Line data Source code
1 : // Copyright 2012 the V8 project authors. All rights reserved.
2 : // Use of this source code is governed by a BSD-style license that can be
3 : // found in the LICENSE file.
4 :
5 : #include "src/regexp/regexp-macro-assembler.h"
6 :
7 : #include "src/assembler.h"
8 : #include "src/isolate-inl.h"
9 : #include "src/regexp/regexp-stack.h"
10 : #include "src/simulator.h"
11 : #include "src/unicode-inl.h"
12 :
13 : #ifdef V8_INTL_SUPPORT
14 : #include "unicode/uchar.h"
15 : #include "unicode/unistr.h"
16 : #endif // V8_INTL_SUPPORT
17 :
18 : namespace v8 {
19 : namespace internal {
20 :
21 3275 : RegExpMacroAssembler::RegExpMacroAssembler(Isolate* isolate, Zone* zone)
22 : : slow_safe_compiler_(false),
23 : global_mode_(NOT_GLOBAL),
24 : isolate_(isolate),
25 85448 : zone_(zone) {}
26 :
27 : RegExpMacroAssembler::~RegExpMacroAssembler() = default;
28 :
29 416053 : int RegExpMacroAssembler::CaseInsensitiveCompareUC16(Address byte_offset1,
30 : Address byte_offset2,
31 : size_t byte_length,
32 : Isolate* isolate) {
33 : // This function is not allowed to cause a garbage collection.
34 : // A GC might move the calling generated code and invalidate the
35 : // return address on the stack.
36 : DCHECK_EQ(0, byte_length % 2);
37 :
38 : #ifdef V8_INTL_SUPPORT
39 416053 : int32_t length = (int32_t)(byte_length >> 1);
40 : icu::UnicodeString uni_str_1(reinterpret_cast<const char16_t*>(byte_offset1),
41 832106 : length);
42 1248159 : return uni_str_1.caseCompare(reinterpret_cast<const char16_t*>(byte_offset2),
43 832106 : length, U_FOLD_CASE_DEFAULT) == 0;
44 : #else
45 : uc16* substring1 = reinterpret_cast<uc16*>(byte_offset1);
46 : uc16* substring2 = reinterpret_cast<uc16*>(byte_offset2);
47 : size_t length = byte_length >> 1;
48 : DCHECK_NOT_NULL(isolate);
49 : unibrow::Mapping<unibrow::Ecma262Canonicalize>* canonicalize =
50 : isolate->regexp_macro_assembler_canonicalize();
51 : for (size_t i = 0; i < length; i++) {
52 : unibrow::uchar c1 = substring1[i];
53 : unibrow::uchar c2 = substring2[i];
54 : if (c1 != c2) {
55 : unibrow::uchar s1[1] = {c1};
56 : canonicalize->get(c1, '\0', s1);
57 : if (s1[0] != c2) {
58 : unibrow::uchar s2[1] = {c2};
59 : canonicalize->get(c2, '\0', s2);
60 : if (s1[0] != s2[0]) {
61 : return 0;
62 : }
63 : }
64 : }
65 : }
66 : return 1;
67 : #endif // V8_INTL_SUPPORT
68 : }
69 :
70 :
71 84 : void RegExpMacroAssembler::CheckNotInSurrogatePair(int cp_offset,
72 : Label* on_failure) {
73 84 : Label ok;
74 : // Check that current character is not a trail surrogate.
75 84 : LoadCurrentCharacter(cp_offset, &ok);
76 84 : CheckCharacterNotInRange(kTrailSurrogateStart, kTrailSurrogateEnd, &ok);
77 : // Check that previous character is not a lead surrogate.
78 84 : LoadCurrentCharacter(cp_offset - 1, &ok);
79 84 : CheckCharacterInRange(kLeadSurrogateStart, kLeadSurrogateEnd, on_failure);
80 84 : Bind(&ok);
81 84 : }
82 :
83 1288 : void RegExpMacroAssembler::CheckPosition(int cp_offset,
84 : Label* on_outside_input) {
85 1288 : LoadCurrentCharacter(cp_offset, on_outside_input, true);
86 1288 : }
87 :
88 1927 : bool RegExpMacroAssembler::CheckSpecialCharacterClass(uc16 type,
89 : Label* on_no_match) {
90 1927 : return false;
91 : }
92 :
93 82173 : NativeRegExpMacroAssembler::NativeRegExpMacroAssembler(Isolate* isolate,
94 : Zone* zone)
95 82173 : : RegExpMacroAssembler(isolate, zone) {}
96 :
97 : NativeRegExpMacroAssembler::~NativeRegExpMacroAssembler() = default;
98 :
99 185392 : bool NativeRegExpMacroAssembler::CanReadUnaligned() {
100 185392 : return FLAG_enable_regexp_unaligned_accesses && !slow_safe();
101 : }
102 :
103 111568 : const byte* NativeRegExpMacroAssembler::StringCharacterPosition(
104 : String subject, int start_index, const DisallowHeapAllocation& no_gc) {
105 111568 : if (subject->IsConsString()) {
106 0 : subject = ConsString::cast(subject)->first();
107 111568 : } else if (subject->IsSlicedString()) {
108 2 : start_index += SlicedString::cast(subject)->offset();
109 2 : subject = SlicedString::cast(subject)->parent();
110 : }
111 111568 : if (subject->IsThinString()) {
112 0 : subject = ThinString::cast(subject)->actual();
113 : }
114 : DCHECK_LE(0, start_index);
115 : DCHECK_LE(start_index, subject->length());
116 111568 : if (subject->IsSeqOneByteString()) {
117 : return reinterpret_cast<const byte*>(
118 40899 : SeqOneByteString::cast(subject)->GetChars(no_gc) + start_index);
119 70669 : } else if (subject->IsSeqTwoByteString()) {
120 : return reinterpret_cast<const byte*>(
121 70600 : SeqTwoByteString::cast(subject)->GetChars(no_gc) + start_index);
122 69 : } else if (subject->IsExternalOneByteString()) {
123 : return reinterpret_cast<const byte*>(
124 24 : ExternalOneByteString::cast(subject)->GetChars() + start_index);
125 : } else {
126 : DCHECK(subject->IsExternalTwoByteString());
127 : return reinterpret_cast<const byte*>(
128 45 : ExternalTwoByteString::cast(subject)->GetChars() + start_index);
129 : }
130 : }
131 :
132 413 : int NativeRegExpMacroAssembler::CheckStackGuardState(
133 : Isolate* isolate, int start_index, bool is_direct_call,
134 : Address* return_address, Code re_code, Address* subject,
135 : const byte** input_start, const byte** input_end) {
136 : AllowHeapAllocation allow_allocation;
137 : DCHECK(re_code->raw_instruction_start() <= *return_address);
138 : DCHECK(*return_address <= re_code->raw_instruction_end());
139 : int return_value = 0;
140 : // Prepare for possible GC.
141 : HandleScope handles(isolate);
142 : Handle<Code> code_handle(re_code, isolate);
143 413 : Handle<String> subject_handle(String::cast(Object(*subject)), isolate);
144 413 : bool is_one_byte = String::IsOneByteRepresentationUnderneath(*subject_handle);
145 :
146 : StackLimitCheck check(isolate);
147 413 : bool js_has_overflowed = check.JsHasOverflowed();
148 :
149 413 : if (is_direct_call) {
150 : // Direct calls from JavaScript can be interrupted in two ways:
151 : // 1. A real stack overflow, in which case we let the caller throw the
152 : // exception.
153 : // 2. The stack guard was used to interrupt execution for another purpose,
154 : // forcing the call through the runtime system.
155 158 : return_value = js_has_overflowed ? EXCEPTION : RETRY;
156 255 : } else if (js_has_overflowed) {
157 80 : isolate->StackOverflow();
158 : return_value = EXCEPTION;
159 : } else {
160 175 : Object result = isolate->stack_guard()->HandleInterrupts();
161 175 : if (result->IsException(isolate)) return_value = EXCEPTION;
162 : }
163 :
164 : DisallowHeapAllocation no_gc;
165 :
166 413 : if (*code_handle != re_code) { // Return address no longer valid
167 1 : intptr_t delta = code_handle->address() - re_code->address();
168 : // Overwrite the return address on the stack.
169 1 : *return_address += delta;
170 : }
171 :
172 : // If we continue, we need to update the subject string addresses.
173 413 : if (return_value == 0) {
174 : // String encoding might have changed.
175 159 : if (String::IsOneByteRepresentationUnderneath(*subject_handle) !=
176 : is_one_byte) {
177 : // If we changed between an LATIN1 and an UC16 string, the specialized
178 : // code cannot be used, and we need to restart regexp matching from
179 : // scratch (including, potentially, compiling a new version of the code).
180 : return_value = RETRY;
181 : } else {
182 155 : *subject = subject_handle->ptr();
183 155 : intptr_t byte_length = *input_end - *input_start;
184 : *input_start =
185 155 : StringCharacterPosition(*subject_handle, start_index, no_gc);
186 155 : *input_end = *input_start + byte_length;
187 : }
188 : }
189 413 : return return_value;
190 : }
191 :
192 : // Returns a {Result} sentinel, or the number of successful matches.
193 111413 : int NativeRegExpMacroAssembler::Match(Handle<Code> regexp_code,
194 : Handle<String> subject,
195 : int* offsets_vector,
196 : int offsets_vector_length,
197 : int previous_index, Isolate* isolate) {
198 : DCHECK(subject->IsFlat());
199 : DCHECK_LE(0, previous_index);
200 : DCHECK_LE(previous_index, subject->length());
201 :
202 : // No allocations before calling the regexp, but we can't use
203 : // DisallowHeapAllocation, since regexps might be preempted, and another
204 : // thread might do allocation anyway.
205 :
206 : String subject_ptr = *subject;
207 : // Character offsets into string.
208 : int start_offset = previous_index;
209 111413 : int char_length = subject_ptr->length() - start_offset;
210 : int slice_offset = 0;
211 :
212 : // The string has been flattened, so if it is a cons string it contains the
213 : // full string in the first part.
214 111413 : if (StringShape(subject_ptr).IsCons()) {
215 : DCHECK_EQ(0, ConsString::cast(subject_ptr)->second()->length());
216 : subject_ptr = ConsString::cast(subject_ptr)->first();
217 111413 : } else if (StringShape(subject_ptr).IsSliced()) {
218 : SlicedString slice = SlicedString::cast(subject_ptr);
219 : subject_ptr = slice->parent();
220 : slice_offset = slice->offset();
221 : }
222 111413 : if (StringShape(subject_ptr).IsThin()) {
223 : subject_ptr = ThinString::cast(subject_ptr)->actual();
224 : }
225 : // Ensure that an underlying string has the same representation.
226 : bool is_one_byte = subject_ptr->IsOneByteRepresentation();
227 : DCHECK(subject_ptr->IsExternalString() || subject_ptr->IsSeqString());
228 : // String is now either Sequential or External
229 111413 : int char_size_shift = is_one_byte ? 0 : 1;
230 :
231 : DisallowHeapAllocation no_gc;
232 : const byte* input_start =
233 111413 : StringCharacterPosition(subject_ptr, start_offset + slice_offset, no_gc);
234 111413 : int byte_length = char_length << char_size_shift;
235 111413 : const byte* input_end = input_start + byte_length;
236 : return Execute(*regexp_code, *subject, start_offset, input_start, input_end,
237 111413 : offsets_vector, offsets_vector_length, isolate);
238 : }
239 :
240 : // Returns a {Result} sentinel, or the number of successful matches.
241 111469 : int NativeRegExpMacroAssembler::Execute(
242 : Code code,
243 : String input, // This needs to be the unpacked (sliced, cons) string.
244 : int start_offset, const byte* input_start, const byte* input_end,
245 : int* output, int output_size, Isolate* isolate) {
246 : // Ensure that the minimum stack has been allocated.
247 222938 : RegExpStackScope stack_scope(isolate);
248 : Address stack_base = stack_scope.stack()->stack_base();
249 :
250 : int direct_call = 0;
251 :
252 : using RegexpMatcherSig = int(
253 : Address input_string, int start_offset, // NOLINT(readability/casting)
254 : const byte* input_start, const byte* input_end, int* output,
255 : int output_size, Address stack_base, int direct_call, Isolate* isolate);
256 :
257 : auto fn = GeneratedCode<RegexpMatcherSig>::FromCode(code);
258 : int result =
259 : fn.CallIrregexp(input.ptr(), start_offset, input_start, input_end, output,
260 : output_size, stack_base, direct_call, isolate);
261 : DCHECK(result >= RETRY);
262 :
263 111569 : if (result == EXCEPTION && !isolate->has_pending_exception()) {
264 : // We detected a stack overflow (on the backtrack stack) in RegExp code,
265 : // but haven't created the exception yet. Additionally, we allow heap
266 : // allocation because even though it invalidates {input_start} and
267 : // {input_end}, we are about to return anyway.
268 : AllowHeapAllocation allow_allocation;
269 4 : isolate->StackOverflow();
270 : }
271 111469 : return result;
272 : }
273 :
274 : // clang-format off
275 : const byte NativeRegExpMacroAssembler::word_character_map[] = {
276 : 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u,
277 : 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u,
278 : 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u,
279 : 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u,
280 :
281 : 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u,
282 : 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u,
283 : 0xFFu, 0xFFu, 0xFFu, 0xFFu, 0xFFu, 0xFFu, 0xFFu, 0xFFu, // '0' - '7'
284 : 0xFFu, 0xFFu, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, // '8' - '9'
285 :
286 : 0x00u, 0xFFu, 0xFFu, 0xFFu, 0xFFu, 0xFFu, 0xFFu, 0xFFu, // 'A' - 'G'
287 : 0xFFu, 0xFFu, 0xFFu, 0xFFu, 0xFFu, 0xFFu, 0xFFu, 0xFFu, // 'H' - 'O'
288 : 0xFFu, 0xFFu, 0xFFu, 0xFFu, 0xFFu, 0xFFu, 0xFFu, 0xFFu, // 'P' - 'W'
289 : 0xFFu, 0xFFu, 0xFFu, 0x00u, 0x00u, 0x00u, 0x00u, 0xFFu, // 'X' - 'Z', '_'
290 :
291 : 0x00u, 0xFFu, 0xFFu, 0xFFu, 0xFFu, 0xFFu, 0xFFu, 0xFFu, // 'a' - 'g'
292 : 0xFFu, 0xFFu, 0xFFu, 0xFFu, 0xFFu, 0xFFu, 0xFFu, 0xFFu, // 'h' - 'o'
293 : 0xFFu, 0xFFu, 0xFFu, 0xFFu, 0xFFu, 0xFFu, 0xFFu, 0xFFu, // 'p' - 'w'
294 : 0xFFu, 0xFFu, 0xFFu, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, // 'x' - 'z'
295 : // Latin-1 range
296 : 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u,
297 : 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u,
298 : 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u,
299 : 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u,
300 :
301 : 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u,
302 : 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u,
303 : 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u,
304 : 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u,
305 :
306 : 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u,
307 : 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u,
308 : 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u,
309 : 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u,
310 :
311 : 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u,
312 : 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u,
313 : 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u,
314 : 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u,
315 : };
316 : // clang-format on
317 :
318 350 : Address NativeRegExpMacroAssembler::GrowStack(Address stack_pointer,
319 : Address* stack_base,
320 : Isolate* isolate) {
321 : RegExpStack* regexp_stack = isolate->regexp_stack();
322 : size_t size = regexp_stack->stack_capacity();
323 : Address old_stack_base = regexp_stack->stack_base();
324 : DCHECK(old_stack_base == *stack_base);
325 : DCHECK(stack_pointer <= old_stack_base);
326 : DCHECK(static_cast<size_t>(old_stack_base - stack_pointer) <= size);
327 350 : Address new_stack_base = regexp_stack->EnsureCapacity(size * 2);
328 350 : if (new_stack_base == kNullAddress) {
329 : return kNullAddress;
330 : }
331 346 : *stack_base = new_stack_base;
332 346 : intptr_t stack_content_size = old_stack_base - stack_pointer;
333 346 : return new_stack_base - stack_content_size;
334 : }
335 :
336 : } // namespace internal
337 121996 : } // namespace v8
|