utils.cc revision ea46f950e7a51585db293cd7f047de190a482414
1/* 2 * Copyright (C) 2011 The Android Open Source Project 3 * 4 * Licensed under the Apache License, Version 2.0 (the "License"); 5 * you may not use this file except in compliance with the License. 6 * You may obtain a copy of the License at 7 * 8 * http://www.apache.org/licenses/LICENSE-2.0 9 * 10 * Unless required by applicable law or agreed to in writing, software 11 * distributed under the License is distributed on an "AS IS" BASIS, 12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 13 * See the License for the specific language governing permissions and 14 * limitations under the License. 15 */ 16 17#include "utils.h" 18 19#include <pthread.h> 20#include <sys/stat.h> 21#include <sys/syscall.h> 22#include <sys/types.h> 23#include <unistd.h> 24 25#include "UniquePtr.h" 26#include "base/unix_file/fd_file.h" 27#include "dex_file-inl.h" 28#include "mirror/art_field-inl.h" 29#include "mirror/art_method-inl.h" 30#include "mirror/class-inl.h" 31#include "mirror/class_loader.h" 32#include "mirror/object-inl.h" 33#include "mirror/object_array-inl.h" 34#include "mirror/string.h" 35#include "object_utils.h" 36#include "os.h" 37#include "utf.h" 38 39#if !defined(HAVE_POSIX_CLOCKS) 40#include <sys/time.h> 41#endif 42 43#if defined(HAVE_PRCTL) 44#include <sys/prctl.h> 45#endif 46 47#if defined(__APPLE__) 48#include "AvailabilityMacros.h" // For MAC_OS_X_VERSION_MAX_ALLOWED 49#include <sys/syscall.h> 50#endif 51 52#include <corkscrew/backtrace.h> // For DumpNativeStack. 53#include <corkscrew/demangle.h> // For DumpNativeStack. 54 55#if defined(__linux__) 56#include <linux/unistd.h> 57#endif 58 59namespace art { 60 61pid_t GetTid() { 62#if defined(__APPLE__) 63 uint64_t owner; 64 CHECK_PTHREAD_CALL(pthread_threadid_np, (NULL, &owner), __FUNCTION__); // Requires Mac OS 10.6 65 return owner; 66#else 67 // Neither bionic nor glibc exposes gettid(2). 68 return syscall(__NR_gettid); 69#endif 70} 71 72std::string GetThreadName(pid_t tid) { 73 std::string result; 74 if (ReadFileToString(StringPrintf("/proc/self/task/%d/comm", tid), &result)) { 75 result.resize(result.size() - 1); // Lose the trailing '\n'. 76 } else { 77 result = "<unknown>"; 78 } 79 return result; 80} 81 82void GetThreadStack(pthread_t thread, void*& stack_base, size_t& stack_size) { 83#if defined(__APPLE__) 84 stack_size = pthread_get_stacksize_np(thread); 85 void* stack_addr = pthread_get_stackaddr_np(thread); 86 87 // Check whether stack_addr is the base or end of the stack. 88 // (On Mac OS 10.7, it's the end.) 89 int stack_variable; 90 if (stack_addr > &stack_variable) { 91 stack_base = reinterpret_cast<byte*>(stack_addr) - stack_size; 92 } else { 93 stack_base = stack_addr; 94 } 95#else 96 pthread_attr_t attributes; 97 CHECK_PTHREAD_CALL(pthread_getattr_np, (thread, &attributes), __FUNCTION__); 98 CHECK_PTHREAD_CALL(pthread_attr_getstack, (&attributes, &stack_base, &stack_size), __FUNCTION__); 99 CHECK_PTHREAD_CALL(pthread_attr_destroy, (&attributes), __FUNCTION__); 100#endif 101} 102 103bool ReadFileToString(const std::string& file_name, std::string* result) { 104 UniquePtr<File> file(new File); 105 if (!file->Open(file_name, O_RDONLY)) { 106 return false; 107 } 108 109 std::vector<char> buf(8 * KB); 110 while (true) { 111 int64_t n = TEMP_FAILURE_RETRY(read(file->Fd(), &buf[0], buf.size())); 112 if (n == -1) { 113 return false; 114 } 115 if (n == 0) { 116 return true; 117 } 118 result->append(&buf[0], n); 119 } 120} 121 122std::string GetIsoDate() { 123 time_t now = time(NULL); 124 tm tmbuf; 125 tm* ptm = localtime_r(&now, &tmbuf); 126 return StringPrintf("%04d-%02d-%02d %02d:%02d:%02d", 127 ptm->tm_year + 1900, ptm->tm_mon+1, ptm->tm_mday, 128 ptm->tm_hour, ptm->tm_min, ptm->tm_sec); 129} 130 131uint64_t MilliTime() { 132#if defined(HAVE_POSIX_CLOCKS) 133 timespec now; 134 clock_gettime(CLOCK_MONOTONIC, &now); 135 return static_cast<uint64_t>(now.tv_sec) * 1000LL + now.tv_nsec / 1000000LL; 136#else 137 timeval now; 138 gettimeofday(&now, NULL); 139 return static_cast<uint64_t>(now.tv_sec) * 1000LL + now.tv_usec / 1000LL; 140#endif 141} 142 143uint64_t MicroTime() { 144#if defined(HAVE_POSIX_CLOCKS) 145 timespec now; 146 clock_gettime(CLOCK_MONOTONIC, &now); 147 return static_cast<uint64_t>(now.tv_sec) * 1000000LL + now.tv_nsec / 1000LL; 148#else 149 timeval now; 150 gettimeofday(&now, NULL); 151 return static_cast<uint64_t>(now.tv_sec) * 1000000LL + now.tv_usec; 152#endif 153} 154 155uint64_t NanoTime() { 156#if defined(HAVE_POSIX_CLOCKS) 157 timespec now; 158 clock_gettime(CLOCK_MONOTONIC, &now); 159 return static_cast<uint64_t>(now.tv_sec) * 1000000000LL + now.tv_nsec; 160#else 161 timeval now; 162 gettimeofday(&now, NULL); 163 return static_cast<uint64_t>(now.tv_sec) * 1000000000LL + now.tv_usec * 1000LL; 164#endif 165} 166 167uint64_t ThreadCpuMicroTime() { 168#if defined(HAVE_POSIX_CLOCKS) 169 timespec now; 170 clock_gettime(CLOCK_THREAD_CPUTIME_ID, &now); 171 return static_cast<uint64_t>(now.tv_sec) * 1000000LL + now.tv_nsec / 1000LL; 172#else 173 UNIMPLEMENTED(WARNING); 174 return -1; 175#endif 176} 177 178uint64_t ThreadCpuNanoTime() { 179#if defined(HAVE_POSIX_CLOCKS) 180 timespec now; 181 clock_gettime(CLOCK_THREAD_CPUTIME_ID, &now); 182 return static_cast<uint64_t>(now.tv_sec) * 1000000000LL + now.tv_nsec; 183#else 184 UNIMPLEMENTED(WARNING); 185 return -1; 186#endif 187} 188 189void NanoSleep(uint64_t ns) { 190 timespec tm; 191 tm.tv_sec = 0; 192 tm.tv_nsec = ns; 193 nanosleep(&tm, NULL); 194} 195 196void InitTimeSpec(bool absolute, int clock, int64_t ms, int32_t ns, timespec* ts) { 197 int64_t endSec; 198 199 if (absolute) { 200#if !defined(__APPLE__) 201 clock_gettime(clock, ts); 202#else 203 UNUSED(clock); 204 timeval tv; 205 gettimeofday(&tv, NULL); 206 ts->tv_sec = tv.tv_sec; 207 ts->tv_nsec = tv.tv_usec * 1000; 208#endif 209 } else { 210 ts->tv_sec = 0; 211 ts->tv_nsec = 0; 212 } 213 endSec = ts->tv_sec + ms / 1000; 214 if (UNLIKELY(endSec >= 0x7fffffff)) { 215 std::ostringstream ss; 216 LOG(INFO) << "Note: end time exceeds epoch: " << ss.str(); 217 endSec = 0x7ffffffe; 218 } 219 ts->tv_sec = endSec; 220 ts->tv_nsec = (ts->tv_nsec + (ms % 1000) * 1000000) + ns; 221 222 // Catch rollover. 223 if (ts->tv_nsec >= 1000000000L) { 224 ts->tv_sec++; 225 ts->tv_nsec -= 1000000000L; 226 } 227} 228 229std::string PrettyDescriptor(const mirror::String* java_descriptor) { 230 if (java_descriptor == NULL) { 231 return "null"; 232 } 233 return PrettyDescriptor(java_descriptor->ToModifiedUtf8()); 234} 235 236std::string PrettyDescriptor(const mirror::Class* klass) { 237 if (klass == NULL) { 238 return "null"; 239 } 240 return PrettyDescriptor(ClassHelper(klass).GetDescriptor()); 241} 242 243std::string PrettyDescriptor(const std::string& descriptor) { 244 // Count the number of '['s to get the dimensionality. 245 const char* c = descriptor.c_str(); 246 size_t dim = 0; 247 while (*c == '[') { 248 dim++; 249 c++; 250 } 251 252 // Reference or primitive? 253 if (*c == 'L') { 254 // "[[La/b/C;" -> "a.b.C[][]". 255 c++; // Skip the 'L'. 256 } else { 257 // "[[B" -> "byte[][]". 258 // To make life easier, we make primitives look like unqualified 259 // reference types. 260 switch (*c) { 261 case 'B': c = "byte;"; break; 262 case 'C': c = "char;"; break; 263 case 'D': c = "double;"; break; 264 case 'F': c = "float;"; break; 265 case 'I': c = "int;"; break; 266 case 'J': c = "long;"; break; 267 case 'S': c = "short;"; break; 268 case 'Z': c = "boolean;"; break; 269 case 'V': c = "void;"; break; // Used when decoding return types. 270 default: return descriptor; 271 } 272 } 273 274 // At this point, 'c' is a string of the form "fully/qualified/Type;" 275 // or "primitive;". Rewrite the type with '.' instead of '/': 276 std::string result; 277 const char* p = c; 278 while (*p != ';') { 279 char ch = *p++; 280 if (ch == '/') { 281 ch = '.'; 282 } 283 result.push_back(ch); 284 } 285 // ...and replace the semicolon with 'dim' "[]" pairs: 286 while (dim--) { 287 result += "[]"; 288 } 289 return result; 290} 291 292std::string PrettyDescriptor(Primitive::Type type) { 293 std::string descriptor_string(Primitive::Descriptor(type)); 294 return PrettyDescriptor(descriptor_string); 295} 296 297std::string PrettyField(const mirror::ArtField* f, bool with_type) { 298 if (f == NULL) { 299 return "null"; 300 } 301 FieldHelper fh(f); 302 std::string result; 303 if (with_type) { 304 result += PrettyDescriptor(fh.GetTypeDescriptor()); 305 result += ' '; 306 } 307 result += PrettyDescriptor(fh.GetDeclaringClassDescriptor()); 308 result += '.'; 309 result += fh.GetName(); 310 return result; 311} 312 313std::string PrettyField(uint32_t field_idx, const DexFile& dex_file, bool with_type) { 314 if (field_idx >= dex_file.NumFieldIds()) { 315 return StringPrintf("<<invalid-field-idx-%d>>", field_idx); 316 } 317 const DexFile::FieldId& field_id = dex_file.GetFieldId(field_idx); 318 std::string result; 319 if (with_type) { 320 result += dex_file.GetFieldTypeDescriptor(field_id); 321 result += ' '; 322 } 323 result += PrettyDescriptor(dex_file.GetFieldDeclaringClassDescriptor(field_id)); 324 result += '.'; 325 result += dex_file.GetFieldName(field_id); 326 return result; 327} 328 329std::string PrettyType(uint32_t type_idx, const DexFile& dex_file) { 330 if (type_idx >= dex_file.NumTypeIds()) { 331 return StringPrintf("<<invalid-type-idx-%d>>", type_idx); 332 } 333 const DexFile::TypeId& type_id = dex_file.GetTypeId(type_idx); 334 return PrettyDescriptor(dex_file.GetTypeDescriptor(type_id)); 335} 336 337std::string PrettyArguments(const char* signature) { 338 std::string result; 339 result += '('; 340 CHECK_EQ(*signature, '('); 341 ++signature; // Skip the '('. 342 while (*signature != ')') { 343 size_t argument_length = 0; 344 while (signature[argument_length] == '[') { 345 ++argument_length; 346 } 347 if (signature[argument_length] == 'L') { 348 argument_length = (strchr(signature, ';') - signature + 1); 349 } else { 350 ++argument_length; 351 } 352 std::string argument_descriptor(signature, argument_length); 353 result += PrettyDescriptor(argument_descriptor); 354 if (signature[argument_length] != ')') { 355 result += ", "; 356 } 357 signature += argument_length; 358 } 359 CHECK_EQ(*signature, ')'); 360 ++signature; // Skip the ')'. 361 result += ')'; 362 return result; 363} 364 365std::string PrettyReturnType(const char* signature) { 366 const char* return_type = strchr(signature, ')'); 367 CHECK(return_type != NULL); 368 ++return_type; // Skip ')'. 369 return PrettyDescriptor(return_type); 370} 371 372std::string PrettyMethod(const mirror::ArtMethod* m, bool with_signature) { 373 if (m == NULL) { 374 return "null"; 375 } 376 MethodHelper mh(m); 377 std::string result(PrettyDescriptor(mh.GetDeclaringClassDescriptor())); 378 result += '.'; 379 result += mh.GetName(); 380 if (with_signature) { 381 std::string signature(mh.GetSignature()); 382 if (signature == "<no signature>") { 383 return result + signature; 384 } 385 result = PrettyReturnType(signature.c_str()) + " " + result + PrettyArguments(signature.c_str()); 386 } 387 return result; 388} 389 390std::string PrettyMethod(uint32_t method_idx, const DexFile& dex_file, bool with_signature) { 391 if (method_idx >= dex_file.NumMethodIds()) { 392 return StringPrintf("<<invalid-method-idx-%d>>", method_idx); 393 } 394 const DexFile::MethodId& method_id = dex_file.GetMethodId(method_idx); 395 std::string result(PrettyDescriptor(dex_file.GetMethodDeclaringClassDescriptor(method_id))); 396 result += '.'; 397 result += dex_file.GetMethodName(method_id); 398 if (with_signature) { 399 std::string signature(dex_file.GetMethodSignature(method_id)); 400 if (signature == "<no signature>") { 401 return result + signature; 402 } 403 result = PrettyReturnType(signature.c_str()) + " " + result + PrettyArguments(signature.c_str()); 404 } 405 return result; 406} 407 408std::string PrettyTypeOf(const mirror::Object* obj) { 409 if (obj == NULL) { 410 return "null"; 411 } 412 if (obj->GetClass() == NULL) { 413 return "(raw)"; 414 } 415 ClassHelper kh(obj->GetClass()); 416 std::string result(PrettyDescriptor(kh.GetDescriptor())); 417 if (obj->IsClass()) { 418 kh.ChangeClass(obj->AsClass()); 419 result += "<" + PrettyDescriptor(kh.GetDescriptor()) + ">"; 420 } 421 return result; 422} 423 424std::string PrettyClass(const mirror::Class* c) { 425 if (c == NULL) { 426 return "null"; 427 } 428 std::string result; 429 result += "java.lang.Class<"; 430 result += PrettyDescriptor(c); 431 result += ">"; 432 return result; 433} 434 435std::string PrettyClassAndClassLoader(const mirror::Class* c) { 436 if (c == NULL) { 437 return "null"; 438 } 439 std::string result; 440 result += "java.lang.Class<"; 441 result += PrettyDescriptor(c); 442 result += ","; 443 result += PrettyTypeOf(c->GetClassLoader()); 444 // TODO: add an identifying hash value for the loader 445 result += ">"; 446 return result; 447} 448 449std::string PrettySize(size_t byte_count) { 450 // The byte thresholds at which we display amounts. A byte count is displayed 451 // in unit U when kUnitThresholds[U] <= bytes < kUnitThresholds[U+1]. 452 static const size_t kUnitThresholds[] = { 453 0, // B up to... 454 3*1024, // KB up to... 455 2*1024*1024, // MB up to... 456 1024*1024*1024 // GB from here. 457 }; 458 static const size_t kBytesPerUnit[] = { 1, KB, MB, GB }; 459 static const char* const kUnitStrings[] = { "B", "KB", "MB", "GB" }; 460 461 int i = arraysize(kUnitThresholds); 462 while (--i > 0) { 463 if (byte_count >= kUnitThresholds[i]) { 464 break; 465 } 466 } 467 468 return StringPrintf("%zd%s", byte_count / kBytesPerUnit[i], kUnitStrings[i]); 469} 470 471std::string PrettyDuration(uint64_t nano_duration) { 472 if (nano_duration == 0) { 473 return "0"; 474 } else { 475 return FormatDuration(nano_duration, GetAppropriateTimeUnit(nano_duration)); 476 } 477} 478 479TimeUnit GetAppropriateTimeUnit(uint64_t nano_duration) { 480 const uint64_t one_sec = 1000 * 1000 * 1000; 481 const uint64_t one_ms = 1000 * 1000; 482 const uint64_t one_us = 1000; 483 if (nano_duration >= one_sec) { 484 return kTimeUnitSecond; 485 } else if (nano_duration >= one_ms) { 486 return kTimeUnitMillisecond; 487 } else if (nano_duration >= one_us) { 488 return kTimeUnitMicrosecond; 489 } else { 490 return kTimeUnitNanosecond; 491 } 492} 493 494uint64_t GetNsToTimeUnitDivisor(TimeUnit time_unit) { 495 const uint64_t one_sec = 1000 * 1000 * 1000; 496 const uint64_t one_ms = 1000 * 1000; 497 const uint64_t one_us = 1000; 498 499 switch (time_unit) { 500 case kTimeUnitSecond: 501 return one_sec; 502 case kTimeUnitMillisecond: 503 return one_ms; 504 case kTimeUnitMicrosecond: 505 return one_us; 506 case kTimeUnitNanosecond: 507 return 1; 508 } 509 return 0; 510} 511 512std::string FormatDuration(uint64_t nano_duration, TimeUnit time_unit) { 513 const char* unit = NULL; 514 uint64_t divisor = GetNsToTimeUnitDivisor(time_unit); 515 uint32_t zero_fill = 1; 516 switch (time_unit) { 517 case kTimeUnitSecond: 518 unit = "s"; 519 zero_fill = 9; 520 break; 521 case kTimeUnitMillisecond: 522 unit = "ms"; 523 zero_fill = 6; 524 break; 525 case kTimeUnitMicrosecond: 526 unit = "us"; 527 zero_fill = 3; 528 break; 529 case kTimeUnitNanosecond: 530 unit = "ns"; 531 zero_fill = 0; 532 break; 533 } 534 535 uint64_t whole_part = nano_duration / divisor; 536 uint64_t fractional_part = nano_duration % divisor; 537 if (fractional_part == 0) { 538 return StringPrintf("%llu%s", whole_part, unit); 539 } else { 540 while ((fractional_part % 1000) == 0) { 541 zero_fill -= 3; 542 fractional_part /= 1000; 543 } 544 if (zero_fill == 3) { 545 return StringPrintf("%llu.%03llu%s", whole_part, fractional_part, unit); 546 } else if (zero_fill == 6) { 547 return StringPrintf("%llu.%06llu%s", whole_part, fractional_part, unit); 548 } else { 549 return StringPrintf("%llu.%09llu%s", whole_part, fractional_part, unit); 550 } 551 } 552} 553 554std::string PrintableString(const std::string& utf) { 555 std::string result; 556 result += '"'; 557 const char* p = utf.c_str(); 558 size_t char_count = CountModifiedUtf8Chars(p); 559 for (size_t i = 0; i < char_count; ++i) { 560 uint16_t ch = GetUtf16FromUtf8(&p); 561 if (ch == '\\') { 562 result += "\\\\"; 563 } else if (ch == '\n') { 564 result += "\\n"; 565 } else if (ch == '\r') { 566 result += "\\r"; 567 } else if (ch == '\t') { 568 result += "\\t"; 569 } else if (NeedsEscaping(ch)) { 570 StringAppendF(&result, "\\u%04x", ch); 571 } else { 572 result += ch; 573 } 574 } 575 result += '"'; 576 return result; 577} 578 579// See http://java.sun.com/j2se/1.5.0/docs/guide/jni/spec/design.html#wp615 for the full rules. 580std::string MangleForJni(const std::string& s) { 581 std::string result; 582 size_t char_count = CountModifiedUtf8Chars(s.c_str()); 583 const char* cp = &s[0]; 584 for (size_t i = 0; i < char_count; ++i) { 585 uint16_t ch = GetUtf16FromUtf8(&cp); 586 if ((ch >= 'A' && ch <= 'Z') || (ch >= 'a' && ch <= 'z') || (ch >= '0' && ch <= '9')) { 587 result.push_back(ch); 588 } else if (ch == '.' || ch == '/') { 589 result += "_"; 590 } else if (ch == '_') { 591 result += "_1"; 592 } else if (ch == ';') { 593 result += "_2"; 594 } else if (ch == '[') { 595 result += "_3"; 596 } else { 597 StringAppendF(&result, "_0%04x", ch); 598 } 599 } 600 return result; 601} 602 603std::string DotToDescriptor(const char* class_name) { 604 std::string descriptor(class_name); 605 std::replace(descriptor.begin(), descriptor.end(), '.', '/'); 606 if (descriptor.length() > 0 && descriptor[0] != '[') { 607 descriptor = "L" + descriptor + ";"; 608 } 609 return descriptor; 610} 611 612std::string DescriptorToDot(const char* descriptor) { 613 size_t length = strlen(descriptor); 614 if (descriptor[0] == 'L' && descriptor[length - 1] == ';') { 615 std::string result(descriptor + 1, length - 2); 616 std::replace(result.begin(), result.end(), '/', '.'); 617 return result; 618 } 619 return descriptor; 620} 621 622std::string DescriptorToName(const char* descriptor) { 623 size_t length = strlen(descriptor); 624 if (descriptor[0] == 'L' && descriptor[length - 1] == ';') { 625 std::string result(descriptor + 1, length - 2); 626 return result; 627 } 628 return descriptor; 629} 630 631std::string JniShortName(const mirror::ArtMethod* m) { 632 MethodHelper mh(m); 633 std::string class_name(mh.GetDeclaringClassDescriptor()); 634 // Remove the leading 'L' and trailing ';'... 635 CHECK_EQ(class_name[0], 'L') << class_name; 636 CHECK_EQ(class_name[class_name.size() - 1], ';') << class_name; 637 class_name.erase(0, 1); 638 class_name.erase(class_name.size() - 1, 1); 639 640 std::string method_name(mh.GetName()); 641 642 std::string short_name; 643 short_name += "Java_"; 644 short_name += MangleForJni(class_name); 645 short_name += "_"; 646 short_name += MangleForJni(method_name); 647 return short_name; 648} 649 650std::string JniLongName(const mirror::ArtMethod* m) { 651 std::string long_name; 652 long_name += JniShortName(m); 653 long_name += "__"; 654 655 std::string signature(MethodHelper(m).GetSignature()); 656 signature.erase(0, 1); 657 signature.erase(signature.begin() + signature.find(')'), signature.end()); 658 659 long_name += MangleForJni(signature); 660 661 return long_name; 662} 663 664// Helper for IsValidPartOfMemberNameUtf8(), a bit vector indicating valid low ascii. 665uint32_t DEX_MEMBER_VALID_LOW_ASCII[4] = { 666 0x00000000, // 00..1f low control characters; nothing valid 667 0x03ff2010, // 20..3f digits and symbols; valid: '0'..'9', '$', '-' 668 0x87fffffe, // 40..5f uppercase etc.; valid: 'A'..'Z', '_' 669 0x07fffffe // 60..7f lowercase etc.; valid: 'a'..'z' 670}; 671 672// Helper for IsValidPartOfMemberNameUtf8(); do not call directly. 673bool IsValidPartOfMemberNameUtf8Slow(const char** pUtf8Ptr) { 674 /* 675 * It's a multibyte encoded character. Decode it and analyze. We 676 * accept anything that isn't (a) an improperly encoded low value, 677 * (b) an improper surrogate pair, (c) an encoded '\0', (d) a high 678 * control character, or (e) a high space, layout, or special 679 * character (U+00a0, U+2000..U+200f, U+2028..U+202f, 680 * U+fff0..U+ffff). This is all specified in the dex format 681 * document. 682 */ 683 684 uint16_t utf16 = GetUtf16FromUtf8(pUtf8Ptr); 685 686 // Perform follow-up tests based on the high 8 bits. 687 switch (utf16 >> 8) { 688 case 0x00: 689 // It's only valid if it's above the ISO-8859-1 high space (0xa0). 690 return (utf16 > 0x00a0); 691 case 0xd8: 692 case 0xd9: 693 case 0xda: 694 case 0xdb: 695 // It's a leading surrogate. Check to see that a trailing 696 // surrogate follows. 697 utf16 = GetUtf16FromUtf8(pUtf8Ptr); 698 return (utf16 >= 0xdc00) && (utf16 <= 0xdfff); 699 case 0xdc: 700 case 0xdd: 701 case 0xde: 702 case 0xdf: 703 // It's a trailing surrogate, which is not valid at this point. 704 return false; 705 case 0x20: 706 case 0xff: 707 // It's in the range that has spaces, controls, and specials. 708 switch (utf16 & 0xfff8) { 709 case 0x2000: 710 case 0x2008: 711 case 0x2028: 712 case 0xfff0: 713 case 0xfff8: 714 return false; 715 } 716 break; 717 } 718 return true; 719} 720 721/* Return whether the pointed-at modified-UTF-8 encoded character is 722 * valid as part of a member name, updating the pointer to point past 723 * the consumed character. This will consume two encoded UTF-16 code 724 * points if the character is encoded as a surrogate pair. Also, if 725 * this function returns false, then the given pointer may only have 726 * been partially advanced. 727 */ 728bool IsValidPartOfMemberNameUtf8(const char** pUtf8Ptr) { 729 uint8_t c = (uint8_t) **pUtf8Ptr; 730 if (c <= 0x7f) { 731 // It's low-ascii, so check the table. 732 uint32_t wordIdx = c >> 5; 733 uint32_t bitIdx = c & 0x1f; 734 (*pUtf8Ptr)++; 735 return (DEX_MEMBER_VALID_LOW_ASCII[wordIdx] & (1 << bitIdx)) != 0; 736 } 737 738 // It's a multibyte encoded character. Call a non-inline function 739 // for the heavy lifting. 740 return IsValidPartOfMemberNameUtf8Slow(pUtf8Ptr); 741} 742 743bool IsValidMemberName(const char* s) { 744 bool angle_name = false; 745 746 switch (*s) { 747 case '\0': 748 // The empty string is not a valid name. 749 return false; 750 case '<': 751 angle_name = true; 752 s++; 753 break; 754 } 755 756 while (true) { 757 switch (*s) { 758 case '\0': 759 return !angle_name; 760 case '>': 761 return angle_name && s[1] == '\0'; 762 } 763 764 if (!IsValidPartOfMemberNameUtf8(&s)) { 765 return false; 766 } 767 } 768} 769 770enum ClassNameType { kName, kDescriptor }; 771bool IsValidClassName(const char* s, ClassNameType type, char separator) { 772 int arrayCount = 0; 773 while (*s == '[') { 774 arrayCount++; 775 s++; 776 } 777 778 if (arrayCount > 255) { 779 // Arrays may have no more than 255 dimensions. 780 return false; 781 } 782 783 if (arrayCount != 0) { 784 /* 785 * If we're looking at an array of some sort, then it doesn't 786 * matter if what is being asked for is a class name; the 787 * format looks the same as a type descriptor in that case, so 788 * treat it as such. 789 */ 790 type = kDescriptor; 791 } 792 793 if (type == kDescriptor) { 794 /* 795 * We are looking for a descriptor. Either validate it as a 796 * single-character primitive type, or continue on to check the 797 * embedded class name (bracketed by "L" and ";"). 798 */ 799 switch (*(s++)) { 800 case 'B': 801 case 'C': 802 case 'D': 803 case 'F': 804 case 'I': 805 case 'J': 806 case 'S': 807 case 'Z': 808 // These are all single-character descriptors for primitive types. 809 return (*s == '\0'); 810 case 'V': 811 // Non-array void is valid, but you can't have an array of void. 812 return (arrayCount == 0) && (*s == '\0'); 813 case 'L': 814 // Class name: Break out and continue below. 815 break; 816 default: 817 // Oddball descriptor character. 818 return false; 819 } 820 } 821 822 /* 823 * We just consumed the 'L' that introduces a class name as part 824 * of a type descriptor, or we are looking for an unadorned class 825 * name. 826 */ 827 828 bool sepOrFirst = true; // first character or just encountered a separator. 829 for (;;) { 830 uint8_t c = (uint8_t) *s; 831 switch (c) { 832 case '\0': 833 /* 834 * Premature end for a type descriptor, but valid for 835 * a class name as long as we haven't encountered an 836 * empty component (including the degenerate case of 837 * the empty string ""). 838 */ 839 return (type == kName) && !sepOrFirst; 840 case ';': 841 /* 842 * Invalid character for a class name, but the 843 * legitimate end of a type descriptor. In the latter 844 * case, make sure that this is the end of the string 845 * and that it doesn't end with an empty component 846 * (including the degenerate case of "L;"). 847 */ 848 return (type == kDescriptor) && !sepOrFirst && (s[1] == '\0'); 849 case '/': 850 case '.': 851 if (c != separator) { 852 // The wrong separator character. 853 return false; 854 } 855 if (sepOrFirst) { 856 // Separator at start or two separators in a row. 857 return false; 858 } 859 sepOrFirst = true; 860 s++; 861 break; 862 default: 863 if (!IsValidPartOfMemberNameUtf8(&s)) { 864 return false; 865 } 866 sepOrFirst = false; 867 break; 868 } 869 } 870} 871 872bool IsValidBinaryClassName(const char* s) { 873 return IsValidClassName(s, kName, '.'); 874} 875 876bool IsValidJniClassName(const char* s) { 877 return IsValidClassName(s, kName, '/'); 878} 879 880bool IsValidDescriptor(const char* s) { 881 return IsValidClassName(s, kDescriptor, '/'); 882} 883 884void Split(const std::string& s, char separator, std::vector<std::string>& result) { 885 const char* p = s.data(); 886 const char* end = p + s.size(); 887 while (p != end) { 888 if (*p == separator) { 889 ++p; 890 } else { 891 const char* start = p; 892 while (++p != end && *p != separator) { 893 // Skip to the next occurrence of the separator. 894 } 895 result.push_back(std::string(start, p - start)); 896 } 897 } 898} 899 900template <typename StringT> 901std::string Join(std::vector<StringT>& strings, char separator) { 902 if (strings.empty()) { 903 return ""; 904 } 905 906 std::string result(strings[0]); 907 for (size_t i = 1; i < strings.size(); ++i) { 908 result += separator; 909 result += strings[i]; 910 } 911 return result; 912} 913 914// Explicit instantiations. 915template std::string Join<std::string>(std::vector<std::string>& strings, char separator); 916template std::string Join<const char*>(std::vector<const char*>& strings, char separator); 917template std::string Join<char*>(std::vector<char*>& strings, char separator); 918 919bool StartsWith(const std::string& s, const char* prefix) { 920 return s.compare(0, strlen(prefix), prefix) == 0; 921} 922 923bool EndsWith(const std::string& s, const char* suffix) { 924 size_t suffix_length = strlen(suffix); 925 size_t string_length = s.size(); 926 if (suffix_length > string_length) { 927 return false; 928 } 929 size_t offset = string_length - suffix_length; 930 return s.compare(offset, suffix_length, suffix) == 0; 931} 932 933void SetThreadName(const char* thread_name) { 934 int hasAt = 0; 935 int hasDot = 0; 936 const char* s = thread_name; 937 while (*s) { 938 if (*s == '.') { 939 hasDot = 1; 940 } else if (*s == '@') { 941 hasAt = 1; 942 } 943 s++; 944 } 945 int len = s - thread_name; 946 if (len < 15 || hasAt || !hasDot) { 947 s = thread_name; 948 } else { 949 s = thread_name + len - 15; 950 } 951#if defined(HAVE_ANDROID_PTHREAD_SETNAME_NP) 952 // pthread_setname_np fails rather than truncating long strings. 953 char buf[16]; // MAX_TASK_COMM_LEN=16 is hard-coded into bionic 954 strncpy(buf, s, sizeof(buf)-1); 955 buf[sizeof(buf)-1] = '\0'; 956 errno = pthread_setname_np(pthread_self(), buf); 957 if (errno != 0) { 958 PLOG(WARNING) << "Unable to set the name of current thread to '" << buf << "'"; 959 } 960#elif defined(__APPLE__) && MAC_OS_X_VERSION_MAX_ALLOWED >= 1060 961 pthread_setname_np(thread_name); 962#elif defined(HAVE_PRCTL) 963 prctl(PR_SET_NAME, (unsigned long) s, 0, 0, 0); // NOLINT (unsigned long) 964#else 965 UNIMPLEMENTED(WARNING) << thread_name; 966#endif 967} 968 969void GetTaskStats(pid_t tid, char& state, int& utime, int& stime, int& task_cpu) { 970 utime = stime = task_cpu = 0; 971 std::string stats; 972 if (!ReadFileToString(StringPrintf("/proc/self/task/%d/stat", tid), &stats)) { 973 return; 974 } 975 // Skip the command, which may contain spaces. 976 stats = stats.substr(stats.find(')') + 2); 977 // Extract the three fields we care about. 978 std::vector<std::string> fields; 979 Split(stats, ' ', fields); 980 state = fields[0][0]; 981 utime = strtoull(fields[11].c_str(), NULL, 10); 982 stime = strtoull(fields[12].c_str(), NULL, 10); 983 task_cpu = strtoull(fields[36].c_str(), NULL, 10); 984} 985 986std::string GetSchedulerGroupName(pid_t tid) { 987 // /proc/<pid>/cgroup looks like this: 988 // 2:devices:/ 989 // 1:cpuacct,cpu:/ 990 // We want the third field from the line whose second field contains the "cpu" token. 991 std::string cgroup_file; 992 if (!ReadFileToString(StringPrintf("/proc/self/task/%d/cgroup", tid), &cgroup_file)) { 993 return ""; 994 } 995 std::vector<std::string> cgroup_lines; 996 Split(cgroup_file, '\n', cgroup_lines); 997 for (size_t i = 0; i < cgroup_lines.size(); ++i) { 998 std::vector<std::string> cgroup_fields; 999 Split(cgroup_lines[i], ':', cgroup_fields); 1000 std::vector<std::string> cgroups; 1001 Split(cgroup_fields[1], ',', cgroups); 1002 for (size_t i = 0; i < cgroups.size(); ++i) { 1003 if (cgroups[i] == "cpu") { 1004 return cgroup_fields[2].substr(1); // Skip the leading slash. 1005 } 1006 } 1007 } 1008 return ""; 1009} 1010 1011static const char* CleanMapName(const backtrace_symbol_t* symbol) { 1012 const char* map_name = symbol->map_name; 1013 if (map_name == NULL) { 1014 map_name = "???"; 1015 } 1016 // Turn "/usr/local/google/home/enh/clean-dalvik-dev/out/host/linux-x86/lib/libartd.so" 1017 // into "libartd.so". 1018 const char* last_slash = strrchr(map_name, '/'); 1019 if (last_slash != NULL) { 1020 map_name = last_slash + 1; 1021 } 1022 return map_name; 1023} 1024 1025static void FindSymbolInElf(const backtrace_frame_t* frame, const backtrace_symbol_t* symbol, 1026 std::string& symbol_name, uint32_t& pc_offset) { 1027 symbol_table_t* symbol_table = NULL; 1028 if (symbol->map_name != NULL) { 1029 symbol_table = load_symbol_table(symbol->map_name); 1030 } 1031 const symbol_t* elf_symbol = NULL; 1032 bool was_relative = true; 1033 if (symbol_table != NULL) { 1034 elf_symbol = find_symbol(symbol_table, symbol->relative_pc); 1035 if (elf_symbol == NULL) { 1036 elf_symbol = find_symbol(symbol_table, frame->absolute_pc); 1037 was_relative = false; 1038 } 1039 } 1040 if (elf_symbol != NULL) { 1041 const char* demangled_symbol_name = demangle_symbol_name(elf_symbol->name); 1042 if (demangled_symbol_name != NULL) { 1043 symbol_name = demangled_symbol_name; 1044 } else { 1045 symbol_name = elf_symbol->name; 1046 } 1047 1048 // TODO: is it a libcorkscrew bug that we have to do this? 1049 pc_offset = (was_relative ? symbol->relative_pc : frame->absolute_pc) - elf_symbol->start; 1050 } else { 1051 symbol_name = "???"; 1052 } 1053 free_symbol_table(symbol_table); 1054} 1055 1056void DumpNativeStack(std::ostream& os, pid_t tid, const char* prefix, bool include_count) { 1057 // Ensure libcorkscrew doesn't use a stale cache of /proc/self/maps. 1058 flush_my_map_info_list(); 1059 1060 const size_t MAX_DEPTH = 32; 1061 UniquePtr<backtrace_frame_t[]> frames(new backtrace_frame_t[MAX_DEPTH]); 1062 size_t ignore_count = 2; // Don't include unwind_backtrace_thread or DumpNativeStack. 1063 ssize_t frame_count = unwind_backtrace_thread(tid, frames.get(), ignore_count, MAX_DEPTH); 1064 if (frame_count == -1) { 1065 os << prefix << "(unwind_backtrace_thread failed for thread " << tid << ")\n"; 1066 return; 1067 } else if (frame_count == 0) { 1068 os << prefix << "(no native stack frames for thread " << tid << ")\n"; 1069 return; 1070 } 1071 1072 UniquePtr<backtrace_symbol_t[]> backtrace_symbols(new backtrace_symbol_t[frame_count]); 1073 get_backtrace_symbols(frames.get(), frame_count, backtrace_symbols.get()); 1074 1075 for (size_t i = 0; i < static_cast<size_t>(frame_count); ++i) { 1076 const backtrace_frame_t* frame = &frames[i]; 1077 const backtrace_symbol_t* symbol = &backtrace_symbols[i]; 1078 1079 // We produce output like this: 1080 // ] #00 unwind_backtrace_thread+536 [0x55d75bb8] (libcorkscrew.so) 1081 1082 std::string symbol_name; 1083 uint32_t pc_offset = 0; 1084 if (symbol->demangled_name != NULL) { 1085 symbol_name = symbol->demangled_name; 1086 pc_offset = symbol->relative_pc - symbol->relative_symbol_addr; 1087 } else if (symbol->symbol_name != NULL) { 1088 symbol_name = symbol->symbol_name; 1089 pc_offset = symbol->relative_pc - symbol->relative_symbol_addr; 1090 } else { 1091 // dladdr(3) didn't find a symbol; maybe it's static? Look in the ELF file... 1092 FindSymbolInElf(frame, symbol, symbol_name, pc_offset); 1093 } 1094 1095 os << prefix; 1096 if (include_count) { 1097 os << StringPrintf("#%02zd ", i); 1098 } 1099 os << symbol_name; 1100 if (pc_offset != 0) { 1101 os << "+" << pc_offset; 1102 } 1103 os << StringPrintf(" [%p] (%s)\n", 1104 reinterpret_cast<void*>(frame->absolute_pc), CleanMapName(symbol)); 1105 } 1106 1107 free_backtrace_symbols(backtrace_symbols.get(), frame_count); 1108} 1109 1110#if defined(__APPLE__) 1111 1112// TODO: is there any way to get the kernel stack on Mac OS? 1113void DumpKernelStack(std::ostream&, pid_t, const char*, bool) {} 1114 1115#else 1116 1117void DumpKernelStack(std::ostream& os, pid_t tid, const char* prefix, bool include_count) { 1118 if (tid == GetTid()) { 1119 // There's no point showing that we're reading our stack out of /proc! 1120 return; 1121 } 1122 1123 std::string kernel_stack_filename(StringPrintf("/proc/self/task/%d/stack", tid)); 1124 std::string kernel_stack; 1125 if (!ReadFileToString(kernel_stack_filename, &kernel_stack)) { 1126 os << prefix << "(couldn't read " << kernel_stack_filename << ")\n"; 1127 return; 1128 } 1129 1130 std::vector<std::string> kernel_stack_frames; 1131 Split(kernel_stack, '\n', kernel_stack_frames); 1132 // We skip the last stack frame because it's always equivalent to "[<ffffffff>] 0xffffffff", 1133 // which looking at the source appears to be the kernel's way of saying "that's all, folks!". 1134 kernel_stack_frames.pop_back(); 1135 for (size_t i = 0; i < kernel_stack_frames.size(); ++i) { 1136 // Turn "[<ffffffff8109156d>] futex_wait_queue_me+0xcd/0x110" into "futex_wait_queue_me+0xcd/0x110". 1137 const char* text = kernel_stack_frames[i].c_str(); 1138 const char* close_bracket = strchr(text, ']'); 1139 if (close_bracket != NULL) { 1140 text = close_bracket + 2; 1141 } 1142 os << prefix; 1143 if (include_count) { 1144 os << StringPrintf("#%02zd ", i); 1145 } 1146 os << text << "\n"; 1147 } 1148} 1149 1150#endif 1151 1152const char* GetAndroidRoot() { 1153 const char* android_root = getenv("ANDROID_ROOT"); 1154 if (android_root == NULL) { 1155 if (OS::DirectoryExists("/system")) { 1156 android_root = "/system"; 1157 } else { 1158 LOG(FATAL) << "ANDROID_ROOT not set and /system does not exist"; 1159 return ""; 1160 } 1161 } 1162 if (!OS::DirectoryExists(android_root)) { 1163 LOG(FATAL) << "Failed to find ANDROID_ROOT directory " << android_root; 1164 return ""; 1165 } 1166 return android_root; 1167} 1168 1169const char* GetAndroidData() { 1170 const char* android_data = getenv("ANDROID_DATA"); 1171 if (android_data == NULL) { 1172 if (OS::DirectoryExists("/data")) { 1173 android_data = "/data"; 1174 } else { 1175 LOG(FATAL) << "ANDROID_DATA not set and /data does not exist"; 1176 return ""; 1177 } 1178 } 1179 if (!OS::DirectoryExists(android_data)) { 1180 LOG(FATAL) << "Failed to find ANDROID_DATA directory " << android_data; 1181 return ""; 1182 } 1183 return android_data; 1184} 1185 1186std::string GetDalvikCacheOrDie(const char* android_data) { 1187 std::string dalvik_cache(StringPrintf("%s/dalvik-cache", android_data)); 1188 1189 if (!OS::DirectoryExists(dalvik_cache.c_str())) { 1190 if (StartsWith(dalvik_cache, "/tmp/")) { 1191 int result = mkdir(dalvik_cache.c_str(), 0700); 1192 if (result != 0) { 1193 LOG(FATAL) << "Failed to create dalvik-cache directory " << dalvik_cache; 1194 return ""; 1195 } 1196 } else { 1197 LOG(FATAL) << "Failed to find dalvik-cache directory " << dalvik_cache; 1198 return ""; 1199 } 1200 } 1201 return dalvik_cache; 1202} 1203 1204std::string GetDalvikCacheFilenameOrDie(const std::string& location) { 1205 std::string dalvik_cache(GetDalvikCacheOrDie(GetAndroidData())); 1206 if (location[0] != '/') { 1207 LOG(FATAL) << "Expected path in location to be absolute: "<< location; 1208 } 1209 std::string cache_file(location, 1); // skip leading slash 1210 if (!IsValidDexFilename(location)) { 1211 cache_file += "/"; 1212 cache_file += DexFile::kClassesDex; 1213 } 1214 std::replace(cache_file.begin(), cache_file.end(), '/', '@'); 1215 return dalvik_cache + "/" + cache_file; 1216} 1217 1218bool IsValidZipFilename(const std::string& filename) { 1219 if (filename.size() < 4) { 1220 return false; 1221 } 1222 std::string suffix(filename.substr(filename.size() - 4)); 1223 return (suffix == ".zip" || suffix == ".jar" || suffix == ".apk"); 1224} 1225 1226bool IsValidDexFilename(const std::string& filename) { 1227 return EndsWith(filename, ".dex"); 1228} 1229 1230bool IsValidOatFilename(const std::string& filename) { 1231 return (EndsWith(filename, ".odex") || 1232 EndsWith(filename, ".oat") || 1233 EndsWith(filename, DexFile::kClassesDex)); 1234} 1235 1236} // namespace art 1237