1 | // Copyright 2003-2009 Google Inc. All rights reserved. |
2 | // Use of this source code is governed by a BSD-style |
3 | // license that can be found in the LICENSE file. |
4 | |
5 | // This is a variant of PCRE's pcrecpp.cc, originally written at Google. |
6 | // The main changes are the addition of the HitLimit method and |
7 | // compilation as PCRE in namespace re2. |
8 | |
9 | #include <assert.h> |
10 | #include <ctype.h> |
11 | #include <errno.h> |
12 | #include <stdlib.h> |
13 | #include <string.h> |
14 | #include <limits> |
15 | #include <string> |
16 | #include <utility> |
17 | |
18 | #include "util/util.h" |
19 | #include "util/flags.h" |
20 | #include "util/logging.h" |
21 | #include "util/pcre.h" |
22 | #include "util/strutil.h" |
23 | |
24 | // Silence warnings about the wacky formatting in the operator() functions. |
25 | // Note that we test for Clang first because it defines __GNUC__ as well. |
26 | #if defined(__clang__) |
27 | #elif defined(__GNUC__) && __GNUC__ >= 6 |
28 | #pragma GCC diagnostic ignored "-Wmisleading-indentation" |
29 | #endif |
30 | |
31 | #define PCREPORT(level) LOG(level) |
32 | |
33 | // Default PCRE limits. |
34 | // Defaults chosen to allow a plausible amount of CPU and |
35 | // not exceed main thread stacks. Note that other threads |
36 | // often have smaller stacks, and therefore tightening |
37 | // regexp_stack_limit may frequently be necessary. |
38 | DEFINE_FLAG(int, regexp_stack_limit, 256 << 10, |
39 | "default PCRE stack limit (bytes)" ); |
40 | DEFINE_FLAG(int, regexp_match_limit, 1000000, |
41 | "default PCRE match limit (function calls)" ); |
42 | |
43 | #ifndef USEPCRE |
44 | |
45 | // Fake just enough of the PCRE API to allow this file to build. :) |
46 | |
47 | struct { |
48 | int ; |
49 | int ; |
50 | int ; |
51 | }; |
52 | |
53 | #define 0 |
54 | #define 0 |
55 | #define PCRE_ANCHORED 0 |
56 | #define PCRE_NOTEMPTY 0 |
57 | #define PCRE_ERROR_NOMATCH 1 |
58 | #define PCRE_ERROR_MATCHLIMIT 2 |
59 | #define PCRE_ERROR_RECURSIONLIMIT 3 |
60 | #define PCRE_INFO_CAPTURECOUNT 0 |
61 | |
62 | void pcre_free(void*) { |
63 | } |
64 | |
65 | pcre* pcre_compile(const char*, int, const char**, int*, const unsigned char*) { |
66 | return NULL; |
67 | } |
68 | |
69 | int (const pcre*, const pcre_extra*, const char*, int, int, int, int*, int) { |
70 | return 0; |
71 | } |
72 | |
73 | int (const pcre*, const pcre_extra*, int, void*) { |
74 | return 0; |
75 | } |
76 | |
77 | #endif |
78 | |
79 | namespace re2 { |
80 | |
81 | // Maximum number of args we can set |
82 | static const int kMaxArgs = 16; |
83 | static const int kVecSize = (1 + kMaxArgs) * 3; // results + PCRE workspace |
84 | |
85 | // Approximate size of a recursive invocation of PCRE's |
86 | // internal "match()" frame. This varies depending on the |
87 | // compiler and architecture, of course, so the constant is |
88 | // just a conservative estimate. To find the exact number, |
89 | // run regexp_unittest with --regexp_stack_limit=0 under |
90 | // a debugger and look at the frames when it crashes. |
91 | // The exact frame size was 656 in production on 2008/02/03. |
92 | static const int kPCREFrameSize = 700; |
93 | |
94 | // Special name for missing C++ arguments. |
95 | PCRE::Arg PCRE::no_more_args((void*)NULL); |
96 | |
97 | const PCRE::PartialMatchFunctor PCRE::PartialMatch = { }; |
98 | const PCRE::FullMatchFunctor PCRE::FullMatch = { } ; |
99 | const PCRE::ConsumeFunctor PCRE::Consume = { }; |
100 | const PCRE::FindAndConsumeFunctor PCRE::FindAndConsume = { }; |
101 | |
102 | // If a regular expression has no error, its error_ field points here |
103 | static const std::string empty_string; |
104 | |
105 | void PCRE::Init(const char* pattern, Option options, int match_limit, |
106 | int stack_limit, bool report_errors) { |
107 | pattern_ = pattern; |
108 | options_ = options; |
109 | match_limit_ = match_limit; |
110 | stack_limit_ = stack_limit; |
111 | hit_limit_ = false; |
112 | error_ = &empty_string; |
113 | report_errors_ = report_errors; |
114 | re_full_ = NULL; |
115 | re_partial_ = NULL; |
116 | |
117 | if (options & ~(EnabledCompileOptions | EnabledExecOptions)) { |
118 | error_ = new std::string("illegal regexp option" ); |
119 | PCREPORT(ERROR) |
120 | << "Error compiling '" << pattern << "': illegal regexp option" ; |
121 | } else { |
122 | re_partial_ = Compile(UNANCHORED); |
123 | if (re_partial_ != NULL) { |
124 | re_full_ = Compile(ANCHOR_BOTH); |
125 | } |
126 | } |
127 | } |
128 | |
129 | PCRE::PCRE(const char* pattern) { |
130 | Init(pattern, None, 0, 0, true); |
131 | } |
132 | PCRE::PCRE(const char* pattern, Option option) { |
133 | Init(pattern, option, 0, 0, true); |
134 | } |
135 | PCRE::PCRE(const std::string& pattern) { |
136 | Init(pattern.c_str(), None, 0, 0, true); |
137 | } |
138 | PCRE::PCRE(const std::string& pattern, Option option) { |
139 | Init(pattern.c_str(), option, 0, 0, true); |
140 | } |
141 | PCRE::PCRE(const std::string& pattern, const PCRE_Options& re_option) { |
142 | Init(pattern.c_str(), re_option.option(), re_option.match_limit(), |
143 | re_option.stack_limit(), re_option.report_errors()); |
144 | } |
145 | |
146 | PCRE::PCRE(const char *pattern, const PCRE_Options& re_option) { |
147 | Init(pattern, re_option.option(), re_option.match_limit(), |
148 | re_option.stack_limit(), re_option.report_errors()); |
149 | } |
150 | |
151 | PCRE::~PCRE() { |
152 | if (re_full_ != NULL) pcre_free(re_full_); |
153 | if (re_partial_ != NULL) pcre_free(re_partial_); |
154 | if (error_ != &empty_string) delete error_; |
155 | } |
156 | |
157 | pcre* PCRE::Compile(Anchor anchor) { |
158 | // Special treatment for anchoring. This is needed because at |
159 | // runtime pcre only provides an option for anchoring at the |
160 | // beginning of a string. |
161 | // |
162 | // There are three types of anchoring we want: |
163 | // UNANCHORED Compile the original pattern, and use |
164 | // a pcre unanchored match. |
165 | // ANCHOR_START Compile the original pattern, and use |
166 | // a pcre anchored match. |
167 | // ANCHOR_BOTH Tack a "\z" to the end of the original pattern |
168 | // and use a pcre anchored match. |
169 | |
170 | const char* error = "" ; |
171 | int eoffset; |
172 | pcre* re; |
173 | if (anchor != ANCHOR_BOTH) { |
174 | re = pcre_compile(pattern_.c_str(), |
175 | (options_ & EnabledCompileOptions), |
176 | &error, &eoffset, NULL); |
177 | } else { |
178 | // Tack a '\z' at the end of PCRE. Parenthesize it first so that |
179 | // the '\z' applies to all top-level alternatives in the regexp. |
180 | std::string wrapped = "(?:" ; // A non-counting grouping operator |
181 | wrapped += pattern_; |
182 | wrapped += ")\\z" ; |
183 | re = pcre_compile(wrapped.c_str(), |
184 | (options_ & EnabledCompileOptions), |
185 | &error, &eoffset, NULL); |
186 | } |
187 | if (re == NULL) { |
188 | if (error_ == &empty_string) error_ = new std::string(error); |
189 | PCREPORT(ERROR) << "Error compiling '" << pattern_ << "': " << error; |
190 | } |
191 | return re; |
192 | } |
193 | |
194 | /***** Convenience interfaces *****/ |
195 | |
196 | bool PCRE::FullMatchFunctor::operator ()(const StringPiece& text, |
197 | const PCRE& re, |
198 | const Arg& a0, |
199 | const Arg& a1, |
200 | const Arg& a2, |
201 | const Arg& a3, |
202 | const Arg& a4, |
203 | const Arg& a5, |
204 | const Arg& a6, |
205 | const Arg& a7, |
206 | const Arg& a8, |
207 | const Arg& a9, |
208 | const Arg& a10, |
209 | const Arg& a11, |
210 | const Arg& a12, |
211 | const Arg& a13, |
212 | const Arg& a14, |
213 | const Arg& a15) const { |
214 | const Arg* args[kMaxArgs]; |
215 | int n = 0; |
216 | if (&a0 == &no_more_args) goto done; args[n++] = &a0; |
217 | if (&a1 == &no_more_args) goto done; args[n++] = &a1; |
218 | if (&a2 == &no_more_args) goto done; args[n++] = &a2; |
219 | if (&a3 == &no_more_args) goto done; args[n++] = &a3; |
220 | if (&a4 == &no_more_args) goto done; args[n++] = &a4; |
221 | if (&a5 == &no_more_args) goto done; args[n++] = &a5; |
222 | if (&a6 == &no_more_args) goto done; args[n++] = &a6; |
223 | if (&a7 == &no_more_args) goto done; args[n++] = &a7; |
224 | if (&a8 == &no_more_args) goto done; args[n++] = &a8; |
225 | if (&a9 == &no_more_args) goto done; args[n++] = &a9; |
226 | if (&a10 == &no_more_args) goto done; args[n++] = &a10; |
227 | if (&a11 == &no_more_args) goto done; args[n++] = &a11; |
228 | if (&a12 == &no_more_args) goto done; args[n++] = &a12; |
229 | if (&a13 == &no_more_args) goto done; args[n++] = &a13; |
230 | if (&a14 == &no_more_args) goto done; args[n++] = &a14; |
231 | if (&a15 == &no_more_args) goto done; args[n++] = &a15; |
232 | done: |
233 | |
234 | size_t consumed; |
235 | int vec[kVecSize] = {}; |
236 | return re.DoMatchImpl(text, ANCHOR_BOTH, &consumed, args, n, vec, kVecSize); |
237 | } |
238 | |
239 | bool PCRE::PartialMatchFunctor::operator ()(const StringPiece& text, |
240 | const PCRE& re, |
241 | const Arg& a0, |
242 | const Arg& a1, |
243 | const Arg& a2, |
244 | const Arg& a3, |
245 | const Arg& a4, |
246 | const Arg& a5, |
247 | const Arg& a6, |
248 | const Arg& a7, |
249 | const Arg& a8, |
250 | const Arg& a9, |
251 | const Arg& a10, |
252 | const Arg& a11, |
253 | const Arg& a12, |
254 | const Arg& a13, |
255 | const Arg& a14, |
256 | const Arg& a15) const { |
257 | const Arg* args[kMaxArgs]; |
258 | int n = 0; |
259 | if (&a0 == &no_more_args) goto done; args[n++] = &a0; |
260 | if (&a1 == &no_more_args) goto done; args[n++] = &a1; |
261 | if (&a2 == &no_more_args) goto done; args[n++] = &a2; |
262 | if (&a3 == &no_more_args) goto done; args[n++] = &a3; |
263 | if (&a4 == &no_more_args) goto done; args[n++] = &a4; |
264 | if (&a5 == &no_more_args) goto done; args[n++] = &a5; |
265 | if (&a6 == &no_more_args) goto done; args[n++] = &a6; |
266 | if (&a7 == &no_more_args) goto done; args[n++] = &a7; |
267 | if (&a8 == &no_more_args) goto done; args[n++] = &a8; |
268 | if (&a9 == &no_more_args) goto done; args[n++] = &a9; |
269 | if (&a10 == &no_more_args) goto done; args[n++] = &a10; |
270 | if (&a11 == &no_more_args) goto done; args[n++] = &a11; |
271 | if (&a12 == &no_more_args) goto done; args[n++] = &a12; |
272 | if (&a13 == &no_more_args) goto done; args[n++] = &a13; |
273 | if (&a14 == &no_more_args) goto done; args[n++] = &a14; |
274 | if (&a15 == &no_more_args) goto done; args[n++] = &a15; |
275 | done: |
276 | |
277 | size_t consumed; |
278 | int vec[kVecSize] = {}; |
279 | return re.DoMatchImpl(text, UNANCHORED, &consumed, args, n, vec, kVecSize); |
280 | } |
281 | |
282 | bool PCRE::ConsumeFunctor::operator ()(StringPiece* input, |
283 | const PCRE& pattern, |
284 | const Arg& a0, |
285 | const Arg& a1, |
286 | const Arg& a2, |
287 | const Arg& a3, |
288 | const Arg& a4, |
289 | const Arg& a5, |
290 | const Arg& a6, |
291 | const Arg& a7, |
292 | const Arg& a8, |
293 | const Arg& a9, |
294 | const Arg& a10, |
295 | const Arg& a11, |
296 | const Arg& a12, |
297 | const Arg& a13, |
298 | const Arg& a14, |
299 | const Arg& a15) const { |
300 | const Arg* args[kMaxArgs]; |
301 | int n = 0; |
302 | if (&a0 == &no_more_args) goto done; args[n++] = &a0; |
303 | if (&a1 == &no_more_args) goto done; args[n++] = &a1; |
304 | if (&a2 == &no_more_args) goto done; args[n++] = &a2; |
305 | if (&a3 == &no_more_args) goto done; args[n++] = &a3; |
306 | if (&a4 == &no_more_args) goto done; args[n++] = &a4; |
307 | if (&a5 == &no_more_args) goto done; args[n++] = &a5; |
308 | if (&a6 == &no_more_args) goto done; args[n++] = &a6; |
309 | if (&a7 == &no_more_args) goto done; args[n++] = &a7; |
310 | if (&a8 == &no_more_args) goto done; args[n++] = &a8; |
311 | if (&a9 == &no_more_args) goto done; args[n++] = &a9; |
312 | if (&a10 == &no_more_args) goto done; args[n++] = &a10; |
313 | if (&a11 == &no_more_args) goto done; args[n++] = &a11; |
314 | if (&a12 == &no_more_args) goto done; args[n++] = &a12; |
315 | if (&a13 == &no_more_args) goto done; args[n++] = &a13; |
316 | if (&a14 == &no_more_args) goto done; args[n++] = &a14; |
317 | if (&a15 == &no_more_args) goto done; args[n++] = &a15; |
318 | done: |
319 | |
320 | size_t consumed; |
321 | int vec[kVecSize] = {}; |
322 | if (pattern.DoMatchImpl(*input, ANCHOR_START, &consumed, |
323 | args, n, vec, kVecSize)) { |
324 | input->remove_prefix(consumed); |
325 | return true; |
326 | } else { |
327 | return false; |
328 | } |
329 | } |
330 | |
331 | bool PCRE::FindAndConsumeFunctor::operator ()(StringPiece* input, |
332 | const PCRE& pattern, |
333 | const Arg& a0, |
334 | const Arg& a1, |
335 | const Arg& a2, |
336 | const Arg& a3, |
337 | const Arg& a4, |
338 | const Arg& a5, |
339 | const Arg& a6, |
340 | const Arg& a7, |
341 | const Arg& a8, |
342 | const Arg& a9, |
343 | const Arg& a10, |
344 | const Arg& a11, |
345 | const Arg& a12, |
346 | const Arg& a13, |
347 | const Arg& a14, |
348 | const Arg& a15) const { |
349 | const Arg* args[kMaxArgs]; |
350 | int n = 0; |
351 | if (&a0 == &no_more_args) goto done; args[n++] = &a0; |
352 | if (&a1 == &no_more_args) goto done; args[n++] = &a1; |
353 | if (&a2 == &no_more_args) goto done; args[n++] = &a2; |
354 | if (&a3 == &no_more_args) goto done; args[n++] = &a3; |
355 | if (&a4 == &no_more_args) goto done; args[n++] = &a4; |
356 | if (&a5 == &no_more_args) goto done; args[n++] = &a5; |
357 | if (&a6 == &no_more_args) goto done; args[n++] = &a6; |
358 | if (&a7 == &no_more_args) goto done; args[n++] = &a7; |
359 | if (&a8 == &no_more_args) goto done; args[n++] = &a8; |
360 | if (&a9 == &no_more_args) goto done; args[n++] = &a9; |
361 | if (&a10 == &no_more_args) goto done; args[n++] = &a10; |
362 | if (&a11 == &no_more_args) goto done; args[n++] = &a11; |
363 | if (&a12 == &no_more_args) goto done; args[n++] = &a12; |
364 | if (&a13 == &no_more_args) goto done; args[n++] = &a13; |
365 | if (&a14 == &no_more_args) goto done; args[n++] = &a14; |
366 | if (&a15 == &no_more_args) goto done; args[n++] = &a15; |
367 | done: |
368 | |
369 | size_t consumed; |
370 | int vec[kVecSize] = {}; |
371 | if (pattern.DoMatchImpl(*input, UNANCHORED, &consumed, |
372 | args, n, vec, kVecSize)) { |
373 | input->remove_prefix(consumed); |
374 | return true; |
375 | } else { |
376 | return false; |
377 | } |
378 | } |
379 | |
380 | bool PCRE::Replace(std::string *str, |
381 | const PCRE& pattern, |
382 | const StringPiece& rewrite) { |
383 | int vec[kVecSize] = {}; |
384 | int matches = pattern.TryMatch(*str, 0, UNANCHORED, true, vec, kVecSize); |
385 | if (matches == 0) |
386 | return false; |
387 | |
388 | std::string s; |
389 | if (!pattern.Rewrite(&s, rewrite, *str, vec, matches)) |
390 | return false; |
391 | |
392 | assert(vec[0] >= 0); |
393 | assert(vec[1] >= 0); |
394 | str->replace(vec[0], vec[1] - vec[0], s); |
395 | return true; |
396 | } |
397 | |
398 | int PCRE::GlobalReplace(std::string *str, |
399 | const PCRE& pattern, |
400 | const StringPiece& rewrite) { |
401 | int count = 0; |
402 | int vec[kVecSize] = {}; |
403 | std::string out; |
404 | size_t start = 0; |
405 | bool last_match_was_empty_string = false; |
406 | |
407 | while (start <= str->size()) { |
408 | // If the previous match was for the empty string, we shouldn't |
409 | // just match again: we'll match in the same way and get an |
410 | // infinite loop. Instead, we do the match in a special way: |
411 | // anchored -- to force another try at the same position -- |
412 | // and with a flag saying that this time, ignore empty matches. |
413 | // If this special match returns, that means there's a non-empty |
414 | // match at this position as well, and we can continue. If not, |
415 | // we do what perl does, and just advance by one. |
416 | // Notice that perl prints '@@@' for this; |
417 | // perl -le '$_ = "aa"; s/b*|aa/@/g; print' |
418 | int matches; |
419 | if (last_match_was_empty_string) { |
420 | matches = pattern.TryMatch(*str, start, ANCHOR_START, false, |
421 | vec, kVecSize); |
422 | if (matches <= 0) { |
423 | if (start < str->size()) |
424 | out.push_back((*str)[start]); |
425 | start++; |
426 | last_match_was_empty_string = false; |
427 | continue; |
428 | } |
429 | } else { |
430 | matches = pattern.TryMatch(*str, start, UNANCHORED, true, |
431 | vec, kVecSize); |
432 | if (matches <= 0) |
433 | break; |
434 | } |
435 | size_t matchstart = vec[0], matchend = vec[1]; |
436 | assert(matchstart >= start); |
437 | assert(matchend >= matchstart); |
438 | |
439 | out.append(*str, start, matchstart - start); |
440 | pattern.Rewrite(&out, rewrite, *str, vec, matches); |
441 | start = matchend; |
442 | count++; |
443 | last_match_was_empty_string = (matchstart == matchend); |
444 | } |
445 | |
446 | if (count == 0) |
447 | return 0; |
448 | |
449 | if (start < str->size()) |
450 | out.append(*str, start, str->size() - start); |
451 | using std::swap; |
452 | swap(out, *str); |
453 | return count; |
454 | } |
455 | |
456 | bool PCRE::(const StringPiece &text, |
457 | const PCRE& pattern, |
458 | const StringPiece &rewrite, |
459 | std::string *out) { |
460 | int vec[kVecSize] = {}; |
461 | int matches = pattern.TryMatch(text, 0, UNANCHORED, true, vec, kVecSize); |
462 | if (matches == 0) |
463 | return false; |
464 | out->clear(); |
465 | return pattern.Rewrite(out, rewrite, text, vec, matches); |
466 | } |
467 | |
468 | std::string PCRE::QuoteMeta(const StringPiece& unquoted) { |
469 | std::string result; |
470 | result.reserve(unquoted.size() << 1); |
471 | |
472 | // Escape any ascii character not in [A-Za-z_0-9]. |
473 | // |
474 | // Note that it's legal to escape a character even if it has no |
475 | // special meaning in a regular expression -- so this function does |
476 | // that. (This also makes it identical to the perl function of the |
477 | // same name except for the null-character special case; |
478 | // see `perldoc -f quotemeta`.) |
479 | for (size_t ii = 0; ii < unquoted.size(); ++ii) { |
480 | // Note that using 'isalnum' here raises the benchmark time from |
481 | // 32ns to 58ns: |
482 | if ((unquoted[ii] < 'a' || unquoted[ii] > 'z') && |
483 | (unquoted[ii] < 'A' || unquoted[ii] > 'Z') && |
484 | (unquoted[ii] < '0' || unquoted[ii] > '9') && |
485 | unquoted[ii] != '_' && |
486 | // If this is the part of a UTF8 or Latin1 character, we need |
487 | // to copy this byte without escaping. Experimentally this is |
488 | // what works correctly with the regexp library. |
489 | !(unquoted[ii] & 128)) { |
490 | if (unquoted[ii] == '\0') { // Special handling for null chars. |
491 | // Can't use "\\0" since the next character might be a digit. |
492 | result += "\\x00" ; |
493 | continue; |
494 | } |
495 | result += '\\'; |
496 | } |
497 | result += unquoted[ii]; |
498 | } |
499 | |
500 | return result; |
501 | } |
502 | |
503 | /***** Actual matching and rewriting code *****/ |
504 | |
505 | bool PCRE::HitLimit() { |
506 | return hit_limit_ != 0; |
507 | } |
508 | |
509 | void PCRE::ClearHitLimit() { |
510 | hit_limit_ = 0; |
511 | } |
512 | |
513 | int PCRE::TryMatch(const StringPiece& text, |
514 | size_t startpos, |
515 | Anchor anchor, |
516 | bool empty_ok, |
517 | int *vec, |
518 | int vecsize) const { |
519 | pcre* re = (anchor == ANCHOR_BOTH) ? re_full_ : re_partial_; |
520 | if (re == NULL) { |
521 | PCREPORT(ERROR) << "Matching against invalid re: " << *error_; |
522 | return 0; |
523 | } |
524 | |
525 | int match_limit = match_limit_; |
526 | if (match_limit <= 0) { |
527 | match_limit = GetFlag(FLAGS_regexp_match_limit); |
528 | } |
529 | |
530 | int stack_limit = stack_limit_; |
531 | if (stack_limit <= 0) { |
532 | stack_limit = GetFlag(FLAGS_regexp_stack_limit); |
533 | } |
534 | |
535 | pcre_extra = { 0 }; |
536 | if (match_limit > 0) { |
537 | extra.flags |= PCRE_EXTRA_MATCH_LIMIT; |
538 | extra.match_limit = match_limit; |
539 | } |
540 | if (stack_limit > 0) { |
541 | extra.flags |= PCRE_EXTRA_MATCH_LIMIT_RECURSION; |
542 | extra.match_limit_recursion = stack_limit / kPCREFrameSize; |
543 | } |
544 | |
545 | int options = 0; |
546 | if (anchor != UNANCHORED) |
547 | options |= PCRE_ANCHORED; |
548 | if (!empty_ok) |
549 | options |= PCRE_NOTEMPTY; |
550 | |
551 | int rc = pcre_exec(re, // The regular expression object |
552 | &extra, |
553 | (text.data() == NULL) ? "" : text.data(), |
554 | static_cast<int>(text.size()), |
555 | static_cast<int>(startpos), |
556 | options, |
557 | vec, |
558 | vecsize); |
559 | |
560 | // Handle errors |
561 | if (rc == 0) { |
562 | // pcre_exec() returns 0 as a special case when the number of |
563 | // capturing subpatterns exceeds the size of the vector. |
564 | // When this happens, there is a match and the output vector |
565 | // is filled, but we miss out on the positions of the extra subpatterns. |
566 | rc = vecsize / 2; |
567 | } else if (rc < 0) { |
568 | switch (rc) { |
569 | case PCRE_ERROR_NOMATCH: |
570 | return 0; |
571 | case PCRE_ERROR_MATCHLIMIT: |
572 | // Writing to hit_limit is not safe if multiple threads |
573 | // are using the PCRE, but the flag is only intended |
574 | // for use by unit tests anyway, so we let it go. |
575 | hit_limit_ = true; |
576 | PCREPORT(WARNING) << "Exceeded match limit of " << match_limit |
577 | << " when matching '" << pattern_ << "'" |
578 | << " against text that is " << text.size() << " bytes." ; |
579 | return 0; |
580 | case PCRE_ERROR_RECURSIONLIMIT: |
581 | // See comment about hit_limit above. |
582 | hit_limit_ = true; |
583 | PCREPORT(WARNING) << "Exceeded stack limit of " << stack_limit |
584 | << " when matching '" << pattern_ << "'" |
585 | << " against text that is " << text.size() << " bytes." ; |
586 | return 0; |
587 | default: |
588 | // There are other return codes from pcre.h : |
589 | // PCRE_ERROR_NULL (-2) |
590 | // PCRE_ERROR_BADOPTION (-3) |
591 | // PCRE_ERROR_BADMAGIC (-4) |
592 | // PCRE_ERROR_UNKNOWN_NODE (-5) |
593 | // PCRE_ERROR_NOMEMORY (-6) |
594 | // PCRE_ERROR_NOSUBSTRING (-7) |
595 | // ... |
596 | PCREPORT(ERROR) << "Unexpected return code: " << rc |
597 | << " when matching '" << pattern_ << "'" |
598 | << ", re=" << re |
599 | << ", text=" << text |
600 | << ", vec=" << vec |
601 | << ", vecsize=" << vecsize; |
602 | return 0; |
603 | } |
604 | } |
605 | |
606 | return rc; |
607 | } |
608 | |
609 | bool PCRE::DoMatchImpl(const StringPiece& text, |
610 | Anchor anchor, |
611 | size_t* consumed, |
612 | const Arg* const* args, |
613 | int n, |
614 | int* vec, |
615 | int vecsize) const { |
616 | assert((1 + n) * 3 <= vecsize); // results + PCRE workspace |
617 | if (NumberOfCapturingGroups() < n) { |
618 | // RE has fewer capturing groups than number of Arg pointers passed in. |
619 | return false; |
620 | } |
621 | |
622 | int matches = TryMatch(text, 0, anchor, true, vec, vecsize); |
623 | assert(matches >= 0); // TryMatch never returns negatives |
624 | if (matches == 0) |
625 | return false; |
626 | |
627 | *consumed = vec[1]; |
628 | |
629 | if (n == 0 || args == NULL) { |
630 | // We are not interested in results |
631 | return true; |
632 | } |
633 | |
634 | // If we got here, we must have matched the whole pattern. |
635 | // We do not need (can not do) any more checks on the value of 'matches' here |
636 | // -- see the comment for TryMatch. |
637 | for (int i = 0; i < n; i++) { |
638 | const int start = vec[2*(i+1)]; |
639 | const int limit = vec[2*(i+1)+1]; |
640 | |
641 | // Avoid invoking undefined behavior when text.data() happens |
642 | // to be null and start happens to be -1, the latter being the |
643 | // case for an unmatched subexpression. Even if text.data() is |
644 | // not null, pointing one byte before was a longstanding bug. |
645 | const char* addr = NULL; |
646 | if (start != -1) { |
647 | addr = text.data() + start; |
648 | } |
649 | |
650 | if (!args[i]->Parse(addr, limit-start)) { |
651 | // TODO: Should we indicate what the error was? |
652 | return false; |
653 | } |
654 | } |
655 | |
656 | return true; |
657 | } |
658 | |
659 | bool PCRE::DoMatch(const StringPiece& text, |
660 | Anchor anchor, |
661 | size_t* consumed, |
662 | const Arg* const args[], |
663 | int n) const { |
664 | assert(n >= 0); |
665 | const int vecsize = (1 + n) * 3; // results + PCRE workspace |
666 | // (as for kVecSize) |
667 | int* vec = new int[vecsize]; |
668 | bool b = DoMatchImpl(text, anchor, consumed, args, n, vec, vecsize); |
669 | delete[] vec; |
670 | return b; |
671 | } |
672 | |
673 | bool PCRE::Rewrite(std::string *out, const StringPiece &rewrite, |
674 | const StringPiece &text, int *vec, int veclen) const { |
675 | int number_of_capturing_groups = NumberOfCapturingGroups(); |
676 | for (const char *s = rewrite.data(), *end = s + rewrite.size(); |
677 | s < end; s++) { |
678 | int c = *s; |
679 | if (c == '\\') { |
680 | c = *++s; |
681 | if (isdigit(c)) { |
682 | int n = (c - '0'); |
683 | if (n >= veclen) { |
684 | if (n <= number_of_capturing_groups) { |
685 | // unmatched optional capturing group. treat |
686 | // its value as empty string; i.e., nothing to append. |
687 | } else { |
688 | PCREPORT(ERROR) << "requested group " << n |
689 | << " in regexp " << rewrite.data(); |
690 | return false; |
691 | } |
692 | } |
693 | int start = vec[2 * n]; |
694 | if (start >= 0) |
695 | out->append(text.data() + start, vec[2 * n + 1] - start); |
696 | } else if (c == '\\') { |
697 | out->push_back('\\'); |
698 | } else { |
699 | PCREPORT(ERROR) << "invalid rewrite pattern: " << rewrite.data(); |
700 | return false; |
701 | } |
702 | } else { |
703 | out->push_back(c); |
704 | } |
705 | } |
706 | return true; |
707 | } |
708 | |
709 | bool PCRE::CheckRewriteString(const StringPiece& rewrite, |
710 | std::string* error) const { |
711 | int max_token = -1; |
712 | for (const char *s = rewrite.data(), *end = s + rewrite.size(); |
713 | s < end; s++) { |
714 | int c = *s; |
715 | if (c != '\\') { |
716 | continue; |
717 | } |
718 | if (++s == end) { |
719 | *error = "Rewrite schema error: '\\' not allowed at end." ; |
720 | return false; |
721 | } |
722 | c = *s; |
723 | if (c == '\\') { |
724 | continue; |
725 | } |
726 | if (!isdigit(c)) { |
727 | *error = "Rewrite schema error: " |
728 | "'\\' must be followed by a digit or '\\'." ; |
729 | return false; |
730 | } |
731 | int n = (c - '0'); |
732 | if (max_token < n) { |
733 | max_token = n; |
734 | } |
735 | } |
736 | |
737 | if (max_token > NumberOfCapturingGroups()) { |
738 | *error = StringPrintf( |
739 | "Rewrite schema requests %d matches, but the regexp only has %d " |
740 | "parenthesized subexpressions." , |
741 | max_token, NumberOfCapturingGroups()); |
742 | return false; |
743 | } |
744 | return true; |
745 | } |
746 | |
747 | |
748 | // Return the number of capturing subpatterns, or -1 if the |
749 | // regexp wasn't valid on construction. |
750 | int PCRE::NumberOfCapturingGroups() const { |
751 | if (re_partial_ == NULL) return -1; |
752 | |
753 | int result; |
754 | int rc = pcre_fullinfo(re_partial_, // The regular expression object |
755 | NULL, // We did not study the pattern |
756 | PCRE_INFO_CAPTURECOUNT, |
757 | &result); |
758 | if (rc != 0) { |
759 | PCREPORT(ERROR) << "Unexpected return code: " << rc; |
760 | return -1; |
761 | } |
762 | return result; |
763 | } |
764 | |
765 | |
766 | /***** Parsers for various types *****/ |
767 | |
768 | bool PCRE::Arg::parse_null(const char* str, size_t n, void* dest) { |
769 | // We fail if somebody asked us to store into a non-NULL void* pointer |
770 | return (dest == NULL); |
771 | } |
772 | |
773 | bool PCRE::Arg::parse_string(const char* str, size_t n, void* dest) { |
774 | if (dest == NULL) return true; |
775 | reinterpret_cast<std::string*>(dest)->assign(str, n); |
776 | return true; |
777 | } |
778 | |
779 | bool PCRE::Arg::parse_stringpiece(const char* str, size_t n, void* dest) { |
780 | if (dest == NULL) return true; |
781 | *(reinterpret_cast<StringPiece*>(dest)) = StringPiece(str, n); |
782 | return true; |
783 | } |
784 | |
785 | bool PCRE::Arg::parse_char(const char* str, size_t n, void* dest) { |
786 | if (n != 1) return false; |
787 | if (dest == NULL) return true; |
788 | *(reinterpret_cast<char*>(dest)) = str[0]; |
789 | return true; |
790 | } |
791 | |
792 | bool PCRE::Arg::parse_schar(const char* str, size_t n, void* dest) { |
793 | if (n != 1) return false; |
794 | if (dest == NULL) return true; |
795 | *(reinterpret_cast<signed char*>(dest)) = str[0]; |
796 | return true; |
797 | } |
798 | |
799 | bool PCRE::Arg::parse_uchar(const char* str, size_t n, void* dest) { |
800 | if (n != 1) return false; |
801 | if (dest == NULL) return true; |
802 | *(reinterpret_cast<unsigned char*>(dest)) = str[0]; |
803 | return true; |
804 | } |
805 | |
806 | // Largest number spec that we are willing to parse |
807 | static const int kMaxNumberLength = 32; |
808 | |
809 | // PCREQUIPCRES "buf" must have length at least kMaxNumberLength+1 |
810 | // PCREQUIPCRES "n > 0" |
811 | // Copies "str" into "buf" and null-terminates if necessary. |
812 | // Returns one of: |
813 | // a. "str" if no termination is needed |
814 | // b. "buf" if the string was copied and null-terminated |
815 | // c. "" if the input was invalid and has no hope of being parsed |
816 | static const char* TerminateNumber(char* buf, const char* str, size_t n) { |
817 | if ((n > 0) && isspace(*str)) { |
818 | // We are less forgiving than the strtoxxx() routines and do not |
819 | // allow leading spaces. |
820 | return "" ; |
821 | } |
822 | |
823 | // See if the character right after the input text may potentially |
824 | // look like a digit. |
825 | if (isdigit(str[n]) || |
826 | ((str[n] >= 'a') && (str[n] <= 'f')) || |
827 | ((str[n] >= 'A') && (str[n] <= 'F'))) { |
828 | if (n > kMaxNumberLength) return "" ; // Input too big to be a valid number |
829 | memcpy(buf, str, n); |
830 | buf[n] = '\0'; |
831 | return buf; |
832 | } else { |
833 | // We can parse right out of the supplied string, so return it. |
834 | return str; |
835 | } |
836 | } |
837 | |
838 | bool PCRE::Arg::parse_long_radix(const char* str, |
839 | size_t n, |
840 | void* dest, |
841 | int radix) { |
842 | if (n == 0) return false; |
843 | char buf[kMaxNumberLength+1]; |
844 | str = TerminateNumber(buf, str, n); |
845 | char* end; |
846 | errno = 0; |
847 | long r = strtol(str, &end, radix); |
848 | if (end != str + n) return false; // Leftover junk |
849 | if (errno) return false; |
850 | if (dest == NULL) return true; |
851 | *(reinterpret_cast<long*>(dest)) = r; |
852 | return true; |
853 | } |
854 | |
855 | bool PCRE::Arg::parse_ulong_radix(const char* str, |
856 | size_t n, |
857 | void* dest, |
858 | int radix) { |
859 | if (n == 0) return false; |
860 | char buf[kMaxNumberLength+1]; |
861 | str = TerminateNumber(buf, str, n); |
862 | if (str[0] == '-') { |
863 | // strtoul() will silently accept negative numbers and parse |
864 | // them. This module is more strict and treats them as errors. |
865 | return false; |
866 | } |
867 | |
868 | char* end; |
869 | errno = 0; |
870 | unsigned long r = strtoul(str, &end, radix); |
871 | if (end != str + n) return false; // Leftover junk |
872 | if (errno) return false; |
873 | if (dest == NULL) return true; |
874 | *(reinterpret_cast<unsigned long*>(dest)) = r; |
875 | return true; |
876 | } |
877 | |
878 | bool PCRE::Arg::parse_short_radix(const char* str, |
879 | size_t n, |
880 | void* dest, |
881 | int radix) { |
882 | long r; |
883 | if (!parse_long_radix(str, n, &r, radix)) return false; // Could not parse |
884 | if ((short)r != r) return false; // Out of range |
885 | if (dest == NULL) return true; |
886 | *(reinterpret_cast<short*>(dest)) = (short)r; |
887 | return true; |
888 | } |
889 | |
890 | bool PCRE::Arg::parse_ushort_radix(const char* str, |
891 | size_t n, |
892 | void* dest, |
893 | int radix) { |
894 | unsigned long r; |
895 | if (!parse_ulong_radix(str, n, &r, radix)) return false; // Could not parse |
896 | if ((unsigned short)r != r) return false; // Out of range |
897 | if (dest == NULL) return true; |
898 | *(reinterpret_cast<unsigned short*>(dest)) = (unsigned short)r; |
899 | return true; |
900 | } |
901 | |
902 | bool PCRE::Arg::parse_int_radix(const char* str, |
903 | size_t n, |
904 | void* dest, |
905 | int radix) { |
906 | long r; |
907 | if (!parse_long_radix(str, n, &r, radix)) return false; // Could not parse |
908 | if ((int)r != r) return false; // Out of range |
909 | if (dest == NULL) return true; |
910 | *(reinterpret_cast<int*>(dest)) = (int)r; |
911 | return true; |
912 | } |
913 | |
914 | bool PCRE::Arg::parse_uint_radix(const char* str, |
915 | size_t n, |
916 | void* dest, |
917 | int radix) { |
918 | unsigned long r; |
919 | if (!parse_ulong_radix(str, n, &r, radix)) return false; // Could not parse |
920 | if ((unsigned int)r != r) return false; // Out of range |
921 | if (dest == NULL) return true; |
922 | *(reinterpret_cast<unsigned int*>(dest)) = (unsigned int)r; |
923 | return true; |
924 | } |
925 | |
926 | bool PCRE::Arg::parse_longlong_radix(const char* str, |
927 | size_t n, |
928 | void* dest, |
929 | int radix) { |
930 | if (n == 0) return false; |
931 | char buf[kMaxNumberLength+1]; |
932 | str = TerminateNumber(buf, str, n); |
933 | char* end; |
934 | errno = 0; |
935 | long long r = strtoll(str, &end, radix); |
936 | if (end != str + n) return false; // Leftover junk |
937 | if (errno) return false; |
938 | if (dest == NULL) return true; |
939 | *(reinterpret_cast<long long*>(dest)) = r; |
940 | return true; |
941 | } |
942 | |
943 | bool PCRE::Arg::parse_ulonglong_radix(const char* str, |
944 | size_t n, |
945 | void* dest, |
946 | int radix) { |
947 | if (n == 0) return false; |
948 | char buf[kMaxNumberLength+1]; |
949 | str = TerminateNumber(buf, str, n); |
950 | if (str[0] == '-') { |
951 | // strtoull() will silently accept negative numbers and parse |
952 | // them. This module is more strict and treats them as errors. |
953 | return false; |
954 | } |
955 | char* end; |
956 | errno = 0; |
957 | unsigned long long r = strtoull(str, &end, radix); |
958 | if (end != str + n) return false; // Leftover junk |
959 | if (errno) return false; |
960 | if (dest == NULL) return true; |
961 | *(reinterpret_cast<unsigned long long*>(dest)) = r; |
962 | return true; |
963 | } |
964 | |
965 | static bool parse_double_float(const char* str, size_t n, bool isfloat, |
966 | void* dest) { |
967 | if (n == 0) return false; |
968 | static const int kMaxLength = 200; |
969 | char buf[kMaxLength]; |
970 | if (n >= kMaxLength) return false; |
971 | memcpy(buf, str, n); |
972 | buf[n] = '\0'; |
973 | char* end; |
974 | errno = 0; |
975 | double r; |
976 | if (isfloat) { |
977 | r = strtof(buf, &end); |
978 | } else { |
979 | r = strtod(buf, &end); |
980 | } |
981 | if (end != buf + n) return false; // Leftover junk |
982 | if (errno) return false; |
983 | if (dest == NULL) return true; |
984 | if (isfloat) { |
985 | *(reinterpret_cast<float*>(dest)) = (float)r; |
986 | } else { |
987 | *(reinterpret_cast<double*>(dest)) = r; |
988 | } |
989 | return true; |
990 | } |
991 | |
992 | bool PCRE::Arg::parse_double(const char* str, size_t n, void* dest) { |
993 | return parse_double_float(str, n, false, dest); |
994 | } |
995 | |
996 | bool PCRE::Arg::parse_float(const char* str, size_t n, void* dest) { |
997 | return parse_double_float(str, n, true, dest); |
998 | } |
999 | |
1000 | #define DEFINE_INTEGER_PARSER(name) \ |
1001 | bool PCRE::Arg::parse_##name(const char* str, size_t n, void* dest) { \ |
1002 | return parse_##name##_radix(str, n, dest, 10); \ |
1003 | } \ |
1004 | bool PCRE::Arg::parse_##name##_hex(const char* str, size_t n, void* dest) { \ |
1005 | return parse_##name##_radix(str, n, dest, 16); \ |
1006 | } \ |
1007 | bool PCRE::Arg::parse_##name##_octal(const char* str, size_t n, \ |
1008 | void* dest) { \ |
1009 | return parse_##name##_radix(str, n, dest, 8); \ |
1010 | } \ |
1011 | bool PCRE::Arg::parse_##name##_cradix(const char* str, size_t n, \ |
1012 | void* dest) { \ |
1013 | return parse_##name##_radix(str, n, dest, 0); \ |
1014 | } |
1015 | |
1016 | DEFINE_INTEGER_PARSER(short); |
1017 | DEFINE_INTEGER_PARSER(ushort); |
1018 | DEFINE_INTEGER_PARSER(int); |
1019 | DEFINE_INTEGER_PARSER(uint); |
1020 | DEFINE_INTEGER_PARSER(long); |
1021 | DEFINE_INTEGER_PARSER(ulong); |
1022 | DEFINE_INTEGER_PARSER(longlong); |
1023 | DEFINE_INTEGER_PARSER(ulonglong); |
1024 | |
1025 | #undef DEFINE_INTEGER_PARSER |
1026 | |
1027 | } // namespace re2 |
1028 | |