| 1 | // fileread.h -- read files for gold -*- C++ -*- |
| 2 | |
| 3 | // Copyright 2006, 2007, 2008, 2009, 2010, 2011 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 | // Classes used to read data from binary input files. |
| 24 | |
| 25 | #ifndef GOLD_FILEREAD_H |
| 26 | #define GOLD_FILEREAD_H |
| 27 | |
| 28 | #include <list> |
| 29 | #include <map> |
| 30 | #include <string> |
| 31 | #include <vector> |
| 32 | |
| 33 | #include "token.h" |
| 34 | |
| 35 | namespace gold |
| 36 | { |
| 37 | |
| 38 | // Since not all system supports stat.st_mtim and struct timespec, |
| 39 | // we define our own structure and fill the nanoseconds if we can. |
| 40 | |
| 41 | struct Timespec |
| 42 | { |
| 43 | Timespec() |
| 44 | : seconds(0), nanoseconds(0) |
| 45 | { } |
| 46 | |
| 47 | Timespec(time_t a_seconds, int a_nanoseconds) |
| 48 | : seconds(a_seconds), nanoseconds(a_nanoseconds) |
| 49 | { } |
| 50 | |
| 51 | time_t seconds; |
| 52 | int nanoseconds; |
| 53 | }; |
| 54 | |
| 55 | // Get the last modified time of an unopened file. Returns false if the |
| 56 | // file does not exist. |
| 57 | |
| 58 | bool |
| 59 | get_mtime(const char* filename, Timespec* mtime); |
| 60 | |
| 61 | class Position_dependent_options; |
| 62 | class Input_file_argument; |
| 63 | class Dirsearch; |
| 64 | class File_view; |
| 65 | |
| 66 | // File_read manages a file descriptor and mappings for a file we are |
| 67 | // reading. |
| 68 | |
| 69 | class File_read |
| 70 | { |
| 71 | public: |
| 72 | File_read() |
| 73 | : name_(), descriptor_(-1), is_descriptor_opened_(false), object_count_(0), |
| 74 | size_(0), token_(false), views_(), saved_views_(), mapped_bytes_(0), |
| 75 | released_(true), whole_file_view_(NULL) |
| 76 | { } |
| 77 | |
| 78 | ~File_read(); |
| 79 | |
| 80 | // Open a file. |
| 81 | bool |
| 82 | open(const Task*, const std::string& name); |
| 83 | |
| 84 | // Pretend to open the file, but provide the file contents. No |
| 85 | // actual file system activity will occur. This is used for |
| 86 | // testing. |
| 87 | bool |
| 88 | open(const Task*, const std::string& name, const unsigned char* contents, |
| 89 | off_t size); |
| 90 | |
| 91 | // Return the file name. |
| 92 | const std::string& |
| 93 | filename() const |
| 94 | { return this->name_; } |
| 95 | |
| 96 | // Add an object associated with a file. |
| 97 | void |
| 98 | add_object() |
| 99 | { ++this->object_count_; } |
| 100 | |
| 101 | // Remove an object associated with a file. |
| 102 | void |
| 103 | remove_object() |
| 104 | { --this->object_count_; } |
| 105 | |
| 106 | // Lock the file for exclusive access within a particular Task::run |
| 107 | // execution. This routine may only be called when the workqueue |
| 108 | // lock is held. |
| 109 | void |
| 110 | lock(const Task* t); |
| 111 | |
| 112 | // Unlock the file. |
| 113 | void |
| 114 | unlock(const Task* t); |
| 115 | |
| 116 | // Test whether the object is locked. |
| 117 | bool |
| 118 | is_locked() const; |
| 119 | |
| 120 | // Return the token, so that the task can be queued. |
| 121 | Task_token* |
| 122 | token() |
| 123 | { return &this->token_; } |
| 124 | |
| 125 | // Release the file. This indicates that we aren't going to do |
| 126 | // anything further with it until it is unlocked. This is used |
| 127 | // because a Task which locks the file never calls either lock or |
| 128 | // unlock; it just locks the token. The basic rule is that a Task |
| 129 | // which locks a file via the Task::locks interface must explicitly |
| 130 | // call release() when it is done. This is not necessary for code |
| 131 | // which calls unlock() on the file. |
| 132 | void |
| 133 | release(); |
| 134 | |
| 135 | // Return the size of the file. |
| 136 | off_t |
| 137 | filesize() const |
| 138 | { return this->size_; } |
| 139 | |
| 140 | // Return a view into the file starting at file offset START for |
| 141 | // SIZE bytes. OFFSET is the offset into the input file for the |
| 142 | // file we are reading; this is zero for a normal object file, |
| 143 | // non-zero for an object file in an archive. ALIGNED is true if |
| 144 | // the data must be naturally aligned; this only matters when OFFSET |
| 145 | // is not zero. The pointer will remain valid until the File_read |
| 146 | // is unlocked. It is an error if we can not read enough data from |
| 147 | // the file. The CACHE parameter is a hint as to whether it will be |
| 148 | // useful to cache this data for later accesses--i.e., later calls |
| 149 | // to get_view, read, or get_lasting_view which retrieve the same |
| 150 | // data. |
| 151 | const unsigned char* |
| 152 | get_view(off_t offset, off_t start, section_size_type size, bool aligned, |
| 153 | bool cache); |
| 154 | |
| 155 | // Read data from the file into the buffer P starting at file offset |
| 156 | // START for SIZE bytes. |
| 157 | void |
| 158 | read(off_t start, section_size_type size, void* p); |
| 159 | |
| 160 | // Return a lasting view into the file starting at file offset START |
| 161 | // for SIZE bytes. This is allocated with new, and the caller is |
| 162 | // responsible for deleting it when done. The data associated with |
| 163 | // this view will remain valid until the view is deleted. It is an |
| 164 | // error if we can not read enough data from the file. The OFFSET, |
| 165 | // ALIGNED and CACHE parameters are as in get_view. |
| 166 | File_view* |
| 167 | get_lasting_view(off_t offset, off_t start, section_size_type size, |
| 168 | bool aligned, bool cache); |
| 169 | |
| 170 | // Mark all views as no longer cached. |
| 171 | void |
| 172 | clear_view_cache_marks(); |
| 173 | |
| 174 | // Discard all uncached views. This is normally done by release(), |
| 175 | // but not for objects in archives. FIXME: This is a complicated |
| 176 | // interface, and it would be nice to have something more automatic. |
| 177 | void |
| 178 | clear_uncached_views() |
| 179 | { this->clear_views(CLEAR_VIEWS_ARCHIVE); } |
| 180 | |
| 181 | // A struct used to do a multiple read. |
| 182 | struct Read_multiple_entry |
| 183 | { |
| 184 | // The file offset of the data to read. |
| 185 | off_t file_offset; |
| 186 | // The amount of data to read. |
| 187 | section_size_type size; |
| 188 | // The buffer where the data should be placed. |
| 189 | unsigned char* buffer; |
| 190 | |
| 191 | Read_multiple_entry(off_t o, section_size_type s, unsigned char* b) |
| 192 | : file_offset(o), size(s), buffer(b) |
| 193 | { } |
| 194 | }; |
| 195 | |
| 196 | typedef std::vector<Read_multiple_entry> Read_multiple; |
| 197 | |
| 198 | // Read a bunch of data from the file into various different |
| 199 | // locations. The vector must be sorted by ascending file_offset. |
| 200 | // BASE is a base offset to be added to all the offsets in the |
| 201 | // vector. |
| 202 | void |
| 203 | read_multiple(off_t base, const Read_multiple&); |
| 204 | |
| 205 | // Dump statistical information to stderr. |
| 206 | static void |
| 207 | print_stats(); |
| 208 | |
| 209 | // Return the open file descriptor (for plugins). |
| 210 | int |
| 211 | descriptor() |
| 212 | { |
| 213 | this->reopen_descriptor(); |
| 214 | return this->descriptor_; |
| 215 | } |
| 216 | |
| 217 | // Return the file last modification time. Calls gold_fatal if the stat |
| 218 | // system call failed. |
| 219 | Timespec |
| 220 | get_mtime(); |
| 221 | |
| 222 | private: |
| 223 | // Control for what views to clear. |
| 224 | enum Clear_views_mode |
| 225 | { |
| 226 | // Clear uncached views not used by an archive. |
| 227 | CLEAR_VIEWS_NORMAL, |
| 228 | // Clear all uncached views (including in an archive). |
| 229 | CLEAR_VIEWS_ARCHIVE, |
| 230 | // Clear all views (i.e., we're destroying the file). |
| 231 | CLEAR_VIEWS_ALL |
| 232 | }; |
| 233 | |
| 234 | // This class may not be copied. |
| 235 | File_read(const File_read&); |
| 236 | File_read& operator=(const File_read&); |
| 237 | |
| 238 | // Total bytes mapped into memory during the link if --stats. |
| 239 | static unsigned long long total_mapped_bytes; |
| 240 | |
| 241 | // Current number of bytes mapped into memory during the link if |
| 242 | // --stats. |
| 243 | static unsigned long long current_mapped_bytes; |
| 244 | |
| 245 | // High water mark of bytes mapped into memory during the link if |
| 246 | // --stats. |
| 247 | static unsigned long long maximum_mapped_bytes; |
| 248 | |
| 249 | // A view into the file. |
| 250 | class View |
| 251 | { |
| 252 | public: |
| 253 | // Specifies how to dispose the data on destruction of the view. |
| 254 | enum Data_ownership |
| 255 | { |
| 256 | // Data owned by File object - nothing done in destructor. |
| 257 | DATA_NOT_OWNED, |
| 258 | // Data allocated with new[] and owned by this object - should |
| 259 | // use delete[]. |
| 260 | DATA_ALLOCATED_ARRAY, |
| 261 | // Data mmapped and owned by this object - should munmap. |
| 262 | DATA_MMAPPED |
| 263 | }; |
| 264 | |
| 265 | View(off_t start, section_size_type size, const unsigned char* data, |
| 266 | unsigned int byteshift, bool cache, Data_ownership data_ownership) |
| 267 | : start_(start), size_(size), data_(data), lock_count_(0), |
| 268 | byteshift_(byteshift), cache_(cache), data_ownership_(data_ownership), |
| 269 | accessed_(true) |
| 270 | { } |
| 271 | |
| 272 | ~View(); |
| 273 | |
| 274 | off_t |
| 275 | start() const |
| 276 | { return this->start_; } |
| 277 | |
| 278 | section_size_type |
| 279 | size() const |
| 280 | { return this->size_; } |
| 281 | |
| 282 | const unsigned char* |
| 283 | data() const |
| 284 | { return this->data_; } |
| 285 | |
| 286 | void |
| 287 | lock(); |
| 288 | |
| 289 | void |
| 290 | unlock(); |
| 291 | |
| 292 | bool |
| 293 | is_locked(); |
| 294 | |
| 295 | unsigned int |
| 296 | byteshift() const |
| 297 | { return this->byteshift_; } |
| 298 | |
| 299 | void |
| 300 | set_cache() |
| 301 | { this->cache_ = true; } |
| 302 | |
| 303 | void |
| 304 | clear_cache() |
| 305 | { this->cache_ = false; } |
| 306 | |
| 307 | bool |
| 308 | should_cache() const |
| 309 | { return this->cache_; } |
| 310 | |
| 311 | void |
| 312 | set_accessed() |
| 313 | { this->accessed_ = true; } |
| 314 | |
| 315 | void |
| 316 | clear_accessed() |
| 317 | { this->accessed_= false; } |
| 318 | |
| 319 | bool |
| 320 | accessed() const |
| 321 | { return this->accessed_; } |
| 322 | |
| 323 | // Returns TRUE if this view contains permanent data -- e.g., data that |
| 324 | // was supplied by the owner of the File object. |
| 325 | bool |
| 326 | is_permanent_view() const |
| 327 | { return this->data_ownership_ == DATA_NOT_OWNED; } |
| 328 | |
| 329 | private: |
| 330 | View(const View&); |
| 331 | View& operator=(const View&); |
| 332 | |
| 333 | // The file offset of the start of the view. |
| 334 | off_t start_; |
| 335 | // The size of the view. |
| 336 | section_size_type size_; |
| 337 | // A pointer to the actual bytes. |
| 338 | const unsigned char* data_; |
| 339 | // The number of locks on this view. |
| 340 | int lock_count_; |
| 341 | // The number of bytes that the view is shifted relative to the |
| 342 | // underlying file. This is used to align data. This is normally |
| 343 | // zero, except possibly for an object in an archive. |
| 344 | unsigned int byteshift_; |
| 345 | // Whether the view is cached. |
| 346 | bool cache_; |
| 347 | // Whether the view is mapped into memory. If not, data_ points |
| 348 | // to memory allocated using new[]. |
| 349 | Data_ownership data_ownership_; |
| 350 | // Whether the view has been accessed recently. |
| 351 | bool accessed_; |
| 352 | }; |
| 353 | |
| 354 | friend class View; |
| 355 | friend class File_view; |
| 356 | |
| 357 | // The type of a mapping from page start and byte shift to views. |
| 358 | typedef std::map<std::pair<off_t, unsigned int>, View*> Views; |
| 359 | |
| 360 | // A simple list of Views. |
| 361 | typedef std::list<View*> Saved_views; |
| 362 | |
| 363 | // Open the descriptor if necessary. |
| 364 | void |
| 365 | reopen_descriptor(); |
| 366 | |
| 367 | // Find a view into the file. |
| 368 | View* |
| 369 | find_view(off_t start, section_size_type size, unsigned int byteshift, |
| 370 | View** vshifted) const; |
| 371 | |
| 372 | // Read data from the file into a buffer. |
| 373 | void |
| 374 | do_read(off_t start, section_size_type size, void* p); |
| 375 | |
| 376 | // Add a view. |
| 377 | void |
| 378 | add_view(View*); |
| 379 | |
| 380 | // Make a view into the file. |
| 381 | View* |
| 382 | make_view(off_t start, section_size_type size, unsigned int byteshift, |
| 383 | bool cache); |
| 384 | |
| 385 | // Find or make a view into the file. |
| 386 | View* |
| 387 | find_or_make_view(off_t offset, off_t start, section_size_type size, |
| 388 | bool aligned, bool cache); |
| 389 | |
| 390 | // Clear the file views. |
| 391 | void |
| 392 | clear_views(Clear_views_mode); |
| 393 | |
| 394 | // The size of a file page for buffering data. |
| 395 | static const off_t page_size = 8192; |
| 396 | |
| 397 | // Given a file offset, return the page offset. |
| 398 | static off_t |
| 399 | page_offset(off_t file_offset) |
| 400 | { return file_offset & ~ (page_size - 1); } |
| 401 | |
| 402 | // Given a file size, return the size to read integral pages. |
| 403 | static off_t |
| 404 | pages(off_t file_size) |
| 405 | { return (file_size + (page_size - 1)) & ~ (page_size - 1); } |
| 406 | |
| 407 | // The maximum number of entries we will pass to ::readv. |
| 408 | static const size_t max_readv_entries = 128; |
| 409 | |
| 410 | // Use readv to read data. |
| 411 | void |
| 412 | do_readv(off_t base, const Read_multiple&, size_t start, size_t count); |
| 413 | |
| 414 | // File name. |
| 415 | std::string name_; |
| 416 | // File descriptor. |
| 417 | int descriptor_; |
| 418 | // Whether we have regained the descriptor after releasing the file. |
| 419 | bool is_descriptor_opened_; |
| 420 | // The number of objects associated with this file. This will be |
| 421 | // more than 1 in the case of an archive. |
| 422 | int object_count_; |
| 423 | // File size. |
| 424 | off_t size_; |
| 425 | // A token used to lock the file. |
| 426 | Task_token token_; |
| 427 | // Buffered views into the file. |
| 428 | Views views_; |
| 429 | // List of views which were locked but had to be removed from views_ |
| 430 | // because they were not large enough. |
| 431 | Saved_views saved_views_; |
| 432 | // Total amount of space mapped into memory. This is only changed |
| 433 | // while the file is locked. When we unlock the file, we transfer |
| 434 | // the total to total_mapped_bytes, and reset this to zero. |
| 435 | size_t mapped_bytes_; |
| 436 | // Whether the file was released. |
| 437 | bool released_; |
| 438 | // A view containing the whole file. May be NULL if we mmap only |
| 439 | // the relevant parts of the file. Not NULL if: |
| 440 | // - Flag --mmap_whole_files is set (default on 64-bit hosts). |
| 441 | // - The contents was specified in the constructor. Used only for |
| 442 | // testing purposes). |
| 443 | View* whole_file_view_; |
| 444 | }; |
| 445 | |
| 446 | // A view of file data that persists even when the file is unlocked. |
| 447 | // Callers should destroy these when no longer required. These are |
| 448 | // obtained form File_read::get_lasting_view. They may only be |
| 449 | // destroyed when the underlying File_read is locked. |
| 450 | |
| 451 | class File_view |
| 452 | { |
| 453 | public: |
| 454 | // This may only be called when the underlying File_read is locked. |
| 455 | ~File_view(); |
| 456 | |
| 457 | // Return a pointer to the data associated with this view. |
| 458 | const unsigned char* |
| 459 | data() const |
| 460 | { return this->data_; } |
| 461 | |
| 462 | private: |
| 463 | File_view(const File_view&); |
| 464 | File_view& operator=(const File_view&); |
| 465 | |
| 466 | friend class File_read; |
| 467 | |
| 468 | // Callers have to get these via File_read::get_lasting_view. |
| 469 | File_view(File_read& file, File_read::View* view, const unsigned char* data) |
| 470 | : file_(file), view_(view), data_(data) |
| 471 | { } |
| 472 | |
| 473 | File_read& file_; |
| 474 | File_read::View* view_; |
| 475 | const unsigned char* data_; |
| 476 | }; |
| 477 | |
| 478 | // All the information we hold for a single input file. This can be |
| 479 | // an object file, a shared library, or an archive. |
| 480 | |
| 481 | class Input_file |
| 482 | { |
| 483 | public: |
| 484 | enum Format |
| 485 | { |
| 486 | FORMAT_NONE, |
| 487 | FORMAT_ELF, |
| 488 | FORMAT_BINARY |
| 489 | }; |
| 490 | |
| 491 | Input_file(const Input_file_argument* input_argument) |
| 492 | : input_argument_(input_argument), found_name_(), file_(), |
| 493 | is_in_sysroot_(false), format_(FORMAT_NONE) |
| 494 | { } |
| 495 | |
| 496 | // Create an input file with the contents already provided. This is |
| 497 | // only used for testing. With this path, don't call the open |
| 498 | // method. |
| 499 | Input_file(const Task*, const char* name, const unsigned char* contents, |
| 500 | off_t size); |
| 501 | |
| 502 | // Return the command line argument. |
| 503 | const Input_file_argument* |
| 504 | input_file_argument() const |
| 505 | { return this->input_argument_; } |
| 506 | |
| 507 | // Return whether this is a file that we will search for in the list |
| 508 | // of directories. |
| 509 | bool |
| 510 | will_search_for() const; |
| 511 | |
| 512 | // Open the file. If the open fails, this will report an error and |
| 513 | // return false. If there is a search, it starts at directory |
| 514 | // *PINDEX. *PINDEX should be initialized to zero. It may be |
| 515 | // restarted to find the next file with a matching name by |
| 516 | // incrementing the result and calling this again. |
| 517 | bool |
| 518 | open(const Dirsearch&, const Task*, int* pindex); |
| 519 | |
| 520 | // Return the name given by the user. For -lc this will return "c". |
| 521 | const char* |
| 522 | name() const; |
| 523 | |
| 524 | // Return the file name. For -lc this will return something like |
| 525 | // "/usr/lib/libc.so". |
| 526 | const std::string& |
| 527 | filename() const |
| 528 | { return this->file_.filename(); } |
| 529 | |
| 530 | // Return the name under which we found the file, corresponding to |
| 531 | // the command line. For -lc this will return something like |
| 532 | // "libc.so". |
| 533 | const std::string& |
| 534 | found_name() const |
| 535 | { return this->found_name_; } |
| 536 | |
| 537 | // Return the position dependent options. |
| 538 | const Position_dependent_options& |
| 539 | options() const; |
| 540 | |
| 541 | // Return the file. |
| 542 | File_read& |
| 543 | file() |
| 544 | { return this->file_; } |
| 545 | |
| 546 | const File_read& |
| 547 | file() const |
| 548 | { return this->file_; } |
| 549 | |
| 550 | // Whether we found the file in a directory in the system root. |
| 551 | bool |
| 552 | is_in_sysroot() const |
| 553 | { return this->is_in_sysroot_; } |
| 554 | |
| 555 | // Whether this file is in a system directory. |
| 556 | bool |
| 557 | is_in_system_directory() const; |
| 558 | |
| 559 | // Return whether this file is to be read only for its symbols. |
| 560 | bool |
| 561 | just_symbols() const; |
| 562 | |
| 563 | // Return the format of the unconverted input file. |
| 564 | Format |
| 565 | format() const |
| 566 | { return this->format_; } |
| 567 | |
| 568 | // Try to find a file in the extra search dirs. Returns true on success. |
| 569 | static bool |
| 570 | try_extra_search_path(int* pindex, |
| 571 | const Input_file_argument* input_argument, |
| 572 | std::string filename, std::string* found_name, |
| 573 | std::string* namep); |
| 574 | |
| 575 | // Find the actual file. |
| 576 | static bool |
| 577 | find_file(const Dirsearch& dirpath, int* pindex, |
| 578 | const Input_file_argument* input_argument, |
| 579 | bool* is_in_sysroot, |
| 580 | std::string* found_name, std::string* namep); |
| 581 | |
| 582 | private: |
| 583 | Input_file(const Input_file&); |
| 584 | Input_file& operator=(const Input_file&); |
| 585 | |
| 586 | // Open a binary file. |
| 587 | bool |
| 588 | open_binary(const Task* task, const std::string& name); |
| 589 | |
| 590 | // The argument from the command line. |
| 591 | const Input_file_argument* input_argument_; |
| 592 | // The name under which we opened the file. This is like the name |
| 593 | // on the command line, but -lc turns into libc.so (or whatever). |
| 594 | // It only includes the full path if the path was on the command |
| 595 | // line. |
| 596 | std::string found_name_; |
| 597 | // The file after we open it. |
| 598 | File_read file_; |
| 599 | // Whether we found the file in a directory in the system root. |
| 600 | bool is_in_sysroot_; |
| 601 | // Format of unconverted input file. |
| 602 | Format format_; |
| 603 | }; |
| 604 | |
| 605 | } // end namespace gold |
| 606 | |
| 607 | #endif // !defined(GOLD_FILEREAD_H) |