script.h revision 1.1
1// script.h -- handle linker scripts for gold -*- C++ -*- 2 3// Copyright 2006, 2007, 2008, 2009, 2010 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// We implement a subset of the original GNU ld linker script language 24// for compatibility. The goal is not to implement the entire 25// language. It is merely to implement enough to handle common uses. 26// In particular we need to handle /usr/lib/libc.so on a typical 27// GNU/Linux system, and we want to handle linker scripts used by the 28// Linux kernel build. 29 30#ifndef GOLD_SCRIPT_H 31#define GOLD_SCRIPT_H 32 33#include <cstdio> 34#include <string> 35#include <vector> 36 37#include "script-sections.h" 38 39namespace gold 40{ 41 42class General_options; 43class Command_line; 44class Symbol_table; 45class Layout; 46class Mapfile; 47class Input_argument; 48class Input_arguments; 49class Input_objects; 50class Input_group; 51class Input_file; 52class Output_segment; 53class Task_token; 54class Workqueue; 55struct Version_dependency_list; 56struct Version_expression_list; 57struct Version_tree; 58struct Version_expression; 59class Lazy_demangler; 60class Incremental_script_entry; 61 62// This class represents an expression in a linker script. 63 64class Expression 65{ 66 protected: 67 // These should only be created by child classes. 68 Expression() 69 { } 70 71 public: 72 virtual ~Expression() 73 { } 74 75 // Return the value of the expression which is not permitted to 76 // refer to the dot symbol. CHECK_ASSERTIONS is true if we should 77 // check whether assertions are true. 78 uint64_t 79 eval(const Symbol_table*, const Layout*, bool check_assertions); 80 81 // Return the value of an expression which is permitted to refer to 82 // the dot symbol. DOT_VALUE is the absolute value of the dot 83 // symbol. DOT_SECTION is the section in which dot is defined; it 84 // should be NULL if the dot symbol has an absolute value (e.g., is 85 // defined in a SECTIONS clause outside of any output section 86 // definition). This sets *RESULT_SECTION to indicate where the 87 // value is defined. If the value is absolute *RESULT_SECTION will 88 // be NULL. Note that the returned value is still an absolute 89 // value; to get a section relative value the caller must subtract 90 // the section address. If RESULT_ALIGNMENT is not NULL, this sets 91 // *RESULT_ALIGNMENT to the alignment of the value of that alignment 92 // is larger than *RESULT_ALIGNMENT; this will only be non-zero if 93 // this is an ALIGN expression. If IS_SECTION_DOT_ASSIGMENT is true, 94 // we are evaluating an assignment to dot within an output section, 95 // and an absolute value should be interpreted as an offset within 96 // the section. 97 uint64_t 98 eval_with_dot(const Symbol_table*, const Layout*, bool check_assertions, 99 uint64_t dot_value, Output_section* dot_section, 100 Output_section** result_section, uint64_t* result_alignment, 101 bool is_section_dot_assignment); 102 103 // Return the value of an expression which may or may not be 104 // permitted to refer to the dot symbol, depending on 105 // is_dot_available. If IS_SECTION_DOT_ASSIGMENT is true, 106 // we are evaluating an assignment to dot within an output section, 107 // and an absolute value should be interpreted as an offset within 108 // the section. 109 uint64_t 110 eval_maybe_dot(const Symbol_table*, const Layout*, bool check_assertions, 111 bool is_dot_available, uint64_t dot_value, 112 Output_section* dot_section, 113 Output_section** result_section, uint64_t* result_alignment, 114 bool is_section_dot_assignment); 115 116 // Print the expression to the FILE. This is for debugging. 117 virtual void 118 print(FILE*) const = 0; 119 120 protected: 121 struct Expression_eval_info; 122 123 public: 124 // Compute the value of the expression (implemented by child class). 125 // This is public rather than protected because it is called 126 // directly by children of Expression on other Expression objects. 127 virtual uint64_t 128 value(const Expression_eval_info*) = 0; 129 130 private: 131 // May not be copied. 132 Expression(const Expression&); 133 Expression& operator=(const Expression&); 134}; 135 136 137// Version_script_info stores information parsed from the version 138// script, either provided by --version-script or as part of a linker 139// script. A single Version_script_info object per target is owned by 140// Script_options. 141 142class Version_script_info 143{ 144 public: 145 // The languages which can be specified in a versionn script. 146 enum Language 147 { 148 LANGUAGE_C, // No demangling. 149 LANGUAGE_CXX, // C++ demangling. 150 LANGUAGE_JAVA, // Java demangling. 151 LANGUAGE_COUNT 152 }; 153 154 Version_script_info(); 155 156 ~Version_script_info(); 157 158 // Clear everything. 159 void 160 clear(); 161 162 // Finalize the version control information. 163 void 164 finalize(); 165 166 // Return whether the information is finalized. 167 bool 168 is_finalized() const 169 { return this->is_finalized_; } 170 171 // Return whether any version were defined in the version script. 172 bool 173 empty() const 174 { return this->version_trees_.empty(); } 175 176 // If there is a version associated with SYMBOL, return true, and 177 // set *VERSION to the version, and *IS_GLOBAL to whether the symbol 178 // should be global. Otherwise, return false. 179 bool 180 get_symbol_version(const char* symbol, std::string* version, 181 bool* is_global) const; 182 183 // Return whether this symbol matches the local: section of some 184 // version. 185 bool 186 symbol_is_local(const char* symbol) const 187 { 188 bool is_global; 189 return (this->get_symbol_version(symbol, NULL, &is_global) 190 && !is_global); 191 } 192 193 // Return the names of versions defined in the version script. 194 std::vector<std::string> 195 get_versions() const; 196 197 // Return the list of dependencies for this version. 198 std::vector<std::string> 199 get_dependencies(const char* version) const; 200 201 // The following functions should only be used by the bison helper 202 // functions. They allocate new structs whose memory belongs to 203 // Version_script_info. The bison functions copy the information 204 // from the version script into these structs. 205 struct Version_dependency_list* 206 allocate_dependency_list(); 207 208 struct Version_expression_list* 209 allocate_expression_list(); 210 211 struct Version_tree* 212 allocate_version_tree(); 213 214 // Build the lookup tables after all data have been read. 215 void 216 build_lookup_tables(); 217 218 // Give an error if there are any unmatched names in the version 219 // script. 220 void 221 check_unmatched_names(const Symbol_table*) const; 222 223 // Print contents to the FILE. This is for debugging. 224 void 225 print(FILE*) const; 226 227 private: 228 void 229 print_expression_list(FILE* f, const Version_expression_list*) const; 230 231 bool 232 get_symbol_version_helper(const char* symbol, 233 bool check_global, 234 std::string* pversion) const; 235 236 // Fast lookup information for a given language. 237 238 // We map from exact match strings to Version_tree's. Historically 239 // version scripts sometimes have the same symbol multiple times, 240 // which is ambiguous. We warn about that case by storing the 241 // second Version_tree we see. 242 struct Version_tree_match 243 { 244 Version_tree_match(const Version_tree* r, bool ig, 245 const Version_expression* e) 246 : real(r), is_global(ig), expression(e), ambiguous(NULL) 247 { } 248 249 // The Version_tree that we return. 250 const Version_tree* real; 251 // True if this is a global match for the REAL member, false if it 252 // is a local match. 253 bool is_global; 254 // Point back to the Version_expression for which we created this 255 // match. 256 const Version_expression* expression; 257 // If not NULL, another Version_tree that defines the symbol. 258 const Version_tree* ambiguous; 259 }; 260 261 // Map from an exact match string to a Version_tree. 262 263 typedef Unordered_map<std::string, Version_tree_match> Exact; 264 265 // Fast lookup information for a glob pattern. 266 struct Glob 267 { 268 Glob() 269 : expression(NULL), version(NULL), is_global(false) 270 { } 271 272 Glob(const Version_expression* e, const Version_tree* v, bool ig) 273 : expression(e), version(v), is_global(ig) 274 { } 275 276 // A pointer to the version expression holding the pattern to 277 // match and the language to use for demangling the symbol before 278 // doing the match. 279 const Version_expression* expression; 280 // The Version_tree we use if this pattern matches. 281 const Version_tree* version; 282 // True if this is a global symbol. 283 bool is_global; 284 }; 285 286 typedef std::vector<Glob> Globs; 287 288 bool 289 unquote(std::string*) const; 290 291 void 292 add_exact_match(const std::string&, const Version_tree*, bool is_global, 293 const Version_expression*, Exact*); 294 295 void 296 build_expression_list_lookup(const Version_expression_list*, 297 const Version_tree*, bool); 298 299 const char* 300 get_name_to_match(const char*, int, 301 Lazy_demangler*, Lazy_demangler*) const; 302 303 // All the version dependencies we allocate. 304 std::vector<Version_dependency_list*> dependency_lists_; 305 // All the version expressions we allocate. 306 std::vector<Version_expression_list*> expression_lists_; 307 // The list of versions. 308 std::vector<Version_tree*> version_trees_; 309 // Exact matches for global symbols, by language. 310 Exact* exact_[LANGUAGE_COUNT]; 311 // A vector of glob patterns mapping to Version_trees. 312 Globs globs_; 313 // The default version to use, if there is one. This is from a 314 // pattern of "*". 315 const Version_tree* default_version_; 316 // True if the default version is global. 317 bool default_is_global_; 318 // Whether this has been finalized. 319 bool is_finalized_; 320}; 321 322// This class manages assignments to symbols. These can appear in 323// three different locations in scripts: outside of a SECTIONS clause, 324// within a SECTIONS clause, and within an output section definition 325// within a SECTIONS clause. This can also appear on the command line 326// via the --defsym command line option. 327 328class Symbol_assignment 329{ 330 public: 331 Symbol_assignment(const char* name, size_t namelen, bool is_defsym, 332 Expression* val, bool provide, bool hidden) 333 : name_(name, namelen), val_(val), is_defsym_(is_defsym), 334 provide_(provide), hidden_(hidden), sym_(NULL) 335 { } 336 337 // Add the symbol to the symbol table. 338 void 339 add_to_table(Symbol_table*); 340 341 // Finalize the symbol value. 342 void 343 finalize(Symbol_table*, const Layout*); 344 345 // Finalize the symbol value when it can refer to the dot symbol. 346 void 347 finalize_with_dot(Symbol_table*, const Layout*, uint64_t dot_value, 348 Output_section* dot_section); 349 350 // Set the symbol value, but only if the value is absolute or relative to 351 // DOT_SECTION. This is used while processing a SECTIONS clause. 352 // We assume that dot is an absolute value here. We do not check assertions. 353 void 354 set_if_absolute(Symbol_table*, const Layout*, bool is_dot_available, 355 uint64_t dot_value, Output_section* dot_section); 356 357 const std::string& 358 name() const 359 { return this->name_; } 360 361 // Print the assignment to the FILE. This is for debugging. 362 void 363 print(FILE*) const; 364 365 private: 366 // Shared by finalize and finalize_with_dot. 367 void 368 finalize_maybe_dot(Symbol_table*, const Layout*, bool is_dot_available, 369 uint64_t dot_value, Output_section* dot_section); 370 371 // Sized version of finalize. 372 template<int size> 373 void 374 sized_finalize(Symbol_table*, const Layout*, bool is_dot_available, 375 uint64_t dot_value, Output_section*); 376 377 // Symbol name. 378 std::string name_; 379 // Expression to assign to symbol. 380 Expression* val_; 381 // True if this symbol is defined by a --defsym, false if it is 382 // defined in a linker script. 383 bool is_defsym_; 384 // Whether the assignment should be provided (only set if there is 385 // an undefined reference to the symbol. 386 bool provide_; 387 // Whether the assignment should be hidden. 388 bool hidden_; 389 // The entry in the symbol table. 390 Symbol* sym_; 391}; 392 393// This class manages assertions in linker scripts. These can appear 394// in all the places where a Symbol_assignment can appear. 395 396class Script_assertion 397{ 398 public: 399 Script_assertion(Expression* check, const char* message, 400 size_t messagelen) 401 : check_(check), message_(message, messagelen) 402 { } 403 404 // Check the assertion. 405 void 406 check(const Symbol_table*, const Layout*); 407 408 // Print the assertion to the FILE. This is for debugging. 409 void 410 print(FILE*) const; 411 412 private: 413 // The expression to check. 414 Expression* check_; 415 // The message to issue if the expression fails. 416 std::string message_; 417}; 418 419// We can read a linker script in two different contexts: when 420// initially parsing the command line, and when we find an input file 421// which is actually a linker script. Also some of the data which can 422// be set by a linker script can also be set via command line options 423// like -e and --defsym. This means that we have a type of data which 424// can be set both during command line option parsing and while 425// reading input files. We store that data in an instance of this 426// object. We will keep pointers to that instance in both the 427// Command_line and Layout objects. 428 429class Script_options 430{ 431 public: 432 Script_options(); 433 434 // Add a symbol to be defined. 435 void 436 add_symbol_assignment(const char* name, size_t length, bool is_defsym, 437 Expression* value, bool provide, bool hidden); 438 439 // Look for an assigned symbol. 440 bool 441 is_pending_assignment(const char* name); 442 443 // Add a reference to a symbol. 444 void 445 add_symbol_reference(const char* name, size_t length); 446 447 // Add an assertion. 448 void 449 add_assertion(Expression* check, const char* message, size_t messagelen); 450 451 // Define a symbol from the command line. 452 bool 453 define_symbol(const char* definition); 454 455 // Create sections required by any linker scripts. 456 void 457 create_script_sections(Layout*); 458 459 // Add all symbol definitions to the symbol table. 460 void 461 add_symbols_to_table(Symbol_table*); 462 463 // Used to iterate over symbols which are referenced in expressions 464 // but not defined. 465 typedef Unordered_set<std::string>::const_iterator referenced_const_iterator; 466 467 referenced_const_iterator 468 referenced_begin() const 469 { return this->symbol_references_.begin(); } 470 471 referenced_const_iterator 472 referenced_end() const 473 { return this->symbol_references_.end(); } 474 475 // Return whether a symbol is referenced but not defined. 476 bool 477 is_referenced(const std::string& name) const 478 { 479 return (this->symbol_references_.find(name) 480 != this->symbol_references_.end()); 481 } 482 483 // Return whether there are any symbols which were referenced but 484 // not defined. 485 bool 486 any_unreferenced() const 487 { return !this->symbol_references_.empty(); } 488 489 // Finalize the symbol values. Also check assertions. 490 void 491 finalize_symbols(Symbol_table*, const Layout*); 492 493 // Version information parsed from a version script. Everything 494 // else has a pointer to this object. 495 Version_script_info* 496 version_script_info() 497 { return &this->version_script_info_; } 498 499 const Version_script_info* 500 version_script_info() const 501 { return &this->version_script_info_; } 502 503 // A SECTIONS clause parsed from a linker script. Everything else 504 // has a pointer to this object. 505 Script_sections* 506 script_sections() 507 { return &this->script_sections_; } 508 509 const Script_sections* 510 script_sections() const 511 { return &this->script_sections_; } 512 513 // Whether we saw a SECTIONS clause. 514 bool 515 saw_sections_clause() const 516 { return this->script_sections_.saw_sections_clause(); } 517 518 // Whether we saw a PHDRS clause. 519 bool 520 saw_phdrs_clause() const 521 { return this->script_sections_.saw_phdrs_clause(); } 522 523 // Set section addresses using a SECTIONS clause. Return the 524 // segment which should hold the file header and segment headers; 525 // this may return NULL, in which case the headers are not in a 526 // loadable segment. 527 Output_segment* 528 set_section_addresses(Symbol_table*, Layout*); 529 530 // Print the script to the FILE. This is for debugging. 531 void 532 print(FILE*) const; 533 534 private: 535 // We keep a list of symbol assignments which occur outside of a 536 // SECTIONS clause. 537 typedef std::vector<Symbol_assignment*> Symbol_assignments; 538 539 // We keep a list of all assertions whcih occur outside of a 540 // SECTIONS clause. 541 typedef std::vector<Script_assertion*> Assertions; 542 543 // The entry address. This will be empty if not set. 544 std::string entry_; 545 // Symbols to set. 546 Symbol_assignments symbol_assignments_; 547 // Symbols defined in an expression, for faster lookup. 548 Unordered_set<std::string> symbol_definitions_; 549 // Symbols referenced in an expression. 550 Unordered_set<std::string> symbol_references_; 551 // Assertions to check. 552 Assertions assertions_; 553 // Version information parsed from a version script. 554 Version_script_info version_script_info_; 555 // Information from any SECTIONS clauses. 556 Script_sections script_sections_; 557}; 558 559// FILE was found as an argument on the command line, but was not 560// recognized as an ELF file. Try to read it as a script. Return 561// true if the file was handled. This has to handle /usr/lib/libc.so 562// on a GNU/Linux system. *USED_NEXT_BLOCKER is set to indicate 563// whether the function took over NEXT_BLOCKER. 564 565bool 566read_input_script(Workqueue*, Symbol_table*, Layout*, Dirsearch*, int, 567 Input_objects*, Mapfile*, Input_group*, 568 const Input_argument*, Input_file*, 569 Task_token* next_blocker, bool* used_next_blocker); 570 571// FILE was found as an argument to --script (-T). 572// Read it as a script, and execute its contents immediately. 573 574bool 575read_commandline_script(const char* filename, Command_line* cmdline); 576 577// FILE was found as an argument to --version-script. Read it as a 578// version script, and store its contents in 579// cmdline->script_options()->version_script_info(). 580 581bool 582read_version_script(const char* filename, Command_line* cmdline); 583 584// FILENAME was found as an argument to --dynamic-list. Read it as a 585// version script (actually, a versym_node from a version script), and 586// store its contents in DYNAMIC_LIST. 587 588bool 589read_dynamic_list(const char* filename, Command_line* cmdline, 590 Script_options* dynamic_list); 591 592} // End namespace gold. 593 594#endif // !defined(GOLD_SCRIPT_H) 595