Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | #include <string.h> |
| 3 | #include "util.h" |
| 4 | #include "debug.h" |
| 5 | |
| 6 | #include "demangle-rust.h" |
| 7 | |
| 8 | /* |
| 9 | * Mangled Rust symbols look like this: |
| 10 | * |
| 11 | * _$LT$std..sys..fd..FileDesc$u20$as$u20$core..ops..Drop$GT$::drop::hc68340e1baa4987a |
| 12 | * |
| 13 | * The original symbol is: |
| 14 | * |
| 15 | * <std::sys::fd::FileDesc as core::ops::Drop>::drop |
| 16 | * |
| 17 | * The last component of the path is a 64-bit hash in lowercase hex, prefixed |
| 18 | * with "h". Rust does not have a global namespace between crates, an illusion |
| 19 | * which Rust maintains by using the hash to distinguish things that would |
| 20 | * otherwise have the same symbol. |
| 21 | * |
| 22 | * Any path component not starting with a XID_Start character is prefixed with |
| 23 | * "_". |
| 24 | * |
| 25 | * The following escape sequences are used: |
| 26 | * |
| 27 | * "," => $C$ |
| 28 | * "@" => $SP$ |
| 29 | * "*" => $BP$ |
| 30 | * "&" => $RF$ |
| 31 | * "<" => $LT$ |
| 32 | * ">" => $GT$ |
| 33 | * "(" => $LP$ |
| 34 | * ")" => $RP$ |
| 35 | * " " => $u20$ |
| 36 | * "'" => $u27$ |
| 37 | * "[" => $u5b$ |
| 38 | * "]" => $u5d$ |
| 39 | * "~" => $u7e$ |
| 40 | * |
| 41 | * A double ".." means "::" and a single "." means "-". |
| 42 | * |
| 43 | * The only characters allowed in the mangled symbol are a-zA-Z0-9 and _.:$ |
| 44 | */ |
| 45 | |
| 46 | static const char *hash_prefix = "::h"; |
| 47 | static const size_t hash_prefix_len = 3; |
| 48 | static const size_t hash_len = 16; |
| 49 | |
| 50 | static bool is_prefixed_hash(const char *start); |
| 51 | static bool looks_like_rust(const char *sym, size_t len); |
| 52 | static bool unescape(const char **in, char **out, const char *seq, char value); |
| 53 | |
| 54 | /* |
| 55 | * INPUT: |
| 56 | * sym: symbol that has been through BFD-demangling |
| 57 | * |
| 58 | * This function looks for the following indicators: |
| 59 | * |
| 60 | * 1. The hash must consist of "h" followed by 16 lowercase hex digits. |
| 61 | * |
| 62 | * 2. As a sanity check, the hash must use between 5 and 15 of the 16 possible |
| 63 | * hex digits. This is true of 99.9998% of hashes so once in your life you |
| 64 | * may see a false negative. The point is to notice path components that |
| 65 | * could be Rust hashes but are probably not, like "haaaaaaaaaaaaaaaa". In |
| 66 | * this case a false positive (non-Rust symbol has an important path |
| 67 | * component removed because it looks like a Rust hash) is worse than a |
| 68 | * false negative (the rare Rust symbol is not demangled) so this sets the |
| 69 | * balance in favor of false negatives. |
| 70 | * |
| 71 | * 3. There must be no characters other than a-zA-Z0-9 and _.:$ |
| 72 | * |
| 73 | * 4. There must be no unrecognized $-sign sequences. |
| 74 | * |
| 75 | * 5. There must be no sequence of three or more dots in a row ("..."). |
| 76 | */ |
| 77 | bool |
| 78 | rust_is_mangled(const char *sym) |
| 79 | { |
| 80 | size_t len, len_without_hash; |
| 81 | |
| 82 | if (!sym) |
| 83 | return false; |
| 84 | |
| 85 | len = strlen(sym); |
| 86 | if (len <= hash_prefix_len + hash_len) |
| 87 | /* Not long enough to contain "::h" + hash + something else */ |
| 88 | return false; |
| 89 | |
| 90 | len_without_hash = len - (hash_prefix_len + hash_len); |
| 91 | if (!is_prefixed_hash(sym + len_without_hash)) |
| 92 | return false; |
| 93 | |
| 94 | return looks_like_rust(sym, len_without_hash); |
| 95 | } |
| 96 | |
| 97 | /* |
| 98 | * A hash is the prefix "::h" followed by 16 lowercase hex digits. The hex |
| 99 | * digits must comprise between 5 and 15 (inclusive) distinct digits. |
| 100 | */ |
| 101 | static bool is_prefixed_hash(const char *str) |
| 102 | { |
| 103 | const char *end; |
| 104 | bool seen[16]; |
| 105 | size_t i; |
| 106 | int count; |
| 107 | |
| 108 | if (strncmp(str, hash_prefix, hash_prefix_len)) |
| 109 | return false; |
| 110 | str += hash_prefix_len; |
| 111 | |
| 112 | memset(seen, false, sizeof(seen)); |
| 113 | for (end = str + hash_len; str < end; str++) |
| 114 | if (*str >= '0' && *str <= '9') |
| 115 | seen[*str - '0'] = true; |
| 116 | else if (*str >= 'a' && *str <= 'f') |
| 117 | seen[*str - 'a' + 10] = true; |
| 118 | else |
| 119 | return false; |
| 120 | |
| 121 | /* Count how many distinct digits seen */ |
| 122 | count = 0; |
| 123 | for (i = 0; i < 16; i++) |
| 124 | if (seen[i]) |
| 125 | count++; |
| 126 | |
| 127 | return count >= 5 && count <= 15; |
| 128 | } |
| 129 | |
| 130 | static bool looks_like_rust(const char *str, size_t len) |
| 131 | { |
| 132 | const char *end = str + len; |
| 133 | |
| 134 | while (str < end) |
| 135 | switch (*str) { |
| 136 | case '$': |
| 137 | if (!strncmp(str, "$C$", 3)) |
| 138 | str += 3; |
| 139 | else if (!strncmp(str, "$SP$", 4) |
| 140 | || !strncmp(str, "$BP$", 4) |
| 141 | || !strncmp(str, "$RF$", 4) |
| 142 | || !strncmp(str, "$LT$", 4) |
| 143 | || !strncmp(str, "$GT$", 4) |
| 144 | || !strncmp(str, "$LP$", 4) |
| 145 | || !strncmp(str, "$RP$", 4)) |
| 146 | str += 4; |
| 147 | else if (!strncmp(str, "$u20$", 5) |
| 148 | || !strncmp(str, "$u27$", 5) |
| 149 | || !strncmp(str, "$u5b$", 5) |
| 150 | || !strncmp(str, "$u5d$", 5) |
| 151 | || !strncmp(str, "$u7e$", 5)) |
| 152 | str += 5; |
| 153 | else |
| 154 | return false; |
| 155 | break; |
| 156 | case '.': |
| 157 | /* Do not allow three or more consecutive dots */ |
| 158 | if (!strncmp(str, "...", 3)) |
| 159 | return false; |
| 160 | /* Fall through */ |
| 161 | case 'a' ... 'z': |
| 162 | case 'A' ... 'Z': |
| 163 | case '0' ... '9': |
| 164 | case '_': |
| 165 | case ':': |
| 166 | str++; |
| 167 | break; |
| 168 | default: |
| 169 | return false; |
| 170 | } |
| 171 | |
| 172 | return true; |
| 173 | } |
| 174 | |
| 175 | /* |
| 176 | * INPUT: |
| 177 | * sym: symbol for which rust_is_mangled(sym) returns true |
| 178 | * |
| 179 | * The input is demangled in-place because the mangled name is always longer |
| 180 | * than the demangled one. |
| 181 | */ |
| 182 | void |
| 183 | rust_demangle_sym(char *sym) |
| 184 | { |
| 185 | const char *in; |
| 186 | char *out; |
| 187 | const char *end; |
| 188 | |
| 189 | if (!sym) |
| 190 | return; |
| 191 | |
| 192 | in = sym; |
| 193 | out = sym; |
| 194 | end = sym + strlen(sym) - (hash_prefix_len + hash_len); |
| 195 | |
| 196 | while (in < end) |
| 197 | switch (*in) { |
| 198 | case '$': |
| 199 | if (!(unescape(&in, &out, "$C$", ',') |
| 200 | || unescape(&in, &out, "$SP$", '@') |
| 201 | || unescape(&in, &out, "$BP$", '*') |
| 202 | || unescape(&in, &out, "$RF$", '&') |
| 203 | || unescape(&in, &out, "$LT$", '<') |
| 204 | || unescape(&in, &out, "$GT$", '>') |
| 205 | || unescape(&in, &out, "$LP$", '(') |
| 206 | || unescape(&in, &out, "$RP$", ')') |
| 207 | || unescape(&in, &out, "$u20$", ' ') |
| 208 | || unescape(&in, &out, "$u27$", '\'') |
| 209 | || unescape(&in, &out, "$u5b$", '[') |
| 210 | || unescape(&in, &out, "$u5d$", ']') |
| 211 | || unescape(&in, &out, "$u7e$", '~'))) { |
| 212 | pr_err("demangle-rust: unexpected escape sequence"); |
| 213 | goto done; |
| 214 | } |
| 215 | break; |
| 216 | case '_': |
| 217 | /* |
| 218 | * If this is the start of a path component and the next |
| 219 | * character is an escape sequence, ignore the |
| 220 | * underscore. The mangler inserts an underscore to make |
| 221 | * sure the path component begins with a XID_Start |
| 222 | * character. |
| 223 | */ |
| 224 | if ((in == sym || in[-1] == ':') && in[1] == '$') |
| 225 | in++; |
| 226 | else |
| 227 | *out++ = *in++; |
| 228 | break; |
| 229 | case '.': |
| 230 | if (in[1] == '.') { |
| 231 | /* ".." becomes "::" */ |
| 232 | *out++ = ':'; |
| 233 | *out++ = ':'; |
| 234 | in += 2; |
| 235 | } else { |
| 236 | /* "." becomes "-" */ |
| 237 | *out++ = '-'; |
| 238 | in++; |
| 239 | } |
| 240 | break; |
| 241 | case 'a' ... 'z': |
| 242 | case 'A' ... 'Z': |
| 243 | case '0' ... '9': |
| 244 | case ':': |
| 245 | *out++ = *in++; |
| 246 | break; |
| 247 | default: |
| 248 | pr_err("demangle-rust: unexpected character '%c' in symbol\n", |
| 249 | *in); |
| 250 | goto done; |
| 251 | } |
| 252 | |
| 253 | done: |
| 254 | *out = '\0'; |
| 255 | } |
| 256 | |
| 257 | static bool unescape(const char **in, char **out, const char *seq, char value) |
| 258 | { |
| 259 | size_t len = strlen(seq); |
| 260 | |
| 261 | if (strncmp(*in, seq, len)) |
| 262 | return false; |
| 263 | |
| 264 | **out = value; |
| 265 | |
| 266 | *in += len; |
| 267 | *out += 1; |
| 268 | |
| 269 | return true; |
| 270 | } |