OLD | NEW |
(Empty) | |
| 1 // Copyright (c) 2006, Google Inc. |
| 2 // All rights reserved. |
| 3 // |
| 4 // Redistribution and use in source and binary forms, with or without |
| 5 // modification, are permitted provided that the following conditions are |
| 6 // met: |
| 7 // |
| 8 // * Redistributions of source code must retain the above copyright |
| 9 // notice, this list of conditions and the following disclaimer. |
| 10 // * Redistributions in binary form must reproduce the above |
| 11 // copyright notice, this list of conditions and the following disclaimer |
| 12 // in the documentation and/or other materials provided with the |
| 13 // distribution. |
| 14 // * Neither the name of Google Inc. nor the names of its |
| 15 // contributors may be used to endorse or promote products derived from |
| 16 // this software without specific prior written permission. |
| 17 // |
| 18 // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
| 19 // "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
| 20 // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR |
| 21 // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT |
| 22 // OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, |
| 23 // SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT |
| 24 // LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, |
| 25 // DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY |
| 26 // THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
| 27 // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE |
| 28 // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
| 29 // |
| 30 // Author: Satoru Takabayashi |
| 31 // Stack-footprint reduction work done by Raksit Ashok |
| 32 // |
| 33 // Implementation note: |
| 34 // |
| 35 // We don't use heaps but only use stacks. We want to reduce the |
| 36 // stack consumption so that the symbolizer can run on small stacks. |
| 37 // |
| 38 // Here are some numbers collected with GCC 4.1.0 on x86: |
| 39 // - sizeof(Elf32_Sym) = 16 |
| 40 // - sizeof(Elf32_Shdr) = 40 |
| 41 // - sizeof(Elf64_Sym) = 24 |
| 42 // - sizeof(Elf64_Shdr) = 64 |
| 43 // |
| 44 // This implementation is intended to be async-signal-safe but uses |
| 45 // some functions which are not guaranteed to be so, such as memchr() |
| 46 // and memmove(). We assume they are async-signal-safe. |
| 47 // |
| 48 |
| 49 #include "utilities.h" |
| 50 |
| 51 #if defined(HAVE_SYMBOLIZE) |
| 52 |
| 53 #include <limits> |
| 54 |
| 55 #include "symbolize.h" |
| 56 #include "demangle.h" |
| 57 |
| 58 _START_GOOGLE_NAMESPACE_ |
| 59 |
| 60 // We don't use assert() since it's not guaranteed to be |
| 61 // async-signal-safe. Instead we define a minimal assertion |
| 62 // macro. So far, we don't need pretty printing for __FILE__, etc. |
| 63 |
| 64 // A wrapper for abort() to make it callable in ? :. |
| 65 static int AssertFail() { |
| 66 abort(); |
| 67 return 0; // Should not reach. |
| 68 } |
| 69 |
| 70 #define SAFE_ASSERT(expr) ((expr) ? 0 : AssertFail()) |
| 71 |
| 72 static SymbolizeCallback g_symbolize_callback = NULL; |
| 73 void InstallSymbolizeCallback(SymbolizeCallback callback) { |
| 74 g_symbolize_callback = callback; |
| 75 } |
| 76 |
| 77 // This function wraps the Demangle function to provide an interface |
| 78 // where the input symbol is demangled in-place. |
| 79 // To keep stack consumption low, we would like this function to not |
| 80 // get inlined. |
| 81 static ATTRIBUTE_NOINLINE void DemangleInplace(char *out, int out_size) { |
| 82 char demangled[256]; // Big enough for sane demangled symbols. |
| 83 if (Demangle(out, demangled, sizeof(demangled))) { |
| 84 // Demangling succeeded. Copy to out if the space allows. |
| 85 int len = strlen(demangled); |
| 86 if (len + 1 <= out_size) { // +1 for '\0'. |
| 87 SAFE_ASSERT(len < sizeof(demangled)); |
| 88 memmove(out, demangled, len + 1); |
| 89 } |
| 90 } |
| 91 } |
| 92 |
| 93 _END_GOOGLE_NAMESPACE_ |
| 94 |
| 95 #if defined(__ELF__) |
| 96 |
| 97 #include <dlfcn.h> |
| 98 #include <elf.h> |
| 99 #include <errno.h> |
| 100 #include <fcntl.h> |
| 101 #include <limits.h> |
| 102 #include <link.h> // For ElfW() macro. |
| 103 #include <stdint.h> |
| 104 #include <stdio.h> |
| 105 #include <stdlib.h> |
| 106 #include <stddef.h> |
| 107 #include <string.h> |
| 108 #include <sys/stat.h> |
| 109 #include <sys/types.h> |
| 110 #include <unistd.h> |
| 111 |
| 112 #include "symbolize.h" |
| 113 #include "config.h" |
| 114 #include "glog/raw_logging.h" |
| 115 |
| 116 // Re-runs fn until it doesn't cause EINTR. |
| 117 #define NO_INTR(fn) do {} while ((fn) < 0 && errno == EINTR) |
| 118 |
| 119 _START_GOOGLE_NAMESPACE_ |
| 120 |
| 121 // Read up to "count" bytes from file descriptor "fd" into the buffer |
| 122 // starting at "buf" while handling short reads and EINTR. On |
| 123 // success, return the number of bytes read. Otherwise, return -1. |
| 124 static ssize_t ReadPersistent(const int fd, void *buf, const size_t count) { |
| 125 SAFE_ASSERT(fd >= 0); |
| 126 SAFE_ASSERT(count >= 0 && count <= std::numeric_limits<ssize_t>::max()); |
| 127 char *buf0 = reinterpret_cast<char *>(buf); |
| 128 ssize_t num_bytes = 0; |
| 129 while (num_bytes < count) { |
| 130 ssize_t len; |
| 131 NO_INTR(len = read(fd, buf0 + num_bytes, count - num_bytes)); |
| 132 if (len < 0) { // There was an error other than EINTR. |
| 133 return -1; |
| 134 } |
| 135 if (len == 0) { // Reached EOF. |
| 136 break; |
| 137 } |
| 138 num_bytes += len; |
| 139 } |
| 140 SAFE_ASSERT(num_bytes <= count); |
| 141 return num_bytes; |
| 142 } |
| 143 |
| 144 // Read up to "count" bytes from "offset" in the file pointed by file |
| 145 // descriptor "fd" into the buffer starting at "buf". On success, |
| 146 // return the number of bytes read. Otherwise, return -1. |
| 147 static ssize_t ReadFromOffset(const int fd, void *buf, |
| 148 const size_t count, const off_t offset) { |
| 149 off_t off = lseek(fd, offset, SEEK_SET); |
| 150 if (off == (off_t)-1) { |
| 151 return -1; |
| 152 } |
| 153 return ReadPersistent(fd, buf, count); |
| 154 } |
| 155 |
| 156 // Try reading exactly "count" bytes from "offset" bytes in a file |
| 157 // pointed by "fd" into the buffer starting at "buf" while handling |
| 158 // short reads and EINTR. On success, return true. Otherwise, return |
| 159 // false. |
| 160 static bool ReadFromOffsetExact(const int fd, void *buf, |
| 161 const size_t count, const off_t offset) { |
| 162 ssize_t len = ReadFromOffset(fd, buf, count, offset); |
| 163 return len == count; |
| 164 } |
| 165 |
| 166 // Returns elf_header.e_type if the file pointed by fd is an ELF binary. |
| 167 static int FileGetElfType(const int fd) { |
| 168 ElfW(Ehdr) elf_header; |
| 169 if (!ReadFromOffsetExact(fd, &elf_header, sizeof(elf_header), 0)) { |
| 170 return -1; |
| 171 } |
| 172 if (memcmp(elf_header.e_ident, ELFMAG, SELFMAG) != 0) { |
| 173 return -1; |
| 174 } |
| 175 return elf_header.e_type; |
| 176 } |
| 177 |
| 178 // Read the section headers in the given ELF binary, and if a section |
| 179 // of the specified type is found, set the output to this section header |
| 180 // and return true. Otherwise, return false. |
| 181 // To keep stack consumption low, we would like this function to not get |
| 182 // inlined. |
| 183 static ATTRIBUTE_NOINLINE bool |
| 184 GetSectionHeaderByType(const int fd, ElfW(Half) sh_num, const off_t sh_offset, |
| 185 ElfW(Word) type, ElfW(Shdr) *out) { |
| 186 // Read at most 16 section headers at a time to save read calls. |
| 187 ElfW(Shdr) buf[16]; |
| 188 for (int i = 0; i < sh_num;) { |
| 189 const ssize_t num_bytes_left = (sh_num - i) * sizeof(buf[0]); |
| 190 const ssize_t num_bytes_to_read = |
| 191 (sizeof(buf) > num_bytes_left) ? num_bytes_left : sizeof(buf); |
| 192 const ssize_t len = ReadFromOffset(fd, buf, num_bytes_to_read, |
| 193 sh_offset + i * sizeof(buf[0])); |
| 194 SAFE_ASSERT(len % sizeof(buf[0]) == 0); |
| 195 const ssize_t num_headers_in_buf = len / sizeof(buf[0]); |
| 196 SAFE_ASSERT(num_headers_in_buf <= sizeof(buf) / sizeof(buf[0])); |
| 197 for (int j = 0; j < num_headers_in_buf; ++j) { |
| 198 if (buf[j].sh_type == type) { |
| 199 *out = buf[j]; |
| 200 return true; |
| 201 } |
| 202 } |
| 203 i += num_headers_in_buf; |
| 204 } |
| 205 return false; |
| 206 } |
| 207 |
| 208 // There is no particular reason to limit section name to 63 characters, |
| 209 // but there has (as yet) been no need for anything longer either. |
| 210 const int kMaxSectionNameLen = 64; |
| 211 |
| 212 // name_len should include terminating '\0'. |
| 213 bool GetSectionHeaderByName(int fd, const char *name, size_t name_len, |
| 214 ElfW(Shdr) *out) { |
| 215 ElfW(Ehdr) elf_header; |
| 216 if (!ReadFromOffsetExact(fd, &elf_header, sizeof(elf_header), 0)) { |
| 217 return false; |
| 218 } |
| 219 |
| 220 ElfW(Shdr) shstrtab; |
| 221 off_t shstrtab_offset = (elf_header.e_shoff + |
| 222 elf_header.e_shentsize * elf_header.e_shstrndx); |
| 223 if (!ReadFromOffsetExact(fd, &shstrtab, sizeof(shstrtab), shstrtab_offset)) { |
| 224 return false; |
| 225 } |
| 226 |
| 227 for (int i = 0; i < elf_header.e_shnum; ++i) { |
| 228 off_t section_header_offset = (elf_header.e_shoff + |
| 229 elf_header.e_shentsize * i); |
| 230 if (!ReadFromOffsetExact(fd, out, sizeof(*out), section_header_offset)) { |
| 231 return false; |
| 232 } |
| 233 char header_name[kMaxSectionNameLen]; |
| 234 if (sizeof(header_name) < name_len) { |
| 235 RAW_LOG(WARNING, "Section name '%s' is too long (%"PRIuS"); " |
| 236 "section will not be found (even if present).", name, name_len); |
| 237 // No point in even trying. |
| 238 return false; |
| 239 } |
| 240 off_t name_offset = shstrtab.sh_offset + out->sh_name; |
| 241 ssize_t n_read = ReadFromOffset(fd, &header_name, name_len, name_offset); |
| 242 if (n_read == -1) { |
| 243 return false; |
| 244 } else if (n_read != name_len) { |
| 245 // Short read -- name could be at end of file. |
| 246 continue; |
| 247 } |
| 248 if (memcmp(header_name, name, name_len) == 0) { |
| 249 return true; |
| 250 } |
| 251 } |
| 252 return false; |
| 253 } |
| 254 |
| 255 // Read a symbol table and look for the symbol containing the |
| 256 // pc. Iterate over symbols in a symbol table and look for the symbol |
| 257 // containing "pc". On success, return true and write the symbol name |
| 258 // to out. Otherwise, return false. |
| 259 // To keep stack consumption low, we would like this function to not get |
| 260 // inlined. |
| 261 static ATTRIBUTE_NOINLINE bool |
| 262 FindSymbol(uint64_t pc, const int fd, char *out, int out_size, |
| 263 uint64_t symbol_offset, const ElfW(Shdr) *strtab, |
| 264 const ElfW(Shdr) *symtab) { |
| 265 if (symtab == NULL) { |
| 266 return false; |
| 267 } |
| 268 const int num_symbols = symtab->sh_size / symtab->sh_entsize; |
| 269 for (int i = 0; i < num_symbols;) { |
| 270 off_t offset = symtab->sh_offset + i * symtab->sh_entsize; |
| 271 |
| 272 // If we are reading Elf64_Sym's, we want to limit this array to |
| 273 // 32 elements (to keep stack consumption low), otherwise we can |
| 274 // have a 64 element Elf32_Sym array. |
| 275 #if __WORDSIZE == 64 |
| 276 #define NUM_SYMBOLS 32 |
| 277 #else |
| 278 #define NUM_SYMBOLS 64 |
| 279 #endif |
| 280 |
| 281 // Read at most NUM_SYMBOLS symbols at once to save read() calls. |
| 282 ElfW(Sym) buf[NUM_SYMBOLS]; |
| 283 const ssize_t len = ReadFromOffset(fd, &buf, sizeof(buf), offset); |
| 284 SAFE_ASSERT(len % sizeof(buf[0]) == 0); |
| 285 const ssize_t num_symbols_in_buf = len / sizeof(buf[0]); |
| 286 SAFE_ASSERT(num_symbols_in_buf <= sizeof(buf)/sizeof(buf[0])); |
| 287 for (int j = 0; j < num_symbols_in_buf; ++j) { |
| 288 const ElfW(Sym)& symbol = buf[j]; |
| 289 uint64_t start_address = symbol.st_value; |
| 290 start_address += symbol_offset; |
| 291 uint64_t end_address = start_address + symbol.st_size; |
| 292 if (symbol.st_value != 0 && // Skip null value symbols. |
| 293 symbol.st_shndx != 0 && // Skip undefined symbols. |
| 294 start_address <= pc && pc < end_address) { |
| 295 ssize_t len1 = ReadFromOffset(fd, out, out_size, |
| 296 strtab->sh_offset + symbol.st_name); |
| 297 if (len1 <= 0 || memchr(out, '\0', out_size) == NULL) { |
| 298 return false; |
| 299 } |
| 300 return true; // Obtained the symbol name. |
| 301 } |
| 302 } |
| 303 i += num_symbols_in_buf; |
| 304 } |
| 305 return false; |
| 306 } |
| 307 |
| 308 // Get the symbol name of "pc" from the file pointed by "fd". Process |
| 309 // both regular and dynamic symbol tables if necessary. On success, |
| 310 // write the symbol name to "out" and return true. Otherwise, return |
| 311 // false. |
| 312 static bool GetSymbolFromObjectFile(const int fd, uint64_t pc, |
| 313 char *out, int out_size, |
| 314 uint64_t map_start_address) { |
| 315 // Read the ELF header. |
| 316 ElfW(Ehdr) elf_header; |
| 317 if (!ReadFromOffsetExact(fd, &elf_header, sizeof(elf_header), 0)) { |
| 318 return false; |
| 319 } |
| 320 |
| 321 uint64_t symbol_offset = 0; |
| 322 if (elf_header.e_type == ET_DYN) { // DSO needs offset adjustment. |
| 323 symbol_offset = map_start_address; |
| 324 } |
| 325 |
| 326 ElfW(Shdr) symtab, strtab; |
| 327 |
| 328 // Consult a regular symbol table first. |
| 329 if (!GetSectionHeaderByType(fd, elf_header.e_shnum, elf_header.e_shoff, |
| 330 SHT_SYMTAB, &symtab)) { |
| 331 return false; |
| 332 } |
| 333 if (!ReadFromOffsetExact(fd, &strtab, sizeof(strtab), elf_header.e_shoff + |
| 334 symtab.sh_link * sizeof(symtab))) { |
| 335 return false; |
| 336 } |
| 337 if (FindSymbol(pc, fd, out, out_size, symbol_offset, |
| 338 &strtab, &symtab)) { |
| 339 return true; // Found the symbol in a regular symbol table. |
| 340 } |
| 341 |
| 342 // If the symbol is not found, then consult a dynamic symbol table. |
| 343 if (!GetSectionHeaderByType(fd, elf_header.e_shnum, elf_header.e_shoff, |
| 344 SHT_DYNSYM, &symtab)) { |
| 345 return false; |
| 346 } |
| 347 if (!ReadFromOffsetExact(fd, &strtab, sizeof(strtab), elf_header.e_shoff + |
| 348 symtab.sh_link * sizeof(symtab))) { |
| 349 return false; |
| 350 } |
| 351 if (FindSymbol(pc, fd, out, out_size, symbol_offset, |
| 352 &strtab, &symtab)) { |
| 353 return true; // Found the symbol in a dynamic symbol table. |
| 354 } |
| 355 |
| 356 return false; |
| 357 } |
| 358 |
| 359 namespace { |
| 360 // Thin wrapper around a file descriptor so that the file descriptor |
| 361 // gets closed for sure. |
| 362 struct FileDescriptor { |
| 363 const int fd_; |
| 364 explicit FileDescriptor(int fd) : fd_(fd) {} |
| 365 ~FileDescriptor() { |
| 366 if (fd_ >= 0) { |
| 367 NO_INTR(close(fd_)); |
| 368 } |
| 369 } |
| 370 int get() { return fd_; } |
| 371 |
| 372 private: |
| 373 explicit FileDescriptor(const FileDescriptor&); |
| 374 void operator=(const FileDescriptor&); |
| 375 }; |
| 376 |
| 377 // Helper class for reading lines from file. |
| 378 // |
| 379 // Note: we don't use ProcMapsIterator since the object is big (it has |
| 380 // a 5k array member) and uses async-unsafe functions such as sscanf() |
| 381 // and snprintf(). |
| 382 class LineReader { |
| 383 public: |
| 384 explicit LineReader(int fd, char *buf, int buf_len) : fd_(fd), |
| 385 buf_(buf), buf_len_(buf_len), bol_(buf), eol_(buf), eod_(buf) { |
| 386 } |
| 387 |
| 388 // Read '\n'-terminated line from file. On success, modify "bol" |
| 389 // and "eol", then return true. Otherwise, return false. |
| 390 // |
| 391 // Note: if the last line doesn't end with '\n', the line will be |
| 392 // dropped. It's an intentional behavior to make the code simple. |
| 393 bool ReadLine(const char **bol, const char **eol) { |
| 394 if (BufferIsEmpty()) { // First time. |
| 395 const ssize_t num_bytes = ReadPersistent(fd_, buf_, buf_len_); |
| 396 if (num_bytes <= 0) { // EOF or error. |
| 397 return false; |
| 398 } |
| 399 eod_ = buf_ + num_bytes; |
| 400 bol_ = buf_; |
| 401 } else { |
| 402 bol_ = eol_ + 1; // Advance to the next line in the buffer. |
| 403 SAFE_ASSERT(bol_ <= eod_); // "bol_" can point to "eod_". |
| 404 if (!HasCompleteLine()) { |
| 405 const int incomplete_line_length = eod_ - bol_; |
| 406 // Move the trailing incomplete line to the beginning. |
| 407 memmove(buf_, bol_, incomplete_line_length); |
| 408 // Read text from file and append it. |
| 409 char * const append_pos = buf_ + incomplete_line_length; |
| 410 const int capacity_left = buf_len_ - incomplete_line_length; |
| 411 const ssize_t num_bytes = ReadPersistent(fd_, append_pos, |
| 412 capacity_left); |
| 413 if (num_bytes <= 0) { // EOF or error. |
| 414 return false; |
| 415 } |
| 416 eod_ = append_pos + num_bytes; |
| 417 bol_ = buf_; |
| 418 } |
| 419 } |
| 420 eol_ = FindLineFeed(); |
| 421 if (eol_ == NULL) { // '\n' not found. Malformed line. |
| 422 return false; |
| 423 } |
| 424 *eol_ = '\0'; // Replace '\n' with '\0'. |
| 425 |
| 426 *bol = bol_; |
| 427 *eol = eol_; |
| 428 return true; |
| 429 } |
| 430 |
| 431 // Beginning of line. |
| 432 const char *bol() { |
| 433 return bol_; |
| 434 } |
| 435 |
| 436 // End of line. |
| 437 const char *eol() { |
| 438 return eol_; |
| 439 } |
| 440 |
| 441 private: |
| 442 explicit LineReader(const LineReader&); |
| 443 void operator=(const LineReader&); |
| 444 |
| 445 char *FindLineFeed() { |
| 446 return reinterpret_cast<char *>(memchr(bol_, '\n', eod_ - bol_)); |
| 447 } |
| 448 |
| 449 bool BufferIsEmpty() { |
| 450 return buf_ == eod_; |
| 451 } |
| 452 |
| 453 bool HasCompleteLine() { |
| 454 return !BufferIsEmpty() && FindLineFeed() != NULL; |
| 455 } |
| 456 |
| 457 const int fd_; |
| 458 char * const buf_; |
| 459 const int buf_len_; |
| 460 char *bol_; |
| 461 char *eol_; |
| 462 const char *eod_; // End of data in "buf_". |
| 463 }; |
| 464 } // namespace |
| 465 |
| 466 // Place the hex number read from "start" into "*hex". The pointer to |
| 467 // the first non-hex character or "end" is returned. |
| 468 static char *GetHex(const char *start, const char *end, uint64_t *hex) { |
| 469 *hex = 0; |
| 470 const char *p; |
| 471 for (p = start; p < end; ++p) { |
| 472 int ch = *p; |
| 473 if ((ch >= '0' && ch <= '9') || |
| 474 (ch >= 'A' && ch <= 'F') || (ch >= 'a' && ch <= 'f')) { |
| 475 *hex = (*hex << 4) | (ch < 'A' ? ch - '0' : (ch & 0xF) + 9); |
| 476 } else { // Encountered the first non-hex character. |
| 477 break; |
| 478 } |
| 479 } |
| 480 SAFE_ASSERT(p <= end); |
| 481 return const_cast<char *>(p); |
| 482 } |
| 483 |
| 484 // Search for the object file (from /proc/self/maps) that contains |
| 485 // the specified pc. If found, open this file and return the file handle, |
| 486 // and also set start_address to the start address of where this object |
| 487 // file is mapped to in memory. Otherwise, return -1. |
| 488 static ATTRIBUTE_NOINLINE int |
| 489 OpenObjectFileContainingPcAndGetStartAddress(uint64_t pc, |
| 490 uint64_t &start_address) { |
| 491 int object_fd; |
| 492 |
| 493 // Open /proc/self/maps. |
| 494 int maps_fd; |
| 495 NO_INTR(maps_fd = open("/proc/self/maps", O_RDONLY)); |
| 496 FileDescriptor wrapped_maps_fd(maps_fd); |
| 497 if (wrapped_maps_fd.get() < 0) { |
| 498 return -1; |
| 499 } |
| 500 |
| 501 // Iterate over maps and look for the map containing the pc. Then |
| 502 // look into the symbol tables inside. |
| 503 char buf[1024]; // Big enough for line of sane /proc/self/maps |
| 504 LineReader reader(wrapped_maps_fd.get(), buf, sizeof(buf)); |
| 505 while (true) { |
| 506 const char *cursor; |
| 507 const char *eol; |
| 508 if (!reader.ReadLine(&cursor, &eol)) { // EOF or malformed line. |
| 509 return -1; |
| 510 } |
| 511 |
| 512 // Start parsing line in /proc/self/maps. Here is an example: |
| 513 // |
| 514 // 08048000-0804c000 r-xp 00000000 08:01 2142121 /bin/cat |
| 515 // |
| 516 // We want start address (08048000), end address (0804c000), flags |
| 517 // (r-xp) and file name (/bin/cat). |
| 518 |
| 519 // Read start address. |
| 520 cursor = GetHex(cursor, eol, &start_address); |
| 521 if (cursor == eol || *cursor != '-') { |
| 522 return -1; // Malformed line. |
| 523 } |
| 524 ++cursor; // Skip '-'. |
| 525 |
| 526 // Read end address. |
| 527 uint64_t end_address; |
| 528 cursor = GetHex(cursor, eol, &end_address); |
| 529 if (cursor == eol || *cursor != ' ') { |
| 530 return -1; // Malformed line. |
| 531 } |
| 532 ++cursor; // Skip ' '. |
| 533 |
| 534 // Check start and end addresses. |
| 535 if (!(start_address <= pc && pc < end_address)) { |
| 536 continue; // We skip this map. PC isn't in this map. |
| 537 } |
| 538 |
| 539 // Read flags. Skip flags until we encounter a space or eol. |
| 540 const char * const flags_start = cursor; |
| 541 while (cursor < eol && *cursor != ' ') { |
| 542 ++cursor; |
| 543 } |
| 544 // We expect at least four letters for flags (ex. "r-xp"). |
| 545 if (cursor == eol || cursor < flags_start + 4) { |
| 546 return -1; // Malformed line. |
| 547 } |
| 548 |
| 549 // Check flags. We are only interested in "r-x" maps. |
| 550 if (memcmp(flags_start, "r-x", 3) != 0) { // Not a "r-x" map. |
| 551 continue; // We skip this map. |
| 552 } |
| 553 ++cursor; // Skip ' '. |
| 554 |
| 555 // Skip to file name. "cursor" now points to file offset. We need to |
| 556 // skip at least three spaces for file offset, dev, and inode. |
| 557 int num_spaces = 0; |
| 558 while (cursor < eol) { |
| 559 if (*cursor == ' ') { |
| 560 ++num_spaces; |
| 561 } else if (num_spaces >= 3) { |
| 562 // The first non-space character after skipping three spaces |
| 563 // is the beginning of the file name. |
| 564 break; |
| 565 } |
| 566 ++cursor; |
| 567 } |
| 568 if (cursor == eol) { |
| 569 return -1; // Malformed line. |
| 570 } |
| 571 |
| 572 // Finally, "cursor" now points to file name of our interest. |
| 573 NO_INTR(object_fd = open(cursor, O_RDONLY)); |
| 574 if (object_fd < 0) { |
| 575 return -1; |
| 576 } |
| 577 return object_fd; |
| 578 } |
| 579 } |
| 580 |
| 581 // The implementation of our symbolization routine. If it |
| 582 // successfully finds the symbol containing "pc" and obtains the |
| 583 // symbol name, returns true and write the symbol name to "out". |
| 584 // Otherwise, returns false. If Callback function is installed via |
| 585 // InstallSymbolizeCallback(), the function is also called in this function, |
| 586 // and "out" is used as its output. |
| 587 // To keep stack consumption low, we would like this function to not |
| 588 // get inlined. |
| 589 static ATTRIBUTE_NOINLINE bool SymbolizeAndDemangle(void *pc, char *out, |
| 590 int out_size) { |
| 591 uint64_t pc0 = reinterpret_cast<uintptr_t>(pc); |
| 592 uint64_t start_address = 0; |
| 593 |
| 594 int object_fd = OpenObjectFileContainingPcAndGetStartAddress(pc0, |
| 595 start_address); |
| 596 if (object_fd == -1) { |
| 597 return false; |
| 598 } |
| 599 FileDescriptor wrapped_object_fd(object_fd); |
| 600 int elf_type = FileGetElfType(wrapped_object_fd.get()); |
| 601 if (elf_type == -1) { |
| 602 return false; |
| 603 } |
| 604 if (g_symbolize_callback) { |
| 605 // Run the call back if it's installed. |
| 606 // Note: relocation (and much of the rest of this code) will be |
| 607 // wrong for prelinked shared libraries and PIE executables. |
| 608 uint64 relocation = (elf_type == ET_DYN) ? start_address : 0; |
| 609 int num_bytes_written = g_symbolize_callback(wrapped_object_fd.get(), |
| 610 pc, out, out_size, |
| 611 relocation); |
| 612 if (num_bytes_written > 0) { |
| 613 out += num_bytes_written; |
| 614 out_size -= num_bytes_written; |
| 615 } |
| 616 } |
| 617 if (!GetSymbolFromObjectFile(wrapped_object_fd.get(), pc0, |
| 618 out, out_size, start_address)) { |
| 619 return false; |
| 620 } |
| 621 |
| 622 // Symbolization succeeded. Now we try to demangle the symbol. |
| 623 DemangleInplace(out, out_size); |
| 624 return true; |
| 625 } |
| 626 |
| 627 _END_GOOGLE_NAMESPACE_ |
| 628 |
| 629 #elif defined(OS_MACOSX) && defined(HAVE_DLADDR) |
| 630 |
| 631 #include <dlfcn.h> |
| 632 #include <string.h> |
| 633 |
| 634 _START_GOOGLE_NAMESPACE_ |
| 635 |
| 636 static ATTRIBUTE_NOINLINE bool SymbolizeAndDemangle(void *pc, char *out, |
| 637 int out_size) { |
| 638 Dl_info info; |
| 639 if (dladdr(pc, &info)) { |
| 640 if (strlen(info.dli_sname) < out_size) { |
| 641 strcpy(out, info.dli_sname); |
| 642 // Symbolization succeeded. Now we try to demangle the symbol. |
| 643 DemangleInplace(out, out_size); |
| 644 return true; |
| 645 } |
| 646 } |
| 647 return false; |
| 648 } |
| 649 |
| 650 _END_GOOGLE_NAMESPACE_ |
| 651 |
| 652 #else |
| 653 # error BUG: HAVE_SYMBOLIZE was wrongly set |
| 654 #endif |
| 655 |
| 656 _START_GOOGLE_NAMESPACE_ |
| 657 |
| 658 bool Symbolize(void *pc, char *out, int out_size) { |
| 659 SAFE_ASSERT(out_size >= 0); |
| 660 return SymbolizeAndDemangle(pc, out, out_size); |
| 661 } |
| 662 |
| 663 _END_GOOGLE_NAMESPACE_ |
| 664 |
| 665 #else /* HAVE_SYMBOLIZE */ |
| 666 |
| 667 #include <assert.h> |
| 668 |
| 669 #include "config.h" |
| 670 |
| 671 _START_GOOGLE_NAMESPACE_ |
| 672 |
| 673 // TODO: Support other environments. |
| 674 bool Symbolize(void *pc, char *out, int out_size) { |
| 675 assert(0); |
| 676 return false; |
| 677 } |
| 678 |
| 679 _END_GOOGLE_NAMESPACE_ |
| 680 |
| 681 #endif |
OLD | NEW |