1/************************************************* 2* Perl-Compatible Regular Expressions * 3*************************************************/ 4 5/* PCRE is a library of functions to support regular expressions whose syntax 6and semantics are as close as possible to those of the Perl 5 language. 7 8 Written by Philip Hazel 9 Copyright (c) 1997-2010 University of Cambridge 10 11----------------------------------------------------------------------------- 12Redistribution and use in source and binary forms, with or without 13modification, are permitted provided that the following conditions are met: 14 15 * Redistributions of source code must retain the above copyright notice, 16 this list of conditions and the following disclaimer. 17 18 * Redistributions in binary form must reproduce the above copyright 19 notice, this list of conditions and the following disclaimer in the 20 documentation and/or other materials provided with the distribution. 21 22 * Neither the name of the University of Cambridge nor the names of its 23 contributors may be used to endorse or promote products derived from 24 this software without specific prior written permission. 25 26THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" 27AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE 30LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 31CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 32SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 33INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 34CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 35ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 36POSSIBILITY OF SUCH DAMAGE. 37----------------------------------------------------------------------------- 38*/ 39 40 41/* This module contains the external function pcre_study(), along with local 42supporting functions. */ 43 44 45#ifdef HAVE_CONFIG_H 46#include "config.h" 47#endif 48 49#include "pcre_internal.h" 50 51 52/* Returns from set_start_bits() */ 53 54enum { SSB_FAIL, SSB_DONE, SSB_CONTINUE }; 55 56 57 58/************************************************* 59* Find the minimum subject length for a group * 60*************************************************/ 61 62/* Scan a parenthesized group and compute the minimum length of subject that 63is needed to match it. This is a lower bound; it does not mean there is a 64string of that length that matches. In UTF8 mode, the result is in characters 65rather than bytes. 66 67Arguments: 68 code pointer to start of group (the bracket) 69 startcode pointer to start of the whole pattern 70 options the compiling options 71 72Returns: the minimum length 73 -1 if \C was encountered 74 -2 internal error (missing capturing bracket) 75*/ 76 77static int 78find_minlength(const uschar *code, const uschar *startcode, int options) 79{ 80int length = -1; 81BOOL utf8 = (options & PCRE_UTF8) != 0; 82BOOL had_recurse = FALSE; 83register int branchlength = 0; 84register uschar *cc = (uschar *)code + 1 + LINK_SIZE; 85 86if (*code == OP_CBRA || *code == OP_SCBRA) cc += 2; 87 88/* Scan along the opcodes for this branch. If we get to the end of the 89branch, check the length against that of the other branches. */ 90 91for (;;) 92 { 93 int d, min; 94 uschar *cs, *ce; 95 register int op = *cc; 96 97 switch (op) 98 { 99 case OP_COND: 100 case OP_SCOND: 101 102 /* If there is only one branch in a condition, the implied branch has zero 103 length, so we don't add anything. This covers the DEFINE "condition" 104 automatically. */ 105 106 cs = cc + GET(cc, 1); 107 if (*cs != OP_ALT) 108 { 109 cc = cs + 1 + LINK_SIZE; 110 break; 111 } 112 113 /* Otherwise we can fall through and treat it the same as any other 114 subpattern. */ 115 116 case OP_CBRA: 117 case OP_SCBRA: 118 case OP_BRA: 119 case OP_SBRA: 120 case OP_ONCE: 121 d = find_minlength(cc, startcode, options); 122 if (d < 0) return d; 123 branchlength += d; 124 do cc += GET(cc, 1); while (*cc == OP_ALT); 125 cc += 1 + LINK_SIZE; 126 break; 127 128 /* Reached end of a branch; if it's a ket it is the end of a nested 129 call. If it's ALT it is an alternation in a nested call. If it is 130 END it's the end of the outer call. All can be handled by the same code. */ 131 132 case OP_ALT: 133 case OP_KET: 134 case OP_KETRMAX: 135 case OP_KETRMIN: 136 case OP_END: 137 if (length < 0 || (!had_recurse && branchlength < length)) 138 length = branchlength; 139 if (*cc != OP_ALT) return length; 140 cc += 1 + LINK_SIZE; 141 branchlength = 0; 142 had_recurse = FALSE; 143 break; 144 145 /* Skip over assertive subpatterns */ 146 147 case OP_ASSERT: 148 case OP_ASSERT_NOT: 149 case OP_ASSERTBACK: 150 case OP_ASSERTBACK_NOT: 151 do cc += GET(cc, 1); while (*cc == OP_ALT); 152 /* Fall through */ 153 154 /* Skip over things that don't match chars */ 155 156 case OP_REVERSE: 157 case OP_CREF: 158 case OP_NCREF: 159 case OP_RREF: 160 case OP_NRREF: 161 case OP_DEF: 162 case OP_OPT: 163 case OP_CALLOUT: 164 case OP_SOD: 165 case OP_SOM: 166 case OP_EOD: 167 case OP_EODN: 168 case OP_CIRC: 169 case OP_DOLL: 170 case OP_NOT_WORD_BOUNDARY: 171 case OP_WORD_BOUNDARY: 172 cc += _pcre_OP_lengths[*cc]; 173 break; 174 175 /* Skip over a subpattern that has a {0} or {0,x} quantifier */ 176 177 case OP_BRAZERO: 178 case OP_BRAMINZERO: 179 case OP_SKIPZERO: 180 cc += _pcre_OP_lengths[*cc]; 181 do cc += GET(cc, 1); while (*cc == OP_ALT); 182 cc += 1 + LINK_SIZE; 183 break; 184 185 /* Handle literal characters and + repetitions */ 186 187 case OP_CHAR: 188 case OP_CHARNC: 189 case OP_NOT: 190 case OP_PLUS: 191 case OP_MINPLUS: 192 case OP_POSPLUS: 193 case OP_NOTPLUS: 194 case OP_NOTMINPLUS: 195 case OP_NOTPOSPLUS: 196 branchlength++; 197 cc += 2; 198#ifdef SUPPORT_UTF8 199 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f]; 200#endif 201 break; 202 203 case OP_TYPEPLUS: 204 case OP_TYPEMINPLUS: 205 case OP_TYPEPOSPLUS: 206 branchlength++; 207 cc += (cc[1] == OP_PROP || cc[1] == OP_NOTPROP)? 4 : 2; 208 break; 209 210 /* Handle exact repetitions. The count is already in characters, but we 211 need to skip over a multibyte character in UTF8 mode. */ 212 213 case OP_EXACT: 214 case OP_NOTEXACT: 215 branchlength += GET2(cc,1); 216 cc += 4; 217#ifdef SUPPORT_UTF8 218 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f]; 219#endif 220 break; 221 222 case OP_TYPEEXACT: 223 branchlength += GET2(cc,1); 224 cc += (cc[3] == OP_PROP || cc[3] == OP_NOTPROP)? 6 : 4; 225 break; 226 227 /* Handle single-char non-literal matchers */ 228 229 case OP_PROP: 230 case OP_NOTPROP: 231 cc += 2; 232 /* Fall through */ 233 234 case OP_NOT_DIGIT: 235 case OP_DIGIT: 236 case OP_NOT_WHITESPACE: 237 case OP_WHITESPACE: 238 case OP_NOT_WORDCHAR: 239 case OP_WORDCHAR: 240 case OP_ANY: 241 case OP_ALLANY: 242 case OP_EXTUNI: 243 case OP_HSPACE: 244 case OP_NOT_HSPACE: 245 case OP_VSPACE: 246 case OP_NOT_VSPACE: 247 branchlength++; 248 cc++; 249 break; 250 251 /* "Any newline" might match two characters */ 252 253 case OP_ANYNL: 254 branchlength += 2; 255 cc++; 256 break; 257 258 /* The single-byte matcher means we can't proceed in UTF-8 mode */ 259 260 case OP_ANYBYTE: 261#ifdef SUPPORT_UTF8 262 if (utf8) return -1; 263#endif 264 branchlength++; 265 cc++; 266 break; 267 268 /* For repeated character types, we have to test for \p and \P, which have 269 an extra two bytes of parameters. */ 270 271 case OP_TYPESTAR: 272 case OP_TYPEMINSTAR: 273 case OP_TYPEQUERY: 274 case OP_TYPEMINQUERY: 275 case OP_TYPEPOSSTAR: 276 case OP_TYPEPOSQUERY: 277 if (cc[1] == OP_PROP || cc[1] == OP_NOTPROP) cc += 2; 278 cc += _pcre_OP_lengths[op]; 279 break; 280 281 case OP_TYPEUPTO: 282 case OP_TYPEMINUPTO: 283 case OP_TYPEPOSUPTO: 284 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2; 285 cc += _pcre_OP_lengths[op]; 286 break; 287 288 /* Check a class for variable quantification */ 289 290#ifdef SUPPORT_UTF8 291 case OP_XCLASS: 292 cc += GET(cc, 1) - 33; 293 /* Fall through */ 294#endif 295 296 case OP_CLASS: 297 case OP_NCLASS: 298 cc += 33; 299 300 switch (*cc) 301 { 302 case OP_CRPLUS: 303 case OP_CRMINPLUS: 304 branchlength++; 305 /* Fall through */ 306 307 case OP_CRSTAR: 308 case OP_CRMINSTAR: 309 case OP_CRQUERY: 310 case OP_CRMINQUERY: 311 cc++; 312 break; 313 314 case OP_CRRANGE: 315 case OP_CRMINRANGE: 316 branchlength += GET2(cc,1); 317 cc += 5; 318 break; 319 320 default: 321 branchlength++; 322 break; 323 } 324 break; 325 326 /* Backreferences and subroutine calls are treated in the same way: we find 327 the minimum length for the subpattern. A recursion, however, causes an 328 a flag to be set that causes the length of this branch to be ignored. The 329 logic is that a recursion can only make sense if there is another 330 alternation that stops the recursing. That will provide the minimum length 331 (when no recursion happens). A backreference within the group that it is 332 referencing behaves in the same way. 333 334 If PCRE_JAVASCRIPT_COMPAT is set, a backreference to an unset bracket 335 matches an empty string (by default it causes a matching failure), so in 336 that case we must set the minimum length to zero. */ 337 338 case OP_REF: 339 if ((options & PCRE_JAVASCRIPT_COMPAT) == 0) 340 { 341 ce = cs = (uschar *)_pcre_find_bracket(startcode, utf8, GET2(cc, 1)); 342 if (cs == NULL) return -2; 343 do ce += GET(ce, 1); while (*ce == OP_ALT); 344 if (cc > cs && cc < ce) 345 { 346 d = 0; 347 had_recurse = TRUE; 348 } 349 else d = find_minlength(cs, startcode, options); 350 } 351 else d = 0; 352 cc += 3; 353 354 /* Handle repeated back references */ 355 356 switch (*cc) 357 { 358 case OP_CRSTAR: 359 case OP_CRMINSTAR: 360 case OP_CRQUERY: 361 case OP_CRMINQUERY: 362 min = 0; 363 cc++; 364 break; 365 366 case OP_CRRANGE: 367 case OP_CRMINRANGE: 368 min = GET2(cc, 1); 369 cc += 5; 370 break; 371 372 default: 373 min = 1; 374 break; 375 } 376 377 branchlength += min * d; 378 break; 379 380 case OP_RECURSE: 381 cs = ce = (uschar *)startcode + GET(cc, 1); 382 if (cs == NULL) return -2; 383 do ce += GET(ce, 1); while (*ce == OP_ALT); 384 if (cc > cs && cc < ce) 385 had_recurse = TRUE; 386 else 387 branchlength += find_minlength(cs, startcode, options); 388 cc += 1 + LINK_SIZE; 389 break; 390 391 /* Anything else does not or need not match a character. We can get the 392 item's length from the table, but for those that can match zero occurrences 393 of a character, we must take special action for UTF-8 characters. */ 394 395 case OP_UPTO: 396 case OP_NOTUPTO: 397 case OP_MINUPTO: 398 case OP_NOTMINUPTO: 399 case OP_POSUPTO: 400 case OP_STAR: 401 case OP_MINSTAR: 402 case OP_NOTMINSTAR: 403 case OP_POSSTAR: 404 case OP_NOTPOSSTAR: 405 case OP_QUERY: 406 case OP_MINQUERY: 407 case OP_NOTMINQUERY: 408 case OP_POSQUERY: 409 case OP_NOTPOSQUERY: 410 cc += _pcre_OP_lengths[op]; 411#ifdef SUPPORT_UTF8 412 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f]; 413#endif 414 break; 415 416 /* For the record, these are the opcodes that are matched by "default": 417 OP_ACCEPT, OP_CLOSE, OP_COMMIT, OP_FAIL, OP_PRUNE, OP_SET_SOM, OP_SKIP, 418 OP_THEN. */ 419 420 default: 421 cc += _pcre_OP_lengths[op]; 422 break; 423 } 424 } 425/* Control never gets here */ 426} 427 428 429 430/************************************************* 431* Set a bit and maybe its alternate case * 432*************************************************/ 433 434/* Given a character, set its bit in the table, and also the bit for the other 435version of a letter if we are caseless. 436 437Arguments: 438 start_bits points to the bit map 439 c is the character 440 caseless the caseless flag 441 cd the block with char table pointers 442 443Returns: nothing 444*/ 445 446static void 447set_table_bit(uschar *start_bits, unsigned int c, BOOL caseless, 448 compile_data *cd) 449{ 450start_bits[c/8] |= (1 << (c&7)); 451if (caseless && (cd->ctypes[c] & ctype_letter) != 0) 452 start_bits[cd->fcc[c]/8] |= (1 << (cd->fcc[c]&7)); 453} 454 455 456 457/************************************************* 458* Create bitmap of starting bytes * 459*************************************************/ 460 461/* This function scans a compiled unanchored expression recursively and 462attempts to build a bitmap of the set of possible starting bytes. As time goes 463by, we may be able to get more clever at doing this. The SSB_CONTINUE return is 464useful for parenthesized groups in patterns such as (a*)b where the group 465provides some optional starting bytes but scanning must continue at the outer 466level to find at least one mandatory byte. At the outermost level, this 467function fails unless the result is SSB_DONE. 468 469Arguments: 470 code points to an expression 471 start_bits points to a 32-byte table, initialized to 0 472 caseless the current state of the caseless flag 473 utf8 TRUE if in UTF-8 mode 474 cd the block with char table pointers 475 476Returns: SSB_FAIL => Failed to find any starting bytes 477 SSB_DONE => Found mandatory starting bytes 478 SSB_CONTINUE => Found optional starting bytes 479*/ 480 481static int 482set_start_bits(const uschar *code, uschar *start_bits, BOOL caseless, 483 BOOL utf8, compile_data *cd) 484{ 485register int c; 486int yield = SSB_DONE; 487 488#if 0 489/* ========================================================================= */ 490/* The following comment and code was inserted in January 1999. In May 2006, 491when it was observed to cause compiler warnings about unused values, I took it 492out again. If anybody is still using OS/2, they will have to put it back 493manually. */ 494 495/* This next statement and the later reference to dummy are here in order to 496trick the optimizer of the IBM C compiler for OS/2 into generating correct 497code. Apparently IBM isn't going to fix the problem, and we would rather not 498disable optimization (in this module it actually makes a big difference, and 499the pcre module can use all the optimization it can get). */ 500 501volatile int dummy; 502/* ========================================================================= */ 503#endif 504 505do 506 { 507 const uschar *tcode = code + (((int)*code == OP_CBRA)? 3:1) + LINK_SIZE; 508 BOOL try_next = TRUE; 509 510 while (try_next) /* Loop for items in this branch */ 511 { 512 int rc; 513 switch(*tcode) 514 { 515 /* Fail if we reach something we don't understand */ 516 517 default: 518 return SSB_FAIL; 519 520 /* If we hit a bracket or a positive lookahead assertion, recurse to set 521 bits from within the subpattern. If it can't find anything, we have to 522 give up. If it finds some mandatory character(s), we are done for this 523 branch. Otherwise, carry on scanning after the subpattern. */ 524 525 case OP_BRA: 526 case OP_SBRA: 527 case OP_CBRA: 528 case OP_SCBRA: 529 case OP_ONCE: 530 case OP_ASSERT: 531 rc = set_start_bits(tcode, start_bits, caseless, utf8, cd); 532 if (rc == SSB_FAIL) return SSB_FAIL; 533 if (rc == SSB_DONE) try_next = FALSE; else 534 { 535 do tcode += GET(tcode, 1); while (*tcode == OP_ALT); 536 tcode += 1 + LINK_SIZE; 537 } 538 break; 539 540 /* If we hit ALT or KET, it means we haven't found anything mandatory in 541 this branch, though we might have found something optional. For ALT, we 542 continue with the next alternative, but we have to arrange that the final 543 result from subpattern is SSB_CONTINUE rather than SSB_DONE. For KET, 544 return SSB_CONTINUE: if this is the top level, that indicates failure, 545 but after a nested subpattern, it causes scanning to continue. */ 546 547 case OP_ALT: 548 yield = SSB_CONTINUE; 549 try_next = FALSE; 550 break; 551 552 case OP_KET: 553 case OP_KETRMAX: 554 case OP_KETRMIN: 555 return SSB_CONTINUE; 556 557 /* Skip over callout */ 558 559 case OP_CALLOUT: 560 tcode += 2 + 2*LINK_SIZE; 561 break; 562 563 /* Skip over lookbehind and negative lookahead assertions */ 564 565 case OP_ASSERT_NOT: 566 case OP_ASSERTBACK: 567 case OP_ASSERTBACK_NOT: 568 do tcode += GET(tcode, 1); while (*tcode == OP_ALT); 569 tcode += 1 + LINK_SIZE; 570 break; 571 572 /* Skip over an option setting, changing the caseless flag */ 573 574 case OP_OPT: 575 caseless = (tcode[1] & PCRE_CASELESS) != 0; 576 tcode += 2; 577 break; 578 579 /* BRAZERO does the bracket, but carries on. */ 580 581 case OP_BRAZERO: 582 case OP_BRAMINZERO: 583 if (set_start_bits(++tcode, start_bits, caseless, utf8, cd) == SSB_FAIL) 584 return SSB_FAIL; 585/* ========================================================================= 586 See the comment at the head of this function concerning the next line, 587 which was an old fudge for the benefit of OS/2. 588 dummy = 1; 589 ========================================================================= */ 590 do tcode += GET(tcode,1); while (*tcode == OP_ALT); 591 tcode += 1 + LINK_SIZE; 592 break; 593 594 /* SKIPZERO skips the bracket. */ 595 596 case OP_SKIPZERO: 597 tcode++; 598 do tcode += GET(tcode,1); while (*tcode == OP_ALT); 599 tcode += 1 + LINK_SIZE; 600 break; 601 602 /* Single-char * or ? sets the bit and tries the next item */ 603 604 case OP_STAR: 605 case OP_MINSTAR: 606 case OP_POSSTAR: 607 case OP_QUERY: 608 case OP_MINQUERY: 609 case OP_POSQUERY: 610 set_table_bit(start_bits, tcode[1], caseless, cd); 611 tcode += 2; 612#ifdef SUPPORT_UTF8 613 if (utf8 && tcode[-1] >= 0xc0) 614 tcode += _pcre_utf8_table4[tcode[-1] & 0x3f]; 615#endif 616 break; 617 618 /* Single-char upto sets the bit and tries the next */ 619 620 case OP_UPTO: 621 case OP_MINUPTO: 622 case OP_POSUPTO: 623 set_table_bit(start_bits, tcode[3], caseless, cd); 624 tcode += 4; 625#ifdef SUPPORT_UTF8 626 if (utf8 && tcode[-1] >= 0xc0) 627 tcode += _pcre_utf8_table4[tcode[-1] & 0x3f]; 628#endif 629 break; 630 631 /* At least one single char sets the bit and stops */ 632 633 case OP_EXACT: /* Fall through */ 634 tcode += 2; 635 636 case OP_CHAR: 637 case OP_CHARNC: 638 case OP_PLUS: 639 case OP_MINPLUS: 640 case OP_POSPLUS: 641 set_table_bit(start_bits, tcode[1], caseless, cd); 642 try_next = FALSE; 643 break; 644 645 /* Single character type sets the bits and stops */ 646 647 case OP_NOT_DIGIT: 648 for (c = 0; c < 32; c++) 649 start_bits[c] |= ~cd->cbits[c+cbit_digit]; 650 try_next = FALSE; 651 break; 652 653 case OP_DIGIT: 654 for (c = 0; c < 32; c++) 655 start_bits[c] |= cd->cbits[c+cbit_digit]; 656 try_next = FALSE; 657 break; 658 659 /* The cbit_space table has vertical tab as whitespace; we have to 660 discard it. */ 661 662 case OP_NOT_WHITESPACE: 663 for (c = 0; c < 32; c++) 664 { 665 int d = cd->cbits[c+cbit_space]; 666 if (c == 1) d &= ~0x08; 667 start_bits[c] |= ~d; 668 } 669 try_next = FALSE; 670 break; 671 672 /* The cbit_space table has vertical tab as whitespace; we have to 673 discard it. */ 674 675 case OP_WHITESPACE: 676 for (c = 0; c < 32; c++) 677 { 678 int d = cd->cbits[c+cbit_space]; 679 if (c == 1) d &= ~0x08; 680 start_bits[c] |= d; 681 } 682 try_next = FALSE; 683 break; 684 685 case OP_NOT_WORDCHAR: 686 for (c = 0; c < 32; c++) 687 start_bits[c] |= ~cd->cbits[c+cbit_word]; 688 try_next = FALSE; 689 break; 690 691 case OP_WORDCHAR: 692 for (c = 0; c < 32; c++) 693 start_bits[c] |= cd->cbits[c+cbit_word]; 694 try_next = FALSE; 695 break; 696 697 /* One or more character type fudges the pointer and restarts, knowing 698 it will hit a single character type and stop there. */ 699 700 case OP_TYPEPLUS: 701 case OP_TYPEMINPLUS: 702 tcode++; 703 break; 704 705 case OP_TYPEEXACT: 706 tcode += 3; 707 break; 708 709 /* Zero or more repeats of character types set the bits and then 710 try again. */ 711 712 case OP_TYPEUPTO: 713 case OP_TYPEMINUPTO: 714 case OP_TYPEPOSUPTO: 715 tcode += 2; /* Fall through */ 716 717 case OP_TYPESTAR: 718 case OP_TYPEMINSTAR: 719 case OP_TYPEPOSSTAR: 720 case OP_TYPEQUERY: 721 case OP_TYPEMINQUERY: 722 case OP_TYPEPOSQUERY: 723 switch(tcode[1]) 724 { 725 case OP_ANY: 726 case OP_ALLANY: 727 return SSB_FAIL; 728 729 case OP_NOT_DIGIT: 730 for (c = 0; c < 32; c++) 731 start_bits[c] |= ~cd->cbits[c+cbit_digit]; 732 break; 733 734 case OP_DIGIT: 735 for (c = 0; c < 32; c++) 736 start_bits[c] |= cd->cbits[c+cbit_digit]; 737 break; 738 739 /* The cbit_space table has vertical tab as whitespace; we have to 740 discard it. */ 741 742 case OP_NOT_WHITESPACE: 743 for (c = 0; c < 32; c++) 744 { 745 int d = cd->cbits[c+cbit_space]; 746 if (c == 1) d &= ~0x08; 747 start_bits[c] |= ~d; 748 } 749 break; 750 751 /* The cbit_space table has vertical tab as whitespace; we have to 752 discard it. */ 753 754 case OP_WHITESPACE: 755 for (c = 0; c < 32; c++) 756 { 757 int d = cd->cbits[c+cbit_space]; 758 if (c == 1) d &= ~0x08; 759 start_bits[c] |= d; 760 } 761 break; 762 763 case OP_NOT_WORDCHAR: 764 for (c = 0; c < 32; c++) 765 start_bits[c] |= ~cd->cbits[c+cbit_word]; 766 break; 767 768 case OP_WORDCHAR: 769 for (c = 0; c < 32; c++) 770 start_bits[c] |= cd->cbits[c+cbit_word]; 771 break; 772 } 773 774 tcode += 2; 775 break; 776 777 /* Character class where all the information is in a bit map: set the 778 bits and either carry on or not, according to the repeat count. If it was 779 a negative class, and we are operating with UTF-8 characters, any byte 780 with a value >= 0xc4 is a potentially valid starter because it starts a 781 character with a value > 255. */ 782 783 case OP_NCLASS: 784#ifdef SUPPORT_UTF8 785 if (utf8) 786 { 787 start_bits[24] |= 0xf0; /* Bits for 0xc4 - 0xc8 */ 788 memset(start_bits+25, 0xff, 7); /* Bits for 0xc9 - 0xff */ 789 } 790#endif 791 /* Fall through */ 792 793 case OP_CLASS: 794 { 795 tcode++; 796 797 /* In UTF-8 mode, the bits in a bit map correspond to character 798 values, not to byte values. However, the bit map we are constructing is 799 for byte values. So we have to do a conversion for characters whose 800 value is > 127. In fact, there are only two possible starting bytes for 801 characters in the range 128 - 255. */ 802 803#ifdef SUPPORT_UTF8 804 if (utf8) 805 { 806 for (c = 0; c < 16; c++) start_bits[c] |= tcode[c]; 807 for (c = 128; c < 256; c++) 808 { 809 if ((tcode[c/8] && (1 << (c&7))) != 0) 810 { 811 int d = (c >> 6) | 0xc0; /* Set bit for this starter */ 812 start_bits[d/8] |= (1 << (d&7)); /* and then skip on to the */ 813 c = (c & 0xc0) + 0x40 - 1; /* next relevant character. */ 814 } 815 } 816 } 817 818 /* In non-UTF-8 mode, the two bit maps are completely compatible. */ 819 820 else 821#endif 822 { 823 for (c = 0; c < 32; c++) start_bits[c] |= tcode[c]; 824 } 825 826 /* Advance past the bit map, and act on what follows */ 827 828 tcode += 32; 829 switch (*tcode) 830 { 831 case OP_CRSTAR: 832 case OP_CRMINSTAR: 833 case OP_CRQUERY: 834 case OP_CRMINQUERY: 835 tcode++; 836 break; 837 838 case OP_CRRANGE: 839 case OP_CRMINRANGE: 840 if (((tcode[1] << 8) + tcode[2]) == 0) tcode += 5; 841 else try_next = FALSE; 842 break; 843 844 default: 845 try_next = FALSE; 846 break; 847 } 848 } 849 break; /* End of bitmap class handling */ 850 851 } /* End of switch */ 852 } /* End of try_next loop */ 853 854 code += GET(code, 1); /* Advance to next branch */ 855 } 856while (*code == OP_ALT); 857return yield; 858} 859 860 861 862/************************************************* 863* Study a compiled expression * 864*************************************************/ 865 866/* This function is handed a compiled expression that it must study to produce 867information that will speed up the matching. It returns a pcre_extra block 868which then gets handed back to pcre_exec(). 869 870Arguments: 871 re points to the compiled expression 872 options contains option bits 873 errorptr points to where to place error messages; 874 set NULL unless error 875 876Returns: pointer to a pcre_extra block, with study_data filled in and the 877 appropriate flags set; 878 NULL on error or if no optimization possible 879*/ 880 881PCRE_EXP_DEFN pcre_extra * PCRE_CALL_CONVENTION 882pcre_study(const pcre *external_re, int options, const char **errorptr) 883{ 884int min; 885BOOL bits_set = FALSE; 886uschar start_bits[32]; 887pcre_extra *extra; 888pcre_study_data *study; 889const uschar *tables; 890uschar *code; 891compile_data compile_block; 892const real_pcre *re = (const real_pcre *)external_re; 893 894*errorptr = NULL; 895 896if (re == NULL || re->magic_number != MAGIC_NUMBER) 897 { 898 *errorptr = "argument is not a compiled regular expression"; 899 return NULL; 900 } 901 902if ((options & ~PUBLIC_STUDY_OPTIONS) != 0) 903 { 904 *errorptr = "unknown or incorrect option bit(s) set"; 905 return NULL; 906 } 907 908code = (uschar *)re + re->name_table_offset + 909 (re->name_count * re->name_entry_size); 910 911/* For an anchored pattern, or an unanchored pattern that has a first char, or 912a multiline pattern that matches only at "line starts", there is no point in 913seeking a list of starting bytes. */ 914 915if ((re->options & PCRE_ANCHORED) == 0 && 916 (re->flags & (PCRE_FIRSTSET|PCRE_STARTLINE)) == 0) 917 { 918 /* Set the character tables in the block that is passed around */ 919 920 tables = re->tables; 921 if (tables == NULL) 922 (void)pcre_fullinfo(external_re, NULL, PCRE_INFO_DEFAULT_TABLES, 923 (void *)(&tables)); 924 925 compile_block.lcc = tables + lcc_offset; 926 compile_block.fcc = tables + fcc_offset; 927 compile_block.cbits = tables + cbits_offset; 928 compile_block.ctypes = tables + ctypes_offset; 929 930 /* See if we can find a fixed set of initial characters for the pattern. */ 931 932 memset(start_bits, 0, 32 * sizeof(uschar)); 933 bits_set = set_start_bits(code, start_bits, 934 (re->options & PCRE_CASELESS) != 0, (re->options & PCRE_UTF8) != 0, 935 &compile_block) == SSB_DONE; 936 } 937 938/* Find the minimum length of subject string. */ 939 940min = find_minlength(code, code, re->options); 941 942/* Return NULL if no optimization is possible. */ 943 944if (!bits_set && min < 0) return NULL; 945 946/* Get a pcre_extra block and a pcre_study_data block. The study data is put in 947the latter, which is pointed to by the former, which may also get additional 948data set later by the calling program. At the moment, the size of 949pcre_study_data is fixed. We nevertheless save it in a field for returning via 950the pcre_fullinfo() function so that if it becomes variable in the future, we 951don't have to change that code. */ 952 953extra = (pcre_extra *)(pcre_malloc) 954 (sizeof(pcre_extra) + sizeof(pcre_study_data)); 955 956if (extra == NULL) 957 { 958 *errorptr = "failed to get memory"; 959 return NULL; 960 } 961 962study = (pcre_study_data *)((char *)extra + sizeof(pcre_extra)); 963extra->flags = PCRE_EXTRA_STUDY_DATA; 964extra->study_data = study; 965 966study->size = sizeof(pcre_study_data); 967study->flags = 0; 968 969if (bits_set) 970 { 971 study->flags |= PCRE_STUDY_MAPPED; 972 memcpy(study->start_bits, start_bits, sizeof(start_bits)); 973 } 974 975if (min >= 0) 976 { 977 study->flags |= PCRE_STUDY_MINLEN; 978 study->minlength = min; 979 } 980 981return extra; 982} 983 984/* End of pcre_study.c */ 985