2 * IRC - Internet Relay Chat, common/match.c
3 * Copyright (C) 1990 Jarkko Oikarinen
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 1, or (at your option)
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
24 #include "ircd_chattr.h"
25 #include "ircd_string.h"
26 #include "ircd_snprintf.h"
32 * Written by Run (carlo@runaway.xs4all.nl), 25-10-96
35 * From: Carlo Wood <carlo@runaway.xs4all.nl>
36 * Message-Id: <199609021026.MAA02393@runaway.xs4all.nl>
37 * Subject: [C-Com] Analysis for `mmatch' (was: gline4 problem)
38 * To: coder-com@mail.undernet.org (coder committee)
39 * Date: Mon, 2 Sep 1996 12:26:01 +0200 (MET DST)
41 * We need a new function `mmatch(const char *old_mask, const char *new_mask)'
42 * which returns `true' likewise the current `match' (start with copying it),
43 * but which treats '*' and '?' in `new_mask' differently (not "\*" and "\?" !)
44 * as follows: a '*' in `new_mask' does not match a '?' in `old_mask' and
45 * a '?' in `new_mask' does not match a '\?' in `old_mask'.
46 * And ofcourse... a '*' in `new_mask' does not match a '\*' in `old_mask'...
47 * And last but not least, '\?' and '\*' in `new_mask' now become one character.
50 int mmatch(const char *old_mask, const char *new_mask)
52 const char *m = old_mask;
53 const char *n = new_mask;
74 for (m--; (m > old_mask) && (*m == '?'); m--)
76 if ((*m == '*') && (m > old_mask) && (m[-1] != '\\'))
82 /* Added to `mmatch' : Because '\?' and '\*' now is one character: */
83 if ((*na == '\\') && ((na[1] == '*') || (na[1] == '?')))
94 if ((*m == '\\') && ((m[1] == '*') || (m[1] == '?')))
102 /* Added to `mmatch' : Because '\?' and '\*' now is one character: */
103 if ((*n == '\\') && ((n[1] == '*') || (n[1] == '?')))
112 * This `if' has been changed compared to match() to do the following:
114 * old (m) new (n) boolean expression
115 * * any (*m == '*' && !mq) ||
116 * ? any except '*' (*m == '?' && !mq && (*n != '*' || nq)) ||
117 * any except * or ? same as m (!((*m == '*' || *m == '?') && !mq) &&
118 * ToLower(*m) == ToLower(*n) &&
119 * !((mq && !nq) || (!mq && nq)))
121 * Here `any' also includes \* and \? !
123 * After reworking the boolean expressions, we get:
124 * (Optimized to use boolean shortcircuits, with most frequently occuring
125 * cases upfront (which took 2 hours!)).
127 if ((*m == '*' && !mq) ||
128 ((!mq || nq) && ToLower(*m) == ToLower(*n)) ||
129 (*m == '?' && !mq && (*n != '*' || nq)))
142 /* Added to `mmatch' : Because '\?' and '\*' now is one character: */
143 if ((*na == '\\') && ((na[1] == '*') || (na[1] == '?')))
152 * Compare if a given string (name) matches the given
153 * mask (which can contain wild cards: '*' - match any
154 * number of chars, '?' - match any single character.
159 * Originally by Douglas A Lewis (dalewis@acsu.buffalo.edu)
160 * Rewritten by Timothy Vogelsang (netski), net@astrolink.org
163 int match(const char *mask, const char *name)
165 const char *m = mask, *n = name;
166 const char *m_tmp = mask, *n_tmp = name;
172 while (*m == '*') /* clean up any additional wildcards */
179 if (*m == '\\') /* next wildcard is disregarded */
184 return 0; /* match */
186 for (m--; (m > mask) && (*m == '?'); m--);
189 if (*m == '*' && (m > mask))
190 return 0; /* match */
199 while (*m == '*') /* clean up any additional wildcards */
204 if (ToLower(*m) != ToLower(*n) && *m != '?') {
206 return 1; /* failure! */
219 return 1; /* no match! */
224 * Collapse a pattern string into minimal components.
225 * This particular version is "in place", so that it changes the pattern
226 * which is to be reduced to a "minimal" size.
228 * (C) Carlo Wood - 6 Oct 1998
229 * Speedup rewrite by Andrea Cocito, December 1998.
230 * Note that this new optimized alghoritm can *only* work in place.
233 char *collapse(char *mask)
243 if ((*m == '*') && ((m[1] == '*') || (m[1] == '?')))
252 if (star && (*m != '?'))
258 if ((*m == '\\') && ((m[1] == '*') || (m[1] == '?')))
267 if ((*m == '\\') && ((m[1] == '*') || (m[1] == '?')))
277 ***************** Nemesi's matchcomp() / matchexec() **************
280 /* These functions allow the use of "compiled" masks, you compile a mask
281 * by means of matchcomp() that gets the plain text mask as input and writes
282 * its result in the memory locations addressed by the 3 parameters:
283 * - *cmask will contain the text of the compiled mask
284 * - *minlen will contain the lenght of the shortest string that can match
286 * - *charset will contain the minimal set of chars needed to match the mask
287 * You can pass NULL as *charset and it will be simply not returned, but you
288 * MUST pass valid pointers for *minlen and *cmask (wich must be big enough
289 * to contain the compiled mask text that is in the worst case as long as the
290 * text of the mask itself in plaintext format) and the return value of
291 * matchcomp() will be the number of chars actually written there (excluded
292 * the trailing zero). cmask can be == mask, matchcomp() can work in place.
293 * The {cmask, minlen} couple of values make the real compiled mask and
294 * need to be passed to the functions that use the compiled mask, if you pass
295 * the wrong minlen or something wrong in cmask to one of these expect a
296 * coredump. This means that when you record a compiled mask you must store
297 * *both* these values.
298 * Once compiled the mask can be used to match a string by means of
299 * matchexec(), it can be printed back to human-readable format by means
300 * of sprintmatch() or it can be compared to another compiled mask by means
301 * of mmexec() that will tell if it completely overrides that mask (a lot like
302 * what mmatch() does for plain text masks).
303 * You can gain a lot of speed in many situations avoiding to matchexec() when:
304 * - The maximum lenght of the field you are about to match() the mask to is
305 * shorter than minlen, in example when matching abc*def*ghil with a nick:
306 * It just cannot match since a nick is at most 9 chars long and the mask
307 * needs at least 10 chars (10 will be the value returned in minlen).
308 * - The charset allowed for the field you are about to match to doesn't
309 * "contain" the charset returned by matchcomp(), in example when you
310 * have *.* as mask it makes no sense to try to match it against a nick
311 * because, again, a nick can't contain a '.', you can check this with
312 * a simple (charset & NTL_IRCNK) in this case.
313 * - As a special case, since compiled masks are forced to lowercase,
314 * it would make no sense to use the NTL_LOWER and NTL_UPPER on a compiled
315 * mask, thus they are reused as follows: if the NTL_LOWER bit of charset
316 * is set it means that the mask contains only non-wilds chars (i.e. you can
317 * use strCasecmp() to match it or a direct hash lookup), if the NTL_UPPER
318 * bit is set it means that it contains only wild chars (and you can
319 * match it with strlen(field)>=minlen).
320 * Do these optimizations ONLY when the data you are about to pass to
321 * matchexec() are *known* to be invalid in advance, using strChattr()
322 * or strlen() on the text would be slower than calling matchexec() directly
324 * Internally a compiled mask contain in the *cmask area the text of
325 * the plain text form of the mask itself with applied the following hacks:
326 * - All characters are forced to lowercase (so that uppercase letters and
327 * specifically the symbols 'A' and 'Z' are reserved for special use)
328 * - All non-escaped stars '*' are replaced by the letter 'Z'
329 * - All non-escaped question marks '?' are replaced by the letter 'A'
330 * - All escape characters are removed, the wilds escaped by them are
331 * then passed by without the escape since they don't collide anymore
332 * with the real wilds (encoded as A/Z)
333 * - Finally the part of the mask that follows the last asterisk is
334 * reversed (byte order mirroring) and moved right after the first
336 * After all this a mask like: Head*CHUNK1*chu\*nK2*ch??k3*TaIl
337 * .... becomes: headZliatZchunk1Zchu*nk2ZchAAk3
338 * This can still be printed on a console, more or less understood by an
339 * human and handled with the usual str*() library functions.
340 * When you store somewhere the compiled mask you can avoid storing the
341 * textform of it since it can be "decompiled" by means of sprintmatch(),
342 * but at that time the following things are changed in the mask:
343 * - All chars have been forced to lowercase.
344 * - The mask is collapsed.
345 * The balance point of using compiled masks in terms of CPU is when you expect
346 * to use matchexec() instead of match() at least 20 times on the same mask
347 * or when you expect to use mmexec() instead of mmatch() 3 times.
353 * Compiles a mask into a form suitable for using in matchexec().
356 int matchcomp(char *cmask, int *minlen, int *charset, const char *mask)
358 const char *m = mask;
363 int l1, l2, lmin, loop, sign;
368 int chset2 = (NTL_LOWER | NTL_UPPER);
380 chset2 &= ~NTL_LOWER;
383 if ((*m == '?') || (*m == '*'))
391 chset2 &= ~NTL_LOWER;
396 chset &= IRCD_CharAttrTab[*b++ - CHAR_MIN];
397 chset2 &= ~NTL_UPPER;
401 *charset = (chset | chset2);
412 for (x1 = ls + 1, x2 = (b - 1); x1 < x2; x1++, x2--)
421 while ((lmin = (l1 < l2) ? l1 : l2))
424 for (loop = 0; loop < lmin; loop++)
432 l1 -= (sign < 0) ? 0 : lmin;
433 l2 -= (sign > 0) ? 0 : lmin;
446 * Executes a match with a mask previosuly compiled with matchcomp()
447 * Note 1: If the mask isn't correctly produced by matchcomp() I will core
448 * Note 2: 'min' MUST be the value returned by matchcomp on that mask,
449 * or.... I will core even faster :-)
450 * Note 3: This piece of code is not intended to be nice but efficient.
453 int matchexec(const char *string, const char *cmask, int minlen)
455 const char *s = string - 1;
456 const char *b = cmask - 1;
462 while ((ToLower(*++s) == *++b) && *s);
464 return ((*b != '\0') && ((*b++ != 'Z') || (*b != '\0')));
475 if ((trash = (s - string - minlen)) < 0)
479 while ((ToLower(*--s) == *++b) && *b && (ToLower(*--s) == *++b) && *b
480 && (ToLower(*--s) == *++b) && *b && (ToLower(*--s) == *++b) && *b);
493 while ((ToLower(*++s) != ch))
499 while ((ToLower(*++s) == *++b) && *b);
522 * Prints the human readable version of *cmask into *mask, (decompiles
524 * The area pointed by *mask MUST be big enough (the mask might be up to
525 * twice the size of its compiled form if it's made all of \? or \*, and
526 * this function can NOT work in place since it might enflate the mask)
527 * The printed mask is not identical to the one that was compiled to cmask,
528 * infact it is 1) forced to all lowercase, 2) collapsed, both things
529 * are supposed to NOT change it's meaning.
530 * It returns the number of chars actually written to *mask;
533 int matchdecomp(char *mask, const char *cmask)
536 const char *rcm = cmask;
537 const char *begtail, *endtail;
545 for (; (*rcm != 'Z'); rcm++, rtb++)
547 if ((*rcm == '?') || (*rcm == '*'))
549 if (!((*rtb = ((*rcm == 'A') ? '?' : *rcm))))
556 while (*rcm && (*rcm != 'Z'))
581 for (rcm = endtail; (--rcm) > begtail; *rtb++ = ((*rcm == 'A') ? '?' : *rcm))
582 if ((*rcm == '?') || (*rcm == '*'))
591 * Checks if a wider compiled mask (wcm/wminlen) completely overrides
592 * a more restrict one (rcm/rminlen), basically what mmatch() does for
593 * non-compiled masks, returns 0 if the override is true (like mmatch()).
594 * "the wider overrides the restrict" means that any string that matches
595 * the restrict one _will_ also match the wider one, always.
596 * In this we behave differently from mmatch() because in example we return
597 * true for " a?*cd overrides a*bcd " for wich the override happens for how
598 * we literally defined it, here mmatch() would have returned false.
599 * The original concepts and the base alghoritm are copied from mmatch()
600 * written by Run (Carlo Wood), this function is written by
601 * Nemesi (Andrea Cocito)
604 int mmexec(const char *wcm, int wminlen, const char *rcm, int rminlen)
606 const char *w, *r, *br, *bw, *rx, *rz;
609 /* First of all rm must have enough non-stars to 'contain' wm */
610 if ((trash = rminlen - wminlen) < 0)
616 /* Let's start the game, remember that '*' is mapped to 'Z', '?'
617 is mapped to 'A' and that head?*??*?chunk*???*tail becomes
618 headAAAAZliatAAAZchunk for compiled masks */
620 /* Match the head of wm with the head of rm */
621 for (; (*r) && (*r != 'Z') && ((*w == *r) || (*w == 'A')); r++, w++);
623 while (*w == 'A') /* Eat extra '?' before '*' in wm if got '*' in rm */
625 if (*w != 'Z') /* head1<any>.. can't match head2<any>.. */
626 return ((*w) || (*r)) ? 1 : 0; /* and head<nul> matches only head<nul> */
628 return 0; /* headZ<nul> matches head<anything> */
630 /* Does rm have any stars in it ? let's check */
631 for (rx = r; *r && (*r != 'Z'); r++);
634 /* rm has no stars and thus isn't a mask but it's just a flat
635 string: special handling occurs here, note that eat must be 0 here */
640 for (; r--, (*w) && ((*w == *r) || (*w == 'A')); w++);
641 if (*w != 'Z') /* headZliat1<any> fails on head<any>2tail */
642 return (*w) ? 1 : 0; /* but headZliat<nul> matches head<any>tail */
645 /* match the chunks */
647 { /* This loop can't break but only return */
649 for (bw = w++; (*w != *rx); rx++) /* Seek the 1st char of the chunk */
650 if (--trash < 0) /* See if we can trash one more char of rm */
651 return 1; /* If not we can only fail of course */
652 for (r = ++rx, w++; (*w) && ((*w == *r) || (*w == 'A')); r++, w++);
653 if (!*w) /* Did last loop match the rest of chunk ? */
654 return 0; /* ... Yes, end of wm, matched ! */
656 { /* ... No, hitted non-star */
657 w = bw; /* Rollback at beginning of chunk */
658 if (--trash < 0) /* Trashed the char where this try started */
659 return 1; /* if we can't trash more chars fail */
663 rx = r; /* Successfully matched a chunk, move rx */
664 } /* and go on with the next one */
668 /* rm has at least one '*' and thus is a 'real' mask */
669 rz = r++; /* rx = unused of head, rz = beg-tail */
671 /* Match the tail of wm (if any) against the tail of rm */
674 for (; (*w) && (*r != 'Z') && ((*w == *r) || (*w == 'A')); w++, r++);
675 if (*r == 'Z') /* extra '?' before tail are fluff, just flush 'em */
678 if (*w != 'Z') /* We aren't matching a chunk, can't rollback */
682 /* Match the chunks of wm against what remains of the head of rm */
686 for (bw++; (rx < rz) && (*bw != *rx); rx++) /* Seek the first */
687 if (--trash < 0) /* waste some trash reserve */
689 if (!(rx < rz)) /* head finished */
691 for (bw++, (br = ++rx);
692 (br < rz) && (*bw) && ((*bw == *br) || (*bw == 'A')); br++, bw++);
693 if (!(br < rz)) /* Note that we didn't use any 'eat' char yet, if */
694 while (*bw == 'A') /* there were eat-en chars the head would be over */
695 bw++, eat++; /* Happens only at end of head, and eat is still 0 */
702 { /* If we failed because we got the end of head */
703 trash -= (br - rx); /* it makes no sense to rollback, just trash */
704 if (--trash < 0) /* all the rest of the head wich isn't long */
705 return 1; /* enough for this chunk and go out of this */
706 break; /* loop, then we try with the chunks of rm */
718 /* Match the unused chunks of wm against the chunks of rm */
720 for (; *r && (*r != 'Z'); r++);
727 while (eat && *r) /* the '?' we had eated make us skip as many chars */
728 if (*r++ != 'Z') /* here, but can't skip stars or trailing zero */
730 for (bw++; (*r) && (*bw != *r); r++)
731 if ((*r != 'Z') && (--trash < 0))
735 for ((br = ++r), bw++;
736 (*br) && (*br != 'Z') && ((*bw == *br) || (*bw == 'A')); br++, bw++);
745 if ((!*br) || (*r == 'Z'))
746 { /* If we hit the end of rm or a star in it */
747 trash -= (br - r); /* makes no sense to rollback within this */
748 if (trash < 0) /* same chunk of br, skip it all and then */
749 return 1; /* either rollback or break this loop if */
750 if (!*br) /* it was the end of rm */
765 /* match the remaining chunks of wm against what remains of the tail of rm */
766 r = rz - eat - 1; /* can't have <nul> or 'Z'within the tail, so just move r */
770 for (bw++; (*bw != *r); r--)
775 for ((br = --r), bw++;
776 (*bw) && (br >= rx) && ((*bw == *br) || (*bw == 'A')); br--, bw++);
792 return 1; /* Auch... something left out ? Fail */
798 #include <sys/socket.h>
799 #include <netinet/in.h>
800 #include <arpa/inet.h>
802 static int ipmask_parse_ipv4(const char *in, struct in_addr *out)
809 class = sscanf(in, "%d.%d.%d.%d/%d", &ad[0], &ad[1], &ad[2], &ad[3], &bits);
812 ircd_snprintf(0, ipname, sizeof(ipname), "%d.%d.%d.%d", ad[0], ad[1], ad[2], ad[3]);
813 out->s_addr = inet_addr(ipname);
817 int ipmask_parse(const char *in, struct irc_in_addr *mask, unsigned char *bits_ptr)
823 if (check_if_ipmask(in)) {
824 bits = ipmask_parse_ipv4(in, &ipv4);
825 mask->in6_16[0] = mask->in6_16[1] = mask->in6_16[2] = 0;
826 mask->in6_16[3] = mask->in6_16[4] = mask->in6_16[5] = 0;
827 memcpy(&mask->in6_16[6], &ipv4.s_addr, sizeof(ipv4.s_addr));
830 if (!(p = strchr(in, '/')))
834 if (!ircd_aton(mask, in)) {
850 int ipmask_check(const struct irc_in_addr *addr, const struct irc_in_addr *mask, unsigned char bits)
854 for (k = 0; k < 8; k++) {
856 return (addr->in6_16[k] & ((unsigned char) (0xffff << (16-bits)))) == mask->in6_16[k];
857 if (addr->in6_16[k] != mask->in6_16[k])