Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 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 | // Additional header can be specified by the GLOG_BUILD_CONFIG_INCLUDE |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 49 | // macro to add platform specific defines (e.g. GLOG_OS_OPENBSD). |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 50 | |
| 51 | #ifdef GLOG_BUILD_CONFIG_INCLUDE |
| 52 | #include GLOG_BUILD_CONFIG_INCLUDE |
| 53 | #endif // GLOG_BUILD_CONFIG_INCLUDE |
| 54 | |
| 55 | #include "utilities.h" |
| 56 | |
| 57 | #if defined(HAVE_SYMBOLIZE) |
| 58 | |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 59 | #include <cstring> |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 60 | |
| 61 | #include <algorithm> |
| 62 | #include <limits> |
| 63 | |
| 64 | #include "symbolize.h" |
| 65 | #include "demangle.h" |
| 66 | |
| 67 | _START_GOOGLE_NAMESPACE_ |
| 68 | |
| 69 | // We don't use assert() since it's not guaranteed to be |
| 70 | // async-signal-safe. Instead we define a minimal assertion |
| 71 | // macro. So far, we don't need pretty printing for __FILE__, etc. |
| 72 | |
| 73 | // A wrapper for abort() to make it callable in ? :. |
| 74 | static int AssertFail() { |
| 75 | abort(); |
| 76 | return 0; // Should not reach. |
| 77 | } |
| 78 | |
| 79 | #define SAFE_ASSERT(expr) ((expr) ? 0 : AssertFail()) |
| 80 | |
| 81 | static SymbolizeCallback g_symbolize_callback = NULL; |
| 82 | void InstallSymbolizeCallback(SymbolizeCallback callback) { |
| 83 | g_symbolize_callback = callback; |
| 84 | } |
| 85 | |
| 86 | static SymbolizeOpenObjectFileCallback g_symbolize_open_object_file_callback = |
| 87 | NULL; |
| 88 | void InstallSymbolizeOpenObjectFileCallback( |
| 89 | SymbolizeOpenObjectFileCallback callback) { |
| 90 | g_symbolize_open_object_file_callback = callback; |
| 91 | } |
| 92 | |
| 93 | // This function wraps the Demangle function to provide an interface |
| 94 | // where the input symbol is demangled in-place. |
| 95 | // To keep stack consumption low, we would like this function to not |
| 96 | // get inlined. |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 97 | static ATTRIBUTE_NOINLINE void DemangleInplace(char *out, size_t out_size) { |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 98 | char demangled[256]; // Big enough for sane demangled symbols. |
| 99 | if (Demangle(out, demangled, sizeof(demangled))) { |
| 100 | // Demangling succeeded. Copy to out if the space allows. |
| 101 | size_t len = strlen(demangled); |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 102 | if (len + 1 <= out_size) { // +1 for '\0'. |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 103 | SAFE_ASSERT(len < sizeof(demangled)); |
| 104 | memmove(out, demangled, len + 1); |
| 105 | } |
| 106 | } |
| 107 | } |
| 108 | |
| 109 | _END_GOOGLE_NAMESPACE_ |
| 110 | |
| 111 | #if defined(__ELF__) |
| 112 | |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 113 | #if defined(HAVE_DLFCN_H) |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 114 | #include <dlfcn.h> |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 115 | #endif |
| 116 | #if defined(GLOG_OS_OPENBSD) |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 117 | #include <sys/exec_elf.h> |
| 118 | #else |
| 119 | #include <elf.h> |
| 120 | #endif |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 121 | #include <cerrno> |
| 122 | #include <climits> |
| 123 | #include <cstddef> |
| 124 | #include <cstdio> |
| 125 | #include <cstdlib> |
| 126 | #include <cstring> |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 127 | #include <fcntl.h> |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 128 | #include <stdint.h> |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 129 | #include <sys/stat.h> |
| 130 | #include <sys/types.h> |
| 131 | #include <unistd.h> |
| 132 | |
| 133 | #include "symbolize.h" |
| 134 | #include "config.h" |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 135 | #include <glog/raw_logging.h> |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 136 | |
| 137 | // Re-runs fn until it doesn't cause EINTR. |
| 138 | #define NO_INTR(fn) do {} while ((fn) < 0 && errno == EINTR) |
| 139 | |
| 140 | _START_GOOGLE_NAMESPACE_ |
| 141 | |
| 142 | // Read up to "count" bytes from "offset" in the file pointed by file |
| 143 | // descriptor "fd" into the buffer starting at "buf" while handling short reads |
| 144 | // and EINTR. On success, return the number of bytes read. Otherwise, return |
| 145 | // -1. |
| 146 | static ssize_t ReadFromOffset(const int fd, void *buf, const size_t count, |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 147 | const size_t offset) { |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 148 | SAFE_ASSERT(fd >= 0); |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 149 | SAFE_ASSERT(count <= static_cast<size_t>(std::numeric_limits<ssize_t>::max())); |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 150 | char *buf0 = reinterpret_cast<char *>(buf); |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 151 | size_t num_bytes = 0; |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 152 | while (num_bytes < count) { |
| 153 | ssize_t len; |
| 154 | NO_INTR(len = pread(fd, buf0 + num_bytes, count - num_bytes, |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 155 | static_cast<off_t>(offset + num_bytes))); |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 156 | if (len < 0) { // There was an error other than EINTR. |
| 157 | return -1; |
| 158 | } |
| 159 | if (len == 0) { // Reached EOF. |
| 160 | break; |
| 161 | } |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 162 | num_bytes += static_cast<size_t>(len); |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 163 | } |
| 164 | SAFE_ASSERT(num_bytes <= count); |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 165 | return static_cast<ssize_t>(num_bytes); |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 166 | } |
| 167 | |
| 168 | // Try reading exactly "count" bytes from "offset" bytes in a file |
| 169 | // pointed by "fd" into the buffer starting at "buf" while handling |
| 170 | // short reads and EINTR. On success, return true. Otherwise, return |
| 171 | // false. |
| 172 | static bool ReadFromOffsetExact(const int fd, void *buf, |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 173 | const size_t count, const size_t offset) { |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 174 | ssize_t len = ReadFromOffset(fd, buf, count, offset); |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 175 | return static_cast<size_t>(len) == count; |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 176 | } |
| 177 | |
| 178 | // Returns elf_header.e_type if the file pointed by fd is an ELF binary. |
| 179 | static int FileGetElfType(const int fd) { |
| 180 | ElfW(Ehdr) elf_header; |
| 181 | if (!ReadFromOffsetExact(fd, &elf_header, sizeof(elf_header), 0)) { |
| 182 | return -1; |
| 183 | } |
| 184 | if (memcmp(elf_header.e_ident, ELFMAG, SELFMAG) != 0) { |
| 185 | return -1; |
| 186 | } |
| 187 | return elf_header.e_type; |
| 188 | } |
| 189 | |
| 190 | // Read the section headers in the given ELF binary, and if a section |
| 191 | // of the specified type is found, set the output to this section header |
| 192 | // and return true. Otherwise, return false. |
| 193 | // To keep stack consumption low, we would like this function to not get |
| 194 | // inlined. |
| 195 | static ATTRIBUTE_NOINLINE bool |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 196 | GetSectionHeaderByType(const int fd, ElfW(Half) sh_num, const size_t sh_offset, |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 197 | ElfW(Word) type, ElfW(Shdr) *out) { |
| 198 | // Read at most 16 section headers at a time to save read calls. |
| 199 | ElfW(Shdr) buf[16]; |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 200 | for (size_t i = 0; i < sh_num;) { |
| 201 | const size_t num_bytes_left = (sh_num - i) * sizeof(buf[0]); |
| 202 | const size_t num_bytes_to_read = |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 203 | (sizeof(buf) > num_bytes_left) ? num_bytes_left : sizeof(buf); |
| 204 | const ssize_t len = ReadFromOffset(fd, buf, num_bytes_to_read, |
| 205 | sh_offset + i * sizeof(buf[0])); |
| 206 | if (len == -1) { |
| 207 | return false; |
| 208 | } |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 209 | SAFE_ASSERT(static_cast<size_t>(len) % sizeof(buf[0]) == 0); |
| 210 | const size_t num_headers_in_buf = static_cast<size_t>(len) / sizeof(buf[0]); |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 211 | SAFE_ASSERT(num_headers_in_buf <= sizeof(buf) / sizeof(buf[0])); |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 212 | for (size_t j = 0; j < num_headers_in_buf; ++j) { |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 213 | if (buf[j].sh_type == type) { |
| 214 | *out = buf[j]; |
| 215 | return true; |
| 216 | } |
| 217 | } |
| 218 | i += num_headers_in_buf; |
| 219 | } |
| 220 | return false; |
| 221 | } |
| 222 | |
| 223 | // There is no particular reason to limit section name to 63 characters, |
| 224 | // but there has (as yet) been no need for anything longer either. |
| 225 | const int kMaxSectionNameLen = 64; |
| 226 | |
| 227 | // name_len should include terminating '\0'. |
| 228 | bool GetSectionHeaderByName(int fd, const char *name, size_t name_len, |
| 229 | ElfW(Shdr) *out) { |
| 230 | ElfW(Ehdr) elf_header; |
| 231 | if (!ReadFromOffsetExact(fd, &elf_header, sizeof(elf_header), 0)) { |
| 232 | return false; |
| 233 | } |
| 234 | |
| 235 | ElfW(Shdr) shstrtab; |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 236 | size_t shstrtab_offset = |
| 237 | (elf_header.e_shoff + static_cast<size_t>(elf_header.e_shentsize) * |
| 238 | static_cast<size_t>(elf_header.e_shstrndx)); |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 239 | if (!ReadFromOffsetExact(fd, &shstrtab, sizeof(shstrtab), shstrtab_offset)) { |
| 240 | return false; |
| 241 | } |
| 242 | |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 243 | for (size_t i = 0; i < elf_header.e_shnum; ++i) { |
| 244 | size_t section_header_offset = (elf_header.e_shoff + |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 245 | elf_header.e_shentsize * i); |
| 246 | if (!ReadFromOffsetExact(fd, out, sizeof(*out), section_header_offset)) { |
| 247 | return false; |
| 248 | } |
| 249 | char header_name[kMaxSectionNameLen]; |
| 250 | if (sizeof(header_name) < name_len) { |
| 251 | RAW_LOG(WARNING, "Section name '%s' is too long (%" PRIuS "); " |
| 252 | "section will not be found (even if present).", name, name_len); |
| 253 | // No point in even trying. |
| 254 | return false; |
| 255 | } |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 256 | size_t name_offset = shstrtab.sh_offset + out->sh_name; |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 257 | ssize_t n_read = ReadFromOffset(fd, &header_name, name_len, name_offset); |
| 258 | if (n_read == -1) { |
| 259 | return false; |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 260 | } else if (static_cast<size_t>(n_read) != name_len) { |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 261 | // Short read -- name could be at end of file. |
| 262 | continue; |
| 263 | } |
| 264 | if (memcmp(header_name, name, name_len) == 0) { |
| 265 | return true; |
| 266 | } |
| 267 | } |
| 268 | return false; |
| 269 | } |
| 270 | |
| 271 | // Read a symbol table and look for the symbol containing the |
| 272 | // pc. Iterate over symbols in a symbol table and look for the symbol |
| 273 | // containing "pc". On success, return true and write the symbol name |
| 274 | // to out. Otherwise, return false. |
| 275 | // To keep stack consumption low, we would like this function to not get |
| 276 | // inlined. |
| 277 | static ATTRIBUTE_NOINLINE bool |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 278 | FindSymbol(uint64_t pc, const int fd, char *out, size_t out_size, |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 279 | uint64_t symbol_offset, const ElfW(Shdr) *strtab, |
| 280 | const ElfW(Shdr) *symtab) { |
| 281 | if (symtab == NULL) { |
| 282 | return false; |
| 283 | } |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 284 | const size_t num_symbols = symtab->sh_size / symtab->sh_entsize; |
| 285 | for (unsigned i = 0; i < num_symbols;) { |
| 286 | size_t offset = symtab->sh_offset + i * symtab->sh_entsize; |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 287 | |
| 288 | // If we are reading Elf64_Sym's, we want to limit this array to |
| 289 | // 32 elements (to keep stack consumption low), otherwise we can |
| 290 | // have a 64 element Elf32_Sym array. |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 291 | #if defined(__WORDSIZE) && __WORDSIZE == 64 |
| 292 | const size_t NUM_SYMBOLS = 32U; |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 293 | #else |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 294 | const size_t NUM_SYMBOLS = 64U; |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 295 | #endif |
| 296 | |
| 297 | // Read at most NUM_SYMBOLS symbols at once to save read() calls. |
| 298 | ElfW(Sym) buf[NUM_SYMBOLS]; |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 299 | size_t num_symbols_to_read = std::min(NUM_SYMBOLS, num_symbols - i); |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 300 | const ssize_t len = |
| 301 | ReadFromOffset(fd, &buf, sizeof(buf[0]) * num_symbols_to_read, offset); |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 302 | SAFE_ASSERT(static_cast<size_t>(len) % sizeof(buf[0]) == 0); |
| 303 | const size_t num_symbols_in_buf = static_cast<size_t>(len) / sizeof(buf[0]); |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 304 | SAFE_ASSERT(num_symbols_in_buf <= num_symbols_to_read); |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 305 | for (unsigned j = 0; j < num_symbols_in_buf; ++j) { |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 306 | const ElfW(Sym)& symbol = buf[j]; |
| 307 | uint64_t start_address = symbol.st_value; |
| 308 | start_address += symbol_offset; |
| 309 | uint64_t end_address = start_address + symbol.st_size; |
| 310 | if (symbol.st_value != 0 && // Skip null value symbols. |
| 311 | symbol.st_shndx != 0 && // Skip undefined symbols. |
| 312 | start_address <= pc && pc < end_address) { |
| 313 | ssize_t len1 = ReadFromOffset(fd, out, out_size, |
| 314 | strtab->sh_offset + symbol.st_name); |
| 315 | if (len1 <= 0 || memchr(out, '\0', out_size) == NULL) { |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 316 | memset(out, 0, out_size); |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 317 | return false; |
| 318 | } |
| 319 | return true; // Obtained the symbol name. |
| 320 | } |
| 321 | } |
| 322 | i += num_symbols_in_buf; |
| 323 | } |
| 324 | return false; |
| 325 | } |
| 326 | |
| 327 | // Get the symbol name of "pc" from the file pointed by "fd". Process |
| 328 | // both regular and dynamic symbol tables if necessary. On success, |
| 329 | // write the symbol name to "out" and return true. Otherwise, return |
| 330 | // false. |
| 331 | static bool GetSymbolFromObjectFile(const int fd, |
| 332 | uint64_t pc, |
| 333 | char* out, |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 334 | size_t out_size, |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 335 | uint64_t base_address) { |
| 336 | // Read the ELF header. |
| 337 | ElfW(Ehdr) elf_header; |
| 338 | if (!ReadFromOffsetExact(fd, &elf_header, sizeof(elf_header), 0)) { |
| 339 | return false; |
| 340 | } |
| 341 | |
| 342 | ElfW(Shdr) symtab, strtab; |
| 343 | |
| 344 | // Consult a regular symbol table first. |
| 345 | if (GetSectionHeaderByType(fd, elf_header.e_shnum, elf_header.e_shoff, |
| 346 | SHT_SYMTAB, &symtab)) { |
| 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, base_address, &strtab, &symtab)) { |
| 352 | return true; // Found the symbol in a regular symbol table. |
| 353 | } |
| 354 | } |
| 355 | |
| 356 | // If the symbol is not found, then consult a dynamic symbol table. |
| 357 | if (GetSectionHeaderByType(fd, elf_header.e_shnum, elf_header.e_shoff, |
| 358 | SHT_DYNSYM, &symtab)) { |
| 359 | if (!ReadFromOffsetExact(fd, &strtab, sizeof(strtab), elf_header.e_shoff + |
| 360 | symtab.sh_link * sizeof(symtab))) { |
| 361 | return false; |
| 362 | } |
| 363 | if (FindSymbol(pc, fd, out, out_size, base_address, &strtab, &symtab)) { |
| 364 | return true; // Found the symbol in a dynamic symbol table. |
| 365 | } |
| 366 | } |
| 367 | |
| 368 | return false; |
| 369 | } |
| 370 | |
| 371 | namespace { |
| 372 | // Thin wrapper around a file descriptor so that the file descriptor |
| 373 | // gets closed for sure. |
| 374 | struct FileDescriptor { |
| 375 | const int fd_; |
| 376 | explicit FileDescriptor(int fd) : fd_(fd) {} |
| 377 | ~FileDescriptor() { |
| 378 | if (fd_ >= 0) { |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 379 | close(fd_); |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 380 | } |
| 381 | } |
| 382 | int get() { return fd_; } |
| 383 | |
| 384 | private: |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 385 | FileDescriptor(const FileDescriptor &); |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 386 | void operator=(const FileDescriptor&); |
| 387 | }; |
| 388 | |
| 389 | // Helper class for reading lines from file. |
| 390 | // |
| 391 | // Note: we don't use ProcMapsIterator since the object is big (it has |
| 392 | // a 5k array member) and uses async-unsafe functions such as sscanf() |
| 393 | // and snprintf(). |
| 394 | class LineReader { |
| 395 | public: |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 396 | explicit LineReader(int fd, char *buf, size_t buf_len, size_t offset) |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 397 | : fd_(fd), |
| 398 | buf_(buf), |
| 399 | buf_len_(buf_len), |
| 400 | offset_(offset), |
| 401 | bol_(buf), |
| 402 | eol_(buf), |
| 403 | eod_(buf) {} |
| 404 | |
| 405 | // Read '\n'-terminated line from file. On success, modify "bol" |
| 406 | // and "eol", then return true. Otherwise, return false. |
| 407 | // |
| 408 | // Note: if the last line doesn't end with '\n', the line will be |
| 409 | // dropped. It's an intentional behavior to make the code simple. |
| 410 | bool ReadLine(const char **bol, const char **eol) { |
| 411 | if (BufferIsEmpty()) { // First time. |
| 412 | const ssize_t num_bytes = ReadFromOffset(fd_, buf_, buf_len_, offset_); |
| 413 | if (num_bytes <= 0) { // EOF or error. |
| 414 | return false; |
| 415 | } |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 416 | offset_ += static_cast<size_t>(num_bytes); |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 417 | eod_ = buf_ + num_bytes; |
| 418 | bol_ = buf_; |
| 419 | } else { |
| 420 | bol_ = eol_ + 1; // Advance to the next line in the buffer. |
| 421 | SAFE_ASSERT(bol_ <= eod_); // "bol_" can point to "eod_". |
| 422 | if (!HasCompleteLine()) { |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 423 | const size_t incomplete_line_length = static_cast<size_t>(eod_ - bol_); |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 424 | // Move the trailing incomplete line to the beginning. |
| 425 | memmove(buf_, bol_, incomplete_line_length); |
| 426 | // Read text from file and append it. |
| 427 | char * const append_pos = buf_ + incomplete_line_length; |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 428 | const size_t capacity_left = buf_len_ - incomplete_line_length; |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 429 | const ssize_t num_bytes = |
| 430 | ReadFromOffset(fd_, append_pos, capacity_left, offset_); |
| 431 | if (num_bytes <= 0) { // EOF or error. |
| 432 | return false; |
| 433 | } |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 434 | offset_ += static_cast<size_t>(num_bytes); |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 435 | eod_ = append_pos + num_bytes; |
| 436 | bol_ = buf_; |
| 437 | } |
| 438 | } |
| 439 | eol_ = FindLineFeed(); |
| 440 | if (eol_ == NULL) { // '\n' not found. Malformed line. |
| 441 | return false; |
| 442 | } |
| 443 | *eol_ = '\0'; // Replace '\n' with '\0'. |
| 444 | |
| 445 | *bol = bol_; |
| 446 | *eol = eol_; |
| 447 | return true; |
| 448 | } |
| 449 | |
| 450 | // Beginning of line. |
| 451 | const char *bol() { |
| 452 | return bol_; |
| 453 | } |
| 454 | |
| 455 | // End of line. |
| 456 | const char *eol() { |
| 457 | return eol_; |
| 458 | } |
| 459 | |
| 460 | private: |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 461 | LineReader(const LineReader &); |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 462 | void operator=(const LineReader&); |
| 463 | |
| 464 | char *FindLineFeed() { |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 465 | return reinterpret_cast<char *>(memchr(bol_, '\n', static_cast<size_t>(eod_ - bol_))); |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 466 | } |
| 467 | |
| 468 | bool BufferIsEmpty() { |
| 469 | return buf_ == eod_; |
| 470 | } |
| 471 | |
| 472 | bool HasCompleteLine() { |
| 473 | return !BufferIsEmpty() && FindLineFeed() != NULL; |
| 474 | } |
| 475 | |
| 476 | const int fd_; |
| 477 | char * const buf_; |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 478 | const size_t buf_len_; |
| 479 | size_t offset_; |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 480 | char *bol_; |
| 481 | char *eol_; |
| 482 | const char *eod_; // End of data in "buf_". |
| 483 | }; |
| 484 | } // namespace |
| 485 | |
| 486 | // Place the hex number read from "start" into "*hex". The pointer to |
| 487 | // the first non-hex character or "end" is returned. |
| 488 | static char *GetHex(const char *start, const char *end, uint64_t *hex) { |
| 489 | *hex = 0; |
| 490 | const char *p; |
| 491 | for (p = start; p < end; ++p) { |
| 492 | int ch = *p; |
| 493 | if ((ch >= '0' && ch <= '9') || |
| 494 | (ch >= 'A' && ch <= 'F') || (ch >= 'a' && ch <= 'f')) { |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 495 | *hex = (*hex << 4U) | (ch < 'A' ? static_cast<uint64_t>(ch - '0') : (ch & 0xF) + 9U); |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 496 | } else { // Encountered the first non-hex character. |
| 497 | break; |
| 498 | } |
| 499 | } |
| 500 | SAFE_ASSERT(p <= end); |
| 501 | return const_cast<char *>(p); |
| 502 | } |
| 503 | |
| 504 | // Searches for the object file (from /proc/self/maps) that contains |
| 505 | // the specified pc. If found, sets |start_address| to the start address |
| 506 | // of where this object file is mapped in memory, sets the module base |
| 507 | // address into |base_address|, copies the object file name into |
| 508 | // |out_file_name|, and attempts to open the object file. If the object |
| 509 | // file is opened successfully, returns the file descriptor. Otherwise, |
| 510 | // returns -1. |out_file_name_size| is the size of the file name buffer |
| 511 | // (including the null-terminator). |
| 512 | static ATTRIBUTE_NOINLINE int |
| 513 | OpenObjectFileContainingPcAndGetStartAddress(uint64_t pc, |
| 514 | uint64_t &start_address, |
| 515 | uint64_t &base_address, |
| 516 | char *out_file_name, |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 517 | size_t out_file_name_size) { |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 518 | int object_fd; |
| 519 | |
| 520 | int maps_fd; |
| 521 | NO_INTR(maps_fd = open("/proc/self/maps", O_RDONLY)); |
| 522 | FileDescriptor wrapped_maps_fd(maps_fd); |
| 523 | if (wrapped_maps_fd.get() < 0) { |
| 524 | return -1; |
| 525 | } |
| 526 | |
| 527 | int mem_fd; |
| 528 | NO_INTR(mem_fd = open("/proc/self/mem", O_RDONLY)); |
| 529 | FileDescriptor wrapped_mem_fd(mem_fd); |
| 530 | if (wrapped_mem_fd.get() < 0) { |
| 531 | return -1; |
| 532 | } |
| 533 | |
| 534 | // Iterate over maps and look for the map containing the pc. Then |
| 535 | // look into the symbol tables inside. |
| 536 | char buf[1024]; // Big enough for line of sane /proc/self/maps |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 537 | unsigned num_maps = 0; |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 538 | LineReader reader(wrapped_maps_fd.get(), buf, sizeof(buf), 0); |
| 539 | while (true) { |
| 540 | num_maps++; |
| 541 | const char *cursor; |
| 542 | const char *eol; |
| 543 | if (!reader.ReadLine(&cursor, &eol)) { // EOF or malformed line. |
| 544 | return -1; |
| 545 | } |
| 546 | |
| 547 | // Start parsing line in /proc/self/maps. Here is an example: |
| 548 | // |
| 549 | // 08048000-0804c000 r-xp 00000000 08:01 2142121 /bin/cat |
| 550 | // |
| 551 | // We want start address (08048000), end address (0804c000), flags |
| 552 | // (r-xp) and file name (/bin/cat). |
| 553 | |
| 554 | // Read start address. |
| 555 | cursor = GetHex(cursor, eol, &start_address); |
| 556 | if (cursor == eol || *cursor != '-') { |
| 557 | return -1; // Malformed line. |
| 558 | } |
| 559 | ++cursor; // Skip '-'. |
| 560 | |
| 561 | // Read end address. |
| 562 | uint64_t end_address; |
| 563 | cursor = GetHex(cursor, eol, &end_address); |
| 564 | if (cursor == eol || *cursor != ' ') { |
| 565 | return -1; // Malformed line. |
| 566 | } |
| 567 | ++cursor; // Skip ' '. |
| 568 | |
| 569 | // Read flags. Skip flags until we encounter a space or eol. |
| 570 | const char * const flags_start = cursor; |
| 571 | while (cursor < eol && *cursor != ' ') { |
| 572 | ++cursor; |
| 573 | } |
| 574 | // We expect at least four letters for flags (ex. "r-xp"). |
| 575 | if (cursor == eol || cursor < flags_start + 4) { |
| 576 | return -1; // Malformed line. |
| 577 | } |
| 578 | |
| 579 | // Determine the base address by reading ELF headers in process memory. |
| 580 | ElfW(Ehdr) ehdr; |
| 581 | // Skip non-readable maps. |
| 582 | if (flags_start[0] == 'r' && |
| 583 | ReadFromOffsetExact(mem_fd, &ehdr, sizeof(ElfW(Ehdr)), start_address) && |
| 584 | memcmp(ehdr.e_ident, ELFMAG, SELFMAG) == 0) { |
| 585 | switch (ehdr.e_type) { |
| 586 | case ET_EXEC: |
| 587 | base_address = 0; |
| 588 | break; |
| 589 | case ET_DYN: |
| 590 | // Find the segment containing file offset 0. This will correspond |
| 591 | // to the ELF header that we just read. Normally this will have |
| 592 | // virtual address 0, but this is not guaranteed. We must subtract |
| 593 | // the virtual address from the address where the ELF header was |
| 594 | // mapped to get the base address. |
| 595 | // |
| 596 | // If we fail to find a segment for file offset 0, use the address |
| 597 | // of the ELF header as the base address. |
| 598 | base_address = start_address; |
| 599 | for (unsigned i = 0; i != ehdr.e_phnum; ++i) { |
| 600 | ElfW(Phdr) phdr; |
| 601 | if (ReadFromOffsetExact( |
| 602 | mem_fd, &phdr, sizeof(phdr), |
| 603 | start_address + ehdr.e_phoff + i * sizeof(phdr)) && |
| 604 | phdr.p_type == PT_LOAD && phdr.p_offset == 0) { |
| 605 | base_address = start_address - phdr.p_vaddr; |
| 606 | break; |
| 607 | } |
| 608 | } |
| 609 | break; |
| 610 | default: |
| 611 | // ET_REL or ET_CORE. These aren't directly executable, so they don't |
| 612 | // affect the base address. |
| 613 | break; |
| 614 | } |
| 615 | } |
| 616 | |
| 617 | // Check start and end addresses. |
| 618 | if (!(start_address <= pc && pc < end_address)) { |
| 619 | continue; // We skip this map. PC isn't in this map. |
| 620 | } |
| 621 | |
| 622 | // Check flags. We are only interested in "r*x" maps. |
| 623 | if (flags_start[0] != 'r' || flags_start[2] != 'x') { |
| 624 | continue; // We skip this map. |
| 625 | } |
| 626 | ++cursor; // Skip ' '. |
| 627 | |
| 628 | // Read file offset. |
| 629 | uint64_t file_offset; |
| 630 | cursor = GetHex(cursor, eol, &file_offset); |
| 631 | if (cursor == eol || *cursor != ' ') { |
| 632 | return -1; // Malformed line. |
| 633 | } |
| 634 | ++cursor; // Skip ' '. |
| 635 | |
| 636 | // Skip to file name. "cursor" now points to dev. We need to |
| 637 | // skip at least two spaces for dev and inode. |
| 638 | int num_spaces = 0; |
| 639 | while (cursor < eol) { |
| 640 | if (*cursor == ' ') { |
| 641 | ++num_spaces; |
| 642 | } else if (num_spaces >= 2) { |
| 643 | // The first non-space character after skipping two spaces |
| 644 | // is the beginning of the file name. |
| 645 | break; |
| 646 | } |
| 647 | ++cursor; |
| 648 | } |
| 649 | if (cursor == eol) { |
| 650 | return -1; // Malformed line. |
| 651 | } |
| 652 | |
| 653 | // Finally, "cursor" now points to file name of our interest. |
| 654 | NO_INTR(object_fd = open(cursor, O_RDONLY)); |
| 655 | if (object_fd < 0) { |
| 656 | // Failed to open object file. Copy the object file name to |
| 657 | // |out_file_name|. |
| 658 | strncpy(out_file_name, cursor, out_file_name_size); |
| 659 | // Making sure |out_file_name| is always null-terminated. |
| 660 | out_file_name[out_file_name_size - 1] = '\0'; |
| 661 | return -1; |
| 662 | } |
| 663 | return object_fd; |
| 664 | } |
| 665 | } |
| 666 | |
| 667 | // POSIX doesn't define any async-signal safe function for converting |
| 668 | // an integer to ASCII. We'll have to define our own version. |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 669 | // itoa_r() converts an (unsigned) integer to ASCII. It returns "buf", if the |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 670 | // conversion was successful or NULL otherwise. It never writes more than "sz" |
| 671 | // bytes. Output will be truncated as needed, and a NUL character is always |
| 672 | // appended. |
| 673 | // NOTE: code from sandbox/linux/seccomp-bpf/demo.cc. |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 674 | static char *itoa_r(uintptr_t i, char *buf, size_t sz, unsigned base, size_t padding) { |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 675 | // Make sure we can write at least one NUL byte. |
| 676 | size_t n = 1; |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 677 | if (n > sz) { |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 678 | return NULL; |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 679 | } |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 680 | |
| 681 | if (base < 2 || base > 16) { |
| 682 | buf[0] = '\000'; |
| 683 | return NULL; |
| 684 | } |
| 685 | |
| 686 | char *start = buf; |
| 687 | |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 688 | // Loop until we have converted the entire number. Output at least one |
| 689 | // character (i.e. '0'). |
| 690 | char *ptr = start; |
| 691 | do { |
| 692 | // Make sure there is still enough space left in our output buffer. |
| 693 | if (++n > sz) { |
| 694 | buf[0] = '\000'; |
| 695 | return NULL; |
| 696 | } |
| 697 | |
| 698 | // Output the next digit. |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 699 | *ptr++ = "0123456789abcdef"[i % base]; |
| 700 | i /= base; |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 701 | |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 702 | if (padding > 0) { |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 703 | padding--; |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 704 | } |
| 705 | } while (i > 0 || padding > 0); |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 706 | |
| 707 | // Terminate the output with a NUL character. |
| 708 | *ptr = '\000'; |
| 709 | |
| 710 | // Conversion to ASCII actually resulted in the digits being in reverse |
| 711 | // order. We can't easily generate them in forward order, as we can't tell |
| 712 | // the number of characters needed until we are done converting. |
| 713 | // So, now, we reverse the string (except for the possible "-" sign). |
| 714 | while (--ptr > start) { |
| 715 | char ch = *ptr; |
| 716 | *ptr = *start; |
| 717 | *start++ = ch; |
| 718 | } |
| 719 | return buf; |
| 720 | } |
| 721 | |
| 722 | // Safely appends string |source| to string |dest|. Never writes past the |
| 723 | // buffer size |dest_size| and guarantees that |dest| is null-terminated. |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 724 | static void SafeAppendString(const char* source, char* dest, size_t dest_size) { |
| 725 | size_t dest_string_length = strlen(dest); |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 726 | SAFE_ASSERT(dest_string_length < dest_size); |
| 727 | dest += dest_string_length; |
| 728 | dest_size -= dest_string_length; |
| 729 | strncpy(dest, source, dest_size); |
| 730 | // Making sure |dest| is always null-terminated. |
| 731 | dest[dest_size - 1] = '\0'; |
| 732 | } |
| 733 | |
| 734 | // Converts a 64-bit value into a hex string, and safely appends it to |dest|. |
| 735 | // Never writes past the buffer size |dest_size| and guarantees that |dest| is |
| 736 | // null-terminated. |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 737 | static void SafeAppendHexNumber(uint64_t value, char* dest, size_t dest_size) { |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 738 | // 64-bit numbers in hex can have up to 16 digits. |
| 739 | char buf[17] = {'\0'}; |
| 740 | SafeAppendString(itoa_r(value, buf, sizeof(buf), 16, 0), dest, dest_size); |
| 741 | } |
| 742 | |
| 743 | // The implementation of our symbolization routine. If it |
| 744 | // successfully finds the symbol containing "pc" and obtains the |
| 745 | // symbol name, returns true and write the symbol name to "out". |
| 746 | // Otherwise, returns false. If Callback function is installed via |
| 747 | // InstallSymbolizeCallback(), the function is also called in this function, |
| 748 | // and "out" is used as its output. |
| 749 | // To keep stack consumption low, we would like this function to not |
| 750 | // get inlined. |
| 751 | static ATTRIBUTE_NOINLINE bool SymbolizeAndDemangle(void *pc, char *out, |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 752 | size_t out_size) { |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 753 | uint64_t pc0 = reinterpret_cast<uintptr_t>(pc); |
| 754 | uint64_t start_address = 0; |
| 755 | uint64_t base_address = 0; |
| 756 | int object_fd = -1; |
| 757 | |
| 758 | if (out_size < 1) { |
| 759 | return false; |
| 760 | } |
| 761 | out[0] = '\0'; |
| 762 | SafeAppendString("(", out, out_size); |
| 763 | |
| 764 | if (g_symbolize_open_object_file_callback) { |
| 765 | object_fd = g_symbolize_open_object_file_callback(pc0, start_address, |
| 766 | base_address, out + 1, |
| 767 | out_size - 1); |
| 768 | } else { |
| 769 | object_fd = OpenObjectFileContainingPcAndGetStartAddress(pc0, start_address, |
| 770 | base_address, |
| 771 | out + 1, |
| 772 | out_size - 1); |
| 773 | } |
| 774 | |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 775 | FileDescriptor wrapped_object_fd(object_fd); |
| 776 | |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 777 | #if defined(PRINT_UNSYMBOLIZED_STACK_TRACES) |
| 778 | { |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 779 | #else |
| 780 | // Check whether a file name was returned. |
| 781 | if (object_fd < 0) { |
| 782 | #endif |
| 783 | if (out[1]) { |
| 784 | // The object file containing PC was determined successfully however the |
| 785 | // object file was not opened successfully. This is still considered |
| 786 | // success because the object file name and offset are known and tools |
| 787 | // like asan_symbolize.py can be used for the symbolization. |
| 788 | out[out_size - 1] = '\0'; // Making sure |out| is always null-terminated. |
| 789 | SafeAppendString("+0x", out, out_size); |
| 790 | SafeAppendHexNumber(pc0 - base_address, out, out_size); |
| 791 | SafeAppendString(")", out, out_size); |
| 792 | return true; |
| 793 | } |
| 794 | // Failed to determine the object file containing PC. Bail out. |
| 795 | return false; |
| 796 | } |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 797 | int elf_type = FileGetElfType(wrapped_object_fd.get()); |
| 798 | if (elf_type == -1) { |
| 799 | return false; |
| 800 | } |
| 801 | if (g_symbolize_callback) { |
| 802 | // Run the call back if it's installed. |
| 803 | // Note: relocation (and much of the rest of this code) will be |
| 804 | // wrong for prelinked shared libraries and PIE executables. |
| 805 | uint64_t relocation = (elf_type == ET_DYN) ? start_address : 0; |
| 806 | int num_bytes_written = g_symbolize_callback(wrapped_object_fd.get(), |
| 807 | pc, out, out_size, |
| 808 | relocation); |
| 809 | if (num_bytes_written > 0) { |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 810 | out += static_cast<size_t>(num_bytes_written); |
| 811 | out_size -= static_cast<size_t>(num_bytes_written); |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 812 | } |
| 813 | } |
| 814 | if (!GetSymbolFromObjectFile(wrapped_object_fd.get(), pc0, |
| 815 | out, out_size, base_address)) { |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 816 | if (out[1] && !g_symbolize_callback) { |
| 817 | // The object file containing PC was opened successfully however the |
| 818 | // symbol was not found. The object may have been stripped. This is still |
| 819 | // considered success because the object file name and offset are known |
| 820 | // and tools like asan_symbolize.py can be used for the symbolization. |
| 821 | out[out_size - 1] = '\0'; // Making sure |out| is always null-terminated. |
| 822 | SafeAppendString("+0x", out, out_size); |
| 823 | SafeAppendHexNumber(pc0 - base_address, out, out_size); |
| 824 | SafeAppendString(")", out, out_size); |
| 825 | return true; |
| 826 | } |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 827 | return false; |
| 828 | } |
| 829 | |
| 830 | // Symbolization succeeded. Now we try to demangle the symbol. |
| 831 | DemangleInplace(out, out_size); |
| 832 | return true; |
| 833 | } |
| 834 | |
| 835 | _END_GOOGLE_NAMESPACE_ |
| 836 | |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 837 | #elif (defined(GLOG_OS_MACOSX) || defined(GLOG_OS_EMSCRIPTEN)) && defined(HAVE_DLADDR) |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 838 | |
| 839 | #include <dlfcn.h> |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 840 | #include <cstring> |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 841 | |
| 842 | _START_GOOGLE_NAMESPACE_ |
| 843 | |
| 844 | static ATTRIBUTE_NOINLINE bool SymbolizeAndDemangle(void *pc, char *out, |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 845 | size_t out_size) { |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 846 | Dl_info info; |
| 847 | if (dladdr(pc, &info)) { |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 848 | if (info.dli_sname) { |
| 849 | if (strlen(info.dli_sname) < out_size) { |
| 850 | strcpy(out, info.dli_sname); |
| 851 | // Symbolization succeeded. Now we try to demangle the symbol. |
| 852 | DemangleInplace(out, out_size); |
| 853 | return true; |
| 854 | } |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 855 | } |
| 856 | } |
| 857 | return false; |
| 858 | } |
| 859 | |
| 860 | _END_GOOGLE_NAMESPACE_ |
| 861 | |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 862 | #elif defined(GLOG_OS_WINDOWS) || defined(GLOG_OS_CYGWIN) |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 863 | |
| 864 | #include <windows.h> |
| 865 | #include <dbghelp.h> |
| 866 | |
| 867 | #ifdef _MSC_VER |
| 868 | #pragma comment(lib, "dbghelp") |
| 869 | #endif |
| 870 | |
| 871 | _START_GOOGLE_NAMESPACE_ |
| 872 | |
| 873 | class SymInitializer { |
| 874 | public: |
| 875 | HANDLE process; |
| 876 | bool ready; |
| 877 | SymInitializer() : process(NULL), ready(false) { |
| 878 | // Initialize the symbol handler. |
| 879 | // https://msdn.microsoft.com/en-us/library/windows/desktop/ms680344(v=vs.85).aspx |
| 880 | process = GetCurrentProcess(); |
| 881 | // Defer symbol loading. |
| 882 | // We do not request undecorated symbols with SYMOPT_UNDNAME |
| 883 | // because the mangling library calls UnDecorateSymbolName. |
| 884 | SymSetOptions(SYMOPT_DEFERRED_LOADS); |
| 885 | if (SymInitialize(process, NULL, true)) { |
| 886 | ready = true; |
| 887 | } |
| 888 | } |
| 889 | ~SymInitializer() { |
| 890 | SymCleanup(process); |
| 891 | // We do not need to close `HANDLE process` because it's a "pseudo handle." |
| 892 | } |
| 893 | private: |
| 894 | SymInitializer(const SymInitializer&); |
| 895 | SymInitializer& operator=(const SymInitializer&); |
| 896 | }; |
| 897 | |
| 898 | static ATTRIBUTE_NOINLINE bool SymbolizeAndDemangle(void *pc, char *out, |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 899 | size_t out_size) { |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 900 | const static SymInitializer symInitializer; |
| 901 | if (!symInitializer.ready) { |
| 902 | return false; |
| 903 | } |
| 904 | // Resolve symbol information from address. |
| 905 | // https://msdn.microsoft.com/en-us/library/windows/desktop/ms680578(v=vs.85).aspx |
| 906 | char buf[sizeof(SYMBOL_INFO) + MAX_SYM_NAME]; |
| 907 | SYMBOL_INFO *symbol = reinterpret_cast<SYMBOL_INFO *>(buf); |
| 908 | symbol->SizeOfStruct = sizeof(SYMBOL_INFO); |
| 909 | symbol->MaxNameLen = MAX_SYM_NAME; |
| 910 | // We use the ANSI version to ensure the string type is always `char *`. |
| 911 | // This could break if a symbol has Unicode in it. |
| 912 | BOOL ret = SymFromAddr(symInitializer.process, |
| 913 | reinterpret_cast<DWORD64>(pc), 0, symbol); |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 914 | if (ret == 1 && static_cast<ssize_t>(symbol->NameLen) < out_size) { |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 915 | // `NameLen` does not include the null terminating character. |
| 916 | strncpy(out, symbol->Name, static_cast<size_t>(symbol->NameLen) + 1); |
| 917 | out[static_cast<size_t>(symbol->NameLen)] = '\0'; |
| 918 | // Symbolization succeeded. Now we try to demangle the symbol. |
| 919 | DemangleInplace(out, out_size); |
| 920 | return true; |
| 921 | } |
| 922 | return false; |
| 923 | } |
| 924 | |
| 925 | _END_GOOGLE_NAMESPACE_ |
| 926 | |
| 927 | #else |
| 928 | # error BUG: HAVE_SYMBOLIZE was wrongly set |
| 929 | #endif |
| 930 | |
| 931 | _START_GOOGLE_NAMESPACE_ |
| 932 | |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 933 | bool Symbolize(void *pc, char *out, size_t out_size) { |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 934 | return SymbolizeAndDemangle(pc, out, out_size); |
| 935 | } |
| 936 | |
| 937 | _END_GOOGLE_NAMESPACE_ |
| 938 | |
| 939 | #else /* HAVE_SYMBOLIZE */ |
| 940 | |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 941 | #include <cassert> |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 942 | |
| 943 | #include "config.h" |
| 944 | |
| 945 | _START_GOOGLE_NAMESPACE_ |
| 946 | |
| 947 | // TODO: Support other environments. |
James Kuszmaul | ba0ac1a | 2022-08-12 16:29:30 -0700 | [diff] [blame] | 948 | bool Symbolize(void* /*pc*/, char* /*out*/, size_t /*out_size*/) { |
Austin Schuh | 906616c | 2019-01-21 20:25:11 -0800 | [diff] [blame] | 949 | assert(0); |
| 950 | return false; |
| 951 | } |
| 952 | |
| 953 | _END_GOOGLE_NAMESPACE_ |
| 954 | |
| 955 | #endif |