script.cc revision 1.1
1// script.cc -- handle linker scripts for gold. 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#include "gold.h" 24 25#include <cstdio> 26#include <cstdlib> 27#include <cstring> 28#include <fnmatch.h> 29#include <string> 30#include <vector> 31#include "filenames.h" 32 33#include "elfcpp.h" 34#include "demangle.h" 35#include "dirsearch.h" 36#include "options.h" 37#include "fileread.h" 38#include "workqueue.h" 39#include "readsyms.h" 40#include "parameters.h" 41#include "layout.h" 42#include "symtab.h" 43#include "target-select.h" 44#include "script.h" 45#include "script-c.h" 46#include "incremental.h" 47 48namespace gold 49{ 50 51// A token read from a script file. We don't implement keywords here; 52// all keywords are simply represented as a string. 53 54class Token 55{ 56 public: 57 // Token classification. 58 enum Classification 59 { 60 // Token is invalid. 61 TOKEN_INVALID, 62 // Token indicates end of input. 63 TOKEN_EOF, 64 // Token is a string of characters. 65 TOKEN_STRING, 66 // Token is a quoted string of characters. 67 TOKEN_QUOTED_STRING, 68 // Token is an operator. 69 TOKEN_OPERATOR, 70 // Token is a number (an integer). 71 TOKEN_INTEGER 72 }; 73 74 // We need an empty constructor so that we can put this STL objects. 75 Token() 76 : classification_(TOKEN_INVALID), value_(NULL), value_length_(0), 77 opcode_(0), lineno_(0), charpos_(0) 78 { } 79 80 // A general token with no value. 81 Token(Classification classification, int lineno, int charpos) 82 : classification_(classification), value_(NULL), value_length_(0), 83 opcode_(0), lineno_(lineno), charpos_(charpos) 84 { 85 gold_assert(classification == TOKEN_INVALID 86 || classification == TOKEN_EOF); 87 } 88 89 // A general token with a value. 90 Token(Classification classification, const char* value, size_t length, 91 int lineno, int charpos) 92 : classification_(classification), value_(value), value_length_(length), 93 opcode_(0), lineno_(lineno), charpos_(charpos) 94 { 95 gold_assert(classification != TOKEN_INVALID 96 && classification != TOKEN_EOF); 97 } 98 99 // A token representing an operator. 100 Token(int opcode, int lineno, int charpos) 101 : classification_(TOKEN_OPERATOR), value_(NULL), value_length_(0), 102 opcode_(opcode), lineno_(lineno), charpos_(charpos) 103 { } 104 105 // Return whether the token is invalid. 106 bool 107 is_invalid() const 108 { return this->classification_ == TOKEN_INVALID; } 109 110 // Return whether this is an EOF token. 111 bool 112 is_eof() const 113 { return this->classification_ == TOKEN_EOF; } 114 115 // Return the token classification. 116 Classification 117 classification() const 118 { return this->classification_; } 119 120 // Return the line number at which the token starts. 121 int 122 lineno() const 123 { return this->lineno_; } 124 125 // Return the character position at this the token starts. 126 int 127 charpos() const 128 { return this->charpos_; } 129 130 // Get the value of a token. 131 132 const char* 133 string_value(size_t* length) const 134 { 135 gold_assert(this->classification_ == TOKEN_STRING 136 || this->classification_ == TOKEN_QUOTED_STRING); 137 *length = this->value_length_; 138 return this->value_; 139 } 140 141 int 142 operator_value() const 143 { 144 gold_assert(this->classification_ == TOKEN_OPERATOR); 145 return this->opcode_; 146 } 147 148 uint64_t 149 integer_value() const; 150 151 private: 152 // The token classification. 153 Classification classification_; 154 // The token value, for TOKEN_STRING or TOKEN_QUOTED_STRING or 155 // TOKEN_INTEGER. 156 const char* value_; 157 // The length of the token value. 158 size_t value_length_; 159 // The token value, for TOKEN_OPERATOR. 160 int opcode_; 161 // The line number where this token started (one based). 162 int lineno_; 163 // The character position within the line where this token started 164 // (one based). 165 int charpos_; 166}; 167 168// Return the value of a TOKEN_INTEGER. 169 170uint64_t 171Token::integer_value() const 172{ 173 gold_assert(this->classification_ == TOKEN_INTEGER); 174 175 size_t len = this->value_length_; 176 177 uint64_t multiplier = 1; 178 char last = this->value_[len - 1]; 179 if (last == 'm' || last == 'M') 180 { 181 multiplier = 1024 * 1024; 182 --len; 183 } 184 else if (last == 'k' || last == 'K') 185 { 186 multiplier = 1024; 187 --len; 188 } 189 190 char *end; 191 uint64_t ret = strtoull(this->value_, &end, 0); 192 gold_assert(static_cast<size_t>(end - this->value_) == len); 193 194 return ret * multiplier; 195} 196 197// This class handles lexing a file into a sequence of tokens. 198 199class Lex 200{ 201 public: 202 // We unfortunately have to support different lexing modes, because 203 // when reading different parts of a linker script we need to parse 204 // things differently. 205 enum Mode 206 { 207 // Reading an ordinary linker script. 208 LINKER_SCRIPT, 209 // Reading an expression in a linker script. 210 EXPRESSION, 211 // Reading a version script. 212 VERSION_SCRIPT, 213 // Reading a --dynamic-list file. 214 DYNAMIC_LIST 215 }; 216 217 Lex(const char* input_string, size_t input_length, int parsing_token) 218 : input_string_(input_string), input_length_(input_length), 219 current_(input_string), mode_(LINKER_SCRIPT), 220 first_token_(parsing_token), token_(), 221 lineno_(1), linestart_(input_string) 222 { } 223 224 // Read a file into a string. 225 static void 226 read_file(Input_file*, std::string*); 227 228 // Return the next token. 229 const Token* 230 next_token(); 231 232 // Return the current lexing mode. 233 Lex::Mode 234 mode() const 235 { return this->mode_; } 236 237 // Set the lexing mode. 238 void 239 set_mode(Mode mode) 240 { this->mode_ = mode; } 241 242 private: 243 Lex(const Lex&); 244 Lex& operator=(const Lex&); 245 246 // Make a general token with no value at the current location. 247 Token 248 make_token(Token::Classification c, const char* start) const 249 { return Token(c, this->lineno_, start - this->linestart_ + 1); } 250 251 // Make a general token with a value at the current location. 252 Token 253 make_token(Token::Classification c, const char* v, size_t len, 254 const char* start) 255 const 256 { return Token(c, v, len, this->lineno_, start - this->linestart_ + 1); } 257 258 // Make an operator token at the current location. 259 Token 260 make_token(int opcode, const char* start) const 261 { return Token(opcode, this->lineno_, start - this->linestart_ + 1); } 262 263 // Make an invalid token at the current location. 264 Token 265 make_invalid_token(const char* start) 266 { return this->make_token(Token::TOKEN_INVALID, start); } 267 268 // Make an EOF token at the current location. 269 Token 270 make_eof_token(const char* start) 271 { return this->make_token(Token::TOKEN_EOF, start); } 272 273 // Return whether C can be the first character in a name. C2 is the 274 // next character, since we sometimes need that. 275 inline bool 276 can_start_name(char c, char c2); 277 278 // If C can appear in a name which has already started, return a 279 // pointer to a character later in the token or just past 280 // it. Otherwise, return NULL. 281 inline const char* 282 can_continue_name(const char* c); 283 284 // Return whether C, C2, C3 can start a hex number. 285 inline bool 286 can_start_hex(char c, char c2, char c3); 287 288 // If C can appear in a hex number which has already started, return 289 // a pointer to a character later in the token or just past 290 // it. Otherwise, return NULL. 291 inline const char* 292 can_continue_hex(const char* c); 293 294 // Return whether C can start a non-hex number. 295 static inline bool 296 can_start_number(char c); 297 298 // If C can appear in a decimal number which has already started, 299 // return a pointer to a character later in the token or just past 300 // it. Otherwise, return NULL. 301 inline const char* 302 can_continue_number(const char* c) 303 { return Lex::can_start_number(*c) ? c + 1 : NULL; } 304 305 // If C1 C2 C3 form a valid three character operator, return the 306 // opcode. Otherwise return 0. 307 static inline int 308 three_char_operator(char c1, char c2, char c3); 309 310 // If C1 C2 form a valid two character operator, return the opcode. 311 // Otherwise return 0. 312 static inline int 313 two_char_operator(char c1, char c2); 314 315 // If C1 is a valid one character operator, return the opcode. 316 // Otherwise return 0. 317 static inline int 318 one_char_operator(char c1); 319 320 // Read the next token. 321 Token 322 get_token(const char**); 323 324 // Skip a C style /* */ comment. Return false if the comment did 325 // not end. 326 bool 327 skip_c_comment(const char**); 328 329 // Skip a line # comment. Return false if there was no newline. 330 bool 331 skip_line_comment(const char**); 332 333 // Build a token CLASSIFICATION from all characters that match 334 // CAN_CONTINUE_FN. The token starts at START. Start matching from 335 // MATCH. Set *PP to the character following the token. 336 inline Token 337 gather_token(Token::Classification, 338 const char* (Lex::*can_continue_fn)(const char*), 339 const char* start, const char* match, const char** pp); 340 341 // Build a token from a quoted string. 342 Token 343 gather_quoted_string(const char** pp); 344 345 // The string we are tokenizing. 346 const char* input_string_; 347 // The length of the string. 348 size_t input_length_; 349 // The current offset into the string. 350 const char* current_; 351 // The current lexing mode. 352 Mode mode_; 353 // The code to use for the first token. This is set to 0 after it 354 // is used. 355 int first_token_; 356 // The current token. 357 Token token_; 358 // The current line number. 359 int lineno_; 360 // The start of the current line in the string. 361 const char* linestart_; 362}; 363 364// Read the whole file into memory. We don't expect linker scripts to 365// be large, so we just use a std::string as a buffer. We ignore the 366// data we've already read, so that we read aligned buffers. 367 368void 369Lex::read_file(Input_file* input_file, std::string* contents) 370{ 371 off_t filesize = input_file->file().filesize(); 372 contents->clear(); 373 contents->reserve(filesize); 374 375 off_t off = 0; 376 unsigned char buf[BUFSIZ]; 377 while (off < filesize) 378 { 379 off_t get = BUFSIZ; 380 if (get > filesize - off) 381 get = filesize - off; 382 input_file->file().read(off, get, buf); 383 contents->append(reinterpret_cast<char*>(&buf[0]), get); 384 off += get; 385 } 386} 387 388// Return whether C can be the start of a name, if the next character 389// is C2. A name can being with a letter, underscore, period, or 390// dollar sign. Because a name can be a file name, we also permit 391// forward slash, backslash, and tilde. Tilde is the tricky case 392// here; GNU ld also uses it as a bitwise not operator. It is only 393// recognized as the operator if it is not immediately followed by 394// some character which can appear in a symbol. That is, when we 395// don't know that we are looking at an expression, "~0" is a file 396// name, and "~ 0" is an expression using bitwise not. We are 397// compatible. 398 399inline bool 400Lex::can_start_name(char c, char c2) 401{ 402 switch (c) 403 { 404 case 'A': case 'B': case 'C': case 'D': case 'E': case 'F': 405 case 'G': case 'H': case 'I': case 'J': case 'K': case 'L': 406 case 'M': case 'N': case 'O': case 'Q': case 'P': case 'R': 407 case 'S': case 'T': case 'U': case 'V': case 'W': case 'X': 408 case 'Y': case 'Z': 409 case 'a': case 'b': case 'c': case 'd': case 'e': case 'f': 410 case 'g': case 'h': case 'i': case 'j': case 'k': case 'l': 411 case 'm': case 'n': case 'o': case 'q': case 'p': case 'r': 412 case 's': case 't': case 'u': case 'v': case 'w': case 'x': 413 case 'y': case 'z': 414 case '_': case '.': case '$': 415 return true; 416 417 case '/': case '\\': 418 return this->mode_ == LINKER_SCRIPT; 419 420 case '~': 421 return this->mode_ == LINKER_SCRIPT && can_continue_name(&c2); 422 423 case '*': case '[': 424 return (this->mode_ == VERSION_SCRIPT 425 || this->mode_ == DYNAMIC_LIST 426 || (this->mode_ == LINKER_SCRIPT 427 && can_continue_name(&c2))); 428 429 default: 430 return false; 431 } 432} 433 434// Return whether C can continue a name which has already started. 435// Subsequent characters in a name are the same as the leading 436// characters, plus digits and "=+-:[],?*". So in general the linker 437// script language requires spaces around operators, unless we know 438// that we are parsing an expression. 439 440inline const char* 441Lex::can_continue_name(const char* c) 442{ 443 switch (*c) 444 { 445 case 'A': case 'B': case 'C': case 'D': case 'E': case 'F': 446 case 'G': case 'H': case 'I': case 'J': case 'K': case 'L': 447 case 'M': case 'N': case 'O': case 'Q': case 'P': case 'R': 448 case 'S': case 'T': case 'U': case 'V': case 'W': case 'X': 449 case 'Y': case 'Z': 450 case 'a': case 'b': case 'c': case 'd': case 'e': case 'f': 451 case 'g': case 'h': case 'i': case 'j': case 'k': case 'l': 452 case 'm': case 'n': case 'o': case 'q': case 'p': case 'r': 453 case 's': case 't': case 'u': case 'v': case 'w': case 'x': 454 case 'y': case 'z': 455 case '_': case '.': case '$': 456 case '0': case '1': case '2': case '3': case '4': 457 case '5': case '6': case '7': case '8': case '9': 458 return c + 1; 459 460 // TODO(csilvers): why not allow ~ in names for version-scripts? 461 case '/': case '\\': case '~': 462 case '=': case '+': 463 case ',': 464 if (this->mode_ == LINKER_SCRIPT) 465 return c + 1; 466 return NULL; 467 468 case '[': case ']': case '*': case '?': case '-': 469 if (this->mode_ == LINKER_SCRIPT || this->mode_ == VERSION_SCRIPT 470 || this->mode_ == DYNAMIC_LIST) 471 return c + 1; 472 return NULL; 473 474 // TODO(csilvers): why allow this? ^ is meaningless in version scripts. 475 case '^': 476 if (this->mode_ == VERSION_SCRIPT || this->mode_ == DYNAMIC_LIST) 477 return c + 1; 478 return NULL; 479 480 case ':': 481 if (this->mode_ == LINKER_SCRIPT) 482 return c + 1; 483 else if ((this->mode_ == VERSION_SCRIPT || this->mode_ == DYNAMIC_LIST) 484 && (c[1] == ':')) 485 { 486 // A name can have '::' in it, as that's a c++ namespace 487 // separator. But a single colon is not part of a name. 488 return c + 2; 489 } 490 return NULL; 491 492 default: 493 return NULL; 494 } 495} 496 497// For a number we accept 0x followed by hex digits, or any sequence 498// of digits. The old linker accepts leading '$' for hex, and 499// trailing HXBOD. Those are for MRI compatibility and we don't 500// accept them. 501 502// Return whether C1 C2 C3 can start a hex number. 503 504inline bool 505Lex::can_start_hex(char c1, char c2, char c3) 506{ 507 if (c1 == '0' && (c2 == 'x' || c2 == 'X')) 508 return this->can_continue_hex(&c3); 509 return false; 510} 511 512// Return whether C can appear in a hex number. 513 514inline const char* 515Lex::can_continue_hex(const char* c) 516{ 517 switch (*c) 518 { 519 case '0': case '1': case '2': case '3': case '4': 520 case '5': case '6': case '7': case '8': case '9': 521 case 'A': case 'B': case 'C': case 'D': case 'E': case 'F': 522 case 'a': case 'b': case 'c': case 'd': case 'e': case 'f': 523 return c + 1; 524 525 default: 526 return NULL; 527 } 528} 529 530// Return whether C can start a non-hex number. 531 532inline bool 533Lex::can_start_number(char c) 534{ 535 switch (c) 536 { 537 case '0': case '1': case '2': case '3': case '4': 538 case '5': case '6': case '7': case '8': case '9': 539 return true; 540 541 default: 542 return false; 543 } 544} 545 546// If C1 C2 C3 form a valid three character operator, return the 547// opcode (defined in the yyscript.h file generated from yyscript.y). 548// Otherwise return 0. 549 550inline int 551Lex::three_char_operator(char c1, char c2, char c3) 552{ 553 switch (c1) 554 { 555 case '<': 556 if (c2 == '<' && c3 == '=') 557 return LSHIFTEQ; 558 break; 559 case '>': 560 if (c2 == '>' && c3 == '=') 561 return RSHIFTEQ; 562 break; 563 default: 564 break; 565 } 566 return 0; 567} 568 569// If C1 C2 form a valid two character operator, return the opcode 570// (defined in the yyscript.h file generated from yyscript.y). 571// Otherwise return 0. 572 573inline int 574Lex::two_char_operator(char c1, char c2) 575{ 576 switch (c1) 577 { 578 case '=': 579 if (c2 == '=') 580 return EQ; 581 break; 582 case '!': 583 if (c2 == '=') 584 return NE; 585 break; 586 case '+': 587 if (c2 == '=') 588 return PLUSEQ; 589 break; 590 case '-': 591 if (c2 == '=') 592 return MINUSEQ; 593 break; 594 case '*': 595 if (c2 == '=') 596 return MULTEQ; 597 break; 598 case '/': 599 if (c2 == '=') 600 return DIVEQ; 601 break; 602 case '|': 603 if (c2 == '=') 604 return OREQ; 605 if (c2 == '|') 606 return OROR; 607 break; 608 case '&': 609 if (c2 == '=') 610 return ANDEQ; 611 if (c2 == '&') 612 return ANDAND; 613 break; 614 case '>': 615 if (c2 == '=') 616 return GE; 617 if (c2 == '>') 618 return RSHIFT; 619 break; 620 case '<': 621 if (c2 == '=') 622 return LE; 623 if (c2 == '<') 624 return LSHIFT; 625 break; 626 default: 627 break; 628 } 629 return 0; 630} 631 632// If C1 is a valid operator, return the opcode. Otherwise return 0. 633 634inline int 635Lex::one_char_operator(char c1) 636{ 637 switch (c1) 638 { 639 case '+': 640 case '-': 641 case '*': 642 case '/': 643 case '%': 644 case '!': 645 case '&': 646 case '|': 647 case '^': 648 case '~': 649 case '<': 650 case '>': 651 case '=': 652 case '?': 653 case ',': 654 case '(': 655 case ')': 656 case '{': 657 case '}': 658 case '[': 659 case ']': 660 case ':': 661 case ';': 662 return c1; 663 default: 664 return 0; 665 } 666} 667 668// Skip a C style comment. *PP points to just after the "/*". Return 669// false if the comment did not end. 670 671bool 672Lex::skip_c_comment(const char** pp) 673{ 674 const char* p = *pp; 675 while (p[0] != '*' || p[1] != '/') 676 { 677 if (*p == '\0') 678 { 679 *pp = p; 680 return false; 681 } 682 683 if (*p == '\n') 684 { 685 ++this->lineno_; 686 this->linestart_ = p + 1; 687 } 688 ++p; 689 } 690 691 *pp = p + 2; 692 return true; 693} 694 695// Skip a line # comment. Return false if there was no newline. 696 697bool 698Lex::skip_line_comment(const char** pp) 699{ 700 const char* p = *pp; 701 size_t skip = strcspn(p, "\n"); 702 if (p[skip] == '\0') 703 { 704 *pp = p + skip; 705 return false; 706 } 707 708 p += skip + 1; 709 ++this->lineno_; 710 this->linestart_ = p; 711 *pp = p; 712 713 return true; 714} 715 716// Build a token CLASSIFICATION from all characters that match 717// CAN_CONTINUE_FN. Update *PP. 718 719inline Token 720Lex::gather_token(Token::Classification classification, 721 const char* (Lex::*can_continue_fn)(const char*), 722 const char* start, 723 const char* match, 724 const char** pp) 725{ 726 const char* new_match = NULL; 727 while ((new_match = (this->*can_continue_fn)(match)) != NULL) 728 match = new_match; 729 730 // A special case: integers may be followed by a single M or K, 731 // case-insensitive. 732 if (classification == Token::TOKEN_INTEGER 733 && (*match == 'm' || *match == 'M' || *match == 'k' || *match == 'K')) 734 ++match; 735 736 *pp = match; 737 return this->make_token(classification, start, match - start, start); 738} 739 740// Build a token from a quoted string. 741 742Token 743Lex::gather_quoted_string(const char** pp) 744{ 745 const char* start = *pp; 746 const char* p = start; 747 ++p; 748 size_t skip = strcspn(p, "\"\n"); 749 if (p[skip] != '"') 750 return this->make_invalid_token(start); 751 *pp = p + skip + 1; 752 return this->make_token(Token::TOKEN_QUOTED_STRING, p, skip, start); 753} 754 755// Return the next token at *PP. Update *PP. General guideline: we 756// require linker scripts to be simple ASCII. No unicode linker 757// scripts. In particular we can assume that any '\0' is the end of 758// the input. 759 760Token 761Lex::get_token(const char** pp) 762{ 763 const char* p = *pp; 764 765 while (true) 766 { 767 if (*p == '\0') 768 { 769 *pp = p; 770 return this->make_eof_token(p); 771 } 772 773 // Skip whitespace quickly. 774 while (*p == ' ' || *p == '\t' || *p == '\r') 775 ++p; 776 777 if (*p == '\n') 778 { 779 ++p; 780 ++this->lineno_; 781 this->linestart_ = p; 782 continue; 783 } 784 785 // Skip C style comments. 786 if (p[0] == '/' && p[1] == '*') 787 { 788 int lineno = this->lineno_; 789 int charpos = p - this->linestart_ + 1; 790 791 *pp = p + 2; 792 if (!this->skip_c_comment(pp)) 793 return Token(Token::TOKEN_INVALID, lineno, charpos); 794 p = *pp; 795 796 continue; 797 } 798 799 // Skip line comments. 800 if (*p == '#') 801 { 802 *pp = p + 1; 803 if (!this->skip_line_comment(pp)) 804 return this->make_eof_token(p); 805 p = *pp; 806 continue; 807 } 808 809 // Check for a name. 810 if (this->can_start_name(p[0], p[1])) 811 return this->gather_token(Token::TOKEN_STRING, 812 &Lex::can_continue_name, 813 p, p + 1, pp); 814 815 // We accept any arbitrary name in double quotes, as long as it 816 // does not cross a line boundary. 817 if (*p == '"') 818 { 819 *pp = p; 820 return this->gather_quoted_string(pp); 821 } 822 823 // Check for a number. 824 825 if (this->can_start_hex(p[0], p[1], p[2])) 826 return this->gather_token(Token::TOKEN_INTEGER, 827 &Lex::can_continue_hex, 828 p, p + 3, pp); 829 830 if (Lex::can_start_number(p[0])) 831 return this->gather_token(Token::TOKEN_INTEGER, 832 &Lex::can_continue_number, 833 p, p + 1, pp); 834 835 // Check for operators. 836 837 int opcode = Lex::three_char_operator(p[0], p[1], p[2]); 838 if (opcode != 0) 839 { 840 *pp = p + 3; 841 return this->make_token(opcode, p); 842 } 843 844 opcode = Lex::two_char_operator(p[0], p[1]); 845 if (opcode != 0) 846 { 847 *pp = p + 2; 848 return this->make_token(opcode, p); 849 } 850 851 opcode = Lex::one_char_operator(p[0]); 852 if (opcode != 0) 853 { 854 *pp = p + 1; 855 return this->make_token(opcode, p); 856 } 857 858 return this->make_token(Token::TOKEN_INVALID, p); 859 } 860} 861 862// Return the next token. 863 864const Token* 865Lex::next_token() 866{ 867 // The first token is special. 868 if (this->first_token_ != 0) 869 { 870 this->token_ = Token(this->first_token_, 0, 0); 871 this->first_token_ = 0; 872 return &this->token_; 873 } 874 875 this->token_ = this->get_token(&this->current_); 876 877 // Don't let an early null byte fool us into thinking that we've 878 // reached the end of the file. 879 if (this->token_.is_eof() 880 && (static_cast<size_t>(this->current_ - this->input_string_) 881 < this->input_length_)) 882 this->token_ = this->make_invalid_token(this->current_); 883 884 return &this->token_; 885} 886 887// class Symbol_assignment. 888 889// Add the symbol to the symbol table. This makes sure the symbol is 890// there and defined. The actual value is stored later. We can't 891// determine the actual value at this point, because we can't 892// necessarily evaluate the expression until all ordinary symbols have 893// been finalized. 894 895// The GNU linker lets symbol assignments in the linker script 896// silently override defined symbols in object files. We are 897// compatible. FIXME: Should we issue a warning? 898 899void 900Symbol_assignment::add_to_table(Symbol_table* symtab) 901{ 902 elfcpp::STV vis = this->hidden_ ? elfcpp::STV_HIDDEN : elfcpp::STV_DEFAULT; 903 this->sym_ = symtab->define_as_constant(this->name_.c_str(), 904 NULL, // version 905 (this->is_defsym_ 906 ? Symbol_table::DEFSYM 907 : Symbol_table::SCRIPT), 908 0, // value 909 0, // size 910 elfcpp::STT_NOTYPE, 911 elfcpp::STB_GLOBAL, 912 vis, 913 0, // nonvis 914 this->provide_, 915 true); // force_override 916} 917 918// Finalize a symbol value. 919 920void 921Symbol_assignment::finalize(Symbol_table* symtab, const Layout* layout) 922{ 923 this->finalize_maybe_dot(symtab, layout, false, 0, NULL); 924} 925 926// Finalize a symbol value which can refer to the dot symbol. 927 928void 929Symbol_assignment::finalize_with_dot(Symbol_table* symtab, 930 const Layout* layout, 931 uint64_t dot_value, 932 Output_section* dot_section) 933{ 934 this->finalize_maybe_dot(symtab, layout, true, dot_value, dot_section); 935} 936 937// Finalize a symbol value, internal version. 938 939void 940Symbol_assignment::finalize_maybe_dot(Symbol_table* symtab, 941 const Layout* layout, 942 bool is_dot_available, 943 uint64_t dot_value, 944 Output_section* dot_section) 945{ 946 // If we were only supposed to provide this symbol, the sym_ field 947 // will be NULL if the symbol was not referenced. 948 if (this->sym_ == NULL) 949 { 950 gold_assert(this->provide_); 951 return; 952 } 953 954 if (parameters->target().get_size() == 32) 955 { 956#if defined(HAVE_TARGET_32_LITTLE) || defined(HAVE_TARGET_32_BIG) 957 this->sized_finalize<32>(symtab, layout, is_dot_available, dot_value, 958 dot_section); 959#else 960 gold_unreachable(); 961#endif 962 } 963 else if (parameters->target().get_size() == 64) 964 { 965#if defined(HAVE_TARGET_64_LITTLE) || defined(HAVE_TARGET_64_BIG) 966 this->sized_finalize<64>(symtab, layout, is_dot_available, dot_value, 967 dot_section); 968#else 969 gold_unreachable(); 970#endif 971 } 972 else 973 gold_unreachable(); 974} 975 976template<int size> 977void 978Symbol_assignment::sized_finalize(Symbol_table* symtab, const Layout* layout, 979 bool is_dot_available, uint64_t dot_value, 980 Output_section* dot_section) 981{ 982 Output_section* section; 983 uint64_t final_val = this->val_->eval_maybe_dot(symtab, layout, true, 984 is_dot_available, 985 dot_value, dot_section, 986 §ion, NULL, false); 987 Sized_symbol<size>* ssym = symtab->get_sized_symbol<size>(this->sym_); 988 ssym->set_value(final_val); 989 if (section != NULL) 990 ssym->set_output_section(section); 991} 992 993// Set the symbol value if the expression yields an absolute value or 994// a value relative to DOT_SECTION. 995 996void 997Symbol_assignment::set_if_absolute(Symbol_table* symtab, const Layout* layout, 998 bool is_dot_available, uint64_t dot_value, 999 Output_section* dot_section) 1000{ 1001 if (this->sym_ == NULL) 1002 return; 1003 1004 Output_section* val_section; 1005 uint64_t val = this->val_->eval_maybe_dot(symtab, layout, false, 1006 is_dot_available, dot_value, 1007 dot_section, &val_section, NULL, 1008 false); 1009 if (val_section != NULL && val_section != dot_section) 1010 return; 1011 1012 if (parameters->target().get_size() == 32) 1013 { 1014#if defined(HAVE_TARGET_32_LITTLE) || defined(HAVE_TARGET_32_BIG) 1015 Sized_symbol<32>* ssym = symtab->get_sized_symbol<32>(this->sym_); 1016 ssym->set_value(val); 1017#else 1018 gold_unreachable(); 1019#endif 1020 } 1021 else if (parameters->target().get_size() == 64) 1022 { 1023#if defined(HAVE_TARGET_64_LITTLE) || defined(HAVE_TARGET_64_BIG) 1024 Sized_symbol<64>* ssym = symtab->get_sized_symbol<64>(this->sym_); 1025 ssym->set_value(val); 1026#else 1027 gold_unreachable(); 1028#endif 1029 } 1030 else 1031 gold_unreachable(); 1032 if (val_section != NULL) 1033 this->sym_->set_output_section(val_section); 1034} 1035 1036// Print for debugging. 1037 1038void 1039Symbol_assignment::print(FILE* f) const 1040{ 1041 if (this->provide_ && this->hidden_) 1042 fprintf(f, "PROVIDE_HIDDEN("); 1043 else if (this->provide_) 1044 fprintf(f, "PROVIDE("); 1045 else if (this->hidden_) 1046 gold_unreachable(); 1047 1048 fprintf(f, "%s = ", this->name_.c_str()); 1049 this->val_->print(f); 1050 1051 if (this->provide_ || this->hidden_) 1052 fprintf(f, ")"); 1053 1054 fprintf(f, "\n"); 1055} 1056 1057// Class Script_assertion. 1058 1059// Check the assertion. 1060 1061void 1062Script_assertion::check(const Symbol_table* symtab, const Layout* layout) 1063{ 1064 if (!this->check_->eval(symtab, layout, true)) 1065 gold_error("%s", this->message_.c_str()); 1066} 1067 1068// Print for debugging. 1069 1070void 1071Script_assertion::print(FILE* f) const 1072{ 1073 fprintf(f, "ASSERT("); 1074 this->check_->print(f); 1075 fprintf(f, ", \"%s\")\n", this->message_.c_str()); 1076} 1077 1078// Class Script_options. 1079 1080Script_options::Script_options() 1081 : entry_(), symbol_assignments_(), symbol_definitions_(), 1082 symbol_references_(), version_script_info_(), script_sections_() 1083{ 1084} 1085 1086// Returns true if NAME is on the list of symbol assignments waiting 1087// to be processed. 1088 1089bool 1090Script_options::is_pending_assignment(const char* name) 1091{ 1092 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin(); 1093 p != this->symbol_assignments_.end(); 1094 ++p) 1095 if ((*p)->name() == name) 1096 return true; 1097 return false; 1098} 1099 1100// Add a symbol to be defined. 1101 1102void 1103Script_options::add_symbol_assignment(const char* name, size_t length, 1104 bool is_defsym, Expression* value, 1105 bool provide, bool hidden) 1106{ 1107 if (length != 1 || name[0] != '.') 1108 { 1109 if (this->script_sections_.in_sections_clause()) 1110 { 1111 gold_assert(!is_defsym); 1112 this->script_sections_.add_symbol_assignment(name, length, value, 1113 provide, hidden); 1114 } 1115 else 1116 { 1117 Symbol_assignment* p = new Symbol_assignment(name, length, is_defsym, 1118 value, provide, hidden); 1119 this->symbol_assignments_.push_back(p); 1120 } 1121 1122 if (!provide) 1123 { 1124 std::string n(name, length); 1125 this->symbol_definitions_.insert(n); 1126 this->symbol_references_.erase(n); 1127 } 1128 } 1129 else 1130 { 1131 if (provide || hidden) 1132 gold_error(_("invalid use of PROVIDE for dot symbol")); 1133 1134 // The GNU linker permits assignments to dot outside of SECTIONS 1135 // clauses and treats them as occurring inside, so we don't 1136 // check in_sections_clause here. 1137 this->script_sections_.add_dot_assignment(value); 1138 } 1139} 1140 1141// Add a reference to a symbol. 1142 1143void 1144Script_options::add_symbol_reference(const char* name, size_t length) 1145{ 1146 if (length != 1 || name[0] != '.') 1147 { 1148 std::string n(name, length); 1149 if (this->symbol_definitions_.find(n) == this->symbol_definitions_.end()) 1150 this->symbol_references_.insert(n); 1151 } 1152} 1153 1154// Add an assertion. 1155 1156void 1157Script_options::add_assertion(Expression* check, const char* message, 1158 size_t messagelen) 1159{ 1160 if (this->script_sections_.in_sections_clause()) 1161 this->script_sections_.add_assertion(check, message, messagelen); 1162 else 1163 { 1164 Script_assertion* p = new Script_assertion(check, message, messagelen); 1165 this->assertions_.push_back(p); 1166 } 1167} 1168 1169// Create sections required by any linker scripts. 1170 1171void 1172Script_options::create_script_sections(Layout* layout) 1173{ 1174 if (this->saw_sections_clause()) 1175 this->script_sections_.create_sections(layout); 1176} 1177 1178// Add any symbols we are defining to the symbol table. 1179 1180void 1181Script_options::add_symbols_to_table(Symbol_table* symtab) 1182{ 1183 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin(); 1184 p != this->symbol_assignments_.end(); 1185 ++p) 1186 (*p)->add_to_table(symtab); 1187 this->script_sections_.add_symbols_to_table(symtab); 1188} 1189 1190// Finalize symbol values. Also check assertions. 1191 1192void 1193Script_options::finalize_symbols(Symbol_table* symtab, const Layout* layout) 1194{ 1195 // We finalize the symbols defined in SECTIONS first, because they 1196 // are the ones which may have changed. This way if symbol outside 1197 // SECTIONS are defined in terms of symbols inside SECTIONS, they 1198 // will get the right value. 1199 this->script_sections_.finalize_symbols(symtab, layout); 1200 1201 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin(); 1202 p != this->symbol_assignments_.end(); 1203 ++p) 1204 (*p)->finalize(symtab, layout); 1205 1206 for (Assertions::iterator p = this->assertions_.begin(); 1207 p != this->assertions_.end(); 1208 ++p) 1209 (*p)->check(symtab, layout); 1210} 1211 1212// Set section addresses. We set all the symbols which have absolute 1213// values. Then we let the SECTIONS clause do its thing. This 1214// returns the segment which holds the file header and segment 1215// headers, if any. 1216 1217Output_segment* 1218Script_options::set_section_addresses(Symbol_table* symtab, Layout* layout) 1219{ 1220 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin(); 1221 p != this->symbol_assignments_.end(); 1222 ++p) 1223 (*p)->set_if_absolute(symtab, layout, false, 0, NULL); 1224 1225 return this->script_sections_.set_section_addresses(symtab, layout); 1226} 1227 1228// This class holds data passed through the parser to the lexer and to 1229// the parser support functions. This avoids global variables. We 1230// can't use global variables because we need not be called by a 1231// singleton thread. 1232 1233class Parser_closure 1234{ 1235 public: 1236 Parser_closure(const char* filename, 1237 const Position_dependent_options& posdep_options, 1238 bool parsing_defsym, bool in_group, bool is_in_sysroot, 1239 Command_line* command_line, 1240 Script_options* script_options, 1241 Lex* lex, 1242 bool skip_on_incompatible_target, 1243 Script_info* script_info) 1244 : filename_(filename), posdep_options_(posdep_options), 1245 parsing_defsym_(parsing_defsym), in_group_(in_group), 1246 is_in_sysroot_(is_in_sysroot), 1247 skip_on_incompatible_target_(skip_on_incompatible_target), 1248 found_incompatible_target_(false), 1249 command_line_(command_line), script_options_(script_options), 1250 version_script_info_(script_options->version_script_info()), 1251 lex_(lex), lineno_(0), charpos_(0), lex_mode_stack_(), inputs_(NULL), 1252 script_info_(script_info) 1253 { 1254 // We start out processing C symbols in the default lex mode. 1255 this->language_stack_.push_back(Version_script_info::LANGUAGE_C); 1256 this->lex_mode_stack_.push_back(lex->mode()); 1257 } 1258 1259 // Return the file name. 1260 const char* 1261 filename() const 1262 { return this->filename_; } 1263 1264 // Return the position dependent options. The caller may modify 1265 // this. 1266 Position_dependent_options& 1267 position_dependent_options() 1268 { return this->posdep_options_; } 1269 1270 // Whether we are parsing a --defsym. 1271 bool 1272 parsing_defsym() const 1273 { return this->parsing_defsym_; } 1274 1275 // Return whether this script is being run in a group. 1276 bool 1277 in_group() const 1278 { return this->in_group_; } 1279 1280 // Return whether this script was found using a directory in the 1281 // sysroot. 1282 bool 1283 is_in_sysroot() const 1284 { return this->is_in_sysroot_; } 1285 1286 // Whether to skip to the next file with the same name if we find an 1287 // incompatible target in an OUTPUT_FORMAT statement. 1288 bool 1289 skip_on_incompatible_target() const 1290 { return this->skip_on_incompatible_target_; } 1291 1292 // Stop skipping to the next file on an incompatible target. This 1293 // is called when we make some unrevocable change to the data 1294 // structures. 1295 void 1296 clear_skip_on_incompatible_target() 1297 { this->skip_on_incompatible_target_ = false; } 1298 1299 // Whether we found an incompatible target in an OUTPUT_FORMAT 1300 // statement. 1301 bool 1302 found_incompatible_target() const 1303 { return this->found_incompatible_target_; } 1304 1305 // Note that we found an incompatible target. 1306 void 1307 set_found_incompatible_target() 1308 { this->found_incompatible_target_ = true; } 1309 1310 // Returns the Command_line structure passed in at constructor time. 1311 // This value may be NULL. The caller may modify this, which modifies 1312 // the passed-in Command_line object (not a copy). 1313 Command_line* 1314 command_line() 1315 { return this->command_line_; } 1316 1317 // Return the options which may be set by a script. 1318 Script_options* 1319 script_options() 1320 { return this->script_options_; } 1321 1322 // Return the object in which version script information should be stored. 1323 Version_script_info* 1324 version_script() 1325 { return this->version_script_info_; } 1326 1327 // Return the next token, and advance. 1328 const Token* 1329 next_token() 1330 { 1331 const Token* token = this->lex_->next_token(); 1332 this->lineno_ = token->lineno(); 1333 this->charpos_ = token->charpos(); 1334 return token; 1335 } 1336 1337 // Set a new lexer mode, pushing the current one. 1338 void 1339 push_lex_mode(Lex::Mode mode) 1340 { 1341 this->lex_mode_stack_.push_back(this->lex_->mode()); 1342 this->lex_->set_mode(mode); 1343 } 1344 1345 // Pop the lexer mode. 1346 void 1347 pop_lex_mode() 1348 { 1349 gold_assert(!this->lex_mode_stack_.empty()); 1350 this->lex_->set_mode(this->lex_mode_stack_.back()); 1351 this->lex_mode_stack_.pop_back(); 1352 } 1353 1354 // Return the current lexer mode. 1355 Lex::Mode 1356 lex_mode() const 1357 { return this->lex_mode_stack_.back(); } 1358 1359 // Return the line number of the last token. 1360 int 1361 lineno() const 1362 { return this->lineno_; } 1363 1364 // Return the character position in the line of the last token. 1365 int 1366 charpos() const 1367 { return this->charpos_; } 1368 1369 // Return the list of input files, creating it if necessary. This 1370 // is a space leak--we never free the INPUTS_ pointer. 1371 Input_arguments* 1372 inputs() 1373 { 1374 if (this->inputs_ == NULL) 1375 this->inputs_ = new Input_arguments(); 1376 return this->inputs_; 1377 } 1378 1379 // Return whether we saw any input files. 1380 bool 1381 saw_inputs() const 1382 { return this->inputs_ != NULL && !this->inputs_->empty(); } 1383 1384 // Return the current language being processed in a version script 1385 // (eg, "C++"). The empty string represents unmangled C names. 1386 Version_script_info::Language 1387 get_current_language() const 1388 { return this->language_stack_.back(); } 1389 1390 // Push a language onto the stack when entering an extern block. 1391 void 1392 push_language(Version_script_info::Language lang) 1393 { this->language_stack_.push_back(lang); } 1394 1395 // Pop a language off of the stack when exiting an extern block. 1396 void 1397 pop_language() 1398 { 1399 gold_assert(!this->language_stack_.empty()); 1400 this->language_stack_.pop_back(); 1401 } 1402 1403 // Return a pointer to the incremental info. 1404 Script_info* 1405 script_info() 1406 { return this->script_info_; } 1407 1408 private: 1409 // The name of the file we are reading. 1410 const char* filename_; 1411 // The position dependent options. 1412 Position_dependent_options posdep_options_; 1413 // True if we are parsing a --defsym. 1414 bool parsing_defsym_; 1415 // Whether we are currently in a --start-group/--end-group. 1416 bool in_group_; 1417 // Whether the script was found in a sysrooted directory. 1418 bool is_in_sysroot_; 1419 // If this is true, then if we find an OUTPUT_FORMAT with an 1420 // incompatible target, then we tell the parser to abort so that we 1421 // can search for the next file with the same name. 1422 bool skip_on_incompatible_target_; 1423 // True if we found an OUTPUT_FORMAT with an incompatible target. 1424 bool found_incompatible_target_; 1425 // May be NULL if the user chooses not to pass one in. 1426 Command_line* command_line_; 1427 // Options which may be set from any linker script. 1428 Script_options* script_options_; 1429 // Information parsed from a version script. 1430 Version_script_info* version_script_info_; 1431 // The lexer. 1432 Lex* lex_; 1433 // The line number of the last token returned by next_token. 1434 int lineno_; 1435 // The column number of the last token returned by next_token. 1436 int charpos_; 1437 // A stack of lexer modes. 1438 std::vector<Lex::Mode> lex_mode_stack_; 1439 // A stack of which extern/language block we're inside. Can be C++, 1440 // java, or empty for C. 1441 std::vector<Version_script_info::Language> language_stack_; 1442 // New input files found to add to the link. 1443 Input_arguments* inputs_; 1444 // Pointer to incremental linking info. 1445 Script_info* script_info_; 1446}; 1447 1448// FILE was found as an argument on the command line. Try to read it 1449// as a script. Return true if the file was handled. 1450 1451bool 1452read_input_script(Workqueue* workqueue, Symbol_table* symtab, Layout* layout, 1453 Dirsearch* dirsearch, int dirindex, 1454 Input_objects* input_objects, Mapfile* mapfile, 1455 Input_group* input_group, 1456 const Input_argument* input_argument, 1457 Input_file* input_file, Task_token* next_blocker, 1458 bool* used_next_blocker) 1459{ 1460 *used_next_blocker = false; 1461 1462 std::string input_string; 1463 Lex::read_file(input_file, &input_string); 1464 1465 Lex lex(input_string.c_str(), input_string.length(), PARSING_LINKER_SCRIPT); 1466 1467 Script_info* script_info = NULL; 1468 if (layout->incremental_inputs() != NULL) 1469 { 1470 const std::string& filename = input_file->filename(); 1471 Timespec mtime = input_file->file().get_mtime(); 1472 unsigned int arg_serial = input_argument->file().arg_serial(); 1473 script_info = new Script_info(filename); 1474 layout->incremental_inputs()->report_script(script_info, arg_serial, 1475 mtime); 1476 } 1477 1478 Parser_closure closure(input_file->filename().c_str(), 1479 input_argument->file().options(), 1480 false, 1481 input_group != NULL, 1482 input_file->is_in_sysroot(), 1483 NULL, 1484 layout->script_options(), 1485 &lex, 1486 input_file->will_search_for(), 1487 script_info); 1488 1489 bool old_saw_sections_clause = 1490 layout->script_options()->saw_sections_clause(); 1491 1492 if (yyparse(&closure) != 0) 1493 { 1494 if (closure.found_incompatible_target()) 1495 { 1496 Read_symbols::incompatible_warning(input_argument, input_file); 1497 Read_symbols::requeue(workqueue, input_objects, symtab, layout, 1498 dirsearch, dirindex, mapfile, input_argument, 1499 input_group, next_blocker); 1500 return true; 1501 } 1502 return false; 1503 } 1504 1505 if (!old_saw_sections_clause 1506 && layout->script_options()->saw_sections_clause() 1507 && layout->have_added_input_section()) 1508 gold_error(_("%s: SECTIONS seen after other input files; try -T/--script"), 1509 input_file->filename().c_str()); 1510 1511 if (!closure.saw_inputs()) 1512 return true; 1513 1514 Task_token* this_blocker = NULL; 1515 for (Input_arguments::const_iterator p = closure.inputs()->begin(); 1516 p != closure.inputs()->end(); 1517 ++p) 1518 { 1519 Task_token* nb; 1520 if (p + 1 == closure.inputs()->end()) 1521 nb = next_blocker; 1522 else 1523 { 1524 nb = new Task_token(true); 1525 nb->add_blocker(); 1526 } 1527 workqueue->queue_soon(new Read_symbols(input_objects, symtab, 1528 layout, dirsearch, 0, mapfile, &*p, 1529 input_group, NULL, this_blocker, nb)); 1530 this_blocker = nb; 1531 } 1532 1533 *used_next_blocker = true; 1534 1535 return true; 1536} 1537 1538// Helper function for read_version_script(), read_commandline_script() and 1539// script_include_directive(). Processes the given file in the mode indicated 1540// by first_token and lex_mode. 1541 1542static bool 1543read_script_file(const char* filename, Command_line* cmdline, 1544 Script_options* script_options, 1545 int first_token, Lex::Mode lex_mode) 1546{ 1547 Dirsearch dirsearch; 1548 std::string name = filename; 1549 1550 // If filename is a relative filename, search for it manually using "." + 1551 // cmdline->options()->library_path() -- not dirsearch. 1552 if (!IS_ABSOLUTE_PATH(filename)) 1553 { 1554 const General_options::Dir_list& search_path = 1555 cmdline->options().library_path(); 1556 name = Dirsearch::find_file_in_dir_list(name, search_path, "."); 1557 } 1558 1559 // The file locking code wants to record a Task, but we haven't 1560 // started the workqueue yet. This is only for debugging purposes, 1561 // so we invent a fake value. 1562 const Task* task = reinterpret_cast<const Task*>(-1); 1563 1564 // We don't want this file to be opened in binary mode. 1565 Position_dependent_options posdep = cmdline->position_dependent_options(); 1566 if (posdep.format_enum() == General_options::OBJECT_FORMAT_BINARY) 1567 posdep.set_format_enum(General_options::OBJECT_FORMAT_ELF); 1568 Input_file_argument input_argument(name.c_str(), 1569 Input_file_argument::INPUT_FILE_TYPE_FILE, 1570 "", false, posdep); 1571 Input_file input_file(&input_argument); 1572 int dummy = 0; 1573 if (!input_file.open(dirsearch, task, &dummy)) 1574 return false; 1575 1576 std::string input_string; 1577 Lex::read_file(&input_file, &input_string); 1578 1579 Lex lex(input_string.c_str(), input_string.length(), first_token); 1580 lex.set_mode(lex_mode); 1581 1582 Parser_closure closure(filename, 1583 cmdline->position_dependent_options(), 1584 first_token == Lex::DYNAMIC_LIST, 1585 false, 1586 input_file.is_in_sysroot(), 1587 cmdline, 1588 script_options, 1589 &lex, 1590 false, 1591 NULL); 1592 if (yyparse(&closure) != 0) 1593 { 1594 input_file.file().unlock(task); 1595 return false; 1596 } 1597 1598 input_file.file().unlock(task); 1599 1600 gold_assert(!closure.saw_inputs()); 1601 1602 return true; 1603} 1604 1605// FILENAME was found as an argument to --script (-T). 1606// Read it as a script, and execute its contents immediately. 1607 1608bool 1609read_commandline_script(const char* filename, Command_line* cmdline) 1610{ 1611 return read_script_file(filename, cmdline, &cmdline->script_options(), 1612 PARSING_LINKER_SCRIPT, Lex::LINKER_SCRIPT); 1613} 1614 1615// FILENAME was found as an argument to --version-script. Read it as 1616// a version script, and store its contents in 1617// cmdline->script_options()->version_script_info(). 1618 1619bool 1620read_version_script(const char* filename, Command_line* cmdline) 1621{ 1622 return read_script_file(filename, cmdline, &cmdline->script_options(), 1623 PARSING_VERSION_SCRIPT, Lex::VERSION_SCRIPT); 1624} 1625 1626// FILENAME was found as an argument to --dynamic-list. Read it as a 1627// list of symbols, and store its contents in DYNAMIC_LIST. 1628 1629bool 1630read_dynamic_list(const char* filename, Command_line* cmdline, 1631 Script_options* dynamic_list) 1632{ 1633 return read_script_file(filename, cmdline, dynamic_list, 1634 PARSING_DYNAMIC_LIST, Lex::DYNAMIC_LIST); 1635} 1636 1637// Implement the --defsym option on the command line. Return true if 1638// all is well. 1639 1640bool 1641Script_options::define_symbol(const char* definition) 1642{ 1643 Lex lex(definition, strlen(definition), PARSING_DEFSYM); 1644 lex.set_mode(Lex::EXPRESSION); 1645 1646 // Dummy value. 1647 Position_dependent_options posdep_options; 1648 1649 Parser_closure closure("command line", posdep_options, true, 1650 false, false, NULL, this, &lex, false, NULL); 1651 1652 if (yyparse(&closure) != 0) 1653 return false; 1654 1655 gold_assert(!closure.saw_inputs()); 1656 1657 return true; 1658} 1659 1660// Print the script to F for debugging. 1661 1662void 1663Script_options::print(FILE* f) const 1664{ 1665 fprintf(f, "%s: Dumping linker script\n", program_name); 1666 1667 if (!this->entry_.empty()) 1668 fprintf(f, "ENTRY(%s)\n", this->entry_.c_str()); 1669 1670 for (Symbol_assignments::const_iterator p = 1671 this->symbol_assignments_.begin(); 1672 p != this->symbol_assignments_.end(); 1673 ++p) 1674 (*p)->print(f); 1675 1676 for (Assertions::const_iterator p = this->assertions_.begin(); 1677 p != this->assertions_.end(); 1678 ++p) 1679 (*p)->print(f); 1680 1681 this->script_sections_.print(f); 1682 1683 this->version_script_info_.print(f); 1684} 1685 1686// Manage mapping from keywords to the codes expected by the bison 1687// parser. We construct one global object for each lex mode with 1688// keywords. 1689 1690class Keyword_to_parsecode 1691{ 1692 public: 1693 // The structure which maps keywords to parsecodes. 1694 struct Keyword_parsecode 1695 { 1696 // Keyword. 1697 const char* keyword; 1698 // Corresponding parsecode. 1699 int parsecode; 1700 }; 1701 1702 Keyword_to_parsecode(const Keyword_parsecode* keywords, 1703 int keyword_count) 1704 : keyword_parsecodes_(keywords), keyword_count_(keyword_count) 1705 { } 1706 1707 // Return the parsecode corresponding KEYWORD, or 0 if it is not a 1708 // keyword. 1709 int 1710 keyword_to_parsecode(const char* keyword, size_t len) const; 1711 1712 private: 1713 const Keyword_parsecode* keyword_parsecodes_; 1714 const int keyword_count_; 1715}; 1716 1717// Mapping from keyword string to keyword parsecode. This array must 1718// be kept in sorted order. Parsecodes are looked up using bsearch. 1719// This array must correspond to the list of parsecodes in yyscript.y. 1720 1721static const Keyword_to_parsecode::Keyword_parsecode 1722script_keyword_parsecodes[] = 1723{ 1724 { "ABSOLUTE", ABSOLUTE }, 1725 { "ADDR", ADDR }, 1726 { "ALIGN", ALIGN_K }, 1727 { "ALIGNOF", ALIGNOF }, 1728 { "ASSERT", ASSERT_K }, 1729 { "AS_NEEDED", AS_NEEDED }, 1730 { "AT", AT }, 1731 { "BIND", BIND }, 1732 { "BLOCK", BLOCK }, 1733 { "BYTE", BYTE }, 1734 { "CONSTANT", CONSTANT }, 1735 { "CONSTRUCTORS", CONSTRUCTORS }, 1736 { "COPY", COPY }, 1737 { "CREATE_OBJECT_SYMBOLS", CREATE_OBJECT_SYMBOLS }, 1738 { "DATA_SEGMENT_ALIGN", DATA_SEGMENT_ALIGN }, 1739 { "DATA_SEGMENT_END", DATA_SEGMENT_END }, 1740 { "DATA_SEGMENT_RELRO_END", DATA_SEGMENT_RELRO_END }, 1741 { "DEFINED", DEFINED }, 1742 { "DSECT", DSECT }, 1743 { "ENTRY", ENTRY }, 1744 { "EXCLUDE_FILE", EXCLUDE_FILE }, 1745 { "EXTERN", EXTERN }, 1746 { "FILL", FILL }, 1747 { "FLOAT", FLOAT }, 1748 { "FORCE_COMMON_ALLOCATION", FORCE_COMMON_ALLOCATION }, 1749 { "GROUP", GROUP }, 1750 { "HLL", HLL }, 1751 { "INCLUDE", INCLUDE }, 1752 { "INFO", INFO }, 1753 { "INHIBIT_COMMON_ALLOCATION", INHIBIT_COMMON_ALLOCATION }, 1754 { "INPUT", INPUT }, 1755 { "KEEP", KEEP }, 1756 { "LENGTH", LENGTH }, 1757 { "LOADADDR", LOADADDR }, 1758 { "LONG", LONG }, 1759 { "MAP", MAP }, 1760 { "MAX", MAX_K }, 1761 { "MEMORY", MEMORY }, 1762 { "MIN", MIN_K }, 1763 { "NEXT", NEXT }, 1764 { "NOCROSSREFS", NOCROSSREFS }, 1765 { "NOFLOAT", NOFLOAT }, 1766 { "NOLOAD", NOLOAD }, 1767 { "ONLY_IF_RO", ONLY_IF_RO }, 1768 { "ONLY_IF_RW", ONLY_IF_RW }, 1769 { "OPTION", OPTION }, 1770 { "ORIGIN", ORIGIN }, 1771 { "OUTPUT", OUTPUT }, 1772 { "OUTPUT_ARCH", OUTPUT_ARCH }, 1773 { "OUTPUT_FORMAT", OUTPUT_FORMAT }, 1774 { "OVERLAY", OVERLAY }, 1775 { "PHDRS", PHDRS }, 1776 { "PROVIDE", PROVIDE }, 1777 { "PROVIDE_HIDDEN", PROVIDE_HIDDEN }, 1778 { "QUAD", QUAD }, 1779 { "SEARCH_DIR", SEARCH_DIR }, 1780 { "SECTIONS", SECTIONS }, 1781 { "SEGMENT_START", SEGMENT_START }, 1782 { "SHORT", SHORT }, 1783 { "SIZEOF", SIZEOF }, 1784 { "SIZEOF_HEADERS", SIZEOF_HEADERS }, 1785 { "SORT", SORT_BY_NAME }, 1786 { "SORT_BY_ALIGNMENT", SORT_BY_ALIGNMENT }, 1787 { "SORT_BY_NAME", SORT_BY_NAME }, 1788 { "SPECIAL", SPECIAL }, 1789 { "SQUAD", SQUAD }, 1790 { "STARTUP", STARTUP }, 1791 { "SUBALIGN", SUBALIGN }, 1792 { "SYSLIB", SYSLIB }, 1793 { "TARGET", TARGET_K }, 1794 { "TRUNCATE", TRUNCATE }, 1795 { "VERSION", VERSIONK }, 1796 { "global", GLOBAL }, 1797 { "l", LENGTH }, 1798 { "len", LENGTH }, 1799 { "local", LOCAL }, 1800 { "o", ORIGIN }, 1801 { "org", ORIGIN }, 1802 { "sizeof_headers", SIZEOF_HEADERS }, 1803}; 1804 1805static const Keyword_to_parsecode 1806script_keywords(&script_keyword_parsecodes[0], 1807 (sizeof(script_keyword_parsecodes) 1808 / sizeof(script_keyword_parsecodes[0]))); 1809 1810static const Keyword_to_parsecode::Keyword_parsecode 1811version_script_keyword_parsecodes[] = 1812{ 1813 { "extern", EXTERN }, 1814 { "global", GLOBAL }, 1815 { "local", LOCAL }, 1816}; 1817 1818static const Keyword_to_parsecode 1819version_script_keywords(&version_script_keyword_parsecodes[0], 1820 (sizeof(version_script_keyword_parsecodes) 1821 / sizeof(version_script_keyword_parsecodes[0]))); 1822 1823static const Keyword_to_parsecode::Keyword_parsecode 1824dynamic_list_keyword_parsecodes[] = 1825{ 1826 { "extern", EXTERN }, 1827}; 1828 1829static const Keyword_to_parsecode 1830dynamic_list_keywords(&dynamic_list_keyword_parsecodes[0], 1831 (sizeof(dynamic_list_keyword_parsecodes) 1832 / sizeof(dynamic_list_keyword_parsecodes[0]))); 1833 1834 1835 1836// Comparison function passed to bsearch. 1837 1838extern "C" 1839{ 1840 1841struct Ktt_key 1842{ 1843 const char* str; 1844 size_t len; 1845}; 1846 1847static int 1848ktt_compare(const void* keyv, const void* kttv) 1849{ 1850 const Ktt_key* key = static_cast<const Ktt_key*>(keyv); 1851 const Keyword_to_parsecode::Keyword_parsecode* ktt = 1852 static_cast<const Keyword_to_parsecode::Keyword_parsecode*>(kttv); 1853 int i = strncmp(key->str, ktt->keyword, key->len); 1854 if (i != 0) 1855 return i; 1856 if (ktt->keyword[key->len] != '\0') 1857 return -1; 1858 return 0; 1859} 1860 1861} // End extern "C". 1862 1863int 1864Keyword_to_parsecode::keyword_to_parsecode(const char* keyword, 1865 size_t len) const 1866{ 1867 Ktt_key key; 1868 key.str = keyword; 1869 key.len = len; 1870 void* kttv = bsearch(&key, 1871 this->keyword_parsecodes_, 1872 this->keyword_count_, 1873 sizeof(this->keyword_parsecodes_[0]), 1874 ktt_compare); 1875 if (kttv == NULL) 1876 return 0; 1877 Keyword_parsecode* ktt = static_cast<Keyword_parsecode*>(kttv); 1878 return ktt->parsecode; 1879} 1880 1881// The following structs are used within the VersionInfo class as well 1882// as in the bison helper functions. They store the information 1883// parsed from the version script. 1884 1885// A single version expression. 1886// For example, pattern="std::map*" and language="C++". 1887struct Version_expression 1888{ 1889 Version_expression(const std::string& a_pattern, 1890 Version_script_info::Language a_language, 1891 bool a_exact_match) 1892 : pattern(a_pattern), language(a_language), exact_match(a_exact_match), 1893 was_matched_by_symbol(false) 1894 { } 1895 1896 std::string pattern; 1897 Version_script_info::Language language; 1898 // If false, we use glob() to match pattern. If true, we use strcmp(). 1899 bool exact_match; 1900 // True if --no-undefined-version is in effect and we found this 1901 // version in get_symbol_version. We use mutable because this 1902 // struct is generally not modifiable after it has been created. 1903 mutable bool was_matched_by_symbol; 1904}; 1905 1906// A list of expressions. 1907struct Version_expression_list 1908{ 1909 std::vector<struct Version_expression> expressions; 1910}; 1911 1912// A list of which versions upon which another version depends. 1913// Strings should be from the Stringpool. 1914struct Version_dependency_list 1915{ 1916 std::vector<std::string> dependencies; 1917}; 1918 1919// The total definition of a version. It includes the tag for the 1920// version, its global and local expressions, and any dependencies. 1921struct Version_tree 1922{ 1923 Version_tree() 1924 : tag(), global(NULL), local(NULL), dependencies(NULL) 1925 { } 1926 1927 std::string tag; 1928 const struct Version_expression_list* global; 1929 const struct Version_expression_list* local; 1930 const struct Version_dependency_list* dependencies; 1931}; 1932 1933// Helper class that calls cplus_demangle when needed and takes care of freeing 1934// the result. 1935 1936class Lazy_demangler 1937{ 1938 public: 1939 Lazy_demangler(const char* symbol, int options) 1940 : symbol_(symbol), options_(options), demangled_(NULL), did_demangle_(false) 1941 { } 1942 1943 ~Lazy_demangler() 1944 { free(this->demangled_); } 1945 1946 // Return the demangled name. The actual demangling happens on the first call, 1947 // and the result is later cached. 1948 inline char* 1949 get(); 1950 1951 private: 1952 // The symbol to demangle. 1953 const char* symbol_; 1954 // Option flags to pass to cplus_demagle. 1955 const int options_; 1956 // The cached demangled value, or NULL if demangling didn't happen yet or 1957 // failed. 1958 char* demangled_; 1959 // Whether we already called cplus_demangle 1960 bool did_demangle_; 1961}; 1962 1963// Return the demangled name. The actual demangling happens on the first call, 1964// and the result is later cached. Returns NULL if the symbol cannot be 1965// demangled. 1966 1967inline char* 1968Lazy_demangler::get() 1969{ 1970 if (!this->did_demangle_) 1971 { 1972 this->demangled_ = cplus_demangle(this->symbol_, this->options_); 1973 this->did_demangle_ = true; 1974 } 1975 return this->demangled_; 1976} 1977 1978// Class Version_script_info. 1979 1980Version_script_info::Version_script_info() 1981 : dependency_lists_(), expression_lists_(), version_trees_(), globs_(), 1982 default_version_(NULL), default_is_global_(false), is_finalized_(false) 1983{ 1984 for (int i = 0; i < LANGUAGE_COUNT; ++i) 1985 this->exact_[i] = NULL; 1986} 1987 1988Version_script_info::~Version_script_info() 1989{ 1990} 1991 1992// Forget all the known version script information. 1993 1994void 1995Version_script_info::clear() 1996{ 1997 for (size_t k = 0; k < this->dependency_lists_.size(); ++k) 1998 delete this->dependency_lists_[k]; 1999 this->dependency_lists_.clear(); 2000 for (size_t k = 0; k < this->version_trees_.size(); ++k) 2001 delete this->version_trees_[k]; 2002 this->version_trees_.clear(); 2003 for (size_t k = 0; k < this->expression_lists_.size(); ++k) 2004 delete this->expression_lists_[k]; 2005 this->expression_lists_.clear(); 2006} 2007 2008// Finalize the version script information. 2009 2010void 2011Version_script_info::finalize() 2012{ 2013 if (!this->is_finalized_) 2014 { 2015 this->build_lookup_tables(); 2016 this->is_finalized_ = true; 2017 } 2018} 2019 2020// Return all the versions. 2021 2022std::vector<std::string> 2023Version_script_info::get_versions() const 2024{ 2025 std::vector<std::string> ret; 2026 for (size_t j = 0; j < this->version_trees_.size(); ++j) 2027 if (!this->version_trees_[j]->tag.empty()) 2028 ret.push_back(this->version_trees_[j]->tag); 2029 return ret; 2030} 2031 2032// Return the dependencies of VERSION. 2033 2034std::vector<std::string> 2035Version_script_info::get_dependencies(const char* version) const 2036{ 2037 std::vector<std::string> ret; 2038 for (size_t j = 0; j < this->version_trees_.size(); ++j) 2039 if (this->version_trees_[j]->tag == version) 2040 { 2041 const struct Version_dependency_list* deps = 2042 this->version_trees_[j]->dependencies; 2043 if (deps != NULL) 2044 for (size_t k = 0; k < deps->dependencies.size(); ++k) 2045 ret.push_back(deps->dependencies[k]); 2046 return ret; 2047 } 2048 return ret; 2049} 2050 2051// A version script essentially maps a symbol name to a version tag 2052// and an indication of whether symbol is global or local within that 2053// version tag. Each symbol maps to at most one version tag. 2054// Unfortunately, in practice, version scripts are ambiguous, and list 2055// symbols multiple times. Thus, we have to document the matching 2056// process. 2057 2058// This is a description of what the GNU linker does as of 2010-01-11. 2059// It walks through the version tags in the order in which they appear 2060// in the version script. For each tag, it first walks through the 2061// global patterns for that tag, then the local patterns. When 2062// looking at a single pattern, it first applies any language specific 2063// demangling as specified for the pattern, and then matches the 2064// resulting symbol name to the pattern. If it finds an exact match 2065// for a literal pattern (a pattern enclosed in quotes or with no 2066// wildcard characters), then that is the match that it uses. If 2067// finds a match with a wildcard pattern, then it saves it and 2068// continues searching. Wildcard patterns that are exactly "*" are 2069// saved separately. 2070 2071// If no exact match with a literal pattern is ever found, then if a 2072// wildcard match with a global pattern was found it is used, 2073// otherwise if a wildcard match with a local pattern was found it is 2074// used. 2075 2076// This is the result: 2077// * If there is an exact match, then we use the first tag in the 2078// version script where it matches. 2079// + If the exact match in that tag is global, it is used. 2080// + Otherwise the exact match in that tag is local, and is used. 2081// * Otherwise, if there is any match with a global wildcard pattern: 2082// + If there is any match with a wildcard pattern which is not 2083// "*", then we use the tag in which the *last* such pattern 2084// appears. 2085// + Otherwise, we matched "*". If there is no match with a local 2086// wildcard pattern which is not "*", then we use the *last* 2087// match with a global "*". Otherwise, continue. 2088// * Otherwise, if there is any match with a local wildcard pattern: 2089// + If there is any match with a wildcard pattern which is not 2090// "*", then we use the tag in which the *last* such pattern 2091// appears. 2092// + Otherwise, we matched "*", and we use the tag in which the 2093// *last* such match occurred. 2094 2095// There is an additional wrinkle. When the GNU linker finds a symbol 2096// with a version defined in an object file due to a .symver 2097// directive, it looks up that symbol name in that version tag. If it 2098// finds it, it matches the symbol name against the patterns for that 2099// version. If there is no match with a global pattern, but there is 2100// a match with a local pattern, then the GNU linker marks the symbol 2101// as local. 2102 2103// We want gold to be generally compatible, but we also want gold to 2104// be fast. These are the rules that gold implements: 2105// * If there is an exact match for the mangled name, we use it. 2106// + If there is more than one exact match, we give a warning, and 2107// we use the first tag in the script which matches. 2108// + If a symbol has an exact match as both global and local for 2109// the same version tag, we give an error. 2110// * Otherwise, we look for an extern C++ or an extern Java exact 2111// match. If we find an exact match, we use it. 2112// + If there is more than one exact match, we give a warning, and 2113// we use the first tag in the script which matches. 2114// + If a symbol has an exact match as both global and local for 2115// the same version tag, we give an error. 2116// * Otherwise, we look through the wildcard patterns, ignoring "*" 2117// patterns. We look through the version tags in reverse order. 2118// For each version tag, we look through the global patterns and 2119// then the local patterns. We use the first match we find (i.e., 2120// the last matching version tag in the file). 2121// * Otherwise, we use the "*" pattern if there is one. We give an 2122// error if there are multiple "*" patterns. 2123 2124// At least for now, gold does not look up the version tag for a 2125// symbol version found in an object file to see if it should be 2126// forced local. There are other ways to force a symbol to be local, 2127// and I don't understand why this one is useful. 2128 2129// Build a set of fast lookup tables for a version script. 2130 2131void 2132Version_script_info::build_lookup_tables() 2133{ 2134 size_t size = this->version_trees_.size(); 2135 for (size_t j = 0; j < size; ++j) 2136 { 2137 const Version_tree* v = this->version_trees_[j]; 2138 this->build_expression_list_lookup(v->local, v, false); 2139 this->build_expression_list_lookup(v->global, v, true); 2140 } 2141} 2142 2143// If a pattern has backlashes but no unquoted wildcard characters, 2144// then we apply backslash unquoting and look for an exact match. 2145// Otherwise we treat it as a wildcard pattern. This function returns 2146// true for a wildcard pattern. Otherwise, it does backslash 2147// unquoting on *PATTERN and returns false. If this returns true, 2148// *PATTERN may have been partially unquoted. 2149 2150bool 2151Version_script_info::unquote(std::string* pattern) const 2152{ 2153 bool saw_backslash = false; 2154 size_t len = pattern->length(); 2155 size_t j = 0; 2156 for (size_t i = 0; i < len; ++i) 2157 { 2158 if (saw_backslash) 2159 saw_backslash = false; 2160 else 2161 { 2162 switch ((*pattern)[i]) 2163 { 2164 case '?': case '[': case '*': 2165 return true; 2166 case '\\': 2167 saw_backslash = true; 2168 continue; 2169 default: 2170 break; 2171 } 2172 } 2173 2174 if (i != j) 2175 (*pattern)[j] = (*pattern)[i]; 2176 ++j; 2177 } 2178 return false; 2179} 2180 2181// Add an exact match for MATCH to *PE. The result of the match is 2182// V/IS_GLOBAL. 2183 2184void 2185Version_script_info::add_exact_match(const std::string& match, 2186 const Version_tree* v, bool is_global, 2187 const Version_expression* ve, 2188 Exact* pe) 2189{ 2190 std::pair<Exact::iterator, bool> ins = 2191 pe->insert(std::make_pair(match, Version_tree_match(v, is_global, ve))); 2192 if (ins.second) 2193 { 2194 // This is the first time we have seen this match. 2195 return; 2196 } 2197 2198 Version_tree_match& vtm(ins.first->second); 2199 if (vtm.real->tag != v->tag) 2200 { 2201 // This is an ambiguous match. We still return the 2202 // first version that we found in the script, but we 2203 // record the new version to issue a warning if we 2204 // wind up looking up this symbol. 2205 if (vtm.ambiguous == NULL) 2206 vtm.ambiguous = v; 2207 } 2208 else if (is_global != vtm.is_global) 2209 { 2210 // We have a match for both the global and local entries for a 2211 // version tag. That's got to be wrong. 2212 gold_error(_("'%s' appears as both a global and a local symbol " 2213 "for version '%s' in script"), 2214 match.c_str(), v->tag.c_str()); 2215 } 2216} 2217 2218// Build fast lookup information for EXPLIST and store it in LOOKUP. 2219// All matches go to V, and IS_GLOBAL is true if they are global 2220// matches. 2221 2222void 2223Version_script_info::build_expression_list_lookup( 2224 const Version_expression_list* explist, 2225 const Version_tree* v, 2226 bool is_global) 2227{ 2228 if (explist == NULL) 2229 return; 2230 size_t size = explist->expressions.size(); 2231 for (size_t i = 0; i < size; ++i) 2232 { 2233 const Version_expression& exp(explist->expressions[i]); 2234 2235 if (exp.pattern.length() == 1 && exp.pattern[0] == '*') 2236 { 2237 if (this->default_version_ != NULL 2238 && this->default_version_->tag != v->tag) 2239 gold_warning(_("wildcard match appears in both version '%s' " 2240 "and '%s' in script"), 2241 this->default_version_->tag.c_str(), v->tag.c_str()); 2242 else if (this->default_version_ != NULL 2243 && this->default_is_global_ != is_global) 2244 gold_error(_("wildcard match appears as both global and local " 2245 "in version '%s' in script"), 2246 v->tag.c_str()); 2247 this->default_version_ = v; 2248 this->default_is_global_ = is_global; 2249 continue; 2250 } 2251 2252 std::string pattern = exp.pattern; 2253 if (!exp.exact_match) 2254 { 2255 if (this->unquote(&pattern)) 2256 { 2257 this->globs_.push_back(Glob(&exp, v, is_global)); 2258 continue; 2259 } 2260 } 2261 2262 if (this->exact_[exp.language] == NULL) 2263 this->exact_[exp.language] = new Exact(); 2264 this->add_exact_match(pattern, v, is_global, &exp, 2265 this->exact_[exp.language]); 2266 } 2267} 2268 2269// Return the name to match given a name, a language code, and two 2270// lazy demanglers. 2271 2272const char* 2273Version_script_info::get_name_to_match(const char* name, 2274 int language, 2275 Lazy_demangler* cpp_demangler, 2276 Lazy_demangler* java_demangler) const 2277{ 2278 switch (language) 2279 { 2280 case LANGUAGE_C: 2281 return name; 2282 case LANGUAGE_CXX: 2283 return cpp_demangler->get(); 2284 case LANGUAGE_JAVA: 2285 return java_demangler->get(); 2286 default: 2287 gold_unreachable(); 2288 } 2289} 2290 2291// Look up SYMBOL_NAME in the list of versions. Return true if the 2292// symbol is found, false if not. If the symbol is found, then if 2293// PVERSION is not NULL, set *PVERSION to the version tag, and if 2294// P_IS_GLOBAL is not NULL, set *P_IS_GLOBAL according to whether the 2295// symbol is global or not. 2296 2297bool 2298Version_script_info::get_symbol_version(const char* symbol_name, 2299 std::string* pversion, 2300 bool* p_is_global) const 2301{ 2302 Lazy_demangler cpp_demangled_name(symbol_name, DMGL_ANSI | DMGL_PARAMS); 2303 Lazy_demangler java_demangled_name(symbol_name, 2304 DMGL_ANSI | DMGL_PARAMS | DMGL_JAVA); 2305 2306 gold_assert(this->is_finalized_); 2307 for (int i = 0; i < LANGUAGE_COUNT; ++i) 2308 { 2309 Exact* exact = this->exact_[i]; 2310 if (exact == NULL) 2311 continue; 2312 2313 const char* name_to_match = this->get_name_to_match(symbol_name, i, 2314 &cpp_demangled_name, 2315 &java_demangled_name); 2316 if (name_to_match == NULL) 2317 { 2318 // If the name can not be demangled, the GNU linker goes 2319 // ahead and tries to match it anyhow. That does not 2320 // make sense to me and I have not implemented it. 2321 continue; 2322 } 2323 2324 Exact::const_iterator pe = exact->find(name_to_match); 2325 if (pe != exact->end()) 2326 { 2327 const Version_tree_match& vtm(pe->second); 2328 if (vtm.ambiguous != NULL) 2329 gold_warning(_("using '%s' as version for '%s' which is also " 2330 "named in version '%s' in script"), 2331 vtm.real->tag.c_str(), name_to_match, 2332 vtm.ambiguous->tag.c_str()); 2333 2334 if (pversion != NULL) 2335 *pversion = vtm.real->tag; 2336 if (p_is_global != NULL) 2337 *p_is_global = vtm.is_global; 2338 2339 // If we are using --no-undefined-version, and this is a 2340 // global symbol, we have to record that we have found this 2341 // symbol, so that we don't warn about it. We have to do 2342 // this now, because otherwise we have no way to get from a 2343 // non-C language back to the demangled name that we 2344 // matched. 2345 if (p_is_global != NULL && vtm.is_global) 2346 vtm.expression->was_matched_by_symbol = true; 2347 2348 return true; 2349 } 2350 } 2351 2352 // Look through the glob patterns in reverse order. 2353 2354 for (Globs::const_reverse_iterator p = this->globs_.rbegin(); 2355 p != this->globs_.rend(); 2356 ++p) 2357 { 2358 int language = p->expression->language; 2359 const char* name_to_match = this->get_name_to_match(symbol_name, 2360 language, 2361 &cpp_demangled_name, 2362 &java_demangled_name); 2363 if (name_to_match == NULL) 2364 continue; 2365 2366 if (fnmatch(p->expression->pattern.c_str(), name_to_match, 2367 FNM_NOESCAPE) == 0) 2368 { 2369 if (pversion != NULL) 2370 *pversion = p->version->tag; 2371 if (p_is_global != NULL) 2372 *p_is_global = p->is_global; 2373 return true; 2374 } 2375 } 2376 2377 // Finally, there may be a wildcard. 2378 if (this->default_version_ != NULL) 2379 { 2380 if (pversion != NULL) 2381 *pversion = this->default_version_->tag; 2382 if (p_is_global != NULL) 2383 *p_is_global = this->default_is_global_; 2384 return true; 2385 } 2386 2387 return false; 2388} 2389 2390// Give an error if any exact symbol names (not wildcards) appear in a 2391// version script, but there is no such symbol. 2392 2393void 2394Version_script_info::check_unmatched_names(const Symbol_table* symtab) const 2395{ 2396 for (size_t i = 0; i < this->version_trees_.size(); ++i) 2397 { 2398 const Version_tree* vt = this->version_trees_[i]; 2399 if (vt->global == NULL) 2400 continue; 2401 for (size_t j = 0; j < vt->global->expressions.size(); ++j) 2402 { 2403 const Version_expression& expression(vt->global->expressions[j]); 2404 2405 // Ignore cases where we used the version because we saw a 2406 // symbol that we looked up. Note that 2407 // WAS_MATCHED_BY_SYMBOL will be true even if the symbol was 2408 // not a definition. That's OK as in that case we most 2409 // likely gave an undefined symbol error anyhow. 2410 if (expression.was_matched_by_symbol) 2411 continue; 2412 2413 // Just ignore names which are in languages other than C. 2414 // We have no way to look them up in the symbol table. 2415 if (expression.language != LANGUAGE_C) 2416 continue; 2417 2418 // Remove backslash quoting, and ignore wildcard patterns. 2419 std::string pattern = expression.pattern; 2420 if (!expression.exact_match) 2421 { 2422 if (this->unquote(&pattern)) 2423 continue; 2424 } 2425 2426 if (symtab->lookup(pattern.c_str(), vt->tag.c_str()) == NULL) 2427 gold_error(_("version script assignment of %s to symbol %s " 2428 "failed: symbol not defined"), 2429 vt->tag.c_str(), pattern.c_str()); 2430 } 2431 } 2432} 2433 2434struct Version_dependency_list* 2435Version_script_info::allocate_dependency_list() 2436{ 2437 dependency_lists_.push_back(new Version_dependency_list); 2438 return dependency_lists_.back(); 2439} 2440 2441struct Version_expression_list* 2442Version_script_info::allocate_expression_list() 2443{ 2444 expression_lists_.push_back(new Version_expression_list); 2445 return expression_lists_.back(); 2446} 2447 2448struct Version_tree* 2449Version_script_info::allocate_version_tree() 2450{ 2451 version_trees_.push_back(new Version_tree); 2452 return version_trees_.back(); 2453} 2454 2455// Print for debugging. 2456 2457void 2458Version_script_info::print(FILE* f) const 2459{ 2460 if (this->empty()) 2461 return; 2462 2463 fprintf(f, "VERSION {"); 2464 2465 for (size_t i = 0; i < this->version_trees_.size(); ++i) 2466 { 2467 const Version_tree* vt = this->version_trees_[i]; 2468 2469 if (vt->tag.empty()) 2470 fprintf(f, " {\n"); 2471 else 2472 fprintf(f, " %s {\n", vt->tag.c_str()); 2473 2474 if (vt->global != NULL) 2475 { 2476 fprintf(f, " global :\n"); 2477 this->print_expression_list(f, vt->global); 2478 } 2479 2480 if (vt->local != NULL) 2481 { 2482 fprintf(f, " local :\n"); 2483 this->print_expression_list(f, vt->local); 2484 } 2485 2486 fprintf(f, " }"); 2487 if (vt->dependencies != NULL) 2488 { 2489 const Version_dependency_list* deps = vt->dependencies; 2490 for (size_t j = 0; j < deps->dependencies.size(); ++j) 2491 { 2492 if (j < deps->dependencies.size() - 1) 2493 fprintf(f, "\n"); 2494 fprintf(f, " %s", deps->dependencies[j].c_str()); 2495 } 2496 } 2497 fprintf(f, ";\n"); 2498 } 2499 2500 fprintf(f, "}\n"); 2501} 2502 2503void 2504Version_script_info::print_expression_list( 2505 FILE* f, 2506 const Version_expression_list* vel) const 2507{ 2508 Version_script_info::Language current_language = LANGUAGE_C; 2509 for (size_t i = 0; i < vel->expressions.size(); ++i) 2510 { 2511 const Version_expression& ve(vel->expressions[i]); 2512 2513 if (ve.language != current_language) 2514 { 2515 if (current_language != LANGUAGE_C) 2516 fprintf(f, " }\n"); 2517 switch (ve.language) 2518 { 2519 case LANGUAGE_C: 2520 break; 2521 case LANGUAGE_CXX: 2522 fprintf(f, " extern \"C++\" {\n"); 2523 break; 2524 case LANGUAGE_JAVA: 2525 fprintf(f, " extern \"Java\" {\n"); 2526 break; 2527 default: 2528 gold_unreachable(); 2529 } 2530 current_language = ve.language; 2531 } 2532 2533 fprintf(f, " "); 2534 if (current_language != LANGUAGE_C) 2535 fprintf(f, " "); 2536 2537 if (ve.exact_match) 2538 fprintf(f, "\""); 2539 fprintf(f, "%s", ve.pattern.c_str()); 2540 if (ve.exact_match) 2541 fprintf(f, "\""); 2542 2543 fprintf(f, "\n"); 2544 } 2545 2546 if (current_language != LANGUAGE_C) 2547 fprintf(f, " }\n"); 2548} 2549 2550} // End namespace gold. 2551 2552// The remaining functions are extern "C", so it's clearer to not put 2553// them in namespace gold. 2554 2555using namespace gold; 2556 2557// This function is called by the bison parser to return the next 2558// token. 2559 2560extern "C" int 2561yylex(YYSTYPE* lvalp, void* closurev) 2562{ 2563 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2564 const Token* token = closure->next_token(); 2565 switch (token->classification()) 2566 { 2567 default: 2568 gold_unreachable(); 2569 2570 case Token::TOKEN_INVALID: 2571 yyerror(closurev, "invalid character"); 2572 return 0; 2573 2574 case Token::TOKEN_EOF: 2575 return 0; 2576 2577 case Token::TOKEN_STRING: 2578 { 2579 // This is either a keyword or a STRING. 2580 size_t len; 2581 const char* str = token->string_value(&len); 2582 int parsecode = 0; 2583 switch (closure->lex_mode()) 2584 { 2585 case Lex::LINKER_SCRIPT: 2586 parsecode = script_keywords.keyword_to_parsecode(str, len); 2587 break; 2588 case Lex::VERSION_SCRIPT: 2589 parsecode = version_script_keywords.keyword_to_parsecode(str, len); 2590 break; 2591 case Lex::DYNAMIC_LIST: 2592 parsecode = dynamic_list_keywords.keyword_to_parsecode(str, len); 2593 break; 2594 default: 2595 break; 2596 } 2597 if (parsecode != 0) 2598 return parsecode; 2599 lvalp->string.value = str; 2600 lvalp->string.length = len; 2601 return STRING; 2602 } 2603 2604 case Token::TOKEN_QUOTED_STRING: 2605 lvalp->string.value = token->string_value(&lvalp->string.length); 2606 return QUOTED_STRING; 2607 2608 case Token::TOKEN_OPERATOR: 2609 return token->operator_value(); 2610 2611 case Token::TOKEN_INTEGER: 2612 lvalp->integer = token->integer_value(); 2613 return INTEGER; 2614 } 2615} 2616 2617// This function is called by the bison parser to report an error. 2618 2619extern "C" void 2620yyerror(void* closurev, const char* message) 2621{ 2622 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2623 gold_error(_("%s:%d:%d: %s"), closure->filename(), closure->lineno(), 2624 closure->charpos(), message); 2625} 2626 2627// Called by the bison parser to add an external symbol to the link. 2628 2629extern "C" void 2630script_add_extern(void* closurev, const char* name, size_t length) 2631{ 2632 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2633 closure->script_options()->add_symbol_reference(name, length); 2634} 2635 2636// Called by the bison parser to add a file to the link. 2637 2638extern "C" void 2639script_add_file(void* closurev, const char* name, size_t length) 2640{ 2641 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2642 2643 // If this is an absolute path, and we found the script in the 2644 // sysroot, then we want to prepend the sysroot to the file name. 2645 // For example, this is how we handle a cross link to the x86_64 2646 // libc.so, which refers to /lib/libc.so.6. 2647 std::string name_string(name, length); 2648 const char* extra_search_path = "."; 2649 std::string script_directory; 2650 if (IS_ABSOLUTE_PATH(name_string.c_str())) 2651 { 2652 if (closure->is_in_sysroot()) 2653 { 2654 const std::string& sysroot(parameters->options().sysroot()); 2655 gold_assert(!sysroot.empty()); 2656 name_string = sysroot + name_string; 2657 } 2658 } 2659 else 2660 { 2661 // In addition to checking the normal library search path, we 2662 // also want to check in the script-directory. 2663 const char* slash = strrchr(closure->filename(), '/'); 2664 if (slash != NULL) 2665 { 2666 script_directory.assign(closure->filename(), 2667 slash - closure->filename() + 1); 2668 extra_search_path = script_directory.c_str(); 2669 } 2670 } 2671 2672 Input_file_argument file(name_string.c_str(), 2673 Input_file_argument::INPUT_FILE_TYPE_FILE, 2674 extra_search_path, false, 2675 closure->position_dependent_options()); 2676 Input_argument& arg = closure->inputs()->add_file(file); 2677 arg.set_script_info(closure->script_info()); 2678} 2679 2680// Called by the bison parser to add a library to the link. 2681 2682extern "C" void 2683script_add_library(void* closurev, const char* name, size_t length) 2684{ 2685 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2686 std::string name_string(name, length); 2687 2688 if (name_string[0] != 'l') 2689 gold_error(_("library name must be prefixed with -l")); 2690 2691 Input_file_argument file(name_string.c_str() + 1, 2692 Input_file_argument::INPUT_FILE_TYPE_LIBRARY, 2693 "", false, 2694 closure->position_dependent_options()); 2695 Input_argument& arg = closure->inputs()->add_file(file); 2696 arg.set_script_info(closure->script_info()); 2697} 2698 2699// Called by the bison parser to start a group. If we are already in 2700// a group, that means that this script was invoked within a 2701// --start-group --end-group sequence on the command line, or that 2702// this script was found in a GROUP of another script. In that case, 2703// we simply continue the existing group, rather than starting a new 2704// one. It is possible to construct a case in which this will do 2705// something other than what would happen if we did a recursive group, 2706// but it's hard to imagine why the different behaviour would be 2707// useful for a real program. Avoiding recursive groups is simpler 2708// and more efficient. 2709 2710extern "C" void 2711script_start_group(void* closurev) 2712{ 2713 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2714 if (!closure->in_group()) 2715 closure->inputs()->start_group(); 2716} 2717 2718// Called by the bison parser at the end of a group. 2719 2720extern "C" void 2721script_end_group(void* closurev) 2722{ 2723 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2724 if (!closure->in_group()) 2725 closure->inputs()->end_group(); 2726} 2727 2728// Called by the bison parser to start an AS_NEEDED list. 2729 2730extern "C" void 2731script_start_as_needed(void* closurev) 2732{ 2733 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2734 closure->position_dependent_options().set_as_needed(true); 2735} 2736 2737// Called by the bison parser at the end of an AS_NEEDED list. 2738 2739extern "C" void 2740script_end_as_needed(void* closurev) 2741{ 2742 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2743 closure->position_dependent_options().set_as_needed(false); 2744} 2745 2746// Called by the bison parser to set the entry symbol. 2747 2748extern "C" void 2749script_set_entry(void* closurev, const char* entry, size_t length) 2750{ 2751 // We'll parse this exactly the same as --entry=ENTRY on the commandline 2752 // TODO(csilvers): FIXME -- call set_entry directly. 2753 std::string arg("--entry="); 2754 arg.append(entry, length); 2755 script_parse_option(closurev, arg.c_str(), arg.size()); 2756} 2757 2758// Called by the bison parser to set whether to define common symbols. 2759 2760extern "C" void 2761script_set_common_allocation(void* closurev, int set) 2762{ 2763 const char* arg = set != 0 ? "--define-common" : "--no-define-common"; 2764 script_parse_option(closurev, arg, strlen(arg)); 2765} 2766 2767// Called by the bison parser to refer to a symbol. 2768 2769extern "C" Expression* 2770script_symbol(void* closurev, const char* name, size_t length) 2771{ 2772 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2773 if (length != 1 || name[0] != '.') 2774 closure->script_options()->add_symbol_reference(name, length); 2775 return script_exp_string(name, length); 2776} 2777 2778// Called by the bison parser to define a symbol. 2779 2780extern "C" void 2781script_set_symbol(void* closurev, const char* name, size_t length, 2782 Expression* value, int providei, int hiddeni) 2783{ 2784 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2785 const bool provide = providei != 0; 2786 const bool hidden = hiddeni != 0; 2787 closure->script_options()->add_symbol_assignment(name, length, 2788 closure->parsing_defsym(), 2789 value, provide, hidden); 2790 closure->clear_skip_on_incompatible_target(); 2791} 2792 2793// Called by the bison parser to add an assertion. 2794 2795extern "C" void 2796script_add_assertion(void* closurev, Expression* check, const char* message, 2797 size_t messagelen) 2798{ 2799 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2800 closure->script_options()->add_assertion(check, message, messagelen); 2801 closure->clear_skip_on_incompatible_target(); 2802} 2803 2804// Called by the bison parser to parse an OPTION. 2805 2806extern "C" void 2807script_parse_option(void* closurev, const char* option, size_t length) 2808{ 2809 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2810 // We treat the option as a single command-line option, even if 2811 // it has internal whitespace. 2812 if (closure->command_line() == NULL) 2813 { 2814 // There are some options that we could handle here--e.g., 2815 // -lLIBRARY. Should we bother? 2816 gold_warning(_("%s:%d:%d: ignoring command OPTION; OPTION is only valid" 2817 " for scripts specified via -T/--script"), 2818 closure->filename(), closure->lineno(), closure->charpos()); 2819 } 2820 else 2821 { 2822 bool past_a_double_dash_option = false; 2823 const char* mutable_option = strndup(option, length); 2824 gold_assert(mutable_option != NULL); 2825 closure->command_line()->process_one_option(1, &mutable_option, 0, 2826 &past_a_double_dash_option); 2827 // The General_options class will quite possibly store a pointer 2828 // into mutable_option, so we can't free it. In cases the class 2829 // does not store such a pointer, this is a memory leak. Alas. :( 2830 } 2831 closure->clear_skip_on_incompatible_target(); 2832} 2833 2834// Called by the bison parser to handle OUTPUT_FORMAT. OUTPUT_FORMAT 2835// takes either one or three arguments. In the three argument case, 2836// the format depends on the endianness option, which we don't 2837// currently support (FIXME). If we see an OUTPUT_FORMAT for the 2838// wrong format, then we want to search for a new file. Returning 0 2839// here will cause the parser to immediately abort. 2840 2841extern "C" int 2842script_check_output_format(void* closurev, 2843 const char* default_name, size_t default_length, 2844 const char*, size_t, const char*, size_t) 2845{ 2846 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2847 std::string name(default_name, default_length); 2848 Target* target = select_target_by_bfd_name(name.c_str()); 2849 if (target == NULL || !parameters->is_compatible_target(target)) 2850 { 2851 if (closure->skip_on_incompatible_target()) 2852 { 2853 closure->set_found_incompatible_target(); 2854 return 0; 2855 } 2856 // FIXME: Should we warn about the unknown target? 2857 } 2858 return 1; 2859} 2860 2861// Called by the bison parser to handle TARGET. 2862 2863extern "C" void 2864script_set_target(void* closurev, const char* target, size_t len) 2865{ 2866 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2867 std::string s(target, len); 2868 General_options::Object_format format_enum; 2869 format_enum = General_options::string_to_object_format(s.c_str()); 2870 closure->position_dependent_options().set_format_enum(format_enum); 2871} 2872 2873// Called by the bison parser to handle SEARCH_DIR. This is handled 2874// exactly like a -L option. 2875 2876extern "C" void 2877script_add_search_dir(void* closurev, const char* option, size_t length) 2878{ 2879 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2880 if (closure->command_line() == NULL) 2881 gold_warning(_("%s:%d:%d: ignoring SEARCH_DIR; SEARCH_DIR is only valid" 2882 " for scripts specified via -T/--script"), 2883 closure->filename(), closure->lineno(), closure->charpos()); 2884 else if (!closure->command_line()->options().nostdlib()) 2885 { 2886 std::string s = "-L" + std::string(option, length); 2887 script_parse_option(closurev, s.c_str(), s.size()); 2888 } 2889} 2890 2891/* Called by the bison parser to push the lexer into expression 2892 mode. */ 2893 2894extern "C" void 2895script_push_lex_into_expression_mode(void* closurev) 2896{ 2897 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2898 closure->push_lex_mode(Lex::EXPRESSION); 2899} 2900 2901/* Called by the bison parser to push the lexer into version 2902 mode. */ 2903 2904extern "C" void 2905script_push_lex_into_version_mode(void* closurev) 2906{ 2907 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2908 if (closure->version_script()->is_finalized()) 2909 gold_error(_("%s:%d:%d: invalid use of VERSION in input file"), 2910 closure->filename(), closure->lineno(), closure->charpos()); 2911 closure->push_lex_mode(Lex::VERSION_SCRIPT); 2912} 2913 2914/* Called by the bison parser to pop the lexer mode. */ 2915 2916extern "C" void 2917script_pop_lex_mode(void* closurev) 2918{ 2919 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2920 closure->pop_lex_mode(); 2921} 2922 2923// Register an entire version node. For example: 2924// 2925// GLIBC_2.1 { 2926// global: foo; 2927// } GLIBC_2.0; 2928// 2929// - tag is "GLIBC_2.1" 2930// - tree contains the information "global: foo" 2931// - deps contains "GLIBC_2.0" 2932 2933extern "C" void 2934script_register_vers_node(void*, 2935 const char* tag, 2936 int taglen, 2937 struct Version_tree* tree, 2938 struct Version_dependency_list* deps) 2939{ 2940 gold_assert(tree != NULL); 2941 tree->dependencies = deps; 2942 if (tag != NULL) 2943 tree->tag = std::string(tag, taglen); 2944} 2945 2946// Add a dependencies to the list of existing dependencies, if any, 2947// and return the expanded list. 2948 2949extern "C" struct Version_dependency_list* 2950script_add_vers_depend(void* closurev, 2951 struct Version_dependency_list* all_deps, 2952 const char* depend_to_add, int deplen) 2953{ 2954 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2955 if (all_deps == NULL) 2956 all_deps = closure->version_script()->allocate_dependency_list(); 2957 all_deps->dependencies.push_back(std::string(depend_to_add, deplen)); 2958 return all_deps; 2959} 2960 2961// Add a pattern expression to an existing list of expressions, if any. 2962 2963extern "C" struct Version_expression_list* 2964script_new_vers_pattern(void* closurev, 2965 struct Version_expression_list* expressions, 2966 const char* pattern, int patlen, int exact_match) 2967{ 2968 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2969 if (expressions == NULL) 2970 expressions = closure->version_script()->allocate_expression_list(); 2971 expressions->expressions.push_back( 2972 Version_expression(std::string(pattern, patlen), 2973 closure->get_current_language(), 2974 static_cast<bool>(exact_match))); 2975 return expressions; 2976} 2977 2978// Attaches b to the end of a, and clears b. So a = a + b and b = {}. 2979 2980extern "C" struct Version_expression_list* 2981script_merge_expressions(struct Version_expression_list* a, 2982 struct Version_expression_list* b) 2983{ 2984 a->expressions.insert(a->expressions.end(), 2985 b->expressions.begin(), b->expressions.end()); 2986 // We could delete b and remove it from expressions_lists_, but 2987 // that's a lot of work. This works just as well. 2988 b->expressions.clear(); 2989 return a; 2990} 2991 2992// Combine the global and local expressions into a a Version_tree. 2993 2994extern "C" struct Version_tree* 2995script_new_vers_node(void* closurev, 2996 struct Version_expression_list* global, 2997 struct Version_expression_list* local) 2998{ 2999 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3000 Version_tree* tree = closure->version_script()->allocate_version_tree(); 3001 tree->global = global; 3002 tree->local = local; 3003 return tree; 3004} 3005 3006// Handle a transition in language, such as at the 3007// start or end of 'extern "C++"' 3008 3009extern "C" void 3010version_script_push_lang(void* closurev, const char* lang, int langlen) 3011{ 3012 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3013 std::string language(lang, langlen); 3014 Version_script_info::Language code; 3015 if (language.empty() || language == "C") 3016 code = Version_script_info::LANGUAGE_C; 3017 else if (language == "C++") 3018 code = Version_script_info::LANGUAGE_CXX; 3019 else if (language == "Java") 3020 code = Version_script_info::LANGUAGE_JAVA; 3021 else 3022 { 3023 char* buf = new char[langlen + 100]; 3024 snprintf(buf, langlen + 100, 3025 _("unrecognized version script language '%s'"), 3026 language.c_str()); 3027 yyerror(closurev, buf); 3028 delete[] buf; 3029 code = Version_script_info::LANGUAGE_C; 3030 } 3031 closure->push_language(code); 3032} 3033 3034extern "C" void 3035version_script_pop_lang(void* closurev) 3036{ 3037 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3038 closure->pop_language(); 3039} 3040 3041// Called by the bison parser to start a SECTIONS clause. 3042 3043extern "C" void 3044script_start_sections(void* closurev) 3045{ 3046 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3047 closure->script_options()->script_sections()->start_sections(); 3048 closure->clear_skip_on_incompatible_target(); 3049} 3050 3051// Called by the bison parser to finish a SECTIONS clause. 3052 3053extern "C" void 3054script_finish_sections(void* closurev) 3055{ 3056 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3057 closure->script_options()->script_sections()->finish_sections(); 3058} 3059 3060// Start processing entries for an output section. 3061 3062extern "C" void 3063script_start_output_section(void* closurev, const char* name, size_t namelen, 3064 const struct Parser_output_section_header* header) 3065{ 3066 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3067 closure->script_options()->script_sections()->start_output_section(name, 3068 namelen, 3069 header); 3070} 3071 3072// Finish processing entries for an output section. 3073 3074extern "C" void 3075script_finish_output_section(void* closurev, 3076 const struct Parser_output_section_trailer* trail) 3077{ 3078 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3079 closure->script_options()->script_sections()->finish_output_section(trail); 3080} 3081 3082// Add a data item (e.g., "WORD (0)") to the current output section. 3083 3084extern "C" void 3085script_add_data(void* closurev, int data_token, Expression* val) 3086{ 3087 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3088 int size; 3089 bool is_signed = true; 3090 switch (data_token) 3091 { 3092 case QUAD: 3093 size = 8; 3094 is_signed = false; 3095 break; 3096 case SQUAD: 3097 size = 8; 3098 break; 3099 case LONG: 3100 size = 4; 3101 break; 3102 case SHORT: 3103 size = 2; 3104 break; 3105 case BYTE: 3106 size = 1; 3107 break; 3108 default: 3109 gold_unreachable(); 3110 } 3111 closure->script_options()->script_sections()->add_data(size, is_signed, val); 3112} 3113 3114// Add a clause setting the fill value to the current output section. 3115 3116extern "C" void 3117script_add_fill(void* closurev, Expression* val) 3118{ 3119 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3120 closure->script_options()->script_sections()->add_fill(val); 3121} 3122 3123// Add a new input section specification to the current output 3124// section. 3125 3126extern "C" void 3127script_add_input_section(void* closurev, 3128 const struct Input_section_spec* spec, 3129 int keepi) 3130{ 3131 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3132 bool keep = keepi != 0; 3133 closure->script_options()->script_sections()->add_input_section(spec, keep); 3134} 3135 3136// When we see DATA_SEGMENT_ALIGN we record that following output 3137// sections may be relro. 3138 3139extern "C" void 3140script_data_segment_align(void* closurev) 3141{ 3142 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3143 if (!closure->script_options()->saw_sections_clause()) 3144 gold_error(_("%s:%d:%d: DATA_SEGMENT_ALIGN not in SECTIONS clause"), 3145 closure->filename(), closure->lineno(), closure->charpos()); 3146 else 3147 closure->script_options()->script_sections()->data_segment_align(); 3148} 3149 3150// When we see DATA_SEGMENT_RELRO_END we know that all output sections 3151// since DATA_SEGMENT_ALIGN should be relro. 3152 3153extern "C" void 3154script_data_segment_relro_end(void* closurev) 3155{ 3156 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3157 if (!closure->script_options()->saw_sections_clause()) 3158 gold_error(_("%s:%d:%d: DATA_SEGMENT_ALIGN not in SECTIONS clause"), 3159 closure->filename(), closure->lineno(), closure->charpos()); 3160 else 3161 closure->script_options()->script_sections()->data_segment_relro_end(); 3162} 3163 3164// Create a new list of string/sort pairs. 3165 3166extern "C" String_sort_list_ptr 3167script_new_string_sort_list(const struct Wildcard_section* string_sort) 3168{ 3169 return new String_sort_list(1, *string_sort); 3170} 3171 3172// Add an entry to a list of string/sort pairs. The way the parser 3173// works permits us to simply modify the first parameter, rather than 3174// copy the vector. 3175 3176extern "C" String_sort_list_ptr 3177script_string_sort_list_add(String_sort_list_ptr pv, 3178 const struct Wildcard_section* string_sort) 3179{ 3180 if (pv == NULL) 3181 return script_new_string_sort_list(string_sort); 3182 else 3183 { 3184 pv->push_back(*string_sort); 3185 return pv; 3186 } 3187} 3188 3189// Create a new list of strings. 3190 3191extern "C" String_list_ptr 3192script_new_string_list(const char* str, size_t len) 3193{ 3194 return new String_list(1, std::string(str, len)); 3195} 3196 3197// Add an element to a list of strings. The way the parser works 3198// permits us to simply modify the first parameter, rather than copy 3199// the vector. 3200 3201extern "C" String_list_ptr 3202script_string_list_push_back(String_list_ptr pv, const char* str, size_t len) 3203{ 3204 if (pv == NULL) 3205 return script_new_string_list(str, len); 3206 else 3207 { 3208 pv->push_back(std::string(str, len)); 3209 return pv; 3210 } 3211} 3212 3213// Concatenate two string lists. Either or both may be NULL. The way 3214// the parser works permits us to modify the parameters, rather than 3215// copy the vector. 3216 3217extern "C" String_list_ptr 3218script_string_list_append(String_list_ptr pv1, String_list_ptr pv2) 3219{ 3220 if (pv1 == NULL) 3221 return pv2; 3222 if (pv2 == NULL) 3223 return pv1; 3224 pv1->insert(pv1->end(), pv2->begin(), pv2->end()); 3225 return pv1; 3226} 3227 3228// Add a new program header. 3229 3230extern "C" void 3231script_add_phdr(void* closurev, const char* name, size_t namelen, 3232 unsigned int type, const Phdr_info* info) 3233{ 3234 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3235 bool includes_filehdr = info->includes_filehdr != 0; 3236 bool includes_phdrs = info->includes_phdrs != 0; 3237 bool is_flags_valid = info->is_flags_valid != 0; 3238 Script_sections* ss = closure->script_options()->script_sections(); 3239 ss->add_phdr(name, namelen, type, includes_filehdr, includes_phdrs, 3240 is_flags_valid, info->flags, info->load_address); 3241 closure->clear_skip_on_incompatible_target(); 3242} 3243 3244// Convert a program header string to a type. 3245 3246#define PHDR_TYPE(NAME) { #NAME, sizeof(#NAME) - 1, elfcpp::NAME } 3247 3248static struct 3249{ 3250 const char* name; 3251 size_t namelen; 3252 unsigned int val; 3253} phdr_type_names[] = 3254{ 3255 PHDR_TYPE(PT_NULL), 3256 PHDR_TYPE(PT_LOAD), 3257 PHDR_TYPE(PT_DYNAMIC), 3258 PHDR_TYPE(PT_INTERP), 3259 PHDR_TYPE(PT_NOTE), 3260 PHDR_TYPE(PT_SHLIB), 3261 PHDR_TYPE(PT_PHDR), 3262 PHDR_TYPE(PT_TLS), 3263 PHDR_TYPE(PT_GNU_EH_FRAME), 3264 PHDR_TYPE(PT_GNU_STACK), 3265 PHDR_TYPE(PT_GNU_RELRO) 3266}; 3267 3268extern "C" unsigned int 3269script_phdr_string_to_type(void* closurev, const char* name, size_t namelen) 3270{ 3271 for (unsigned int i = 0; 3272 i < sizeof(phdr_type_names) / sizeof(phdr_type_names[0]); 3273 ++i) 3274 if (namelen == phdr_type_names[i].namelen 3275 && strncmp(name, phdr_type_names[i].name, namelen) == 0) 3276 return phdr_type_names[i].val; 3277 yyerror(closurev, _("unknown PHDR type (try integer)")); 3278 return elfcpp::PT_NULL; 3279} 3280 3281extern "C" void 3282script_saw_segment_start_expression(void* closurev) 3283{ 3284 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3285 Script_sections* ss = closure->script_options()->script_sections(); 3286 ss->set_saw_segment_start_expression(true); 3287} 3288 3289extern "C" void 3290script_set_section_region(void* closurev, const char* name, size_t namelen, 3291 int set_vma) 3292{ 3293 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3294 if (!closure->script_options()->saw_sections_clause()) 3295 { 3296 gold_error(_("%s:%d:%d: MEMORY region '%.*s' referred to outside of " 3297 "SECTIONS clause"), 3298 closure->filename(), closure->lineno(), closure->charpos(), 3299 static_cast<int>(namelen), name); 3300 return; 3301 } 3302 3303 Script_sections* ss = closure->script_options()->script_sections(); 3304 Memory_region* mr = ss->find_memory_region(name, namelen); 3305 if (mr == NULL) 3306 { 3307 gold_error(_("%s:%d:%d: MEMORY region '%.*s' not declared"), 3308 closure->filename(), closure->lineno(), closure->charpos(), 3309 static_cast<int>(namelen), name); 3310 return; 3311 } 3312 3313 ss->set_memory_region(mr, set_vma); 3314} 3315 3316extern "C" void 3317script_add_memory(void* closurev, const char* name, size_t namelen, 3318 unsigned int attrs, Expression* origin, Expression* length) 3319{ 3320 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3321 Script_sections* ss = closure->script_options()->script_sections(); 3322 ss->add_memory_region(name, namelen, attrs, origin, length); 3323} 3324 3325extern "C" unsigned int 3326script_parse_memory_attr(void* closurev, const char* attrs, size_t attrlen, 3327 int invert) 3328{ 3329 int attributes = 0; 3330 3331 while (attrlen--) 3332 switch (*attrs++) 3333 { 3334 case 'R': 3335 case 'r': 3336 attributes |= MEM_READABLE; break; 3337 case 'W': 3338 case 'w': 3339 attributes |= MEM_READABLE | MEM_WRITEABLE; break; 3340 case 'X': 3341 case 'x': 3342 attributes |= MEM_EXECUTABLE; break; 3343 case 'A': 3344 case 'a': 3345 attributes |= MEM_ALLOCATABLE; break; 3346 case 'I': 3347 case 'i': 3348 case 'L': 3349 case 'l': 3350 attributes |= MEM_INITIALIZED; break; 3351 default: 3352 yyerror(closurev, _("unknown MEMORY attribute")); 3353 } 3354 3355 if (invert) 3356 attributes = (~ attributes) & MEM_ATTR_MASK; 3357 3358 return attributes; 3359} 3360 3361extern "C" void 3362script_include_directive(void* closurev, const char* filename, size_t length) 3363{ 3364 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3365 std::string name(filename, length); 3366 Command_line* cmdline = closure->command_line(); 3367 read_script_file(name.c_str(), cmdline, &cmdline->script_options(), 3368 PARSING_LINKER_SCRIPT, Lex::LINKER_SCRIPT); 3369} 3370 3371// Functions for memory regions. 3372 3373extern "C" Expression* 3374script_exp_function_origin(void* closurev, const char* name, size_t namelen) 3375{ 3376 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3377 Script_sections* ss = closure->script_options()->script_sections(); 3378 Expression* origin = ss->find_memory_region_origin(name, namelen); 3379 3380 if (origin == NULL) 3381 { 3382 gold_error(_("undefined memory region '%s' referenced " 3383 "in ORIGIN expression"), 3384 name); 3385 // Create a dummy expression to prevent crashes later on. 3386 origin = script_exp_integer(0); 3387 } 3388 3389 return origin; 3390} 3391 3392extern "C" Expression* 3393script_exp_function_length(void* closurev, const char* name, size_t namelen) 3394{ 3395 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3396 Script_sections* ss = closure->script_options()->script_sections(); 3397 Expression* length = ss->find_memory_region_length(name, namelen); 3398 3399 if (length == NULL) 3400 { 3401 gold_error(_("undefined memory region '%s' referenced " 3402 "in LENGTH expression"), 3403 name); 3404 // Create a dummy expression to prevent crashes later on. 3405 length = script_exp_integer(0); 3406 } 3407 3408 return length; 3409} 3410