193
2194 /* add char to the number and check overflow. */
2195 tmp = i * 10;
2196 if (tmp / 10 != i) {
2197 i = ULLONG_MAX;
2198 goto read_uint64_eat;
2199 }
2200 if (ULLONG_MAX - tmp < j) {
2201 i = ULLONG_MAX;
2202 goto read_uint64_eat;
2203 }
2204 i = tmp + j;
2205 ptr++;
2206 }
2207read_uint64_eat:
2208 /* eat each numeric char */
2209 while (ptr < end) {
2210 if ((unsigned int)(*ptr - '0') > 9)
2211 break;
2212 ptr++;
2213 }
2214read_uint64_end:
2215 *s = ptr;
2216 return i;
2217}
2218
2219/* This function reads an integer from the string pointed to by <s> and returns
2220 * it. The <s> pointer is adjusted to point to the first unread char. The function
2221 * automatically stops at <end>. Il the number is bigger than 2^63-2, the 2^63-1
2222 * value is returned. If the number is lowest than -2^63-1, the -2^63 value is
2223 * returned.
2224 */
2225long long int read_int64(const char **s, const char *end)
2226{
2227 unsigned long long int i = 0;
2228 int neg = 0;
2229
2230 /* Look for minus char. */
2231 if (**s == '-') {
2232 neg = 1;
2233 (*s)++;
2234 }
2235 else if (**s == '+')
2236 (*s)++;
2237
2238 /* convert as positive number. */
2239 i = read_uint64(s, end);
2240
2241 if (neg) {
2242 if (i > 0x8000000000000000ULL)
2243 return LLONG_MIN;
2244 return -i;
2245 }
2246 if (i > 0x7fffffffffffffffULL)
2247 return LLONG_MAX;
2248 return i;
2249}
2250
Willy Tarreau6911fa42007-03-04 18:06:08 +01002251/* This one is 7 times faster than strtol() on athlon with checks.
2252 * It returns the value of the number composed of all valid digits read,
2253 * and can process negative numbers too.
2254 */
2255int strl2ic(const char *s, int len)
2256{
2257 int i = 0;
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002258 int j, k;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002259
2260 if (len > 0) {
2261 if (*s != '-') {
2262 /* positive number */
2263 while (len-- > 0) {
2264 j = (*s++) - '0';
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002265 k = i * 10;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002266 if (j > 9)
2267 break;
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002268 i = k + j;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002269 }
2270 } else {
2271 /* negative number */
2272 s++;
2273 while (--len > 0) {
2274 j = (*s++) - '0';
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002275 k = i * 10;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002276 if (j > 9)
2277 break;
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002278 i = k - j;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002279 }
2280 }
2281 }
2282 return i;
2283}
2284
2285
2286/* This function reads exactly <len> chars from <s> and converts them to a
2287 * signed integer which it stores into <ret>. It accurately detects any error
2288 * (truncated string, invalid chars, overflows). It is meant to be used in
2289 * applications designed for hostile environments. It returns zero when the
2290 * number has successfully been converted, non-zero otherwise. When an error
2291 * is returned, the <ret> value is left untouched. It is yet 5 to 40 times
2292 * faster than strtol().
2293 */
2294int strl2irc(const char *s, int len, int *ret)
2295{
2296 int i = 0;
2297 int j;
2298
2299 if (!len)
2300 return 1;
2301
2302 if (*s != '-') {
2303 /* positive number */
2304 while (len-- > 0) {
2305 j = (*s++) - '0';
2306 if (j > 9) return 1; /* invalid char */
2307 if (i > INT_MAX / 10) return 1; /* check for multiply overflow */
2308 i = i * 10;
2309 if (i + j < i) return 1; /* check for addition overflow */
2310 i = i + j;
2311 }
2312 } else {
2313 /* negative number */
2314 s++;
2315 while (--len > 0) {
2316 j = (*s++) - '0';
2317 if (j > 9) return 1; /* invalid char */
2318 if (i < INT_MIN / 10) return 1; /* check for multiply overflow */
2319 i = i * 10;
2320 if (i - j > i) return 1; /* check for subtract overflow */
2321 i = i - j;
2322 }
2323 }
2324 *ret = i;
2325 return 0;
2326}
2327
2328
2329/* This function reads exactly <len> chars from <s> and converts them to a
2330 * signed integer which it stores into <ret>. It accurately detects any error
2331 * (truncated string, invalid chars, overflows). It is meant to be used in
2332 * applications designed for hostile environments. It returns zero when the
2333 * number has successfully been converted, non-zero otherwise. When an error
2334 * is returned, the <ret> value is left untouched. It is about 3 times slower
William Dauchy060ffc82021-02-06 20:47:51 +01002335 * than strl2irc().
Willy Tarreau6911fa42007-03-04 18:06:08 +01002336 */
Willy Tarreau6911fa42007-03-04 18:06:08 +01002337
2338int strl2llrc(const char *s, int len, long long *ret)
2339{
2340 long long i = 0;
2341 int j;
2342
2343 if (!len)
2344 return 1;
2345
2346 if (*s != '-') {
2347 /* positive number */
2348 while (len-- > 0) {
2349 j = (*s++) - '0';
2350 if (j > 9) return 1; /* invalid char */
2351 if (i > LLONG_MAX / 10LL) return 1; /* check for multiply overflow */
2352 i = i * 10LL;
2353 if (i + j < i) return 1; /* check for addition overflow */
2354 i = i + j;
2355 }
2356 } else {
2357 /* negative number */
2358 s++;
2359 while (--len > 0) {
2360 j = (*s++) - '0';
2361 if (j > 9) return 1; /* invalid char */
2362 if (i < LLONG_MIN / 10LL) return 1; /* check for multiply overflow */
2363 i = i * 10LL;
2364 if (i - j > i) return 1; /* check for subtract overflow */
2365 i = i - j;
2366 }
2367 }
2368 *ret = i;
2369 return 0;
2370}
2371
Thierry FOURNIER511e9472014-01-23 17:40:34 +01002372/* This function is used with pat_parse_dotted_ver(). It converts a string
2373 * composed by two number separated by a dot. Each part must contain in 16 bits
2374 * because internally they will be represented as a 32-bit quantity stored in
2375 * a 64-bit integer. It returns zero when the number has successfully been
2376 * converted, non-zero otherwise. When an error is returned, the <ret> value
2377 * is left untouched.
2378 *
2379 * "1.3" -> 0x0000000000010003
2380 * "65535.65535" -> 0x00000000ffffffff
2381 */
2382int strl2llrc_dotted(const char *text, int len, long long *ret)
2383{
2384 const char *end = &text[len];
2385 const char *p;
2386 long long major, minor;
2387
2388 /* Look for dot. */
2389 for (p = text; p < end; p++)
2390 if (*p == '.')
2391 break;
2392
2393 /* Convert major. */
2394 if (strl2llrc(text, p - text, &major) != 0)
2395 return 1;
2396
2397 /* Check major. */
2398 if (major >= 65536)
2399 return 1;
2400
2401 /* Convert minor. */
2402 minor = 0;
2403 if (p < end)
2404 if (strl2llrc(p + 1, end - (p + 1), &minor) != 0)
2405 return 1;
2406
2407 /* Check minor. */
2408 if (minor >= 65536)
2409 return 1;
2410
2411 /* Compose value. */
2412 *ret = (major << 16) | (minor & 0xffff);
2413 return 0;
2414}
2415
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002416/* This function parses a time value optionally followed by a unit suffix among
2417 * "d", "h", "m", "s", "ms" or "us". It converts the value into the unit
2418 * expected by the caller. The computation does its best to avoid overflows.
2419 * The value is returned in <ret> if everything is fine, and a NULL is returned
2420 * by the function. In case of error, a pointer to the error is returned and
2421 * <ret> is left untouched. Values are automatically rounded up when needed.
Willy Tarreau9faebe32019-06-07 19:00:37 +02002422 * Values resulting in values larger than or equal to 2^31 after conversion are
2423 * reported as an overflow as value PARSE_TIME_OVER. Non-null values resulting
2424 * in an underflow are reported as an underflow as value PARSE_TIME_UNDER.
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002425 */
2426const char *parse_time_err(const char *text, unsigned *ret, unsigned unit_flags)
2427{
Willy Tarreau9faebe32019-06-07 19:00:37 +02002428 unsigned long long imult, idiv;
2429 unsigned long long omult, odiv;
2430 unsigned long long value, result;
Christopher Fauletc20ad0d2020-12-11 09:23:07 +01002431 const char *str = text;
2432
2433 if (!isdigit((unsigned char)*text))
2434 return text;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002435
2436 omult = odiv = 1;
2437
2438 switch (unit_flags & TIME_UNIT_MASK) {
2439 case TIME_UNIT_US: omult = 1000000; break;
2440 case TIME_UNIT_MS: omult = 1000; break;
2441 case TIME_UNIT_S: break;
2442 case TIME_UNIT_MIN: odiv = 60; break;
2443 case TIME_UNIT_HOUR: odiv = 3600; break;
2444 case TIME_UNIT_DAY: odiv = 86400; break;
2445 default: break;
2446 }
2447
2448 value = 0;
2449
2450 while (1) {
2451 unsigned int j;
2452
2453 j = *text - '0';
2454 if (j > 9)
2455 break;
2456 text++;
2457 value *= 10;
2458 value += j;
2459 }
2460
2461 imult = idiv = 1;
2462 switch (*text) {
2463 case '\0': /* no unit = default unit */
2464 imult = omult = idiv = odiv = 1;
Christopher Fauletc20ad0d2020-12-11 09:23:07 +01002465 goto end;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002466 case 's': /* second = unscaled unit */
2467 break;
2468 case 'u': /* microsecond : "us" */
2469 if (text[1] == 's') {
2470 idiv = 1000000;
2471 text++;
Thayne McCombsa6838052021-04-02 14:12:43 -06002472 break;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002473 }
Christopher Fauletc20ad0d2020-12-11 09:23:07 +01002474 return text;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002475 case 'm': /* millisecond : "ms" or minute: "m" */
2476 if (text[1] == 's') {
2477 idiv = 1000;
2478 text++;
2479 } else
2480 imult = 60;
2481 break;
2482 case 'h': /* hour : "h" */
2483 imult = 3600;
2484 break;
2485 case 'd': /* day : "d" */
2486 imult = 86400;
2487 break;
2488 default:
2489 return text;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002490 }
Christopher Fauletc20ad0d2020-12-11 09:23:07 +01002491 if (*(++text) != '\0') {
2492 ha_warning("unexpected character '%c' after the timer value '%s', only "
2493 "(us=microseconds,ms=milliseconds,s=seconds,m=minutes,h=hours,d=days) are supported."
2494 " This will be reported as an error in next versions.\n", *text, str);
2495 }
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002496
Christopher Fauletc20ad0d2020-12-11 09:23:07 +01002497 end:
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002498 if (omult % idiv == 0) { omult /= idiv; idiv = 1; }
2499 if (idiv % omult == 0) { idiv /= omult; omult = 1; }
2500 if (imult % odiv == 0) { imult /= odiv; odiv = 1; }
2501 if (odiv % imult == 0) { odiv /= imult; imult = 1; }
2502
Willy Tarreau9faebe32019-06-07 19:00:37 +02002503 result = (value * (imult * omult) + (idiv * odiv - 1)) / (idiv * odiv);
2504 if (result >= 0x80000000)
2505 return PARSE_TIME_OVER;
2506 if (!result && value)
2507 return PARSE_TIME_UNDER;
2508 *ret = result;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002509 return NULL;
2510}
Willy Tarreau6911fa42007-03-04 18:06:08 +01002511
Emeric Brun39132b22010-01-04 14:57:24 +01002512/* this function converts the string starting at <text> to an unsigned int
2513 * stored in <ret>. If an error is detected, the pointer to the unexpected
Joseph Herlant32b83272018-11-15 11:58:28 -08002514 * character is returned. If the conversion is successful, NULL is returned.
Emeric Brun39132b22010-01-04 14:57:24 +01002515 */
2516const char *parse_size_err(const char *text, unsigned *ret) {
2517 unsigned value = 0;
2518
Christopher Faulet82635a02020-12-11 09:30:45 +01002519 if (!isdigit((unsigned char)*text))
2520 return text;
2521
Emeric Brun39132b22010-01-04 14:57:24 +01002522 while (1) {
2523 unsigned int j;
2524
2525 j = *text - '0';
2526 if (j > 9)
2527 break;
2528 if (value > ~0U / 10)
2529 return text;
2530 value *= 10;
2531 if (value > (value + j))
2532 return text;
2533 value += j;
2534 text++;
2535 }
2536
2537 switch (*text) {
2538 case '\0':
2539 break;
2540 case 'K':
2541 case 'k':
2542 if (value > ~0U >> 10)
2543 return text;
2544 value = value << 10;
2545 break;
2546 case 'M':
2547 case 'm':
2548 if (value > ~0U >> 20)
2549 return text;
2550 value = value << 20;
2551 break;
2552 case 'G':
2553 case 'g':
2554 if (value > ~0U >> 30)
2555 return text;
2556 value = value << 30;
2557 break;
2558 default:
2559 return text;
2560 }
2561
Godbach58048a22015-01-28 17:36:16 +08002562 if (*text != '\0' && *++text != '\0')
2563 return text;
2564
Emeric Brun39132b22010-01-04 14:57:24 +01002565 *ret = value;
2566 return NULL;
2567}
2568
Willy Tarreau126d4062013-12-03 17:50:47 +01002569/*
2570 * Parse binary string written in hexadecimal (source) and store the decoded
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05002571 * result into binstr and set binstrlen to the length of binstr. Memory for
Willy Tarreau126d4062013-12-03 17:50:47 +01002572 * binstr is allocated by the function. In case of error, returns 0 with an
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05002573 * error message in err. In success case, it returns the consumed length.
Willy Tarreau126d4062013-12-03 17:50:47 +01002574 */
2575int parse_binary(const char *source, char **binstr, int *binstrlen, char **err)
2576{
2577 int len;
2578 const char *p = source;
2579 int i,j;
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002580 int alloc;
Willy Tarreau126d4062013-12-03 17:50:47 +01002581
2582 len = strlen(source);
2583 if (len % 2) {
2584 memprintf(err, "an even number of hex digit is expected");
2585 return 0;
2586 }
2587
2588 len = len >> 1;
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002589
Willy Tarreau126d4062013-12-03 17:50:47 +01002590 if (!*binstr) {
Tim Duesterhuse52b6e52020-09-12 20:26:43 +02002591 *binstr = calloc(len, sizeof(**binstr));
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002592 if (!*binstr) {
2593 memprintf(err, "out of memory while loading string pattern");
2594 return 0;
2595 }
2596 alloc = 1;
Willy Tarreau126d4062013-12-03 17:50:47 +01002597 }
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002598 else {
2599 if (*binstrlen < len) {
Joseph Herlant76dbe782018-11-15 12:01:22 -08002600 memprintf(err, "no space available in the buffer. expect %d, provides %d",
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002601 len, *binstrlen);
2602 return 0;
2603 }
2604 alloc = 0;
2605 }
2606 *binstrlen = len;
Willy Tarreau126d4062013-12-03 17:50:47 +01002607
2608 i = j = 0;
2609 while (j < len) {
2610 if (!ishex(p[i++]))
2611 goto bad_input;
2612 if (!ishex(p[i++]))
2613 goto bad_input;
2614 (*binstr)[j++] = (hex2i(p[i-2]) << 4) + hex2i(p[i-1]);
2615 }
Thierry FOURNIERee330af2014-01-21 11:36:14 +01002616 return len << 1;
Willy Tarreau126d4062013-12-03 17:50:47 +01002617
2618bad_input:
2619 memprintf(err, "an hex digit is expected (found '%c')", p[i-1]);
Willy Tarreau61cfdf42021-02-20 10:46:51 +01002620 if (alloc)
2621 ha_free(binstr);
Willy Tarreau126d4062013-12-03 17:50:47 +01002622 return 0;
2623}
2624
Willy Tarreau946ba592009-05-10 15:41:18 +02002625/* copies at most <n> characters from <src> and always terminates with '\0' */
2626char *my_strndup(const char *src, int n)
2627{
2628 int len = 0;
2629 char *ret;
2630
2631 while (len < n && src[len])
2632 len++;
2633
Vincent Bernat3c2f2f22016-04-03 13:48:42 +02002634 ret = malloc(len + 1);
Willy Tarreau946ba592009-05-10 15:41:18 +02002635 if (!ret)
2636 return ret;
2637 memcpy(ret, src, len);
2638 ret[len] = '\0';
2639 return ret;
2640}
2641
Baptiste Assmannbb77c8e2013-10-06 23:24:13 +02002642/*
2643 * search needle in haystack
2644 * returns the pointer if found, returns NULL otherwise
2645 */
2646const void *my_memmem(const void *haystack, size_t haystacklen, const void *needle, size_t needlelen)
2647{
2648 const void *c = NULL;
2649 unsigned char f;
2650
2651 if ((haystack == NULL) || (needle == NULL) || (haystacklen < needlelen))
2652 return NULL;
2653
2654 f = *(char *)needle;
2655 c = haystack;
2656 while ((c = memchr(c, f, haystacklen - (c - haystack))) != NULL) {
2657 if ((haystacklen - (c - haystack)) < needlelen)
2658 return NULL;
2659
2660 if (memcmp(c, needle, needlelen) == 0)
2661 return c;
2662 ++c;
2663 }
2664 return NULL;
2665}
2666
Ilya Shipitsinc02a23f2020-05-06 00:53:22 +05002667/* get length of the initial segment consisting entirely of bytes in <accept> */
Christopher Faulet5eb96cb2020-04-15 10:23:01 +02002668size_t my_memspn(const void *str, size_t len, const void *accept, size_t acceptlen)
2669{
2670 size_t ret = 0;
2671
2672 while (ret < len && memchr(accept, *((int *)str), acceptlen)) {
2673 str++;
2674 ret++;
2675 }
2676 return ret;
2677}
2678
Ilya Shipitsinc02a23f2020-05-06 00:53:22 +05002679/* get length of the initial segment consisting entirely of bytes not in <rejcet> */
Christopher Faulet5eb96cb2020-04-15 10:23:01 +02002680size_t my_memcspn(const void *str, size_t len, const void *reject, size_t rejectlen)
2681{
2682 size_t ret = 0;
2683
2684 while (ret < len) {
2685 if(memchr(reject, *((int *)str), rejectlen))
2686 return ret;
2687 str++;
2688 ret++;
2689 }
2690 return ret;
2691}
2692
Willy Tarreau482b00d2009-10-04 22:48:42 +02002693/* This function returns the first unused key greater than or equal to <key> in
2694 * ID tree <root>. Zero is returned if no place is found.
2695 */
2696unsigned int get_next_id(struct eb_root *root, unsigned int key)
2697{
2698 struct eb32_node *used;
2699
2700 do {
2701 used = eb32_lookup_ge(root, key);
2702 if (!used || used->key > key)
2703 return key; /* key is available */
2704 key++;
2705 } while (key);
2706 return key;
2707}
2708
Willy Tarreau9c1e15d2017-11-15 18:51:29 +01002709/* dump the full tree to <file> in DOT format for debugging purposes. Will
2710 * optionally highlight node <subj> if found, depending on operation <op> :
2711 * 0 : nothing
2712 * >0 : insertion, node/leaf are surrounded in red
2713 * <0 : removal, node/leaf are dashed with no background
2714 * Will optionally add "desc" as a label on the graph if set and non-null.
2715 */
2716void 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 +01002717{
2718 struct eb32sc_node *node;
2719 unsigned long scope = -1;
2720
Willy Tarreau9c1e15d2017-11-15 18:51:29 +01002721 fprintf(file, "digraph ebtree {\n");
2722
2723 if (desc && *desc) {
2724 fprintf(file,
2725 " fontname=\"fixed\";\n"
2726 " fontsize=8;\n"
2727 " label=\"%s\";\n", desc);
2728 }
2729
Willy Tarreaued3cda02017-11-15 15:04:05 +01002730 fprintf(file,
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002731 " node [fontname=\"fixed\" fontsize=8 shape=\"box\" style=\"filled\" color=\"black\" fillcolor=\"white\"];\n"
2732 " edge [fontname=\"fixed\" fontsize=8 style=\"solid\" color=\"magenta\" dir=\"forward\"];\n"
Willy Tarreaued3cda02017-11-15 15:04:05 +01002733 " \"%lx_n\" [label=\"root\\n%lx\"]\n", (long)eb_root_to_node(root), (long)root
2734 );
2735
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002736 fprintf(file, " \"%lx_n\" -> \"%lx_%c\" [taillabel=\"L\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002737 (long)eb_root_to_node(root),
2738 (long)eb_root_to_node(eb_clrtag(root->b[0])),
Willy Tarreaued3cda02017-11-15 15:04:05 +01002739 eb_gettag(root->b[0]) == EB_LEAF ? 'l' : 'n');
2740
2741 node = eb32sc_first(root, scope);
2742 while (node) {
2743 if (node->node.node_p) {
2744 /* node part is used */
Willy Tarreau9c1e15d2017-11-15 18:51:29 +01002745 fprintf(file, " \"%lx_n\" [label=\"%lx\\nkey=%u\\nscope=%lx\\nbit=%d\" fillcolor=\"lightskyblue1\" %s];\n",
2746 (long)node, (long)node, node->key, node->node_s, node->node.bit,
2747 (node == subj) ? (op < 0 ? "color=\"red\" style=\"dashed\"" : op > 0 ? "color=\"red\"" : "") : "");
Willy Tarreaued3cda02017-11-15 15:04:05 +01002748
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002749 fprintf(file, " \"%lx_n\" -> \"%lx_n\" [taillabel=\"%c\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002750 (long)node,
2751 (long)eb_root_to_node(eb_clrtag(node->node.node_p)),
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002752 eb_gettag(node->node.node_p) ? 'R' : 'L');
Willy Tarreaued3cda02017-11-15 15:04:05 +01002753
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002754 fprintf(file, " \"%lx_n\" -> \"%lx_%c\" [taillabel=\"L\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002755 (long)node,
2756 (long)eb_root_to_node(eb_clrtag(node->node.branches.b[0])),
Willy Tarreaued3cda02017-11-15 15:04:05 +01002757 eb_gettag(node->node.branches.b[0]) == EB_LEAF ? 'l' : 'n');
2758
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002759 fprintf(file, " \"%lx_n\" -> \"%lx_%c\" [taillabel=\"R\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002760 (long)node,
2761 (long)eb_root_to_node(eb_clrtag(node->node.branches.b[1])),
Willy Tarreaued3cda02017-11-15 15:04:05 +01002762 eb_gettag(node->node.branches.b[1]) == EB_LEAF ? 'l' : 'n');
2763 }
2764
Willy Tarreau9c1e15d2017-11-15 18:51:29 +01002765 fprintf(file, " \"%lx_l\" [label=\"%lx\\nkey=%u\\nscope=%lx\\npfx=%u\" fillcolor=\"yellow\" %s];\n",
2766 (long)node, (long)node, node->key, node->leaf_s, node->node.pfx,
2767 (node == subj) ? (op < 0 ? "color=\"red\" style=\"dashed\"" : op > 0 ? "color=\"red\"" : "") : "");
Willy Tarreaued3cda02017-11-15 15:04:05 +01002768
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002769 fprintf(file, " \"%lx_l\" -> \"%lx_n\" [taillabel=\"%c\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002770 (long)node,
2771 (long)eb_root_to_node(eb_clrtag(node->node.leaf_p)),
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002772 eb_gettag(node->node.leaf_p) ? 'R' : 'L');
Willy Tarreaued3cda02017-11-15 15:04:05 +01002773 node = eb32sc_next(node, scope);
2774 }
2775 fprintf(file, "}\n");
2776}
2777
Willy Tarreau0dc9e6d2022-08-01 11:55:57 +02002778/* dump the full tree to <file> in DOT format for debugging purposes. Will
2779 * optionally highlight node <subj> if found, depending on operation <op> :
2780 * 0 : nothing
2781 * >0 : insertion, node/leaf are surrounded in red
2782 * <0 : removal, node/leaf are dashed with no background
2783 * Will optionally add "desc" as a label on the graph if set and non-null. The
2784 * key is printed as a u32 hex value. A full-sized hex dump would be better but
2785 * is left to be implemented.
2786 */
2787void ebmb_to_file(FILE *file, struct eb_root *root, const struct ebmb_node *subj, int op, const char *desc)
2788{
2789 struct ebmb_node *node;
2790
2791 fprintf(file, "digraph ebtree {\n");
2792
2793 if (desc && *desc) {
2794 fprintf(file,
2795 " fontname=\"fixed\";\n"
2796 " fontsize=8;\n"
2797 " label=\"%s\";\n", desc);
2798 }
2799
2800 fprintf(file,
2801 " node [fontname=\"fixed\" fontsize=8 shape=\"box\" style=\"filled\" color=\"black\" fillcolor=\"white\"];\n"
2802 " edge [fontname=\"fixed\" fontsize=8 style=\"solid\" color=\"magenta\" dir=\"forward\"];\n"
2803 " \"%lx_n\" [label=\"root\\n%lx\"]\n", (long)eb_root_to_node(root), (long)root
2804 );
2805
2806 fprintf(file, " \"%lx_n\" -> \"%lx_%c\" [taillabel=\"L\"];\n",
2807 (long)eb_root_to_node(root),
2808 (long)eb_root_to_node(eb_clrtag(root->b[0])),
2809 eb_gettag(root->b[0]) == EB_LEAF ? 'l' : 'n');
2810
2811 node = ebmb_first(root);
2812 while (node) {
2813 if (node->node.node_p) {
2814 /* node part is used */
2815 fprintf(file, " \"%lx_n\" [label=\"%lx\\nkey=%#x\\nbit=%d\" fillcolor=\"lightskyblue1\" %s];\n",
2816 (long)node, (long)node, read_u32(node->key), node->node.bit,
2817 (node == subj) ? (op < 0 ? "color=\"red\" style=\"dashed\"" : op > 0 ? "color=\"red\"" : "") : "");
2818
2819 fprintf(file, " \"%lx_n\" -> \"%lx_n\" [taillabel=\"%c\"];\n",
2820 (long)node,
2821 (long)eb_root_to_node(eb_clrtag(node->node.node_p)),
2822 eb_gettag(node->node.node_p) ? 'R' : 'L');
2823
2824 fprintf(file, " \"%lx_n\" -> \"%lx_%c\" [taillabel=\"L\"];\n",
2825 (long)node,
2826 (long)eb_root_to_node(eb_clrtag(node->node.branches.b[0])),
2827 eb_gettag(node->node.branches.b[0]) == EB_LEAF ? 'l' : 'n');
2828
2829 fprintf(file, " \"%lx_n\" -> \"%lx_%c\" [taillabel=\"R\"];\n",
2830 (long)node,
2831 (long)eb_root_to_node(eb_clrtag(node->node.branches.b[1])),
2832 eb_gettag(node->node.branches.b[1]) == EB_LEAF ? 'l' : 'n');
2833 }
2834
2835 fprintf(file, " \"%lx_l\" [label=\"%lx\\nkey=%#x\\npfx=%u\" fillcolor=\"yellow\" %s];\n",
2836 (long)node, (long)node, read_u32(node->key), node->node.pfx,
2837 (node == subj) ? (op < 0 ? "color=\"red\" style=\"dashed\"" : op > 0 ? "color=\"red\"" : "") : "");
2838
2839 fprintf(file, " \"%lx_l\" -> \"%lx_n\" [taillabel=\"%c\"];\n",
2840 (long)node,
2841 (long)eb_root_to_node(eb_clrtag(node->node.leaf_p)),
2842 eb_gettag(node->node.leaf_p) ? 'R' : 'L');
2843 node = ebmb_next(node);
2844 }
2845 fprintf(file, "}\n");
2846}
2847
Willy Tarreau348238b2010-01-18 15:05:57 +01002848/* This function compares a sample word possibly followed by blanks to another
2849 * clean word. The compare is case-insensitive. 1 is returned if both are equal,
2850 * otherwise zero. This intends to be used when checking HTTP headers for some
2851 * values. Note that it validates a word followed only by blanks but does not
2852 * validate a word followed by blanks then other chars.
2853 */
2854int word_match(const char *sample, int slen, const char *word, int wlen)
2855{
2856 if (slen < wlen)
2857 return 0;
2858
2859 while (wlen) {
2860 char c = *sample ^ *word;
2861 if (c && c != ('A' ^ 'a'))
2862 return 0;
2863 sample++;
2864 word++;
2865 slen--;
2866 wlen--;
2867 }
2868
2869 while (slen) {
2870 if (*sample != ' ' && *sample != '\t')
2871 return 0;
2872 sample++;
2873 slen--;
2874 }
2875 return 1;
2876}
Willy Tarreau482b00d2009-10-04 22:48:42 +02002877
Willy Tarreaud54bbdc2009-09-07 11:00:31 +02002878/* Converts any text-formatted IPv4 address to a host-order IPv4 address. It
2879 * is particularly fast because it avoids expensive operations such as
2880 * multiplies, which are optimized away at the end. It requires a properly
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05002881 * formatted address though (3 points).
Willy Tarreaud54bbdc2009-09-07 11:00:31 +02002882 */
2883unsigned int inetaddr_host(const char *text)
2884{
2885 const unsigned int ascii_zero = ('0' << 24) | ('0' << 16) | ('0' << 8) | '0';
2886 register unsigned int dig100, dig10, dig1;
2887 int s;
2888 const char *p, *d;
2889
2890 dig1 = dig10 = dig100 = ascii_zero;
2891 s = 24;
2892
2893 p = text;
2894 while (1) {
2895 if (((unsigned)(*p - '0')) <= 9) {
2896 p++;
2897 continue;
2898 }
2899
2900 /* here, we have a complete byte between <text> and <p> (exclusive) */
2901 if (p == text)
2902 goto end;
2903
2904 d = p - 1;
2905 dig1 |= (unsigned int)(*d << s);
2906 if (d == text)
2907 goto end;
2908
2909 d--;
2910 dig10 |= (unsigned int)(*d << s);
2911 if (d == text)
2912 goto end;
2913
2914 d--;
2915 dig100 |= (unsigned int)(*d << s);
2916 end:
2917 if (!s || *p != '.')
2918 break;
2919
2920 s -= 8;
2921 text = ++p;
2922 }
2923
2924 dig100 -= ascii_zero;
2925 dig10 -= ascii_zero;
2926 dig1 -= ascii_zero;
2927 return ((dig100 * 10) + dig10) * 10 + dig1;
2928}
2929
2930/*
2931 * Idem except the first unparsed character has to be passed in <stop>.
2932 */
2933unsigned int inetaddr_host_lim(const char *text, const char *stop)
2934{
2935 const unsigned int ascii_zero = ('0' << 24) | ('0' << 16) | ('0' << 8) | '0';
2936 register unsigned int dig100, dig10, dig1;
2937 int s;
2938 const char *p, *d;
2939
2940 dig1 = dig10 = dig100 = ascii_zero;
2941 s = 24;
2942
2943 p = text;
2944 while (1) {
2945 if (((unsigned)(*p - '0')) <= 9 && p < stop) {
2946 p++;
2947 continue;
2948 }
2949
2950 /* here, we have a complete byte between <text> and <p> (exclusive) */
2951 if (p == text)
2952 goto end;
2953
2954 d = p - 1;
2955 dig1 |= (unsigned int)(*d << s);
2956 if (d == text)
2957 goto end;
2958
2959 d--;
2960 dig10 |= (unsigned int)(*d << s);
2961 if (d == text)
2962 goto end;
2963
2964 d--;
2965 dig100 |= (unsigned int)(*d << s);
2966 end:
2967 if (!s || p == stop || *p != '.')
2968 break;
2969
2970 s -= 8;
2971 text = ++p;
2972 }
2973
2974 dig100 -= ascii_zero;
2975 dig10 -= ascii_zero;
2976 dig1 -= ascii_zero;
2977 return ((dig100 * 10) + dig10) * 10 + dig1;
2978}
2979
2980/*
2981 * Idem except the pointer to first unparsed byte is returned into <ret> which
2982 * must not be NULL.
2983 */
Willy Tarreau74172752010-10-15 23:21:42 +02002984unsigned int inetaddr_host_lim_ret(char *text, char *stop, char **ret)
Willy Tarreaud54bbdc2009-09-07 11:00:31 +02002985{
2986 const unsigned int ascii_zero = ('0' << 24) | ('0' << 16) | ('0' << 8) | '0';
2987 register unsigned int dig100, dig10, dig1;
2988 int s;
Willy Tarreau74172752010-10-15 23:21:42 +02002989 char *p, *d;
Willy Tarreaud54bbdc2009-09-07 11:00:31 +02002990
2991 dig1 = dig10 = dig100 = ascii_zero;
2992 s = 24;
2993
2994 p = text;
2995 while (1) {
2996 if (((unsigned)(*p - '0')) <= 9 && p < stop) {
2997 p++;
2998 continue;
2999 }
3000
3001 /* here, we have a complete byte between <text> and <p> (exclusive) */
3002 if (p == text)
3003 goto end;
3004
3005 d = p - 1;
3006 dig1 |= (unsigned int)(*d << s);
3007 if (d == text)
3008 goto end;
3009
3010 d--;
3011 dig10 |= (unsigned int)(*d << s);
3012 if (d == text)
3013 goto end;
3014
3015 d--;
3016 dig100 |= (unsigned int)(*d << s);
3017 end:
3018 if (!s || p == stop || *p != '.')
3019 break;
3020
3021 s -= 8;
3022 text = ++p;
3023 }
3024
3025 *ret = p;
3026 dig100 -= ascii_zero;
3027 dig10 -= ascii_zero;
3028 dig1 -= ascii_zero;
3029 return ((dig100 * 10) + dig10) * 10 + dig1;
3030}
3031
Willy Tarreauf0b38bf2010-06-06 13:22:23 +02003032/* Convert a fixed-length string to an IP address. Returns 0 in case of error,
3033 * or the number of chars read in case of success. Maybe this could be replaced
3034 * by one of the functions above. Also, apparently this function does not support
3035 * hosts above 255 and requires exactly 4 octets.
Willy Tarreau075415a2013-12-12 11:29:39 +01003036 * The destination is only modified on success.
Willy Tarreauf0b38bf2010-06-06 13:22:23 +02003037 */
3038int buf2ip(const char *buf, size_t len, struct in_addr *dst)
3039{
3040 const char *addr;
3041 int saw_digit, octets, ch;
3042 u_char tmp[4], *tp;
3043 const char *cp = buf;
3044
3045 saw_digit = 0;
3046 octets = 0;
3047 *(tp = tmp) = 0;
3048
3049 for (addr = buf; addr - buf < len; addr++) {
3050 unsigned char digit = (ch = *addr) - '0';
3051
3052 if (digit > 9 && ch != '.')
3053 break;
3054
3055 if (digit <= 9) {
3056 u_int new = *tp * 10 + digit;
3057
3058 if (new > 255)
3059 return 0;
3060
3061 *tp = new;
3062
3063 if (!saw_digit) {
3064 if (++octets > 4)
3065 return 0;
3066 saw_digit = 1;
3067 }
3068 } else if (ch == '.' && saw_digit) {
3069 if (octets == 4)
3070 return 0;
3071
3072 *++tp = 0;
3073 saw_digit = 0;
3074 } else
3075 return 0;
3076 }
3077
3078 if (octets < 4)
3079 return 0;
3080
3081 memcpy(&dst->s_addr, tmp, 4);
3082 return addr - cp;
3083}
3084
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01003085/* This function converts the string in <buf> of the len <len> to
3086 * struct in6_addr <dst> which must be allocated by the caller.
3087 * This function returns 1 in success case, otherwise zero.
Willy Tarreau075415a2013-12-12 11:29:39 +01003088 * The destination is only modified on success.
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01003089 */
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01003090int buf2ip6(const char *buf, size_t len, struct in6_addr *dst)
3091{
Thierry FOURNIERcd659912013-12-11 12:33:54 +01003092 char null_term_ip6[INET6_ADDRSTRLEN + 1];
Willy Tarreau075415a2013-12-12 11:29:39 +01003093 struct in6_addr out;
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01003094
Thierry FOURNIERcd659912013-12-11 12:33:54 +01003095 if (len > INET6_ADDRSTRLEN)
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01003096 return 0;
3097
3098 memcpy(null_term_ip6, buf, len);
3099 null_term_ip6[len] = '\0';
3100
Willy Tarreau075415a2013-12-12 11:29:39 +01003101 if (!inet_pton(AF_INET6, null_term_ip6, &out))
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01003102 return 0;
3103
Willy Tarreau075415a2013-12-12 11:29:39 +01003104 *dst = out;
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01003105 return 1;
3106}
3107
Willy Tarreauacf95772010-06-14 19:09:21 +02003108/* To be used to quote config arg positions. Returns the short string at <ptr>
3109 * surrounded by simple quotes if <ptr> is valid and non-empty, or "end of line"
3110 * if ptr is NULL or empty. The string is locally allocated.
3111 */
3112const char *quote_arg(const char *ptr)
3113{
Christopher Faulet1bc04c72017-10-29 20:14:08 +01003114 static THREAD_LOCAL char val[32];
Willy Tarreauacf95772010-06-14 19:09:21 +02003115 int i;
3116
3117 if (!ptr || !*ptr)
3118 return "end of line";
3119 val[0] = '\'';
Willy Tarreaude2dd6b2013-01-24 02:14:42 +01003120 for (i = 1; i < sizeof(val) - 2 && *ptr; i++)
Willy Tarreauacf95772010-06-14 19:09:21 +02003121 val[i] = *ptr++;
3122 val[i++] = '\'';
3123 val[i] = '\0';
3124 return val;
3125}
3126
Willy Tarreau5b180202010-07-18 10:40:48 +02003127/* returns an operator among STD_OP_* for string <str> or < 0 if unknown */
3128int get_std_op(const char *str)
3129{
3130 int ret = -1;
3131
3132 if (*str == 'e' && str[1] == 'q')
3133 ret = STD_OP_EQ;
3134 else if (*str == 'n' && str[1] == 'e')
3135 ret = STD_OP_NE;
3136 else if (*str == 'l') {
3137 if (str[1] == 'e') ret = STD_OP_LE;
3138 else if (str[1] == 't') ret = STD_OP_LT;
3139 }
3140 else if (*str == 'g') {
3141 if (str[1] == 'e') ret = STD_OP_GE;
3142 else if (str[1] == 't') ret = STD_OP_GT;
3143 }
3144
3145 if (ret == -1 || str[2] != '\0')
3146 return -1;
3147 return ret;
3148}
3149
Willy Tarreau4c14eaa2010-11-24 14:01:45 +01003150/* hash a 32-bit integer to another 32-bit integer */
3151unsigned int full_hash(unsigned int a)
3152{
3153 return __full_hash(a);
3154}
3155
Willy Tarreauf3241112019-02-26 09:56:22 +01003156/* Return the bit position in mask <m> of the nth bit set of rank <r>, between
3157 * 0 and LONGBITS-1 included, starting from the left. For example ranks 0,1,2,3
3158 * for mask 0x55 will be 6, 4, 2 and 0 respectively. This algorithm is based on
3159 * a popcount variant and is described here :
3160 * https://graphics.stanford.edu/~seander/bithacks.html
3161 */
3162unsigned int mask_find_rank_bit(unsigned int r, unsigned long m)
3163{
3164 unsigned long a, b, c, d;
3165 unsigned int s;
3166 unsigned int t;
3167
3168 a = m - ((m >> 1) & ~0UL/3);
3169 b = (a & ~0UL/5) + ((a >> 2) & ~0UL/5);
3170 c = (b + (b >> 4)) & ~0UL/0x11;
3171 d = (c + (c >> 8)) & ~0UL/0x101;
3172
3173 r++; // make r be 1..64
3174
3175 t = 0;
3176 s = LONGBITS;
3177 if (s > 32) {
Willy Tarreau9b6be3b2019-03-18 16:31:18 +01003178 unsigned long d2 = (d >> 16) >> 16;
3179 t = d2 + (d2 >> 16);
Willy Tarreauf3241112019-02-26 09:56:22 +01003180 s -= ((t - r) & 256) >> 3; r -= (t & ((t - r) >> 8));
3181 }
3182
3183 t = (d >> (s - 16)) & 0xff;
3184 s -= ((t - r) & 256) >> 4; r -= (t & ((t - r) >> 8));
3185 t = (c >> (s - 8)) & 0xf;
3186 s -= ((t - r) & 256) >> 5; r -= (t & ((t - r) >> 8));
3187 t = (b >> (s - 4)) & 0x7;
3188 s -= ((t - r) & 256) >> 6; r -= (t & ((t - r) >> 8));
3189 t = (a >> (s - 2)) & 0x3;
3190 s -= ((t - r) & 256) >> 7; r -= (t & ((t - r) >> 8));
3191 t = (m >> (s - 1)) & 0x1;
3192 s -= ((t - r) & 256) >> 8;
3193
3194 return s - 1;
3195}
3196
3197/* Same as mask_find_rank_bit() above but makes use of pre-computed bitmaps
3198 * based on <m>, in <a..d>. These ones must be updated whenever <m> changes
3199 * using mask_prep_rank_map() below.
3200 */
3201unsigned int mask_find_rank_bit_fast(unsigned int r, unsigned long m,
3202 unsigned long a, unsigned long b,
3203 unsigned long c, unsigned long d)
3204{
3205 unsigned int s;
3206 unsigned int t;
3207
3208 r++; // make r be 1..64
3209
3210 t = 0;
3211 s = LONGBITS;
3212 if (s > 32) {
Willy Tarreau9b6be3b2019-03-18 16:31:18 +01003213 unsigned long d2 = (d >> 16) >> 16;
3214 t = d2 + (d2 >> 16);
Willy Tarreauf3241112019-02-26 09:56:22 +01003215 s -= ((t - r) & 256) >> 3; r -= (t & ((t - r) >> 8));
3216 }
3217
3218 t = (d >> (s - 16)) & 0xff;
3219 s -= ((t - r) & 256) >> 4; r -= (t & ((t - r) >> 8));
3220 t = (c >> (s - 8)) & 0xf;
3221 s -= ((t - r) & 256) >> 5; r -= (t & ((t - r) >> 8));
3222 t = (b >> (s - 4)) & 0x7;
3223 s -= ((t - r) & 256) >> 6; r -= (t & ((t - r) >> 8));
3224 t = (a >> (s - 2)) & 0x3;
3225 s -= ((t - r) & 256) >> 7; r -= (t & ((t - r) >> 8));
3226 t = (m >> (s - 1)) & 0x1;
3227 s -= ((t - r) & 256) >> 8;
3228
3229 return s - 1;
3230}
3231
3232/* Prepare the bitmaps used by the fast implementation of the find_rank_bit()
3233 * above.
3234 */
3235void mask_prep_rank_map(unsigned long m,
3236 unsigned long *a, unsigned long *b,
3237 unsigned long *c, unsigned long *d)
3238{
3239 *a = m - ((m >> 1) & ~0UL/3);
3240 *b = (*a & ~0UL/5) + ((*a >> 2) & ~0UL/5);
3241 *c = (*b + (*b >> 4)) & ~0UL/0x11;
3242 *d = (*c + (*c >> 8)) & ~0UL/0x101;
3243}
3244
Willy Tarreauc7a8a3c2022-06-21 20:19:54 +02003245/* Returns the position of one bit set in <v>, starting at position <bit>, and
3246 * searching in other halves if not found. This is intended to be used to
3247 * report the position of one bit set among several based on a counter or a
3248 * random generator while preserving a relatively good distribution so that
3249 * values made of holes in the middle do not see one of the bits around the
3250 * hole being returned much more often than the other one. It can be seen as a
3251 * disturbed ffsl() where the initial search starts at bit <bit>. The look up
3252 * is performed in O(logN) time for N bit words, yielding a bit among 64 in
3253 * about 16 cycles. Its usage differs from the rank find function in that the
3254 * bit passed doesn't need to be limited to the value's popcount, making the
3255 * function easier to use for random picking, and twice as fast. Passing value
3256 * 0 for <v> makes no sense and -1 is returned in this case.
3257 */
3258int one_among_mask(unsigned long v, int bit)
3259{
3260 /* note, these masks may be produced by ~0UL/((1UL<<scale)+1) but
3261 * that's more expensive.
3262 */
3263 static const unsigned long halves[] = {
3264 (unsigned long)0x5555555555555555ULL,
3265 (unsigned long)0x3333333333333333ULL,
3266 (unsigned long)0x0F0F0F0F0F0F0F0FULL,
3267 (unsigned long)0x00FF00FF00FF00FFULL,
3268 (unsigned long)0x0000FFFF0000FFFFULL,
3269 (unsigned long)0x00000000FFFFFFFFULL
3270 };
3271 unsigned long halfword = ~0UL;
3272 int scope = 0;
3273 int mirror;
3274 int scale;
3275
3276 if (!v)
3277 return -1;
3278
3279 /* we check if the exact bit is set or if it's present in a mirror
3280 * position based on the current scale we're checking, in which case
3281 * it's returned with its current (or mirrored) value. Otherwise we'll
3282 * make sure there's at least one bit in the half we're in, and will
3283 * scale down to a smaller scope and try again, until we find the
3284 * closest bit.
3285 */
3286 for (scale = (sizeof(long) > 4) ? 5 : 4; scale >= 0; scale--) {
3287 halfword >>= (1UL << scale);
3288 scope |= (1UL << scale);
3289 mirror = bit ^ (1UL << scale);
3290 if (v & ((1UL << bit) | (1UL << mirror)))
3291 return (v & (1UL << bit)) ? bit : mirror;
3292
3293 if (!((v >> (bit & scope)) & halves[scale] & halfword))
3294 bit = mirror;
3295 }
3296 return bit;
3297}
3298
David du Colombier4f92d322011-03-24 11:09:31 +01003299/* Return non-zero if IPv4 address is part of the network,
Willy Tarreaueec1d382016-07-13 11:59:39 +02003300 * otherwise zero. Note that <addr> may not necessarily be aligned
3301 * while the two other ones must.
David du Colombier4f92d322011-03-24 11:09:31 +01003302 */
Willy Tarreaueec1d382016-07-13 11:59:39 +02003303int in_net_ipv4(const void *addr, const struct in_addr *mask, const struct in_addr *net)
David du Colombier4f92d322011-03-24 11:09:31 +01003304{
Willy Tarreaueec1d382016-07-13 11:59:39 +02003305 struct in_addr addr_copy;
3306
3307 memcpy(&addr_copy, addr, sizeof(addr_copy));
3308 return((addr_copy.s_addr & mask->s_addr) == (net->s_addr & mask->s_addr));
David du Colombier4f92d322011-03-24 11:09:31 +01003309}
3310
3311/* Return non-zero if IPv6 address is part of the network,
Willy Tarreaueec1d382016-07-13 11:59:39 +02003312 * otherwise zero. Note that <addr> may not necessarily be aligned
3313 * while the two other ones must.
David du Colombier4f92d322011-03-24 11:09:31 +01003314 */
Willy Tarreaueec1d382016-07-13 11:59:39 +02003315int in_net_ipv6(const void *addr, const struct in6_addr *mask, const struct in6_addr *net)
David du Colombier4f92d322011-03-24 11:09:31 +01003316{
3317 int i;
Willy Tarreaueec1d382016-07-13 11:59:39 +02003318 struct in6_addr addr_copy;
David du Colombier4f92d322011-03-24 11:09:31 +01003319
Willy Tarreaueec1d382016-07-13 11:59:39 +02003320 memcpy(&addr_copy, addr, sizeof(addr_copy));
David du Colombier4f92d322011-03-24 11:09:31 +01003321 for (i = 0; i < sizeof(struct in6_addr) / sizeof(int); i++)
Willy Tarreaueec1d382016-07-13 11:59:39 +02003322 if (((((int *)&addr_copy)[i] & ((int *)mask)[i])) !=
David du Colombier4f92d322011-03-24 11:09:31 +01003323 (((int *)net)[i] & ((int *)mask)[i]))
3324 return 0;
3325 return 1;
3326}
3327
3328/* RFC 4291 prefix */
3329const char rfc4291_pfx[] = { 0x00, 0x00, 0x00, 0x00,
3330 0x00, 0x00, 0x00, 0x00,
3331 0x00, 0x00, 0xFF, 0xFF };
3332
Joseph Herlant32b83272018-11-15 11:58:28 -08003333/* Map IPv4 address on IPv6 address, as specified in RFC 3513.
Thierry FOURNIER4a04dc32013-11-28 16:33:15 +01003334 * Input and output may overlap.
3335 */
David du Colombier4f92d322011-03-24 11:09:31 +01003336void v4tov6(struct in6_addr *sin6_addr, struct in_addr *sin_addr)
3337{
Thierry FOURNIER4a04dc32013-11-28 16:33:15 +01003338 struct in_addr tmp_addr;
3339
3340 tmp_addr.s_addr = sin_addr->s_addr;
David du Colombier4f92d322011-03-24 11:09:31 +01003341 memcpy(sin6_addr->s6_addr, rfc4291_pfx, sizeof(rfc4291_pfx));
Thierry FOURNIER4a04dc32013-11-28 16:33:15 +01003342 memcpy(sin6_addr->s6_addr+12, &tmp_addr.s_addr, 4);
David du Colombier4f92d322011-03-24 11:09:31 +01003343}
3344
Joseph Herlant32b83272018-11-15 11:58:28 -08003345/* Map IPv6 address on IPv4 address, as specified in RFC 3513.
David du Colombier4f92d322011-03-24 11:09:31 +01003346 * Return true if conversion is possible and false otherwise.
3347 */
3348int v6tov4(struct in_addr *sin_addr, struct in6_addr *sin6_addr)
3349{
3350 if (memcmp(sin6_addr->s6_addr, rfc4291_pfx, sizeof(rfc4291_pfx)) == 0) {
3351 memcpy(&(sin_addr->s_addr), &(sin6_addr->s6_addr[12]),
3352 sizeof(struct in_addr));
3353 return 1;
3354 }
3355
3356 return 0;
3357}
3358
Amaury Denoyelle21e611d2022-12-01 17:46:45 +01003359/* compare two struct sockaddr_storage, including port if <check_port> is true,
3360 * and return:
Baptiste Assmann08b24cf2016-01-23 23:39:12 +01003361 * 0 (true) if the addr is the same in both
3362 * 1 (false) if the addr is not the same in both
3363 * -1 (unable) if one of the addr is not AF_INET*
3364 */
Amaury Denoyelle21e611d2022-12-01 17:46:45 +01003365int ipcmp(struct sockaddr_storage *ss1, struct sockaddr_storage *ss2, int check_port)
Baptiste Assmann08b24cf2016-01-23 23:39:12 +01003366{
3367 if ((ss1->ss_family != AF_INET) && (ss1->ss_family != AF_INET6))
3368 return -1;
3369
3370 if ((ss2->ss_family != AF_INET) && (ss2->ss_family != AF_INET6))
3371 return -1;
3372
3373 if (ss1->ss_family != ss2->ss_family)
3374 return 1;
3375
3376 switch (ss1->ss_family) {
3377 case AF_INET:
Amaury Denoyelle21e611d2022-12-01 17:46:45 +01003378 return (memcmp(&((struct sockaddr_in *)ss1)->sin_addr,
Baptiste Assmann08b24cf2016-01-23 23:39:12 +01003379 &((struct sockaddr_in *)ss2)->sin_addr,
Amaury Denoyelle21e611d2022-12-01 17:46:45 +01003380 sizeof(struct in_addr)) != 0) ||
3381 (check_port && get_net_port(ss1) != get_net_port(ss2));
Baptiste Assmann08b24cf2016-01-23 23:39:12 +01003382 case AF_INET6:
Amaury Denoyelle21e611d2022-12-01 17:46:45 +01003383 return (memcmp(&((struct sockaddr_in6 *)ss1)->sin6_addr,
Baptiste Assmann08b24cf2016-01-23 23:39:12 +01003384 &((struct sockaddr_in6 *)ss2)->sin6_addr,
Amaury Denoyelle21e611d2022-12-01 17:46:45 +01003385 sizeof(struct in6_addr)) != 0) ||
3386 (check_port && get_net_port(ss1) != get_net_port(ss2));
Baptiste Assmann08b24cf2016-01-23 23:39:12 +01003387 }
3388
3389 return 1;
3390}
3391
Christopher Faulet9553de72021-02-26 09:12:50 +01003392/* compare a struct sockaddr_storage to a struct net_addr and return :
3393 * 0 (true) if <addr> is matching <net>
3394 * 1 (false) if <addr> is not matching <net>
3395 * -1 (unable) if <addr> or <net> is not AF_INET*
3396 */
3397int ipcmp2net(const struct sockaddr_storage *addr, const struct net_addr *net)
3398{
3399 if ((addr->ss_family != AF_INET) && (addr->ss_family != AF_INET6))
3400 return -1;
3401
3402 if ((net->family != AF_INET) && (net->family != AF_INET6))
3403 return -1;
3404
3405 if (addr->ss_family != net->family)
3406 return 1;
3407
3408 if (addr->ss_family == AF_INET &&
3409 (((struct sockaddr_in *)addr)->sin_addr.s_addr & net->addr.v4.mask.s_addr) == net->addr.v4.ip.s_addr)
3410 return 0;
3411 else {
3412 const struct in6_addr *addr6 = &(((const struct sockaddr_in6*)addr)->sin6_addr);
3413 const struct in6_addr *nip6 = &net->addr.v6.ip;
3414 const struct in6_addr *nmask6 = &net->addr.v6.mask;
3415
3416 if ((read_u32(&addr6->s6_addr[0]) & read_u32(&nmask6->s6_addr[0])) == read_u32(&nip6->s6_addr[0]) &&
3417 (read_u32(&addr6->s6_addr[4]) & read_u32(&nmask6->s6_addr[4])) == read_u32(&nip6->s6_addr[4]) &&
3418 (read_u32(&addr6->s6_addr[8]) & read_u32(&nmask6->s6_addr[8])) == read_u32(&nip6->s6_addr[8]) &&
3419 (read_u32(&addr6->s6_addr[12]) & read_u32(&nmask6->s6_addr[12])) == read_u32(&nip6->s6_addr[12]))
3420 return 0;
3421 }
3422
3423 return 1;
3424}
3425
Baptiste Assmann08396c82016-01-31 00:27:17 +01003426/* copy IP address from <source> into <dest>
Willy Tarreaudc3a9e82016-11-04 18:47:01 +01003427 * The caller must allocate and clear <dest> before calling.
3428 * The source must be in either AF_INET or AF_INET6 family, or the destination
3429 * address will be undefined. If the destination address used to hold a port,
3430 * it is preserved, so that this function can be used to switch to another
3431 * address family with no risk. Returns a pointer to the destination.
Baptiste Assmann08396c82016-01-31 00:27:17 +01003432 */
3433struct sockaddr_storage *ipcpy(struct sockaddr_storage *source, struct sockaddr_storage *dest)
3434{
Willy Tarreaudc3a9e82016-11-04 18:47:01 +01003435 int prev_port;
3436
3437 prev_port = get_net_port(dest);
3438 memset(dest, 0, sizeof(*dest));
Baptiste Assmann08396c82016-01-31 00:27:17 +01003439 dest->ss_family = source->ss_family;
3440
3441 /* copy new addr and apply it */
3442 switch (source->ss_family) {
3443 case AF_INET:
3444 ((struct sockaddr_in *)dest)->sin_addr.s_addr = ((struct sockaddr_in *)source)->sin_addr.s_addr;
Willy Tarreaudc3a9e82016-11-04 18:47:01 +01003445 ((struct sockaddr_in *)dest)->sin_port = prev_port;
Baptiste Assmann08396c82016-01-31 00:27:17 +01003446 break;
3447 case AF_INET6:
3448 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 +01003449 ((struct sockaddr_in6 *)dest)->sin6_port = prev_port;
Baptiste Assmann08396c82016-01-31 00:27:17 +01003450 break;
3451 }
3452
3453 return dest;
3454}
3455
William Lallemand421f5b52012-02-06 18:15:57 +01003456char *human_time(int t, short hz_div) {
3457 static char rv[sizeof("24855d23h")+1]; // longest of "23h59m" and "59m59s"
3458 char *p = rv;
Willy Tarreau761b3d52014-04-14 14:53:06 +02003459 char *end = rv + sizeof(rv);
William Lallemand421f5b52012-02-06 18:15:57 +01003460 int cnt=2; // print two numbers
3461
3462 if (unlikely(t < 0 || hz_div <= 0)) {
Willy Tarreau761b3d52014-04-14 14:53:06 +02003463 snprintf(p, end - p, "?");
William Lallemand421f5b52012-02-06 18:15:57 +01003464 return rv;
3465 }
3466
3467 if (unlikely(hz_div > 1))
3468 t /= hz_div;
3469
3470 if (t >= DAY) {
Willy Tarreau761b3d52014-04-14 14:53:06 +02003471 p += snprintf(p, end - p, "%dd", t / DAY);
William Lallemand421f5b52012-02-06 18:15:57 +01003472 cnt--;
3473 }
3474
3475 if (cnt && t % DAY / HOUR) {
Willy Tarreau761b3d52014-04-14 14:53:06 +02003476 p += snprintf(p, end - p, "%dh", t % DAY / HOUR);
William Lallemand421f5b52012-02-06 18:15:57 +01003477 cnt--;
3478 }
3479
3480 if (cnt && t % HOUR / MINUTE) {
Willy Tarreau761b3d52014-04-14 14:53:06 +02003481 p += snprintf(p, end - p, "%dm", t % HOUR / MINUTE);
William Lallemand421f5b52012-02-06 18:15:57 +01003482 cnt--;
3483 }
3484
3485 if ((cnt && t % MINUTE) || !t) // also display '0s'
Willy Tarreau761b3d52014-04-14 14:53:06 +02003486 p += snprintf(p, end - p, "%ds", t % MINUTE / SEC);
William Lallemand421f5b52012-02-06 18:15:57 +01003487
3488 return rv;
3489}
3490
3491const char *monthname[12] = {
3492 "Jan", "Feb", "Mar", "Apr", "May", "Jun",
3493 "Jul", "Aug", "Sep", "Oct", "Nov", "Dec"
3494};
3495
3496/* date2str_log: write a date in the format :
3497 * sprintf(str, "%02d/%s/%04d:%02d:%02d:%02d.%03d",
3498 * tm.tm_mday, monthname[tm.tm_mon], tm.tm_year+1900,
3499 * tm.tm_hour, tm.tm_min, tm.tm_sec, (int)date.tv_usec/1000);
3500 *
3501 * without using sprintf. return a pointer to the last char written (\0) or
3502 * NULL if there isn't enough space.
3503 */
Willy Tarreauf16cb412018-09-04 19:08:48 +02003504char *date2str_log(char *dst, const struct tm *tm, const struct timeval *date, size_t size)
William Lallemand421f5b52012-02-06 18:15:57 +01003505{
3506
3507 if (size < 25) /* the size is fixed: 24 chars + \0 */
3508 return NULL;
3509
3510 dst = utoa_pad((unsigned int)tm->tm_mday, dst, 3); // day
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003511 if (!dst)
3512 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003513 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003514
William Lallemand421f5b52012-02-06 18:15:57 +01003515 memcpy(dst, monthname[tm->tm_mon], 3); // month
3516 dst += 3;
3517 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003518
William Lallemand421f5b52012-02-06 18:15:57 +01003519 dst = utoa_pad((unsigned int)tm->tm_year+1900, dst, 5); // year
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003520 if (!dst)
3521 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003522 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003523
William Lallemand421f5b52012-02-06 18:15:57 +01003524 dst = utoa_pad((unsigned int)tm->tm_hour, dst, 3); // hour
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003525 if (!dst)
3526 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003527 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003528
William Lallemand421f5b52012-02-06 18:15:57 +01003529 dst = utoa_pad((unsigned int)tm->tm_min, dst, 3); // minutes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003530 if (!dst)
3531 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003532 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003533
William Lallemand421f5b52012-02-06 18:15:57 +01003534 dst = utoa_pad((unsigned int)tm->tm_sec, dst, 3); // secondes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003535 if (!dst)
3536 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003537 *dst++ = '.';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003538
Willy Tarreau7d9421d2020-02-29 09:08:02 +01003539 dst = utoa_pad((unsigned int)(date->tv_usec/1000)%1000, dst, 4); // milliseconds
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003540 if (!dst)
3541 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003542 *dst = '\0';
3543
3544 return dst;
3545}
3546
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003547/* Base year used to compute leap years */
3548#define TM_YEAR_BASE 1900
3549
3550/* Return the difference in seconds between two times (leap seconds are ignored).
3551 * Retrieved from glibc 2.18 source code.
3552 */
3553static int my_tm_diff(const struct tm *a, const struct tm *b)
3554{
3555 /* Compute intervening leap days correctly even if year is negative.
3556 * Take care to avoid int overflow in leap day calculations,
3557 * but it's OK to assume that A and B are close to each other.
3558 */
3559 int a4 = (a->tm_year >> 2) + (TM_YEAR_BASE >> 2) - ! (a->tm_year & 3);
3560 int b4 = (b->tm_year >> 2) + (TM_YEAR_BASE >> 2) - ! (b->tm_year & 3);
3561 int a100 = a4 / 25 - (a4 % 25 < 0);
3562 int b100 = b4 / 25 - (b4 % 25 < 0);
3563 int a400 = a100 >> 2;
3564 int b400 = b100 >> 2;
3565 int intervening_leap_days = (a4 - b4) - (a100 - b100) + (a400 - b400);
3566 int years = a->tm_year - b->tm_year;
3567 int days = (365 * years + intervening_leap_days
3568 + (a->tm_yday - b->tm_yday));
3569 return (60 * (60 * (24 * days + (a->tm_hour - b->tm_hour))
3570 + (a->tm_min - b->tm_min))
3571 + (a->tm_sec - b->tm_sec));
3572}
3573
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003574/* Return the GMT offset for a specific local time.
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003575 * Both t and tm must represent the same time.
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003576 * The string returned has the same format as returned by strftime(... "%z", tm).
3577 * Offsets are kept in an internal cache for better performances.
3578 */
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003579const char *get_gmt_offset(time_t t, struct tm *tm)
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003580{
3581 /* Cache offsets from GMT (depending on whether DST is active or not) */
Christopher Faulet1bc04c72017-10-29 20:14:08 +01003582 static THREAD_LOCAL char gmt_offsets[2][5+1] = { "", "" };
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003583
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003584 char *gmt_offset;
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003585 struct tm tm_gmt;
3586 int diff;
3587 int isdst = tm->tm_isdst;
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003588
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003589 /* Pretend DST not active if its status is unknown */
3590 if (isdst < 0)
3591 isdst = 0;
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003592
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003593 /* Fetch the offset and initialize it if needed */
3594 gmt_offset = gmt_offsets[isdst & 0x01];
3595 if (unlikely(!*gmt_offset)) {
3596 get_gmtime(t, &tm_gmt);
3597 diff = my_tm_diff(tm, &tm_gmt);
3598 if (diff < 0) {
3599 diff = -diff;
3600 *gmt_offset = '-';
3601 } else {
3602 *gmt_offset = '+';
3603 }
Willy Tarreaue112c8a2019-10-29 10:16:11 +01003604 diff %= 86400U;
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003605 diff /= 60; /* Convert to minutes */
3606 snprintf(gmt_offset+1, 4+1, "%02d%02d", diff/60, diff%60);
3607 }
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003608
Willy Tarreaue112c8a2019-10-29 10:16:11 +01003609 return gmt_offset;
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003610}
3611
William Lallemand421f5b52012-02-06 18:15:57 +01003612/* gmt2str_log: write a date in the format :
3613 * "%02d/%s/%04d:%02d:%02d:%02d +0000" without using snprintf
3614 * return a pointer to the last char written (\0) or
3615 * NULL if there isn't enough space.
3616 */
3617char *gmt2str_log(char *dst, struct tm *tm, size_t size)
3618{
Yuxans Yao4e25b012012-10-19 10:36:09 +08003619 if (size < 27) /* the size is fixed: 26 chars + \0 */
William Lallemand421f5b52012-02-06 18:15:57 +01003620 return NULL;
3621
3622 dst = utoa_pad((unsigned int)tm->tm_mday, dst, 3); // day
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003623 if (!dst)
3624 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003625 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003626
William Lallemand421f5b52012-02-06 18:15:57 +01003627 memcpy(dst, monthname[tm->tm_mon], 3); // month
3628 dst += 3;
3629 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003630
William Lallemand421f5b52012-02-06 18:15:57 +01003631 dst = utoa_pad((unsigned int)tm->tm_year+1900, dst, 5); // year
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003632 if (!dst)
3633 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003634 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003635
William Lallemand421f5b52012-02-06 18:15:57 +01003636 dst = utoa_pad((unsigned int)tm->tm_hour, dst, 3); // hour
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003637 if (!dst)
3638 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003639 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003640
William Lallemand421f5b52012-02-06 18:15:57 +01003641 dst = utoa_pad((unsigned int)tm->tm_min, dst, 3); // minutes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003642 if (!dst)
3643 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003644 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003645
William Lallemand421f5b52012-02-06 18:15:57 +01003646 dst = utoa_pad((unsigned int)tm->tm_sec, dst, 3); // secondes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003647 if (!dst)
3648 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003649 *dst++ = ' ';
3650 *dst++ = '+';
3651 *dst++ = '0';
3652 *dst++ = '0';
3653 *dst++ = '0';
3654 *dst++ = '0';
3655 *dst = '\0';
3656
3657 return dst;
3658}
3659
Yuxans Yao4e25b012012-10-19 10:36:09 +08003660/* localdate2str_log: write a date in the format :
3661 * "%02d/%s/%04d:%02d:%02d:%02d +0000(local timezone)" without using snprintf
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003662 * Both t and tm must represent the same time.
3663 * return a pointer to the last char written (\0) or
3664 * NULL if there isn't enough space.
Yuxans Yao4e25b012012-10-19 10:36:09 +08003665 */
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003666char *localdate2str_log(char *dst, time_t t, struct tm *tm, size_t size)
Yuxans Yao4e25b012012-10-19 10:36:09 +08003667{
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003668 const char *gmt_offset;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003669 if (size < 27) /* the size is fixed: 26 chars + \0 */
3670 return NULL;
3671
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003672 gmt_offset = get_gmt_offset(t, tm);
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003673
Yuxans Yao4e25b012012-10-19 10:36:09 +08003674 dst = utoa_pad((unsigned int)tm->tm_mday, dst, 3); // day
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003675 if (!dst)
3676 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003677 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003678
Yuxans Yao4e25b012012-10-19 10:36:09 +08003679 memcpy(dst, monthname[tm->tm_mon], 3); // month
3680 dst += 3;
3681 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003682
Yuxans Yao4e25b012012-10-19 10:36:09 +08003683 dst = utoa_pad((unsigned int)tm->tm_year+1900, dst, 5); // year
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003684 if (!dst)
3685 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003686 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003687
Yuxans Yao4e25b012012-10-19 10:36:09 +08003688 dst = utoa_pad((unsigned int)tm->tm_hour, dst, 3); // hour
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003689 if (!dst)
3690 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003691 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003692
Yuxans Yao4e25b012012-10-19 10:36:09 +08003693 dst = utoa_pad((unsigned int)tm->tm_min, dst, 3); // minutes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003694 if (!dst)
3695 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003696 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003697
Yuxans Yao4e25b012012-10-19 10:36:09 +08003698 dst = utoa_pad((unsigned int)tm->tm_sec, dst, 3); // secondes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003699 if (!dst)
3700 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003701 *dst++ = ' ';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003702
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003703 memcpy(dst, gmt_offset, 5); // Offset from local time to GMT
Yuxans Yao4e25b012012-10-19 10:36:09 +08003704 dst += 5;
3705 *dst = '\0';
3706
3707 return dst;
3708}
3709
Willy Tarreaucb1949b2017-07-19 19:05:29 +02003710/* Returns the number of seconds since 01/01/1970 0:0:0 GMT for GMT date <tm>.
3711 * It is meant as a portable replacement for timegm() for use with valid inputs.
3712 * Returns undefined results for invalid dates (eg: months out of range 0..11).
3713 */
3714time_t my_timegm(const struct tm *tm)
3715{
3716 /* Each month has 28, 29, 30 or 31 days, or 28+N. The date in the year
3717 * is thus (current month - 1)*28 + cumulated_N[month] to count the
3718 * sum of the extra N days for elapsed months. The sum of all these N
3719 * days doesn't exceed 30 for a complete year (366-12*28) so it fits
3720 * in a 5-bit word. This means that with 60 bits we can represent a
3721 * matrix of all these values at once, which is fast and efficient to
3722 * access. The extra February day for leap years is not counted here.
3723 *
3724 * Jan : none = 0 (0)
3725 * Feb : Jan = 3 (3)
3726 * Mar : Jan..Feb = 3 (3 + 0)
3727 * Apr : Jan..Mar = 6 (3 + 0 + 3)
3728 * May : Jan..Apr = 8 (3 + 0 + 3 + 2)
3729 * Jun : Jan..May = 11 (3 + 0 + 3 + 2 + 3)
3730 * Jul : Jan..Jun = 13 (3 + 0 + 3 + 2 + 3 + 2)
3731 * Aug : Jan..Jul = 16 (3 + 0 + 3 + 2 + 3 + 2 + 3)
3732 * Sep : Jan..Aug = 19 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3)
3733 * Oct : Jan..Sep = 21 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3 + 2)
3734 * Nov : Jan..Oct = 24 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3 + 2 + 3)
3735 * Dec : Jan..Nov = 26 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3 + 2 + 3 + 2)
3736 */
3737 uint64_t extra =
3738 ( 0ULL << 0*5) + ( 3ULL << 1*5) + ( 3ULL << 2*5) + /* Jan, Feb, Mar, */
3739 ( 6ULL << 3*5) + ( 8ULL << 4*5) + (11ULL << 5*5) + /* Apr, May, Jun, */
3740 (13ULL << 6*5) + (16ULL << 7*5) + (19ULL << 8*5) + /* Jul, Aug, Sep, */
3741 (21ULL << 9*5) + (24ULL << 10*5) + (26ULL << 11*5); /* Oct, Nov, Dec, */
3742
3743 unsigned int y = tm->tm_year + 1900;
3744 unsigned int m = tm->tm_mon;
3745 unsigned long days = 0;
3746
3747 /* days since 1/1/1970 for full years */
3748 days += days_since_zero(y) - days_since_zero(1970);
3749
3750 /* days for full months in the current year */
3751 days += 28 * m + ((extra >> (m * 5)) & 0x1f);
3752
3753 /* count + 1 after March for leap years. A leap year is a year multiple
3754 * of 4, unless it's multiple of 100 without being multiple of 400. 2000
3755 * is leap, 1900 isn't, 1904 is.
3756 */
3757 if ((m > 1) && !(y & 3) && ((y % 100) || !(y % 400)))
3758 days++;
3759
3760 days += tm->tm_mday - 1;
3761 return days * 86400ULL + tm->tm_hour * 3600 + tm->tm_min * 60 + tm->tm_sec;
3762}
3763
Thierry Fournier93127942016-01-20 18:49:45 +01003764/* This function check a char. It returns true and updates
3765 * <date> and <len> pointer to the new position if the
3766 * character is found.
3767 */
3768static inline int parse_expect_char(const char **date, int *len, char c)
3769{
3770 if (*len < 1 || **date != c)
3771 return 0;
3772 (*len)--;
3773 (*date)++;
3774 return 1;
3775}
3776
3777/* This function expects a string <str> of len <l>. It return true and updates.
3778 * <date> and <len> if the string matches, otherwise, it returns false.
3779 */
3780static inline int parse_strcmp(const char **date, int *len, char *str, int l)
3781{
3782 if (*len < l || strncmp(*date, str, l) != 0)
3783 return 0;
3784 (*len) -= l;
3785 (*date) += l;
3786 return 1;
3787}
3788
3789/* This macro converts 3 chars name in integer. */
3790#define STR2I3(__a, __b, __c) ((__a) * 65536 + (__b) * 256 + (__c))
3791
3792/* day-name = %x4D.6F.6E ; "Mon", case-sensitive
3793 * / %x54.75.65 ; "Tue", case-sensitive
3794 * / %x57.65.64 ; "Wed", case-sensitive
3795 * / %x54.68.75 ; "Thu", case-sensitive
3796 * / %x46.72.69 ; "Fri", case-sensitive
3797 * / %x53.61.74 ; "Sat", case-sensitive
3798 * / %x53.75.6E ; "Sun", case-sensitive
3799 *
3800 * This array must be alphabetically sorted
3801 */
3802static inline int parse_http_dayname(const char **date, int *len, struct tm *tm)
3803{
3804 if (*len < 3)
3805 return 0;
3806 switch (STR2I3((*date)[0], (*date)[1], (*date)[2])) {
3807 case STR2I3('M','o','n'): tm->tm_wday = 1; break;
3808 case STR2I3('T','u','e'): tm->tm_wday = 2; break;
3809 case STR2I3('W','e','d'): tm->tm_wday = 3; break;
3810 case STR2I3('T','h','u'): tm->tm_wday = 4; break;
3811 case STR2I3('F','r','i'): tm->tm_wday = 5; break;
3812 case STR2I3('S','a','t'): tm->tm_wday = 6; break;
3813 case STR2I3('S','u','n'): tm->tm_wday = 7; break;
3814 default: return 0;
3815 }
3816 *len -= 3;
3817 *date += 3;
3818 return 1;
3819}
3820
3821/* month = %x4A.61.6E ; "Jan", case-sensitive
3822 * / %x46.65.62 ; "Feb", case-sensitive
3823 * / %x4D.61.72 ; "Mar", case-sensitive
3824 * / %x41.70.72 ; "Apr", case-sensitive
3825 * / %x4D.61.79 ; "May", case-sensitive
3826 * / %x4A.75.6E ; "Jun", case-sensitive
3827 * / %x4A.75.6C ; "Jul", case-sensitive
3828 * / %x41.75.67 ; "Aug", case-sensitive
3829 * / %x53.65.70 ; "Sep", case-sensitive
3830 * / %x4F.63.74 ; "Oct", case-sensitive
3831 * / %x4E.6F.76 ; "Nov", case-sensitive
3832 * / %x44.65.63 ; "Dec", case-sensitive
3833 *
3834 * This array must be alphabetically sorted
3835 */
3836static inline int parse_http_monthname(const char **date, int *len, struct tm *tm)
3837{
3838 if (*len < 3)
3839 return 0;
3840 switch (STR2I3((*date)[0], (*date)[1], (*date)[2])) {
3841 case STR2I3('J','a','n'): tm->tm_mon = 0; break;
3842 case STR2I3('F','e','b'): tm->tm_mon = 1; break;
3843 case STR2I3('M','a','r'): tm->tm_mon = 2; break;
3844 case STR2I3('A','p','r'): tm->tm_mon = 3; break;
3845 case STR2I3('M','a','y'): tm->tm_mon = 4; break;
3846 case STR2I3('J','u','n'): tm->tm_mon = 5; break;
3847 case STR2I3('J','u','l'): tm->tm_mon = 6; break;
3848 case STR2I3('A','u','g'): tm->tm_mon = 7; break;
3849 case STR2I3('S','e','p'): tm->tm_mon = 8; break;
3850 case STR2I3('O','c','t'): tm->tm_mon = 9; break;
3851 case STR2I3('N','o','v'): tm->tm_mon = 10; break;
3852 case STR2I3('D','e','c'): tm->tm_mon = 11; break;
3853 default: return 0;
3854 }
3855 *len -= 3;
3856 *date += 3;
3857 return 1;
3858}
3859
3860/* day-name-l = %x4D.6F.6E.64.61.79 ; "Monday", case-sensitive
3861 * / %x54.75.65.73.64.61.79 ; "Tuesday", case-sensitive
3862 * / %x57.65.64.6E.65.73.64.61.79 ; "Wednesday", case-sensitive
3863 * / %x54.68.75.72.73.64.61.79 ; "Thursday", case-sensitive
3864 * / %x46.72.69.64.61.79 ; "Friday", case-sensitive
3865 * / %x53.61.74.75.72.64.61.79 ; "Saturday", case-sensitive
3866 * / %x53.75.6E.64.61.79 ; "Sunday", case-sensitive
3867 *
3868 * This array must be alphabetically sorted
3869 */
3870static inline int parse_http_ldayname(const char **date, int *len, struct tm *tm)
3871{
3872 if (*len < 6) /* Minimum length. */
3873 return 0;
3874 switch (STR2I3((*date)[0], (*date)[1], (*date)[2])) {
3875 case STR2I3('M','o','n'):
3876 RET0_UNLESS(parse_strcmp(date, len, "Monday", 6));
3877 tm->tm_wday = 1;
3878 return 1;
3879 case STR2I3('T','u','e'):
3880 RET0_UNLESS(parse_strcmp(date, len, "Tuesday", 7));
3881 tm->tm_wday = 2;
3882 return 1;
3883 case STR2I3('W','e','d'):
3884 RET0_UNLESS(parse_strcmp(date, len, "Wednesday", 9));
3885 tm->tm_wday = 3;
3886 return 1;
3887 case STR2I3('T','h','u'):
3888 RET0_UNLESS(parse_strcmp(date, len, "Thursday", 8));
3889 tm->tm_wday = 4;
3890 return 1;
3891 case STR2I3('F','r','i'):
3892 RET0_UNLESS(parse_strcmp(date, len, "Friday", 6));
3893 tm->tm_wday = 5;
3894 return 1;
3895 case STR2I3('S','a','t'):
3896 RET0_UNLESS(parse_strcmp(date, len, "Saturday", 8));
3897 tm->tm_wday = 6;
3898 return 1;
3899 case STR2I3('S','u','n'):
3900 RET0_UNLESS(parse_strcmp(date, len, "Sunday", 6));
3901 tm->tm_wday = 7;
3902 return 1;
3903 }
3904 return 0;
3905}
3906
3907/* This function parses exactly 1 digit and returns the numeric value in "digit". */
3908static inline int parse_digit(const char **date, int *len, int *digit)
3909{
3910 if (*len < 1 || **date < '0' || **date > '9')
3911 return 0;
3912 *digit = (**date - '0');
3913 (*date)++;
3914 (*len)--;
3915 return 1;
3916}
3917
3918/* This function parses exactly 2 digits and returns the numeric value in "digit". */
3919static inline int parse_2digit(const char **date, int *len, int *digit)
3920{
3921 int value;
3922
3923 RET0_UNLESS(parse_digit(date, len, &value));
3924 (*digit) = value * 10;
3925 RET0_UNLESS(parse_digit(date, len, &value));
3926 (*digit) += value;
3927
3928 return 1;
3929}
3930
3931/* This function parses exactly 4 digits and returns the numeric value in "digit". */
3932static inline int parse_4digit(const char **date, int *len, int *digit)
3933{
3934 int value;
3935
3936 RET0_UNLESS(parse_digit(date, len, &value));
3937 (*digit) = value * 1000;
3938
3939 RET0_UNLESS(parse_digit(date, len, &value));
3940 (*digit) += value * 100;
3941
3942 RET0_UNLESS(parse_digit(date, len, &value));
3943 (*digit) += value * 10;
3944
3945 RET0_UNLESS(parse_digit(date, len, &value));
3946 (*digit) += value;
3947
3948 return 1;
3949}
3950
3951/* time-of-day = hour ":" minute ":" second
3952 * ; 00:00:00 - 23:59:60 (leap second)
3953 *
3954 * hour = 2DIGIT
3955 * minute = 2DIGIT
3956 * second = 2DIGIT
3957 */
3958static inline int parse_http_time(const char **date, int *len, struct tm *tm)
3959{
3960 RET0_UNLESS(parse_2digit(date, len, &tm->tm_hour)); /* hour 2DIGIT */
3961 RET0_UNLESS(parse_expect_char(date, len, ':')); /* expect ":" */
3962 RET0_UNLESS(parse_2digit(date, len, &tm->tm_min)); /* min 2DIGIT */
3963 RET0_UNLESS(parse_expect_char(date, len, ':')); /* expect ":" */
3964 RET0_UNLESS(parse_2digit(date, len, &tm->tm_sec)); /* sec 2DIGIT */
3965 return 1;
3966}
3967
3968/* From RFC7231
3969 * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
3970 *
3971 * IMF-fixdate = day-name "," SP date1 SP time-of-day SP GMT
3972 * ; fixed length/zone/capitalization subset of the format
3973 * ; see Section 3.3 of [RFC5322]
3974 *
3975 *
3976 * date1 = day SP month SP year
3977 * ; e.g., 02 Jun 1982
3978 *
3979 * day = 2DIGIT
3980 * year = 4DIGIT
3981 *
3982 * GMT = %x47.4D.54 ; "GMT", case-sensitive
3983 *
3984 * time-of-day = hour ":" minute ":" second
3985 * ; 00:00:00 - 23:59:60 (leap second)
3986 *
3987 * hour = 2DIGIT
3988 * minute = 2DIGIT
3989 * second = 2DIGIT
3990 *
3991 * DIGIT = decimal 0-9
3992 */
3993int parse_imf_date(const char *date, int len, struct tm *tm)
3994{
David Carlier327298c2016-11-20 10:42:38 +00003995 /* tm_gmtoff, if present, ought to be zero'ed */
3996 memset(tm, 0, sizeof(*tm));
3997
Thierry Fournier93127942016-01-20 18:49:45 +01003998 RET0_UNLESS(parse_http_dayname(&date, &len, tm)); /* day-name */
3999 RET0_UNLESS(parse_expect_char(&date, &len, ',')); /* expect "," */
4000 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
4001 RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_mday)); /* day 2DIGIT */
4002 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
4003 RET0_UNLESS(parse_http_monthname(&date, &len, tm)); /* Month */
4004 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
4005 RET0_UNLESS(parse_4digit(&date, &len, &tm->tm_year)); /* year = 4DIGIT */
4006 tm->tm_year -= 1900;
4007 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
4008 RET0_UNLESS(parse_http_time(&date, &len, tm)); /* Parse time. */
4009 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
4010 RET0_UNLESS(parse_strcmp(&date, &len, "GMT", 3)); /* GMT = %x47.4D.54 ; "GMT", case-sensitive */
4011 tm->tm_isdst = -1;
Thierry Fournier93127942016-01-20 18:49:45 +01004012 return 1;
4013}
4014
4015/* From RFC7231
4016 * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
4017 *
4018 * rfc850-date = day-name-l "," SP date2 SP time-of-day SP GMT
4019 * date2 = day "-" month "-" 2DIGIT
4020 * ; e.g., 02-Jun-82
4021 *
4022 * day = 2DIGIT
4023 */
4024int parse_rfc850_date(const char *date, int len, struct tm *tm)
4025{
4026 int year;
4027
David Carlier327298c2016-11-20 10:42:38 +00004028 /* tm_gmtoff, if present, ought to be zero'ed */
4029 memset(tm, 0, sizeof(*tm));
4030
Thierry Fournier93127942016-01-20 18:49:45 +01004031 RET0_UNLESS(parse_http_ldayname(&date, &len, tm)); /* Read the day name */
4032 RET0_UNLESS(parse_expect_char(&date, &len, ',')); /* expect "," */
4033 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
4034 RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_mday)); /* day 2DIGIT */
4035 RET0_UNLESS(parse_expect_char(&date, &len, '-')); /* expect "-" */
4036 RET0_UNLESS(parse_http_monthname(&date, &len, tm)); /* Month */
4037 RET0_UNLESS(parse_expect_char(&date, &len, '-')); /* expect "-" */
4038
4039 /* year = 2DIGIT
4040 *
4041 * Recipients of a timestamp value in rfc850-(*date) format, which uses a
4042 * two-digit year, MUST interpret a timestamp that appears to be more
4043 * than 50 years in the future as representing the most recent year in
4044 * the past that had the same last two digits.
4045 */
4046 RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_year));
4047
4048 /* expect SP */
4049 if (!parse_expect_char(&date, &len, ' ')) {
4050 /* Maybe we have the date with 4 digits. */
4051 RET0_UNLESS(parse_2digit(&date, &len, &year));
4052 tm->tm_year = (tm->tm_year * 100 + year) - 1900;
4053 /* expect SP */
4054 RET0_UNLESS(parse_expect_char(&date, &len, ' '));
4055 } else {
4056 /* I fix 60 as pivot: >60: +1900, <60: +2000. Note that the
4057 * tm_year is the number of year since 1900, so for +1900, we
4058 * do nothing, and for +2000, we add 100.
4059 */
4060 if (tm->tm_year <= 60)
4061 tm->tm_year += 100;
4062 }
4063
4064 RET0_UNLESS(parse_http_time(&date, &len, tm)); /* Parse time. */
4065 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
4066 RET0_UNLESS(parse_strcmp(&date, &len, "GMT", 3)); /* GMT = %x47.4D.54 ; "GMT", case-sensitive */
4067 tm->tm_isdst = -1;
Thierry Fournier93127942016-01-20 18:49:45 +01004068
4069 return 1;
4070}
4071
4072/* From RFC7231
4073 * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
4074 *
4075 * asctime-date = day-name SP date3 SP time-of-day SP year
4076 * date3 = month SP ( 2DIGIT / ( SP 1DIGIT ))
4077 * ; e.g., Jun 2
4078 *
4079 * HTTP-date is case sensitive. A sender MUST NOT generate additional
4080 * whitespace in an HTTP-date beyond that specifically included as SP in
4081 * the grammar.
4082 */
4083int parse_asctime_date(const char *date, int len, struct tm *tm)
4084{
David Carlier327298c2016-11-20 10:42:38 +00004085 /* tm_gmtoff, if present, ought to be zero'ed */
4086 memset(tm, 0, sizeof(*tm));
4087
Thierry Fournier93127942016-01-20 18:49:45 +01004088 RET0_UNLESS(parse_http_dayname(&date, &len, tm)); /* day-name */
4089 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
4090 RET0_UNLESS(parse_http_monthname(&date, &len, tm)); /* expect month */
4091 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
4092
4093 /* expect SP and 1DIGIT or 2DIGIT */
4094 if (parse_expect_char(&date, &len, ' '))
4095 RET0_UNLESS(parse_digit(&date, &len, &tm->tm_mday));
4096 else
4097 RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_mday));
4098
4099 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
4100 RET0_UNLESS(parse_http_time(&date, &len, tm)); /* Parse time. */
4101 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
4102 RET0_UNLESS(parse_4digit(&date, &len, &tm->tm_year)); /* year = 4DIGIT */
4103 tm->tm_year -= 1900;
4104 tm->tm_isdst = -1;
Thierry Fournier93127942016-01-20 18:49:45 +01004105 return 1;
4106}
4107
4108/* From RFC7231
4109 * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
4110 *
4111 * HTTP-date = IMF-fixdate / obs-date
4112 * obs-date = rfc850-date / asctime-date
4113 *
4114 * parses an HTTP date in the RFC format and is accepted
4115 * alternatives. <date> is the strinf containing the date,
4116 * len is the len of the string. <tm> is filled with the
4117 * parsed time. We must considers this time as GMT.
4118 */
4119int parse_http_date(const char *date, int len, struct tm *tm)
4120{
4121 if (parse_imf_date(date, len, tm))
4122 return 1;
4123
4124 if (parse_rfc850_date(date, len, tm))
4125 return 1;
4126
4127 if (parse_asctime_date(date, len, tm))
4128 return 1;
4129
4130 return 0;
4131}
4132
Willy Tarreau4deeb102021-01-29 10:47:52 +01004133/* print the time <ns> in a short form (exactly 7 chars) at the end of buffer
4134 * <out>. "-" is printed if the value is zero, "inf" if larger than 1000 years.
4135 * It returns the new buffer length, or 0 if it doesn't fit. The value will be
4136 * surrounded by <pfx> and <sfx> respectively if not NULL.
4137 */
4138int print_time_short(struct buffer *out, const char *pfx, uint64_t ns, const char *sfx)
4139{
4140 double val = ns; // 52 bits of mantissa keep ns accuracy over 52 days
4141 const char *unit;
4142
4143 if (!pfx)
4144 pfx = "";
4145 if (!sfx)
4146 sfx = "";
4147
4148 do {
4149 unit = " - "; if (val <= 0.0) break;
4150 unit = "ns"; if (val < 1000.0) break;
4151 unit = "us"; val /= 1000.0; if (val < 1000.0) break;
4152 unit = "ms"; val /= 1000.0; if (val < 1000.0) break;
4153 unit = "s "; val /= 1000.0; if (val < 60.0) break;
4154 unit = "m "; val /= 60.0; if (val < 60.0) break;
4155 unit = "h "; val /= 60.0; if (val < 24.0) break;
4156 unit = "d "; val /= 24.0; if (val < 365.0) break;
4157 unit = "yr"; val /= 365.0; if (val < 1000.0) break;
4158 unit = " inf "; val = 0.0; break;
4159 } while (0);
4160
4161 if (val <= 0.0)
4162 return chunk_appendf(out, "%s%7s%s", pfx, unit, sfx);
4163 else if (val < 10.0)
4164 return chunk_appendf(out, "%s%1.3f%s%s", pfx, val, unit, sfx);
4165 else if (val < 100.0)
4166 return chunk_appendf(out, "%s%2.2f%s%s", pfx, val, unit, sfx);
4167 else
4168 return chunk_appendf(out, "%s%3.1f%s%s", pfx, val, unit, sfx);
4169}
4170
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004171/* Dynamically allocates a string of the proper length to hold the formatted
4172 * output. NULL is returned on error. The caller is responsible for freeing the
4173 * memory area using free(). The resulting string is returned in <out> if the
4174 * pointer is not NULL. A previous version of <out> might be used to build the
4175 * new string, and it will be freed before returning if it is not NULL, which
4176 * makes it possible to build complex strings from iterative calls without
4177 * having to care about freeing intermediate values, as in the example below :
4178 *
4179 * memprintf(&err, "invalid argument: '%s'", arg);
4180 * ...
4181 * memprintf(&err, "parser said : <%s>\n", *err);
4182 * ...
4183 * free(*err);
4184 *
4185 * This means that <err> must be initialized to NULL before first invocation.
4186 * The return value also holds the allocated string, which eases error checking
4187 * and immediate consumption. If the output pointer is not used, NULL must be
Willy Tarreaueb6cead2012-09-20 19:43:14 +02004188 * passed instead and it will be ignored. The returned message will then also
4189 * be NULL so that the caller does not have to bother with freeing anything.
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004190 *
4191 * It is also convenient to use it without any free except the last one :
4192 * err = NULL;
4193 * if (!fct1(err)) report(*err);
4194 * if (!fct2(err)) report(*err);
4195 * if (!fct3(err)) report(*err);
4196 * free(*err);
Christopher Faulet93a518f2017-10-24 11:25:33 +02004197 *
4198 * memprintf relies on memvprintf. This last version can be called from any
4199 * function with variadic arguments.
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004200 */
Christopher Faulet93a518f2017-10-24 11:25:33 +02004201char *memvprintf(char **out, const char *format, va_list orig_args)
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004202{
4203 va_list args;
4204 char *ret = NULL;
4205 int allocated = 0;
4206 int needed = 0;
4207
Willy Tarreaueb6cead2012-09-20 19:43:14 +02004208 if (!out)
4209 return NULL;
4210
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004211 do {
Willy Tarreaue0609f52019-03-29 19:13:23 +01004212 char buf1;
4213
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004214 /* vsnprintf() will return the required length even when the
4215 * target buffer is NULL. We do this in a loop just in case
4216 * intermediate evaluations get wrong.
4217 */
Christopher Faulet93a518f2017-10-24 11:25:33 +02004218 va_copy(args, orig_args);
Willy Tarreaue0609f52019-03-29 19:13:23 +01004219 needed = vsnprintf(ret ? ret : &buf1, allocated, format, args);
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004220 va_end(args);
Willy Tarreau1b2fed62013-04-01 22:48:54 +02004221 if (needed < allocated) {
4222 /* Note: on Solaris 8, the first iteration always
4223 * returns -1 if allocated is zero, so we force a
4224 * retry.
4225 */
4226 if (!allocated)
4227 needed = 0;
4228 else
4229 break;
4230 }
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004231
Willy Tarreau1b2fed62013-04-01 22:48:54 +02004232 allocated = needed + 1;
Hubert Verstraete831962e2016-06-28 22:44:26 +02004233 ret = my_realloc2(ret, allocated);
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004234 } while (ret);
4235
4236 if (needed < 0) {
4237 /* an error was encountered */
Willy Tarreau61cfdf42021-02-20 10:46:51 +01004238 ha_free(&ret);
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004239 }
4240
4241 if (out) {
4242 free(*out);
4243 *out = ret;
4244 }
4245
4246 return ret;
4247}
William Lallemand421f5b52012-02-06 18:15:57 +01004248
Christopher Faulet93a518f2017-10-24 11:25:33 +02004249char *memprintf(char **out, const char *format, ...)
4250{
4251 va_list args;
4252 char *ret = NULL;
4253
4254 va_start(args, format);
4255 ret = memvprintf(out, format, args);
4256 va_end(args);
4257
4258 return ret;
4259}
4260
Willy Tarreau21c705b2012-09-14 11:40:36 +02004261/* Used to add <level> spaces before each line of <out>, unless there is only one line.
4262 * The input argument is automatically freed and reassigned. The result will have to be
Willy Tarreau70eec382012-10-10 08:56:47 +02004263 * freed by the caller. It also supports being passed a NULL which results in the same
4264 * output.
Willy Tarreau21c705b2012-09-14 11:40:36 +02004265 * Example of use :
4266 * parse(cmd, &err); (callee: memprintf(&err, ...))
4267 * fprintf(stderr, "Parser said: %s\n", indent_error(&err));
4268 * free(err);
4269 */
4270char *indent_msg(char **out, int level)
4271{
4272 char *ret, *in, *p;
4273 int needed = 0;
4274 int lf = 0;
4275 int lastlf = 0;
4276 int len;
4277
Willy Tarreau70eec382012-10-10 08:56:47 +02004278 if (!out || !*out)
4279 return NULL;
4280
Willy Tarreau21c705b2012-09-14 11:40:36 +02004281 in = *out - 1;
4282 while ((in = strchr(in + 1, '\n')) != NULL) {
4283 lastlf = in - *out;
4284 lf++;
4285 }
4286
4287 if (!lf) /* single line, no LF, return it as-is */
4288 return *out;
4289
4290 len = strlen(*out);
4291
4292 if (lf == 1 && lastlf == len - 1) {
4293 /* single line, LF at end, strip it and return as-is */
4294 (*out)[lastlf] = 0;
4295 return *out;
4296 }
4297
4298 /* OK now we have at least one LF, we need to process the whole string
4299 * as a multi-line string. What we'll do :
4300 * - prefix with an LF if there is none
4301 * - add <level> spaces before each line
4302 * This means at most ( 1 + level + (len-lf) + lf*<1+level) ) =
4303 * 1 + level + len + lf * level = 1 + level * (lf + 1) + len.
4304 */
4305
4306 needed = 1 + level * (lf + 1) + len + 1;
4307 p = ret = malloc(needed);
4308 in = *out;
4309
4310 /* skip initial LFs */
4311 while (*in == '\n')
4312 in++;
4313
4314 /* copy each line, prefixed with LF and <level> spaces, and without the trailing LF */
4315 while (*in) {
4316 *p++ = '\n';
4317 memset(p, ' ', level);
4318 p += level;
4319 do {
4320 *p++ = *in++;
4321 } while (*in && *in != '\n');
4322 if (*in)
4323 in++;
4324 }
4325 *p = 0;
4326
4327 free(*out);
4328 *out = ret;
4329
4330 return ret;
4331}
4332
Willy Tarreaua2c99112019-08-21 13:17:37 +02004333/* makes a copy of message <in> into <out>, with each line prefixed with <pfx>
4334 * and end of lines replaced with <eol> if not 0. The first line to indent has
4335 * to be indicated in <first> (starts at zero), so that it is possible to skip
4336 * indenting the first line if it has to be appended after an existing message.
4337 * Empty strings are never indented, and NULL strings are considered empty both
4338 * for <in> and <pfx>. It returns non-zero if an EOL was appended as the last
4339 * character, non-zero otherwise.
4340 */
4341int append_prefixed_str(struct buffer *out, const char *in, const char *pfx, char eol, int first)
4342{
4343 int bol, lf;
4344 int pfxlen = pfx ? strlen(pfx) : 0;
4345
4346 if (!in)
4347 return 0;
4348
4349 bol = 1;
4350 lf = 0;
4351 while (*in) {
4352 if (bol && pfxlen) {
4353 if (first > 0)
4354 first--;
4355 else
4356 b_putblk(out, pfx, pfxlen);
4357 bol = 0;
4358 }
4359
4360 lf = (*in == '\n');
4361 bol |= lf;
4362 b_putchr(out, (lf && eol) ? eol : *in);
4363 in++;
4364 }
4365 return lf;
4366}
4367
Willy Tarreau9d22e562019-03-29 18:49:09 +01004368/* removes environment variable <name> from the environment as found in
4369 * environ. This is only provided as an alternative for systems without
4370 * unsetenv() (old Solaris and AIX versions). THIS IS NOT THREAD SAFE.
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05004371 * The principle is to scan environ for each occurrence of variable name
Willy Tarreau9d22e562019-03-29 18:49:09 +01004372 * <name> and to replace the matching pointers with the last pointer of
4373 * the array (since variables are not ordered).
4374 * It always returns 0 (success).
4375 */
4376int my_unsetenv(const char *name)
4377{
4378 extern char **environ;
4379 char **p = environ;
4380 int vars;
4381 int next;
4382 int len;
4383
4384 len = strlen(name);
4385 for (vars = 0; p[vars]; vars++)
4386 ;
4387 next = 0;
4388 while (next < vars) {
4389 if (strncmp(p[next], name, len) != 0 || p[next][len] != '=') {
4390 next++;
4391 continue;
4392 }
4393 if (next < vars - 1)
4394 p[next] = p[vars - 1];
4395 p[--vars] = NULL;
4396 }
4397 return 0;
4398}
4399
Willy Tarreaudad36a32013-03-11 01:20:04 +01004400/* Convert occurrences of environment variables in the input string to their
4401 * corresponding value. A variable is identified as a series of alphanumeric
4402 * characters or underscores following a '$' sign. The <in> string must be
4403 * free()able. NULL returns NULL. The resulting string might be reallocated if
4404 * some expansion is made. Variable names may also be enclosed into braces if
4405 * needed (eg: to concatenate alphanum characters).
4406 */
4407char *env_expand(char *in)
4408{
4409 char *txt_beg;
4410 char *out;
4411 char *txt_end;
4412 char *var_beg;
4413 char *var_end;
4414 char *value;
4415 char *next;
4416 int out_len;
4417 int val_len;
4418
4419 if (!in)
4420 return in;
4421
4422 value = out = NULL;
4423 out_len = 0;
4424
4425 txt_beg = in;
4426 do {
4427 /* look for next '$' sign in <in> */
4428 for (txt_end = txt_beg; *txt_end && *txt_end != '$'; txt_end++);
4429
4430 if (!*txt_end && !out) /* end and no expansion performed */
4431 return in;
4432
4433 val_len = 0;
4434 next = txt_end;
4435 if (*txt_end == '$') {
4436 char save;
4437
4438 var_beg = txt_end + 1;
4439 if (*var_beg == '{')
4440 var_beg++;
4441
4442 var_end = var_beg;
Willy Tarreau90807112020-02-25 08:16:33 +01004443 while (isalnum((unsigned char)*var_end) || *var_end == '_') {
Willy Tarreaudad36a32013-03-11 01:20:04 +01004444 var_end++;
4445 }
4446
4447 next = var_end;
4448 if (*var_end == '}' && (var_beg > txt_end + 1))
4449 next++;
4450
4451 /* get value of the variable name at this location */
4452 save = *var_end;
4453 *var_end = '\0';
4454 value = getenv(var_beg);
4455 *var_end = save;
4456 val_len = value ? strlen(value) : 0;
4457 }
4458
Hubert Verstraete831962e2016-06-28 22:44:26 +02004459 out = my_realloc2(out, out_len + (txt_end - txt_beg) + val_len + 1);
Willy Tarreaudad36a32013-03-11 01:20:04 +01004460 if (txt_end > txt_beg) {
4461 memcpy(out + out_len, txt_beg, txt_end - txt_beg);
4462 out_len += txt_end - txt_beg;
4463 }
4464 if (val_len) {
4465 memcpy(out + out_len, value, val_len);
4466 out_len += val_len;
4467 }
4468 out[out_len] = 0;
4469 txt_beg = next;
4470 } while (*txt_beg);
4471
4472 /* here we know that <out> was allocated and that we don't need <in> anymore */
4473 free(in);
4474 return out;
4475}
4476
de Lafond Guillaume88c278f2013-04-15 19:27:10 +02004477
4478/* same as strstr() but case-insensitive and with limit length */
4479const char *strnistr(const char *str1, int len_str1, const char *str2, int len_str2)
4480{
4481 char *pptr, *sptr, *start;
Willy Tarreauc8746532014-05-28 23:05:07 +02004482 unsigned int slen, plen;
4483 unsigned int tmp1, tmp2;
de Lafond Guillaume88c278f2013-04-15 19:27:10 +02004484
4485 if (str1 == NULL || len_str1 == 0) // search pattern into an empty string => search is not found
4486 return NULL;
4487
4488 if (str2 == NULL || len_str2 == 0) // pattern is empty => every str1 match
4489 return str1;
4490
4491 if (len_str1 < len_str2) // pattern is longer than string => search is not found
4492 return NULL;
4493
4494 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 +02004495 while (toupper((unsigned char)*start) != toupper((unsigned char)*str2)) {
de Lafond Guillaume88c278f2013-04-15 19:27:10 +02004496 start++;
4497 slen--;
4498 tmp1++;
4499
4500 if (tmp1 >= len_str1)
4501 return NULL;
4502
4503 /* if pattern longer than string */
4504 if (slen < plen)
4505 return NULL;
4506 }
4507
4508 sptr = start;
4509 pptr = (char *)str2;
4510
4511 tmp2 = 0;
Willy Tarreauf278eec2020-07-05 21:46:32 +02004512 while (toupper((unsigned char)*sptr) == toupper((unsigned char)*pptr)) {
de Lafond Guillaume88c278f2013-04-15 19:27:10 +02004513 sptr++;
4514 pptr++;
4515 tmp2++;
4516
4517 if (*pptr == '\0' || tmp2 == len_str2) /* end of pattern found */
4518 return start;
4519 if (*sptr == '\0' || tmp2 == len_str1) /* end of string found and the pattern is not fully found */
4520 return NULL;
4521 }
4522 }
4523 return NULL;
4524}
4525
Willy Tarreau3ff476e2022-03-30 10:02:56 +02004526/* Returns true if s1 < s2 < s3 otherwise zero. Both s1 and s3 may be NULL and
4527 * in this case only non-null strings are compared. This allows to pass initial
4528 * values in iterators and in sort functions.
4529 */
4530int strordered(const char *s1, const char *s2, const char *s3)
4531{
4532 return (!s1 || strcmp(s1, s2) < 0) && (!s3 || strcmp(s2, s3) < 0);
4533}
4534
Thierry FOURNIER317e1c42014-08-12 10:20:47 +02004535/* This function read the next valid utf8 char.
4536 * <s> is the byte srray to be decode, <len> is its length.
4537 * The function returns decoded char encoded like this:
4538 * The 4 msb are the return code (UTF8_CODE_*), the 4 lsb
4539 * are the length read. The decoded character is stored in <c>.
4540 */
4541unsigned char utf8_next(const char *s, int len, unsigned int *c)
4542{
4543 const unsigned char *p = (unsigned char *)s;
4544 int dec;
4545 unsigned char code = UTF8_CODE_OK;
4546
4547 if (len < 1)
4548 return UTF8_CODE_OK;
4549
4550 /* Check the type of UTF8 sequence
4551 *
4552 * 0... .... 0x00 <= x <= 0x7f : 1 byte: ascii char
4553 * 10.. .... 0x80 <= x <= 0xbf : invalid sequence
4554 * 110. .... 0xc0 <= x <= 0xdf : 2 bytes
4555 * 1110 .... 0xe0 <= x <= 0xef : 3 bytes
4556 * 1111 0... 0xf0 <= x <= 0xf7 : 4 bytes
4557 * 1111 10.. 0xf8 <= x <= 0xfb : 5 bytes
4558 * 1111 110. 0xfc <= x <= 0xfd : 6 bytes
4559 * 1111 111. 0xfe <= x <= 0xff : invalid sequence
4560 */
4561 switch (*p) {
4562 case 0x00 ... 0x7f:
4563 *c = *p;
4564 return UTF8_CODE_OK | 1;
4565
4566 case 0x80 ... 0xbf:
4567 *c = *p;
4568 return UTF8_CODE_BADSEQ | 1;
4569
4570 case 0xc0 ... 0xdf:
4571 if (len < 2) {
4572 *c = *p;
4573 return UTF8_CODE_BADSEQ | 1;
4574 }
4575 *c = *p & 0x1f;
4576 dec = 1;
4577 break;
4578
4579 case 0xe0 ... 0xef:
4580 if (len < 3) {
4581 *c = *p;
4582 return UTF8_CODE_BADSEQ | 1;
4583 }
4584 *c = *p & 0x0f;
4585 dec = 2;
4586 break;
4587
4588 case 0xf0 ... 0xf7:
4589 if (len < 4) {
4590 *c = *p;
4591 return UTF8_CODE_BADSEQ | 1;
4592 }
4593 *c = *p & 0x07;
4594 dec = 3;
4595 break;
4596
4597 case 0xf8 ... 0xfb:
4598 if (len < 5) {
4599 *c = *p;
4600 return UTF8_CODE_BADSEQ | 1;
4601 }
4602 *c = *p & 0x03;
4603 dec = 4;
4604 break;
4605
4606 case 0xfc ... 0xfd:
4607 if (len < 6) {
4608 *c = *p;
4609 return UTF8_CODE_BADSEQ | 1;
4610 }
4611 *c = *p & 0x01;
4612 dec = 5;
4613 break;
4614
4615 case 0xfe ... 0xff:
4616 default:
4617 *c = *p;
4618 return UTF8_CODE_BADSEQ | 1;
4619 }
4620
4621 p++;
4622
4623 while (dec > 0) {
4624
4625 /* need 0x10 for the 2 first bits */
4626 if ( ( *p & 0xc0 ) != 0x80 )
4627 return UTF8_CODE_BADSEQ | ((p-(unsigned char *)s)&0xffff);
4628
4629 /* add data at char */
4630 *c = ( *c << 6 ) | ( *p & 0x3f );
4631
4632 dec--;
4633 p++;
4634 }
4635
4636 /* Check ovelong encoding.
4637 * 1 byte : 5 + 6 : 11 : 0x80 ... 0x7ff
4638 * 2 bytes : 4 + 6 + 6 : 16 : 0x800 ... 0xffff
4639 * 3 bytes : 3 + 6 + 6 + 6 : 21 : 0x10000 ... 0x1fffff
4640 */
Thierry FOURNIER9e7ec082015-03-12 19:32:38 +01004641 if (( *c <= 0x7f && (p-(unsigned char *)s) > 1) ||
Thierry FOURNIER317e1c42014-08-12 10:20:47 +02004642 (*c >= 0x80 && *c <= 0x7ff && (p-(unsigned char *)s) > 2) ||
4643 (*c >= 0x800 && *c <= 0xffff && (p-(unsigned char *)s) > 3) ||
4644 (*c >= 0x10000 && *c <= 0x1fffff && (p-(unsigned char *)s) > 4))
4645 code |= UTF8_CODE_OVERLONG;
4646
4647 /* Check invalid UTF8 range. */
4648 if ((*c >= 0xd800 && *c <= 0xdfff) ||
4649 (*c >= 0xfffe && *c <= 0xffff))
4650 code |= UTF8_CODE_INVRANGE;
4651
4652 return code | ((p-(unsigned char *)s)&0x0f);
4653}
4654
Maxime de Roucydc887852016-05-13 23:52:54 +02004655/* append a copy of string <str> (in a wordlist) at the end of the list <li>
4656 * On failure : return 0 and <err> filled with an error message.
4657 * The caller is responsible for freeing the <err> and <str> copy
4658 * memory area using free()
4659 */
4660int list_append_word(struct list *li, const char *str, char **err)
4661{
4662 struct wordlist *wl;
4663
4664 wl = calloc(1, sizeof(*wl));
4665 if (!wl) {
4666 memprintf(err, "out of memory");
4667 goto fail_wl;
4668 }
4669
4670 wl->s = strdup(str);
4671 if (!wl->s) {
4672 memprintf(err, "out of memory");
4673 goto fail_wl_s;
4674 }
4675
Willy Tarreau2b718102021-04-21 07:32:39 +02004676 LIST_APPEND(li, &wl->list);
Maxime de Roucydc887852016-05-13 23:52:54 +02004677
4678 return 1;
4679
4680fail_wl_s:
4681 free(wl->s);
4682fail_wl:
4683 free(wl);
4684 return 0;
4685}
4686
Willy Tarreau37101052019-05-20 16:48:20 +02004687/* indicates if a memory location may safely be read or not. The trick consists
4688 * in performing a harmless syscall using this location as an input and letting
4689 * the operating system report whether it's OK or not. For this we have the
4690 * stat() syscall, which will return EFAULT when the memory location supposed
4691 * to contain the file name is not readable. If it is readable it will then
4692 * either return 0 if the area contains an existing file name, or -1 with
4693 * another code. This must not be abused, and some audit systems might detect
4694 * this as abnormal activity. It's used only for unsafe dumps.
4695 */
4696int may_access(const void *ptr)
4697{
4698 struct stat buf;
4699
4700 if (stat(ptr, &buf) == 0)
4701 return 1;
4702 if (errno == EFAULT)
4703 return 0;
4704 return 1;
4705}
4706
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004707/* print a string of text buffer to <out>. The format is :
4708 * Non-printable chars \t, \n, \r and \e are * encoded in C format.
4709 * Other non-printable chars are encoded "\xHH". Space, '\', and '=' are also escaped.
4710 * Print stopped if null char or <bsize> is reached, or if no more place in the chunk.
4711 */
Willy Tarreau83061a82018-07-13 11:56:34 +02004712int dump_text(struct buffer *out, const char *buf, int bsize)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004713{
4714 unsigned char c;
Tim Duesterhus18795d42021-08-29 00:58:22 +02004715 size_t ptr = 0;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004716
Tim Duesterhus18795d42021-08-29 00:58:22 +02004717 while (ptr < bsize && buf[ptr]) {
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004718 c = buf[ptr];
Willy Tarreau90807112020-02-25 08:16:33 +01004719 if (isprint((unsigned char)c) && isascii((unsigned char)c) && c != '\\' && c != ' ' && c != '=') {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004720 if (out->data > out->size - 1)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004721 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004722 out->area[out->data++] = c;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004723 }
4724 else if (c == '\t' || c == '\n' || c == '\r' || c == '\e' || c == '\\' || c == ' ' || c == '=') {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004725 if (out->data > out->size - 2)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004726 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004727 out->area[out->data++] = '\\';
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004728 switch (c) {
4729 case ' ': c = ' '; break;
4730 case '\t': c = 't'; break;
4731 case '\n': c = 'n'; break;
4732 case '\r': c = 'r'; break;
4733 case '\e': c = 'e'; break;
4734 case '\\': c = '\\'; break;
4735 case '=': c = '='; break;
4736 }
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004737 out->area[out->data++] = c;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004738 }
4739 else {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004740 if (out->data > out->size - 4)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004741 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004742 out->area[out->data++] = '\\';
4743 out->area[out->data++] = 'x';
4744 out->area[out->data++] = hextab[(c >> 4) & 0xF];
4745 out->area[out->data++] = hextab[c & 0xF];
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004746 }
4747 ptr++;
4748 }
4749
4750 return ptr;
4751}
4752
4753/* print a buffer in hexa.
4754 * Print stopped if <bsize> is reached, or if no more place in the chunk.
4755 */
Willy Tarreau83061a82018-07-13 11:56:34 +02004756int dump_binary(struct buffer *out, const char *buf, int bsize)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004757{
4758 unsigned char c;
4759 int ptr = 0;
4760
4761 while (ptr < bsize) {
4762 c = buf[ptr];
4763
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004764 if (out->data > out->size - 2)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004765 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004766 out->area[out->data++] = hextab[(c >> 4) & 0xF];
4767 out->area[out->data++] = hextab[c & 0xF];
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004768
4769 ptr++;
4770 }
4771 return ptr;
4772}
4773
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004774/* Appends into buffer <out> a hex dump of memory area <buf> for <len> bytes,
4775 * prepending each line with prefix <pfx>. The output is *not* initialized.
4776 * The output will not wrap pas the buffer's end so it is more optimal if the
4777 * caller makes sure the buffer is aligned first. A trailing zero will always
4778 * be appended (and not counted) if there is room for it. The caller must make
Willy Tarreau37101052019-05-20 16:48:20 +02004779 * sure that the area is dumpable first. If <unsafe> is non-null, the memory
4780 * locations are checked first for being readable.
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004781 */
Willy Tarreau37101052019-05-20 16:48:20 +02004782void dump_hex(struct buffer *out, const char *pfx, const void *buf, int len, int unsafe)
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004783{
4784 const unsigned char *d = buf;
4785 int i, j, start;
4786
4787 d = (const unsigned char *)(((unsigned long)buf) & -16);
4788 start = ((unsigned long)buf) & 15;
4789
4790 for (i = 0; i < start + len; i += 16) {
4791 chunk_appendf(out, (sizeof(void *) == 4) ? "%s%8p: " : "%s%16p: ", pfx, d + i);
4792
Willy Tarreau37101052019-05-20 16:48:20 +02004793 // 0: unchecked, 1: checked safe, 2: danger
4794 unsafe = !!unsafe;
4795 if (unsafe && !may_access(d + i))
4796 unsafe = 2;
4797
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004798 for (j = 0; j < 16; j++) {
Willy Tarreau37101052019-05-20 16:48:20 +02004799 if ((i + j < start) || (i + j >= start + len))
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004800 chunk_strcat(out, "'' ");
Willy Tarreau37101052019-05-20 16:48:20 +02004801 else if (unsafe > 1)
4802 chunk_strcat(out, "** ");
4803 else
4804 chunk_appendf(out, "%02x ", d[i + j]);
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004805
4806 if (j == 7)
4807 chunk_strcat(out, "- ");
4808 }
4809 chunk_strcat(out, " ");
4810 for (j = 0; j < 16; j++) {
Willy Tarreau37101052019-05-20 16:48:20 +02004811 if ((i + j < start) || (i + j >= start + len))
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004812 chunk_strcat(out, "'");
Willy Tarreau37101052019-05-20 16:48:20 +02004813 else if (unsafe > 1)
4814 chunk_strcat(out, "*");
Willy Tarreau90807112020-02-25 08:16:33 +01004815 else if (isprint((unsigned char)d[i + j]))
Willy Tarreau37101052019-05-20 16:48:20 +02004816 chunk_appendf(out, "%c", d[i + j]);
4817 else
4818 chunk_strcat(out, ".");
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004819 }
4820 chunk_strcat(out, "\n");
4821 }
4822}
4823
Willy Tarreau762fb3e2020-03-03 15:57:10 +01004824/* dumps <pfx> followed by <n> bytes from <addr> in hex form into buffer <buf>
4825 * enclosed in brackets after the address itself, formatted on 14 chars
4826 * including the "0x" prefix. This is meant to be used as a prefix for code
4827 * areas. For example:
4828 * "0x7f10b6557690 [48 c7 c0 0f 00 00 00 0f]"
4829 * It relies on may_access() to know if the bytes are dumpable, otherwise "--"
4830 * is emitted. A NULL <pfx> will be considered empty.
4831 */
4832void dump_addr_and_bytes(struct buffer *buf, const char *pfx, const void *addr, int n)
4833{
4834 int ok = 0;
4835 int i;
4836
4837 chunk_appendf(buf, "%s%#14lx [", pfx ? pfx : "", (long)addr);
4838
4839 for (i = 0; i < n; i++) {
4840 if (i == 0 || (((long)(addr + i) ^ (long)(addr)) & 4096))
4841 ok = may_access(addr + i);
4842 if (ok)
4843 chunk_appendf(buf, "%02x%s", ((uint8_t*)addr)[i], (i<n-1) ? " " : "]");
4844 else
4845 chunk_appendf(buf, "--%s", (i<n-1) ? " " : "]");
4846 }
4847}
4848
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004849/* print a line of text buffer (limited to 70 bytes) to <out>. The format is :
4850 * <2 spaces> <offset=5 digits> <space or plus> <space> <70 chars max> <\n>
4851 * which is 60 chars per line. Non-printable chars \t, \n, \r and \e are
4852 * encoded in C format. Other non-printable chars are encoded "\xHH". Original
4853 * lines are respected within the limit of 70 output chars. Lines that are
4854 * continuation of a previous truncated line begin with "+" instead of " "
4855 * after the offset. The new pointer is returned.
4856 */
Willy Tarreau83061a82018-07-13 11:56:34 +02004857int dump_text_line(struct buffer *out, const char *buf, int bsize, int len,
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004858 int *line, int ptr)
4859{
4860 int end;
4861 unsigned char c;
4862
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004863 end = out->data + 80;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004864 if (end > out->size)
4865 return ptr;
4866
4867 chunk_appendf(out, " %05d%c ", ptr, (ptr == *line) ? ' ' : '+');
4868
4869 while (ptr < len && ptr < bsize) {
4870 c = buf[ptr];
Willy Tarreau90807112020-02-25 08:16:33 +01004871 if (isprint((unsigned char)c) && isascii((unsigned char)c) && c != '\\') {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004872 if (out->data > end - 2)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004873 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004874 out->area[out->data++] = c;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004875 } else if (c == '\t' || c == '\n' || c == '\r' || c == '\e' || c == '\\') {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004876 if (out->data > end - 3)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004877 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004878 out->area[out->data++] = '\\';
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004879 switch (c) {
4880 case '\t': c = 't'; break;
4881 case '\n': c = 'n'; break;
4882 case '\r': c = 'r'; break;
4883 case '\e': c = 'e'; break;
4884 case '\\': c = '\\'; break;
4885 }
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004886 out->area[out->data++] = c;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004887 } else {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004888 if (out->data > end - 5)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004889 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004890 out->area[out->data++] = '\\';
4891 out->area[out->data++] = 'x';
4892 out->area[out->data++] = hextab[(c >> 4) & 0xF];
4893 out->area[out->data++] = hextab[c & 0xF];
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004894 }
4895 if (buf[ptr++] == '\n') {
4896 /* we had a line break, let's return now */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004897 out->area[out->data++] = '\n';
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004898 *line = ptr;
4899 return ptr;
4900 }
4901 }
4902 /* we have an incomplete line, we return it as-is */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004903 out->area[out->data++] = '\n';
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004904 return ptr;
4905}
4906
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004907/* displays a <len> long memory block at <buf>, assuming first byte of <buf>
Willy Tarreaued936c52017-04-27 18:03:20 +02004908 * has address <baseaddr>. String <pfx> may be placed as a prefix in front of
4909 * each line. It may be NULL if unused. The output is emitted to file <out>.
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004910 */
Willy Tarreaued936c52017-04-27 18:03:20 +02004911void debug_hexdump(FILE *out, const char *pfx, const char *buf,
4912 unsigned int baseaddr, int len)
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004913{
Willy Tarreau73459792017-04-11 07:58:08 +02004914 unsigned int i;
4915 int b, j;
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004916
4917 for (i = 0; i < (len + (baseaddr & 15)); i += 16) {
4918 b = i - (baseaddr & 15);
Willy Tarreaued936c52017-04-27 18:03:20 +02004919 fprintf(out, "%s%08x: ", pfx ? pfx : "", i + (baseaddr & ~15));
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004920 for (j = 0; j < 8; j++) {
4921 if (b + j >= 0 && b + j < len)
4922 fprintf(out, "%02x ", (unsigned char)buf[b + j]);
4923 else
4924 fprintf(out, " ");
4925 }
4926
4927 if (b + j >= 0 && b + j < len)
4928 fputc('-', out);
4929 else
4930 fputc(' ', out);
4931
4932 for (j = 8; j < 16; j++) {
4933 if (b + j >= 0 && b + j < len)
4934 fprintf(out, " %02x", (unsigned char)buf[b + j]);
4935 else
4936 fprintf(out, " ");
4937 }
4938
4939 fprintf(out, " ");
4940 for (j = 0; j < 16; j++) {
4941 if (b + j >= 0 && b + j < len) {
4942 if (isprint((unsigned char)buf[b + j]))
4943 fputc((unsigned char)buf[b + j], out);
4944 else
4945 fputc('.', out);
4946 }
4947 else
4948 fputc(' ', out);
4949 }
4950 fputc('\n', out);
4951 }
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004952}
4953
Willy Tarreaubb869862020-04-16 10:52:41 +02004954/* Tries to report the executable path name on platforms supporting this. If
4955 * not found or not possible, returns NULL.
4956 */
4957const char *get_exec_path()
4958{
4959 const char *ret = NULL;
4960
David Carlier43a56852022-03-04 15:50:48 +00004961#if defined(__linux__) && defined(__GLIBC__) && (__GLIBC__ > 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 16))
Willy Tarreaubb869862020-04-16 10:52:41 +02004962 long execfn = getauxval(AT_EXECFN);
4963
4964 if (execfn && execfn != ENOENT)
4965 ret = (const char *)execfn;
devnexen@gmail.comc4e52322021-08-17 12:55:49 +01004966#elif defined(__FreeBSD__)
4967 Elf_Auxinfo *auxv;
4968 for (auxv = __elf_aux_vector; auxv->a_type != AT_NULL; ++auxv) {
4969 if (auxv->a_type == AT_EXECPATH) {
4970 ret = (const char *)auxv->a_un.a_ptr;
4971 break;
4972 }
4973 }
David Carlierbd2cced2021-08-17 08:44:25 +01004974#elif defined(__NetBSD__)
4975 AuxInfo *auxv;
4976 for (auxv = _dlauxinfo(); auxv->a_type != AT_NULL; ++auxv) {
4977 if (auxv->a_type == AT_SUN_EXECNAME) {
4978 ret = (const char *)auxv->a_v;
4979 break;
4980 }
4981 }
David Carlier7198c702022-05-14 17:15:49 +01004982#elif defined(__sun)
4983 ret = getexecname();
Willy Tarreaubb869862020-04-16 10:52:41 +02004984#endif
4985 return ret;
4986}
4987
Baruch Siache1651b22020-07-24 07:52:20 +03004988#if (defined(__ELF__) && !defined(__linux__)) || defined(USE_DL)
Willy Tarreau9133e482020-03-04 10:19:36 +01004989/* calls dladdr() or dladdr1() on <addr> and <dli>. If dladdr1 is available,
4990 * also returns the symbol size in <size>, otherwise returns 0 there.
4991 */
4992static int dladdr_and_size(const void *addr, Dl_info *dli, size_t *size)
4993{
4994 int ret;
Willy Tarreau7b2108c2021-08-30 10:15:35 +02004995#if defined(__GLIBC__) && (__GLIBC__ > 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 3)) // most detailed one
Willy Tarreauf3d5c4b2022-01-28 09:42:29 +01004996 const ElfW(Sym) *sym __attribute__((may_alias));
Willy Tarreau9133e482020-03-04 10:19:36 +01004997
4998 ret = dladdr1(addr, dli, (void **)&sym, RTLD_DL_SYMENT);
4999 if (ret)
5000 *size = sym ? sym->st_size : 0;
5001#else
David Carlierae5c42f2021-12-31 08:15:29 +00005002#if defined(__sun)
5003 ret = dladdr((void *)addr, dli);
5004#else
Willy Tarreau9133e482020-03-04 10:19:36 +01005005 ret = dladdr(addr, dli);
David Carlierae5c42f2021-12-31 08:15:29 +00005006#endif
Willy Tarreau9133e482020-03-04 10:19:36 +01005007 *size = 0;
5008#endif
5009 return ret;
5010}
Willy Tarreau64192392021-05-05 09:06:21 +02005011
Willy Tarreau5b3cd952022-07-18 13:58:17 +02005012/* Sets build_is_static to true if we detect a static build. Some older glibcs
5013 * tend to crash inside dlsym() in static builds, but tests show that at least
5014 * dladdr() still works (and will fail to resolve anything of course). Thus we
5015 * try to determine if we're on a static build to avoid calling dlsym() in this
5016 * case.
Willy Tarreau288dc1d2022-07-16 13:49:34 +02005017 */
Willy Tarreau5b3cd952022-07-18 13:58:17 +02005018void check_if_static_build()
Willy Tarreau288dc1d2022-07-16 13:49:34 +02005019{
Willy Tarreau5b3cd952022-07-18 13:58:17 +02005020 Dl_info dli = { };
5021 size_t size = 0;
5022
5023 /* Now let's try to be smarter */
5024 if (!dladdr_and_size(&main, &dli, &size))
5025 build_is_static = 1;
5026 else
5027 build_is_static = 0;
Willy Tarreau288dc1d2022-07-16 13:49:34 +02005028}
5029
Willy Tarreau5b3cd952022-07-18 13:58:17 +02005030INITCALL0(STG_PREPARE, check_if_static_build);
Willy Tarreau288dc1d2022-07-16 13:49:34 +02005031
Willy Tarreau64192392021-05-05 09:06:21 +02005032/* Tries to retrieve the address of the first occurrence symbol <name>.
5033 * Note that NULL in return is not always an error as a symbol may have that
5034 * address in special situations.
5035 */
5036void *get_sym_curr_addr(const char *name)
5037{
5038 void *ptr = NULL;
5039
5040#ifdef RTLD_DEFAULT
Willy Tarreau5b3cd952022-07-18 13:58:17 +02005041 if (!build_is_static)
Willy Tarreau288dc1d2022-07-16 13:49:34 +02005042 ptr = dlsym(RTLD_DEFAULT, name);
Willy Tarreau64192392021-05-05 09:06:21 +02005043#endif
5044 return ptr;
5045}
5046
5047
5048/* Tries to retrieve the address of the next occurrence of symbol <name>
5049 * Note that NULL in return is not always an error as a symbol may have that
5050 * address in special situations.
5051 */
5052void *get_sym_next_addr(const char *name)
5053{
5054 void *ptr = NULL;
5055
5056#ifdef RTLD_NEXT
Willy Tarreau5b3cd952022-07-18 13:58:17 +02005057 if (!build_is_static)
Willy Tarreau288dc1d2022-07-16 13:49:34 +02005058 ptr = dlsym(RTLD_NEXT, name);
Willy Tarreau9133e482020-03-04 10:19:36 +01005059#endif
Willy Tarreau64192392021-05-05 09:06:21 +02005060 return ptr;
5061}
5062
5063#else /* elf & linux & dl */
5064
5065/* no possible resolving on other platforms at the moment */
5066void *get_sym_curr_addr(const char *name)
5067{
5068 return NULL;
5069}
5070
5071void *get_sym_next_addr(const char *name)
5072{
5073 return NULL;
5074}
5075
5076#endif /* elf & linux & dl */
Willy Tarreau9133e482020-03-04 10:19:36 +01005077
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01005078/* Tries to append to buffer <buf> some indications about the symbol at address
5079 * <addr> using the following form:
5080 * lib:+0xoffset (unresolvable address from lib's base)
5081 * main+0xoffset (unresolvable address from main (+/-))
5082 * lib:main+0xoffset (unresolvable lib address from main (+/-))
5083 * name (resolved exact exec address)
5084 * lib:name (resolved exact lib address)
5085 * name+0xoffset/0xsize (resolved address within exec symbol)
5086 * lib:name+0xoffset/0xsize (resolved address within lib symbol)
5087 *
5088 * The file name (lib or executable) is limited to what lies between the last
5089 * '/' and the first following '.'. An optional prefix <pfx> is prepended before
5090 * 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 +03005091 * that contains the "main" symbol, or when __ELF__ && USE_DL are not set.
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01005092 *
5093 * The symbol's base address is returned, or NULL when unresolved, in order to
5094 * allow the caller to match it against known ones.
5095 */
Willy Tarreau45fd1032021-01-20 14:37:59 +01005096const void *resolve_sym_name(struct buffer *buf, const char *pfx, const void *addr)
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01005097{
5098 const struct {
5099 const void *func;
5100 const char *name;
5101 } fcts[] = {
5102 { .func = process_stream, .name = "process_stream" },
5103 { .func = task_run_applet, .name = "task_run_applet" },
Willy Tarreau462b9892022-05-18 18:06:53 +02005104 { .func = sc_conn_io_cb, .name = "sc_conn_io_cb" },
Willy Tarreau586f71b2020-12-11 15:54:36 +01005105 { .func = sock_conn_iocb, .name = "sock_conn_iocb" },
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01005106 { .func = dgram_fd_handler, .name = "dgram_fd_handler" },
5107 { .func = listener_accept, .name = "listener_accept" },
Willy Tarreaud597ec22021-01-29 14:29:06 +01005108 { .func = manage_global_listener_queue, .name = "manage_global_listener_queue" },
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01005109 { .func = poller_pipe_io_handler, .name = "poller_pipe_io_handler" },
5110 { .func = mworker_accept_wrapper, .name = "mworker_accept_wrapper" },
Willy Tarreau02922e12021-01-29 12:27:57 +01005111 { .func = session_expire_embryonic, .name = "session_expire_embryonic" },
Willy Tarreaufb5401f2021-01-29 12:25:23 +01005112#ifdef USE_THREAD
5113 { .func = accept_queue_process, .name = "accept_queue_process" },
5114#endif
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01005115#ifdef USE_LUA
5116 { .func = hlua_process_task, .name = "hlua_process_task" },
5117#endif
Ilya Shipitsinbdec3ba2020-11-14 01:56:34 +05005118#ifdef SSL_MODE_ASYNC
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01005119 { .func = ssl_async_fd_free, .name = "ssl_async_fd_free" },
5120 { .func = ssl_async_fd_handler, .name = "ssl_async_fd_handler" },
5121#endif
Willy Tarreau8f6da642023-03-10 12:04:02 +01005122#ifdef USE_QUIC
5123 { .func = quic_conn_sock_fd_iocb, .name = "quic_conn_sock_fd_iocb" },
5124#endif
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01005125 };
5126
Baruch Siache1651b22020-07-24 07:52:20 +03005127#if (defined(__ELF__) && !defined(__linux__)) || defined(USE_DL)
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01005128 Dl_info dli, dli_main;
Willy Tarreau9133e482020-03-04 10:19:36 +01005129 size_t size;
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01005130 const char *fname, *p;
5131#endif
5132 int i;
5133
5134 if (pfx)
5135 chunk_appendf(buf, "%s", pfx);
5136
5137 for (i = 0; i < sizeof(fcts) / sizeof(fcts[0]); i++) {
5138 if (addr == fcts[i].func) {
5139 chunk_appendf(buf, "%s", fcts[i].name);
5140 return addr;
5141 }
5142 }
5143
Baruch Siache1651b22020-07-24 07:52:20 +03005144#if (defined(__ELF__) && !defined(__linux__)) || defined(USE_DL)
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01005145 /* Now let's try to be smarter */
Willy Tarreau9133e482020-03-04 10:19:36 +01005146 if (!dladdr_and_size(addr, &dli, &size))
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01005147 goto unknown;
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01005148
5149 /* 1. prefix the library name if it's not the same object as the one
5150 * that contains the main function. The name is picked between last '/'
5151 * and first following '.'.
5152 */
5153 if (!dladdr(main, &dli_main))
5154 dli_main.dli_fbase = NULL;
5155
5156 if (dli_main.dli_fbase != dli.dli_fbase) {
5157 fname = dli.dli_fname;
5158 p = strrchr(fname, '/');
5159 if (p++)
5160 fname = p;
5161 p = strchr(fname, '.');
5162 if (!p)
5163 p = fname + strlen(fname);
5164
5165 chunk_appendf(buf, "%.*s:", (int)(long)(p - fname), fname);
5166 }
5167
5168 /* 2. symbol name */
5169 if (dli.dli_sname) {
5170 /* known, dump it and return symbol's address (exact or relative) */
5171 chunk_appendf(buf, "%s", dli.dli_sname);
5172 if (addr != dli.dli_saddr) {
5173 chunk_appendf(buf, "+%#lx", (long)(addr - dli.dli_saddr));
Willy Tarreau9133e482020-03-04 10:19:36 +01005174 if (size)
5175 chunk_appendf(buf, "/%#lx", (long)size);
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01005176 }
5177 return dli.dli_saddr;
5178 }
5179 else if (dli_main.dli_fbase != dli.dli_fbase) {
5180 /* unresolved symbol from a known library, report relative offset */
5181 chunk_appendf(buf, "+%#lx", (long)(addr - dli.dli_fbase));
5182 return NULL;
5183 }
Baruch Siache1651b22020-07-24 07:52:20 +03005184#endif /* __ELF__ && !__linux__ || USE_DL */
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01005185 unknown:
5186 /* unresolved symbol from the main file, report relative offset to main */
5187 if ((void*)addr < (void*)main)
5188 chunk_appendf(buf, "main-%#lx", (long)((void*)main - addr));
5189 else
5190 chunk_appendf(buf, "main+%#lx", (long)(addr - (void*)main));
5191 return NULL;
Willy Tarreau0ebb5112016-12-05 00:10:57 +01005192}
5193
Willy Tarreau6ab7b212021-12-28 09:57:10 +01005194/* On systems where this is supported, let's provide a possibility to enumerate
5195 * the list of object files. The output is appended to a buffer initialized by
5196 * the caller, with one name per line. A trailing zero is always emitted if data
5197 * are written. Only real objects are dumped (executable and .so libs). The
5198 * function returns non-zero if it dumps anything. These functions do not make
5199 * use of the trash so that it is possible for the caller to call them with the
5200 * trash on input. The output format may be platform-specific but at least one
5201 * version must emit raw object file names when argument is zero.
5202 */
5203#if defined(HA_HAVE_DUMP_LIBS)
5204# if defined(HA_HAVE_DL_ITERATE_PHDR)
5205/* the private <data> we pass below is a dump context initialized like this */
5206struct dl_dump_ctx {
5207 struct buffer *buf;
5208 int with_addr;
5209};
5210
5211static int dl_dump_libs_cb(struct dl_phdr_info *info, size_t size, void *data)
5212{
5213 struct dl_dump_ctx *ctx = data;
5214 const char *fname;
5215 size_t p1, p2, beg, end;
5216 int idx;
5217
5218 if (!info || !info->dlpi_name)
5219 goto leave;
5220
5221 if (!*info->dlpi_name)
5222 fname = get_exec_path();
5223 else if (strchr(info->dlpi_name, '/'))
5224 fname = info->dlpi_name;
5225 else
5226 /* else it's a VDSO or similar and we're not interested */
5227 goto leave;
5228
5229 if (!ctx->with_addr)
5230 goto dump_name;
5231
5232 /* virtual addresses are relative to the load address and are per
5233 * pseudo-header, so we have to scan them all to find the furthest
5234 * one from the beginning. In this case we only dump entries if
5235 * they have at least one section.
5236 */
5237 beg = ~0; end = 0;
5238 for (idx = 0; idx < info->dlpi_phnum; idx++) {
5239 if (!info->dlpi_phdr[idx].p_memsz)
5240 continue;
5241 p1 = info->dlpi_phdr[idx].p_vaddr;
5242 if (p1 < beg)
5243 beg = p1;
5244 p2 = p1 + info->dlpi_phdr[idx].p_memsz - 1;
5245 if (p2 > end)
5246 end = p2;
5247 }
5248
5249 if (!idx)
5250 goto leave;
5251
5252 chunk_appendf(ctx->buf, "0x%012llx-0x%012llx (0x%07llx) ",
5253 (ullong)info->dlpi_addr + beg,
5254 (ullong)info->dlpi_addr + end,
5255 (ullong)(end - beg + 1));
5256 dump_name:
5257 chunk_appendf(ctx->buf, "%s\n", fname);
5258 leave:
5259 return 0;
5260}
5261
5262/* dumps lib names and optionally address ranges */
5263int dump_libs(struct buffer *output, int with_addr)
5264{
5265 struct dl_dump_ctx ctx = { .buf = output, .with_addr = with_addr };
5266 size_t old_data = output->data;
5267
5268 dl_iterate_phdr(dl_dump_libs_cb, &ctx);
5269 return output->data != old_data;
5270}
5271# else // no DL_ITERATE_PHDR
5272# error "No dump_libs() function for this platform"
5273# endif
5274#else // no HA_HAVE_DUMP_LIBS
5275
5276/* unsupported platform: do not dump anything */
5277int dump_libs(struct buffer *output, int with_addr)
5278{
5279 return 0;
5280}
5281
5282#endif // HA_HAVE_DUMP_LIBS
5283
Frédéric Lécaille3b717162019-02-25 15:04:22 +01005284/*
5285 * Allocate an array of unsigned int with <nums> as address from <str> string
Ilya Shipitsin46a030c2020-07-05 16:36:08 +05005286 * made of integer separated by dot characters.
Frédéric Lécaille3b717162019-02-25 15:04:22 +01005287 *
5288 * First, initializes the value with <sz> as address to 0 and initializes the
5289 * array with <nums> as address to NULL. Then allocates the array with <nums> as
5290 * address updating <sz> pointed value to the size of this array.
5291 *
5292 * Returns 1 if succeeded, 0 if not.
5293 */
5294int parse_dotted_uints(const char *str, unsigned int **nums, size_t *sz)
5295{
5296 unsigned int *n;
5297 const char *s, *end;
5298
5299 s = str;
5300 *sz = 0;
5301 end = str + strlen(str);
5302 *nums = n = NULL;
5303
5304 while (1) {
5305 unsigned int r;
5306
5307 if (s >= end)
5308 break;
5309
5310 r = read_uint(&s, end);
5311 /* Expected characters after having read an uint: '\0' or '.',
5312 * if '.', must not be terminal.
5313 */
Christopher Faulet4b524122021-02-11 10:42:41 +01005314 if (*s != '\0'&& (*s++ != '.' || s == end)) {
5315 free(n);
Frédéric Lécaille3b717162019-02-25 15:04:22 +01005316 return 0;
Christopher Faulet4b524122021-02-11 10:42:41 +01005317 }
Frédéric Lécaille3b717162019-02-25 15:04:22 +01005318
Frédéric Lécaille12a71842019-02-26 18:19:48 +01005319 n = my_realloc2(n, (*sz + 1) * sizeof *n);
Frédéric Lécaille3b717162019-02-25 15:04:22 +01005320 if (!n)
5321 return 0;
5322
5323 n[(*sz)++] = r;
5324 }
5325 *nums = n;
5326
5327 return 1;
5328}
5329
Willy Tarreau4d589e72019-08-23 19:02:26 +02005330
5331/* returns the number of bytes needed to encode <v> as a varint. An inline
5332 * version exists for use with constants (__varint_bytes()).
5333 */
5334int varint_bytes(uint64_t v)
5335{
5336 int len = 1;
5337
5338 if (v >= 240) {
5339 v = (v - 240) >> 4;
5340 while (1) {
5341 len++;
5342 if (v < 128)
5343 break;
5344 v = (v - 128) >> 7;
5345 }
5346 }
5347 return len;
5348}
5349
Willy Tarreau52bf8392020-03-08 00:42:37 +01005350
5351/* Random number generator state, see below */
Willy Tarreau1544c142020-03-12 00:31:18 +01005352static uint64_t ha_random_state[2] ALIGNED(2*sizeof(uint64_t));
Willy Tarreau52bf8392020-03-08 00:42:37 +01005353
5354/* This is a thread-safe implementation of xoroshiro128** described below:
5355 * http://prng.di.unimi.it/
5356 * It features a 2^128 long sequence, returns 64 high-quality bits on each call,
5357 * supports fast jumps and passes all common quality tests. It is thread-safe,
5358 * uses a double-cas on 64-bit architectures supporting it, and falls back to a
5359 * local lock on other ones.
5360 */
5361uint64_t ha_random64()
5362{
Willy Tarreau1544c142020-03-12 00:31:18 +01005363 uint64_t old[2] ALIGNED(2*sizeof(uint64_t));
5364 uint64_t new[2] ALIGNED(2*sizeof(uint64_t));
Willy Tarreau52bf8392020-03-08 00:42:37 +01005365
5366#if defined(USE_THREAD) && (!defined(HA_CAS_IS_8B) || !defined(HA_HAVE_CAS_DW))
5367 static HA_SPINLOCK_T rand_lock;
5368
5369 HA_SPIN_LOCK(OTHER_LOCK, &rand_lock);
5370#endif
5371
5372 old[0] = ha_random_state[0];
5373 old[1] = ha_random_state[1];
5374
5375#if defined(USE_THREAD) && defined(HA_CAS_IS_8B) && defined(HA_HAVE_CAS_DW)
5376 do {
5377#endif
Willy Tarreau52bf8392020-03-08 00:42:37 +01005378 new[1] = old[0] ^ old[1];
5379 new[0] = rotl64(old[0], 24) ^ new[1] ^ (new[1] << 16); // a, b
5380 new[1] = rotl64(new[1], 37); // c
5381
5382#if defined(USE_THREAD) && defined(HA_CAS_IS_8B) && defined(HA_HAVE_CAS_DW)
5383 } while (unlikely(!_HA_ATOMIC_DWCAS(ha_random_state, old, new)));
5384#else
5385 ha_random_state[0] = new[0];
5386 ha_random_state[1] = new[1];
5387#if defined(USE_THREAD)
5388 HA_SPIN_UNLOCK(OTHER_LOCK, &rand_lock);
5389#endif
5390#endif
Willy Tarreaub2475a12021-05-09 10:26:14 +02005391 return rotl64(old[0] * 5, 7) * 9;
Willy Tarreau52bf8392020-03-08 00:42:37 +01005392}
5393
5394/* seeds the random state using up to <len> bytes from <seed>, starting with
5395 * the first non-zero byte.
5396 */
5397void ha_random_seed(const unsigned char *seed, size_t len)
5398{
5399 size_t pos;
5400
5401 /* the seed must not be all zeroes, so we pre-fill it with alternating
5402 * bits and overwrite part of them with the block starting at the first
5403 * non-zero byte from the seed.
5404 */
5405 memset(ha_random_state, 0x55, sizeof(ha_random_state));
5406
5407 for (pos = 0; pos < len; pos++)
5408 if (seed[pos] != 0)
5409 break;
5410
5411 if (pos == len)
5412 return;
5413
5414 seed += pos;
5415 len -= pos;
5416
5417 if (len > sizeof(ha_random_state))
5418 len = sizeof(ha_random_state);
5419
5420 memcpy(ha_random_state, seed, len);
5421}
5422
5423/* This causes a jump to (dist * 2^96) places in the pseudo-random sequence,
5424 * and is equivalent to calling ha_random64() as many times. It is used to
5425 * provide non-overlapping sequences of 2^96 numbers (~7*10^28) to up to 2^32
5426 * different generators (i.e. different processes after a fork). The <dist>
5427 * argument is the distance to jump to and is used in a loop so it rather not
5428 * be too large if the processing time is a concern.
5429 *
5430 * BEWARE: this function is NOT thread-safe and must not be called during
5431 * concurrent accesses to ha_random64().
5432 */
5433void ha_random_jump96(uint32_t dist)
5434{
5435 while (dist--) {
5436 uint64_t s0 = 0;
5437 uint64_t s1 = 0;
5438 int b;
5439
5440 for (b = 0; b < 64; b++) {
5441 if ((0xd2a98b26625eee7bULL >> b) & 1) {
5442 s0 ^= ha_random_state[0];
5443 s1 ^= ha_random_state[1];
5444 }
5445 ha_random64();
5446 }
5447
5448 for (b = 0; b < 64; b++) {
5449 if ((0xdddf9b1090aa7ac1ULL >> b) & 1) {
5450 s0 ^= ha_random_state[0];
5451 s1 ^= ha_random_state[1];
5452 }
5453 ha_random64();
5454 }
5455 ha_random_state[0] = s0;
5456 ha_random_state[1] = s1;
5457 }
5458}
5459
Willy Tarreauee3bcdd2020-03-08 17:48:17 +01005460/* Generates an RFC4122 UUID into chunk <output> which must be at least 37
5461 * bytes large.
5462 */
5463void ha_generate_uuid(struct buffer *output)
5464{
5465 uint32_t rnd[4];
5466 uint64_t last;
5467
5468 last = ha_random64();
5469 rnd[0] = last;
5470 rnd[1] = last >> 32;
5471
5472 last = ha_random64();
5473 rnd[2] = last;
5474 rnd[3] = last >> 32;
5475
5476 chunk_printf(output, "%8.8x-%4.4x-%4.4x-%4.4x-%12.12llx",
5477 rnd[0],
5478 rnd[1] & 0xFFFF,
5479 ((rnd[1] >> 16u) & 0xFFF) | 0x4000, // highest 4 bits indicate the uuid version
5480 (rnd[2] & 0x3FFF) | 0x8000, // the highest 2 bits indicate the UUID variant (10),
5481 (long long)((rnd[2] >> 14u) | ((uint64_t) rnd[3] << 18u)) & 0xFFFFFFFFFFFFull);
5482}
5483
5484
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005485/* only used by parse_line() below. It supports writing in place provided that
5486 * <in> is updated to the next location before calling it. In that case, the
5487 * char at <in> may be overwritten.
5488 */
5489#define EMIT_CHAR(x) \
5490 do { \
5491 char __c = (char)(x); \
5492 if ((opts & PARSE_OPT_INPLACE) && out+outpos > in) \
5493 err |= PARSE_ERR_OVERLAP; \
5494 if (outpos >= outmax) \
5495 err |= PARSE_ERR_TOOLARGE; \
5496 if (!err) \
5497 out[outpos] = __c; \
5498 outpos++; \
5499 } while (0)
5500
Ilya Shipitsin46a030c2020-07-05 16:36:08 +05005501/* Parse <in>, copy it into <out> split into isolated words whose pointers
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005502 * are put in <args>. If more than <outlen> bytes have to be emitted, the
5503 * extraneous ones are not emitted but <outlen> is updated so that the caller
5504 * knows how much to realloc. Similarly, <args> are not updated beyond <nbargs>
5505 * but the returned <nbargs> indicates how many were found. All trailing args
Willy Tarreau61dd44b2020-06-25 07:35:42 +02005506 * up to <nbargs> point to the trailing zero, and as long as <nbargs> is > 0,
5507 * it is guaranteed that at least one arg will point to the zero. It is safe
5508 * to call it with a NULL <args> if <nbargs> is 0.
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005509 *
5510 * <out> may overlap with <in> provided that it never goes further, in which
5511 * case the parser will accept to perform in-place parsing and unquoting/
5512 * unescaping but only if environment variables do not lead to expansion that
5513 * causes overlapping, otherwise the input string being destroyed, the error
5514 * will not be recoverable. Note that even during out-of-place <in> will
5515 * experience temporary modifications in-place for variable resolution and must
5516 * be writable, and will also receive zeroes to delimit words when using
5517 * in-place copy. Parsing options <opts> taken from PARSE_OPT_*. Return value
5518 * is zero on success otherwise a bitwise-or of PARSE_ERR_*. Upon error, the
5519 * starting point of the first invalid character sequence or unmatched
5520 * quote/brace is reported in <errptr> if not NULL. When using in-place parsing
5521 * error reporting might be difficult since zeroes will have been inserted into
5522 * the string. One solution for the caller may consist in replacing all args
5523 * delimiters with spaces in this case.
5524 */
Maximilian Mader29c6cd72021-06-06 00:50:21 +02005525uint32_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 +02005526{
5527 char *quote = NULL;
5528 char *brace = NULL;
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005529 char *word_expand = NULL;
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005530 unsigned char hex1, hex2;
5531 size_t outmax = *outlen;
Willy Tarreau61dd44b2020-06-25 07:35:42 +02005532 int argsmax = *nbargs - 1;
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005533 size_t outpos = 0;
5534 int squote = 0;
5535 int dquote = 0;
5536 int arg = 0;
5537 uint32_t err = 0;
5538
5539 *nbargs = 0;
5540 *outlen = 0;
5541
Willy Tarreau61dd44b2020-06-25 07:35:42 +02005542 /* argsmax may be -1 here, protecting args[] from any write */
5543 if (arg < argsmax)
5544 args[arg] = out;
5545
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005546 while (1) {
5547 if (*in >= '-' && *in != '\\') {
5548 /* speedup: directly send all regular chars starting
5549 * with '-', '.', '/', alnum etc...
5550 */
5551 EMIT_CHAR(*in++);
5552 continue;
5553 }
5554 else if (*in == '\0' || *in == '\n' || *in == '\r') {
5555 /* end of line */
5556 break;
5557 }
5558 else if (*in == '#' && (opts & PARSE_OPT_SHARP) && !squote && !dquote) {
5559 /* comment */
5560 break;
5561 }
5562 else if (*in == '"' && !squote && (opts & PARSE_OPT_DQUOTE)) { /* double quote outside single quotes */
5563 if (dquote) {
5564 dquote = 0;
5565 quote = NULL;
5566 }
5567 else {
5568 dquote = 1;
5569 quote = in;
5570 }
5571 in++;
5572 continue;
5573 }
5574 else if (*in == '\'' && !dquote && (opts & PARSE_OPT_SQUOTE)) { /* single quote outside double quotes */
5575 if (squote) {
5576 squote = 0;
5577 quote = NULL;
5578 }
5579 else {
5580 squote = 1;
5581 quote = in;
5582 }
5583 in++;
5584 continue;
5585 }
5586 else if (*in == '\\' && !squote && (opts & PARSE_OPT_BKSLASH)) {
5587 /* first, we'll replace \\, \<space>, \#, \r, \n, \t, \xXX with their
5588 * C equivalent value but only when they have a special meaning and within
5589 * double quotes for some of them. Other combinations left unchanged (eg: \1).
5590 */
5591 char tosend = *in;
5592
5593 switch (in[1]) {
5594 case ' ':
5595 case '\\':
5596 tosend = in[1];
5597 in++;
5598 break;
5599
5600 case 't':
5601 tosend = '\t';
5602 in++;
5603 break;
5604
5605 case 'n':
5606 tosend = '\n';
5607 in++;
5608 break;
5609
5610 case 'r':
5611 tosend = '\r';
5612 in++;
5613 break;
5614
5615 case '#':
5616 /* escaping of "#" only if comments are supported */
5617 if (opts & PARSE_OPT_SHARP)
5618 in++;
5619 tosend = *in;
5620 break;
5621
5622 case '\'':
5623 /* escaping of "'" only outside single quotes and only if single quotes are supported */
5624 if (opts & PARSE_OPT_SQUOTE && !squote)
5625 in++;
5626 tosend = *in;
5627 break;
5628
5629 case '"':
5630 /* escaping of '"' only outside single quotes and only if double quotes are supported */
5631 if (opts & PARSE_OPT_DQUOTE && !squote)
5632 in++;
5633 tosend = *in;
5634 break;
5635
5636 case '$':
5637 /* escaping of '$' only inside double quotes and only if env supported */
5638 if (opts & PARSE_OPT_ENV && dquote)
5639 in++;
5640 tosend = *in;
5641 break;
5642
5643 case 'x':
5644 if (!ishex(in[2]) || !ishex(in[3])) {
5645 /* invalid or incomplete hex sequence */
5646 err |= PARSE_ERR_HEX;
5647 if (errptr)
5648 *errptr = in;
5649 goto leave;
5650 }
Willy Tarreauf278eec2020-07-05 21:46:32 +02005651 hex1 = toupper((unsigned char)in[2]) - '0';
5652 hex2 = toupper((unsigned char)in[3]) - '0';
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005653 if (hex1 > 9) hex1 -= 'A' - '9' - 1;
5654 if (hex2 > 9) hex2 -= 'A' - '9' - 1;
5655 tosend = (hex1 << 4) + hex2;
5656 in += 3;
5657 break;
5658
5659 default:
5660 /* other combinations are not escape sequences */
5661 break;
5662 }
5663
5664 in++;
5665 EMIT_CHAR(tosend);
5666 }
5667 else if (isspace((unsigned char)*in) && !squote && !dquote) {
5668 /* a non-escaped space is an argument separator */
5669 while (isspace((unsigned char)*in))
5670 in++;
5671 EMIT_CHAR(0);
5672 arg++;
5673 if (arg < argsmax)
5674 args[arg] = out + outpos;
5675 else
5676 err |= PARSE_ERR_TOOMANY;
5677 }
5678 else if (*in == '$' && (opts & PARSE_OPT_ENV) && (dquote || !(opts & PARSE_OPT_DQUOTE))) {
5679 /* environment variables are evaluated anywhere, or only
5680 * inside double quotes if they are supported.
5681 */
5682 char *var_name;
5683 char save_char;
Willy Tarreaua46f1af2021-05-06 10:25:11 +02005684 const char *value;
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005685
5686 in++;
5687
5688 if (*in == '{')
5689 brace = in++;
5690
Willy Tarreaua46f1af2021-05-06 10:25:11 +02005691 if (!isalpha((unsigned char)*in) && *in != '_' && *in != '.') {
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005692 /* unacceptable character in variable name */
5693 err |= PARSE_ERR_VARNAME;
5694 if (errptr)
5695 *errptr = in;
5696 goto leave;
5697 }
5698
5699 var_name = in;
Willy Tarreaua46f1af2021-05-06 10:25:11 +02005700 if (*in == '.')
5701 in++;
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005702 while (isalnum((unsigned char)*in) || *in == '_')
5703 in++;
5704
5705 save_char = *in;
5706 *in = '\0';
Willy Tarreaua46f1af2021-05-06 10:25:11 +02005707 if (unlikely(*var_name == '.')) {
5708 /* internal pseudo-variables */
5709 if (strcmp(var_name, ".LINE") == 0)
5710 value = ultoa(global.cfg_curr_line);
5711 else if (strcmp(var_name, ".FILE") == 0)
5712 value = global.cfg_curr_file;
5713 else if (strcmp(var_name, ".SECTION") == 0)
5714 value = global.cfg_curr_section;
5715 else {
5716 /* unsupported internal variable name */
5717 err |= PARSE_ERR_VARNAME;
5718 if (errptr)
5719 *errptr = var_name;
5720 goto leave;
5721 }
5722 } else {
5723 value = getenv(var_name);
5724 }
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005725 *in = save_char;
5726
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005727 /* support for '[*]' sequence to force word expansion,
5728 * only available inside braces */
5729 if (*in == '[' && brace && (opts & PARSE_OPT_WORD_EXPAND)) {
5730 word_expand = in++;
5731
5732 if (*in++ != '*' || *in++ != ']') {
5733 err |= PARSE_ERR_WRONG_EXPAND;
5734 if (errptr)
5735 *errptr = word_expand;
5736 goto leave;
5737 }
5738 }
5739
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005740 if (brace) {
Willy Tarreauec347b12021-11-18 17:42:50 +01005741 if (*in == '-') {
5742 /* default value starts just after the '-' */
5743 if (!value)
5744 value = in + 1;
5745
5746 while (*in && *in != '}')
5747 in++;
5748 if (!*in)
5749 goto no_brace;
5750 *in = 0; // terminate the default value
5751 }
5752 else if (*in != '}') {
5753 no_brace:
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005754 /* unmatched brace */
5755 err |= PARSE_ERR_BRACE;
5756 if (errptr)
5757 *errptr = brace;
5758 goto leave;
5759 }
Willy Tarreauec347b12021-11-18 17:42:50 +01005760
5761 /* brace found, skip it */
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005762 in++;
5763 brace = NULL;
5764 }
5765
5766 if (value) {
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005767 while (*value) {
5768 /* expand as individual parameters on a space character */
Willy Tarreaufe2cc412020-10-01 18:04:40 +02005769 if (word_expand && isspace((unsigned char)*value)) {
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005770 EMIT_CHAR(0);
5771 ++arg;
5772 if (arg < argsmax)
5773 args[arg] = out + outpos;
5774 else
5775 err |= PARSE_ERR_TOOMANY;
5776
5777 /* skip consecutive spaces */
Willy Tarreaufe2cc412020-10-01 18:04:40 +02005778 while (isspace((unsigned char)*++value))
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005779 ;
5780 } else {
5781 EMIT_CHAR(*value++);
5782 }
5783 }
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005784 }
Frédéric Lécaille51fad642023-11-30 09:26:05 +01005785 else {
5786 /* An unmatched environment variable was parsed.
5787 * Let's skip the trailing double-quote character
5788 * and spaces.
5789 */
5790 if (likely(*var_name != '.') && *in == '"') {
5791 in++;
5792 while (isspace((unsigned char)*in))
5793 in++;
5794 if (dquote) {
5795 dquote = 0;
5796 quote = NULL;
5797 }
5798 }
5799 }
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005800 word_expand = NULL;
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005801 }
5802 else {
5803 /* any other regular char */
5804 EMIT_CHAR(*in++);
5805 }
5806 }
5807
5808 /* end of output string */
5809 EMIT_CHAR(0);
Erwan Le Goas8a6767d2022-09-23 15:06:34 +02005810
Willy Tarreau94ab1392022-10-03 08:27:55 +02005811 /* Don't add an empty arg after trailing spaces. Note that args[arg]
5812 * may contain some distances relative to NULL if <out> was NULL, or
5813 * pointers beyond the end of <out> in case <outlen> is too short, thus
5814 * we must not dereference it.
Erwan Le Goas8a6767d2022-09-23 15:06:34 +02005815 */
Willy Tarreau94ab1392022-10-03 08:27:55 +02005816 if (arg < argsmax && args[arg] != out + outpos - 1)
Erwan Le Goas8a6767d2022-09-23 15:06:34 +02005817 arg++;
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005818
5819 if (quote) {
5820 /* unmatched quote */
5821 err |= PARSE_ERR_QUOTE;
5822 if (errptr)
5823 *errptr = quote;
5824 goto leave;
5825 }
5826 leave:
5827 *nbargs = arg;
5828 *outlen = outpos;
5829
Willy Tarreau61dd44b2020-06-25 07:35:42 +02005830 /* empty all trailing args by making them point to the trailing zero,
5831 * at least the last one in any case.
5832 */
5833 if (arg > argsmax)
5834 arg = argsmax;
5835
5836 while (arg >= 0 && arg <= argsmax)
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005837 args[arg++] = out + outpos - 1;
5838
5839 return err;
5840}
5841#undef EMIT_CHAR
5842
Willy Tarreauc54e5ad2020-06-25 09:15:40 +02005843/* This is used to sanitize an input line that's about to be used for error reporting.
5844 * It will adjust <line> to print approximately <width> chars around <pos>, trying to
5845 * preserve the beginning, with leading or trailing "..." when the line is truncated.
5846 * If non-printable chars are present in the output. It returns the new offset <pos>
5847 * in the modified line. Non-printable characters are replaced with '?'. <width> must
5848 * be at least 6 to support two "..." otherwise the result is undefined. The line
5849 * itself must have at least 7 chars allocated for the same reason.
5850 */
5851size_t sanitize_for_printing(char *line, size_t pos, size_t width)
5852{
5853 size_t shift = 0;
5854 char *out = line;
5855 char *in = line;
5856 char *end = line + width;
5857
5858 if (pos >= width) {
5859 /* if we have to shift, we'll be out of context, so let's
5860 * try to put <pos> at the center of width.
5861 */
5862 shift = pos - width / 2;
5863 in += shift + 3;
5864 end = out + width - 3;
5865 out[0] = out[1] = out[2] = '.';
5866 out += 3;
5867 }
5868
5869 while (out < end && *in) {
5870 if (isspace((unsigned char)*in))
5871 *out++ = ' ';
5872 else if (isprint((unsigned char)*in))
5873 *out++ = *in;
5874 else
5875 *out++ = '?';
5876 in++;
5877 }
5878
5879 if (end < line + width) {
5880 out[0] = out[1] = out[2] = '.';
5881 out += 3;
5882 }
5883
5884 *out++ = 0;
5885 return pos - shift;
5886}
Willy Tarreau06e69b52021-03-02 14:01:35 +01005887
Willy Tarreaue33c4b32021-03-12 18:59:31 +01005888/* Update array <fp> with the fingerprint of word <word> by counting the
Willy Tarreauba2c4452021-03-12 09:01:52 +01005889 * transitions between characters. <fp> is a 1024-entries array indexed as
5890 * 32*from+to. Positions for 'from' and 'to' are:
Willy Tarreau9294e882021-03-15 09:34:27 +01005891 * 1..26=letter, 27=digit, 28=other/begin/end.
5892 * Row "from=0" is used to mark the character's presence. Others unused.
Willy Tarreauba2c4452021-03-12 09:01:52 +01005893 */
Willy Tarreaue33c4b32021-03-12 18:59:31 +01005894void update_word_fingerprint(uint8_t *fp, const char *word)
Willy Tarreauba2c4452021-03-12 09:01:52 +01005895{
5896 const char *p;
5897 int from, to;
5898 int c;
5899
Willy Tarreauba2c4452021-03-12 09:01:52 +01005900 from = 28; // begin
5901 for (p = word; *p; p++) {
5902 c = tolower(*p);
5903 switch(c) {
Willy Tarreau9294e882021-03-15 09:34:27 +01005904 case 'a'...'z': to = c - 'a' + 1; break;
5905 case 'A'...'Z': to = tolower(c) - 'a' + 1; break;
5906 case '0'...'9': to = 27; break;
5907 default: to = 28; break;
Willy Tarreauba2c4452021-03-12 09:01:52 +01005908 }
Willy Tarreau9294e882021-03-15 09:34:27 +01005909 fp[to] = 1;
Willy Tarreauba2c4452021-03-12 09:01:52 +01005910 fp[32 * from + to]++;
5911 from = to;
5912 }
5913 to = 28; // end
5914 fp[32 * from + to]++;
5915}
5916
Erwan Le Goas9c766372022-09-14 17:40:24 +02005917/* This function hashes a word, scramble is the anonymizing key, returns
5918 * the hashed word when the key (scramble) != 0, else returns the word.
5919 * This function can be called NB_L_HASH_WORD times in a row, don't call
5920 * it if you called it more than NB_L_HASH_WORD.
5921 */
5922const char *hash_anon(uint32_t scramble, const char *string2hash, const char *prefix, const char *suffix)
5923{
5924 index_hash++;
Erwan Le Goasd2605cf2022-09-21 16:24:23 +02005925 if (index_hash == NB_L_HASH_WORD)
Erwan Le Goas9c766372022-09-14 17:40:24 +02005926 index_hash = 0;
5927
5928 /* don't hash empty strings */
5929 if (!string2hash[0] || (string2hash[0] == ' ' && string2hash[1] == 0))
5930 return string2hash;
5931
5932 if (scramble != 0) {
5933 snprintf(hash_word[index_hash], sizeof(hash_word[index_hash]), "%s%06x%s",
5934 prefix, HA_ANON(scramble, string2hash, strlen(string2hash)), suffix);
5935 return hash_word[index_hash];
5936 }
5937 else
5938 return string2hash;
5939}
5940
5941/* This function hashes or not an ip address ipstring, scramble is the anonymizing
5942 * key, returns the hashed ip with his port or ipstring when there is nothing to hash.
Erwan Le Goas5eef1582022-09-29 10:25:31 +02005943 * Put hasport equal 0 to point out ipstring has no port, else put an other int.
5944 * Without port, return a simple hash or ipstring.
Erwan Le Goas9c766372022-09-14 17:40:24 +02005945 */
Erwan Le Goas5eef1582022-09-29 10:25:31 +02005946const char *hash_ipanon(uint32_t scramble, char *ipstring, int hasport)
Erwan Le Goas9c766372022-09-14 17:40:24 +02005947{
5948 char *errmsg = NULL;
5949 struct sockaddr_storage *sa;
Erwan Le Goas5eef1582022-09-29 10:25:31 +02005950 struct sockaddr_storage ss;
Erwan Le Goas9c766372022-09-14 17:40:24 +02005951 char addr[46];
5952 int port;
5953
5954 index_hash++;
Erwan Le Goasd2605cf2022-09-21 16:24:23 +02005955 if (index_hash == NB_L_HASH_WORD) {
Erwan Le Goas9c766372022-09-14 17:40:24 +02005956 index_hash = 0;
5957 }
5958
Erwan Le Goas5eef1582022-09-29 10:25:31 +02005959 if (scramble == 0) {
Erwan Le Goas9c766372022-09-14 17:40:24 +02005960 return ipstring;
5961 }
Christopher Faulet015bbc22022-09-29 11:53:07 +02005962 if (strcmp(ipstring, "localhost") == 0 ||
5963 strcmp(ipstring, "stdout") == 0 ||
5964 strcmp(ipstring, "stderr") == 0 ||
5965 strncmp(ipstring, "fd@", 3) == 0 ||
5966 strncmp(ipstring, "sockpair@", 9) == 0) {
Erwan Le Goas5eef1582022-09-29 10:25:31 +02005967 return ipstring;
5968 }
Erwan Le Goas9c766372022-09-14 17:40:24 +02005969 else {
Erwan Le Goas5eef1582022-09-29 10:25:31 +02005970 if (hasport == 0) {
5971 memset(&ss, 0, sizeof(ss));
5972 if (str2ip2(ipstring, &ss, 1) == NULL) {
5973 return HA_ANON_STR(scramble, ipstring);
5974 }
5975 sa = &ss;
Erwan Le Goas9c766372022-09-14 17:40:24 +02005976 }
5977 else {
Erwan Le Goas5eef1582022-09-29 10:25:31 +02005978 sa = str2sa_range(ipstring, NULL, NULL, NULL, NULL, NULL, &errmsg, NULL, NULL,
Christopher Faulet7e50e4b2022-09-29 11:46:34 +02005979 PA_O_PORT_OK | PA_O_STREAM | PA_O_DGRAM | PA_O_XPRT | PA_O_CONNECT |
5980 PA_O_PORT_RANGE | PA_O_PORT_OFS | PA_O_RESOLVE);
Erwan Le Goas5eef1582022-09-29 10:25:31 +02005981 if (sa == NULL) {
5982 return HA_ANON_STR(scramble, ipstring);
5983 }
5984 }
5985 addr_to_str(sa, addr, sizeof(addr));
5986 port = get_host_port(sa);
Erwan Le Goas9c766372022-09-14 17:40:24 +02005987
Erwan Le Goas5eef1582022-09-29 10:25:31 +02005988 switch(sa->ss_family) {
5989 case AF_INET:
5990 if (strncmp(addr, "127", 3) == 0 || strncmp(addr, "255", 3) == 0 || strncmp(addr, "0", 1) == 0) {
5991 return ipstring;
5992 }
5993 else {
5994 if (port != 0) {
5995 snprintf(hash_word[index_hash], sizeof(hash_word[index_hash]), "IPV4(%06x):%d", HA_ANON(scramble, addr, strlen(addr)), port);
5996 return hash_word[index_hash];
Erwan Le Goas9c766372022-09-14 17:40:24 +02005997 }
5998 else {
Erwan Le Goas5eef1582022-09-29 10:25:31 +02005999 snprintf(hash_word[index_hash], sizeof(hash_word[index_hash]), "IPV4(%06x)", HA_ANON(scramble, addr, strlen(addr)));
6000 return hash_word[index_hash];
Erwan Le Goas9c766372022-09-14 17:40:24 +02006001 }
Erwan Le Goas5eef1582022-09-29 10:25:31 +02006002 }
6003 break;
Erwan Le Goas9c766372022-09-14 17:40:24 +02006004
Erwan Le Goas5eef1582022-09-29 10:25:31 +02006005 case AF_INET6:
6006 if (strcmp(addr, "::1") == 0) {
6007 return ipstring;
6008 }
6009 else {
6010 if (port != 0) {
6011 snprintf(hash_word[index_hash], sizeof(hash_word[index_hash]), "IPV6(%06x):%d", HA_ANON(scramble, addr, strlen(addr)), port);
6012 return hash_word[index_hash];
Erwan Le Goas9c766372022-09-14 17:40:24 +02006013 }
6014 else {
Erwan Le Goas5eef1582022-09-29 10:25:31 +02006015 snprintf(hash_word[index_hash], sizeof(hash_word[index_hash]), "IPV6(%06x)", HA_ANON(scramble, addr, strlen(addr)));
6016 return hash_word[index_hash];
Erwan Le Goas9c766372022-09-14 17:40:24 +02006017 }
Erwan Le Goas5eef1582022-09-29 10:25:31 +02006018 }
6019 break;
Erwan Le Goas9c766372022-09-14 17:40:24 +02006020
Erwan Le Goas5eef1582022-09-29 10:25:31 +02006021 case AF_UNIX:
6022 return HA_ANON_STR(scramble, ipstring);
6023 break;
6024
6025 default:
6026 return ipstring;
6027 break;
6028 };
Erwan Le Goas9c766372022-09-14 17:40:24 +02006029 }
6030 return ipstring;
6031}
6032
Willy Tarreaue33c4b32021-03-12 18:59:31 +01006033/* Initialize array <fp> with the fingerprint of word <word> by counting the
6034 * transitions between characters. <fp> is a 1024-entries array indexed as
6035 * 32*from+to. Positions for 'from' and 'to' are:
6036 * 0..25=letter, 26=digit, 27=other, 28=begin, 29=end, others unused.
6037 */
6038void make_word_fingerprint(uint8_t *fp, const char *word)
6039{
6040 memset(fp, 0, 1024);
6041 update_word_fingerprint(fp, word);
6042}
6043
Willy Tarreauba2c4452021-03-12 09:01:52 +01006044/* Return the distance between two word fingerprints created by function
6045 * make_word_fingerprint(). It's a positive integer calculated as the sum of
Willy Tarreau714c4c12021-03-15 09:44:53 +01006046 * the differences between each location.
Willy Tarreauba2c4452021-03-12 09:01:52 +01006047 */
6048int word_fingerprint_distance(const uint8_t *fp1, const uint8_t *fp2)
6049{
6050 int i, k, dist = 0;
6051
6052 for (i = 0; i < 1024; i++) {
6053 k = (int)fp1[i] - (int)fp2[i];
Willy Tarreau714c4c12021-03-15 09:44:53 +01006054 dist += abs(k);
Willy Tarreauba2c4452021-03-12 09:01:52 +01006055 }
6056 return dist;
6057}
6058
William Lallemand3aeb3f92021-08-21 23:59:56 +02006059/*
6060 * This function compares the loaded openssl version with a string <version>
6061 * This function use the same return code as compare_current_version:
6062 *
6063 * -1 : the version in argument is older than the current openssl version
6064 * 0 : the version in argument is the same as the current openssl version
6065 * 1 : the version in argument is newer than the current openssl version
6066 *
6067 * Or some errors:
6068 * -2 : openssl is not available on this process
6069 * -3 : the version in argument is not parsable
6070 */
6071int openssl_compare_current_version(const char *version)
6072{
6073#ifdef USE_OPENSSL
6074 int numversion;
6075
6076 numversion = openssl_version_parser(version);
6077 if (numversion == 0)
6078 return -3;
6079
6080 if (numversion < OPENSSL_VERSION_NUMBER)
6081 return -1;
6082 else if (numversion > OPENSSL_VERSION_NUMBER)
6083 return 1;
6084 else
6085 return 0;
6086#else
6087 return -2;
6088#endif
6089}
6090
Remi Tricot-Le Bretonb01179a2021-10-11 15:34:12 +02006091/*
6092 * This function compares the loaded openssl name with a string <name>
6093 * This function returns 0 if the OpenSSL name starts like the passed parameter,
6094 * 1 otherwise.
6095 */
6096int openssl_compare_current_name(const char *name)
6097{
6098#ifdef USE_OPENSSL
6099 int name_len = 0;
6100 const char *openssl_version = OpenSSL_version(OPENSSL_VERSION);
6101
6102 if (name) {
6103 name_len = strlen(name);
6104 if (strlen(name) <= strlen(openssl_version))
6105 return strncmp(openssl_version, name, name_len);
6106 }
6107#endif
6108 return 1;
6109}
6110
Willy Tarreau40dde2d2022-06-19 16:41:59 +02006111#if defined(RTLD_DEFAULT) || defined(RTLD_NEXT)
6112/* redefine dlopen() so that we can detect unexpected replacement of some
6113 * critical symbols, typically init/alloc/free functions coming from alternate
6114 * libraries. When called, a tainted flag is set (TAINTED_SHARED_LIBS).
Willy Tarreauc3b297d2023-03-22 17:01:32 +01006115 * It's important to understand that the dynamic linker will present the
6116 * first loaded of each symbol to all libs, so that if haproxy is linked
6117 * with a new lib that uses a static inline or a #define to replace an old
6118 * function, and a dependency was linked against an older version of that
6119 * lib that had a function there, that lib would use all of the newer
6120 * versions of the functions that are already loaded in haproxy, except
6121 * for that unique function which would continue to be the old one. This
6122 * creates all sort of problems when init code allocates smaller structs
6123 * than required for example but uses new functions on them, etc. Thus what
6124 * we do here is to try to detect API consistency: we take a fingerprint of
6125 * a number of known functions, and verify that if they change in a loaded
6126 * library, either there all appeared or all disappeared, but not partially.
6127 * We can check up to 64 symbols that belong to individual groups that are
6128 * checked together.
Willy Tarreau40dde2d2022-06-19 16:41:59 +02006129 */
6130void *dlopen(const char *filename, int flags)
6131{
6132 static void *(*_dlopen)(const char *filename, int flags);
Willy Tarreau177aed52022-06-19 16:49:51 +02006133 struct {
6134 const char *name;
Willy Tarreauc3b297d2023-03-22 17:01:32 +01006135 uint64_t bit, grp;
Willy Tarreau177aed52022-06-19 16:49:51 +02006136 void *curr, *next;
6137 } check_syms[] = {
Willy Tarreau997ad152023-04-22 19:47:19 +02006138 /* openssl's libcrypto checks: group bits 0x1f */
6139 { .name="OPENSSL_init", .bit = 0x0000000000000001, .grp = 0x000000000000001f, }, // openssl 1.0 / 1.1 / 3.0
6140 { .name="OPENSSL_init_crypto", .bit = 0x0000000000000002, .grp = 0x000000000000001f, }, // openssl 1.1 / 3.0
6141 { .name="ENGINE_init", .bit = 0x0000000000000004, .grp = 0x000000000000001f, }, // openssl 1.x / 3.x with engine
6142 { .name="EVP_CIPHER_CTX_init", .bit = 0x0000000000000008, .grp = 0x000000000000001f, }, // openssl 1.0
6143 { .name="HMAC_Init", .bit = 0x0000000000000010, .grp = 0x000000000000001f, }, // openssl 1.x
6144
6145 /* openssl's libssl checks: group bits 0x3e0 */
6146 { .name="OPENSSL_init_ssl", .bit = 0x0000000000000020, .grp = 0x00000000000003e0, }, // openssl 1.1 / 3.0
6147 { .name="SSL_library_init", .bit = 0x0000000000000040, .grp = 0x00000000000003e0, }, // openssl 1.x
6148 { .name="SSL_is_quic", .bit = 0x0000000000000080, .grp = 0x00000000000003e0, }, // quictls
6149 { .name="SSL_CTX_new_ex", .bit = 0x0000000000000100, .grp = 0x00000000000003e0, }, // openssl 3.x
6150 { .name="SSL_CTX_get0_security_ex_data", .bit = 0x0000000000000200, .grp = 0x00000000000003e0, }, // openssl 1.x / 3.x
6151
Willy Tarreau177aed52022-06-19 16:49:51 +02006152 /* insert only above, 0 must be the last one */
6153 { 0 },
6154 };
6155 const char *trace;
Willy Tarreauc3b297d2023-03-22 17:01:32 +01006156 uint64_t own_fp, lib_fp; // symbols fingerprints
Willy Tarreau177aed52022-06-19 16:49:51 +02006157 void *addr;
Willy Tarreau40dde2d2022-06-19 16:41:59 +02006158 void *ret;
Willy Tarreau177aed52022-06-19 16:49:51 +02006159 int sym = 0;
Willy Tarreau40dde2d2022-06-19 16:41:59 +02006160
6161 if (!_dlopen) {
6162 _dlopen = get_sym_next_addr("dlopen");
6163 if (!_dlopen || _dlopen == dlopen) {
6164 _dlopen = NULL;
6165 return NULL;
6166 }
6167 }
6168
Willy Tarreau177aed52022-06-19 16:49:51 +02006169 /* save a few pointers to critical symbols. We keep a copy of both the
6170 * current and the next value, because we might already have replaced
Willy Tarreau58912b82023-03-22 15:47:51 +01006171 * some of them in an inconsistent way (i.e. not all), and we're only
Willy Tarreau177aed52022-06-19 16:49:51 +02006172 * interested in verifying that a loaded library doesn't come with a
Willy Tarreauc3b297d2023-03-22 17:01:32 +01006173 * completely different definition that would be incompatible. We'll
6174 * keep a fingerprint of our own symbols.
Willy Tarreau177aed52022-06-19 16:49:51 +02006175 */
Willy Tarreauc3b297d2023-03-22 17:01:32 +01006176 own_fp = 0;
Willy Tarreau177aed52022-06-19 16:49:51 +02006177 for (sym = 0; check_syms[sym].name; sym++) {
6178 check_syms[sym].curr = get_sym_curr_addr(check_syms[sym].name);
6179 check_syms[sym].next = get_sym_next_addr(check_syms[sym].name);
Willy Tarreauc3b297d2023-03-22 17:01:32 +01006180 if (check_syms[sym].curr || check_syms[sym].next)
6181 own_fp |= check_syms[sym].bit;
Willy Tarreau177aed52022-06-19 16:49:51 +02006182 }
Willy Tarreau40dde2d2022-06-19 16:41:59 +02006183
6184 /* now open the requested lib */
6185 ret = _dlopen(filename, flags);
6186 if (!ret)
6187 return ret;
6188
6189 mark_tainted(TAINTED_SHARED_LIBS);
6190
Willy Tarreau177aed52022-06-19 16:49:51 +02006191 /* and check that critical symbols didn't change */
Willy Tarreauc3b297d2023-03-22 17:01:32 +01006192 lib_fp = 0;
Willy Tarreau177aed52022-06-19 16:49:51 +02006193 for (sym = 0; check_syms[sym].name; sym++) {
Willy Tarreau177aed52022-06-19 16:49:51 +02006194 addr = dlsym(ret, check_syms[sym].name);
Willy Tarreauc3b297d2023-03-22 17:01:32 +01006195 if (addr)
6196 lib_fp |= check_syms[sym].bit;
6197 }
Willy Tarreau177aed52022-06-19 16:49:51 +02006198
Willy Tarreauc3b297d2023-03-22 17:01:32 +01006199 if (lib_fp != own_fp) {
6200 /* let's check what changed: */
6201 uint64_t mask = 0;
Willy Tarreau177aed52022-06-19 16:49:51 +02006202
Willy Tarreauc3b297d2023-03-22 17:01:32 +01006203 for (sym = 0; check_syms[sym].name; sym++) {
6204 mask = check_syms[sym].grp;
6205
6206 /* new group of symbols. If they all appeared together
6207 * their use will be consistent. If none appears, it's
6208 * just that the lib doesn't use them. If some appear
6209 * or disappear, it means the lib relies on a different
6210 * dependency and will end up with a mix.
6211 */
6212 if (!(own_fp & mask) || !(lib_fp & mask) ||
6213 (own_fp & mask) == (lib_fp & mask))
6214 continue;
6215
6216 /* let's report a symbol that really changes */
6217 if (!((own_fp ^ lib_fp) & check_syms[sym].bit))
6218 continue;
6219
6220 /* OK it's clear that this symbol was redefined */
6221 mark_tainted(TAINTED_REDEFINITION);
6222
6223 trace = hlua_show_current_location("\n ");
6224 ha_warning("dlopen(): shared library '%s' brings a different and inconsistent definition of symbol '%s'. The process cannot be trusted anymore!%s%s\n",
6225 filename, check_syms[sym].name,
6226 trace ? " Suspected call location: \n " : "",
6227 trace ? trace : "");
6228 }
Willy Tarreau177aed52022-06-19 16:49:51 +02006229 }
6230
Willy Tarreau40dde2d2022-06-19 16:41:59 +02006231 return ret;
6232}
6233#endif
6234
Willy Tarreau06e69b52021-03-02 14:01:35 +01006235static int init_tools_per_thread()
6236{
6237 /* Let's make each thread start from a different position */
Willy Tarreaue936aaf2024-03-01 16:17:47 +01006238 statistical_prng_state += ha_random32();
Willy Tarreau06e69b52021-03-02 14:01:35 +01006239 if (!statistical_prng_state)
6240 statistical_prng_state++;
6241 return 1;
6242}
6243REGISTER_PER_THREAD_INIT(init_tools_per_thread);
Willy Tarreauc54e5ad2020-06-25 09:15:40 +02006244
Willy Tarreaubaaee002006-06-26 02:48:02 +02006245/*
6246 * Local variables:
6247 * c-indent-level: 8
6248 * c-basic-offset: 8
6249 * End:
6250 */