| 1 | // stringpool.h -- a string pool for gold -*- C++ -*- |
| 2 | |
| 3 | // Copyright (C) 2006-2019 Free Software Foundation, Inc. |
| 4 | // Written by Ian Lance Taylor <iant@google.com>. |
| 5 | |
| 6 | // This file is part of gold. |
| 7 | |
| 8 | // This program is free software; you can redistribute it and/or modify |
| 9 | // it under the terms of the GNU General Public License as published by |
| 10 | // the Free Software Foundation; either version 3 of the License, or |
| 11 | // (at your option) any later version. |
| 12 | |
| 13 | // This program is distributed in the hope that it will be useful, |
| 14 | // but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 15 | // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 16 | // GNU General Public License for more details. |
| 17 | |
| 18 | // You should have received a copy of the GNU General Public License |
| 19 | // along with this program; if not, write to the Free Software |
| 20 | // Foundation, Inc., 51 Franklin Street - Fifth Floor, Boston, |
| 21 | // MA 02110-1301, USA. |
| 22 | |
| 23 | #include <string> |
| 24 | #include <list> |
| 25 | #include <vector> |
| 26 | |
| 27 | #ifndef GOLD_STRINGPOOL_H |
| 28 | #define GOLD_STRINGPOOL_H |
| 29 | |
| 30 | namespace gold |
| 31 | { |
| 32 | |
| 33 | class Output_file; |
| 34 | |
| 35 | // Return the length of a string in units of Char_type. |
| 36 | |
| 37 | template<typename Char_type> |
| 38 | inline size_t |
| 39 | string_length(const Char_type* p) |
| 40 | { |
| 41 | size_t len = 0; |
| 42 | for (; *p != 0; ++p) |
| 43 | ++len; |
| 44 | return len; |
| 45 | } |
| 46 | |
| 47 | // Specialize string_length for char. Maybe we could just use |
| 48 | // std::char_traits<>::length? |
| 49 | |
| 50 | template<> |
| 51 | inline size_t |
| 52 | string_length(const char* p) |
| 53 | { |
| 54 | return strlen(p); |
| 55 | } |
| 56 | |
| 57 | // A Stringpool is a pool of unique strings. It provides the |
| 58 | // following features: |
| 59 | |
| 60 | // Every string in the pool is unique. Thus, if you have two strings |
| 61 | // in the Stringpool, you can compare them for equality by using |
| 62 | // pointer comparison rather than string comparison. |
| 63 | |
| 64 | // There is a key associated with every string in the pool. If you |
| 65 | // add strings to the Stringpool in the same order, then the key for |
| 66 | // each string will always be the same for any run of the linker. |
| 67 | // This is not true of the string pointers themselves, as they may |
| 68 | // change due to address space randomization. Some parts of the |
| 69 | // linker (e.g., the symbol table) use the key value instead of the |
| 70 | // string pointer so that repeated runs of the linker will generate |
| 71 | // precisely the same output. |
| 72 | |
| 73 | // When you add a string to a Stringpool, Stringpool will optionally |
| 74 | // make a copy of it. Thus there is no requirement to keep a copy |
| 75 | // elsewhere. |
| 76 | |
| 77 | // A Stringpool can be turned into a string table, a sequential series |
| 78 | // of null terminated strings. The first string may optionally be a |
| 79 | // single zero byte, as required for SHT_STRTAB sections. This |
| 80 | // conversion is only permitted after all strings have been added to |
| 81 | // the Stringpool. After doing this conversion, you can ask for the |
| 82 | // offset of any string (or any key) in the stringpool in the string |
| 83 | // table, and you can write the resulting string table to an output |
| 84 | // file. |
| 85 | |
| 86 | // When a Stringpool is turned into a string table, then as an |
| 87 | // optimization it will reuse string suffixes to avoid duplicating |
| 88 | // strings. That is, given the strings "abc" and "bc", only the |
| 89 | // string "abc" will be stored, and "bc" will be represented by an |
| 90 | // offset into the middle of the string "abc". |
| 91 | |
| 92 | |
| 93 | // A simple chunked vector class--this is a subset of std::vector |
| 94 | // which stores memory in chunks. We don't provide iterators, because |
| 95 | // we don't need them. |
| 96 | |
| 97 | template<typename Element> |
| 98 | class Chunked_vector |
| 99 | { |
| 100 | public: |
| 101 | Chunked_vector() |
| 102 | : chunks_(), size_(0) |
| 103 | { } |
| 104 | |
| 105 | // Clear the elements. |
| 106 | void |
| 107 | clear() |
| 108 | { |
| 109 | this->chunks_.clear(); |
| 110 | this->size_ = 0; |
| 111 | } |
| 112 | |
| 113 | // Reserve elements. |
| 114 | void |
| 115 | reserve(unsigned int n) |
| 116 | { |
| 117 | if (n > this->chunks_.size() * chunk_size) |
| 118 | { |
| 119 | this->chunks_.resize((n + chunk_size - 1) / chunk_size); |
| 120 | // We need to call reserve() of all chunks since changing |
| 121 | // this->chunks_ causes Element_vectors to be copied. The |
| 122 | // reserved capacity of an Element_vector may be lost in copying. |
| 123 | for (size_t i = 0; i < this->chunks_.size(); ++i) |
| 124 | this->chunks_[i].reserve(chunk_size); |
| 125 | } |
| 126 | } |
| 127 | |
| 128 | // Get the number of elements. |
| 129 | size_t |
| 130 | size() const |
| 131 | { return this->size_; } |
| 132 | |
| 133 | // Push a new element on the back of the vector. |
| 134 | void |
| 135 | push_back(const Element& element) |
| 136 | { |
| 137 | size_t chunk_index = this->size_ / chunk_size; |
| 138 | if (chunk_index >= this->chunks_.size()) |
| 139 | { |
| 140 | this->chunks_.push_back(Element_vector()); |
| 141 | this->chunks_.back().reserve(chunk_size); |
| 142 | gold_assert(chunk_index < this->chunks_.size()); |
| 143 | } |
| 144 | this->chunks_[chunk_index].push_back(element); |
| 145 | this->size_++; |
| 146 | } |
| 147 | |
| 148 | // Return a reference to an entry in the vector. |
| 149 | Element& |
| 150 | operator[](size_t i) |
| 151 | { return this->chunks_[i / chunk_size][i % chunk_size]; } |
| 152 | |
| 153 | const Element& |
| 154 | operator[](size_t i) const |
| 155 | { return this->chunks_[i / chunk_size][i % chunk_size]; } |
| 156 | |
| 157 | private: |
| 158 | static const unsigned int chunk_size = 8192; |
| 159 | |
| 160 | typedef std::vector<Element> Element_vector; |
| 161 | typedef std::vector<Element_vector> Chunk_vector; |
| 162 | |
| 163 | Chunk_vector chunks_; |
| 164 | size_t size_; |
| 165 | }; |
| 166 | |
| 167 | |
| 168 | // Stringpools are implemented in terms of Stringpool_template, which |
| 169 | // is generalized on the type of character used for the strings. Most |
| 170 | // uses will want the Stringpool type which uses char. Other cases |
| 171 | // are used for merging wide string constants. |
| 172 | |
| 173 | template<typename Stringpool_char> |
| 174 | class Stringpool_template |
| 175 | { |
| 176 | public: |
| 177 | // The type of a key into the stringpool. As described above, a key |
| 178 | // value will always be the same during any run of the linker. Zero |
| 179 | // is never a valid key value. |
| 180 | typedef size_t Key; |
| 181 | |
| 182 | // Create a Stringpool. |
| 183 | Stringpool_template(uint64_t addralign = 1); |
| 184 | |
| 185 | ~Stringpool_template(); |
| 186 | |
| 187 | // Clear all the data from the stringpool. |
| 188 | void |
| 189 | clear(); |
| 190 | |
| 191 | // Hint to the stringpool class that you intend to insert n additional |
| 192 | // elements. The stringpool class can use this info however it likes; |
| 193 | // in practice it will resize its internal hashtables to make room. |
| 194 | void |
| 195 | reserve(unsigned int n); |
| 196 | |
| 197 | // Indicate that we should not reserve offset 0 to hold the empty |
| 198 | // string when converting the stringpool to a string table. This |
| 199 | // should not be called for a proper ELF SHT_STRTAB section. |
| 200 | void |
| 201 | set_no_zero_null() |
| 202 | { |
| 203 | gold_assert(this->string_set_.empty() |
| 204 | && this->offset_ == sizeof(Stringpool_char)); |
| 205 | this->zero_null_ = false; |
| 206 | this->offset_ = 0; |
| 207 | } |
| 208 | |
| 209 | // Indicate that this string pool should be optimized, even if not |
| 210 | // running with -O2. |
| 211 | void |
| 212 | set_optimize() |
| 213 | { this->optimize_ = true; } |
| 214 | |
| 215 | // Add the string S to the pool. This returns a canonical permanent |
| 216 | // pointer to the string in the pool. If COPY is true, the string |
| 217 | // is copied into permanent storage. If PKEY is not NULL, this sets |
| 218 | // *PKEY to the key for the string. |
| 219 | const Stringpool_char* |
| 220 | add(const Stringpool_char* s, bool copy, Key* pkey); |
| 221 | |
| 222 | // Add the string S to the pool. |
| 223 | const Stringpool_char* |
| 224 | add(const std::basic_string<Stringpool_char>& s, bool copy, Key* pkey) |
| 225 | { return this->add_with_length(s.data(), s.size(), copy, pkey); } |
| 226 | |
| 227 | // Add string S of length LEN characters to the pool. If COPY is |
| 228 | // true, S need not be null terminated. |
| 229 | const Stringpool_char* |
| 230 | add_with_length(const Stringpool_char* s, size_t len, bool copy, Key* pkey); |
| 231 | |
| 232 | // If the string S is present in the pool, return the canonical |
| 233 | // string pointer. Otherwise, return NULL. If PKEY is not NULL, |
| 234 | // set *PKEY to the key. |
| 235 | const Stringpool_char* |
| 236 | find(const Stringpool_char* s, Key* pkey) const; |
| 237 | |
| 238 | // Turn the stringpool into a string table: determine the offsets of |
| 239 | // all the strings. After this is called, no more strings may be |
| 240 | // added to the stringpool. |
| 241 | void |
| 242 | set_string_offsets(); |
| 243 | |
| 244 | // Get the offset of the string S in the string table. This returns |
| 245 | // the offset in bytes, not in units of Stringpool_char. This may |
| 246 | // only be called after set_string_offsets has been called. |
| 247 | section_offset_type |
| 248 | get_offset(const Stringpool_char* s) const; |
| 249 | |
| 250 | // Get the offset of the string S in the string table. |
| 251 | section_offset_type |
| 252 | get_offset(const std::basic_string<Stringpool_char>& s) const |
| 253 | { return this->get_offset_with_length(s.c_str(), s.size()); } |
| 254 | |
| 255 | // Get the offset of string S, with length LENGTH characters, in the |
| 256 | // string table. |
| 257 | section_offset_type |
| 258 | get_offset_with_length(const Stringpool_char* s, size_t length) const; |
| 259 | |
| 260 | // Get the offset of the string with key K. |
| 261 | section_offset_type |
| 262 | get_offset_from_key(Key k) const |
| 263 | { |
| 264 | gold_assert(k <= this->key_to_offset_.size()); |
| 265 | return this->key_to_offset_[k - 1]; |
| 266 | } |
| 267 | |
| 268 | // Get the size of the string table. This returns the number of |
| 269 | // bytes, not in units of Stringpool_char. |
| 270 | section_size_type |
| 271 | get_strtab_size() const |
| 272 | { |
| 273 | gold_assert(this->strtab_size_ != 0); |
| 274 | return this->strtab_size_; |
| 275 | } |
| 276 | |
| 277 | // Write the string table into the output file at the specified |
| 278 | // offset. |
| 279 | void |
| 280 | write(Output_file*, off_t offset); |
| 281 | |
| 282 | // Write the string table into the specified buffer, of the |
| 283 | // specified size. buffer_size should be at least |
| 284 | // get_strtab_size(). |
| 285 | void |
| 286 | write_to_buffer(unsigned char* buffer, section_size_type buffer_size); |
| 287 | |
| 288 | // Dump statistical information to stderr. |
| 289 | void |
| 290 | print_stats(const char*) const; |
| 291 | |
| 292 | private: |
| 293 | Stringpool_template(const Stringpool_template&); |
| 294 | Stringpool_template& operator=(const Stringpool_template&); |
| 295 | |
| 296 | // Return whether two strings are equal. |
| 297 | static bool |
| 298 | string_equal(const Stringpool_char*, const Stringpool_char*); |
| 299 | |
| 300 | // Compute a hash code for a string. LENGTH is the length of the |
| 301 | // string in characters. |
| 302 | static size_t |
| 303 | string_hash(const Stringpool_char*, size_t length); |
| 304 | |
| 305 | // We store the actual data in a list of these buffers. |
| 306 | struct Stringdata |
| 307 | { |
| 308 | // Length of data in buffer. |
| 309 | size_t len; |
| 310 | // Allocated size of buffer. |
| 311 | size_t alc; |
| 312 | // Buffer. |
| 313 | char data[1]; |
| 314 | }; |
| 315 | |
| 316 | // Add a new key offset entry. |
| 317 | void |
| 318 | new_key_offset(size_t); |
| 319 | |
| 320 | // Copy a string into the buffers, returning a canonical string. |
| 321 | const Stringpool_char* |
| 322 | add_string(const Stringpool_char*, size_t); |
| 323 | |
| 324 | // Return whether s1 is a suffix of s2. |
| 325 | static bool |
| 326 | is_suffix(const Stringpool_char* s1, size_t len1, |
| 327 | const Stringpool_char* s2, size_t len2); |
| 328 | |
| 329 | // The hash table key includes the string, the length of the string, |
| 330 | // and the hash code for the string. We put the hash code |
| 331 | // explicitly into the key so that we can do a find()/insert() |
| 332 | // sequence without having to recompute the hash. Computing the |
| 333 | // hash code is a significant user of CPU time in the linker. |
| 334 | struct Hashkey |
| 335 | { |
| 336 | const Stringpool_char* string; |
| 337 | // Length is in characters, not bytes. |
| 338 | size_t length; |
| 339 | size_t hash_code; |
| 340 | |
| 341 | // This goes in an STL container, so we need a default |
| 342 | // constructor. |
| 343 | Hashkey() |
| 344 | : string(NULL), length(0), hash_code(0) |
| 345 | { } |
| 346 | |
| 347 | // Note that these constructors are relatively expensive, because |
| 348 | // they compute the hash code. |
| 349 | explicit Hashkey(const Stringpool_char* s) |
| 350 | : string(s), length(string_length(s)), hash_code(string_hash(s, length)) |
| 351 | { } |
| 352 | |
| 353 | Hashkey(const Stringpool_char* s, size_t len) |
| 354 | : string(s), length(len), hash_code(string_hash(s, len)) |
| 355 | { } |
| 356 | }; |
| 357 | |
| 358 | // Hash function. This is trivial, since we have already computed |
| 359 | // the hash. |
| 360 | struct Stringpool_hash |
| 361 | { |
| 362 | size_t |
| 363 | operator()(const Hashkey& hk) const |
| 364 | { return hk.hash_code; } |
| 365 | }; |
| 366 | |
| 367 | // Equality comparison function for hash table. |
| 368 | struct Stringpool_eq |
| 369 | { |
| 370 | bool |
| 371 | operator()(const Hashkey&, const Hashkey&) const; |
| 372 | }; |
| 373 | |
| 374 | // The hash table is a map from strings to Keys. |
| 375 | |
| 376 | typedef Key Hashval; |
| 377 | |
| 378 | typedef Unordered_map<Hashkey, Hashval, Stringpool_hash, |
| 379 | Stringpool_eq> String_set_type; |
| 380 | |
| 381 | // Comparison routine used when sorting into a string table. |
| 382 | |
| 383 | typedef typename String_set_type::iterator Stringpool_sort_info; |
| 384 | |
| 385 | struct Stringpool_sort_comparison |
| 386 | { |
| 387 | bool |
| 388 | operator()(const Stringpool_sort_info&, const Stringpool_sort_info&) const; |
| 389 | }; |
| 390 | |
| 391 | // Keys map to offsets via a Chunked_vector. We only use the |
| 392 | // offsets if we turn this into an string table section. |
| 393 | typedef Chunked_vector<section_offset_type> Key_to_offset; |
| 394 | |
| 395 | // List of Stringdata structures. |
| 396 | typedef std::list<Stringdata*> Stringdata_list; |
| 397 | |
| 398 | // Mapping from const char* to namepool entry. |
| 399 | String_set_type string_set_; |
| 400 | // Mapping from Key to string table offset. |
| 401 | Key_to_offset key_to_offset_; |
| 402 | // List of buffers. |
| 403 | Stringdata_list strings_; |
| 404 | // Size of string table. |
| 405 | section_size_type strtab_size_; |
| 406 | // Whether to reserve offset 0 to hold the null string. |
| 407 | bool zero_null_; |
| 408 | // Whether to optimize the string table. |
| 409 | bool optimize_; |
| 410 | // offset of the next string. |
| 411 | section_offset_type offset_; |
| 412 | // The alignment of strings in the stringpool. |
| 413 | uint64_t addralign_; |
| 414 | }; |
| 415 | |
| 416 | // The most common type of Stringpool. |
| 417 | typedef Stringpool_template<char> Stringpool; |
| 418 | |
| 419 | } // End namespace gold. |
| 420 | |
| 421 | #endif // !defined(GOLD_STRINGPOOL_H) |