Ben Murdoch | 4a90d5f | 2016-03-22 12:00:34 +0000 | [diff] [blame] | 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 | |
Ben Murdoch | 097c5b2 | 2016-05-18 11:27:45 +0100 | [diff] [blame] | 12 | #ifdef V8_I18N_SUPPORT |
| 13 | #include "unicode/uchar.h" |
| 14 | #endif // V8_I18N_SUPPORT |
| 15 | |
Ben Murdoch | 4a90d5f | 2016-03-22 12:00:34 +0000 | [diff] [blame] | 16 | namespace v8 { |
| 17 | namespace internal { |
| 18 | |
| 19 | RegExpMacroAssembler::RegExpMacroAssembler(Isolate* isolate, Zone* zone) |
| 20 | : slow_safe_compiler_(false), |
| 21 | global_mode_(NOT_GLOBAL), |
| 22 | isolate_(isolate), |
| 23 | zone_(zone) {} |
| 24 | |
| 25 | |
| 26 | RegExpMacroAssembler::~RegExpMacroAssembler() { |
| 27 | } |
| 28 | |
| 29 | |
Ben Murdoch | 097c5b2 | 2016-05-18 11:27:45 +0100 | [diff] [blame] | 30 | int RegExpMacroAssembler::CaseInsensitiveCompareUC16(Address byte_offset1, |
| 31 | Address byte_offset2, |
| 32 | size_t byte_length, |
| 33 | Isolate* isolate) { |
| 34 | unibrow::Mapping<unibrow::Ecma262Canonicalize>* canonicalize = |
| 35 | isolate->regexp_macro_assembler_canonicalize(); |
| 36 | // This function is not allowed to cause a garbage collection. |
| 37 | // A GC might move the calling generated code and invalidate the |
| 38 | // return address on the stack. |
| 39 | DCHECK(byte_length % 2 == 0); |
| 40 | uc16* substring1 = reinterpret_cast<uc16*>(byte_offset1); |
| 41 | uc16* substring2 = reinterpret_cast<uc16*>(byte_offset2); |
| 42 | size_t length = byte_length >> 1; |
| 43 | |
| 44 | #ifdef V8_I18N_SUPPORT |
| 45 | if (isolate == nullptr) { |
| 46 | for (size_t i = 0; i < length; i++) { |
| 47 | uc32 c1 = substring1[i]; |
| 48 | uc32 c2 = substring2[i]; |
| 49 | if (unibrow::Utf16::IsLeadSurrogate(c1)) { |
| 50 | // Non-BMP characters do not have case-equivalents in the BMP. |
| 51 | // Both have to be non-BMP for them to be able to match. |
| 52 | if (!unibrow::Utf16::IsLeadSurrogate(c2)) return 0; |
| 53 | if (i + 1 < length) { |
| 54 | uc16 c1t = substring1[i + 1]; |
| 55 | uc16 c2t = substring2[i + 1]; |
| 56 | if (unibrow::Utf16::IsTrailSurrogate(c1t) && |
| 57 | unibrow::Utf16::IsTrailSurrogate(c2t)) { |
| 58 | c1 = unibrow::Utf16::CombineSurrogatePair(c1, c1t); |
| 59 | c2 = unibrow::Utf16::CombineSurrogatePair(c2, c2t); |
| 60 | i++; |
| 61 | } |
| 62 | } |
| 63 | } |
| 64 | c1 = u_foldCase(c1, U_FOLD_CASE_DEFAULT); |
| 65 | c2 = u_foldCase(c2, U_FOLD_CASE_DEFAULT); |
| 66 | if (c1 != c2) return 0; |
| 67 | } |
| 68 | return 1; |
| 69 | } |
| 70 | #endif // V8_I18N_SUPPORT |
| 71 | DCHECK_NOT_NULL(isolate); |
| 72 | for (size_t i = 0; i < length; i++) { |
| 73 | unibrow::uchar c1 = substring1[i]; |
| 74 | unibrow::uchar c2 = substring2[i]; |
| 75 | if (c1 != c2) { |
| 76 | unibrow::uchar s1[1] = {c1}; |
| 77 | canonicalize->get(c1, '\0', s1); |
| 78 | if (s1[0] != c2) { |
| 79 | unibrow::uchar s2[1] = {c2}; |
| 80 | canonicalize->get(c2, '\0', s2); |
| 81 | if (s1[0] != s2[0]) { |
| 82 | return 0; |
| 83 | } |
| 84 | } |
| 85 | } |
| 86 | } |
| 87 | return 1; |
| 88 | } |
| 89 | |
| 90 | |
| 91 | void RegExpMacroAssembler::CheckNotInSurrogatePair(int cp_offset, |
| 92 | Label* on_failure) { |
| 93 | Label ok; |
| 94 | // Check that current character is not a trail surrogate. |
| 95 | LoadCurrentCharacter(cp_offset, &ok); |
| 96 | CheckCharacterNotInRange(kTrailSurrogateStart, kTrailSurrogateEnd, &ok); |
| 97 | // Check that previous character is not a lead surrogate. |
| 98 | LoadCurrentCharacter(cp_offset - 1, &ok); |
| 99 | CheckCharacterInRange(kLeadSurrogateStart, kLeadSurrogateEnd, on_failure); |
| 100 | Bind(&ok); |
| 101 | } |
| 102 | |
Ben Murdoch | c561043 | 2016-08-08 18:44:38 +0100 | [diff] [blame] | 103 | void RegExpMacroAssembler::CheckPosition(int cp_offset, |
| 104 | Label* on_outside_input) { |
| 105 | LoadCurrentCharacter(cp_offset, on_outside_input, true); |
| 106 | } |
| 107 | |
| 108 | bool RegExpMacroAssembler::CheckSpecialCharacterClass(uc16 type, |
| 109 | Label* on_no_match) { |
| 110 | return false; |
| 111 | } |
Ben Murdoch | 097c5b2 | 2016-05-18 11:27:45 +0100 | [diff] [blame] | 112 | |
Ben Murdoch | 4a90d5f | 2016-03-22 12:00:34 +0000 | [diff] [blame] | 113 | #ifndef V8_INTERPRETED_REGEXP // Avoid unused code, e.g., on ARM. |
| 114 | |
| 115 | NativeRegExpMacroAssembler::NativeRegExpMacroAssembler(Isolate* isolate, |
| 116 | Zone* zone) |
| 117 | : RegExpMacroAssembler(isolate, zone) {} |
| 118 | |
| 119 | |
| 120 | NativeRegExpMacroAssembler::~NativeRegExpMacroAssembler() { |
| 121 | } |
| 122 | |
| 123 | |
| 124 | bool NativeRegExpMacroAssembler::CanReadUnaligned() { |
| 125 | return FLAG_enable_unaligned_accesses && !slow_safe(); |
| 126 | } |
| 127 | |
| 128 | const byte* NativeRegExpMacroAssembler::StringCharacterPosition( |
| 129 | String* subject, |
| 130 | int start_index) { |
| 131 | if (subject->IsConsString()) { |
| 132 | subject = ConsString::cast(subject)->first(); |
| 133 | } else if (subject->IsSlicedString()) { |
| 134 | start_index += SlicedString::cast(subject)->offset(); |
| 135 | subject = SlicedString::cast(subject)->parent(); |
| 136 | } |
| 137 | DCHECK(start_index >= 0); |
| 138 | DCHECK(start_index <= subject->length()); |
| 139 | if (subject->IsSeqOneByteString()) { |
| 140 | return reinterpret_cast<const byte*>( |
| 141 | SeqOneByteString::cast(subject)->GetChars() + start_index); |
| 142 | } else if (subject->IsSeqTwoByteString()) { |
| 143 | return reinterpret_cast<const byte*>( |
| 144 | SeqTwoByteString::cast(subject)->GetChars() + start_index); |
| 145 | } else if (subject->IsExternalOneByteString()) { |
| 146 | return reinterpret_cast<const byte*>( |
| 147 | ExternalOneByteString::cast(subject)->GetChars() + start_index); |
| 148 | } else { |
| 149 | return reinterpret_cast<const byte*>( |
| 150 | ExternalTwoByteString::cast(subject)->GetChars() + start_index); |
| 151 | } |
| 152 | } |
| 153 | |
| 154 | |
| 155 | int NativeRegExpMacroAssembler::CheckStackGuardState( |
| 156 | Isolate* isolate, int start_index, bool is_direct_call, |
| 157 | Address* return_address, Code* re_code, String** subject, |
| 158 | const byte** input_start, const byte** input_end) { |
| 159 | DCHECK(re_code->instruction_start() <= *return_address); |
| 160 | DCHECK(*return_address <= re_code->instruction_end()); |
| 161 | int return_value = 0; |
| 162 | // Prepare for possible GC. |
| 163 | HandleScope handles(isolate); |
| 164 | Handle<Code> code_handle(re_code); |
| 165 | Handle<String> subject_handle(*subject); |
| 166 | bool is_one_byte = subject_handle->IsOneByteRepresentationUnderneath(); |
| 167 | |
| 168 | StackLimitCheck check(isolate); |
| 169 | if (check.JsHasOverflowed()) { |
| 170 | isolate->StackOverflow(); |
| 171 | return_value = EXCEPTION; |
| 172 | } else if (is_direct_call) { |
| 173 | // If not real stack overflow the stack guard was used to interrupt |
| 174 | // execution for another purpose. If this is a direct call from JavaScript |
| 175 | // retry the RegExp forcing the call through the runtime system. |
| 176 | // Currently the direct call cannot handle a GC. |
| 177 | return_value = RETRY; |
| 178 | } else { |
| 179 | Object* result = isolate->stack_guard()->HandleInterrupts(); |
Ben Murdoch | 61f157c | 2016-09-16 13:49:30 +0100 | [diff] [blame] | 180 | if (result->IsException(isolate)) return_value = EXCEPTION; |
Ben Murdoch | 4a90d5f | 2016-03-22 12:00:34 +0000 | [diff] [blame] | 181 | } |
| 182 | |
| 183 | DisallowHeapAllocation no_gc; |
| 184 | |
| 185 | if (*code_handle != re_code) { // Return address no longer valid |
| 186 | intptr_t delta = code_handle->address() - re_code->address(); |
| 187 | // Overwrite the return address on the stack. |
| 188 | *return_address += delta; |
| 189 | } |
| 190 | |
| 191 | // If we continue, we need to update the subject string addresses. |
| 192 | if (return_value == 0) { |
| 193 | // String encoding might have changed. |
| 194 | if (subject_handle->IsOneByteRepresentationUnderneath() != is_one_byte) { |
| 195 | // If we changed between an LATIN1 and an UC16 string, the specialized |
| 196 | // code cannot be used, and we need to restart regexp matching from |
| 197 | // scratch (including, potentially, compiling a new version of the code). |
| 198 | return_value = RETRY; |
| 199 | } else { |
| 200 | *subject = *subject_handle; |
| 201 | intptr_t byte_length = *input_end - *input_start; |
| 202 | *input_start = StringCharacterPosition(*subject, start_index); |
| 203 | *input_end = *input_start + byte_length; |
| 204 | } |
| 205 | } |
| 206 | return return_value; |
| 207 | } |
| 208 | |
| 209 | |
| 210 | NativeRegExpMacroAssembler::Result NativeRegExpMacroAssembler::Match( |
| 211 | Handle<Code> regexp_code, |
| 212 | Handle<String> subject, |
| 213 | int* offsets_vector, |
| 214 | int offsets_vector_length, |
| 215 | int previous_index, |
| 216 | Isolate* isolate) { |
| 217 | |
| 218 | DCHECK(subject->IsFlat()); |
| 219 | DCHECK(previous_index >= 0); |
| 220 | DCHECK(previous_index <= subject->length()); |
| 221 | |
| 222 | // No allocations before calling the regexp, but we can't use |
| 223 | // DisallowHeapAllocation, since regexps might be preempted, and another |
| 224 | // thread might do allocation anyway. |
| 225 | |
| 226 | String* subject_ptr = *subject; |
| 227 | // Character offsets into string. |
| 228 | int start_offset = previous_index; |
| 229 | int char_length = subject_ptr->length() - start_offset; |
| 230 | int slice_offset = 0; |
| 231 | |
| 232 | // The string has been flattened, so if it is a cons string it contains the |
| 233 | // full string in the first part. |
| 234 | if (StringShape(subject_ptr).IsCons()) { |
| 235 | DCHECK_EQ(0, ConsString::cast(subject_ptr)->second()->length()); |
| 236 | subject_ptr = ConsString::cast(subject_ptr)->first(); |
| 237 | } else if (StringShape(subject_ptr).IsSliced()) { |
| 238 | SlicedString* slice = SlicedString::cast(subject_ptr); |
| 239 | subject_ptr = slice->parent(); |
| 240 | slice_offset = slice->offset(); |
| 241 | } |
| 242 | // Ensure that an underlying string has the same representation. |
| 243 | bool is_one_byte = subject_ptr->IsOneByteRepresentation(); |
| 244 | DCHECK(subject_ptr->IsExternalString() || subject_ptr->IsSeqString()); |
| 245 | // String is now either Sequential or External |
| 246 | int char_size_shift = is_one_byte ? 0 : 1; |
| 247 | |
| 248 | const byte* input_start = |
| 249 | StringCharacterPosition(subject_ptr, start_offset + slice_offset); |
| 250 | int byte_length = char_length << char_size_shift; |
| 251 | const byte* input_end = input_start + byte_length; |
| 252 | Result res = Execute(*regexp_code, |
| 253 | *subject, |
| 254 | start_offset, |
| 255 | input_start, |
| 256 | input_end, |
| 257 | offsets_vector, |
| 258 | offsets_vector_length, |
| 259 | isolate); |
| 260 | return res; |
| 261 | } |
| 262 | |
| 263 | |
| 264 | NativeRegExpMacroAssembler::Result NativeRegExpMacroAssembler::Execute( |
| 265 | Code* code, |
| 266 | String* input, // This needs to be the unpacked (sliced, cons) string. |
| 267 | int start_offset, |
| 268 | const byte* input_start, |
| 269 | const byte* input_end, |
| 270 | int* output, |
| 271 | int output_size, |
| 272 | Isolate* isolate) { |
| 273 | // Ensure that the minimum stack has been allocated. |
| 274 | RegExpStackScope stack_scope(isolate); |
| 275 | Address stack_base = stack_scope.stack()->stack_base(); |
| 276 | |
| 277 | int direct_call = 0; |
| 278 | int result = CALL_GENERATED_REGEXP_CODE( |
| 279 | isolate, code->entry(), input, start_offset, input_start, input_end, |
| 280 | output, output_size, stack_base, direct_call, isolate); |
| 281 | DCHECK(result >= RETRY); |
| 282 | |
| 283 | if (result == EXCEPTION && !isolate->has_pending_exception()) { |
| 284 | // We detected a stack overflow (on the backtrack stack) in RegExp code, |
| 285 | // but haven't created the exception yet. |
| 286 | isolate->StackOverflow(); |
| 287 | } |
| 288 | return static_cast<Result>(result); |
| 289 | } |
| 290 | |
| 291 | |
| 292 | const byte NativeRegExpMacroAssembler::word_character_map[] = { |
| 293 | 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, |
| 294 | 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, |
| 295 | 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, |
| 296 | 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, |
| 297 | |
| 298 | 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, |
| 299 | 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, |
| 300 | 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, // '0' - '7' |
| 301 | 0xffu, 0xffu, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, // '8' - '9' |
| 302 | |
| 303 | 0x00u, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, // 'A' - 'G' |
| 304 | 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, // 'H' - 'O' |
| 305 | 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, // 'P' - 'W' |
| 306 | 0xffu, 0xffu, 0xffu, 0x00u, 0x00u, 0x00u, 0x00u, 0xffu, // 'X' - 'Z', '_' |
| 307 | |
| 308 | 0x00u, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, // 'a' - 'g' |
| 309 | 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, // 'h' - 'o' |
| 310 | 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, 0xffu, // 'p' - 'w' |
| 311 | 0xffu, 0xffu, 0xffu, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, // 'x' - 'z' |
| 312 | // Latin-1 range |
| 313 | 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, |
| 314 | 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, |
| 315 | 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, |
| 316 | 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, |
| 317 | |
| 318 | 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, |
| 319 | 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, |
| 320 | 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, |
| 321 | 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, |
| 322 | |
| 323 | 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, |
| 324 | 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, |
| 325 | 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, |
| 326 | 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, |
| 327 | |
| 328 | 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, |
| 329 | 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, |
| 330 | 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, |
| 331 | 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, 0x00u, |
| 332 | }; |
| 333 | |
| 334 | |
Ben Murdoch | 4a90d5f | 2016-03-22 12:00:34 +0000 | [diff] [blame] | 335 | Address NativeRegExpMacroAssembler::GrowStack(Address stack_pointer, |
| 336 | Address* stack_base, |
| 337 | Isolate* isolate) { |
| 338 | RegExpStack* regexp_stack = isolate->regexp_stack(); |
| 339 | size_t size = regexp_stack->stack_capacity(); |
| 340 | Address old_stack_base = regexp_stack->stack_base(); |
| 341 | DCHECK(old_stack_base == *stack_base); |
| 342 | DCHECK(stack_pointer <= old_stack_base); |
| 343 | DCHECK(static_cast<size_t>(old_stack_base - stack_pointer) <= size); |
| 344 | Address new_stack_base = regexp_stack->EnsureCapacity(size * 2); |
| 345 | if (new_stack_base == NULL) { |
| 346 | return NULL; |
| 347 | } |
| 348 | *stack_base = new_stack_base; |
| 349 | intptr_t stack_content_size = old_stack_base - stack_pointer; |
| 350 | return new_stack_base - stack_content_size; |
| 351 | } |
| 352 | |
| 353 | #endif // V8_INTERPRETED_REGEXP |
| 354 | |
| 355 | } // namespace internal |
| 356 | } // namespace v8 |