blob: 7eaa7433e931daf5578e6d776cd6f5b0eb2f848b [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
David Carlier1b9d57d2021-08-17 08:44:25 +010019#if defined(__NetBSD__)
20#include <sys/exec_elf.h>
21#include <dlfcn.h>
22#endif
23
Willy Tarreau2e74c3f2007-12-02 18:45:09 +010024#include <ctype.h>
Willy Tarreau16e01562016-08-09 16:46:18 +020025#include <errno.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020026#include <netdb.h>
Willy Tarreau9a7bea52012-04-27 11:16:50 +020027#include <stdarg.h>
Willy Tarreaudd2f85e2012-09-02 22:34:23 +020028#include <stdio.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020029#include <stdlib.h>
30#include <string.h>
Thierry Fournier93127942016-01-20 18:49:45 +010031#include <time.h>
Willy Tarreau16e01562016-08-09 16:46:18 +020032#include <unistd.h>
Willy Tarreau127f9662007-12-06 00:53:51 +010033#include <sys/socket.h>
Willy Tarreau37101052019-05-20 16:48:20 +020034#include <sys/stat.h>
35#include <sys/types.h>
Willy Tarreau127f9662007-12-06 00:53:51 +010036#include <sys/un.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020037#include <netinet/in.h>
38#include <arpa/inet.h>
39
Willy Tarreau30053062020-08-20 16:39:14 +020040#if (__GLIBC__ > 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 16))
41#include <sys/auxv.h>
42#endif
43
Willy Tarreau48fbcae2020-06-03 18:09:46 +020044#include <import/eb32sctree.h>
Willy Tarreaub2551052020-06-09 09:07:15 +020045#include <import/eb32tree.h>
Willy Tarreau48fbcae2020-06-03 18:09:46 +020046
Willy Tarreau4c7e4b72020-05-27 12:58:42 +020047#include <haproxy/api.h>
Willy Tarreauc13ed532020-06-02 10:22:45 +020048#include <haproxy/chunk.h>
Willy Tarreau7c18b542020-06-11 09:23:02 +020049#include <haproxy/dgram.h>
Willy Tarreauf268ee82020-06-04 17:05:57 +020050#include <haproxy/global.h>
Willy Tarreau86416052020-06-04 09:20:54 +020051#include <haproxy/hlua.h>
Willy Tarreau213e9902020-06-04 14:58:24 +020052#include <haproxy/listener.h>
Willy Tarreau7a00efb2020-06-02 17:02:59 +020053#include <haproxy/namespace.h>
Christopher Faulet9553de72021-02-26 09:12:50 +010054#include <haproxy/net_helper.h>
Willy Tarreau5fc93282020-09-16 18:25:03 +020055#include <haproxy/protocol.h>
Emeric Brunc9437992021-02-12 19:42:55 +010056#include <haproxy/resolvers.h>
Willy Tarreau586f71b2020-12-11 15:54:36 +010057#include <haproxy/sock.h>
Willy Tarreau209108d2020-06-04 20:30:20 +020058#include <haproxy/ssl_sock.h>
Willy Tarreau5e539c92020-06-04 20:45:39 +020059#include <haproxy/stream_interface.h>
Willy Tarreaucea0e1b2020-06-04 17:25:40 +020060#include <haproxy/task.h>
Willy Tarreau48fbcae2020-06-03 18:09:46 +020061#include <haproxy/tools.h>
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +010062
Thierry Fournier93127942016-01-20 18:49:45 +010063/* This macro returns false if the test __x is false. Many
64 * of the following parsing function must be abort the processing
65 * if it returns 0, so this macro is useful for writing light code.
66 */
67#define RET0_UNLESS(__x) do { if (!(__x)) return 0; } while (0)
68
Willy Tarreau56adcf22012-12-23 18:00:29 +010069/* enough to store NB_ITOA_STR integers of :
Willy Tarreau72d759c2007-10-25 12:14:10 +020070 * 2^64-1 = 18446744073709551615 or
71 * -2^63 = -9223372036854775808
Willy Tarreaue7239b52009-03-29 13:41:58 +020072 *
73 * The HTML version needs room for adding the 25 characters
74 * '<span class="rls"></span>' around digits at positions 3N+1 in order
75 * to add spacing at up to 6 positions : 18 446 744 073 709 551 615
Willy Tarreau72d759c2007-10-25 12:14:10 +020076 */
Christopher Faulet99bca652017-11-14 16:47:26 +010077THREAD_LOCAL char itoa_str[NB_ITOA_STR][171];
78THREAD_LOCAL int itoa_idx = 0; /* index of next itoa_str to use */
Willy Tarreaubaaee002006-06-26 02:48:02 +020079
Willy Tarreau588297f2014-06-16 15:16:40 +020080/* sometimes we'll need to quote strings (eg: in stats), and we don't expect
81 * to quote strings larger than a max configuration line.
82 */
Christopher Faulet99bca652017-11-14 16:47:26 +010083THREAD_LOCAL char quoted_str[NB_QSTR][QSTR_SIZE + 1];
84THREAD_LOCAL int quoted_idx = 0;
Willy Tarreau588297f2014-06-16 15:16:40 +020085
Willy Tarreau06e69b52021-03-02 14:01:35 +010086/* thread-local PRNG state. It's modified to start from a different sequence
87 * on all threads upon startup. It must not be used or anything beyond getting
88 * statistical values as it's 100% predictable.
89 */
90THREAD_LOCAL unsigned int statistical_prng_state = 2463534242U;
91
Willy Tarreaubaaee002006-06-26 02:48:02 +020092/*
William Lallemande7340ec2012-01-24 11:15:39 +010093 * unsigned long long ASCII representation
94 *
95 * return the last char '\0' or NULL if no enough
96 * space in dst
97 */
98char *ulltoa(unsigned long long n, char *dst, size_t size)
99{
100 int i = 0;
101 char *res;
102
103 switch(n) {
104 case 1ULL ... 9ULL:
105 i = 0;
106 break;
107
108 case 10ULL ... 99ULL:
109 i = 1;
110 break;
111
112 case 100ULL ... 999ULL:
113 i = 2;
114 break;
115
116 case 1000ULL ... 9999ULL:
117 i = 3;
118 break;
119
120 case 10000ULL ... 99999ULL:
121 i = 4;
122 break;
123
124 case 100000ULL ... 999999ULL:
125 i = 5;
126 break;
127
128 case 1000000ULL ... 9999999ULL:
129 i = 6;
130 break;
131
132 case 10000000ULL ... 99999999ULL:
133 i = 7;
134 break;
135
136 case 100000000ULL ... 999999999ULL:
137 i = 8;
138 break;
139
140 case 1000000000ULL ... 9999999999ULL:
141 i = 9;
142 break;
143
144 case 10000000000ULL ... 99999999999ULL:
145 i = 10;
146 break;
147
148 case 100000000000ULL ... 999999999999ULL:
149 i = 11;
150 break;
151
152 case 1000000000000ULL ... 9999999999999ULL:
153 i = 12;
154 break;
155
156 case 10000000000000ULL ... 99999999999999ULL:
157 i = 13;
158 break;
159
160 case 100000000000000ULL ... 999999999999999ULL:
161 i = 14;
162 break;
163
164 case 1000000000000000ULL ... 9999999999999999ULL:
165 i = 15;
166 break;
167
168 case 10000000000000000ULL ... 99999999999999999ULL:
169 i = 16;
170 break;
171
172 case 100000000000000000ULL ... 999999999999999999ULL:
173 i = 17;
174 break;
175
176 case 1000000000000000000ULL ... 9999999999999999999ULL:
177 i = 18;
178 break;
179
180 case 10000000000000000000ULL ... ULLONG_MAX:
181 i = 19;
182 break;
183 }
184 if (i + 2 > size) // (i + 1) + '\0'
185 return NULL; // too long
186 res = dst + i + 1;
187 *res = '\0';
188 for (; i >= 0; i--) {
189 dst[i] = n % 10ULL + '0';
190 n /= 10ULL;
191 }
192 return res;
193}
194
195/*
196 * unsigned long ASCII representation
197 *
198 * return the last char '\0' or NULL if no enough
199 * space in dst
200 */
201char *ultoa_o(unsigned long n, char *dst, size_t size)
202{
203 int i = 0;
204 char *res;
205
206 switch (n) {
207 case 0U ... 9UL:
208 i = 0;
209 break;
210
211 case 10U ... 99UL:
212 i = 1;
213 break;
214
215 case 100U ... 999UL:
216 i = 2;
217 break;
218
219 case 1000U ... 9999UL:
220 i = 3;
221 break;
222
223 case 10000U ... 99999UL:
224 i = 4;
225 break;
226
227 case 100000U ... 999999UL:
228 i = 5;
229 break;
230
231 case 1000000U ... 9999999UL:
232 i = 6;
233 break;
234
235 case 10000000U ... 99999999UL:
236 i = 7;
237 break;
238
239 case 100000000U ... 999999999UL:
240 i = 8;
241 break;
242#if __WORDSIZE == 32
243
244 case 1000000000ULL ... ULONG_MAX:
245 i = 9;
246 break;
247
248#elif __WORDSIZE == 64
249
250 case 1000000000ULL ... 9999999999UL:
251 i = 9;
252 break;
253
254 case 10000000000ULL ... 99999999999UL:
255 i = 10;
256 break;
257
258 case 100000000000ULL ... 999999999999UL:
259 i = 11;
260 break;
261
262 case 1000000000000ULL ... 9999999999999UL:
263 i = 12;
264 break;
265
266 case 10000000000000ULL ... 99999999999999UL:
267 i = 13;
268 break;
269
270 case 100000000000000ULL ... 999999999999999UL:
271 i = 14;
272 break;
273
274 case 1000000000000000ULL ... 9999999999999999UL:
275 i = 15;
276 break;
277
278 case 10000000000000000ULL ... 99999999999999999UL:
279 i = 16;
280 break;
281
282 case 100000000000000000ULL ... 999999999999999999UL:
283 i = 17;
284 break;
285
286 case 1000000000000000000ULL ... 9999999999999999999UL:
287 i = 18;
288 break;
289
290 case 10000000000000000000ULL ... ULONG_MAX:
291 i = 19;
292 break;
293
294#endif
295 }
296 if (i + 2 > size) // (i + 1) + '\0'
297 return NULL; // too long
298 res = dst + i + 1;
299 *res = '\0';
300 for (; i >= 0; i--) {
301 dst[i] = n % 10U + '0';
302 n /= 10U;
303 }
304 return res;
305}
306
307/*
308 * signed long ASCII representation
309 *
310 * return the last char '\0' or NULL if no enough
311 * space in dst
312 */
313char *ltoa_o(long int n, char *dst, size_t size)
314{
315 char *pos = dst;
316
317 if (n < 0) {
318 if (size < 3)
319 return NULL; // min size is '-' + digit + '\0' but another test in ultoa
320 *pos = '-';
321 pos++;
322 dst = ultoa_o(-n, pos, size - 1);
323 } else {
324 dst = ultoa_o(n, dst, size);
325 }
326 return dst;
327}
328
329/*
330 * signed long long ASCII representation
331 *
332 * return the last char '\0' or NULL if no enough
333 * space in dst
334 */
335char *lltoa(long long n, char *dst, size_t size)
336{
337 char *pos = dst;
338
339 if (n < 0) {
340 if (size < 3)
341 return NULL; // min size is '-' + digit + '\0' but another test in ulltoa
342 *pos = '-';
343 pos++;
344 dst = ulltoa(-n, pos, size - 1);
345 } else {
346 dst = ulltoa(n, dst, size);
347 }
348 return dst;
349}
350
351/*
352 * write a ascii representation of a unsigned into dst,
353 * return a pointer to the last character
354 * Pad the ascii representation with '0', using size.
355 */
356char *utoa_pad(unsigned int n, char *dst, size_t size)
357{
358 int i = 0;
359 char *ret;
360
361 switch(n) {
362 case 0U ... 9U:
363 i = 0;
364 break;
365
366 case 10U ... 99U:
367 i = 1;
368 break;
369
370 case 100U ... 999U:
371 i = 2;
372 break;
373
374 case 1000U ... 9999U:
375 i = 3;
376 break;
377
378 case 10000U ... 99999U:
379 i = 4;
380 break;
381
382 case 100000U ... 999999U:
383 i = 5;
384 break;
385
386 case 1000000U ... 9999999U:
387 i = 6;
388 break;
389
390 case 10000000U ... 99999999U:
391 i = 7;
392 break;
393
394 case 100000000U ... 999999999U:
395 i = 8;
396 break;
397
398 case 1000000000U ... 4294967295U:
399 i = 9;
400 break;
401 }
402 if (i + 2 > size) // (i + 1) + '\0'
403 return NULL; // too long
404 if (i < size)
405 i = size - 2; // padding - '\0'
406
407 ret = dst + i + 1;
408 *ret = '\0';
409 for (; i >= 0; i--) {
410 dst[i] = n % 10U + '0';
411 n /= 10U;
412 }
413 return ret;
414}
415
416/*
Willy Tarreaubaaee002006-06-26 02:48:02 +0200417 * copies at most <size-1> chars from <src> to <dst>. Last char is always
418 * set to 0, unless <size> is 0. The number of chars copied is returned
419 * (excluding the terminating zero).
420 * This code has been optimized for size and speed : on x86, it's 45 bytes
421 * long, uses only registers, and consumes only 4 cycles per char.
422 */
423int strlcpy2(char *dst, const char *src, int size)
424{
425 char *orig = dst;
426 if (size) {
427 while (--size && (*dst = *src)) {
428 src++; dst++;
429 }
430 *dst = 0;
431 }
432 return dst - orig;
433}
434
435/*
Willy Tarreau72d759c2007-10-25 12:14:10 +0200436 * This function simply returns a locally allocated string containing
Willy Tarreaubaaee002006-06-26 02:48:02 +0200437 * the ascii representation for number 'n' in decimal.
438 */
Emeric Brun3a7fce52010-01-04 14:54:38 +0100439char *ultoa_r(unsigned long n, char *buffer, int size)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200440{
441 char *pos;
442
Willy Tarreau72d759c2007-10-25 12:14:10 +0200443 pos = buffer + size - 1;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200444 *pos-- = '\0';
445
446 do {
447 *pos-- = '0' + n % 10;
448 n /= 10;
Willy Tarreau72d759c2007-10-25 12:14:10 +0200449 } while (n && pos >= buffer);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200450 return pos + 1;
451}
452
Willy Tarreau91092e52007-10-25 16:58:42 +0200453/*
Willy Tarreaue7239b52009-03-29 13:41:58 +0200454 * This function simply returns a locally allocated string containing
Thierry FOURNIER763a5d82015-07-06 23:09:52 +0200455 * the ascii representation for number 'n' in decimal.
456 */
457char *lltoa_r(long long int in, char *buffer, int size)
458{
459 char *pos;
460 int neg = 0;
461 unsigned long long int n;
462
463 pos = buffer + size - 1;
464 *pos-- = '\0';
465
466 if (in < 0) {
467 neg = 1;
468 n = -in;
469 }
470 else
471 n = in;
472
473 do {
474 *pos-- = '0' + n % 10;
475 n /= 10;
476 } while (n && pos >= buffer);
477 if (neg && pos > buffer)
478 *pos-- = '-';
479 return pos + 1;
480}
481
482/*
483 * This function simply returns a locally allocated string containing
Thierry FOURNIER1480bd82015-06-06 19:14:59 +0200484 * the ascii representation for signed number 'n' in decimal.
485 */
486char *sltoa_r(long n, char *buffer, int size)
487{
488 char *pos;
489
490 if (n >= 0)
491 return ultoa_r(n, buffer, size);
492
493 pos = ultoa_r(-n, buffer + 1, size - 1) - 1;
494 *pos = '-';
495 return pos;
496}
497
498/*
499 * This function simply returns a locally allocated string containing
Willy Tarreaue7239b52009-03-29 13:41:58 +0200500 * the ascii representation for number 'n' in decimal, formatted for
501 * HTML output with tags to create visual grouping by 3 digits. The
502 * output needs to support at least 171 characters.
503 */
504const char *ulltoh_r(unsigned long long n, char *buffer, int size)
505{
506 char *start;
507 int digit = 0;
508
509 start = buffer + size;
510 *--start = '\0';
511
512 do {
513 if (digit == 3 && start >= buffer + 7)
514 memcpy(start -= 7, "</span>", 7);
515
516 if (start >= buffer + 1) {
517 *--start = '0' + n % 10;
518 n /= 10;
519 }
520
521 if (digit == 3 && start >= buffer + 18)
522 memcpy(start -= 18, "<span class=\"rls\">", 18);
523
524 if (digit++ == 3)
525 digit = 1;
526 } while (n && start > buffer);
527 return start;
528}
529
530/*
Willy Tarreau91092e52007-10-25 16:58:42 +0200531 * This function simply returns a locally allocated string containing the ascii
532 * representation for number 'n' in decimal, unless n is 0 in which case it
533 * returns the alternate string (or an empty string if the alternate string is
534 * NULL). It use is intended for limits reported in reports, where it's
535 * desirable not to display anything if there is no limit. Warning! it shares
536 * the same vector as ultoa_r().
537 */
538const char *limit_r(unsigned long n, char *buffer, int size, const char *alt)
539{
540 return (n) ? ultoa_r(n, buffer, size) : (alt ? alt : "");
541}
542
Willy Tarreau56d1d8d2021-05-08 10:28:53 +0200543/* Trims the first "%f" float in a string to its minimum number of digits after
544 * the decimal point by trimming trailing zeroes, even dropping the decimal
545 * point if not needed. The string is in <buffer> of length <len>, and the
546 * number is expected to start at or after position <num_start> (the first
547 * point appearing there is considered). A NUL character is always placed at
548 * the end if some trimming occurs. The new buffer length is returned.
549 */
550size_t flt_trim(char *buffer, size_t num_start, size_t len)
551{
552 char *end = buffer + len;
553 char *p = buffer + num_start;
554 char *trim;
555
556 do {
557 if (p >= end)
558 return len;
559 trim = p++;
560 } while (*trim != '.');
561
562 /* For now <trim> is on the decimal point. Let's look for any other
563 * meaningful digit after it.
564 */
565 while (p < end) {
566 if (*p++ != '0')
567 trim = p;
568 }
569
570 if (trim < end)
571 *trim = 0;
572
573 return trim - buffer;
574}
575
Willy Tarreauae03d262021-05-08 07:35:00 +0200576/*
577 * This function simply returns a locally allocated string containing
578 * the ascii representation for number 'n' in decimal with useless trailing
579 * zeroes trimmed.
580 */
581char *ftoa_r(double n, char *buffer, int size)
582{
583 flt_trim(buffer, 0, snprintf(buffer, size, "%f", n));
584 return buffer;
585}
586
Willy Tarreau588297f2014-06-16 15:16:40 +0200587/* returns a locally allocated string containing the quoted encoding of the
588 * input string. The output may be truncated to QSTR_SIZE chars, but it is
589 * guaranteed that the string will always be properly terminated. Quotes are
590 * encoded by doubling them as is commonly done in CSV files. QSTR_SIZE must
591 * always be at least 4 chars.
592 */
593const char *qstr(const char *str)
594{
595 char *ret = quoted_str[quoted_idx];
596 char *p, *end;
597
598 if (++quoted_idx >= NB_QSTR)
599 quoted_idx = 0;
600
601 p = ret;
602 end = ret + QSTR_SIZE;
603
604 *p++ = '"';
605
606 /* always keep 3 chars to support passing "" and the ending " */
607 while (*str && p < end - 3) {
608 if (*str == '"') {
609 *p++ = '"';
610 *p++ = '"';
611 }
612 else
613 *p++ = *str;
614 str++;
615 }
616 *p++ = '"';
617 return ret;
618}
619
Robert Tsai81ae1952007-12-05 10:47:29 +0100620/*
Willy Tarreaubaaee002006-06-26 02:48:02 +0200621 * Returns non-zero if character <s> is a hex digit (0-9, a-f, A-F), else zero.
622 *
623 * It looks like this one would be a good candidate for inlining, but this is
624 * not interesting because it around 35 bytes long and often called multiple
625 * times within the same function.
626 */
627int ishex(char s)
628{
629 s -= '0';
630 if ((unsigned char)s <= 9)
631 return 1;
632 s -= 'A' - '0';
633 if ((unsigned char)s <= 5)
634 return 1;
635 s -= 'a' - 'A';
636 if ((unsigned char)s <= 5)
637 return 1;
638 return 0;
639}
640
Willy Tarreau3ca1a882015-01-15 18:43:49 +0100641/* rounds <i> down to the closest value having max 2 digits */
642unsigned int round_2dig(unsigned int i)
643{
644 unsigned int mul = 1;
645
646 while (i >= 100) {
647 i /= 10;
648 mul *= 10;
649 }
650 return i * mul;
651}
652
Willy Tarreau2e74c3f2007-12-02 18:45:09 +0100653/*
654 * Checks <name> for invalid characters. Valid chars are [A-Za-z0-9_:.-]. If an
655 * invalid character is found, a pointer to it is returned. If everything is
656 * fine, NULL is returned.
657 */
658const char *invalid_char(const char *name)
659{
660 if (!*name)
661 return name;
662
663 while (*name) {
Willy Tarreau90807112020-02-25 08:16:33 +0100664 if (!isalnum((unsigned char)*name) && *name != '.' && *name != ':' &&
Willy Tarreau2e74c3f2007-12-02 18:45:09 +0100665 *name != '_' && *name != '-')
666 return name;
667 name++;
668 }
669 return NULL;
670}
Willy Tarreaubaaee002006-06-26 02:48:02 +0200671
672/*
Frédéric Lécailleb82f7422017-04-13 18:24:23 +0200673 * Checks <name> for invalid characters. Valid chars are [_.-] and those
674 * accepted by <f> function.
Krzysztof Piotr Oledzkiefe3b6f2008-05-23 23:49:32 +0200675 * If an invalid character is found, a pointer to it is returned.
676 * If everything is fine, NULL is returned.
677 */
Frédéric Lécailleb82f7422017-04-13 18:24:23 +0200678static inline const char *__invalid_char(const char *name, int (*f)(int)) {
Krzysztof Piotr Oledzkiefe3b6f2008-05-23 23:49:32 +0200679
680 if (!*name)
681 return name;
682
683 while (*name) {
Willy Tarreau90807112020-02-25 08:16:33 +0100684 if (!f((unsigned char)*name) && *name != '.' &&
Krzysztof Piotr Oledzkiefe3b6f2008-05-23 23:49:32 +0200685 *name != '_' && *name != '-')
686 return name;
687
688 name++;
689 }
690
691 return NULL;
692}
693
694/*
Frédéric Lécailleb82f7422017-04-13 18:24:23 +0200695 * Checks <name> for invalid characters. Valid chars are [A-Za-z0-9_.-].
696 * If an invalid character is found, a pointer to it is returned.
697 * If everything is fine, NULL is returned.
698 */
699const char *invalid_domainchar(const char *name) {
700 return __invalid_char(name, isalnum);
701}
702
703/*
704 * Checks <name> for invalid characters. Valid chars are [A-Za-z_.-].
705 * If an invalid character is found, a pointer to it is returned.
706 * If everything is fine, NULL is returned.
707 */
708const char *invalid_prefix_char(const char *name) {
Thierry Fournierf7b7c3e2018-03-26 11:54:39 +0200709 return __invalid_char(name, isalnum);
Frédéric Lécailleb82f7422017-04-13 18:24:23 +0200710}
711
712/*
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100713 * converts <str> to a struct sockaddr_storage* provided by the caller. The
Willy Tarreau24709282013-03-10 21:32:12 +0100714 * caller must have zeroed <sa> first, and may have set sa->ss_family to force
715 * parse a specific address format. If the ss_family is 0 or AF_UNSPEC, then
716 * the function tries to guess the address family from the syntax. If the
717 * family is forced and the format doesn't match, an error is returned. The
Willy Tarreaufab5a432011-03-04 15:31:53 +0100718 * string is assumed to contain only an address, no port. The address can be a
719 * dotted IPv4 address, an IPv6 address, a host name, or empty or "*" to
720 * indicate INADDR_ANY. NULL is returned if the host part cannot be resolved.
721 * The return address will only have the address family and the address set,
722 * all other fields remain zero. The string is not supposed to be modified.
Thierry FOURNIER58639a02014-11-25 12:02:25 +0100723 * The IPv6 '::' address is IN6ADDR_ANY. If <resolve> is non-zero, the hostname
724 * is resolved, otherwise only IP addresses are resolved, and anything else
Willy Tarreauecde7df2016-11-02 22:37:03 +0100725 * returns NULL. If the address contains a port, this one is preserved.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200726 */
Thierry FOURNIER58639a02014-11-25 12:02:25 +0100727struct sockaddr_storage *str2ip2(const char *str, struct sockaddr_storage *sa, int resolve)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200728{
Willy Tarreaufab5a432011-03-04 15:31:53 +0100729 struct hostent *he;
mildisff5d5102015-10-26 18:50:08 +0100730 /* max IPv6 length, including brackets and terminating NULL */
731 char tmpip[48];
Willy Tarreauecde7df2016-11-02 22:37:03 +0100732 int port = get_host_port(sa);
mildisff5d5102015-10-26 18:50:08 +0100733
734 /* check IPv6 with square brackets */
735 if (str[0] == '[') {
736 size_t iplength = strlen(str);
737
738 if (iplength < 4) {
739 /* minimal size is 4 when using brackets "[::]" */
740 goto fail;
741 }
742 else if (iplength >= sizeof(tmpip)) {
743 /* IPv6 literal can not be larger than tmpip */
744 goto fail;
745 }
746 else {
747 if (str[iplength - 1] != ']') {
748 /* if address started with bracket, it should end with bracket */
749 goto fail;
750 }
751 else {
752 memcpy(tmpip, str + 1, iplength - 2);
753 tmpip[iplength - 2] = '\0';
754 str = tmpip;
755 }
756 }
757 }
Willy Tarreaufab5a432011-03-04 15:31:53 +0100758
Willy Tarreaufab5a432011-03-04 15:31:53 +0100759 /* Any IPv6 address */
760 if (str[0] == ':' && str[1] == ':' && !str[2]) {
Willy Tarreau24709282013-03-10 21:32:12 +0100761 if (!sa->ss_family || sa->ss_family == AF_UNSPEC)
762 sa->ss_family = AF_INET6;
763 else if (sa->ss_family != AF_INET6)
764 goto fail;
Willy Tarreauecde7df2016-11-02 22:37:03 +0100765 set_host_port(sa, port);
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100766 return sa;
Willy Tarreaufab5a432011-03-04 15:31:53 +0100767 }
768
Willy Tarreau24709282013-03-10 21:32:12 +0100769 /* Any address for the family, defaults to IPv4 */
Willy Tarreaufab5a432011-03-04 15:31:53 +0100770 if (!str[0] || (str[0] == '*' && !str[1])) {
Willy Tarreau24709282013-03-10 21:32:12 +0100771 if (!sa->ss_family || sa->ss_family == AF_UNSPEC)
772 sa->ss_family = AF_INET;
Willy Tarreauecde7df2016-11-02 22:37:03 +0100773 set_host_port(sa, port);
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100774 return sa;
Willy Tarreaufab5a432011-03-04 15:31:53 +0100775 }
776
777 /* check for IPv6 first */
Willy Tarreau24709282013-03-10 21:32:12 +0100778 if ((!sa->ss_family || sa->ss_family == AF_UNSPEC || sa->ss_family == AF_INET6) &&
779 inet_pton(AF_INET6, str, &((struct sockaddr_in6 *)sa)->sin6_addr)) {
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100780 sa->ss_family = AF_INET6;
Willy Tarreauecde7df2016-11-02 22:37:03 +0100781 set_host_port(sa, port);
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100782 return sa;
Willy Tarreaufab5a432011-03-04 15:31:53 +0100783 }
784
785 /* then check for IPv4 */
Willy Tarreau24709282013-03-10 21:32:12 +0100786 if ((!sa->ss_family || sa->ss_family == AF_UNSPEC || sa->ss_family == AF_INET) &&
787 inet_pton(AF_INET, str, &((struct sockaddr_in *)sa)->sin_addr)) {
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100788 sa->ss_family = AF_INET;
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
Thierry FOURNIER58639a02014-11-25 12:02:25 +0100793 if (!resolve)
794 return NULL;
795
Emeric Brund30e9a12020-12-23 18:49:16 +0100796 if (!resolv_hostname_validation(str, NULL))
Baptiste Assmanna68ca962015-04-14 01:15:08 +0200797 return NULL;
798
David du Colombierd5f43282011-03-17 10:40:16 +0100799#ifdef USE_GETADDRINFO
Nenad Merdanovic88afe032014-04-14 15:56:58 +0200800 if (global.tune.options & GTUNE_USE_GAI) {
David du Colombierd5f43282011-03-17 10:40:16 +0100801 struct addrinfo hints, *result;
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100802 int success = 0;
David du Colombierd5f43282011-03-17 10:40:16 +0100803
804 memset(&result, 0, sizeof(result));
805 memset(&hints, 0, sizeof(hints));
Willy Tarreau24709282013-03-10 21:32:12 +0100806 hints.ai_family = sa->ss_family ? sa->ss_family : AF_UNSPEC;
David du Colombierd5f43282011-03-17 10:40:16 +0100807 hints.ai_socktype = SOCK_DGRAM;
Dmitry Sivachenkoeab7f392015-10-02 01:01:58 +0200808 hints.ai_flags = 0;
David du Colombierd5f43282011-03-17 10:40:16 +0100809 hints.ai_protocol = 0;
810
811 if (getaddrinfo(str, NULL, &hints, &result) == 0) {
Willy Tarreau24709282013-03-10 21:32:12 +0100812 if (!sa->ss_family || sa->ss_family == AF_UNSPEC)
813 sa->ss_family = result->ai_family;
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100814 else if (sa->ss_family != result->ai_family) {
815 freeaddrinfo(result);
Willy Tarreau24709282013-03-10 21:32:12 +0100816 goto fail;
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100817 }
Willy Tarreau24709282013-03-10 21:32:12 +0100818
David du Colombierd5f43282011-03-17 10:40:16 +0100819 switch (result->ai_family) {
820 case AF_INET:
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100821 memcpy((struct sockaddr_in *)sa, result->ai_addr, result->ai_addrlen);
Willy Tarreauecde7df2016-11-02 22:37:03 +0100822 set_host_port(sa, port);
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100823 success = 1;
824 break;
David du Colombierd5f43282011-03-17 10:40:16 +0100825 case AF_INET6:
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100826 memcpy((struct sockaddr_in6 *)sa, result->ai_addr, result->ai_addrlen);
Willy Tarreauecde7df2016-11-02 22:37:03 +0100827 set_host_port(sa, port);
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100828 success = 1;
829 break;
David du Colombierd5f43282011-03-17 10:40:16 +0100830 }
831 }
832
Sean Carey58ea0392013-02-15 23:39:18 +0100833 if (result)
834 freeaddrinfo(result);
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100835
836 if (success)
837 return sa;
Willy Tarreaufab5a432011-03-04 15:31:53 +0100838 }
David du Colombierd5f43282011-03-17 10:40:16 +0100839#endif
Nenad Merdanovic88afe032014-04-14 15:56:58 +0200840 /* try to resolve an IPv4/IPv6 hostname */
841 he = gethostbyname(str);
842 if (he) {
843 if (!sa->ss_family || sa->ss_family == AF_UNSPEC)
844 sa->ss_family = he->h_addrtype;
845 else if (sa->ss_family != he->h_addrtype)
846 goto fail;
847
848 switch (sa->ss_family) {
849 case AF_INET:
850 ((struct sockaddr_in *)sa)->sin_addr = *(struct in_addr *) *(he->h_addr_list);
Willy Tarreauecde7df2016-11-02 22:37:03 +0100851 set_host_port(sa, port);
Nenad Merdanovic88afe032014-04-14 15:56:58 +0200852 return sa;
853 case AF_INET6:
854 ((struct sockaddr_in6 *)sa)->sin6_addr = *(struct in6_addr *) *(he->h_addr_list);
Willy Tarreauecde7df2016-11-02 22:37:03 +0100855 set_host_port(sa, port);
Nenad Merdanovic88afe032014-04-14 15:56:58 +0200856 return sa;
857 }
858 }
859
David du Colombierd5f43282011-03-17 10:40:16 +0100860 /* unsupported address family */
Willy Tarreau24709282013-03-10 21:32:12 +0100861 fail:
Willy Tarreaufab5a432011-03-04 15:31:53 +0100862 return NULL;
863}
864
865/*
Willy Tarreaud4448bc2013-02-20 15:55:15 +0100866 * Converts <str> to a locally allocated struct sockaddr_storage *, and a port
867 * range or offset consisting in two integers that the caller will have to
868 * check to find the relevant input format. The following format are supported :
869 *
870 * String format | address | port | low | high
871 * addr | <addr> | 0 | 0 | 0
872 * addr: | <addr> | 0 | 0 | 0
873 * addr:port | <addr> | <port> | <port> | <port>
874 * addr:pl-ph | <addr> | <pl> | <pl> | <ph>
875 * addr:+port | <addr> | <port> | 0 | <port>
876 * addr:-port | <addr> |-<port> | <port> | 0
877 *
878 * The detection of a port range or increment by the caller is made by
879 * comparing <low> and <high>. If both are equal, then port 0 means no port
880 * was specified. The caller may pass NULL for <low> and <high> if it is not
881 * interested in retrieving port ranges.
882 *
883 * Note that <addr> above may also be :
884 * - empty ("") => family will be AF_INET and address will be INADDR_ANY
885 * - "*" => family will be AF_INET and address will be INADDR_ANY
886 * - "::" => family will be AF_INET6 and address will be IN6ADDR_ANY
887 * - a host name => family and address will depend on host name resolving.
888 *
Willy Tarreau24709282013-03-10 21:32:12 +0100889 * A prefix may be passed in before the address above to force the family :
890 * - "ipv4@" => force address to resolve as IPv4 and fail if not possible.
891 * - "ipv6@" => force address to resolve as IPv6 and fail if not possible.
892 * - "unix@" => force address to be a path to a UNIX socket even if the
893 * path does not start with a '/'
Willy Tarreauccfccef2014-05-10 01:49:15 +0200894 * - 'abns@' -> force address to belong to the abstract namespace (Linux
895 * only). These sockets are just like Unix sockets but without
896 * the need for an underlying file system. The address is a
897 * string. Technically it's like a Unix socket with a zero in
898 * the first byte of the address.
Willy Tarreau40aa0702013-03-10 23:51:38 +0100899 * - "fd@" => an integer must follow, and is a file descriptor number.
Willy Tarreau24709282013-03-10 21:32:12 +0100900 *
mildisff5d5102015-10-26 18:50:08 +0100901 * IPv6 addresses can be declared with or without square brackets. When using
902 * square brackets for IPv6 addresses, the port separator (colon) is optional.
903 * If not using square brackets, and in order to avoid any ambiguity with
904 * IPv6 addresses, the last colon ':' is mandatory even when no port is specified.
905 * NULL is returned if the address cannot be parsed. The <low> and <high> ports
906 * are always initialized if non-null, even for non-IP families.
Willy Tarreaud393a622013-03-04 18:22:00 +0100907 *
908 * If <pfx> is non-null, it is used as a string prefix before any path-based
909 * address (typically the path to a unix socket).
Willy Tarreau40aa0702013-03-10 23:51:38 +0100910 *
Willy Tarreau72b8c1f2015-09-08 15:50:19 +0200911 * if <fqdn> is non-null, it will be filled with :
912 * - a pointer to the FQDN of the server name to resolve if there's one, and
913 * that the caller will have to free(),
914 * - NULL if there was an explicit address that doesn't require resolution.
915 *
Willy Tarreaucd3a55912020-09-04 15:30:46 +0200916 * Hostnames are only resolved if <opts> has PA_O_RESOLVE. Otherwise <fqdn> is
917 * still honored so it is possible for the caller to know whether a resolution
918 * failed by clearing this flag and checking if <fqdn> was filled, indicating
919 * the need for a resolution.
Thierry FOURNIER7fe3be72015-09-26 20:03:36 +0200920 *
Willy Tarreau40aa0702013-03-10 23:51:38 +0100921 * When a file descriptor is passed, its value is put into the s_addr part of
Willy Tarreaua5b325f2020-09-04 16:44:20 +0200922 * the address when cast to sockaddr_in and the address family is
923 * AF_CUST_EXISTING_FD.
Willy Tarreaua93e5c72020-09-15 14:01:16 +0200924 *
Willy Tarreau5fc93282020-09-16 18:25:03 +0200925 * The matching protocol will be set into <proto> if non-null.
926 *
Willy Tarreaua93e5c72020-09-15 14:01:16 +0200927 * Any known file descriptor is also assigned to <fd> if non-null, otherwise it
928 * is forced to -1.
Willy Tarreaufab5a432011-03-04 15:31:53 +0100929 */
Willy Tarreau5fc93282020-09-16 18:25:03 +0200930struct sockaddr_storage *str2sa_range(const char *str, int *port, int *low, int *high, int *fd,
931 struct protocol **proto, char **err,
932 const char *pfx, char **fqdn, unsigned int opts)
Willy Tarreaufab5a432011-03-04 15:31:53 +0100933{
Christopher Faulet1bc04c72017-10-29 20:14:08 +0100934 static THREAD_LOCAL struct sockaddr_storage ss;
David du Colombier6f5ccb12011-03-10 22:26:24 +0100935 struct sockaddr_storage *ret = NULL;
Willy Tarreau5fc93282020-09-16 18:25:03 +0200936 struct protocol *new_proto = NULL;
Willy Tarreau24709282013-03-10 21:32:12 +0100937 char *back, *str2;
Willy Tarreaud4448bc2013-02-20 15:55:15 +0100938 char *port1, *port2;
939 int portl, porth, porta;
Willy Tarreauccfccef2014-05-10 01:49:15 +0200940 int abstract = 0;
Willy Tarreaua93e5c72020-09-15 14:01:16 +0200941 int new_fd = -1;
Willy Tarreaue835bd82020-09-16 11:35:47 +0200942 int sock_type, ctrl_type;
Willy Tarreaud4448bc2013-02-20 15:55:15 +0100943
944 portl = porth = porta = 0;
Willy Tarreau72b8c1f2015-09-08 15:50:19 +0200945 if (fqdn)
946 *fqdn = NULL;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200947
Willy Tarreaudad36a32013-03-11 01:20:04 +0100948 str2 = back = env_expand(strdup(str));
Willy Tarreaudf350f12013-03-01 20:22:54 +0100949 if (str2 == NULL) {
950 memprintf(err, "out of memory in '%s'\n", __FUNCTION__);
Willy Tarreaud5191e72010-02-09 20:50:45 +0100951 goto out;
Willy Tarreaudf350f12013-03-01 20:22:54 +0100952 }
Willy Tarreaubaaee002006-06-26 02:48:02 +0200953
Willy Tarreau9f69f462015-09-08 16:01:25 +0200954 if (!*str2) {
955 memprintf(err, "'%s' resolves to an empty address (environment variable missing?)\n", str);
956 goto out;
957 }
958
Willy Tarreau24709282013-03-10 21:32:12 +0100959 memset(&ss, 0, sizeof(ss));
960
Willy Tarreaue835bd82020-09-16 11:35:47 +0200961 /* prepare the default socket types */
Willy Tarreauf23b1bc2021-03-23 18:36:37 +0100962 if ((opts & (PA_O_STREAM|PA_O_DGRAM)) == PA_O_DGRAM ||
963 ((opts & (PA_O_STREAM|PA_O_DGRAM)) == (PA_O_DGRAM|PA_O_STREAM) && (opts & PA_O_DEFAULT_DGRAM)))
Willy Tarreaue835bd82020-09-16 11:35:47 +0200964 sock_type = ctrl_type = SOCK_DGRAM;
965 else
966 sock_type = ctrl_type = SOCK_STREAM;
967
968 if (strncmp(str2, "stream+", 7) == 0) {
969 str2 += 7;
970 sock_type = ctrl_type = SOCK_STREAM;
971 }
972 else if (strncmp(str2, "dgram+", 6) == 0) {
973 str2 += 6;
974 sock_type = ctrl_type = SOCK_DGRAM;
975 }
976
Willy Tarreau24709282013-03-10 21:32:12 +0100977 if (strncmp(str2, "unix@", 5) == 0) {
978 str2 += 5;
Willy Tarreauccfccef2014-05-10 01:49:15 +0200979 abstract = 0;
Willy Tarreau24709282013-03-10 21:32:12 +0100980 ss.ss_family = AF_UNIX;
981 }
Emeric Brunce325c42021-04-02 17:05:09 +0200982 else if (strncmp(str2, "uxdg@", 5) == 0) {
983 str2 += 5;
984 abstract = 0;
985 ss.ss_family = AF_UNIX;
986 sock_type = ctrl_type = SOCK_DGRAM;
987 }
988 else if (strncmp(str2, "uxst@", 5) == 0) {
989 str2 += 5;
990 abstract = 0;
991 ss.ss_family = AF_UNIX;
992 sock_type = ctrl_type = SOCK_STREAM;
993 }
Willy Tarreauccfccef2014-05-10 01:49:15 +0200994 else if (strncmp(str2, "abns@", 5) == 0) {
995 str2 += 5;
996 abstract = 1;
997 ss.ss_family = AF_UNIX;
998 }
Emeric Brunce325c42021-04-02 17:05:09 +0200999 else if (strncmp(str2, "ip@", 3) == 0) {
1000 str2 += 3;
1001 ss.ss_family = AF_UNSPEC;
1002 }
Willy Tarreau24709282013-03-10 21:32:12 +01001003 else if (strncmp(str2, "ipv4@", 5) == 0) {
1004 str2 += 5;
1005 ss.ss_family = AF_INET;
1006 }
1007 else if (strncmp(str2, "ipv6@", 5) == 0) {
1008 str2 += 5;
1009 ss.ss_family = AF_INET6;
1010 }
Emeric Brunce325c42021-04-02 17:05:09 +02001011 else if (strncmp(str2, "tcp4@", 5) == 0) {
1012 str2 += 5;
1013 ss.ss_family = AF_INET;
1014 sock_type = ctrl_type = SOCK_STREAM;
1015 }
Emeric Brun3835c0d2020-07-07 09:46:09 +02001016 else if (strncmp(str2, "udp4@", 5) == 0) {
1017 str2 += 5;
1018 ss.ss_family = AF_INET;
Willy Tarreaue835bd82020-09-16 11:35:47 +02001019 sock_type = ctrl_type = SOCK_DGRAM;
Emeric Brun3835c0d2020-07-07 09:46:09 +02001020 }
Emeric Brunce325c42021-04-02 17:05:09 +02001021 else if (strncmp(str2, "tcp6@", 5) == 0) {
1022 str2 += 5;
1023 ss.ss_family = AF_INET6;
1024 sock_type = ctrl_type = SOCK_STREAM;
1025 }
Emeric Brun3835c0d2020-07-07 09:46:09 +02001026 else if (strncmp(str2, "udp6@", 5) == 0) {
1027 str2 += 5;
1028 ss.ss_family = AF_INET6;
Willy Tarreaue835bd82020-09-16 11:35:47 +02001029 sock_type = ctrl_type = SOCK_DGRAM;
Emeric Brun3835c0d2020-07-07 09:46:09 +02001030 }
Emeric Brunce325c42021-04-02 17:05:09 +02001031 else if (strncmp(str2, "tcp@", 4) == 0) {
1032 str2 += 4;
1033 ss.ss_family = AF_UNSPEC;
1034 sock_type = ctrl_type = SOCK_STREAM;
1035 }
Emeric Brun3835c0d2020-07-07 09:46:09 +02001036 else if (strncmp(str2, "udp@", 4) == 0) {
1037 str2 += 4;
1038 ss.ss_family = AF_UNSPEC;
Willy Tarreaue835bd82020-09-16 11:35:47 +02001039 sock_type = ctrl_type = SOCK_DGRAM;
Emeric Brun3835c0d2020-07-07 09:46:09 +02001040 }
Frédéric Lécaille10caf652020-11-23 11:36:57 +01001041 else if (strncmp(str2, "quic4@", 6) == 0) {
1042 str2 += 6;
1043 ss.ss_family = AF_INET;
1044 sock_type = SOCK_DGRAM;
1045 ctrl_type = SOCK_STREAM;
1046 }
1047 else if (strncmp(str2, "quic6@", 6) == 0) {
1048 str2 += 6;
1049 ss.ss_family = AF_INET6;
1050 sock_type = SOCK_DGRAM;
1051 ctrl_type = SOCK_STREAM;
1052 }
Willy Tarreau5a7beed2020-09-04 16:54:05 +02001053 else if (strncmp(str2, "fd@", 3) == 0) {
1054 str2 += 3;
1055 ss.ss_family = AF_CUST_EXISTING_FD;
1056 }
1057 else if (strncmp(str2, "sockpair@", 9) == 0) {
1058 str2 += 9;
1059 ss.ss_family = AF_CUST_SOCKPAIR;
1060 }
Willy Tarreau24709282013-03-10 21:32:12 +01001061 else if (*str2 == '/') {
1062 ss.ss_family = AF_UNIX;
1063 }
1064 else
1065 ss.ss_family = AF_UNSPEC;
1066
Willy Tarreau5a7beed2020-09-04 16:54:05 +02001067 if (ss.ss_family == AF_CUST_SOCKPAIR) {
Willy Tarreaua215be22020-09-16 10:14:16 +02001068 struct sockaddr_storage ss2;
1069 socklen_t addr_len;
William Lallemand2fe7dd02018-09-11 16:51:29 +02001070 char *endptr;
1071
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001072 new_fd = strtol(str2, &endptr, 10);
1073 if (!*str2 || new_fd < 0 || *endptr) {
William Lallemand2fe7dd02018-09-11 16:51:29 +02001074 memprintf(err, "file descriptor '%s' is not a valid integer in '%s'\n", str2, str);
1075 goto out;
1076 }
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001077
Willy Tarreaua215be22020-09-16 10:14:16 +02001078 /* just verify that it's a socket */
1079 addr_len = sizeof(ss2);
1080 if (getsockname(new_fd, (struct sockaddr *)&ss2, &addr_len) == -1) {
1081 memprintf(err, "cannot use file descriptor '%d' : %s.\n", new_fd, strerror(errno));
1082 goto out;
1083 }
1084
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001085 ((struct sockaddr_in *)&ss)->sin_addr.s_addr = new_fd;
1086 ((struct sockaddr_in *)&ss)->sin_port = 0;
William Lallemand2fe7dd02018-09-11 16:51:29 +02001087 }
Willy Tarreau5a7beed2020-09-04 16:54:05 +02001088 else if (ss.ss_family == AF_CUST_EXISTING_FD) {
Willy Tarreau40aa0702013-03-10 23:51:38 +01001089 char *endptr;
1090
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001091 new_fd = strtol(str2, &endptr, 10);
1092 if (!*str2 || new_fd < 0 || *endptr) {
Willy Tarreaudad36a32013-03-11 01:20:04 +01001093 memprintf(err, "file descriptor '%s' is not a valid integer in '%s'\n", str2, str);
Willy Tarreau40aa0702013-03-10 23:51:38 +01001094 goto out;
1095 }
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001096
Willy Tarreau6edc7222020-09-15 17:41:56 +02001097 if (opts & PA_O_SOCKET_FD) {
1098 socklen_t addr_len;
1099 int type;
1100
1101 addr_len = sizeof(ss);
1102 if (getsockname(new_fd, (struct sockaddr *)&ss, &addr_len) == -1) {
1103 memprintf(err, "cannot use file descriptor '%d' : %s.\n", new_fd, strerror(errno));
1104 goto out;
1105 }
1106
1107 addr_len = sizeof(type);
1108 if (getsockopt(new_fd, SOL_SOCKET, SO_TYPE, &type, &addr_len) != 0 ||
Willy Tarreaue835bd82020-09-16 11:35:47 +02001109 (type == SOCK_STREAM) != (sock_type == SOCK_STREAM)) {
Willy Tarreau6edc7222020-09-15 17:41:56 +02001110 memprintf(err, "socket on file descriptor '%d' is of the wrong type.\n", new_fd);
1111 goto out;
1112 }
1113
1114 porta = portl = porth = get_host_port(&ss);
1115 } else if (opts & PA_O_RAW_FD) {
1116 ((struct sockaddr_in *)&ss)->sin_addr.s_addr = new_fd;
1117 ((struct sockaddr_in *)&ss)->sin_port = 0;
1118 } else {
1119 memprintf(err, "a file descriptor is not acceptable here in '%s'\n", str);
1120 goto out;
1121 }
Willy Tarreau40aa0702013-03-10 23:51:38 +01001122 }
1123 else if (ss.ss_family == AF_UNIX) {
Willy Tarreau8daa9202019-06-16 18:16:33 +02001124 struct sockaddr_un *un = (struct sockaddr_un *)&ss;
Willy Tarreau15586382013-03-04 19:48:14 +01001125 int prefix_path_len;
1126 int max_path_len;
Willy Tarreau94ef3f32014-04-14 14:49:00 +02001127 int adr_len;
Willy Tarreau15586382013-03-04 19:48:14 +01001128
1129 /* complete unix socket path name during startup or soft-restart is
1130 * <unix_bind_prefix><path>.<pid>.<bak|tmp>
1131 */
Willy Tarreauccfccef2014-05-10 01:49:15 +02001132 prefix_path_len = (pfx && !abstract) ? strlen(pfx) : 0;
Willy Tarreau8daa9202019-06-16 18:16:33 +02001133 max_path_len = (sizeof(un->sun_path) - 1) -
Willy Tarreau327ea5a2020-02-11 06:43:37 +01001134 (abstract ? 0 : prefix_path_len + 1 + 5 + 1 + 3);
Willy Tarreau15586382013-03-04 19:48:14 +01001135
Willy Tarreau94ef3f32014-04-14 14:49:00 +02001136 adr_len = strlen(str2);
1137 if (adr_len > max_path_len) {
Willy Tarreau15586382013-03-04 19:48:14 +01001138 memprintf(err, "socket path '%s' too long (max %d)\n", str, max_path_len);
1139 goto out;
1140 }
1141
Willy Tarreauccfccef2014-05-10 01:49:15 +02001142 /* when abstract==1, we skip the first zero and copy all bytes except the trailing zero */
Willy Tarreau8daa9202019-06-16 18:16:33 +02001143 memset(un->sun_path, 0, sizeof(un->sun_path));
Willy Tarreau94ef3f32014-04-14 14:49:00 +02001144 if (prefix_path_len)
Willy Tarreau8daa9202019-06-16 18:16:33 +02001145 memcpy(un->sun_path, pfx, prefix_path_len);
1146 memcpy(un->sun_path + prefix_path_len + abstract, str2, adr_len + 1 - abstract);
Willy Tarreau15586382013-03-04 19:48:14 +01001147 }
Willy Tarreau24709282013-03-10 21:32:12 +01001148 else { /* IPv4 and IPv6 */
mildisff5d5102015-10-26 18:50:08 +01001149 char *end = str2 + strlen(str2);
1150 char *chr;
Willy Tarreau72b8c1f2015-09-08 15:50:19 +02001151
mildisff5d5102015-10-26 18:50:08 +01001152 /* search for : or ] whatever comes first */
1153 for (chr = end-1; chr > str2; chr--) {
1154 if (*chr == ']' || *chr == ':')
1155 break;
1156 }
1157
1158 if (*chr == ':') {
1159 /* Found a colon before a closing-bracket, must be a port separator.
1160 * This guarantee backward compatibility.
1161 */
Willy Tarreau7f96a842020-09-15 11:12:44 +02001162 if (!(opts & PA_O_PORT_OK)) {
1163 memprintf(err, "port specification not permitted here in '%s'", str);
1164 goto out;
1165 }
mildisff5d5102015-10-26 18:50:08 +01001166 *chr++ = '\0';
1167 port1 = chr;
1168 }
1169 else {
1170 /* Either no colon and no closing-bracket
1171 * or directly ending with a closing-bracket.
1172 * However, no port.
1173 */
Willy Tarreau7f96a842020-09-15 11:12:44 +02001174 if (opts & PA_O_PORT_MAND) {
1175 memprintf(err, "missing port specification in '%s'", str);
1176 goto out;
1177 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001178 port1 = "";
mildisff5d5102015-10-26 18:50:08 +01001179 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02001180
Willy Tarreau90807112020-02-25 08:16:33 +01001181 if (isdigit((unsigned char)*port1)) { /* single port or range */
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001182 port2 = strchr(port1, '-');
Willy Tarreau7f96a842020-09-15 11:12:44 +02001183 if (port2) {
1184 if (!(opts & PA_O_PORT_RANGE)) {
1185 memprintf(err, "port range not permitted here in '%s'", str);
1186 goto out;
1187 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001188 *port2++ = '\0';
Willy Tarreau7f96a842020-09-15 11:12:44 +02001189 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001190 else
1191 port2 = port1;
1192 portl = atoi(port1);
1193 porth = atoi(port2);
Willy Tarreau7f96a842020-09-15 11:12:44 +02001194
1195 if (portl < !!(opts & PA_O_PORT_MAND) || portl > 65535) {
1196 memprintf(err, "invalid port '%s'", port1);
1197 goto out;
1198 }
1199
1200 if (porth < !!(opts & PA_O_PORT_MAND) || porth > 65535) {
1201 memprintf(err, "invalid port '%s'", port2);
1202 goto out;
1203 }
1204
1205 if (portl > porth) {
1206 memprintf(err, "invalid port range '%d-%d'", portl, porth);
1207 goto out;
1208 }
1209
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001210 porta = portl;
1211 }
1212 else if (*port1 == '-') { /* negative offset */
Willy Tarreau7f96a842020-09-15 11:12:44 +02001213 if (!(opts & PA_O_PORT_OFS)) {
1214 memprintf(err, "port offset not permitted here in '%s'", str);
1215 goto out;
1216 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001217 portl = atoi(port1 + 1);
1218 porta = -portl;
1219 }
1220 else if (*port1 == '+') { /* positive offset */
Willy Tarreau7f96a842020-09-15 11:12:44 +02001221 if (!(opts & PA_O_PORT_OFS)) {
1222 memprintf(err, "port offset not permitted here in '%s'", str);
1223 goto out;
1224 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001225 porth = atoi(port1 + 1);
1226 porta = porth;
1227 }
1228 else if (*port1) { /* other any unexpected char */
Willy Tarreaudad36a32013-03-11 01:20:04 +01001229 memprintf(err, "invalid character '%c' in port number '%s' in '%s'\n", *port1, port1, str);
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001230 goto out;
1231 }
Willy Tarreau7f96a842020-09-15 11:12:44 +02001232 else if (opts & PA_O_PORT_MAND) {
1233 memprintf(err, "missing port specification in '%s'", str);
1234 goto out;
1235 }
Willy Tarreauceccdd72016-11-02 22:27:10 +01001236
1237 /* first try to parse the IP without resolving. If it fails, it
1238 * tells us we need to keep a copy of the FQDN to resolve later
1239 * and to enable DNS. In this case we can proceed if <fqdn> is
Willy Tarreaucd3a55912020-09-04 15:30:46 +02001240 * set or if PA_O_RESOLVE is set, otherwise it's an error.
Willy Tarreauceccdd72016-11-02 22:27:10 +01001241 */
1242 if (str2ip2(str2, &ss, 0) == NULL) {
Willy Tarreaucd3a55912020-09-04 15:30:46 +02001243 if ((!(opts & PA_O_RESOLVE) && !fqdn) ||
1244 ((opts & PA_O_RESOLVE) && str2ip2(str2, &ss, 1) == NULL)) {
Willy Tarreauceccdd72016-11-02 22:27:10 +01001245 memprintf(err, "invalid address: '%s' in '%s'\n", str2, str);
1246 goto out;
1247 }
Willy Tarreau72b8c1f2015-09-08 15:50:19 +02001248
Willy Tarreauceccdd72016-11-02 22:27:10 +01001249 if (fqdn) {
1250 if (str2 != back)
1251 memmove(back, str2, strlen(str2) + 1);
1252 *fqdn = back;
1253 back = NULL;
1254 }
Willy Tarreau72b8c1f2015-09-08 15:50:19 +02001255 }
Willy Tarreauceccdd72016-11-02 22:27:10 +01001256 set_host_port(&ss, porta);
Willy Tarreaue4c58c82013-03-06 15:28:17 +01001257 }
Willy Tarreaufab5a432011-03-04 15:31:53 +01001258
Willy Tarreaue835bd82020-09-16 11:35:47 +02001259 if (ctrl_type == SOCK_STREAM && !(opts & PA_O_STREAM)) {
1260 memprintf(err, "stream-type socket not acceptable in '%s'\n", str);
1261 goto out;
1262 }
1263 else if (ctrl_type == SOCK_DGRAM && !(opts & PA_O_DGRAM)) {
1264 memprintf(err, "dgram-type socket not acceptable in '%s'\n", str);
1265 goto out;
1266 }
1267
Willy Tarreau65ec4e32020-09-16 19:17:08 +02001268 if (proto || (opts & PA_O_CONNECT)) {
Willy Tarreau5fc93282020-09-16 18:25:03 +02001269 /* Note: if the caller asks for a proto, we must find one,
Emeric Brun26754902021-04-07 14:26:44 +02001270 * except if we inherit from a raw FD (family == AF_CUST_EXISTING_FD)
1271 * orif we return with an fqdn that will resolve later,
Willy Tarreau5fc93282020-09-16 18:25:03 +02001272 * in which case the address is not known yet (this is only
1273 * for servers actually).
1274 */
Willy Tarreaub2ffc992020-09-16 21:37:31 +02001275 new_proto = protocol_lookup(ss.ss_family,
Willy Tarreauaf9609b2020-09-16 22:04:46 +02001276 sock_type == SOCK_DGRAM,
1277 ctrl_type == SOCK_DGRAM);
Willy Tarreaub2ffc992020-09-16 21:37:31 +02001278
Emeric Brun26754902021-04-07 14:26:44 +02001279 if (!new_proto && (!fqdn || !*fqdn) && (ss.ss_family != AF_CUST_EXISTING_FD)) {
Willy Tarreau5fc93282020-09-16 18:25:03 +02001280 memprintf(err, "unsupported protocol family %d for address '%s'", ss.ss_family, str);
1281 goto out;
1282 }
Willy Tarreau65ec4e32020-09-16 19:17:08 +02001283
1284 if ((opts & PA_O_CONNECT) && new_proto && !new_proto->connect) {
1285 memprintf(err, "connect() not supported for this protocol family %d used by address '%s'", ss.ss_family, str);
1286 goto out;
1287 }
Willy Tarreau5fc93282020-09-16 18:25:03 +02001288 }
1289
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001290 ret = &ss;
Willy Tarreaud5191e72010-02-09 20:50:45 +01001291 out:
Willy Tarreau48ef4c92017-01-06 18:32:38 +01001292 if (port)
1293 *port = porta;
Willy Tarreaud4448bc2013-02-20 15:55:15 +01001294 if (low)
1295 *low = portl;
1296 if (high)
1297 *high = porth;
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001298 if (fd)
1299 *fd = new_fd;
Willy Tarreau5fc93282020-09-16 18:25:03 +02001300 if (proto)
1301 *proto = new_proto;
Willy Tarreau24709282013-03-10 21:32:12 +01001302 free(back);
Willy Tarreaud5191e72010-02-09 20:50:45 +01001303 return ret;
Willy Tarreauc6f4ce82009-06-10 11:09:37 +02001304}
1305
Thayne McCombs92149f92020-11-20 01:28:26 -07001306/* converts <addr> and <port> into a string representation of the address and port. This is sort
1307 * of an inverse of str2sa_range, with some restrictions. The supported families are AF_INET,
1308 * AF_INET6, AF_UNIX, and AF_CUST_SOCKPAIR. If the family is unsopported NULL is returned.
1309 * If map_ports is true, then the sign of the port is included in the output, to indicate it is
1310 * relative to the incoming port. AF_INET and AF_INET6 will be in the form "<addr>:<port>".
1311 * AF_UNIX will either be just the path (if using a pathname) or "abns@<path>" if it is abstract.
1312 * AF_CUST_SOCKPAIR will be of the form "sockpair@<fd>".
1313 *
1314 * The returned char* is allocated, and it is the responsibility of the caller to free it.
1315 */
1316char * sa2str(const struct sockaddr_storage *addr, int port, int map_ports)
1317{
1318 char buffer[INET6_ADDRSTRLEN];
1319 char *out = NULL;
1320 const void *ptr;
1321 const char *path;
1322
1323 switch (addr->ss_family) {
1324 case AF_INET:
1325 ptr = &((struct sockaddr_in *)addr)->sin_addr;
1326 break;
1327 case AF_INET6:
1328 ptr = &((struct sockaddr_in6 *)addr)->sin6_addr;
1329 break;
1330 case AF_UNIX:
1331 path = ((struct sockaddr_un *)addr)->sun_path;
1332 if (path[0] == '\0') {
1333 const int max_length = sizeof(struct sockaddr_un) - offsetof(struct sockaddr_un, sun_path) - 1;
1334 return memprintf(&out, "abns@%.*s", max_length, path+1);
1335 } else {
1336 return strdup(path);
1337 }
1338 case AF_CUST_SOCKPAIR:
1339 return memprintf(&out, "sockpair@%d", ((struct sockaddr_in *)addr)->sin_addr.s_addr);
1340 default:
1341 return NULL;
1342 }
1343 inet_ntop(addr->ss_family, ptr, buffer, get_addr_len(addr));
1344 if (map_ports)
1345 return memprintf(&out, "%s:%+d", buffer, port);
1346 else
1347 return memprintf(&out, "%s:%d", buffer, port);
1348}
1349
1350
Willy Tarreau2937c0d2010-01-26 17:36:17 +01001351/* converts <str> to a struct in_addr containing a network mask. It can be
1352 * passed in dotted form (255.255.255.0) or in CIDR form (24). It returns 1
Jarno Huuskonen577d5ac2017-05-21 17:32:21 +03001353 * if the conversion succeeds otherwise zero.
Willy Tarreau2937c0d2010-01-26 17:36:17 +01001354 */
1355int str2mask(const char *str, struct in_addr *mask)
1356{
1357 if (strchr(str, '.') != NULL) { /* dotted notation */
1358 if (!inet_pton(AF_INET, str, mask))
1359 return 0;
1360 }
1361 else { /* mask length */
1362 char *err;
1363 unsigned long len = strtol(str, &err, 10);
1364
1365 if (!*str || (err && *err) || (unsigned)len > 32)
1366 return 0;
Tim Duesterhus8575f722018-01-25 16:24:48 +01001367
1368 len2mask4(len, mask);
Willy Tarreau2937c0d2010-01-26 17:36:17 +01001369 }
1370 return 1;
1371}
1372
Tim Duesterhus47185172018-01-25 16:24:49 +01001373/* converts <str> to a struct in6_addr containing a network mask. It can be
Tim Duesterhus5e642862018-02-20 17:02:18 +01001374 * passed in quadruplet form (ffff:ffff::) or in CIDR form (64). It returns 1
Tim Duesterhus47185172018-01-25 16:24:49 +01001375 * if the conversion succeeds otherwise zero.
1376 */
1377int str2mask6(const char *str, struct in6_addr *mask)
1378{
1379 if (strchr(str, ':') != NULL) { /* quadruplet notation */
1380 if (!inet_pton(AF_INET6, str, mask))
1381 return 0;
1382 }
1383 else { /* mask length */
1384 char *err;
1385 unsigned long len = strtol(str, &err, 10);
1386
1387 if (!*str || (err && *err) || (unsigned)len > 128)
1388 return 0;
1389
1390 len2mask6(len, mask);
1391 }
1392 return 1;
1393}
1394
Thierry FOURNIERb0504632013-12-14 15:39:02 +01001395/* convert <cidr> to struct in_addr <mask>. It returns 1 if the conversion
1396 * succeeds otherwise zero.
1397 */
1398int cidr2dotted(int cidr, struct in_addr *mask) {
1399
1400 if (cidr < 0 || cidr > 32)
1401 return 0;
1402
1403 mask->s_addr = cidr ? htonl(~0UL << (32 - cidr)) : 0;
1404 return 1;
1405}
1406
Thierry Fournier70473a52016-02-17 17:12:14 +01001407/* Convert mask from bit length form to in_addr form.
1408 * This function never fails.
1409 */
1410void len2mask4(int len, struct in_addr *addr)
1411{
1412 if (len >= 32) {
1413 addr->s_addr = 0xffffffff;
1414 return;
1415 }
1416 if (len <= 0) {
1417 addr->s_addr = 0x00000000;
1418 return;
1419 }
1420 addr->s_addr = 0xffffffff << (32 - len);
1421 addr->s_addr = htonl(addr->s_addr);
1422}
1423
1424/* Convert mask from bit length form to in6_addr form.
1425 * This function never fails.
1426 */
1427void len2mask6(int len, struct in6_addr *addr)
1428{
1429 len2mask4(len, (struct in_addr *)&addr->s6_addr[0]); /* msb */
1430 len -= 32;
1431 len2mask4(len, (struct in_addr *)&addr->s6_addr[4]);
1432 len -= 32;
1433 len2mask4(len, (struct in_addr *)&addr->s6_addr[8]);
1434 len -= 32;
1435 len2mask4(len, (struct in_addr *)&addr->s6_addr[12]); /* lsb */
1436}
1437
Willy Tarreauc6f4ce82009-06-10 11:09:37 +02001438/*
Willy Tarreaud077a8e2007-05-08 18:28:09 +02001439 * converts <str> to two struct in_addr* which must be pre-allocated.
Willy Tarreaubaaee002006-06-26 02:48:02 +02001440 * The format is "addr[/mask]", where "addr" cannot be empty, and mask
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05001441 * is optional and either in the dotted or CIDR notation.
Willy Tarreaubaaee002006-06-26 02:48:02 +02001442 * Note: "addr" can also be a hostname. Returns 1 if OK, 0 if error.
1443 */
Thierry FOURNIERfc7ac7b2014-02-11 15:23:04 +01001444int str2net(const char *str, int resolve, struct in_addr *addr, struct in_addr *mask)
Willy Tarreaubaaee002006-06-26 02:48:02 +02001445{
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001446 __label__ out_free, out_err;
1447 char *c, *s;
1448 int ret_val;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001449
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001450 s = strdup(str);
1451 if (!s)
1452 return 0;
1453
Willy Tarreaubaaee002006-06-26 02:48:02 +02001454 memset(mask, 0, sizeof(*mask));
1455 memset(addr, 0, sizeof(*addr));
Willy Tarreaubaaee002006-06-26 02:48:02 +02001456
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001457 if ((c = strrchr(s, '/')) != NULL) {
Willy Tarreaubaaee002006-06-26 02:48:02 +02001458 *c++ = '\0';
1459 /* c points to the mask */
Willy Tarreau2937c0d2010-01-26 17:36:17 +01001460 if (!str2mask(c, mask))
1461 goto out_err;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001462 }
1463 else {
Willy Tarreauebd61602006-12-30 11:54:15 +01001464 mask->s_addr = ~0U;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001465 }
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001466 if (!inet_pton(AF_INET, s, addr)) {
Willy Tarreaubaaee002006-06-26 02:48:02 +02001467 struct hostent *he;
1468
Thierry FOURNIERfc7ac7b2014-02-11 15:23:04 +01001469 if (!resolve)
1470 goto out_err;
1471
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001472 if ((he = gethostbyname(s)) == NULL) {
1473 goto out_err;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001474 }
1475 else
1476 *addr = *(struct in_addr *) *(he->h_addr_list);
1477 }
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001478
1479 ret_val = 1;
1480 out_free:
1481 free(s);
1482 return ret_val;
1483 out_err:
1484 ret_val = 0;
1485 goto out_free;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001486}
1487
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001488
1489/*
Willy Tarreau6d20e282012-04-27 22:49:47 +02001490 * converts <str> to two struct in6_addr* which must be pre-allocated.
1491 * The format is "addr[/mask]", where "addr" cannot be empty, and mask
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05001492 * is an optional number of bits (128 being the default).
Willy Tarreau6d20e282012-04-27 22:49:47 +02001493 * Returns 1 if OK, 0 if error.
1494 */
1495int str62net(const char *str, struct in6_addr *addr, unsigned char *mask)
1496{
1497 char *c, *s;
1498 int ret_val = 0;
1499 char *err;
1500 unsigned long len = 128;
1501
1502 s = strdup(str);
1503 if (!s)
1504 return 0;
1505
1506 memset(mask, 0, sizeof(*mask));
1507 memset(addr, 0, sizeof(*addr));
1508
1509 if ((c = strrchr(s, '/')) != NULL) {
1510 *c++ = '\0'; /* c points to the mask */
1511 if (!*c)
1512 goto out_free;
1513
1514 len = strtoul(c, &err, 10);
1515 if ((err && *err) || (unsigned)len > 128)
1516 goto out_free;
1517 }
1518 *mask = len; /* OK we have a valid mask in <len> */
1519
1520 if (!inet_pton(AF_INET6, s, addr))
1521 goto out_free;
1522
1523 ret_val = 1;
1524 out_free:
1525 free(s);
1526 return ret_val;
1527}
1528
1529
1530/*
Willy Tarreau12e10272021-03-25 11:34:40 +01001531 * Parse IPv4 address found in url. Return the number of bytes parsed. It
1532 * expects exactly 4 numbers between 0 and 255 delimited by dots, and returns
1533 * zero in case of mismatch.
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001534 */
David du Colombier6f5ccb12011-03-10 22:26:24 +01001535int url2ipv4(const char *addr, struct in_addr *dst)
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001536{
1537 int saw_digit, octets, ch;
1538 u_char tmp[4], *tp;
1539 const char *cp = addr;
1540
1541 saw_digit = 0;
1542 octets = 0;
1543 *(tp = tmp) = 0;
1544
1545 while (*addr) {
Willy Tarreau12e10272021-03-25 11:34:40 +01001546 unsigned char digit = (ch = *addr) - '0';
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001547 if (digit > 9 && ch != '.')
1548 break;
Willy Tarreau12e10272021-03-25 11:34:40 +01001549 addr++;
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001550 if (digit <= 9) {
1551 u_int new = *tp * 10 + digit;
1552 if (new > 255)
1553 return 0;
1554 *tp = new;
1555 if (!saw_digit) {
1556 if (++octets > 4)
1557 return 0;
1558 saw_digit = 1;
1559 }
1560 } else if (ch == '.' && saw_digit) {
1561 if (octets == 4)
1562 return 0;
1563 *++tp = 0;
1564 saw_digit = 0;
1565 } else
1566 return 0;
1567 }
1568
1569 if (octets < 4)
1570 return 0;
1571
1572 memcpy(&dst->s_addr, tmp, 4);
Willy Tarreau12e10272021-03-25 11:34:40 +01001573 return addr - cp;
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001574}
1575
1576/*
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001577 * Resolve destination server from URL. Convert <str> to a sockaddr_storage.
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05001578 * <out> contain the code of the detected scheme, the start and length of
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001579 * the hostname. Actually only http and https are supported. <out> can be NULL.
1580 * This function returns the consumed length. It is useful if you parse complete
1581 * url like http://host:port/path, because the consumed length corresponds to
1582 * the first character of the path. If the conversion fails, it returns -1.
1583 *
1584 * This function tries to resolve the DNS name if haproxy is in starting mode.
1585 * So, this function may be used during the configuration parsing.
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001586 */
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001587int url2sa(const char *url, int ulen, struct sockaddr_storage *addr, struct split_url *out)
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001588{
1589 const char *curr = url, *cp = url;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001590 const char *end;
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001591 int ret, url_code = 0;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001592 unsigned long long int http_code = 0;
1593 int default_port;
1594 struct hostent *he;
1595 char *p;
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001596
1597 /* Firstly, try to find :// pattern */
1598 while (curr < url+ulen && url_code != 0x3a2f2f) {
1599 url_code = ((url_code & 0xffff) << 8);
1600 url_code += (unsigned char)*curr++;
1601 }
1602
1603 /* Secondly, if :// pattern is found, verify parsed stuff
1604 * before pattern is matching our http pattern.
1605 * If so parse ip address and port in uri.
1606 *
1607 * WARNING: Current code doesn't support dynamic async dns resolver.
1608 */
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001609 if (url_code != 0x3a2f2f)
1610 return -1;
1611
1612 /* Copy scheme, and utrn to lower case. */
1613 while (cp < curr - 3)
1614 http_code = (http_code << 8) + *cp++;
1615 http_code |= 0x2020202020202020ULL; /* Turn everything to lower case */
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001616
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001617 /* HTTP or HTTPS url matching */
1618 if (http_code == 0x2020202068747470ULL) {
1619 default_port = 80;
1620 if (out)
1621 out->scheme = SCH_HTTP;
1622 }
1623 else if (http_code == 0x2020206874747073ULL) {
1624 default_port = 443;
1625 if (out)
1626 out->scheme = SCH_HTTPS;
1627 }
1628 else
1629 return -1;
1630
1631 /* If the next char is '[', the host address is IPv6. */
1632 if (*curr == '[') {
1633 curr++;
1634
1635 /* Check trash size */
1636 if (trash.size < ulen)
1637 return -1;
1638
1639 /* Look for ']' and copy the address in a trash buffer. */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001640 p = trash.area;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001641 for (end = curr;
1642 end < url + ulen && *end != ']';
1643 end++, p++)
1644 *p = *end;
1645 if (*end != ']')
1646 return -1;
1647 *p = '\0';
1648
1649 /* Update out. */
1650 if (out) {
1651 out->host = curr;
1652 out->host_len = end - curr;
1653 }
1654
1655 /* Try IPv6 decoding. */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001656 if (!inet_pton(AF_INET6, trash.area, &((struct sockaddr_in6 *)addr)->sin6_addr))
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001657 return -1;
1658 end++;
1659
1660 /* Decode port. */
1661 if (*end == ':') {
1662 end++;
1663 default_port = read_uint(&end, url + ulen);
1664 }
1665 ((struct sockaddr_in6 *)addr)->sin6_port = htons(default_port);
1666 ((struct sockaddr_in6 *)addr)->sin6_family = AF_INET6;
1667 return end - url;
1668 }
1669 else {
1670 /* We are looking for IP address. If you want to parse and
1671 * resolve hostname found in url, you can use str2sa_range(), but
1672 * be warned this can slow down global daemon performances
1673 * while handling lagging dns responses.
1674 */
1675 ret = url2ipv4(curr, &((struct sockaddr_in *)addr)->sin_addr);
1676 if (ret) {
1677 /* Update out. */
1678 if (out) {
1679 out->host = curr;
1680 out->host_len = ret;
1681 }
1682
1683 curr += ret;
1684
1685 /* Decode port. */
1686 if (*curr == ':') {
1687 curr++;
1688 default_port = read_uint(&curr, url + ulen);
1689 }
1690 ((struct sockaddr_in *)addr)->sin_port = htons(default_port);
1691
1692 /* Set family. */
1693 ((struct sockaddr_in *)addr)->sin_family = AF_INET;
1694 return curr - url;
1695 }
1696 else if (global.mode & MODE_STARTING) {
1697 /* The IPv4 and IPv6 decoding fails, maybe the url contain name. Try to execute
1698 * synchronous DNS request only if HAProxy is in the start state.
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001699 */
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001700
1701 /* look for : or / or end */
1702 for (end = curr;
1703 end < url + ulen && *end != '/' && *end != ':';
1704 end++);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001705 memcpy(trash.area, curr, end - curr);
1706 trash.area[end - curr] = '\0';
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001707
1708 /* try to resolve an IPv4/IPv6 hostname */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001709 he = gethostbyname(trash.area);
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001710 if (!he)
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001711 return -1;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001712
1713 /* Update out. */
1714 if (out) {
1715 out->host = curr;
1716 out->host_len = end - curr;
1717 }
1718
1719 /* Decode port. */
1720 if (*end == ':') {
1721 end++;
1722 default_port = read_uint(&end, url + ulen);
1723 }
1724
1725 /* Copy IP address, set port and family. */
1726 switch (he->h_addrtype) {
1727 case AF_INET:
1728 ((struct sockaddr_in *)addr)->sin_addr = *(struct in_addr *) *(he->h_addr_list);
1729 ((struct sockaddr_in *)addr)->sin_port = htons(default_port);
1730 ((struct sockaddr_in *)addr)->sin_family = AF_INET;
1731 return end - url;
1732
1733 case AF_INET6:
1734 ((struct sockaddr_in6 *)addr)->sin6_addr = *(struct in6_addr *) *(he->h_addr_list);
1735 ((struct sockaddr_in6 *)addr)->sin6_port = htons(default_port);
1736 ((struct sockaddr_in6 *)addr)->sin6_family = AF_INET6;
1737 return end - url;
1738 }
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001739 }
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001740 }
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001741 return -1;
1742}
1743
Willy Tarreau631f01c2011-09-05 00:36:48 +02001744/* Tries to convert a sockaddr_storage address to text form. Upon success, the
1745 * address family is returned so that it's easy for the caller to adapt to the
1746 * output format. Zero is returned if the address family is not supported. -1
1747 * is returned upon error, with errno set. AF_INET, AF_INET6 and AF_UNIX are
1748 * supported.
1749 */
Willy Tarreaud5ec4bf2019-04-25 17:48:16 +02001750int addr_to_str(const struct sockaddr_storage *addr, char *str, int size)
Willy Tarreau631f01c2011-09-05 00:36:48 +02001751{
1752
Willy Tarreaud5ec4bf2019-04-25 17:48:16 +02001753 const void *ptr;
Willy Tarreau631f01c2011-09-05 00:36:48 +02001754
1755 if (size < 5)
1756 return 0;
1757 *str = '\0';
1758
1759 switch (addr->ss_family) {
1760 case AF_INET:
1761 ptr = &((struct sockaddr_in *)addr)->sin_addr;
1762 break;
1763 case AF_INET6:
1764 ptr = &((struct sockaddr_in6 *)addr)->sin6_addr;
1765 break;
1766 case AF_UNIX:
1767 memcpy(str, "unix", 5);
1768 return addr->ss_family;
1769 default:
1770 return 0;
1771 }
1772
1773 if (inet_ntop(addr->ss_family, ptr, str, size))
1774 return addr->ss_family;
1775
1776 /* failed */
1777 return -1;
1778}
1779
Simon Horman75ab8bd2014-06-16 09:39:41 +09001780/* Tries to convert a sockaddr_storage port to text form. Upon success, the
1781 * address family is returned so that it's easy for the caller to adapt to the
1782 * output format. Zero is returned if the address family is not supported. -1
1783 * is returned upon error, with errno set. AF_INET, AF_INET6 and AF_UNIX are
1784 * supported.
1785 */
Willy Tarreaud5ec4bf2019-04-25 17:48:16 +02001786int port_to_str(const struct sockaddr_storage *addr, char *str, int size)
Simon Horman75ab8bd2014-06-16 09:39:41 +09001787{
1788
1789 uint16_t port;
1790
1791
Willy Tarreaud7dad1b2017-01-06 16:46:22 +01001792 if (size < 6)
Simon Horman75ab8bd2014-06-16 09:39:41 +09001793 return 0;
1794 *str = '\0';
1795
1796 switch (addr->ss_family) {
1797 case AF_INET:
1798 port = ((struct sockaddr_in *)addr)->sin_port;
1799 break;
1800 case AF_INET6:
1801 port = ((struct sockaddr_in6 *)addr)->sin6_port;
1802 break;
1803 case AF_UNIX:
1804 memcpy(str, "unix", 5);
1805 return addr->ss_family;
1806 default:
1807 return 0;
1808 }
1809
1810 snprintf(str, size, "%u", ntohs(port));
1811 return addr->ss_family;
1812}
1813
Willy Tarreau16e01562016-08-09 16:46:18 +02001814/* check if the given address is local to the system or not. It will return
1815 * -1 when it's not possible to know, 0 when the address is not local, 1 when
1816 * it is. We don't want to iterate over all interfaces for this (and it is not
1817 * portable). So instead we try to bind in UDP to this address on a free non
1818 * privileged port and to connect to the same address, port 0 (connect doesn't
1819 * care). If it succeeds, we own the address. Note that non-inet addresses are
1820 * considered local since they're most likely AF_UNIX.
1821 */
1822int addr_is_local(const struct netns_entry *ns,
1823 const struct sockaddr_storage *orig)
1824{
1825 struct sockaddr_storage addr;
1826 int result;
1827 int fd;
1828
1829 if (!is_inet_addr(orig))
1830 return 1;
1831
1832 memcpy(&addr, orig, sizeof(addr));
1833 set_host_port(&addr, 0);
1834
1835 fd = my_socketat(ns, addr.ss_family, SOCK_DGRAM, IPPROTO_UDP);
1836 if (fd < 0)
1837 return -1;
1838
1839 result = -1;
1840 if (bind(fd, (struct sockaddr *)&addr, get_addr_len(&addr)) == 0) {
1841 if (connect(fd, (struct sockaddr *)&addr, get_addr_len(&addr)) == -1)
1842 result = 0; // fail, non-local address
1843 else
1844 result = 1; // success, local address
1845 }
1846 else {
1847 if (errno == EADDRNOTAVAIL)
1848 result = 0; // definitely not local :-)
1849 }
1850 close(fd);
1851
1852 return result;
1853}
1854
Willy Tarreaubaaee002006-06-26 02:48:02 +02001855/* will try to encode the string <string> replacing all characters tagged in
1856 * <map> with the hexadecimal representation of their ASCII-code (2 digits)
1857 * prefixed by <escape>, and will store the result between <start> (included)
1858 * and <stop> (excluded), and will always terminate the string with a '\0'
1859 * before <stop>. The position of the '\0' is returned if the conversion
1860 * completes. If bytes are missing between <start> and <stop>, then the
1861 * conversion will be incomplete and truncated. If <stop> <= <start>, the '\0'
1862 * cannot even be stored so we return <start> without writing the 0.
1863 * The input string must also be zero-terminated.
1864 */
1865const char hextab[16] = "0123456789ABCDEF";
1866char *encode_string(char *start, char *stop,
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001867 const char escape, const long *map,
Willy Tarreaubaaee002006-06-26 02:48:02 +02001868 const char *string)
1869{
1870 if (start < stop) {
1871 stop--; /* reserve one byte for the final '\0' */
1872 while (start < stop && *string != '\0') {
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001873 if (!ha_bit_test((unsigned char)(*string), map))
Willy Tarreaubaaee002006-06-26 02:48:02 +02001874 *start++ = *string;
1875 else {
1876 if (start + 3 >= stop)
1877 break;
1878 *start++ = escape;
1879 *start++ = hextab[(*string >> 4) & 15];
1880 *start++ = hextab[*string & 15];
1881 }
1882 string++;
1883 }
1884 *start = '\0';
1885 }
1886 return start;
1887}
1888
Thierry FOURNIERe059ec92014-03-17 12:01:13 +01001889/*
1890 * Same behavior as encode_string() above, except that it encodes chunk
1891 * <chunk> instead of a string.
1892 */
1893char *encode_chunk(char *start, char *stop,
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001894 const char escape, const long *map,
Willy Tarreau83061a82018-07-13 11:56:34 +02001895 const struct buffer *chunk)
Thierry FOURNIERe059ec92014-03-17 12:01:13 +01001896{
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001897 char *str = chunk->area;
1898 char *end = chunk->area + chunk->data;
Thierry FOURNIERe059ec92014-03-17 12:01:13 +01001899
1900 if (start < stop) {
1901 stop--; /* reserve one byte for the final '\0' */
1902 while (start < stop && str < end) {
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001903 if (!ha_bit_test((unsigned char)(*str), map))
Thierry FOURNIERe059ec92014-03-17 12:01:13 +01001904 *start++ = *str;
1905 else {
1906 if (start + 3 >= stop)
1907 break;
1908 *start++ = escape;
1909 *start++ = hextab[(*str >> 4) & 15];
1910 *start++ = hextab[*str & 15];
1911 }
1912 str++;
1913 }
1914 *start = '\0';
1915 }
1916 return start;
1917}
1918
Dragan Dosen0edd1092016-02-12 13:23:02 +01001919/*
1920 * Tries to prefix characters tagged in the <map> with the <escape>
Dragan Dosen1a5d0602016-07-22 16:00:31 +02001921 * character. The input <string> must be zero-terminated. The result will
1922 * be stored between <start> (included) and <stop> (excluded). This
1923 * function will always try to terminate the resulting string with a '\0'
1924 * before <stop>, and will return its position if the conversion
1925 * completes.
1926 */
1927char *escape_string(char *start, char *stop,
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001928 const char escape, const long *map,
Dragan Dosen1a5d0602016-07-22 16:00:31 +02001929 const char *string)
1930{
1931 if (start < stop) {
1932 stop--; /* reserve one byte for the final '\0' */
1933 while (start < stop && *string != '\0') {
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001934 if (!ha_bit_test((unsigned char)(*string), map))
Dragan Dosen1a5d0602016-07-22 16:00:31 +02001935 *start++ = *string;
1936 else {
1937 if (start + 2 >= stop)
1938 break;
1939 *start++ = escape;
1940 *start++ = *string;
1941 }
1942 string++;
1943 }
1944 *start = '\0';
1945 }
1946 return start;
1947}
1948
1949/*
1950 * Tries to prefix characters tagged in the <map> with the <escape>
Dragan Dosen0edd1092016-02-12 13:23:02 +01001951 * character. <chunk> contains the input to be escaped. The result will be
1952 * stored between <start> (included) and <stop> (excluded). The function
1953 * will always try to terminate the resulting string with a '\0' before
1954 * <stop>, and will return its position if the conversion completes.
1955 */
1956char *escape_chunk(char *start, char *stop,
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001957 const char escape, const long *map,
Willy Tarreau83061a82018-07-13 11:56:34 +02001958 const struct buffer *chunk)
Dragan Dosen0edd1092016-02-12 13:23:02 +01001959{
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001960 char *str = chunk->area;
1961 char *end = chunk->area + chunk->data;
Dragan Dosen0edd1092016-02-12 13:23:02 +01001962
1963 if (start < stop) {
1964 stop--; /* reserve one byte for the final '\0' */
1965 while (start < stop && str < end) {
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001966 if (!ha_bit_test((unsigned char)(*str), map))
Dragan Dosen0edd1092016-02-12 13:23:02 +01001967 *start++ = *str;
1968 else {
1969 if (start + 2 >= stop)
1970 break;
1971 *start++ = escape;
1972 *start++ = *str;
1973 }
1974 str++;
1975 }
1976 *start = '\0';
1977 }
1978 return start;
1979}
1980
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001981/* Check a string for using it in a CSV output format. If the string contains
1982 * one of the following four char <">, <,>, CR or LF, the string is
1983 * encapsulated between <"> and the <"> are escaped by a <""> sequence.
1984 * <str> is the input string to be escaped. The function assumes that
1985 * the input string is null-terminated.
1986 *
1987 * If <quote> is 0, the result is returned escaped but without double quote.
Willy Tarreau898529b2016-01-06 18:07:04 +01001988 * It is useful if the escaped string is used between double quotes in the
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001989 * format.
1990 *
Willy Tarreau898529b2016-01-06 18:07:04 +01001991 * printf("..., \"%s\", ...\r\n", csv_enc(str, 0, &trash));
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001992 *
Willy Tarreaub631c292016-01-08 10:04:08 +01001993 * If <quote> is 1, the converter puts the quotes only if any reserved character
1994 * is present. If <quote> is 2, the converter always puts the quotes.
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001995 *
Willy Tarreau83061a82018-07-13 11:56:34 +02001996 * <output> is a struct buffer used for storing the output string.
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001997 *
Willy Tarreau898529b2016-01-06 18:07:04 +01001998 * The function returns the converted string on its output. If an error
1999 * occurs, the function returns an empty string. This type of output is useful
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002000 * for using the function directly as printf() argument.
2001 *
2002 * If the output buffer is too short to contain the input string, the result
2003 * is truncated.
Willy Tarreau898529b2016-01-06 18:07:04 +01002004 *
Willy Tarreaub631c292016-01-08 10:04:08 +01002005 * This function appends the encoding to the existing output chunk, and it
2006 * guarantees that it starts immediately at the first available character of
2007 * the chunk. Please use csv_enc() instead if you want to replace the output
2008 * chunk.
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002009 */
Willy Tarreau83061a82018-07-13 11:56:34 +02002010const char *csv_enc_append(const char *str, int quote, struct buffer *output)
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002011{
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002012 char *end = output->area + output->size;
2013 char *out = output->area + output->data;
Willy Tarreau898529b2016-01-06 18:07:04 +01002014 char *ptr = out;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002015
Willy Tarreaub631c292016-01-08 10:04:08 +01002016 if (quote == 1) {
2017 /* automatic quoting: first verify if we'll have to quote the string */
2018 if (!strpbrk(str, "\n\r,\""))
2019 quote = 0;
2020 }
2021
2022 if (quote)
2023 *ptr++ = '"';
2024
Willy Tarreau898529b2016-01-06 18:07:04 +01002025 while (*str && ptr < end - 2) { /* -2 for reserving space for <"> and \0. */
2026 *ptr = *str;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002027 if (*str == '"') {
Willy Tarreau898529b2016-01-06 18:07:04 +01002028 ptr++;
2029 if (ptr >= end - 2) {
2030 ptr--;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002031 break;
2032 }
Willy Tarreau898529b2016-01-06 18:07:04 +01002033 *ptr = '"';
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002034 }
Willy Tarreau898529b2016-01-06 18:07:04 +01002035 ptr++;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002036 str++;
2037 }
2038
Willy Tarreaub631c292016-01-08 10:04:08 +01002039 if (quote)
2040 *ptr++ = '"';
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002041
Willy Tarreau898529b2016-01-06 18:07:04 +01002042 *ptr = '\0';
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002043 output->data = ptr - output->area;
Willy Tarreau898529b2016-01-06 18:07:04 +01002044 return out;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002045}
2046
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002047/* Decode an URL-encoded string in-place. The resulting string might
2048 * be shorter. If some forbidden characters are found, the conversion is
Thierry FOURNIER5068d962013-10-04 16:27:27 +02002049 * aborted, the string is truncated before the issue and a negative value is
2050 * returned, otherwise the operation returns the length of the decoded string.
Willy Tarreau62ba9ba2020-04-23 17:54:47 +02002051 * If the 'in_form' argument is non-nul the string is assumed to be part of
2052 * an "application/x-www-form-urlencoded" encoded string, and the '+' will be
2053 * turned to a space. If it's zero, this will only be done after a question
2054 * mark ('?').
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002055 */
Willy Tarreau62ba9ba2020-04-23 17:54:47 +02002056int url_decode(char *string, int in_form)
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002057{
2058 char *in, *out;
Thierry FOURNIER5068d962013-10-04 16:27:27 +02002059 int ret = -1;
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002060
2061 in = string;
2062 out = string;
2063 while (*in) {
2064 switch (*in) {
2065 case '+' :
Willy Tarreau62ba9ba2020-04-23 17:54:47 +02002066 *out++ = in_form ? ' ' : *in;
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002067 break;
2068 case '%' :
2069 if (!ishex(in[1]) || !ishex(in[2]))
2070 goto end;
2071 *out++ = (hex2i(in[1]) << 4) + hex2i(in[2]);
2072 in += 2;
2073 break;
Willy Tarreau62ba9ba2020-04-23 17:54:47 +02002074 case '?':
2075 in_form = 1;
2076 /* fall through */
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002077 default:
2078 *out++ = *in;
2079 break;
2080 }
2081 in++;
2082 }
Thierry FOURNIER5068d962013-10-04 16:27:27 +02002083 ret = out - string; /* success */
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002084 end:
2085 *out = 0;
2086 return ret;
2087}
Willy Tarreaubaaee002006-06-26 02:48:02 +02002088
Willy Tarreau6911fa42007-03-04 18:06:08 +01002089unsigned int str2ui(const char *s)
2090{
2091 return __str2ui(s);
2092}
2093
2094unsigned int str2uic(const char *s)
2095{
2096 return __str2uic(s);
2097}
2098
2099unsigned int strl2ui(const char *s, int len)
2100{
2101 return __strl2ui(s, len);
2102}
2103
2104unsigned int strl2uic(const char *s, int len)
2105{
2106 return __strl2uic(s, len);
2107}
2108
Willy Tarreau4ec83cd2010-10-15 23:19:55 +02002109unsigned int read_uint(const char **s, const char *end)
2110{
2111 return __read_uint(s, end);
2112}
2113
Thierry FOURNIER763a5d82015-07-06 23:09:52 +02002114/* This function reads an unsigned integer from the string pointed to by <s> and
2115 * returns it. The <s> pointer is adjusted to point to the first unread char. The
2116 * function automatically stops at <end>. If the number overflows, the 2^64-1
2117 * value is returned.
2118 */
2119unsigned long long int read_uint64(const char **s, const char *end)
2120{
2121 const char *ptr = *s;
2122 unsigned long long int i = 0, tmp;
2123 unsigned int j;
2124
2125 while (ptr < end) {
2126
2127 /* read next char */
2128 j = *ptr - '0';
2129 if (j > 9)
2130 goto read_uint64_end;
2131
2132 /* add char to the number and check overflow. */
2133 tmp = i * 10;
2134 if (tmp / 10 != i) {
2135 i = ULLONG_MAX;
2136 goto read_uint64_eat;
2137 }
2138 if (ULLONG_MAX - tmp < j) {
2139 i = ULLONG_MAX;
2140 goto read_uint64_eat;
2141 }
2142 i = tmp + j;
2143 ptr++;
2144 }
2145read_uint64_eat:
2146 /* eat each numeric char */
2147 while (ptr < end) {
2148 if ((unsigned int)(*ptr - '0') > 9)
2149 break;
2150 ptr++;
2151 }
2152read_uint64_end:
2153 *s = ptr;
2154 return i;
2155}
2156
2157/* This function reads an integer from the string pointed to by <s> and returns
2158 * it. The <s> pointer is adjusted to point to the first unread char. The function
2159 * automatically stops at <end>. Il the number is bigger than 2^63-2, the 2^63-1
2160 * value is returned. If the number is lowest than -2^63-1, the -2^63 value is
2161 * returned.
2162 */
2163long long int read_int64(const char **s, const char *end)
2164{
2165 unsigned long long int i = 0;
2166 int neg = 0;
2167
2168 /* Look for minus char. */
2169 if (**s == '-') {
2170 neg = 1;
2171 (*s)++;
2172 }
2173 else if (**s == '+')
2174 (*s)++;
2175
2176 /* convert as positive number. */
2177 i = read_uint64(s, end);
2178
2179 if (neg) {
2180 if (i > 0x8000000000000000ULL)
2181 return LLONG_MIN;
2182 return -i;
2183 }
2184 if (i > 0x7fffffffffffffffULL)
2185 return LLONG_MAX;
2186 return i;
2187}
2188
Willy Tarreau6911fa42007-03-04 18:06:08 +01002189/* This one is 7 times faster than strtol() on athlon with checks.
2190 * It returns the value of the number composed of all valid digits read,
2191 * and can process negative numbers too.
2192 */
2193int strl2ic(const char *s, int len)
2194{
2195 int i = 0;
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002196 int j, k;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002197
2198 if (len > 0) {
2199 if (*s != '-') {
2200 /* positive number */
2201 while (len-- > 0) {
2202 j = (*s++) - '0';
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002203 k = i * 10;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002204 if (j > 9)
2205 break;
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002206 i = k + j;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002207 }
2208 } else {
2209 /* negative number */
2210 s++;
2211 while (--len > 0) {
2212 j = (*s++) - '0';
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002213 k = i * 10;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002214 if (j > 9)
2215 break;
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002216 i = k - j;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002217 }
2218 }
2219 }
2220 return i;
2221}
2222
2223
2224/* This function reads exactly <len> chars from <s> and converts them to a
2225 * signed integer which it stores into <ret>. It accurately detects any error
2226 * (truncated string, invalid chars, overflows). It is meant to be used in
2227 * applications designed for hostile environments. It returns zero when the
2228 * number has successfully been converted, non-zero otherwise. When an error
2229 * is returned, the <ret> value is left untouched. It is yet 5 to 40 times
2230 * faster than strtol().
2231 */
2232int strl2irc(const char *s, int len, int *ret)
2233{
2234 int i = 0;
2235 int j;
2236
2237 if (!len)
2238 return 1;
2239
2240 if (*s != '-') {
2241 /* positive number */
2242 while (len-- > 0) {
2243 j = (*s++) - '0';
2244 if (j > 9) return 1; /* invalid char */
2245 if (i > INT_MAX / 10) return 1; /* check for multiply overflow */
2246 i = i * 10;
2247 if (i + j < i) return 1; /* check for addition overflow */
2248 i = i + j;
2249 }
2250 } else {
2251 /* negative number */
2252 s++;
2253 while (--len > 0) {
2254 j = (*s++) - '0';
2255 if (j > 9) return 1; /* invalid char */
2256 if (i < INT_MIN / 10) return 1; /* check for multiply overflow */
2257 i = i * 10;
2258 if (i - j > i) return 1; /* check for subtract overflow */
2259 i = i - j;
2260 }
2261 }
2262 *ret = i;
2263 return 0;
2264}
2265
2266
2267/* This function reads exactly <len> chars from <s> and converts them to a
2268 * signed integer which it stores into <ret>. It accurately detects any error
2269 * (truncated string, invalid chars, overflows). It is meant to be used in
2270 * applications designed for hostile environments. It returns zero when the
2271 * number has successfully been converted, non-zero otherwise. When an error
2272 * is returned, the <ret> value is left untouched. It is about 3 times slower
William Dauchy060ffc82021-02-06 20:47:51 +01002273 * than strl2irc().
Willy Tarreau6911fa42007-03-04 18:06:08 +01002274 */
Willy Tarreau6911fa42007-03-04 18:06:08 +01002275
2276int strl2llrc(const char *s, int len, long long *ret)
2277{
2278 long long i = 0;
2279 int j;
2280
2281 if (!len)
2282 return 1;
2283
2284 if (*s != '-') {
2285 /* positive number */
2286 while (len-- > 0) {
2287 j = (*s++) - '0';
2288 if (j > 9) return 1; /* invalid char */
2289 if (i > LLONG_MAX / 10LL) return 1; /* check for multiply overflow */
2290 i = i * 10LL;
2291 if (i + j < i) return 1; /* check for addition overflow */
2292 i = i + j;
2293 }
2294 } else {
2295 /* negative number */
2296 s++;
2297 while (--len > 0) {
2298 j = (*s++) - '0';
2299 if (j > 9) return 1; /* invalid char */
2300 if (i < LLONG_MIN / 10LL) return 1; /* check for multiply overflow */
2301 i = i * 10LL;
2302 if (i - j > i) return 1; /* check for subtract overflow */
2303 i = i - j;
2304 }
2305 }
2306 *ret = i;
2307 return 0;
2308}
2309
Thierry FOURNIER511e9472014-01-23 17:40:34 +01002310/* This function is used with pat_parse_dotted_ver(). It converts a string
2311 * composed by two number separated by a dot. Each part must contain in 16 bits
2312 * because internally they will be represented as a 32-bit quantity stored in
2313 * a 64-bit integer. It returns zero when the number has successfully been
2314 * converted, non-zero otherwise. When an error is returned, the <ret> value
2315 * is left untouched.
2316 *
2317 * "1.3" -> 0x0000000000010003
2318 * "65535.65535" -> 0x00000000ffffffff
2319 */
2320int strl2llrc_dotted(const char *text, int len, long long *ret)
2321{
2322 const char *end = &text[len];
2323 const char *p;
2324 long long major, minor;
2325
2326 /* Look for dot. */
2327 for (p = text; p < end; p++)
2328 if (*p == '.')
2329 break;
2330
2331 /* Convert major. */
2332 if (strl2llrc(text, p - text, &major) != 0)
2333 return 1;
2334
2335 /* Check major. */
2336 if (major >= 65536)
2337 return 1;
2338
2339 /* Convert minor. */
2340 minor = 0;
2341 if (p < end)
2342 if (strl2llrc(p + 1, end - (p + 1), &minor) != 0)
2343 return 1;
2344
2345 /* Check minor. */
2346 if (minor >= 65536)
2347 return 1;
2348
2349 /* Compose value. */
2350 *ret = (major << 16) | (minor & 0xffff);
2351 return 0;
2352}
2353
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002354/* This function parses a time value optionally followed by a unit suffix among
2355 * "d", "h", "m", "s", "ms" or "us". It converts the value into the unit
2356 * expected by the caller. The computation does its best to avoid overflows.
2357 * The value is returned in <ret> if everything is fine, and a NULL is returned
2358 * by the function. In case of error, a pointer to the error is returned and
2359 * <ret> is left untouched. Values are automatically rounded up when needed.
Willy Tarreau9faebe32019-06-07 19:00:37 +02002360 * Values resulting in values larger than or equal to 2^31 after conversion are
2361 * reported as an overflow as value PARSE_TIME_OVER. Non-null values resulting
2362 * in an underflow are reported as an underflow as value PARSE_TIME_UNDER.
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002363 */
2364const char *parse_time_err(const char *text, unsigned *ret, unsigned unit_flags)
2365{
Willy Tarreau9faebe32019-06-07 19:00:37 +02002366 unsigned long long imult, idiv;
2367 unsigned long long omult, odiv;
2368 unsigned long long value, result;
Christopher Fauletc20ad0d2020-12-11 09:23:07 +01002369 const char *str = text;
2370
2371 if (!isdigit((unsigned char)*text))
2372 return text;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002373
2374 omult = odiv = 1;
2375
2376 switch (unit_flags & TIME_UNIT_MASK) {
2377 case TIME_UNIT_US: omult = 1000000; break;
2378 case TIME_UNIT_MS: omult = 1000; break;
2379 case TIME_UNIT_S: break;
2380 case TIME_UNIT_MIN: odiv = 60; break;
2381 case TIME_UNIT_HOUR: odiv = 3600; break;
2382 case TIME_UNIT_DAY: odiv = 86400; break;
2383 default: break;
2384 }
2385
2386 value = 0;
2387
2388 while (1) {
2389 unsigned int j;
2390
2391 j = *text - '0';
2392 if (j > 9)
2393 break;
2394 text++;
2395 value *= 10;
2396 value += j;
2397 }
2398
2399 imult = idiv = 1;
2400 switch (*text) {
2401 case '\0': /* no unit = default unit */
2402 imult = omult = idiv = odiv = 1;
Christopher Fauletc20ad0d2020-12-11 09:23:07 +01002403 goto end;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002404 case 's': /* second = unscaled unit */
2405 break;
2406 case 'u': /* microsecond : "us" */
2407 if (text[1] == 's') {
2408 idiv = 1000000;
2409 text++;
Thayne McCombsa6838052021-04-02 14:12:43 -06002410 break;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002411 }
Christopher Fauletc20ad0d2020-12-11 09:23:07 +01002412 return text;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002413 case 'm': /* millisecond : "ms" or minute: "m" */
2414 if (text[1] == 's') {
2415 idiv = 1000;
2416 text++;
2417 } else
2418 imult = 60;
2419 break;
2420 case 'h': /* hour : "h" */
2421 imult = 3600;
2422 break;
2423 case 'd': /* day : "d" */
2424 imult = 86400;
2425 break;
2426 default:
2427 return text;
2428 break;
2429 }
Christopher Fauletc20ad0d2020-12-11 09:23:07 +01002430 if (*(++text) != '\0') {
2431 ha_warning("unexpected character '%c' after the timer value '%s', only "
2432 "(us=microseconds,ms=milliseconds,s=seconds,m=minutes,h=hours,d=days) are supported."
2433 " This will be reported as an error in next versions.\n", *text, str);
2434 }
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002435
Christopher Fauletc20ad0d2020-12-11 09:23:07 +01002436 end:
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002437 if (omult % idiv == 0) { omult /= idiv; idiv = 1; }
2438 if (idiv % omult == 0) { idiv /= omult; omult = 1; }
2439 if (imult % odiv == 0) { imult /= odiv; odiv = 1; }
2440 if (odiv % imult == 0) { odiv /= imult; imult = 1; }
2441
Willy Tarreau9faebe32019-06-07 19:00:37 +02002442 result = (value * (imult * omult) + (idiv * odiv - 1)) / (idiv * odiv);
2443 if (result >= 0x80000000)
2444 return PARSE_TIME_OVER;
2445 if (!result && value)
2446 return PARSE_TIME_UNDER;
2447 *ret = result;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002448 return NULL;
2449}
Willy Tarreau6911fa42007-03-04 18:06:08 +01002450
Emeric Brun39132b22010-01-04 14:57:24 +01002451/* this function converts the string starting at <text> to an unsigned int
2452 * stored in <ret>. If an error is detected, the pointer to the unexpected
Joseph Herlant32b83272018-11-15 11:58:28 -08002453 * character is returned. If the conversion is successful, NULL is returned.
Emeric Brun39132b22010-01-04 14:57:24 +01002454 */
2455const char *parse_size_err(const char *text, unsigned *ret) {
2456 unsigned value = 0;
2457
Christopher Faulet82635a02020-12-11 09:30:45 +01002458 if (!isdigit((unsigned char)*text))
2459 return text;
2460
Emeric Brun39132b22010-01-04 14:57:24 +01002461 while (1) {
2462 unsigned int j;
2463
2464 j = *text - '0';
2465 if (j > 9)
2466 break;
2467 if (value > ~0U / 10)
2468 return text;
2469 value *= 10;
2470 if (value > (value + j))
2471 return text;
2472 value += j;
2473 text++;
2474 }
2475
2476 switch (*text) {
2477 case '\0':
2478 break;
2479 case 'K':
2480 case 'k':
2481 if (value > ~0U >> 10)
2482 return text;
2483 value = value << 10;
2484 break;
2485 case 'M':
2486 case 'm':
2487 if (value > ~0U >> 20)
2488 return text;
2489 value = value << 20;
2490 break;
2491 case 'G':
2492 case 'g':
2493 if (value > ~0U >> 30)
2494 return text;
2495 value = value << 30;
2496 break;
2497 default:
2498 return text;
2499 }
2500
Godbach58048a22015-01-28 17:36:16 +08002501 if (*text != '\0' && *++text != '\0')
2502 return text;
2503
Emeric Brun39132b22010-01-04 14:57:24 +01002504 *ret = value;
2505 return NULL;
2506}
2507
Willy Tarreau126d4062013-12-03 17:50:47 +01002508/*
2509 * Parse binary string written in hexadecimal (source) and store the decoded
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05002510 * result into binstr and set binstrlen to the length of binstr. Memory for
Willy Tarreau126d4062013-12-03 17:50:47 +01002511 * binstr is allocated by the function. In case of error, returns 0 with an
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05002512 * error message in err. In success case, it returns the consumed length.
Willy Tarreau126d4062013-12-03 17:50:47 +01002513 */
2514int parse_binary(const char *source, char **binstr, int *binstrlen, char **err)
2515{
2516 int len;
2517 const char *p = source;
2518 int i,j;
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002519 int alloc;
Willy Tarreau126d4062013-12-03 17:50:47 +01002520
2521 len = strlen(source);
2522 if (len % 2) {
2523 memprintf(err, "an even number of hex digit is expected");
2524 return 0;
2525 }
2526
2527 len = len >> 1;
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002528
Willy Tarreau126d4062013-12-03 17:50:47 +01002529 if (!*binstr) {
Tim Duesterhuse52b6e52020-09-12 20:26:43 +02002530 *binstr = calloc(len, sizeof(**binstr));
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002531 if (!*binstr) {
2532 memprintf(err, "out of memory while loading string pattern");
2533 return 0;
2534 }
2535 alloc = 1;
Willy Tarreau126d4062013-12-03 17:50:47 +01002536 }
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002537 else {
2538 if (*binstrlen < len) {
Joseph Herlant76dbe782018-11-15 12:01:22 -08002539 memprintf(err, "no space available in the buffer. expect %d, provides %d",
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002540 len, *binstrlen);
2541 return 0;
2542 }
2543 alloc = 0;
2544 }
2545 *binstrlen = len;
Willy Tarreau126d4062013-12-03 17:50:47 +01002546
2547 i = j = 0;
2548 while (j < len) {
2549 if (!ishex(p[i++]))
2550 goto bad_input;
2551 if (!ishex(p[i++]))
2552 goto bad_input;
2553 (*binstr)[j++] = (hex2i(p[i-2]) << 4) + hex2i(p[i-1]);
2554 }
Thierry FOURNIERee330af2014-01-21 11:36:14 +01002555 return len << 1;
Willy Tarreau126d4062013-12-03 17:50:47 +01002556
2557bad_input:
2558 memprintf(err, "an hex digit is expected (found '%c')", p[i-1]);
Willy Tarreau61cfdf42021-02-20 10:46:51 +01002559 if (alloc)
2560 ha_free(binstr);
Willy Tarreau126d4062013-12-03 17:50:47 +01002561 return 0;
2562}
2563
Willy Tarreau946ba592009-05-10 15:41:18 +02002564/* copies at most <n> characters from <src> and always terminates with '\0' */
2565char *my_strndup(const char *src, int n)
2566{
2567 int len = 0;
2568 char *ret;
2569
2570 while (len < n && src[len])
2571 len++;
2572
Vincent Bernat3c2f2f22016-04-03 13:48:42 +02002573 ret = malloc(len + 1);
Willy Tarreau946ba592009-05-10 15:41:18 +02002574 if (!ret)
2575 return ret;
2576 memcpy(ret, src, len);
2577 ret[len] = '\0';
2578 return ret;
2579}
2580
Baptiste Assmannbb77c8e2013-10-06 23:24:13 +02002581/*
2582 * search needle in haystack
2583 * returns the pointer if found, returns NULL otherwise
2584 */
2585const void *my_memmem(const void *haystack, size_t haystacklen, const void *needle, size_t needlelen)
2586{
2587 const void *c = NULL;
2588 unsigned char f;
2589
2590 if ((haystack == NULL) || (needle == NULL) || (haystacklen < needlelen))
2591 return NULL;
2592
2593 f = *(char *)needle;
2594 c = haystack;
2595 while ((c = memchr(c, f, haystacklen - (c - haystack))) != NULL) {
2596 if ((haystacklen - (c - haystack)) < needlelen)
2597 return NULL;
2598
2599 if (memcmp(c, needle, needlelen) == 0)
2600 return c;
2601 ++c;
2602 }
2603 return NULL;
2604}
2605
Ilya Shipitsinc02a23f2020-05-06 00:53:22 +05002606/* get length of the initial segment consisting entirely of bytes in <accept> */
Christopher Faulet5eb96cb2020-04-15 10:23:01 +02002607size_t my_memspn(const void *str, size_t len, const void *accept, size_t acceptlen)
2608{
2609 size_t ret = 0;
2610
2611 while (ret < len && memchr(accept, *((int *)str), acceptlen)) {
2612 str++;
2613 ret++;
2614 }
2615 return ret;
2616}
2617
Ilya Shipitsinc02a23f2020-05-06 00:53:22 +05002618/* get length of the initial segment consisting entirely of bytes not in <rejcet> */
Christopher Faulet5eb96cb2020-04-15 10:23:01 +02002619size_t my_memcspn(const void *str, size_t len, const void *reject, size_t rejectlen)
2620{
2621 size_t ret = 0;
2622
2623 while (ret < len) {
2624 if(memchr(reject, *((int *)str), rejectlen))
2625 return ret;
2626 str++;
2627 ret++;
2628 }
2629 return ret;
2630}
2631
Willy Tarreau482b00d2009-10-04 22:48:42 +02002632/* This function returns the first unused key greater than or equal to <key> in
2633 * ID tree <root>. Zero is returned if no place is found.
2634 */
2635unsigned int get_next_id(struct eb_root *root, unsigned int key)
2636{
2637 struct eb32_node *used;
2638
2639 do {
2640 used = eb32_lookup_ge(root, key);
2641 if (!used || used->key > key)
2642 return key; /* key is available */
2643 key++;
2644 } while (key);
2645 return key;
2646}
2647
Willy Tarreau9c1e15d2017-11-15 18:51:29 +01002648/* dump the full tree to <file> in DOT format for debugging purposes. Will
2649 * optionally highlight node <subj> if found, depending on operation <op> :
2650 * 0 : nothing
2651 * >0 : insertion, node/leaf are surrounded in red
2652 * <0 : removal, node/leaf are dashed with no background
2653 * Will optionally add "desc" as a label on the graph if set and non-null.
2654 */
2655void 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 +01002656{
2657 struct eb32sc_node *node;
2658 unsigned long scope = -1;
2659
Willy Tarreau9c1e15d2017-11-15 18:51:29 +01002660 fprintf(file, "digraph ebtree {\n");
2661
2662 if (desc && *desc) {
2663 fprintf(file,
2664 " fontname=\"fixed\";\n"
2665 " fontsize=8;\n"
2666 " label=\"%s\";\n", desc);
2667 }
2668
Willy Tarreaued3cda02017-11-15 15:04:05 +01002669 fprintf(file,
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002670 " node [fontname=\"fixed\" fontsize=8 shape=\"box\" style=\"filled\" color=\"black\" fillcolor=\"white\"];\n"
2671 " edge [fontname=\"fixed\" fontsize=8 style=\"solid\" color=\"magenta\" dir=\"forward\"];\n"
Willy Tarreaued3cda02017-11-15 15:04:05 +01002672 " \"%lx_n\" [label=\"root\\n%lx\"]\n", (long)eb_root_to_node(root), (long)root
2673 );
2674
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002675 fprintf(file, " \"%lx_n\" -> \"%lx_%c\" [taillabel=\"L\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002676 (long)eb_root_to_node(root),
2677 (long)eb_root_to_node(eb_clrtag(root->b[0])),
Willy Tarreaued3cda02017-11-15 15:04:05 +01002678 eb_gettag(root->b[0]) == EB_LEAF ? 'l' : 'n');
2679
2680 node = eb32sc_first(root, scope);
2681 while (node) {
2682 if (node->node.node_p) {
2683 /* node part is used */
Willy Tarreau9c1e15d2017-11-15 18:51:29 +01002684 fprintf(file, " \"%lx_n\" [label=\"%lx\\nkey=%u\\nscope=%lx\\nbit=%d\" fillcolor=\"lightskyblue1\" %s];\n",
2685 (long)node, (long)node, node->key, node->node_s, node->node.bit,
2686 (node == subj) ? (op < 0 ? "color=\"red\" style=\"dashed\"" : op > 0 ? "color=\"red\"" : "") : "");
Willy Tarreaued3cda02017-11-15 15:04:05 +01002687
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002688 fprintf(file, " \"%lx_n\" -> \"%lx_n\" [taillabel=\"%c\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002689 (long)node,
2690 (long)eb_root_to_node(eb_clrtag(node->node.node_p)),
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002691 eb_gettag(node->node.node_p) ? 'R' : 'L');
Willy Tarreaued3cda02017-11-15 15:04:05 +01002692
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002693 fprintf(file, " \"%lx_n\" -> \"%lx_%c\" [taillabel=\"L\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002694 (long)node,
2695 (long)eb_root_to_node(eb_clrtag(node->node.branches.b[0])),
Willy Tarreaued3cda02017-11-15 15:04:05 +01002696 eb_gettag(node->node.branches.b[0]) == EB_LEAF ? 'l' : 'n');
2697
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002698 fprintf(file, " \"%lx_n\" -> \"%lx_%c\" [taillabel=\"R\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002699 (long)node,
2700 (long)eb_root_to_node(eb_clrtag(node->node.branches.b[1])),
Willy Tarreaued3cda02017-11-15 15:04:05 +01002701 eb_gettag(node->node.branches.b[1]) == EB_LEAF ? 'l' : 'n');
2702 }
2703
Willy Tarreau9c1e15d2017-11-15 18:51:29 +01002704 fprintf(file, " \"%lx_l\" [label=\"%lx\\nkey=%u\\nscope=%lx\\npfx=%u\" fillcolor=\"yellow\" %s];\n",
2705 (long)node, (long)node, node->key, node->leaf_s, node->node.pfx,
2706 (node == subj) ? (op < 0 ? "color=\"red\" style=\"dashed\"" : op > 0 ? "color=\"red\"" : "") : "");
Willy Tarreaued3cda02017-11-15 15:04:05 +01002707
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002708 fprintf(file, " \"%lx_l\" -> \"%lx_n\" [taillabel=\"%c\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002709 (long)node,
2710 (long)eb_root_to_node(eb_clrtag(node->node.leaf_p)),
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002711 eb_gettag(node->node.leaf_p) ? 'R' : 'L');
Willy Tarreaued3cda02017-11-15 15:04:05 +01002712 node = eb32sc_next(node, scope);
2713 }
2714 fprintf(file, "}\n");
2715}
2716
Willy Tarreau348238b2010-01-18 15:05:57 +01002717/* This function compares a sample word possibly followed by blanks to another
2718 * clean word. The compare is case-insensitive. 1 is returned if both are equal,
2719 * otherwise zero. This intends to be used when checking HTTP headers for some
2720 * values. Note that it validates a word followed only by blanks but does not
2721 * validate a word followed by blanks then other chars.
2722 */
2723int word_match(const char *sample, int slen, const char *word, int wlen)
2724{
2725 if (slen < wlen)
2726 return 0;
2727
2728 while (wlen) {
2729 char c = *sample ^ *word;
2730 if (c && c != ('A' ^ 'a'))
2731 return 0;
2732 sample++;
2733 word++;
2734 slen--;
2735 wlen--;
2736 }
2737
2738 while (slen) {
2739 if (*sample != ' ' && *sample != '\t')
2740 return 0;
2741 sample++;
2742 slen--;
2743 }
2744 return 1;
2745}
Willy Tarreau482b00d2009-10-04 22:48:42 +02002746
Willy Tarreaud54bbdc2009-09-07 11:00:31 +02002747/* Converts any text-formatted IPv4 address to a host-order IPv4 address. It
2748 * is particularly fast because it avoids expensive operations such as
2749 * multiplies, which are optimized away at the end. It requires a properly
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05002750 * formatted address though (3 points).
Willy Tarreaud54bbdc2009-09-07 11:00:31 +02002751 */
2752unsigned int inetaddr_host(const char *text)
2753{
2754 const unsigned int ascii_zero = ('0' << 24) | ('0' << 16) | ('0' << 8) | '0';
2755 register unsigned int dig100, dig10, dig1;
2756 int s;
2757 const char *p, *d;
2758
2759 dig1 = dig10 = dig100 = ascii_zero;
2760 s = 24;
2761
2762 p = text;
2763 while (1) {
2764 if (((unsigned)(*p - '0')) <= 9) {
2765 p++;
2766 continue;
2767 }
2768
2769 /* here, we have a complete byte between <text> and <p> (exclusive) */
2770 if (p == text)
2771 goto end;
2772
2773 d = p - 1;
2774 dig1 |= (unsigned int)(*d << s);
2775 if (d == text)
2776 goto end;
2777
2778 d--;
2779 dig10 |= (unsigned int)(*d << s);
2780 if (d == text)
2781 goto end;
2782
2783 d--;
2784 dig100 |= (unsigned int)(*d << s);
2785 end:
2786 if (!s || *p != '.')
2787 break;
2788
2789 s -= 8;
2790 text = ++p;
2791 }
2792
2793 dig100 -= ascii_zero;
2794 dig10 -= ascii_zero;
2795 dig1 -= ascii_zero;
2796 return ((dig100 * 10) + dig10) * 10 + dig1;
2797}
2798
2799/*
2800 * Idem except the first unparsed character has to be passed in <stop>.
2801 */
2802unsigned int inetaddr_host_lim(const char *text, const char *stop)
2803{
2804 const unsigned int ascii_zero = ('0' << 24) | ('0' << 16) | ('0' << 8) | '0';
2805 register unsigned int dig100, dig10, dig1;
2806 int s;
2807 const char *p, *d;
2808
2809 dig1 = dig10 = dig100 = ascii_zero;
2810 s = 24;
2811
2812 p = text;
2813 while (1) {
2814 if (((unsigned)(*p - '0')) <= 9 && p < stop) {
2815 p++;
2816 continue;
2817 }
2818
2819 /* here, we have a complete byte between <text> and <p> (exclusive) */
2820 if (p == text)
2821 goto end;
2822
2823 d = p - 1;
2824 dig1 |= (unsigned int)(*d << s);
2825 if (d == text)
2826 goto end;
2827
2828 d--;
2829 dig10 |= (unsigned int)(*d << s);
2830 if (d == text)
2831 goto end;
2832
2833 d--;
2834 dig100 |= (unsigned int)(*d << s);
2835 end:
2836 if (!s || p == stop || *p != '.')
2837 break;
2838
2839 s -= 8;
2840 text = ++p;
2841 }
2842
2843 dig100 -= ascii_zero;
2844 dig10 -= ascii_zero;
2845 dig1 -= ascii_zero;
2846 return ((dig100 * 10) + dig10) * 10 + dig1;
2847}
2848
2849/*
2850 * Idem except the pointer to first unparsed byte is returned into <ret> which
2851 * must not be NULL.
2852 */
Willy Tarreau74172752010-10-15 23:21:42 +02002853unsigned int inetaddr_host_lim_ret(char *text, char *stop, char **ret)
Willy Tarreaud54bbdc2009-09-07 11:00:31 +02002854{
2855 const unsigned int ascii_zero = ('0' << 24) | ('0' << 16) | ('0' << 8) | '0';
2856 register unsigned int dig100, dig10, dig1;
2857 int s;
Willy Tarreau74172752010-10-15 23:21:42 +02002858 char *p, *d;
Willy Tarreaud54bbdc2009-09-07 11:00:31 +02002859
2860 dig1 = dig10 = dig100 = ascii_zero;
2861 s = 24;
2862
2863 p = text;
2864 while (1) {
2865 if (((unsigned)(*p - '0')) <= 9 && p < stop) {
2866 p++;
2867 continue;
2868 }
2869
2870 /* here, we have a complete byte between <text> and <p> (exclusive) */
2871 if (p == text)
2872 goto end;
2873
2874 d = p - 1;
2875 dig1 |= (unsigned int)(*d << s);
2876 if (d == text)
2877 goto end;
2878
2879 d--;
2880 dig10 |= (unsigned int)(*d << s);
2881 if (d == text)
2882 goto end;
2883
2884 d--;
2885 dig100 |= (unsigned int)(*d << s);
2886 end:
2887 if (!s || p == stop || *p != '.')
2888 break;
2889
2890 s -= 8;
2891 text = ++p;
2892 }
2893
2894 *ret = p;
2895 dig100 -= ascii_zero;
2896 dig10 -= ascii_zero;
2897 dig1 -= ascii_zero;
2898 return ((dig100 * 10) + dig10) * 10 + dig1;
2899}
2900
Willy Tarreauf0b38bf2010-06-06 13:22:23 +02002901/* Convert a fixed-length string to an IP address. Returns 0 in case of error,
2902 * or the number of chars read in case of success. Maybe this could be replaced
2903 * by one of the functions above. Also, apparently this function does not support
2904 * hosts above 255 and requires exactly 4 octets.
Willy Tarreau075415a2013-12-12 11:29:39 +01002905 * The destination is only modified on success.
Willy Tarreauf0b38bf2010-06-06 13:22:23 +02002906 */
2907int buf2ip(const char *buf, size_t len, struct in_addr *dst)
2908{
2909 const char *addr;
2910 int saw_digit, octets, ch;
2911 u_char tmp[4], *tp;
2912 const char *cp = buf;
2913
2914 saw_digit = 0;
2915 octets = 0;
2916 *(tp = tmp) = 0;
2917
2918 for (addr = buf; addr - buf < len; addr++) {
2919 unsigned char digit = (ch = *addr) - '0';
2920
2921 if (digit > 9 && ch != '.')
2922 break;
2923
2924 if (digit <= 9) {
2925 u_int new = *tp * 10 + digit;
2926
2927 if (new > 255)
2928 return 0;
2929
2930 *tp = new;
2931
2932 if (!saw_digit) {
2933 if (++octets > 4)
2934 return 0;
2935 saw_digit = 1;
2936 }
2937 } else if (ch == '.' && saw_digit) {
2938 if (octets == 4)
2939 return 0;
2940
2941 *++tp = 0;
2942 saw_digit = 0;
2943 } else
2944 return 0;
2945 }
2946
2947 if (octets < 4)
2948 return 0;
2949
2950 memcpy(&dst->s_addr, tmp, 4);
2951 return addr - cp;
2952}
2953
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002954/* This function converts the string in <buf> of the len <len> to
2955 * struct in6_addr <dst> which must be allocated by the caller.
2956 * This function returns 1 in success case, otherwise zero.
Willy Tarreau075415a2013-12-12 11:29:39 +01002957 * The destination is only modified on success.
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002958 */
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002959int buf2ip6(const char *buf, size_t len, struct in6_addr *dst)
2960{
Thierry FOURNIERcd659912013-12-11 12:33:54 +01002961 char null_term_ip6[INET6_ADDRSTRLEN + 1];
Willy Tarreau075415a2013-12-12 11:29:39 +01002962 struct in6_addr out;
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002963
Thierry FOURNIERcd659912013-12-11 12:33:54 +01002964 if (len > INET6_ADDRSTRLEN)
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002965 return 0;
2966
2967 memcpy(null_term_ip6, buf, len);
2968 null_term_ip6[len] = '\0';
2969
Willy Tarreau075415a2013-12-12 11:29:39 +01002970 if (!inet_pton(AF_INET6, null_term_ip6, &out))
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002971 return 0;
2972
Willy Tarreau075415a2013-12-12 11:29:39 +01002973 *dst = out;
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002974 return 1;
2975}
2976
Willy Tarreauacf95772010-06-14 19:09:21 +02002977/* To be used to quote config arg positions. Returns the short string at <ptr>
2978 * surrounded by simple quotes if <ptr> is valid and non-empty, or "end of line"
2979 * if ptr is NULL or empty. The string is locally allocated.
2980 */
2981const char *quote_arg(const char *ptr)
2982{
Christopher Faulet1bc04c72017-10-29 20:14:08 +01002983 static THREAD_LOCAL char val[32];
Willy Tarreauacf95772010-06-14 19:09:21 +02002984 int i;
2985
2986 if (!ptr || !*ptr)
2987 return "end of line";
2988 val[0] = '\'';
Willy Tarreaude2dd6b2013-01-24 02:14:42 +01002989 for (i = 1; i < sizeof(val) - 2 && *ptr; i++)
Willy Tarreauacf95772010-06-14 19:09:21 +02002990 val[i] = *ptr++;
2991 val[i++] = '\'';
2992 val[i] = '\0';
2993 return val;
2994}
2995
Willy Tarreau5b180202010-07-18 10:40:48 +02002996/* returns an operator among STD_OP_* for string <str> or < 0 if unknown */
2997int get_std_op(const char *str)
2998{
2999 int ret = -1;
3000
3001 if (*str == 'e' && str[1] == 'q')
3002 ret = STD_OP_EQ;
3003 else if (*str == 'n' && str[1] == 'e')
3004 ret = STD_OP_NE;
3005 else if (*str == 'l') {
3006 if (str[1] == 'e') ret = STD_OP_LE;
3007 else if (str[1] == 't') ret = STD_OP_LT;
3008 }
3009 else if (*str == 'g') {
3010 if (str[1] == 'e') ret = STD_OP_GE;
3011 else if (str[1] == 't') ret = STD_OP_GT;
3012 }
3013
3014 if (ret == -1 || str[2] != '\0')
3015 return -1;
3016 return ret;
3017}
3018
Willy Tarreau4c14eaa2010-11-24 14:01:45 +01003019/* hash a 32-bit integer to another 32-bit integer */
3020unsigned int full_hash(unsigned int a)
3021{
3022 return __full_hash(a);
3023}
3024
Willy Tarreauf3241112019-02-26 09:56:22 +01003025/* Return the bit position in mask <m> of the nth bit set of rank <r>, between
3026 * 0 and LONGBITS-1 included, starting from the left. For example ranks 0,1,2,3
3027 * for mask 0x55 will be 6, 4, 2 and 0 respectively. This algorithm is based on
3028 * a popcount variant and is described here :
3029 * https://graphics.stanford.edu/~seander/bithacks.html
3030 */
3031unsigned int mask_find_rank_bit(unsigned int r, unsigned long m)
3032{
3033 unsigned long a, b, c, d;
3034 unsigned int s;
3035 unsigned int t;
3036
3037 a = m - ((m >> 1) & ~0UL/3);
3038 b = (a & ~0UL/5) + ((a >> 2) & ~0UL/5);
3039 c = (b + (b >> 4)) & ~0UL/0x11;
3040 d = (c + (c >> 8)) & ~0UL/0x101;
3041
3042 r++; // make r be 1..64
3043
3044 t = 0;
3045 s = LONGBITS;
3046 if (s > 32) {
Willy Tarreau9b6be3b2019-03-18 16:31:18 +01003047 unsigned long d2 = (d >> 16) >> 16;
3048 t = d2 + (d2 >> 16);
Willy Tarreauf3241112019-02-26 09:56:22 +01003049 s -= ((t - r) & 256) >> 3; r -= (t & ((t - r) >> 8));
3050 }
3051
3052 t = (d >> (s - 16)) & 0xff;
3053 s -= ((t - r) & 256) >> 4; r -= (t & ((t - r) >> 8));
3054 t = (c >> (s - 8)) & 0xf;
3055 s -= ((t - r) & 256) >> 5; r -= (t & ((t - r) >> 8));
3056 t = (b >> (s - 4)) & 0x7;
3057 s -= ((t - r) & 256) >> 6; r -= (t & ((t - r) >> 8));
3058 t = (a >> (s - 2)) & 0x3;
3059 s -= ((t - r) & 256) >> 7; r -= (t & ((t - r) >> 8));
3060 t = (m >> (s - 1)) & 0x1;
3061 s -= ((t - r) & 256) >> 8;
3062
3063 return s - 1;
3064}
3065
3066/* Same as mask_find_rank_bit() above but makes use of pre-computed bitmaps
3067 * based on <m>, in <a..d>. These ones must be updated whenever <m> changes
3068 * using mask_prep_rank_map() below.
3069 */
3070unsigned int mask_find_rank_bit_fast(unsigned int r, unsigned long m,
3071 unsigned long a, unsigned long b,
3072 unsigned long c, unsigned long d)
3073{
3074 unsigned int s;
3075 unsigned int t;
3076
3077 r++; // make r be 1..64
3078
3079 t = 0;
3080 s = LONGBITS;
3081 if (s > 32) {
Willy Tarreau9b6be3b2019-03-18 16:31:18 +01003082 unsigned long d2 = (d >> 16) >> 16;
3083 t = d2 + (d2 >> 16);
Willy Tarreauf3241112019-02-26 09:56:22 +01003084 s -= ((t - r) & 256) >> 3; r -= (t & ((t - r) >> 8));
3085 }
3086
3087 t = (d >> (s - 16)) & 0xff;
3088 s -= ((t - r) & 256) >> 4; r -= (t & ((t - r) >> 8));
3089 t = (c >> (s - 8)) & 0xf;
3090 s -= ((t - r) & 256) >> 5; r -= (t & ((t - r) >> 8));
3091 t = (b >> (s - 4)) & 0x7;
3092 s -= ((t - r) & 256) >> 6; r -= (t & ((t - r) >> 8));
3093 t = (a >> (s - 2)) & 0x3;
3094 s -= ((t - r) & 256) >> 7; r -= (t & ((t - r) >> 8));
3095 t = (m >> (s - 1)) & 0x1;
3096 s -= ((t - r) & 256) >> 8;
3097
3098 return s - 1;
3099}
3100
3101/* Prepare the bitmaps used by the fast implementation of the find_rank_bit()
3102 * above.
3103 */
3104void mask_prep_rank_map(unsigned long m,
3105 unsigned long *a, unsigned long *b,
3106 unsigned long *c, unsigned long *d)
3107{
3108 *a = m - ((m >> 1) & ~0UL/3);
3109 *b = (*a & ~0UL/5) + ((*a >> 2) & ~0UL/5);
3110 *c = (*b + (*b >> 4)) & ~0UL/0x11;
3111 *d = (*c + (*c >> 8)) & ~0UL/0x101;
3112}
3113
David du Colombier4f92d322011-03-24 11:09:31 +01003114/* Return non-zero if IPv4 address is part of the network,
Willy Tarreaueec1d382016-07-13 11:59:39 +02003115 * otherwise zero. Note that <addr> may not necessarily be aligned
3116 * while the two other ones must.
David du Colombier4f92d322011-03-24 11:09:31 +01003117 */
Willy Tarreaueec1d382016-07-13 11:59:39 +02003118int in_net_ipv4(const void *addr, const struct in_addr *mask, const struct in_addr *net)
David du Colombier4f92d322011-03-24 11:09:31 +01003119{
Willy Tarreaueec1d382016-07-13 11:59:39 +02003120 struct in_addr addr_copy;
3121
3122 memcpy(&addr_copy, addr, sizeof(addr_copy));
3123 return((addr_copy.s_addr & mask->s_addr) == (net->s_addr & mask->s_addr));
David du Colombier4f92d322011-03-24 11:09:31 +01003124}
3125
3126/* Return non-zero if IPv6 address is part of the network,
Willy Tarreaueec1d382016-07-13 11:59:39 +02003127 * otherwise zero. Note that <addr> may not necessarily be aligned
3128 * while the two other ones must.
David du Colombier4f92d322011-03-24 11:09:31 +01003129 */
Willy Tarreaueec1d382016-07-13 11:59:39 +02003130int in_net_ipv6(const void *addr, const struct in6_addr *mask, const struct in6_addr *net)
David du Colombier4f92d322011-03-24 11:09:31 +01003131{
3132 int i;
Willy Tarreaueec1d382016-07-13 11:59:39 +02003133 struct in6_addr addr_copy;
David du Colombier4f92d322011-03-24 11:09:31 +01003134
Willy Tarreaueec1d382016-07-13 11:59:39 +02003135 memcpy(&addr_copy, addr, sizeof(addr_copy));
David du Colombier4f92d322011-03-24 11:09:31 +01003136 for (i = 0; i < sizeof(struct in6_addr) / sizeof(int); i++)
Willy Tarreaueec1d382016-07-13 11:59:39 +02003137 if (((((int *)&addr_copy)[i] & ((int *)mask)[i])) !=
David du Colombier4f92d322011-03-24 11:09:31 +01003138 (((int *)net)[i] & ((int *)mask)[i]))
3139 return 0;
3140 return 1;
3141}
3142
3143/* RFC 4291 prefix */
3144const char rfc4291_pfx[] = { 0x00, 0x00, 0x00, 0x00,
3145 0x00, 0x00, 0x00, 0x00,
3146 0x00, 0x00, 0xFF, 0xFF };
3147
Joseph Herlant32b83272018-11-15 11:58:28 -08003148/* Map IPv4 address on IPv6 address, as specified in RFC 3513.
Thierry FOURNIER4a04dc32013-11-28 16:33:15 +01003149 * Input and output may overlap.
3150 */
David du Colombier4f92d322011-03-24 11:09:31 +01003151void v4tov6(struct in6_addr *sin6_addr, struct in_addr *sin_addr)
3152{
Thierry FOURNIER4a04dc32013-11-28 16:33:15 +01003153 struct in_addr tmp_addr;
3154
3155 tmp_addr.s_addr = sin_addr->s_addr;
David du Colombier4f92d322011-03-24 11:09:31 +01003156 memcpy(sin6_addr->s6_addr, rfc4291_pfx, sizeof(rfc4291_pfx));
Thierry FOURNIER4a04dc32013-11-28 16:33:15 +01003157 memcpy(sin6_addr->s6_addr+12, &tmp_addr.s_addr, 4);
David du Colombier4f92d322011-03-24 11:09:31 +01003158}
3159
Joseph Herlant32b83272018-11-15 11:58:28 -08003160/* Map IPv6 address on IPv4 address, as specified in RFC 3513.
David du Colombier4f92d322011-03-24 11:09:31 +01003161 * Return true if conversion is possible and false otherwise.
3162 */
3163int v6tov4(struct in_addr *sin_addr, struct in6_addr *sin6_addr)
3164{
3165 if (memcmp(sin6_addr->s6_addr, rfc4291_pfx, sizeof(rfc4291_pfx)) == 0) {
3166 memcpy(&(sin_addr->s_addr), &(sin6_addr->s6_addr[12]),
3167 sizeof(struct in_addr));
3168 return 1;
3169 }
3170
3171 return 0;
3172}
3173
Baptiste Assmann08b24cf2016-01-23 23:39:12 +01003174/* compare two struct sockaddr_storage and return:
3175 * 0 (true) if the addr is the same in both
3176 * 1 (false) if the addr is not the same in both
3177 * -1 (unable) if one of the addr is not AF_INET*
3178 */
3179int ipcmp(struct sockaddr_storage *ss1, struct sockaddr_storage *ss2)
3180{
3181 if ((ss1->ss_family != AF_INET) && (ss1->ss_family != AF_INET6))
3182 return -1;
3183
3184 if ((ss2->ss_family != AF_INET) && (ss2->ss_family != AF_INET6))
3185 return -1;
3186
3187 if (ss1->ss_family != ss2->ss_family)
3188 return 1;
3189
3190 switch (ss1->ss_family) {
3191 case AF_INET:
3192 return memcmp(&((struct sockaddr_in *)ss1)->sin_addr,
3193 &((struct sockaddr_in *)ss2)->sin_addr,
3194 sizeof(struct in_addr)) != 0;
3195 case AF_INET6:
3196 return memcmp(&((struct sockaddr_in6 *)ss1)->sin6_addr,
3197 &((struct sockaddr_in6 *)ss2)->sin6_addr,
3198 sizeof(struct in6_addr)) != 0;
3199 }
3200
3201 return 1;
3202}
3203
Christopher Faulet9553de72021-02-26 09:12:50 +01003204/* compare a struct sockaddr_storage to a struct net_addr and return :
3205 * 0 (true) if <addr> is matching <net>
3206 * 1 (false) if <addr> is not matching <net>
3207 * -1 (unable) if <addr> or <net> is not AF_INET*
3208 */
3209int ipcmp2net(const struct sockaddr_storage *addr, const struct net_addr *net)
3210{
3211 if ((addr->ss_family != AF_INET) && (addr->ss_family != AF_INET6))
3212 return -1;
3213
3214 if ((net->family != AF_INET) && (net->family != AF_INET6))
3215 return -1;
3216
3217 if (addr->ss_family != net->family)
3218 return 1;
3219
3220 if (addr->ss_family == AF_INET &&
3221 (((struct sockaddr_in *)addr)->sin_addr.s_addr & net->addr.v4.mask.s_addr) == net->addr.v4.ip.s_addr)
3222 return 0;
3223 else {
3224 const struct in6_addr *addr6 = &(((const struct sockaddr_in6*)addr)->sin6_addr);
3225 const struct in6_addr *nip6 = &net->addr.v6.ip;
3226 const struct in6_addr *nmask6 = &net->addr.v6.mask;
3227
3228 if ((read_u32(&addr6->s6_addr[0]) & read_u32(&nmask6->s6_addr[0])) == read_u32(&nip6->s6_addr[0]) &&
3229 (read_u32(&addr6->s6_addr[4]) & read_u32(&nmask6->s6_addr[4])) == read_u32(&nip6->s6_addr[4]) &&
3230 (read_u32(&addr6->s6_addr[8]) & read_u32(&nmask6->s6_addr[8])) == read_u32(&nip6->s6_addr[8]) &&
3231 (read_u32(&addr6->s6_addr[12]) & read_u32(&nmask6->s6_addr[12])) == read_u32(&nip6->s6_addr[12]))
3232 return 0;
3233 }
3234
3235 return 1;
3236}
3237
Baptiste Assmann08396c82016-01-31 00:27:17 +01003238/* copy IP address from <source> into <dest>
Willy Tarreaudc3a9e82016-11-04 18:47:01 +01003239 * The caller must allocate and clear <dest> before calling.
3240 * The source must be in either AF_INET or AF_INET6 family, or the destination
3241 * address will be undefined. If the destination address used to hold a port,
3242 * it is preserved, so that this function can be used to switch to another
3243 * address family with no risk. Returns a pointer to the destination.
Baptiste Assmann08396c82016-01-31 00:27:17 +01003244 */
3245struct sockaddr_storage *ipcpy(struct sockaddr_storage *source, struct sockaddr_storage *dest)
3246{
Willy Tarreaudc3a9e82016-11-04 18:47:01 +01003247 int prev_port;
3248
3249 prev_port = get_net_port(dest);
3250 memset(dest, 0, sizeof(*dest));
Baptiste Assmann08396c82016-01-31 00:27:17 +01003251 dest->ss_family = source->ss_family;
3252
3253 /* copy new addr and apply it */
3254 switch (source->ss_family) {
3255 case AF_INET:
3256 ((struct sockaddr_in *)dest)->sin_addr.s_addr = ((struct sockaddr_in *)source)->sin_addr.s_addr;
Willy Tarreaudc3a9e82016-11-04 18:47:01 +01003257 ((struct sockaddr_in *)dest)->sin_port = prev_port;
Baptiste Assmann08396c82016-01-31 00:27:17 +01003258 break;
3259 case AF_INET6:
3260 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 +01003261 ((struct sockaddr_in6 *)dest)->sin6_port = prev_port;
Baptiste Assmann08396c82016-01-31 00:27:17 +01003262 break;
3263 }
3264
3265 return dest;
3266}
3267
William Lallemand421f5b52012-02-06 18:15:57 +01003268char *human_time(int t, short hz_div) {
3269 static char rv[sizeof("24855d23h")+1]; // longest of "23h59m" and "59m59s"
3270 char *p = rv;
Willy Tarreau761b3d52014-04-14 14:53:06 +02003271 char *end = rv + sizeof(rv);
William Lallemand421f5b52012-02-06 18:15:57 +01003272 int cnt=2; // print two numbers
3273
3274 if (unlikely(t < 0 || hz_div <= 0)) {
Willy Tarreau761b3d52014-04-14 14:53:06 +02003275 snprintf(p, end - p, "?");
William Lallemand421f5b52012-02-06 18:15:57 +01003276 return rv;
3277 }
3278
3279 if (unlikely(hz_div > 1))
3280 t /= hz_div;
3281
3282 if (t >= DAY) {
Willy Tarreau761b3d52014-04-14 14:53:06 +02003283 p += snprintf(p, end - p, "%dd", t / DAY);
William Lallemand421f5b52012-02-06 18:15:57 +01003284 cnt--;
3285 }
3286
3287 if (cnt && t % DAY / HOUR) {
Willy Tarreau761b3d52014-04-14 14:53:06 +02003288 p += snprintf(p, end - p, "%dh", t % DAY / HOUR);
William Lallemand421f5b52012-02-06 18:15:57 +01003289 cnt--;
3290 }
3291
3292 if (cnt && t % HOUR / MINUTE) {
Willy Tarreau761b3d52014-04-14 14:53:06 +02003293 p += snprintf(p, end - p, "%dm", t % HOUR / MINUTE);
William Lallemand421f5b52012-02-06 18:15:57 +01003294 cnt--;
3295 }
3296
3297 if ((cnt && t % MINUTE) || !t) // also display '0s'
Willy Tarreau761b3d52014-04-14 14:53:06 +02003298 p += snprintf(p, end - p, "%ds", t % MINUTE / SEC);
William Lallemand421f5b52012-02-06 18:15:57 +01003299
3300 return rv;
3301}
3302
3303const char *monthname[12] = {
3304 "Jan", "Feb", "Mar", "Apr", "May", "Jun",
3305 "Jul", "Aug", "Sep", "Oct", "Nov", "Dec"
3306};
3307
3308/* date2str_log: write a date in the format :
3309 * sprintf(str, "%02d/%s/%04d:%02d:%02d:%02d.%03d",
3310 * tm.tm_mday, monthname[tm.tm_mon], tm.tm_year+1900,
3311 * tm.tm_hour, tm.tm_min, tm.tm_sec, (int)date.tv_usec/1000);
3312 *
3313 * without using sprintf. return a pointer to the last char written (\0) or
3314 * NULL if there isn't enough space.
3315 */
Willy Tarreauf16cb412018-09-04 19:08:48 +02003316char *date2str_log(char *dst, const struct tm *tm, const struct timeval *date, size_t size)
William Lallemand421f5b52012-02-06 18:15:57 +01003317{
3318
3319 if (size < 25) /* the size is fixed: 24 chars + \0 */
3320 return NULL;
3321
3322 dst = utoa_pad((unsigned int)tm->tm_mday, dst, 3); // day
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003323 if (!dst)
3324 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003325 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003326
William Lallemand421f5b52012-02-06 18:15:57 +01003327 memcpy(dst, monthname[tm->tm_mon], 3); // month
3328 dst += 3;
3329 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003330
William Lallemand421f5b52012-02-06 18:15:57 +01003331 dst = utoa_pad((unsigned int)tm->tm_year+1900, dst, 5); // year
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003332 if (!dst)
3333 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003334 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003335
William Lallemand421f5b52012-02-06 18:15:57 +01003336 dst = utoa_pad((unsigned int)tm->tm_hour, dst, 3); // hour
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003337 if (!dst)
3338 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003339 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003340
William Lallemand421f5b52012-02-06 18:15:57 +01003341 dst = utoa_pad((unsigned int)tm->tm_min, dst, 3); // minutes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003342 if (!dst)
3343 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003344 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003345
William Lallemand421f5b52012-02-06 18:15:57 +01003346 dst = utoa_pad((unsigned int)tm->tm_sec, dst, 3); // secondes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003347 if (!dst)
3348 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003349 *dst++ = '.';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003350
Willy Tarreau7d9421d2020-02-29 09:08:02 +01003351 dst = utoa_pad((unsigned int)(date->tv_usec/1000)%1000, dst, 4); // milliseconds
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003352 if (!dst)
3353 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003354 *dst = '\0';
3355
3356 return dst;
3357}
3358
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003359/* Base year used to compute leap years */
3360#define TM_YEAR_BASE 1900
3361
3362/* Return the difference in seconds between two times (leap seconds are ignored).
3363 * Retrieved from glibc 2.18 source code.
3364 */
3365static int my_tm_diff(const struct tm *a, const struct tm *b)
3366{
3367 /* Compute intervening leap days correctly even if year is negative.
3368 * Take care to avoid int overflow in leap day calculations,
3369 * but it's OK to assume that A and B are close to each other.
3370 */
3371 int a4 = (a->tm_year >> 2) + (TM_YEAR_BASE >> 2) - ! (a->tm_year & 3);
3372 int b4 = (b->tm_year >> 2) + (TM_YEAR_BASE >> 2) - ! (b->tm_year & 3);
3373 int a100 = a4 / 25 - (a4 % 25 < 0);
3374 int b100 = b4 / 25 - (b4 % 25 < 0);
3375 int a400 = a100 >> 2;
3376 int b400 = b100 >> 2;
3377 int intervening_leap_days = (a4 - b4) - (a100 - b100) + (a400 - b400);
3378 int years = a->tm_year - b->tm_year;
3379 int days = (365 * years + intervening_leap_days
3380 + (a->tm_yday - b->tm_yday));
3381 return (60 * (60 * (24 * days + (a->tm_hour - b->tm_hour))
3382 + (a->tm_min - b->tm_min))
3383 + (a->tm_sec - b->tm_sec));
3384}
3385
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003386/* Return the GMT offset for a specific local time.
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003387 * Both t and tm must represent the same time.
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003388 * The string returned has the same format as returned by strftime(... "%z", tm).
3389 * Offsets are kept in an internal cache for better performances.
3390 */
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003391const char *get_gmt_offset(time_t t, struct tm *tm)
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003392{
3393 /* Cache offsets from GMT (depending on whether DST is active or not) */
Christopher Faulet1bc04c72017-10-29 20:14:08 +01003394 static THREAD_LOCAL char gmt_offsets[2][5+1] = { "", "" };
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003395
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003396 char *gmt_offset;
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003397 struct tm tm_gmt;
3398 int diff;
3399 int isdst = tm->tm_isdst;
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003400
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003401 /* Pretend DST not active if its status is unknown */
3402 if (isdst < 0)
3403 isdst = 0;
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003404
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003405 /* Fetch the offset and initialize it if needed */
3406 gmt_offset = gmt_offsets[isdst & 0x01];
3407 if (unlikely(!*gmt_offset)) {
3408 get_gmtime(t, &tm_gmt);
3409 diff = my_tm_diff(tm, &tm_gmt);
3410 if (diff < 0) {
3411 diff = -diff;
3412 *gmt_offset = '-';
3413 } else {
3414 *gmt_offset = '+';
3415 }
Willy Tarreaue112c8a2019-10-29 10:16:11 +01003416 diff %= 86400U;
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003417 diff /= 60; /* Convert to minutes */
3418 snprintf(gmt_offset+1, 4+1, "%02d%02d", diff/60, diff%60);
3419 }
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003420
Willy Tarreaue112c8a2019-10-29 10:16:11 +01003421 return gmt_offset;
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003422}
3423
William Lallemand421f5b52012-02-06 18:15:57 +01003424/* gmt2str_log: write a date in the format :
3425 * "%02d/%s/%04d:%02d:%02d:%02d +0000" without using snprintf
3426 * return a pointer to the last char written (\0) or
3427 * NULL if there isn't enough space.
3428 */
3429char *gmt2str_log(char *dst, struct tm *tm, size_t size)
3430{
Yuxans Yao4e25b012012-10-19 10:36:09 +08003431 if (size < 27) /* the size is fixed: 26 chars + \0 */
William Lallemand421f5b52012-02-06 18:15:57 +01003432 return NULL;
3433
3434 dst = utoa_pad((unsigned int)tm->tm_mday, dst, 3); // day
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003435 if (!dst)
3436 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003437 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003438
William Lallemand421f5b52012-02-06 18:15:57 +01003439 memcpy(dst, monthname[tm->tm_mon], 3); // month
3440 dst += 3;
3441 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003442
William Lallemand421f5b52012-02-06 18:15:57 +01003443 dst = utoa_pad((unsigned int)tm->tm_year+1900, dst, 5); // year
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003444 if (!dst)
3445 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003446 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003447
William Lallemand421f5b52012-02-06 18:15:57 +01003448 dst = utoa_pad((unsigned int)tm->tm_hour, dst, 3); // hour
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003449 if (!dst)
3450 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003451 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003452
William Lallemand421f5b52012-02-06 18:15:57 +01003453 dst = utoa_pad((unsigned int)tm->tm_min, dst, 3); // minutes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003454 if (!dst)
3455 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003456 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003457
William Lallemand421f5b52012-02-06 18:15:57 +01003458 dst = utoa_pad((unsigned int)tm->tm_sec, dst, 3); // secondes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003459 if (!dst)
3460 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003461 *dst++ = ' ';
3462 *dst++ = '+';
3463 *dst++ = '0';
3464 *dst++ = '0';
3465 *dst++ = '0';
3466 *dst++ = '0';
3467 *dst = '\0';
3468
3469 return dst;
3470}
3471
Yuxans Yao4e25b012012-10-19 10:36:09 +08003472/* localdate2str_log: write a date in the format :
3473 * "%02d/%s/%04d:%02d:%02d:%02d +0000(local timezone)" without using snprintf
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003474 * Both t and tm must represent the same time.
3475 * return a pointer to the last char written (\0) or
3476 * NULL if there isn't enough space.
Yuxans Yao4e25b012012-10-19 10:36:09 +08003477 */
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003478char *localdate2str_log(char *dst, time_t t, struct tm *tm, size_t size)
Yuxans Yao4e25b012012-10-19 10:36:09 +08003479{
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003480 const char *gmt_offset;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003481 if (size < 27) /* the size is fixed: 26 chars + \0 */
3482 return NULL;
3483
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003484 gmt_offset = get_gmt_offset(t, tm);
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003485
Yuxans Yao4e25b012012-10-19 10:36:09 +08003486 dst = utoa_pad((unsigned int)tm->tm_mday, dst, 3); // day
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003487 if (!dst)
3488 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003489 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003490
Yuxans Yao4e25b012012-10-19 10:36:09 +08003491 memcpy(dst, monthname[tm->tm_mon], 3); // month
3492 dst += 3;
3493 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003494
Yuxans Yao4e25b012012-10-19 10:36:09 +08003495 dst = utoa_pad((unsigned int)tm->tm_year+1900, dst, 5); // year
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003496 if (!dst)
3497 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003498 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003499
Yuxans Yao4e25b012012-10-19 10:36:09 +08003500 dst = utoa_pad((unsigned int)tm->tm_hour, dst, 3); // hour
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003501 if (!dst)
3502 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003503 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003504
Yuxans Yao4e25b012012-10-19 10:36:09 +08003505 dst = utoa_pad((unsigned int)tm->tm_min, dst, 3); // minutes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003506 if (!dst)
3507 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003508 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003509
Yuxans Yao4e25b012012-10-19 10:36:09 +08003510 dst = utoa_pad((unsigned int)tm->tm_sec, dst, 3); // secondes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003511 if (!dst)
3512 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003513 *dst++ = ' ';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003514
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003515 memcpy(dst, gmt_offset, 5); // Offset from local time to GMT
Yuxans Yao4e25b012012-10-19 10:36:09 +08003516 dst += 5;
3517 *dst = '\0';
3518
3519 return dst;
3520}
3521
Willy Tarreaucb1949b2017-07-19 19:05:29 +02003522/* Returns the number of seconds since 01/01/1970 0:0:0 GMT for GMT date <tm>.
3523 * It is meant as a portable replacement for timegm() for use with valid inputs.
3524 * Returns undefined results for invalid dates (eg: months out of range 0..11).
3525 */
3526time_t my_timegm(const struct tm *tm)
3527{
3528 /* Each month has 28, 29, 30 or 31 days, or 28+N. The date in the year
3529 * is thus (current month - 1)*28 + cumulated_N[month] to count the
3530 * sum of the extra N days for elapsed months. The sum of all these N
3531 * days doesn't exceed 30 for a complete year (366-12*28) so it fits
3532 * in a 5-bit word. This means that with 60 bits we can represent a
3533 * matrix of all these values at once, which is fast and efficient to
3534 * access. The extra February day for leap years is not counted here.
3535 *
3536 * Jan : none = 0 (0)
3537 * Feb : Jan = 3 (3)
3538 * Mar : Jan..Feb = 3 (3 + 0)
3539 * Apr : Jan..Mar = 6 (3 + 0 + 3)
3540 * May : Jan..Apr = 8 (3 + 0 + 3 + 2)
3541 * Jun : Jan..May = 11 (3 + 0 + 3 + 2 + 3)
3542 * Jul : Jan..Jun = 13 (3 + 0 + 3 + 2 + 3 + 2)
3543 * Aug : Jan..Jul = 16 (3 + 0 + 3 + 2 + 3 + 2 + 3)
3544 * Sep : Jan..Aug = 19 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3)
3545 * Oct : Jan..Sep = 21 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3 + 2)
3546 * Nov : Jan..Oct = 24 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3 + 2 + 3)
3547 * Dec : Jan..Nov = 26 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3 + 2 + 3 + 2)
3548 */
3549 uint64_t extra =
3550 ( 0ULL << 0*5) + ( 3ULL << 1*5) + ( 3ULL << 2*5) + /* Jan, Feb, Mar, */
3551 ( 6ULL << 3*5) + ( 8ULL << 4*5) + (11ULL << 5*5) + /* Apr, May, Jun, */
3552 (13ULL << 6*5) + (16ULL << 7*5) + (19ULL << 8*5) + /* Jul, Aug, Sep, */
3553 (21ULL << 9*5) + (24ULL << 10*5) + (26ULL << 11*5); /* Oct, Nov, Dec, */
3554
3555 unsigned int y = tm->tm_year + 1900;
3556 unsigned int m = tm->tm_mon;
3557 unsigned long days = 0;
3558
3559 /* days since 1/1/1970 for full years */
3560 days += days_since_zero(y) - days_since_zero(1970);
3561
3562 /* days for full months in the current year */
3563 days += 28 * m + ((extra >> (m * 5)) & 0x1f);
3564
3565 /* count + 1 after March for leap years. A leap year is a year multiple
3566 * of 4, unless it's multiple of 100 without being multiple of 400. 2000
3567 * is leap, 1900 isn't, 1904 is.
3568 */
3569 if ((m > 1) && !(y & 3) && ((y % 100) || !(y % 400)))
3570 days++;
3571
3572 days += tm->tm_mday - 1;
3573 return days * 86400ULL + tm->tm_hour * 3600 + tm->tm_min * 60 + tm->tm_sec;
3574}
3575
Thierry Fournier93127942016-01-20 18:49:45 +01003576/* This function check a char. It returns true and updates
3577 * <date> and <len> pointer to the new position if the
3578 * character is found.
3579 */
3580static inline int parse_expect_char(const char **date, int *len, char c)
3581{
3582 if (*len < 1 || **date != c)
3583 return 0;
3584 (*len)--;
3585 (*date)++;
3586 return 1;
3587}
3588
3589/* This function expects a string <str> of len <l>. It return true and updates.
3590 * <date> and <len> if the string matches, otherwise, it returns false.
3591 */
3592static inline int parse_strcmp(const char **date, int *len, char *str, int l)
3593{
3594 if (*len < l || strncmp(*date, str, l) != 0)
3595 return 0;
3596 (*len) -= l;
3597 (*date) += l;
3598 return 1;
3599}
3600
3601/* This macro converts 3 chars name in integer. */
3602#define STR2I3(__a, __b, __c) ((__a) * 65536 + (__b) * 256 + (__c))
3603
3604/* day-name = %x4D.6F.6E ; "Mon", case-sensitive
3605 * / %x54.75.65 ; "Tue", case-sensitive
3606 * / %x57.65.64 ; "Wed", case-sensitive
3607 * / %x54.68.75 ; "Thu", case-sensitive
3608 * / %x46.72.69 ; "Fri", case-sensitive
3609 * / %x53.61.74 ; "Sat", case-sensitive
3610 * / %x53.75.6E ; "Sun", case-sensitive
3611 *
3612 * This array must be alphabetically sorted
3613 */
3614static inline int parse_http_dayname(const char **date, int *len, struct tm *tm)
3615{
3616 if (*len < 3)
3617 return 0;
3618 switch (STR2I3((*date)[0], (*date)[1], (*date)[2])) {
3619 case STR2I3('M','o','n'): tm->tm_wday = 1; break;
3620 case STR2I3('T','u','e'): tm->tm_wday = 2; break;
3621 case STR2I3('W','e','d'): tm->tm_wday = 3; break;
3622 case STR2I3('T','h','u'): tm->tm_wday = 4; break;
3623 case STR2I3('F','r','i'): tm->tm_wday = 5; break;
3624 case STR2I3('S','a','t'): tm->tm_wday = 6; break;
3625 case STR2I3('S','u','n'): tm->tm_wday = 7; break;
3626 default: return 0;
3627 }
3628 *len -= 3;
3629 *date += 3;
3630 return 1;
3631}
3632
3633/* month = %x4A.61.6E ; "Jan", case-sensitive
3634 * / %x46.65.62 ; "Feb", case-sensitive
3635 * / %x4D.61.72 ; "Mar", case-sensitive
3636 * / %x41.70.72 ; "Apr", case-sensitive
3637 * / %x4D.61.79 ; "May", case-sensitive
3638 * / %x4A.75.6E ; "Jun", case-sensitive
3639 * / %x4A.75.6C ; "Jul", case-sensitive
3640 * / %x41.75.67 ; "Aug", case-sensitive
3641 * / %x53.65.70 ; "Sep", case-sensitive
3642 * / %x4F.63.74 ; "Oct", case-sensitive
3643 * / %x4E.6F.76 ; "Nov", case-sensitive
3644 * / %x44.65.63 ; "Dec", case-sensitive
3645 *
3646 * This array must be alphabetically sorted
3647 */
3648static inline int parse_http_monthname(const char **date, int *len, struct tm *tm)
3649{
3650 if (*len < 3)
3651 return 0;
3652 switch (STR2I3((*date)[0], (*date)[1], (*date)[2])) {
3653 case STR2I3('J','a','n'): tm->tm_mon = 0; break;
3654 case STR2I3('F','e','b'): tm->tm_mon = 1; break;
3655 case STR2I3('M','a','r'): tm->tm_mon = 2; break;
3656 case STR2I3('A','p','r'): tm->tm_mon = 3; break;
3657 case STR2I3('M','a','y'): tm->tm_mon = 4; break;
3658 case STR2I3('J','u','n'): tm->tm_mon = 5; break;
3659 case STR2I3('J','u','l'): tm->tm_mon = 6; break;
3660 case STR2I3('A','u','g'): tm->tm_mon = 7; break;
3661 case STR2I3('S','e','p'): tm->tm_mon = 8; break;
3662 case STR2I3('O','c','t'): tm->tm_mon = 9; break;
3663 case STR2I3('N','o','v'): tm->tm_mon = 10; break;
3664 case STR2I3('D','e','c'): tm->tm_mon = 11; break;
3665 default: return 0;
3666 }
3667 *len -= 3;
3668 *date += 3;
3669 return 1;
3670}
3671
3672/* day-name-l = %x4D.6F.6E.64.61.79 ; "Monday", case-sensitive
3673 * / %x54.75.65.73.64.61.79 ; "Tuesday", case-sensitive
3674 * / %x57.65.64.6E.65.73.64.61.79 ; "Wednesday", case-sensitive
3675 * / %x54.68.75.72.73.64.61.79 ; "Thursday", case-sensitive
3676 * / %x46.72.69.64.61.79 ; "Friday", case-sensitive
3677 * / %x53.61.74.75.72.64.61.79 ; "Saturday", case-sensitive
3678 * / %x53.75.6E.64.61.79 ; "Sunday", case-sensitive
3679 *
3680 * This array must be alphabetically sorted
3681 */
3682static inline int parse_http_ldayname(const char **date, int *len, struct tm *tm)
3683{
3684 if (*len < 6) /* Minimum length. */
3685 return 0;
3686 switch (STR2I3((*date)[0], (*date)[1], (*date)[2])) {
3687 case STR2I3('M','o','n'):
3688 RET0_UNLESS(parse_strcmp(date, len, "Monday", 6));
3689 tm->tm_wday = 1;
3690 return 1;
3691 case STR2I3('T','u','e'):
3692 RET0_UNLESS(parse_strcmp(date, len, "Tuesday", 7));
3693 tm->tm_wday = 2;
3694 return 1;
3695 case STR2I3('W','e','d'):
3696 RET0_UNLESS(parse_strcmp(date, len, "Wednesday", 9));
3697 tm->tm_wday = 3;
3698 return 1;
3699 case STR2I3('T','h','u'):
3700 RET0_UNLESS(parse_strcmp(date, len, "Thursday", 8));
3701 tm->tm_wday = 4;
3702 return 1;
3703 case STR2I3('F','r','i'):
3704 RET0_UNLESS(parse_strcmp(date, len, "Friday", 6));
3705 tm->tm_wday = 5;
3706 return 1;
3707 case STR2I3('S','a','t'):
3708 RET0_UNLESS(parse_strcmp(date, len, "Saturday", 8));
3709 tm->tm_wday = 6;
3710 return 1;
3711 case STR2I3('S','u','n'):
3712 RET0_UNLESS(parse_strcmp(date, len, "Sunday", 6));
3713 tm->tm_wday = 7;
3714 return 1;
3715 }
3716 return 0;
3717}
3718
3719/* This function parses exactly 1 digit and returns the numeric value in "digit". */
3720static inline int parse_digit(const char **date, int *len, int *digit)
3721{
3722 if (*len < 1 || **date < '0' || **date > '9')
3723 return 0;
3724 *digit = (**date - '0');
3725 (*date)++;
3726 (*len)--;
3727 return 1;
3728}
3729
3730/* This function parses exactly 2 digits and returns the numeric value in "digit". */
3731static inline int parse_2digit(const char **date, int *len, int *digit)
3732{
3733 int value;
3734
3735 RET0_UNLESS(parse_digit(date, len, &value));
3736 (*digit) = value * 10;
3737 RET0_UNLESS(parse_digit(date, len, &value));
3738 (*digit) += value;
3739
3740 return 1;
3741}
3742
3743/* This function parses exactly 4 digits and returns the numeric value in "digit". */
3744static inline int parse_4digit(const char **date, int *len, int *digit)
3745{
3746 int value;
3747
3748 RET0_UNLESS(parse_digit(date, len, &value));
3749 (*digit) = value * 1000;
3750
3751 RET0_UNLESS(parse_digit(date, len, &value));
3752 (*digit) += value * 100;
3753
3754 RET0_UNLESS(parse_digit(date, len, &value));
3755 (*digit) += value * 10;
3756
3757 RET0_UNLESS(parse_digit(date, len, &value));
3758 (*digit) += value;
3759
3760 return 1;
3761}
3762
3763/* time-of-day = hour ":" minute ":" second
3764 * ; 00:00:00 - 23:59:60 (leap second)
3765 *
3766 * hour = 2DIGIT
3767 * minute = 2DIGIT
3768 * second = 2DIGIT
3769 */
3770static inline int parse_http_time(const char **date, int *len, struct tm *tm)
3771{
3772 RET0_UNLESS(parse_2digit(date, len, &tm->tm_hour)); /* hour 2DIGIT */
3773 RET0_UNLESS(parse_expect_char(date, len, ':')); /* expect ":" */
3774 RET0_UNLESS(parse_2digit(date, len, &tm->tm_min)); /* min 2DIGIT */
3775 RET0_UNLESS(parse_expect_char(date, len, ':')); /* expect ":" */
3776 RET0_UNLESS(parse_2digit(date, len, &tm->tm_sec)); /* sec 2DIGIT */
3777 return 1;
3778}
3779
3780/* From RFC7231
3781 * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
3782 *
3783 * IMF-fixdate = day-name "," SP date1 SP time-of-day SP GMT
3784 * ; fixed length/zone/capitalization subset of the format
3785 * ; see Section 3.3 of [RFC5322]
3786 *
3787 *
3788 * date1 = day SP month SP year
3789 * ; e.g., 02 Jun 1982
3790 *
3791 * day = 2DIGIT
3792 * year = 4DIGIT
3793 *
3794 * GMT = %x47.4D.54 ; "GMT", case-sensitive
3795 *
3796 * time-of-day = hour ":" minute ":" second
3797 * ; 00:00:00 - 23:59:60 (leap second)
3798 *
3799 * hour = 2DIGIT
3800 * minute = 2DIGIT
3801 * second = 2DIGIT
3802 *
3803 * DIGIT = decimal 0-9
3804 */
3805int parse_imf_date(const char *date, int len, struct tm *tm)
3806{
David Carlier327298c2016-11-20 10:42:38 +00003807 /* tm_gmtoff, if present, ought to be zero'ed */
3808 memset(tm, 0, sizeof(*tm));
3809
Thierry Fournier93127942016-01-20 18:49:45 +01003810 RET0_UNLESS(parse_http_dayname(&date, &len, tm)); /* day-name */
3811 RET0_UNLESS(parse_expect_char(&date, &len, ',')); /* expect "," */
3812 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3813 RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_mday)); /* day 2DIGIT */
3814 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3815 RET0_UNLESS(parse_http_monthname(&date, &len, tm)); /* Month */
3816 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3817 RET0_UNLESS(parse_4digit(&date, &len, &tm->tm_year)); /* year = 4DIGIT */
3818 tm->tm_year -= 1900;
3819 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3820 RET0_UNLESS(parse_http_time(&date, &len, tm)); /* Parse time. */
3821 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3822 RET0_UNLESS(parse_strcmp(&date, &len, "GMT", 3)); /* GMT = %x47.4D.54 ; "GMT", case-sensitive */
3823 tm->tm_isdst = -1;
Thierry Fournier93127942016-01-20 18:49:45 +01003824 return 1;
3825}
3826
3827/* From RFC7231
3828 * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
3829 *
3830 * rfc850-date = day-name-l "," SP date2 SP time-of-day SP GMT
3831 * date2 = day "-" month "-" 2DIGIT
3832 * ; e.g., 02-Jun-82
3833 *
3834 * day = 2DIGIT
3835 */
3836int parse_rfc850_date(const char *date, int len, struct tm *tm)
3837{
3838 int year;
3839
David Carlier327298c2016-11-20 10:42:38 +00003840 /* tm_gmtoff, if present, ought to be zero'ed */
3841 memset(tm, 0, sizeof(*tm));
3842
Thierry Fournier93127942016-01-20 18:49:45 +01003843 RET0_UNLESS(parse_http_ldayname(&date, &len, tm)); /* Read the day name */
3844 RET0_UNLESS(parse_expect_char(&date, &len, ',')); /* expect "," */
3845 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3846 RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_mday)); /* day 2DIGIT */
3847 RET0_UNLESS(parse_expect_char(&date, &len, '-')); /* expect "-" */
3848 RET0_UNLESS(parse_http_monthname(&date, &len, tm)); /* Month */
3849 RET0_UNLESS(parse_expect_char(&date, &len, '-')); /* expect "-" */
3850
3851 /* year = 2DIGIT
3852 *
3853 * Recipients of a timestamp value in rfc850-(*date) format, which uses a
3854 * two-digit year, MUST interpret a timestamp that appears to be more
3855 * than 50 years in the future as representing the most recent year in
3856 * the past that had the same last two digits.
3857 */
3858 RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_year));
3859
3860 /* expect SP */
3861 if (!parse_expect_char(&date, &len, ' ')) {
3862 /* Maybe we have the date with 4 digits. */
3863 RET0_UNLESS(parse_2digit(&date, &len, &year));
3864 tm->tm_year = (tm->tm_year * 100 + year) - 1900;
3865 /* expect SP */
3866 RET0_UNLESS(parse_expect_char(&date, &len, ' '));
3867 } else {
3868 /* I fix 60 as pivot: >60: +1900, <60: +2000. Note that the
3869 * tm_year is the number of year since 1900, so for +1900, we
3870 * do nothing, and for +2000, we add 100.
3871 */
3872 if (tm->tm_year <= 60)
3873 tm->tm_year += 100;
3874 }
3875
3876 RET0_UNLESS(parse_http_time(&date, &len, tm)); /* Parse time. */
3877 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3878 RET0_UNLESS(parse_strcmp(&date, &len, "GMT", 3)); /* GMT = %x47.4D.54 ; "GMT", case-sensitive */
3879 tm->tm_isdst = -1;
Thierry Fournier93127942016-01-20 18:49:45 +01003880
3881 return 1;
3882}
3883
3884/* From RFC7231
3885 * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
3886 *
3887 * asctime-date = day-name SP date3 SP time-of-day SP year
3888 * date3 = month SP ( 2DIGIT / ( SP 1DIGIT ))
3889 * ; e.g., Jun 2
3890 *
3891 * HTTP-date is case sensitive. A sender MUST NOT generate additional
3892 * whitespace in an HTTP-date beyond that specifically included as SP in
3893 * the grammar.
3894 */
3895int parse_asctime_date(const char *date, int len, struct tm *tm)
3896{
David Carlier327298c2016-11-20 10:42:38 +00003897 /* tm_gmtoff, if present, ought to be zero'ed */
3898 memset(tm, 0, sizeof(*tm));
3899
Thierry Fournier93127942016-01-20 18:49:45 +01003900 RET0_UNLESS(parse_http_dayname(&date, &len, tm)); /* day-name */
3901 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3902 RET0_UNLESS(parse_http_monthname(&date, &len, tm)); /* expect month */
3903 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3904
3905 /* expect SP and 1DIGIT or 2DIGIT */
3906 if (parse_expect_char(&date, &len, ' '))
3907 RET0_UNLESS(parse_digit(&date, &len, &tm->tm_mday));
3908 else
3909 RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_mday));
3910
3911 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3912 RET0_UNLESS(parse_http_time(&date, &len, tm)); /* Parse time. */
3913 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3914 RET0_UNLESS(parse_4digit(&date, &len, &tm->tm_year)); /* year = 4DIGIT */
3915 tm->tm_year -= 1900;
3916 tm->tm_isdst = -1;
Thierry Fournier93127942016-01-20 18:49:45 +01003917 return 1;
3918}
3919
3920/* From RFC7231
3921 * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
3922 *
3923 * HTTP-date = IMF-fixdate / obs-date
3924 * obs-date = rfc850-date / asctime-date
3925 *
3926 * parses an HTTP date in the RFC format and is accepted
3927 * alternatives. <date> is the strinf containing the date,
3928 * len is the len of the string. <tm> is filled with the
3929 * parsed time. We must considers this time as GMT.
3930 */
3931int parse_http_date(const char *date, int len, struct tm *tm)
3932{
3933 if (parse_imf_date(date, len, tm))
3934 return 1;
3935
3936 if (parse_rfc850_date(date, len, tm))
3937 return 1;
3938
3939 if (parse_asctime_date(date, len, tm))
3940 return 1;
3941
3942 return 0;
3943}
3944
Willy Tarreau4deeb102021-01-29 10:47:52 +01003945/* print the time <ns> in a short form (exactly 7 chars) at the end of buffer
3946 * <out>. "-" is printed if the value is zero, "inf" if larger than 1000 years.
3947 * It returns the new buffer length, or 0 if it doesn't fit. The value will be
3948 * surrounded by <pfx> and <sfx> respectively if not NULL.
3949 */
3950int print_time_short(struct buffer *out, const char *pfx, uint64_t ns, const char *sfx)
3951{
3952 double val = ns; // 52 bits of mantissa keep ns accuracy over 52 days
3953 const char *unit;
3954
3955 if (!pfx)
3956 pfx = "";
3957 if (!sfx)
3958 sfx = "";
3959
3960 do {
3961 unit = " - "; if (val <= 0.0) break;
3962 unit = "ns"; if (val < 1000.0) break;
3963 unit = "us"; val /= 1000.0; if (val < 1000.0) break;
3964 unit = "ms"; val /= 1000.0; if (val < 1000.0) break;
3965 unit = "s "; val /= 1000.0; if (val < 60.0) break;
3966 unit = "m "; val /= 60.0; if (val < 60.0) break;
3967 unit = "h "; val /= 60.0; if (val < 24.0) break;
3968 unit = "d "; val /= 24.0; if (val < 365.0) break;
3969 unit = "yr"; val /= 365.0; if (val < 1000.0) break;
3970 unit = " inf "; val = 0.0; break;
3971 } while (0);
3972
3973 if (val <= 0.0)
3974 return chunk_appendf(out, "%s%7s%s", pfx, unit, sfx);
3975 else if (val < 10.0)
3976 return chunk_appendf(out, "%s%1.3f%s%s", pfx, val, unit, sfx);
3977 else if (val < 100.0)
3978 return chunk_appendf(out, "%s%2.2f%s%s", pfx, val, unit, sfx);
3979 else
3980 return chunk_appendf(out, "%s%3.1f%s%s", pfx, val, unit, sfx);
3981}
3982
Willy Tarreau9a7bea52012-04-27 11:16:50 +02003983/* Dynamically allocates a string of the proper length to hold the formatted
3984 * output. NULL is returned on error. The caller is responsible for freeing the
3985 * memory area using free(). The resulting string is returned in <out> if the
3986 * pointer is not NULL. A previous version of <out> might be used to build the
3987 * new string, and it will be freed before returning if it is not NULL, which
3988 * makes it possible to build complex strings from iterative calls without
3989 * having to care about freeing intermediate values, as in the example below :
3990 *
3991 * memprintf(&err, "invalid argument: '%s'", arg);
3992 * ...
3993 * memprintf(&err, "parser said : <%s>\n", *err);
3994 * ...
3995 * free(*err);
3996 *
3997 * This means that <err> must be initialized to NULL before first invocation.
3998 * The return value also holds the allocated string, which eases error checking
3999 * and immediate consumption. If the output pointer is not used, NULL must be
Willy Tarreaueb6cead2012-09-20 19:43:14 +02004000 * passed instead and it will be ignored. The returned message will then also
4001 * be NULL so that the caller does not have to bother with freeing anything.
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004002 *
4003 * It is also convenient to use it without any free except the last one :
4004 * err = NULL;
4005 * if (!fct1(err)) report(*err);
4006 * if (!fct2(err)) report(*err);
4007 * if (!fct3(err)) report(*err);
4008 * free(*err);
Christopher Faulet93a518f2017-10-24 11:25:33 +02004009 *
4010 * memprintf relies on memvprintf. This last version can be called from any
4011 * function with variadic arguments.
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004012 */
Christopher Faulet93a518f2017-10-24 11:25:33 +02004013char *memvprintf(char **out, const char *format, va_list orig_args)
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004014{
4015 va_list args;
4016 char *ret = NULL;
4017 int allocated = 0;
4018 int needed = 0;
4019
Willy Tarreaueb6cead2012-09-20 19:43:14 +02004020 if (!out)
4021 return NULL;
4022
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004023 do {
Willy Tarreaue0609f52019-03-29 19:13:23 +01004024 char buf1;
4025
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004026 /* vsnprintf() will return the required length even when the
4027 * target buffer is NULL. We do this in a loop just in case
4028 * intermediate evaluations get wrong.
4029 */
Christopher Faulet93a518f2017-10-24 11:25:33 +02004030 va_copy(args, orig_args);
Willy Tarreaue0609f52019-03-29 19:13:23 +01004031 needed = vsnprintf(ret ? ret : &buf1, allocated, format, args);
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004032 va_end(args);
Willy Tarreau1b2fed62013-04-01 22:48:54 +02004033 if (needed < allocated) {
4034 /* Note: on Solaris 8, the first iteration always
4035 * returns -1 if allocated is zero, so we force a
4036 * retry.
4037 */
4038 if (!allocated)
4039 needed = 0;
4040 else
4041 break;
4042 }
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004043
Willy Tarreau1b2fed62013-04-01 22:48:54 +02004044 allocated = needed + 1;
Hubert Verstraete831962e2016-06-28 22:44:26 +02004045 ret = my_realloc2(ret, allocated);
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004046 } while (ret);
4047
4048 if (needed < 0) {
4049 /* an error was encountered */
Willy Tarreau61cfdf42021-02-20 10:46:51 +01004050 ha_free(&ret);
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004051 }
4052
4053 if (out) {
4054 free(*out);
4055 *out = ret;
4056 }
4057
4058 return ret;
4059}
William Lallemand421f5b52012-02-06 18:15:57 +01004060
Christopher Faulet93a518f2017-10-24 11:25:33 +02004061char *memprintf(char **out, const char *format, ...)
4062{
4063 va_list args;
4064 char *ret = NULL;
4065
4066 va_start(args, format);
4067 ret = memvprintf(out, format, args);
4068 va_end(args);
4069
4070 return ret;
4071}
4072
Willy Tarreau21c705b2012-09-14 11:40:36 +02004073/* Used to add <level> spaces before each line of <out>, unless there is only one line.
4074 * The input argument is automatically freed and reassigned. The result will have to be
Willy Tarreau70eec382012-10-10 08:56:47 +02004075 * freed by the caller. It also supports being passed a NULL which results in the same
4076 * output.
Willy Tarreau21c705b2012-09-14 11:40:36 +02004077 * Example of use :
4078 * parse(cmd, &err); (callee: memprintf(&err, ...))
4079 * fprintf(stderr, "Parser said: %s\n", indent_error(&err));
4080 * free(err);
4081 */
4082char *indent_msg(char **out, int level)
4083{
4084 char *ret, *in, *p;
4085 int needed = 0;
4086 int lf = 0;
4087 int lastlf = 0;
4088 int len;
4089
Willy Tarreau70eec382012-10-10 08:56:47 +02004090 if (!out || !*out)
4091 return NULL;
4092
Willy Tarreau21c705b2012-09-14 11:40:36 +02004093 in = *out - 1;
4094 while ((in = strchr(in + 1, '\n')) != NULL) {
4095 lastlf = in - *out;
4096 lf++;
4097 }
4098
4099 if (!lf) /* single line, no LF, return it as-is */
4100 return *out;
4101
4102 len = strlen(*out);
4103
4104 if (lf == 1 && lastlf == len - 1) {
4105 /* single line, LF at end, strip it and return as-is */
4106 (*out)[lastlf] = 0;
4107 return *out;
4108 }
4109
4110 /* OK now we have at least one LF, we need to process the whole string
4111 * as a multi-line string. What we'll do :
4112 * - prefix with an LF if there is none
4113 * - add <level> spaces before each line
4114 * This means at most ( 1 + level + (len-lf) + lf*<1+level) ) =
4115 * 1 + level + len + lf * level = 1 + level * (lf + 1) + len.
4116 */
4117
4118 needed = 1 + level * (lf + 1) + len + 1;
4119 p = ret = malloc(needed);
4120 in = *out;
4121
4122 /* skip initial LFs */
4123 while (*in == '\n')
4124 in++;
4125
4126 /* copy each line, prefixed with LF and <level> spaces, and without the trailing LF */
4127 while (*in) {
4128 *p++ = '\n';
4129 memset(p, ' ', level);
4130 p += level;
4131 do {
4132 *p++ = *in++;
4133 } while (*in && *in != '\n');
4134 if (*in)
4135 in++;
4136 }
4137 *p = 0;
4138
4139 free(*out);
4140 *out = ret;
4141
4142 return ret;
4143}
4144
Willy Tarreaua2c99112019-08-21 13:17:37 +02004145/* makes a copy of message <in> into <out>, with each line prefixed with <pfx>
4146 * and end of lines replaced with <eol> if not 0. The first line to indent has
4147 * to be indicated in <first> (starts at zero), so that it is possible to skip
4148 * indenting the first line if it has to be appended after an existing message.
4149 * Empty strings are never indented, and NULL strings are considered empty both
4150 * for <in> and <pfx>. It returns non-zero if an EOL was appended as the last
4151 * character, non-zero otherwise.
4152 */
4153int append_prefixed_str(struct buffer *out, const char *in, const char *pfx, char eol, int first)
4154{
4155 int bol, lf;
4156 int pfxlen = pfx ? strlen(pfx) : 0;
4157
4158 if (!in)
4159 return 0;
4160
4161 bol = 1;
4162 lf = 0;
4163 while (*in) {
4164 if (bol && pfxlen) {
4165 if (first > 0)
4166 first--;
4167 else
4168 b_putblk(out, pfx, pfxlen);
4169 bol = 0;
4170 }
4171
4172 lf = (*in == '\n');
4173 bol |= lf;
4174 b_putchr(out, (lf && eol) ? eol : *in);
4175 in++;
4176 }
4177 return lf;
4178}
4179
Willy Tarreau9d22e562019-03-29 18:49:09 +01004180/* removes environment variable <name> from the environment as found in
4181 * environ. This is only provided as an alternative for systems without
4182 * unsetenv() (old Solaris and AIX versions). THIS IS NOT THREAD SAFE.
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05004183 * The principle is to scan environ for each occurrence of variable name
Willy Tarreau9d22e562019-03-29 18:49:09 +01004184 * <name> and to replace the matching pointers with the last pointer of
4185 * the array (since variables are not ordered).
4186 * It always returns 0 (success).
4187 */
4188int my_unsetenv(const char *name)
4189{
4190 extern char **environ;
4191 char **p = environ;
4192 int vars;
4193 int next;
4194 int len;
4195
4196 len = strlen(name);
4197 for (vars = 0; p[vars]; vars++)
4198 ;
4199 next = 0;
4200 while (next < vars) {
4201 if (strncmp(p[next], name, len) != 0 || p[next][len] != '=') {
4202 next++;
4203 continue;
4204 }
4205 if (next < vars - 1)
4206 p[next] = p[vars - 1];
4207 p[--vars] = NULL;
4208 }
4209 return 0;
4210}
4211
Willy Tarreaudad36a32013-03-11 01:20:04 +01004212/* Convert occurrences of environment variables in the input string to their
4213 * corresponding value. A variable is identified as a series of alphanumeric
4214 * characters or underscores following a '$' sign. The <in> string must be
4215 * free()able. NULL returns NULL. The resulting string might be reallocated if
4216 * some expansion is made. Variable names may also be enclosed into braces if
4217 * needed (eg: to concatenate alphanum characters).
4218 */
4219char *env_expand(char *in)
4220{
4221 char *txt_beg;
4222 char *out;
4223 char *txt_end;
4224 char *var_beg;
4225 char *var_end;
4226 char *value;
4227 char *next;
4228 int out_len;
4229 int val_len;
4230
4231 if (!in)
4232 return in;
4233
4234 value = out = NULL;
4235 out_len = 0;
4236
4237 txt_beg = in;
4238 do {
4239 /* look for next '$' sign in <in> */
4240 for (txt_end = txt_beg; *txt_end && *txt_end != '$'; txt_end++);
4241
4242 if (!*txt_end && !out) /* end and no expansion performed */
4243 return in;
4244
4245 val_len = 0;
4246 next = txt_end;
4247 if (*txt_end == '$') {
4248 char save;
4249
4250 var_beg = txt_end + 1;
4251 if (*var_beg == '{')
4252 var_beg++;
4253
4254 var_end = var_beg;
Willy Tarreau90807112020-02-25 08:16:33 +01004255 while (isalnum((unsigned char)*var_end) || *var_end == '_') {
Willy Tarreaudad36a32013-03-11 01:20:04 +01004256 var_end++;
4257 }
4258
4259 next = var_end;
4260 if (*var_end == '}' && (var_beg > txt_end + 1))
4261 next++;
4262
4263 /* get value of the variable name at this location */
4264 save = *var_end;
4265 *var_end = '\0';
4266 value = getenv(var_beg);
4267 *var_end = save;
4268 val_len = value ? strlen(value) : 0;
4269 }
4270
Hubert Verstraete831962e2016-06-28 22:44:26 +02004271 out = my_realloc2(out, out_len + (txt_end - txt_beg) + val_len + 1);
Willy Tarreaudad36a32013-03-11 01:20:04 +01004272 if (txt_end > txt_beg) {
4273 memcpy(out + out_len, txt_beg, txt_end - txt_beg);
4274 out_len += txt_end - txt_beg;
4275 }
4276 if (val_len) {
4277 memcpy(out + out_len, value, val_len);
4278 out_len += val_len;
4279 }
4280 out[out_len] = 0;
4281 txt_beg = next;
4282 } while (*txt_beg);
4283
4284 /* here we know that <out> was allocated and that we don't need <in> anymore */
4285 free(in);
4286 return out;
4287}
4288
de Lafond Guillaume88c278f2013-04-15 19:27:10 +02004289
4290/* same as strstr() but case-insensitive and with limit length */
4291const char *strnistr(const char *str1, int len_str1, const char *str2, int len_str2)
4292{
4293 char *pptr, *sptr, *start;
Willy Tarreauc8746532014-05-28 23:05:07 +02004294 unsigned int slen, plen;
4295 unsigned int tmp1, tmp2;
de Lafond Guillaume88c278f2013-04-15 19:27:10 +02004296
4297 if (str1 == NULL || len_str1 == 0) // search pattern into an empty string => search is not found
4298 return NULL;
4299
4300 if (str2 == NULL || len_str2 == 0) // pattern is empty => every str1 match
4301 return str1;
4302
4303 if (len_str1 < len_str2) // pattern is longer than string => search is not found
4304 return NULL;
4305
4306 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 +02004307 while (toupper((unsigned char)*start) != toupper((unsigned char)*str2)) {
de Lafond Guillaume88c278f2013-04-15 19:27:10 +02004308 start++;
4309 slen--;
4310 tmp1++;
4311
4312 if (tmp1 >= len_str1)
4313 return NULL;
4314
4315 /* if pattern longer than string */
4316 if (slen < plen)
4317 return NULL;
4318 }
4319
4320 sptr = start;
4321 pptr = (char *)str2;
4322
4323 tmp2 = 0;
Willy Tarreauf278eec2020-07-05 21:46:32 +02004324 while (toupper((unsigned char)*sptr) == toupper((unsigned char)*pptr)) {
de Lafond Guillaume88c278f2013-04-15 19:27:10 +02004325 sptr++;
4326 pptr++;
4327 tmp2++;
4328
4329 if (*pptr == '\0' || tmp2 == len_str2) /* end of pattern found */
4330 return start;
4331 if (*sptr == '\0' || tmp2 == len_str1) /* end of string found and the pattern is not fully found */
4332 return NULL;
4333 }
4334 }
4335 return NULL;
4336}
4337
Thierry FOURNIER317e1c42014-08-12 10:20:47 +02004338/* This function read the next valid utf8 char.
4339 * <s> is the byte srray to be decode, <len> is its length.
4340 * The function returns decoded char encoded like this:
4341 * The 4 msb are the return code (UTF8_CODE_*), the 4 lsb
4342 * are the length read. The decoded character is stored in <c>.
4343 */
4344unsigned char utf8_next(const char *s, int len, unsigned int *c)
4345{
4346 const unsigned char *p = (unsigned char *)s;
4347 int dec;
4348 unsigned char code = UTF8_CODE_OK;
4349
4350 if (len < 1)
4351 return UTF8_CODE_OK;
4352
4353 /* Check the type of UTF8 sequence
4354 *
4355 * 0... .... 0x00 <= x <= 0x7f : 1 byte: ascii char
4356 * 10.. .... 0x80 <= x <= 0xbf : invalid sequence
4357 * 110. .... 0xc0 <= x <= 0xdf : 2 bytes
4358 * 1110 .... 0xe0 <= x <= 0xef : 3 bytes
4359 * 1111 0... 0xf0 <= x <= 0xf7 : 4 bytes
4360 * 1111 10.. 0xf8 <= x <= 0xfb : 5 bytes
4361 * 1111 110. 0xfc <= x <= 0xfd : 6 bytes
4362 * 1111 111. 0xfe <= x <= 0xff : invalid sequence
4363 */
4364 switch (*p) {
4365 case 0x00 ... 0x7f:
4366 *c = *p;
4367 return UTF8_CODE_OK | 1;
4368
4369 case 0x80 ... 0xbf:
4370 *c = *p;
4371 return UTF8_CODE_BADSEQ | 1;
4372
4373 case 0xc0 ... 0xdf:
4374 if (len < 2) {
4375 *c = *p;
4376 return UTF8_CODE_BADSEQ | 1;
4377 }
4378 *c = *p & 0x1f;
4379 dec = 1;
4380 break;
4381
4382 case 0xe0 ... 0xef:
4383 if (len < 3) {
4384 *c = *p;
4385 return UTF8_CODE_BADSEQ | 1;
4386 }
4387 *c = *p & 0x0f;
4388 dec = 2;
4389 break;
4390
4391 case 0xf0 ... 0xf7:
4392 if (len < 4) {
4393 *c = *p;
4394 return UTF8_CODE_BADSEQ | 1;
4395 }
4396 *c = *p & 0x07;
4397 dec = 3;
4398 break;
4399
4400 case 0xf8 ... 0xfb:
4401 if (len < 5) {
4402 *c = *p;
4403 return UTF8_CODE_BADSEQ | 1;
4404 }
4405 *c = *p & 0x03;
4406 dec = 4;
4407 break;
4408
4409 case 0xfc ... 0xfd:
4410 if (len < 6) {
4411 *c = *p;
4412 return UTF8_CODE_BADSEQ | 1;
4413 }
4414 *c = *p & 0x01;
4415 dec = 5;
4416 break;
4417
4418 case 0xfe ... 0xff:
4419 default:
4420 *c = *p;
4421 return UTF8_CODE_BADSEQ | 1;
4422 }
4423
4424 p++;
4425
4426 while (dec > 0) {
4427
4428 /* need 0x10 for the 2 first bits */
4429 if ( ( *p & 0xc0 ) != 0x80 )
4430 return UTF8_CODE_BADSEQ | ((p-(unsigned char *)s)&0xffff);
4431
4432 /* add data at char */
4433 *c = ( *c << 6 ) | ( *p & 0x3f );
4434
4435 dec--;
4436 p++;
4437 }
4438
4439 /* Check ovelong encoding.
4440 * 1 byte : 5 + 6 : 11 : 0x80 ... 0x7ff
4441 * 2 bytes : 4 + 6 + 6 : 16 : 0x800 ... 0xffff
4442 * 3 bytes : 3 + 6 + 6 + 6 : 21 : 0x10000 ... 0x1fffff
4443 */
Thierry FOURNIER9e7ec082015-03-12 19:32:38 +01004444 if (( *c <= 0x7f && (p-(unsigned char *)s) > 1) ||
Thierry FOURNIER317e1c42014-08-12 10:20:47 +02004445 (*c >= 0x80 && *c <= 0x7ff && (p-(unsigned char *)s) > 2) ||
4446 (*c >= 0x800 && *c <= 0xffff && (p-(unsigned char *)s) > 3) ||
4447 (*c >= 0x10000 && *c <= 0x1fffff && (p-(unsigned char *)s) > 4))
4448 code |= UTF8_CODE_OVERLONG;
4449
4450 /* Check invalid UTF8 range. */
4451 if ((*c >= 0xd800 && *c <= 0xdfff) ||
4452 (*c >= 0xfffe && *c <= 0xffff))
4453 code |= UTF8_CODE_INVRANGE;
4454
4455 return code | ((p-(unsigned char *)s)&0x0f);
4456}
4457
Maxime de Roucydc887852016-05-13 23:52:54 +02004458/* append a copy of string <str> (in a wordlist) at the end of the list <li>
4459 * On failure : return 0 and <err> filled with an error message.
4460 * The caller is responsible for freeing the <err> and <str> copy
4461 * memory area using free()
4462 */
4463int list_append_word(struct list *li, const char *str, char **err)
4464{
4465 struct wordlist *wl;
4466
4467 wl = calloc(1, sizeof(*wl));
4468 if (!wl) {
4469 memprintf(err, "out of memory");
4470 goto fail_wl;
4471 }
4472
4473 wl->s = strdup(str);
4474 if (!wl->s) {
4475 memprintf(err, "out of memory");
4476 goto fail_wl_s;
4477 }
4478
Willy Tarreau2b718102021-04-21 07:32:39 +02004479 LIST_APPEND(li, &wl->list);
Maxime de Roucydc887852016-05-13 23:52:54 +02004480
4481 return 1;
4482
4483fail_wl_s:
4484 free(wl->s);
4485fail_wl:
4486 free(wl);
4487 return 0;
4488}
4489
Willy Tarreau37101052019-05-20 16:48:20 +02004490/* indicates if a memory location may safely be read or not. The trick consists
4491 * in performing a harmless syscall using this location as an input and letting
4492 * the operating system report whether it's OK or not. For this we have the
4493 * stat() syscall, which will return EFAULT when the memory location supposed
4494 * to contain the file name is not readable. If it is readable it will then
4495 * either return 0 if the area contains an existing file name, or -1 with
4496 * another code. This must not be abused, and some audit systems might detect
4497 * this as abnormal activity. It's used only for unsafe dumps.
4498 */
4499int may_access(const void *ptr)
4500{
4501 struct stat buf;
4502
4503 if (stat(ptr, &buf) == 0)
4504 return 1;
4505 if (errno == EFAULT)
4506 return 0;
4507 return 1;
4508}
4509
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004510/* print a string of text buffer to <out>. The format is :
4511 * Non-printable chars \t, \n, \r and \e are * encoded in C format.
4512 * Other non-printable chars are encoded "\xHH". Space, '\', and '=' are also escaped.
4513 * Print stopped if null char or <bsize> is reached, or if no more place in the chunk.
4514 */
Willy Tarreau83061a82018-07-13 11:56:34 +02004515int dump_text(struct buffer *out, const char *buf, int bsize)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004516{
4517 unsigned char c;
Tim Duesterhuscd5521e2021-08-29 00:58:22 +02004518 size_t ptr = 0;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004519
Tim Duesterhuscd5521e2021-08-29 00:58:22 +02004520 while (ptr < bsize && buf[ptr]) {
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004521 c = buf[ptr];
Willy Tarreau90807112020-02-25 08:16:33 +01004522 if (isprint((unsigned char)c) && isascii((unsigned char)c) && c != '\\' && c != ' ' && c != '=') {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004523 if (out->data > out->size - 1)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004524 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004525 out->area[out->data++] = c;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004526 }
4527 else if (c == '\t' || c == '\n' || c == '\r' || c == '\e' || c == '\\' || c == ' ' || c == '=') {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004528 if (out->data > out->size - 2)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004529 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004530 out->area[out->data++] = '\\';
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004531 switch (c) {
4532 case ' ': c = ' '; break;
4533 case '\t': c = 't'; break;
4534 case '\n': c = 'n'; break;
4535 case '\r': c = 'r'; break;
4536 case '\e': c = 'e'; break;
4537 case '\\': c = '\\'; break;
4538 case '=': c = '='; break;
4539 }
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004540 out->area[out->data++] = c;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004541 }
4542 else {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004543 if (out->data > out->size - 4)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004544 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004545 out->area[out->data++] = '\\';
4546 out->area[out->data++] = 'x';
4547 out->area[out->data++] = hextab[(c >> 4) & 0xF];
4548 out->area[out->data++] = hextab[c & 0xF];
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004549 }
4550 ptr++;
4551 }
4552
4553 return ptr;
4554}
4555
4556/* print a buffer in hexa.
4557 * Print stopped if <bsize> is reached, or if no more place in the chunk.
4558 */
Willy Tarreau83061a82018-07-13 11:56:34 +02004559int dump_binary(struct buffer *out, const char *buf, int bsize)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004560{
4561 unsigned char c;
4562 int ptr = 0;
4563
4564 while (ptr < bsize) {
4565 c = buf[ptr];
4566
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004567 if (out->data > out->size - 2)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004568 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004569 out->area[out->data++] = hextab[(c >> 4) & 0xF];
4570 out->area[out->data++] = hextab[c & 0xF];
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004571
4572 ptr++;
4573 }
4574 return ptr;
4575}
4576
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004577/* Appends into buffer <out> a hex dump of memory area <buf> for <len> bytes,
4578 * prepending each line with prefix <pfx>. The output is *not* initialized.
4579 * The output will not wrap pas the buffer's end so it is more optimal if the
4580 * caller makes sure the buffer is aligned first. A trailing zero will always
4581 * be appended (and not counted) if there is room for it. The caller must make
Willy Tarreau37101052019-05-20 16:48:20 +02004582 * sure that the area is dumpable first. If <unsafe> is non-null, the memory
4583 * locations are checked first for being readable.
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004584 */
Willy Tarreau37101052019-05-20 16:48:20 +02004585void dump_hex(struct buffer *out, const char *pfx, const void *buf, int len, int unsafe)
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004586{
4587 const unsigned char *d = buf;
4588 int i, j, start;
4589
4590 d = (const unsigned char *)(((unsigned long)buf) & -16);
4591 start = ((unsigned long)buf) & 15;
4592
4593 for (i = 0; i < start + len; i += 16) {
4594 chunk_appendf(out, (sizeof(void *) == 4) ? "%s%8p: " : "%s%16p: ", pfx, d + i);
4595
Willy Tarreau37101052019-05-20 16:48:20 +02004596 // 0: unchecked, 1: checked safe, 2: danger
4597 unsafe = !!unsafe;
4598 if (unsafe && !may_access(d + i))
4599 unsafe = 2;
4600
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004601 for (j = 0; j < 16; j++) {
Willy Tarreau37101052019-05-20 16:48:20 +02004602 if ((i + j < start) || (i + j >= start + len))
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004603 chunk_strcat(out, "'' ");
Willy Tarreau37101052019-05-20 16:48:20 +02004604 else if (unsafe > 1)
4605 chunk_strcat(out, "** ");
4606 else
4607 chunk_appendf(out, "%02x ", d[i + j]);
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004608
4609 if (j == 7)
4610 chunk_strcat(out, "- ");
4611 }
4612 chunk_strcat(out, " ");
4613 for (j = 0; j < 16; j++) {
Willy Tarreau37101052019-05-20 16:48:20 +02004614 if ((i + j < start) || (i + j >= start + len))
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004615 chunk_strcat(out, "'");
Willy Tarreau37101052019-05-20 16:48:20 +02004616 else if (unsafe > 1)
4617 chunk_strcat(out, "*");
Willy Tarreau90807112020-02-25 08:16:33 +01004618 else if (isprint((unsigned char)d[i + j]))
Willy Tarreau37101052019-05-20 16:48:20 +02004619 chunk_appendf(out, "%c", d[i + j]);
4620 else
4621 chunk_strcat(out, ".");
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004622 }
4623 chunk_strcat(out, "\n");
4624 }
4625}
4626
Willy Tarreau762fb3e2020-03-03 15:57:10 +01004627/* dumps <pfx> followed by <n> bytes from <addr> in hex form into buffer <buf>
4628 * enclosed in brackets after the address itself, formatted on 14 chars
4629 * including the "0x" prefix. This is meant to be used as a prefix for code
4630 * areas. For example:
4631 * "0x7f10b6557690 [48 c7 c0 0f 00 00 00 0f]"
4632 * It relies on may_access() to know if the bytes are dumpable, otherwise "--"
4633 * is emitted. A NULL <pfx> will be considered empty.
4634 */
4635void dump_addr_and_bytes(struct buffer *buf, const char *pfx, const void *addr, int n)
4636{
4637 int ok = 0;
4638 int i;
4639
4640 chunk_appendf(buf, "%s%#14lx [", pfx ? pfx : "", (long)addr);
4641
4642 for (i = 0; i < n; i++) {
4643 if (i == 0 || (((long)(addr + i) ^ (long)(addr)) & 4096))
4644 ok = may_access(addr + i);
4645 if (ok)
4646 chunk_appendf(buf, "%02x%s", ((uint8_t*)addr)[i], (i<n-1) ? " " : "]");
4647 else
4648 chunk_appendf(buf, "--%s", (i<n-1) ? " " : "]");
4649 }
4650}
4651
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004652/* print a line of text buffer (limited to 70 bytes) to <out>. The format is :
4653 * <2 spaces> <offset=5 digits> <space or plus> <space> <70 chars max> <\n>
4654 * which is 60 chars per line. Non-printable chars \t, \n, \r and \e are
4655 * encoded in C format. Other non-printable chars are encoded "\xHH". Original
4656 * lines are respected within the limit of 70 output chars. Lines that are
4657 * continuation of a previous truncated line begin with "+" instead of " "
4658 * after the offset. The new pointer is returned.
4659 */
Willy Tarreau83061a82018-07-13 11:56:34 +02004660int dump_text_line(struct buffer *out, const char *buf, int bsize, int len,
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004661 int *line, int ptr)
4662{
4663 int end;
4664 unsigned char c;
4665
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004666 end = out->data + 80;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004667 if (end > out->size)
4668 return ptr;
4669
4670 chunk_appendf(out, " %05d%c ", ptr, (ptr == *line) ? ' ' : '+');
4671
4672 while (ptr < len && ptr < bsize) {
4673 c = buf[ptr];
Willy Tarreau90807112020-02-25 08:16:33 +01004674 if (isprint((unsigned char)c) && isascii((unsigned char)c) && c != '\\') {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004675 if (out->data > end - 2)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004676 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004677 out->area[out->data++] = c;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004678 } else if (c == '\t' || c == '\n' || c == '\r' || c == '\e' || c == '\\') {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004679 if (out->data > end - 3)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004680 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004681 out->area[out->data++] = '\\';
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004682 switch (c) {
4683 case '\t': c = 't'; break;
4684 case '\n': c = 'n'; break;
4685 case '\r': c = 'r'; break;
4686 case '\e': c = 'e'; break;
4687 case '\\': c = '\\'; break;
4688 }
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004689 out->area[out->data++] = c;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004690 } else {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004691 if (out->data > end - 5)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004692 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004693 out->area[out->data++] = '\\';
4694 out->area[out->data++] = 'x';
4695 out->area[out->data++] = hextab[(c >> 4) & 0xF];
4696 out->area[out->data++] = hextab[c & 0xF];
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004697 }
4698 if (buf[ptr++] == '\n') {
4699 /* we had a line break, let's return now */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004700 out->area[out->data++] = '\n';
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004701 *line = ptr;
4702 return ptr;
4703 }
4704 }
4705 /* we have an incomplete line, we return it as-is */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004706 out->area[out->data++] = '\n';
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004707 return ptr;
4708}
4709
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004710/* displays a <len> long memory block at <buf>, assuming first byte of <buf>
Willy Tarreaued936c52017-04-27 18:03:20 +02004711 * has address <baseaddr>. String <pfx> may be placed as a prefix in front of
4712 * each line. It may be NULL if unused. The output is emitted to file <out>.
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004713 */
Willy Tarreaued936c52017-04-27 18:03:20 +02004714void debug_hexdump(FILE *out, const char *pfx, const char *buf,
4715 unsigned int baseaddr, int len)
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004716{
Willy Tarreau73459792017-04-11 07:58:08 +02004717 unsigned int i;
4718 int b, j;
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004719
4720 for (i = 0; i < (len + (baseaddr & 15)); i += 16) {
4721 b = i - (baseaddr & 15);
Willy Tarreaued936c52017-04-27 18:03:20 +02004722 fprintf(out, "%s%08x: ", pfx ? pfx : "", i + (baseaddr & ~15));
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004723 for (j = 0; j < 8; j++) {
4724 if (b + j >= 0 && b + j < len)
4725 fprintf(out, "%02x ", (unsigned char)buf[b + j]);
4726 else
4727 fprintf(out, " ");
4728 }
4729
4730 if (b + j >= 0 && b + j < len)
4731 fputc('-', out);
4732 else
4733 fputc(' ', out);
4734
4735 for (j = 8; j < 16; j++) {
4736 if (b + j >= 0 && b + j < len)
4737 fprintf(out, " %02x", (unsigned char)buf[b + j]);
4738 else
4739 fprintf(out, " ");
4740 }
4741
4742 fprintf(out, " ");
4743 for (j = 0; j < 16; j++) {
4744 if (b + j >= 0 && b + j < len) {
4745 if (isprint((unsigned char)buf[b + j]))
4746 fputc((unsigned char)buf[b + j], out);
4747 else
4748 fputc('.', out);
4749 }
4750 else
4751 fputc(' ', out);
4752 }
4753 fputc('\n', out);
4754 }
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004755}
4756
Willy Tarreaubb869862020-04-16 10:52:41 +02004757/* Tries to report the executable path name on platforms supporting this. If
4758 * not found or not possible, returns NULL.
4759 */
4760const char *get_exec_path()
4761{
4762 const char *ret = NULL;
4763
4764#if (__GLIBC__ > 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 16))
4765 long execfn = getauxval(AT_EXECFN);
4766
4767 if (execfn && execfn != ENOENT)
4768 ret = (const char *)execfn;
David Carlier1b9d57d2021-08-17 08:44:25 +01004769#elif defined(__NetBSD__)
4770 AuxInfo *auxv;
4771 for (auxv = _dlauxinfo(); auxv->a_type != AT_NULL; ++auxv) {
4772 if (auxv->a_type == AT_SUN_EXECNAME) {
4773 ret = (const char *)auxv->a_v;
4774 break;
4775 }
4776 }
Willy Tarreaubb869862020-04-16 10:52:41 +02004777#endif
4778 return ret;
4779}
4780
Baruch Siache1651b22020-07-24 07:52:20 +03004781#if (defined(__ELF__) && !defined(__linux__)) || defined(USE_DL)
Willy Tarreau9133e482020-03-04 10:19:36 +01004782/* calls dladdr() or dladdr1() on <addr> and <dli>. If dladdr1 is available,
4783 * also returns the symbol size in <size>, otherwise returns 0 there.
4784 */
4785static int dladdr_and_size(const void *addr, Dl_info *dli, size_t *size)
4786{
4787 int ret;
Willy Tarreau62af9c82020-03-10 07:51:48 +01004788#if (__GLIBC__ > 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 3)) // most detailed one
Willy Tarreau9133e482020-03-04 10:19:36 +01004789 const ElfW(Sym) *sym;
4790
4791 ret = dladdr1(addr, dli, (void **)&sym, RTLD_DL_SYMENT);
4792 if (ret)
4793 *size = sym ? sym->st_size : 0;
4794#else
4795 ret = dladdr(addr, dli);
4796 *size = 0;
4797#endif
4798 return ret;
4799}
Willy Tarreau64192392021-05-05 09:06:21 +02004800
4801/* Tries to retrieve the address of the first occurrence symbol <name>.
4802 * Note that NULL in return is not always an error as a symbol may have that
4803 * address in special situations.
4804 */
4805void *get_sym_curr_addr(const char *name)
4806{
4807 void *ptr = NULL;
4808
4809#ifdef RTLD_DEFAULT
4810 ptr = dlsym(RTLD_DEFAULT, name);
4811#endif
4812 return ptr;
4813}
4814
4815
4816/* Tries to retrieve the address of the next occurrence of symbol <name>
4817 * Note that NULL in return is not always an error as a symbol may have that
4818 * address in special situations.
4819 */
4820void *get_sym_next_addr(const char *name)
4821{
4822 void *ptr = NULL;
4823
4824#ifdef RTLD_NEXT
4825 ptr = dlsym(RTLD_NEXT, name);
Willy Tarreau9133e482020-03-04 10:19:36 +01004826#endif
Willy Tarreau64192392021-05-05 09:06:21 +02004827 return ptr;
4828}
4829
4830#else /* elf & linux & dl */
4831
4832/* no possible resolving on other platforms at the moment */
4833void *get_sym_curr_addr(const char *name)
4834{
4835 return NULL;
4836}
4837
4838void *get_sym_next_addr(const char *name)
4839{
4840 return NULL;
4841}
4842
4843#endif /* elf & linux & dl */
Willy Tarreau9133e482020-03-04 10:19:36 +01004844
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004845/* Tries to append to buffer <buf> some indications about the symbol at address
4846 * <addr> using the following form:
4847 * lib:+0xoffset (unresolvable address from lib's base)
4848 * main+0xoffset (unresolvable address from main (+/-))
4849 * lib:main+0xoffset (unresolvable lib address from main (+/-))
4850 * name (resolved exact exec address)
4851 * lib:name (resolved exact lib address)
4852 * name+0xoffset/0xsize (resolved address within exec symbol)
4853 * lib:name+0xoffset/0xsize (resolved address within lib symbol)
4854 *
4855 * The file name (lib or executable) is limited to what lies between the last
4856 * '/' and the first following '.'. An optional prefix <pfx> is prepended before
4857 * 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 +03004858 * that contains the "main" symbol, or when __ELF__ && USE_DL are not set.
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004859 *
4860 * The symbol's base address is returned, or NULL when unresolved, in order to
4861 * allow the caller to match it against known ones.
4862 */
Willy Tarreau45fd1032021-01-20 14:37:59 +01004863const void *resolve_sym_name(struct buffer *buf, const char *pfx, const void *addr)
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004864{
4865 const struct {
4866 const void *func;
4867 const char *name;
4868 } fcts[] = {
4869 { .func = process_stream, .name = "process_stream" },
4870 { .func = task_run_applet, .name = "task_run_applet" },
4871 { .func = si_cs_io_cb, .name = "si_cs_io_cb" },
Willy Tarreau586f71b2020-12-11 15:54:36 +01004872 { .func = sock_conn_iocb, .name = "sock_conn_iocb" },
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004873 { .func = dgram_fd_handler, .name = "dgram_fd_handler" },
4874 { .func = listener_accept, .name = "listener_accept" },
Willy Tarreaud597ec22021-01-29 14:29:06 +01004875 { .func = manage_global_listener_queue, .name = "manage_global_listener_queue" },
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004876 { .func = poller_pipe_io_handler, .name = "poller_pipe_io_handler" },
4877 { .func = mworker_accept_wrapper, .name = "mworker_accept_wrapper" },
Willy Tarreau02922e12021-01-29 12:27:57 +01004878 { .func = session_expire_embryonic, .name = "session_expire_embryonic" },
Willy Tarreaufb5401f2021-01-29 12:25:23 +01004879#ifdef USE_THREAD
4880 { .func = accept_queue_process, .name = "accept_queue_process" },
4881#endif
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004882#ifdef USE_LUA
4883 { .func = hlua_process_task, .name = "hlua_process_task" },
4884#endif
Ilya Shipitsinbdec3ba2020-11-14 01:56:34 +05004885#ifdef SSL_MODE_ASYNC
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004886 { .func = ssl_async_fd_free, .name = "ssl_async_fd_free" },
4887 { .func = ssl_async_fd_handler, .name = "ssl_async_fd_handler" },
4888#endif
4889 };
4890
Baruch Siache1651b22020-07-24 07:52:20 +03004891#if (defined(__ELF__) && !defined(__linux__)) || defined(USE_DL)
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004892 Dl_info dli, dli_main;
Willy Tarreau9133e482020-03-04 10:19:36 +01004893 size_t size;
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004894 const char *fname, *p;
4895#endif
4896 int i;
4897
4898 if (pfx)
4899 chunk_appendf(buf, "%s", pfx);
4900
4901 for (i = 0; i < sizeof(fcts) / sizeof(fcts[0]); i++) {
4902 if (addr == fcts[i].func) {
4903 chunk_appendf(buf, "%s", fcts[i].name);
4904 return addr;
4905 }
4906 }
4907
Baruch Siache1651b22020-07-24 07:52:20 +03004908#if (defined(__ELF__) && !defined(__linux__)) || defined(USE_DL)
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004909 /* Now let's try to be smarter */
Willy Tarreau9133e482020-03-04 10:19:36 +01004910 if (!dladdr_and_size(addr, &dli, &size))
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004911 goto unknown;
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004912
4913 /* 1. prefix the library name if it's not the same object as the one
4914 * that contains the main function. The name is picked between last '/'
4915 * and first following '.'.
4916 */
4917 if (!dladdr(main, &dli_main))
4918 dli_main.dli_fbase = NULL;
4919
4920 if (dli_main.dli_fbase != dli.dli_fbase) {
4921 fname = dli.dli_fname;
4922 p = strrchr(fname, '/');
4923 if (p++)
4924 fname = p;
4925 p = strchr(fname, '.');
4926 if (!p)
4927 p = fname + strlen(fname);
4928
4929 chunk_appendf(buf, "%.*s:", (int)(long)(p - fname), fname);
4930 }
4931
4932 /* 2. symbol name */
4933 if (dli.dli_sname) {
4934 /* known, dump it and return symbol's address (exact or relative) */
4935 chunk_appendf(buf, "%s", dli.dli_sname);
4936 if (addr != dli.dli_saddr) {
4937 chunk_appendf(buf, "+%#lx", (long)(addr - dli.dli_saddr));
Willy Tarreau9133e482020-03-04 10:19:36 +01004938 if (size)
4939 chunk_appendf(buf, "/%#lx", (long)size);
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004940 }
4941 return dli.dli_saddr;
4942 }
4943 else if (dli_main.dli_fbase != dli.dli_fbase) {
4944 /* unresolved symbol from a known library, report relative offset */
4945 chunk_appendf(buf, "+%#lx", (long)(addr - dli.dli_fbase));
4946 return NULL;
4947 }
Baruch Siache1651b22020-07-24 07:52:20 +03004948#endif /* __ELF__ && !__linux__ || USE_DL */
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004949 unknown:
4950 /* unresolved symbol from the main file, report relative offset to main */
4951 if ((void*)addr < (void*)main)
4952 chunk_appendf(buf, "main-%#lx", (long)((void*)main - addr));
4953 else
4954 chunk_appendf(buf, "main+%#lx", (long)(addr - (void*)main));
4955 return NULL;
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004956}
4957
Frédéric Lécaille3b717162019-02-25 15:04:22 +01004958/*
4959 * Allocate an array of unsigned int with <nums> as address from <str> string
Ilya Shipitsin46a030c2020-07-05 16:36:08 +05004960 * made of integer separated by dot characters.
Frédéric Lécaille3b717162019-02-25 15:04:22 +01004961 *
4962 * First, initializes the value with <sz> as address to 0 and initializes the
4963 * array with <nums> as address to NULL. Then allocates the array with <nums> as
4964 * address updating <sz> pointed value to the size of this array.
4965 *
4966 * Returns 1 if succeeded, 0 if not.
4967 */
4968int parse_dotted_uints(const char *str, unsigned int **nums, size_t *sz)
4969{
4970 unsigned int *n;
4971 const char *s, *end;
4972
4973 s = str;
4974 *sz = 0;
4975 end = str + strlen(str);
4976 *nums = n = NULL;
4977
4978 while (1) {
4979 unsigned int r;
4980
4981 if (s >= end)
4982 break;
4983
4984 r = read_uint(&s, end);
4985 /* Expected characters after having read an uint: '\0' or '.',
4986 * if '.', must not be terminal.
4987 */
Christopher Faulet4b524122021-02-11 10:42:41 +01004988 if (*s != '\0'&& (*s++ != '.' || s == end)) {
4989 free(n);
Frédéric Lécaille3b717162019-02-25 15:04:22 +01004990 return 0;
Christopher Faulet4b524122021-02-11 10:42:41 +01004991 }
Frédéric Lécaille3b717162019-02-25 15:04:22 +01004992
Frédéric Lécaille12a71842019-02-26 18:19:48 +01004993 n = my_realloc2(n, (*sz + 1) * sizeof *n);
Frédéric Lécaille3b717162019-02-25 15:04:22 +01004994 if (!n)
4995 return 0;
4996
4997 n[(*sz)++] = r;
4998 }
4999 *nums = n;
5000
5001 return 1;
5002}
5003
Willy Tarreau4d589e72019-08-23 19:02:26 +02005004
5005/* returns the number of bytes needed to encode <v> as a varint. An inline
5006 * version exists for use with constants (__varint_bytes()).
5007 */
5008int varint_bytes(uint64_t v)
5009{
5010 int len = 1;
5011
5012 if (v >= 240) {
5013 v = (v - 240) >> 4;
5014 while (1) {
5015 len++;
5016 if (v < 128)
5017 break;
5018 v = (v - 128) >> 7;
5019 }
5020 }
5021 return len;
5022}
5023
Willy Tarreau52bf8392020-03-08 00:42:37 +01005024
5025/* Random number generator state, see below */
Willy Tarreau1544c142020-03-12 00:31:18 +01005026static uint64_t ha_random_state[2] ALIGNED(2*sizeof(uint64_t));
Willy Tarreau52bf8392020-03-08 00:42:37 +01005027
5028/* This is a thread-safe implementation of xoroshiro128** described below:
5029 * http://prng.di.unimi.it/
5030 * It features a 2^128 long sequence, returns 64 high-quality bits on each call,
5031 * supports fast jumps and passes all common quality tests. It is thread-safe,
5032 * uses a double-cas on 64-bit architectures supporting it, and falls back to a
5033 * local lock on other ones.
5034 */
5035uint64_t ha_random64()
5036{
Willy Tarreau1544c142020-03-12 00:31:18 +01005037 uint64_t old[2] ALIGNED(2*sizeof(uint64_t));
5038 uint64_t new[2] ALIGNED(2*sizeof(uint64_t));
Willy Tarreau52bf8392020-03-08 00:42:37 +01005039
5040#if defined(USE_THREAD) && (!defined(HA_CAS_IS_8B) || !defined(HA_HAVE_CAS_DW))
5041 static HA_SPINLOCK_T rand_lock;
5042
5043 HA_SPIN_LOCK(OTHER_LOCK, &rand_lock);
5044#endif
5045
5046 old[0] = ha_random_state[0];
5047 old[1] = ha_random_state[1];
5048
5049#if defined(USE_THREAD) && defined(HA_CAS_IS_8B) && defined(HA_HAVE_CAS_DW)
5050 do {
5051#endif
Willy Tarreau52bf8392020-03-08 00:42:37 +01005052 new[1] = old[0] ^ old[1];
5053 new[0] = rotl64(old[0], 24) ^ new[1] ^ (new[1] << 16); // a, b
5054 new[1] = rotl64(new[1], 37); // c
5055
5056#if defined(USE_THREAD) && defined(HA_CAS_IS_8B) && defined(HA_HAVE_CAS_DW)
5057 } while (unlikely(!_HA_ATOMIC_DWCAS(ha_random_state, old, new)));
5058#else
5059 ha_random_state[0] = new[0];
5060 ha_random_state[1] = new[1];
5061#if defined(USE_THREAD)
5062 HA_SPIN_UNLOCK(OTHER_LOCK, &rand_lock);
5063#endif
5064#endif
Willy Tarreaub2475a12021-05-09 10:26:14 +02005065 return rotl64(old[0] * 5, 7) * 9;
Willy Tarreau52bf8392020-03-08 00:42:37 +01005066}
5067
5068/* seeds the random state using up to <len> bytes from <seed>, starting with
5069 * the first non-zero byte.
5070 */
5071void ha_random_seed(const unsigned char *seed, size_t len)
5072{
5073 size_t pos;
5074
5075 /* the seed must not be all zeroes, so we pre-fill it with alternating
5076 * bits and overwrite part of them with the block starting at the first
5077 * non-zero byte from the seed.
5078 */
5079 memset(ha_random_state, 0x55, sizeof(ha_random_state));
5080
5081 for (pos = 0; pos < len; pos++)
5082 if (seed[pos] != 0)
5083 break;
5084
5085 if (pos == len)
5086 return;
5087
5088 seed += pos;
5089 len -= pos;
5090
5091 if (len > sizeof(ha_random_state))
5092 len = sizeof(ha_random_state);
5093
5094 memcpy(ha_random_state, seed, len);
5095}
5096
5097/* This causes a jump to (dist * 2^96) places in the pseudo-random sequence,
5098 * and is equivalent to calling ha_random64() as many times. It is used to
5099 * provide non-overlapping sequences of 2^96 numbers (~7*10^28) to up to 2^32
5100 * different generators (i.e. different processes after a fork). The <dist>
5101 * argument is the distance to jump to and is used in a loop so it rather not
5102 * be too large if the processing time is a concern.
5103 *
5104 * BEWARE: this function is NOT thread-safe and must not be called during
5105 * concurrent accesses to ha_random64().
5106 */
5107void ha_random_jump96(uint32_t dist)
5108{
5109 while (dist--) {
5110 uint64_t s0 = 0;
5111 uint64_t s1 = 0;
5112 int b;
5113
5114 for (b = 0; b < 64; b++) {
5115 if ((0xd2a98b26625eee7bULL >> b) & 1) {
5116 s0 ^= ha_random_state[0];
5117 s1 ^= ha_random_state[1];
5118 }
5119 ha_random64();
5120 }
5121
5122 for (b = 0; b < 64; b++) {
5123 if ((0xdddf9b1090aa7ac1ULL >> b) & 1) {
5124 s0 ^= ha_random_state[0];
5125 s1 ^= ha_random_state[1];
5126 }
5127 ha_random64();
5128 }
5129 ha_random_state[0] = s0;
5130 ha_random_state[1] = s1;
5131 }
5132}
5133
Willy Tarreauee3bcdd2020-03-08 17:48:17 +01005134/* Generates an RFC4122 UUID into chunk <output> which must be at least 37
5135 * bytes large.
5136 */
5137void ha_generate_uuid(struct buffer *output)
5138{
5139 uint32_t rnd[4];
5140 uint64_t last;
5141
5142 last = ha_random64();
5143 rnd[0] = last;
5144 rnd[1] = last >> 32;
5145
5146 last = ha_random64();
5147 rnd[2] = last;
5148 rnd[3] = last >> 32;
5149
5150 chunk_printf(output, "%8.8x-%4.4x-%4.4x-%4.4x-%12.12llx",
5151 rnd[0],
5152 rnd[1] & 0xFFFF,
5153 ((rnd[1] >> 16u) & 0xFFF) | 0x4000, // highest 4 bits indicate the uuid version
5154 (rnd[2] & 0x3FFF) | 0x8000, // the highest 2 bits indicate the UUID variant (10),
5155 (long long)((rnd[2] >> 14u) | ((uint64_t) rnd[3] << 18u)) & 0xFFFFFFFFFFFFull);
5156}
5157
5158
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005159/* only used by parse_line() below. It supports writing in place provided that
5160 * <in> is updated to the next location before calling it. In that case, the
5161 * char at <in> may be overwritten.
5162 */
5163#define EMIT_CHAR(x) \
5164 do { \
5165 char __c = (char)(x); \
5166 if ((opts & PARSE_OPT_INPLACE) && out+outpos > in) \
5167 err |= PARSE_ERR_OVERLAP; \
5168 if (outpos >= outmax) \
5169 err |= PARSE_ERR_TOOLARGE; \
5170 if (!err) \
5171 out[outpos] = __c; \
5172 outpos++; \
5173 } while (0)
5174
Ilya Shipitsin46a030c2020-07-05 16:36:08 +05005175/* Parse <in>, copy it into <out> split into isolated words whose pointers
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005176 * are put in <args>. If more than <outlen> bytes have to be emitted, the
5177 * extraneous ones are not emitted but <outlen> is updated so that the caller
5178 * knows how much to realloc. Similarly, <args> are not updated beyond <nbargs>
5179 * but the returned <nbargs> indicates how many were found. All trailing args
Willy Tarreau61dd44b2020-06-25 07:35:42 +02005180 * up to <nbargs> point to the trailing zero, and as long as <nbargs> is > 0,
5181 * it is guaranteed that at least one arg will point to the zero. It is safe
5182 * to call it with a NULL <args> if <nbargs> is 0.
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005183 *
5184 * <out> may overlap with <in> provided that it never goes further, in which
5185 * case the parser will accept to perform in-place parsing and unquoting/
5186 * unescaping but only if environment variables do not lead to expansion that
5187 * causes overlapping, otherwise the input string being destroyed, the error
5188 * will not be recoverable. Note that even during out-of-place <in> will
5189 * experience temporary modifications in-place for variable resolution and must
5190 * be writable, and will also receive zeroes to delimit words when using
5191 * in-place copy. Parsing options <opts> taken from PARSE_OPT_*. Return value
5192 * is zero on success otherwise a bitwise-or of PARSE_ERR_*. Upon error, the
5193 * starting point of the first invalid character sequence or unmatched
5194 * quote/brace is reported in <errptr> if not NULL. When using in-place parsing
5195 * error reporting might be difficult since zeroes will have been inserted into
5196 * the string. One solution for the caller may consist in replacing all args
5197 * delimiters with spaces in this case.
5198 */
5199uint32_t parse_line(char *in, char *out, size_t *outlen, char **args, int *nbargs, uint32_t opts, char **errptr)
5200{
5201 char *quote = NULL;
5202 char *brace = NULL;
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005203 char *word_expand = NULL;
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005204 unsigned char hex1, hex2;
5205 size_t outmax = *outlen;
Willy Tarreau61dd44b2020-06-25 07:35:42 +02005206 int argsmax = *nbargs - 1;
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005207 size_t outpos = 0;
5208 int squote = 0;
5209 int dquote = 0;
5210 int arg = 0;
5211 uint32_t err = 0;
5212
5213 *nbargs = 0;
5214 *outlen = 0;
5215
Willy Tarreau61dd44b2020-06-25 07:35:42 +02005216 /* argsmax may be -1 here, protecting args[] from any write */
5217 if (arg < argsmax)
5218 args[arg] = out;
5219
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005220 while (1) {
5221 if (*in >= '-' && *in != '\\') {
5222 /* speedup: directly send all regular chars starting
5223 * with '-', '.', '/', alnum etc...
5224 */
5225 EMIT_CHAR(*in++);
5226 continue;
5227 }
5228 else if (*in == '\0' || *in == '\n' || *in == '\r') {
5229 /* end of line */
5230 break;
5231 }
5232 else if (*in == '#' && (opts & PARSE_OPT_SHARP) && !squote && !dquote) {
5233 /* comment */
5234 break;
5235 }
5236 else if (*in == '"' && !squote && (opts & PARSE_OPT_DQUOTE)) { /* double quote outside single quotes */
5237 if (dquote) {
5238 dquote = 0;
5239 quote = NULL;
5240 }
5241 else {
5242 dquote = 1;
5243 quote = in;
5244 }
5245 in++;
5246 continue;
5247 }
5248 else if (*in == '\'' && !dquote && (opts & PARSE_OPT_SQUOTE)) { /* single quote outside double quotes */
5249 if (squote) {
5250 squote = 0;
5251 quote = NULL;
5252 }
5253 else {
5254 squote = 1;
5255 quote = in;
5256 }
5257 in++;
5258 continue;
5259 }
5260 else if (*in == '\\' && !squote && (opts & PARSE_OPT_BKSLASH)) {
5261 /* first, we'll replace \\, \<space>, \#, \r, \n, \t, \xXX with their
5262 * C equivalent value but only when they have a special meaning and within
5263 * double quotes for some of them. Other combinations left unchanged (eg: \1).
5264 */
5265 char tosend = *in;
5266
5267 switch (in[1]) {
5268 case ' ':
5269 case '\\':
5270 tosend = in[1];
5271 in++;
5272 break;
5273
5274 case 't':
5275 tosend = '\t';
5276 in++;
5277 break;
5278
5279 case 'n':
5280 tosend = '\n';
5281 in++;
5282 break;
5283
5284 case 'r':
5285 tosend = '\r';
5286 in++;
5287 break;
5288
5289 case '#':
5290 /* escaping of "#" only if comments are supported */
5291 if (opts & PARSE_OPT_SHARP)
5292 in++;
5293 tosend = *in;
5294 break;
5295
5296 case '\'':
5297 /* escaping of "'" only outside single quotes and only if single quotes are supported */
5298 if (opts & PARSE_OPT_SQUOTE && !squote)
5299 in++;
5300 tosend = *in;
5301 break;
5302
5303 case '"':
5304 /* escaping of '"' only outside single quotes and only if double quotes are supported */
5305 if (opts & PARSE_OPT_DQUOTE && !squote)
5306 in++;
5307 tosend = *in;
5308 break;
5309
5310 case '$':
5311 /* escaping of '$' only inside double quotes and only if env supported */
5312 if (opts & PARSE_OPT_ENV && dquote)
5313 in++;
5314 tosend = *in;
5315 break;
5316
5317 case 'x':
5318 if (!ishex(in[2]) || !ishex(in[3])) {
5319 /* invalid or incomplete hex sequence */
5320 err |= PARSE_ERR_HEX;
5321 if (errptr)
5322 *errptr = in;
5323 goto leave;
5324 }
Willy Tarreauf278eec2020-07-05 21:46:32 +02005325 hex1 = toupper((unsigned char)in[2]) - '0';
5326 hex2 = toupper((unsigned char)in[3]) - '0';
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005327 if (hex1 > 9) hex1 -= 'A' - '9' - 1;
5328 if (hex2 > 9) hex2 -= 'A' - '9' - 1;
5329 tosend = (hex1 << 4) + hex2;
5330 in += 3;
5331 break;
5332
5333 default:
5334 /* other combinations are not escape sequences */
5335 break;
5336 }
5337
5338 in++;
5339 EMIT_CHAR(tosend);
5340 }
5341 else if (isspace((unsigned char)*in) && !squote && !dquote) {
5342 /* a non-escaped space is an argument separator */
5343 while (isspace((unsigned char)*in))
5344 in++;
5345 EMIT_CHAR(0);
5346 arg++;
5347 if (arg < argsmax)
5348 args[arg] = out + outpos;
5349 else
5350 err |= PARSE_ERR_TOOMANY;
5351 }
5352 else if (*in == '$' && (opts & PARSE_OPT_ENV) && (dquote || !(opts & PARSE_OPT_DQUOTE))) {
5353 /* environment variables are evaluated anywhere, or only
5354 * inside double quotes if they are supported.
5355 */
5356 char *var_name;
5357 char save_char;
Willy Tarreaua46f1af2021-05-06 10:25:11 +02005358 const char *value;
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005359
5360 in++;
5361
5362 if (*in == '{')
5363 brace = in++;
5364
Willy Tarreaua46f1af2021-05-06 10:25:11 +02005365 if (!isalpha((unsigned char)*in) && *in != '_' && *in != '.') {
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005366 /* unacceptable character in variable name */
5367 err |= PARSE_ERR_VARNAME;
5368 if (errptr)
5369 *errptr = in;
5370 goto leave;
5371 }
5372
5373 var_name = in;
Willy Tarreaua46f1af2021-05-06 10:25:11 +02005374 if (*in == '.')
5375 in++;
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005376 while (isalnum((unsigned char)*in) || *in == '_')
5377 in++;
5378
5379 save_char = *in;
5380 *in = '\0';
Willy Tarreaua46f1af2021-05-06 10:25:11 +02005381 if (unlikely(*var_name == '.')) {
5382 /* internal pseudo-variables */
5383 if (strcmp(var_name, ".LINE") == 0)
5384 value = ultoa(global.cfg_curr_line);
5385 else if (strcmp(var_name, ".FILE") == 0)
5386 value = global.cfg_curr_file;
5387 else if (strcmp(var_name, ".SECTION") == 0)
5388 value = global.cfg_curr_section;
5389 else {
5390 /* unsupported internal variable name */
5391 err |= PARSE_ERR_VARNAME;
5392 if (errptr)
5393 *errptr = var_name;
5394 goto leave;
5395 }
5396 } else {
5397 value = getenv(var_name);
5398 }
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005399 *in = save_char;
5400
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005401 /* support for '[*]' sequence to force word expansion,
5402 * only available inside braces */
5403 if (*in == '[' && brace && (opts & PARSE_OPT_WORD_EXPAND)) {
5404 word_expand = in++;
5405
5406 if (*in++ != '*' || *in++ != ']') {
5407 err |= PARSE_ERR_WRONG_EXPAND;
5408 if (errptr)
5409 *errptr = word_expand;
5410 goto leave;
5411 }
5412 }
5413
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005414 if (brace) {
5415 if (*in != '}') {
5416 /* unmatched brace */
5417 err |= PARSE_ERR_BRACE;
5418 if (errptr)
5419 *errptr = brace;
5420 goto leave;
5421 }
5422 in++;
5423 brace = NULL;
5424 }
5425
5426 if (value) {
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005427 while (*value) {
5428 /* expand as individual parameters on a space character */
Willy Tarreaufe2cc412020-10-01 18:04:40 +02005429 if (word_expand && isspace((unsigned char)*value)) {
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005430 EMIT_CHAR(0);
5431 ++arg;
5432 if (arg < argsmax)
5433 args[arg] = out + outpos;
5434 else
5435 err |= PARSE_ERR_TOOMANY;
5436
5437 /* skip consecutive spaces */
Willy Tarreaufe2cc412020-10-01 18:04:40 +02005438 while (isspace((unsigned char)*++value))
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005439 ;
5440 } else {
5441 EMIT_CHAR(*value++);
5442 }
5443 }
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005444 }
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005445 word_expand = NULL;
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005446 }
5447 else {
5448 /* any other regular char */
5449 EMIT_CHAR(*in++);
5450 }
5451 }
5452
5453 /* end of output string */
5454 EMIT_CHAR(0);
5455 arg++;
5456
5457 if (quote) {
5458 /* unmatched quote */
5459 err |= PARSE_ERR_QUOTE;
5460 if (errptr)
5461 *errptr = quote;
5462 goto leave;
5463 }
5464 leave:
5465 *nbargs = arg;
5466 *outlen = outpos;
5467
Willy Tarreau61dd44b2020-06-25 07:35:42 +02005468 /* empty all trailing args by making them point to the trailing zero,
5469 * at least the last one in any case.
5470 */
5471 if (arg > argsmax)
5472 arg = argsmax;
5473
5474 while (arg >= 0 && arg <= argsmax)
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005475 args[arg++] = out + outpos - 1;
5476
5477 return err;
5478}
5479#undef EMIT_CHAR
5480
Willy Tarreauc54e5ad2020-06-25 09:15:40 +02005481/* This is used to sanitize an input line that's about to be used for error reporting.
5482 * It will adjust <line> to print approximately <width> chars around <pos>, trying to
5483 * preserve the beginning, with leading or trailing "..." when the line is truncated.
5484 * If non-printable chars are present in the output. It returns the new offset <pos>
5485 * in the modified line. Non-printable characters are replaced with '?'. <width> must
5486 * be at least 6 to support two "..." otherwise the result is undefined. The line
5487 * itself must have at least 7 chars allocated for the same reason.
5488 */
5489size_t sanitize_for_printing(char *line, size_t pos, size_t width)
5490{
5491 size_t shift = 0;
5492 char *out = line;
5493 char *in = line;
5494 char *end = line + width;
5495
5496 if (pos >= width) {
5497 /* if we have to shift, we'll be out of context, so let's
5498 * try to put <pos> at the center of width.
5499 */
5500 shift = pos - width / 2;
5501 in += shift + 3;
5502 end = out + width - 3;
5503 out[0] = out[1] = out[2] = '.';
5504 out += 3;
5505 }
5506
5507 while (out < end && *in) {
5508 if (isspace((unsigned char)*in))
5509 *out++ = ' ';
5510 else if (isprint((unsigned char)*in))
5511 *out++ = *in;
5512 else
5513 *out++ = '?';
5514 in++;
5515 }
5516
5517 if (end < line + width) {
5518 out[0] = out[1] = out[2] = '.';
5519 out += 3;
5520 }
5521
5522 *out++ = 0;
5523 return pos - shift;
5524}
Willy Tarreau06e69b52021-03-02 14:01:35 +01005525
Willy Tarreaue33c4b32021-03-12 18:59:31 +01005526/* Update array <fp> with the fingerprint of word <word> by counting the
Willy Tarreauba2c4452021-03-12 09:01:52 +01005527 * transitions between characters. <fp> is a 1024-entries array indexed as
5528 * 32*from+to. Positions for 'from' and 'to' are:
Willy Tarreau9294e882021-03-15 09:34:27 +01005529 * 1..26=letter, 27=digit, 28=other/begin/end.
5530 * Row "from=0" is used to mark the character's presence. Others unused.
Willy Tarreauba2c4452021-03-12 09:01:52 +01005531 */
Willy Tarreaue33c4b32021-03-12 18:59:31 +01005532void update_word_fingerprint(uint8_t *fp, const char *word)
Willy Tarreauba2c4452021-03-12 09:01:52 +01005533{
5534 const char *p;
5535 int from, to;
5536 int c;
5537
Willy Tarreauba2c4452021-03-12 09:01:52 +01005538 from = 28; // begin
5539 for (p = word; *p; p++) {
5540 c = tolower(*p);
5541 switch(c) {
Willy Tarreau9294e882021-03-15 09:34:27 +01005542 case 'a'...'z': to = c - 'a' + 1; break;
5543 case 'A'...'Z': to = tolower(c) - 'a' + 1; break;
5544 case '0'...'9': to = 27; break;
5545 default: to = 28; break;
Willy Tarreauba2c4452021-03-12 09:01:52 +01005546 }
Willy Tarreau9294e882021-03-15 09:34:27 +01005547 fp[to] = 1;
Willy Tarreauba2c4452021-03-12 09:01:52 +01005548 fp[32 * from + to]++;
5549 from = to;
5550 }
5551 to = 28; // end
5552 fp[32 * from + to]++;
5553}
5554
Willy Tarreaue33c4b32021-03-12 18:59:31 +01005555/* Initialize array <fp> with the fingerprint of word <word> by counting the
5556 * transitions between characters. <fp> is a 1024-entries array indexed as
5557 * 32*from+to. Positions for 'from' and 'to' are:
5558 * 0..25=letter, 26=digit, 27=other, 28=begin, 29=end, others unused.
5559 */
5560void make_word_fingerprint(uint8_t *fp, const char *word)
5561{
5562 memset(fp, 0, 1024);
5563 update_word_fingerprint(fp, word);
5564}
5565
Willy Tarreauba2c4452021-03-12 09:01:52 +01005566/* Return the distance between two word fingerprints created by function
5567 * make_word_fingerprint(). It's a positive integer calculated as the sum of
Willy Tarreau714c4c12021-03-15 09:44:53 +01005568 * the differences between each location.
Willy Tarreauba2c4452021-03-12 09:01:52 +01005569 */
5570int word_fingerprint_distance(const uint8_t *fp1, const uint8_t *fp2)
5571{
5572 int i, k, dist = 0;
5573
5574 for (i = 0; i < 1024; i++) {
5575 k = (int)fp1[i] - (int)fp2[i];
Willy Tarreau714c4c12021-03-15 09:44:53 +01005576 dist += abs(k);
Willy Tarreauba2c4452021-03-12 09:01:52 +01005577 }
5578 return dist;
5579}
5580
Willy Tarreau06e69b52021-03-02 14:01:35 +01005581static int init_tools_per_thread()
5582{
5583 /* Let's make each thread start from a different position */
5584 statistical_prng_state += tid * MAX_THREADS;
5585 if (!statistical_prng_state)
5586 statistical_prng_state++;
5587 return 1;
5588}
5589REGISTER_PER_THREAD_INIT(init_tools_per_thread);
Willy Tarreauc54e5ad2020-06-25 09:15:40 +02005590
Willy Tarreaubaaee002006-06-26 02:48:02 +02005591/*
5592 * Local variables:
5593 * c-indent-level: 8
5594 * c-basic-offset: 8
5595 * End:
5596 */