1 // script.cc -- handle linker scripts for gold. 2 3 // Copyright (C) 2006-2014 Free Software Foundation, Inc. 4 // Written by Ian Lance Taylor <iant (at) 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 48 namespace 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 54 class 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. 169 170 uint64_t 171 Token::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. 198 199 class 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. 367 368 void 369 Lex::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. 398 399 inline bool 400 Lex::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 '$': 415 return true; 416 417 case '/': case '\\': 418 return this->mode_ == LINKER_SCRIPT; 419 420 case '~': 421 return this->mode_ == LINKER_SCRIPT && can_continue_name(&c2); 422 423 case '*': case '[': 424 return (this->mode_ == VERSION_SCRIPT 425 || this->mode_ == DYNAMIC_LIST 426 || (this->mode_ == LINKER_SCRIPT 427 && can_continue_name(&c2))); 428 429 default: 430 return false; 431 } 432 } 433 434 // Return whether C can continue a name which has already started. 435 // Subsequent characters in a name are the same as the leading 436 // characters, plus digits and "=+-:[],?*". So in general the linker 437 // script language requires spaces around operators, unless we know 438 // that we are parsing an expression. 439 440 inline const char* 441 Lex::can_continue_name(const char* c) 442 { 443 switch (*c) 444 { 445 case 'A': case 'B': case 'C': case 'D': case 'E': case 'F': 446 case 'G': case 'H': case 'I': case 'J': case 'K': case 'L': 447 case 'M': case 'N': case 'O': case 'Q': case 'P': case 'R': 448 case 'S': case 'T': case 'U': case 'V': case 'W': case 'X': 449 case 'Y': case 'Z': 450 case 'a': case 'b': case 'c': case 'd': case 'e': case 'f': 451 case 'g': case 'h': case 'i': case 'j': case 'k': case 'l': 452 case 'm': case 'n': case 'o': case 'q': case 'p': case 'r': 453 case 's': case 't': case 'u': case 'v': case 'w': case 'x': 454 case 'y': case 'z': 455 case '_': case '.': case '$': 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. 503 504 inline bool 505 Lex::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. 513 514 inline const char* 515 Lex::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. 531 532 inline bool 533 Lex::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. 549 550 inline int 551 Lex::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. 572 573 inline int 574 Lex::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. 633 634 inline int 635 Lex::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. 670 671 bool 672 Lex::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. 696 697 bool 698 Lex::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. 718 719 inline Token 720 Lex::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. 741 742 Token 743 Lex::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. 759 760 Token 761 Lex::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. 863 864 const Token* 865 Lex::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? 898 899 void 900 Symbol_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. 919 920 void 921 Symbol_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. 927 928 void 929 Symbol_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. 938 939 void 940 Symbol_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 } 975 976 template<int size> 977 void 978 Symbol_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 elfcpp::STT type = elfcpp::STT_NOTYPE; 984 elfcpp::STV vis = elfcpp::STV_DEFAULT; 985 unsigned char nonvis = 0; 986 uint64_t final_val = this->val_->eval_maybe_dot(symtab, layout, true, 987 is_dot_available, 988 dot_value, dot_section, 989 §ion, NULL, &type, 990 &vis, &nonvis, false); 991 Sized_symbol<size>* ssym = symtab->get_sized_symbol<size>(this->sym_); 992 ssym->set_value(final_val); 993 ssym->set_type(type); 994 ssym->set_visibility(vis); 995 ssym->set_nonvis(nonvis); 996 if (section != NULL) 997 ssym->set_output_section(section); 998 } 999 1000 // Set the symbol value if the expression yields an absolute value or 1001 // a value relative to DOT_SECTION. 1002 1003 void 1004 Symbol_assignment::set_if_absolute(Symbol_table* symtab, const Layout* layout, 1005 bool is_dot_available, uint64_t dot_value, 1006 Output_section* dot_section) 1007 { 1008 if (this->sym_ == NULL) 1009 return; 1010 1011 Output_section* val_section; 1012 uint64_t val = this->val_->eval_maybe_dot(symtab, layout, false, 1013 is_dot_available, dot_value, 1014 dot_section, &val_section, NULL, 1015 NULL, NULL, NULL, false); 1016 if (val_section != NULL && val_section != dot_section) 1017 return; 1018 1019 if (parameters->target().get_size() == 32) 1020 { 1021 #if defined(HAVE_TARGET_32_LITTLE) || defined(HAVE_TARGET_32_BIG) 1022 Sized_symbol<32>* ssym = symtab->get_sized_symbol<32>(this->sym_); 1023 ssym->set_value(val); 1024 #else 1025 gold_unreachable(); 1026 #endif 1027 } 1028 else if (parameters->target().get_size() == 64) 1029 { 1030 #if defined(HAVE_TARGET_64_LITTLE) || defined(HAVE_TARGET_64_BIG) 1031 Sized_symbol<64>* ssym = symtab->get_sized_symbol<64>(this->sym_); 1032 ssym->set_value(val); 1033 #else 1034 gold_unreachable(); 1035 #endif 1036 } 1037 else 1038 gold_unreachable(); 1039 if (val_section != NULL) 1040 this->sym_->set_output_section(val_section); 1041 } 1042 1043 // Print for debugging. 1044 1045 void 1046 Symbol_assignment::print(FILE* f) const 1047 { 1048 if (this->provide_ && this->hidden_) 1049 fprintf(f, "PROVIDE_HIDDEN("); 1050 else if (this->provide_) 1051 fprintf(f, "PROVIDE("); 1052 else if (this->hidden_) 1053 gold_unreachable(); 1054 1055 fprintf(f, "%s = ", this->name_.c_str()); 1056 this->val_->print(f); 1057 1058 if (this->provide_ || this->hidden_) 1059 fprintf(f, ")"); 1060 1061 fprintf(f, "\n"); 1062 } 1063 1064 // Class Script_assertion. 1065 1066 // Check the assertion. 1067 1068 void 1069 Script_assertion::check(const Symbol_table* symtab, const Layout* layout) 1070 { 1071 if (!this->check_->eval(symtab, layout, true)) 1072 gold_error("%s", this->message_.c_str()); 1073 } 1074 1075 // Print for debugging. 1076 1077 void 1078 Script_assertion::print(FILE* f) const 1079 { 1080 fprintf(f, "ASSERT("); 1081 this->check_->print(f); 1082 fprintf(f, ", \"%s\")\n", this->message_.c_str()); 1083 } 1084 1085 // Class Script_options. 1086 1087 Script_options::Script_options() 1088 : entry_(), symbol_assignments_(), symbol_definitions_(), 1089 symbol_references_(), version_script_info_(), script_sections_() 1090 { 1091 } 1092 1093 // Returns true if NAME is on the list of symbol assignments waiting 1094 // to be processed. 1095 1096 bool 1097 Script_options::is_pending_assignment(const char* name) 1098 { 1099 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin(); 1100 p != this->symbol_assignments_.end(); 1101 ++p) 1102 if ((*p)->name() == name) 1103 return true; 1104 return false; 1105 } 1106 1107 // Add a symbol to be defined. 1108 1109 void 1110 Script_options::add_symbol_assignment(const char* name, size_t length, 1111 bool is_defsym, Expression* value, 1112 bool provide, bool hidden) 1113 { 1114 if (length != 1 || name[0] != '.') 1115 { 1116 if (this->script_sections_.in_sections_clause()) 1117 { 1118 gold_assert(!is_defsym); 1119 this->script_sections_.add_symbol_assignment(name, length, value, 1120 provide, hidden); 1121 } 1122 else 1123 { 1124 Symbol_assignment* p = new Symbol_assignment(name, length, is_defsym, 1125 value, provide, hidden); 1126 this->symbol_assignments_.push_back(p); 1127 } 1128 1129 if (!provide) 1130 { 1131 std::string n(name, length); 1132 this->symbol_definitions_.insert(n); 1133 this->symbol_references_.erase(n); 1134 } 1135 } 1136 else 1137 { 1138 if (provide || hidden) 1139 gold_error(_("invalid use of PROVIDE for dot symbol")); 1140 1141 // The GNU linker permits assignments to dot outside of SECTIONS 1142 // clauses and treats them as occurring inside, so we don't 1143 // check in_sections_clause here. 1144 this->script_sections_.add_dot_assignment(value); 1145 } 1146 } 1147 1148 // Add a reference to a symbol. 1149 1150 void 1151 Script_options::add_symbol_reference(const char* name, size_t length) 1152 { 1153 if (length != 1 || name[0] != '.') 1154 { 1155 std::string n(name, length); 1156 if (this->symbol_definitions_.find(n) == this->symbol_definitions_.end()) 1157 this->symbol_references_.insert(n); 1158 } 1159 } 1160 1161 // Add an assertion. 1162 1163 void 1164 Script_options::add_assertion(Expression* check, const char* message, 1165 size_t messagelen) 1166 { 1167 if (this->script_sections_.in_sections_clause()) 1168 this->script_sections_.add_assertion(check, message, messagelen); 1169 else 1170 { 1171 Script_assertion* p = new Script_assertion(check, message, messagelen); 1172 this->assertions_.push_back(p); 1173 } 1174 } 1175 1176 // Create sections required by any linker scripts. 1177 1178 void 1179 Script_options::create_script_sections(Layout* layout) 1180 { 1181 if (this->saw_sections_clause()) 1182 this->script_sections_.create_sections(layout); 1183 } 1184 1185 // Add any symbols we are defining to the symbol table. 1186 1187 void 1188 Script_options::add_symbols_to_table(Symbol_table* symtab) 1189 { 1190 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin(); 1191 p != this->symbol_assignments_.end(); 1192 ++p) 1193 (*p)->add_to_table(symtab); 1194 this->script_sections_.add_symbols_to_table(symtab); 1195 } 1196 1197 // Finalize symbol values. Also check assertions. 1198 1199 void 1200 Script_options::finalize_symbols(Symbol_table* symtab, const Layout* layout) 1201 { 1202 // We finalize the symbols defined in SECTIONS first, because they 1203 // are the ones which may have changed. This way if symbol outside 1204 // SECTIONS are defined in terms of symbols inside SECTIONS, they 1205 // will get the right value. 1206 this->script_sections_.finalize_symbols(symtab, layout); 1207 1208 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin(); 1209 p != this->symbol_assignments_.end(); 1210 ++p) 1211 (*p)->finalize(symtab, layout); 1212 1213 for (Assertions::iterator p = this->assertions_.begin(); 1214 p != this->assertions_.end(); 1215 ++p) 1216 (*p)->check(symtab, layout); 1217 } 1218 1219 // Set section addresses. We set all the symbols which have absolute 1220 // values. Then we let the SECTIONS clause do its thing. This 1221 // returns the segment which holds the file header and segment 1222 // headers, if any. 1223 1224 Output_segment* 1225 Script_options::set_section_addresses(Symbol_table* symtab, Layout* layout) 1226 { 1227 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin(); 1228 p != this->symbol_assignments_.end(); 1229 ++p) 1230 (*p)->set_if_absolute(symtab, layout, false, 0, NULL); 1231 1232 return this->script_sections_.set_section_addresses(symtab, layout); 1233 } 1234 1235 // This class holds data passed through the parser to the lexer and to 1236 // the parser support functions. This avoids global variables. We 1237 // can't use global variables because we need not be called by a 1238 // singleton thread. 1239 1240 class Parser_closure 1241 { 1242 public: 1243 Parser_closure(const char* filename, 1244 const Position_dependent_options& posdep_options, 1245 bool parsing_defsym, bool in_group, bool is_in_sysroot, 1246 Command_line* command_line, 1247 Script_options* script_options, 1248 Lex* lex, 1249 bool skip_on_incompatible_target, 1250 Script_info* script_info) 1251 : filename_(filename), posdep_options_(posdep_options), 1252 parsing_defsym_(parsing_defsym), in_group_(in_group), 1253 is_in_sysroot_(is_in_sysroot), 1254 skip_on_incompatible_target_(skip_on_incompatible_target), 1255 found_incompatible_target_(false), 1256 command_line_(command_line), script_options_(script_options), 1257 version_script_info_(script_options->version_script_info()), 1258 lex_(lex), lineno_(0), charpos_(0), lex_mode_stack_(), inputs_(NULL), 1259 script_info_(script_info) 1260 { 1261 // We start out processing C symbols in the default lex mode. 1262 this->language_stack_.push_back(Version_script_info::LANGUAGE_C); 1263 this->lex_mode_stack_.push_back(lex->mode()); 1264 } 1265 1266 // Return the file name. 1267 const char* 1268 filename() const 1269 { return this->filename_; } 1270 1271 // Return the position dependent options. The caller may modify 1272 // this. 1273 Position_dependent_options& 1274 position_dependent_options() 1275 { return this->posdep_options_; } 1276 1277 // Whether we are parsing a --defsym. 1278 bool 1279 parsing_defsym() const 1280 { return this->parsing_defsym_; } 1281 1282 // Return whether this script is being run in a group. 1283 bool 1284 in_group() const 1285 { return this->in_group_; } 1286 1287 // Return whether this script was found using a directory in the 1288 // sysroot. 1289 bool 1290 is_in_sysroot() const 1291 { return this->is_in_sysroot_; } 1292 1293 // Whether to skip to the next file with the same name if we find an 1294 // incompatible target in an OUTPUT_FORMAT statement. 1295 bool 1296 skip_on_incompatible_target() const 1297 { return this->skip_on_incompatible_target_; } 1298 1299 // Stop skipping to the next file on an incompatible target. This 1300 // is called when we make some unrevocable change to the data 1301 // structures. 1302 void 1303 clear_skip_on_incompatible_target() 1304 { this->skip_on_incompatible_target_ = false; } 1305 1306 // Whether we found an incompatible target in an OUTPUT_FORMAT 1307 // statement. 1308 bool 1309 found_incompatible_target() const 1310 { return this->found_incompatible_target_; } 1311 1312 // Note that we found an incompatible target. 1313 void 1314 set_found_incompatible_target() 1315 { this->found_incompatible_target_ = true; } 1316 1317 // Returns the Command_line structure passed in at constructor time. 1318 // This value may be NULL. The caller may modify this, which modifies 1319 // the passed-in Command_line object (not a copy). 1320 Command_line* 1321 command_line() 1322 { return this->command_line_; } 1323 1324 // Return the options which may be set by a script. 1325 Script_options* 1326 script_options() 1327 { return this->script_options_; } 1328 1329 // Return the object in which version script information should be stored. 1330 Version_script_info* 1331 version_script() 1332 { return this->version_script_info_; } 1333 1334 // Return the next token, and advance. 1335 const Token* 1336 next_token() 1337 { 1338 const Token* token = this->lex_->next_token(); 1339 this->lineno_ = token->lineno(); 1340 this->charpos_ = token->charpos(); 1341 return token; 1342 } 1343 1344 // Set a new lexer mode, pushing the current one. 1345 void 1346 push_lex_mode(Lex::Mode mode) 1347 { 1348 this->lex_mode_stack_.push_back(this->lex_->mode()); 1349 this->lex_->set_mode(mode); 1350 } 1351 1352 // Pop the lexer mode. 1353 void 1354 pop_lex_mode() 1355 { 1356 gold_assert(!this->lex_mode_stack_.empty()); 1357 this->lex_->set_mode(this->lex_mode_stack_.back()); 1358 this->lex_mode_stack_.pop_back(); 1359 } 1360 1361 // Return the current lexer mode. 1362 Lex::Mode 1363 lex_mode() const 1364 { return this->lex_mode_stack_.back(); } 1365 1366 // Return the line number of the last token. 1367 int 1368 lineno() const 1369 { return this->lineno_; } 1370 1371 // Return the character position in the line of the last token. 1372 int 1373 charpos() const 1374 { return this->charpos_; } 1375 1376 // Return the list of input files, creating it if necessary. This 1377 // is a space leak--we never free the INPUTS_ pointer. 1378 Input_arguments* 1379 inputs() 1380 { 1381 if (this->inputs_ == NULL) 1382 this->inputs_ = new Input_arguments(); 1383 return this->inputs_; 1384 } 1385 1386 // Return whether we saw any input files. 1387 bool 1388 saw_inputs() const 1389 { return this->inputs_ != NULL && !this->inputs_->empty(); } 1390 1391 // Return the current language being processed in a version script 1392 // (eg, "C++"). The empty string represents unmangled C names. 1393 Version_script_info::Language 1394 get_current_language() const 1395 { return this->language_stack_.back(); } 1396 1397 // Push a language onto the stack when entering an extern block. 1398 void 1399 push_language(Version_script_info::Language lang) 1400 { this->language_stack_.push_back(lang); } 1401 1402 // Pop a language off of the stack when exiting an extern block. 1403 void 1404 pop_language() 1405 { 1406 gold_assert(!this->language_stack_.empty()); 1407 this->language_stack_.pop_back(); 1408 } 1409 1410 // Return a pointer to the incremental info. 1411 Script_info* 1412 script_info() 1413 { return this->script_info_; } 1414 1415 private: 1416 // The name of the file we are reading. 1417 const char* filename_; 1418 // The position dependent options. 1419 Position_dependent_options posdep_options_; 1420 // True if we are parsing a --defsym. 1421 bool parsing_defsym_; 1422 // Whether we are currently in a --start-group/--end-group. 1423 bool in_group_; 1424 // Whether the script was found in a sysrooted directory. 1425 bool is_in_sysroot_; 1426 // If this is true, then if we find an OUTPUT_FORMAT with an 1427 // incompatible target, then we tell the parser to abort so that we 1428 // can search for the next file with the same name. 1429 bool skip_on_incompatible_target_; 1430 // True if we found an OUTPUT_FORMAT with an incompatible target. 1431 bool found_incompatible_target_; 1432 // May be NULL if the user chooses not to pass one in. 1433 Command_line* command_line_; 1434 // Options which may be set from any linker script. 1435 Script_options* script_options_; 1436 // Information parsed from a version script. 1437 Version_script_info* version_script_info_; 1438 // The lexer. 1439 Lex* lex_; 1440 // The line number of the last token returned by next_token. 1441 int lineno_; 1442 // The column number of the last token returned by next_token. 1443 int charpos_; 1444 // A stack of lexer modes. 1445 std::vector<Lex::Mode> lex_mode_stack_; 1446 // A stack of which extern/language block we're inside. Can be C++, 1447 // java, or empty for C. 1448 std::vector<Version_script_info::Language> language_stack_; 1449 // New input files found to add to the link. 1450 Input_arguments* inputs_; 1451 // Pointer to incremental linking info. 1452 Script_info* script_info_; 1453 }; 1454 1455 // FILE was found as an argument on the command line. Try to read it 1456 // as a script. Return true if the file was handled. 1457 1458 bool 1459 read_input_script(Workqueue* workqueue, Symbol_table* symtab, Layout* layout, 1460 Dirsearch* dirsearch, int dirindex, 1461 Input_objects* input_objects, Mapfile* mapfile, 1462 Input_group* input_group, 1463 const Input_argument* input_argument, 1464 Input_file* input_file, Task_token* next_blocker, 1465 bool* used_next_blocker) 1466 { 1467 *used_next_blocker = false; 1468 1469 std::string input_string; 1470 Lex::read_file(input_file, &input_string); 1471 1472 Lex lex(input_string.c_str(), input_string.length(), PARSING_LINKER_SCRIPT); 1473 1474 Script_info* script_info = NULL; 1475 if (layout->incremental_inputs() != NULL) 1476 { 1477 const std::string& filename = input_file->filename(); 1478 Timespec mtime = input_file->file().get_mtime(); 1479 unsigned int arg_serial = input_argument->file().arg_serial(); 1480 script_info = new Script_info(filename); 1481 layout->incremental_inputs()->report_script(script_info, arg_serial, 1482 mtime); 1483 } 1484 1485 Parser_closure closure(input_file->filename().c_str(), 1486 input_argument->file().options(), 1487 false, 1488 input_group != NULL, 1489 input_file->is_in_sysroot(), 1490 NULL, 1491 layout->script_options(), 1492 &lex, 1493 input_file->will_search_for(), 1494 script_info); 1495 1496 bool old_saw_sections_clause = 1497 layout->script_options()->saw_sections_clause(); 1498 1499 if (yyparse(&closure) != 0) 1500 { 1501 if (closure.found_incompatible_target()) 1502 { 1503 Read_symbols::incompatible_warning(input_argument, input_file); 1504 Read_symbols::requeue(workqueue, input_objects, symtab, layout, 1505 dirsearch, dirindex, mapfile, input_argument, 1506 input_group, next_blocker); 1507 return true; 1508 } 1509 return false; 1510 } 1511 1512 if (!old_saw_sections_clause 1513 && layout->script_options()->saw_sections_clause() 1514 && layout->have_added_input_section()) 1515 gold_error(_("%s: SECTIONS seen after other input files; try -T/--script"), 1516 input_file->filename().c_str()); 1517 1518 if (!closure.saw_inputs()) 1519 return true; 1520 1521 Task_token* this_blocker = NULL; 1522 for (Input_arguments::const_iterator p = closure.inputs()->begin(); 1523 p != closure.inputs()->end(); 1524 ++p) 1525 { 1526 Task_token* nb; 1527 if (p + 1 == closure.inputs()->end()) 1528 nb = next_blocker; 1529 else 1530 { 1531 nb = new Task_token(true); 1532 nb->add_blocker(); 1533 } 1534 workqueue->queue_soon(new Read_symbols(input_objects, symtab, 1535 layout, dirsearch, 0, mapfile, &*p, 1536 input_group, NULL, this_blocker, nb)); 1537 this_blocker = nb; 1538 } 1539 1540 *used_next_blocker = true; 1541 1542 return true; 1543 } 1544 1545 // Helper function for read_version_script(), read_commandline_script() and 1546 // script_include_directive(). Processes the given file in the mode indicated 1547 // by first_token and lex_mode. 1548 1549 static bool 1550 read_script_file(const char* filename, Command_line* cmdline, 1551 Script_options* script_options, 1552 int first_token, Lex::Mode lex_mode) 1553 { 1554 Dirsearch dirsearch; 1555 std::string name = filename; 1556 1557 // If filename is a relative filename, search for it manually using "." + 1558 // cmdline->options()->library_path() -- not dirsearch. 1559 if (!IS_ABSOLUTE_PATH(filename)) 1560 { 1561 const General_options::Dir_list& search_path = 1562 cmdline->options().library_path(); 1563 name = Dirsearch::find_file_in_dir_list(name, search_path, "."); 1564 } 1565 1566 // The file locking code wants to record a Task, but we haven't 1567 // started the workqueue yet. This is only for debugging purposes, 1568 // so we invent a fake value. 1569 const Task* task = reinterpret_cast<const Task*>(-1); 1570 1571 // We don't want this file to be opened in binary mode. 1572 Position_dependent_options posdep = cmdline->position_dependent_options(); 1573 if (posdep.format_enum() == General_options::OBJECT_FORMAT_BINARY) 1574 posdep.set_format_enum(General_options::OBJECT_FORMAT_ELF); 1575 Input_file_argument input_argument(name.c_str(), 1576 Input_file_argument::INPUT_FILE_TYPE_FILE, 1577 "", false, posdep); 1578 Input_file input_file(&input_argument); 1579 int dummy = 0; 1580 if (!input_file.open(dirsearch, task, &dummy)) 1581 return false; 1582 1583 std::string input_string; 1584 Lex::read_file(&input_file, &input_string); 1585 1586 Lex lex(input_string.c_str(), input_string.length(), first_token); 1587 lex.set_mode(lex_mode); 1588 1589 Parser_closure closure(filename, 1590 cmdline->position_dependent_options(), 1591 first_token == Lex::DYNAMIC_LIST, 1592 false, 1593 input_file.is_in_sysroot(), 1594 cmdline, 1595 script_options, 1596 &lex, 1597 false, 1598 NULL); 1599 if (yyparse(&closure) != 0) 1600 { 1601 input_file.file().unlock(task); 1602 return false; 1603 } 1604 1605 input_file.file().unlock(task); 1606 1607 gold_assert(!closure.saw_inputs()); 1608 1609 return true; 1610 } 1611 1612 // FILENAME was found as an argument to --script (-T). 1613 // Read it as a script, and execute its contents immediately. 1614 1615 bool 1616 read_commandline_script(const char* filename, Command_line* cmdline) 1617 { 1618 return read_script_file(filename, cmdline, &cmdline->script_options(), 1619 PARSING_LINKER_SCRIPT, Lex::LINKER_SCRIPT); 1620 } 1621 1622 // FILENAME was found as an argument to --version-script. Read it as 1623 // a version script, and store its contents in 1624 // cmdline->script_options()->version_script_info(). 1625 1626 bool 1627 read_version_script(const char* filename, Command_line* cmdline) 1628 { 1629 return read_script_file(filename, cmdline, &cmdline->script_options(), 1630 PARSING_VERSION_SCRIPT, Lex::VERSION_SCRIPT); 1631 } 1632 1633 // FILENAME was found as an argument to --dynamic-list. Read it as a 1634 // list of symbols, and store its contents in DYNAMIC_LIST. 1635 1636 bool 1637 read_dynamic_list(const char* filename, Command_line* cmdline, 1638 Script_options* dynamic_list) 1639 { 1640 return read_script_file(filename, cmdline, dynamic_list, 1641 PARSING_DYNAMIC_LIST, Lex::DYNAMIC_LIST); 1642 } 1643 1644 // Implement the --defsym option on the command line. Return true if 1645 // all is well. 1646 1647 bool 1648 Script_options::define_symbol(const char* definition) 1649 { 1650 Lex lex(definition, strlen(definition), PARSING_DEFSYM); 1651 lex.set_mode(Lex::EXPRESSION); 1652 1653 // Dummy value. 1654 Position_dependent_options posdep_options; 1655 1656 Parser_closure closure("command line", posdep_options, true, 1657 false, false, NULL, this, &lex, false, NULL); 1658 1659 if (yyparse(&closure) != 0) 1660 return false; 1661 1662 gold_assert(!closure.saw_inputs()); 1663 1664 return true; 1665 } 1666 1667 // Print the script to F for debugging. 1668 1669 void 1670 Script_options::print(FILE* f) const 1671 { 1672 fprintf(f, "%s: Dumping linker script\n", program_name); 1673 1674 if (!this->entry_.empty()) 1675 fprintf(f, "ENTRY(%s)\n", this->entry_.c_str()); 1676 1677 for (Symbol_assignments::const_iterator p = 1678 this->symbol_assignments_.begin(); 1679 p != this->symbol_assignments_.end(); 1680 ++p) 1681 (*p)->print(f); 1682 1683 for (Assertions::const_iterator p = this->assertions_.begin(); 1684 p != this->assertions_.end(); 1685 ++p) 1686 (*p)->print(f); 1687 1688 this->script_sections_.print(f); 1689 1690 this->version_script_info_.print(f); 1691 } 1692 1693 // Manage mapping from keywords to the codes expected by the bison 1694 // parser. We construct one global object for each lex mode with 1695 // keywords. 1696 1697 class Keyword_to_parsecode 1698 { 1699 public: 1700 // The structure which maps keywords to parsecodes. 1701 struct Keyword_parsecode 1702 { 1703 // Keyword. 1704 const char* keyword; 1705 // Corresponding parsecode. 1706 int parsecode; 1707 }; 1708 1709 Keyword_to_parsecode(const Keyword_parsecode* keywords, 1710 int keyword_count) 1711 : keyword_parsecodes_(keywords), keyword_count_(keyword_count) 1712 { } 1713 1714 // Return the parsecode corresponding KEYWORD, or 0 if it is not a 1715 // keyword. 1716 int 1717 keyword_to_parsecode(const char* keyword, size_t len) const; 1718 1719 private: 1720 const Keyword_parsecode* keyword_parsecodes_; 1721 const int keyword_count_; 1722 }; 1723 1724 // Mapping from keyword string to keyword parsecode. This array must 1725 // be kept in sorted order. Parsecodes are looked up using bsearch. 1726 // This array must correspond to the list of parsecodes in yyscript.y. 1727 1728 static const Keyword_to_parsecode::Keyword_parsecode 1729 script_keyword_parsecodes[] = 1730 { 1731 { "ABSOLUTE", ABSOLUTE }, 1732 { "ADDR", ADDR }, 1733 { "ALIGN", ALIGN_K }, 1734 { "ALIGNOF", ALIGNOF }, 1735 { "ASSERT", ASSERT_K }, 1736 { "AS_NEEDED", AS_NEEDED }, 1737 { "AT", AT }, 1738 { "BIND", BIND }, 1739 { "BLOCK", BLOCK }, 1740 { "BYTE", BYTE }, 1741 { "CONSTANT", CONSTANT }, 1742 { "CONSTRUCTORS", CONSTRUCTORS }, 1743 { "COPY", COPY }, 1744 { "CREATE_OBJECT_SYMBOLS", CREATE_OBJECT_SYMBOLS }, 1745 { "DATA_SEGMENT_ALIGN", DATA_SEGMENT_ALIGN }, 1746 { "DATA_SEGMENT_END", DATA_SEGMENT_END }, 1747 { "DATA_SEGMENT_RELRO_END", DATA_SEGMENT_RELRO_END }, 1748 { "DEFINED", DEFINED }, 1749 { "DSECT", DSECT }, 1750 { "ENTRY", ENTRY }, 1751 { "EXCLUDE_FILE", EXCLUDE_FILE }, 1752 { "EXTERN", EXTERN }, 1753 { "FILL", FILL }, 1754 { "FLOAT", FLOAT }, 1755 { "FORCE_COMMON_ALLOCATION", FORCE_COMMON_ALLOCATION }, 1756 { "GROUP", GROUP }, 1757 { "HLL", HLL }, 1758 { "INCLUDE", INCLUDE }, 1759 { "INFO", INFO }, 1760 { "INHIBIT_COMMON_ALLOCATION", INHIBIT_COMMON_ALLOCATION }, 1761 { "INPUT", INPUT }, 1762 { "KEEP", KEEP }, 1763 { "LENGTH", LENGTH }, 1764 { "LOADADDR", LOADADDR }, 1765 { "LONG", LONG }, 1766 { "MAP", MAP }, 1767 { "MAX", MAX_K }, 1768 { "MEMORY", MEMORY }, 1769 { "MIN", MIN_K }, 1770 { "NEXT", NEXT }, 1771 { "NOCROSSREFS", NOCROSSREFS }, 1772 { "NOFLOAT", NOFLOAT }, 1773 { "NOLOAD", NOLOAD }, 1774 { "ONLY_IF_RO", ONLY_IF_RO }, 1775 { "ONLY_IF_RW", ONLY_IF_RW }, 1776 { "OPTION", OPTION }, 1777 { "ORIGIN", ORIGIN }, 1778 { "OUTPUT", OUTPUT }, 1779 { "OUTPUT_ARCH", OUTPUT_ARCH }, 1780 { "OUTPUT_FORMAT", OUTPUT_FORMAT }, 1781 { "OVERLAY", OVERLAY }, 1782 { "PHDRS", PHDRS }, 1783 { "PROVIDE", PROVIDE }, 1784 { "PROVIDE_HIDDEN", PROVIDE_HIDDEN }, 1785 { "QUAD", QUAD }, 1786 { "SEARCH_DIR", SEARCH_DIR }, 1787 { "SECTIONS", SECTIONS }, 1788 { "SEGMENT_START", SEGMENT_START }, 1789 { "SHORT", SHORT }, 1790 { "SIZEOF", SIZEOF }, 1791 { "SIZEOF_HEADERS", SIZEOF_HEADERS }, 1792 { "SORT", SORT_BY_NAME }, 1793 { "SORT_BY_ALIGNMENT", SORT_BY_ALIGNMENT }, 1794 { "SORT_BY_NAME", SORT_BY_NAME }, 1795 { "SPECIAL", SPECIAL }, 1796 { "SQUAD", SQUAD }, 1797 { "STARTUP", STARTUP }, 1798 { "SUBALIGN", SUBALIGN }, 1799 { "SYSLIB", SYSLIB }, 1800 { "TARGET", TARGET_K }, 1801 { "TRUNCATE", TRUNCATE }, 1802 { "VERSION", VERSIONK }, 1803 { "global", GLOBAL }, 1804 { "l", LENGTH }, 1805 { "len", LENGTH }, 1806 { "local", LOCAL }, 1807 { "o", ORIGIN }, 1808 { "org", ORIGIN }, 1809 { "sizeof_headers", SIZEOF_HEADERS }, 1810 }; 1811 1812 static const Keyword_to_parsecode 1813 script_keywords(&script_keyword_parsecodes[0], 1814 (sizeof(script_keyword_parsecodes) 1815 / sizeof(script_keyword_parsecodes[0]))); 1816 1817 static const Keyword_to_parsecode::Keyword_parsecode 1818 version_script_keyword_parsecodes[] = 1819 { 1820 { "extern", EXTERN }, 1821 { "global", GLOBAL }, 1822 { "local", LOCAL }, 1823 }; 1824 1825 static const Keyword_to_parsecode 1826 version_script_keywords(&version_script_keyword_parsecodes[0], 1827 (sizeof(version_script_keyword_parsecodes) 1828 / sizeof(version_script_keyword_parsecodes[0]))); 1829 1830 static const Keyword_to_parsecode::Keyword_parsecode 1831 dynamic_list_keyword_parsecodes[] = 1832 { 1833 { "extern", EXTERN }, 1834 }; 1835 1836 static const Keyword_to_parsecode 1837 dynamic_list_keywords(&dynamic_list_keyword_parsecodes[0], 1838 (sizeof(dynamic_list_keyword_parsecodes) 1839 / sizeof(dynamic_list_keyword_parsecodes[0]))); 1840 1841 1842 1843 // Comparison function passed to bsearch. 1844 1845 extern "C" 1846 { 1847 1848 struct Ktt_key 1849 { 1850 const char* str; 1851 size_t len; 1852 }; 1853 1854 static int 1855 ktt_compare(const void* keyv, const void* kttv) 1856 { 1857 const Ktt_key* key = static_cast<const Ktt_key*>(keyv); 1858 const Keyword_to_parsecode::Keyword_parsecode* ktt = 1859 static_cast<const Keyword_to_parsecode::Keyword_parsecode*>(kttv); 1860 int i = strncmp(key->str, ktt->keyword, key->len); 1861 if (i != 0) 1862 return i; 1863 if (ktt->keyword[key->len] != '\0') 1864 return -1; 1865 return 0; 1866 } 1867 1868 } // End extern "C". 1869 1870 int 1871 Keyword_to_parsecode::keyword_to_parsecode(const char* keyword, 1872 size_t len) const 1873 { 1874 Ktt_key key; 1875 key.str = keyword; 1876 key.len = len; 1877 void* kttv = bsearch(&key, 1878 this->keyword_parsecodes_, 1879 this->keyword_count_, 1880 sizeof(this->keyword_parsecodes_[0]), 1881 ktt_compare); 1882 if (kttv == NULL) 1883 return 0; 1884 Keyword_parsecode* ktt = static_cast<Keyword_parsecode*>(kttv); 1885 return ktt->parsecode; 1886 } 1887 1888 // The following structs are used within the VersionInfo class as well 1889 // as in the bison helper functions. They store the information 1890 // parsed from the version script. 1891 1892 // A single version expression. 1893 // For example, pattern="std::map*" and language="C++". 1894 struct Version_expression 1895 { 1896 Version_expression(const std::string& a_pattern, 1897 Version_script_info::Language a_language, 1898 bool a_exact_match) 1899 : pattern(a_pattern), language(a_language), exact_match(a_exact_match), 1900 was_matched_by_symbol(false) 1901 { } 1902 1903 std::string pattern; 1904 Version_script_info::Language language; 1905 // If false, we use glob() to match pattern. If true, we use strcmp(). 1906 bool exact_match; 1907 // True if --no-undefined-version is in effect and we found this 1908 // version in get_symbol_version. We use mutable because this 1909 // struct is generally not modifiable after it has been created. 1910 mutable bool was_matched_by_symbol; 1911 }; 1912 1913 // A list of expressions. 1914 struct Version_expression_list 1915 { 1916 std::vector<struct Version_expression> expressions; 1917 }; 1918 1919 // A list of which versions upon which another version depends. 1920 // Strings should be from the Stringpool. 1921 struct Version_dependency_list 1922 { 1923 std::vector<std::string> dependencies; 1924 }; 1925 1926 // The total definition of a version. It includes the tag for the 1927 // version, its global and local expressions, and any dependencies. 1928 struct Version_tree 1929 { 1930 Version_tree() 1931 : tag(), global(NULL), local(NULL), dependencies(NULL) 1932 { } 1933 1934 std::string tag; 1935 const struct Version_expression_list* global; 1936 const struct Version_expression_list* local; 1937 const struct Version_dependency_list* dependencies; 1938 }; 1939 1940 // Helper class that calls cplus_demangle when needed and takes care of freeing 1941 // the result. 1942 1943 class Lazy_demangler 1944 { 1945 public: 1946 Lazy_demangler(const char* symbol, int options) 1947 : symbol_(symbol), options_(options), demangled_(NULL), did_demangle_(false) 1948 { } 1949 1950 ~Lazy_demangler() 1951 { free(this->demangled_); } 1952 1953 // Return the demangled name. The actual demangling happens on the first call, 1954 // and the result is later cached. 1955 inline char* 1956 get(); 1957 1958 private: 1959 // The symbol to demangle. 1960 const char* symbol_; 1961 // Option flags to pass to cplus_demagle. 1962 const int options_; 1963 // The cached demangled value, or NULL if demangling didn't happen yet or 1964 // failed. 1965 char* demangled_; 1966 // Whether we already called cplus_demangle 1967 bool did_demangle_; 1968 }; 1969 1970 // Return the demangled name. The actual demangling happens on the first call, 1971 // and the result is later cached. Returns NULL if the symbol cannot be 1972 // demangled. 1973 1974 inline char* 1975 Lazy_demangler::get() 1976 { 1977 if (!this->did_demangle_) 1978 { 1979 this->demangled_ = cplus_demangle(this->symbol_, this->options_); 1980 this->did_demangle_ = true; 1981 } 1982 return this->demangled_; 1983 } 1984 1985 // Class Version_script_info. 1986 1987 Version_script_info::Version_script_info() 1988 : dependency_lists_(), expression_lists_(), version_trees_(), globs_(), 1989 default_version_(NULL), default_is_global_(false), is_finalized_(false) 1990 { 1991 for (int i = 0; i < LANGUAGE_COUNT; ++i) 1992 this->exact_[i] = NULL; 1993 } 1994 1995 Version_script_info::~Version_script_info() 1996 { 1997 } 1998 1999 // Forget all the known version script information. 2000 2001 void 2002 Version_script_info::clear() 2003 { 2004 for (size_t k = 0; k < this->dependency_lists_.size(); ++k) 2005 delete this->dependency_lists_[k]; 2006 this->dependency_lists_.clear(); 2007 for (size_t k = 0; k < this->version_trees_.size(); ++k) 2008 delete this->version_trees_[k]; 2009 this->version_trees_.clear(); 2010 for (size_t k = 0; k < this->expression_lists_.size(); ++k) 2011 delete this->expression_lists_[k]; 2012 this->expression_lists_.clear(); 2013 } 2014 2015 // Finalize the version script information. 2016 2017 void 2018 Version_script_info::finalize() 2019 { 2020 if (!this->is_finalized_) 2021 { 2022 this->build_lookup_tables(); 2023 this->is_finalized_ = true; 2024 } 2025 } 2026 2027 // Return all the versions. 2028 2029 std::vector<std::string> 2030 Version_script_info::get_versions() const 2031 { 2032 std::vector<std::string> ret; 2033 for (size_t j = 0; j < this->version_trees_.size(); ++j) 2034 if (!this->version_trees_[j]->tag.empty()) 2035 ret.push_back(this->version_trees_[j]->tag); 2036 return ret; 2037 } 2038 2039 // Return the dependencies of VERSION. 2040 2041 std::vector<std::string> 2042 Version_script_info::get_dependencies(const char* version) const 2043 { 2044 std::vector<std::string> ret; 2045 for (size_t j = 0; j < this->version_trees_.size(); ++j) 2046 if (this->version_trees_[j]->tag == version) 2047 { 2048 const struct Version_dependency_list* deps = 2049 this->version_trees_[j]->dependencies; 2050 if (deps != NULL) 2051 for (size_t k = 0; k < deps->dependencies.size(); ++k) 2052 ret.push_back(deps->dependencies[k]); 2053 return ret; 2054 } 2055 return ret; 2056 } 2057 2058 // A version script essentially maps a symbol name to a version tag 2059 // and an indication of whether symbol is global or local within that 2060 // version tag. Each symbol maps to at most one version tag. 2061 // Unfortunately, in practice, version scripts are ambiguous, and list 2062 // symbols multiple times. Thus, we have to document the matching 2063 // process. 2064 2065 // This is a description of what the GNU linker does as of 2010-01-11. 2066 // It walks through the version tags in the order in which they appear 2067 // in the version script. For each tag, it first walks through the 2068 // global patterns for that tag, then the local patterns. When 2069 // looking at a single pattern, it first applies any language specific 2070 // demangling as specified for the pattern, and then matches the 2071 // resulting symbol name to the pattern. If it finds an exact match 2072 // for a literal pattern (a pattern enclosed in quotes or with no 2073 // wildcard characters), then that is the match that it uses. If 2074 // finds a match with a wildcard pattern, then it saves it and 2075 // continues searching. Wildcard patterns that are exactly "*" are 2076 // saved separately. 2077 2078 // If no exact match with a literal pattern is ever found, then if a 2079 // wildcard match with a global pattern was found it is used, 2080 // otherwise if a wildcard match with a local pattern was found it is 2081 // used. 2082 2083 // This is the result: 2084 // * If there is an exact match, then we use the first tag in the 2085 // version script where it matches. 2086 // + If the exact match in that tag is global, it is used. 2087 // + Otherwise the exact match in that tag is local, and is used. 2088 // * Otherwise, if there is any match with a global 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 "*". If there is no match with a local 2093 // wildcard pattern which is not "*", then we use the *last* 2094 // match with a global "*". Otherwise, continue. 2095 // * Otherwise, if there is any match with a local wildcard pattern: 2096 // + If there is any match with a wildcard pattern which is not 2097 // "*", then we use the tag in which the *last* such pattern 2098 // appears. 2099 // + Otherwise, we matched "*", and we use the tag in which the 2100 // *last* such match occurred. 2101 2102 // There is an additional wrinkle. When the GNU linker finds a symbol 2103 // with a version defined in an object file due to a .symver 2104 // directive, it looks up that symbol name in that version tag. If it 2105 // finds it, it matches the symbol name against the patterns for that 2106 // version. If there is no match with a global pattern, but there is 2107 // a match with a local pattern, then the GNU linker marks the symbol 2108 // as local. 2109 2110 // We want gold to be generally compatible, but we also want gold to 2111 // be fast. These are the rules that gold implements: 2112 // * If there is an exact match for the mangled name, we use it. 2113 // + If there is more than one exact match, we give a warning, and 2114 // we use the first tag in the script which matches. 2115 // + If a symbol has an exact match as both global and local for 2116 // the same version tag, we give an error. 2117 // * Otherwise, we look for an extern C++ or an extern Java exact 2118 // match. If we find an exact match, we use it. 2119 // + If there is more than one exact match, we give a warning, and 2120 // we use the first tag in the script which matches. 2121 // + If a symbol has an exact match as both global and local for 2122 // the same version tag, we give an error. 2123 // * Otherwise, we look through the wildcard patterns, ignoring "*" 2124 // patterns. We look through the version tags in reverse order. 2125 // For each version tag, we look through the global patterns and 2126 // then the local patterns. We use the first match we find (i.e., 2127 // the last matching version tag in the file). 2128 // * Otherwise, we use the "*" pattern if there is one. We give an 2129 // error if there are multiple "*" patterns. 2130 2131 // At least for now, gold does not look up the version tag for a 2132 // symbol version found in an object file to see if it should be 2133 // forced local. There are other ways to force a symbol to be local, 2134 // and I don't understand why this one is useful. 2135 2136 // Build a set of fast lookup tables for a version script. 2137 2138 void 2139 Version_script_info::build_lookup_tables() 2140 { 2141 size_t size = this->version_trees_.size(); 2142 for (size_t j = 0; j < size; ++j) 2143 { 2144 const Version_tree* v = this->version_trees_[j]; 2145 this->build_expression_list_lookup(v->local, v, false); 2146 this->build_expression_list_lookup(v->global, v, true); 2147 } 2148 } 2149 2150 // If a pattern has backlashes but no unquoted wildcard characters, 2151 // then we apply backslash unquoting and look for an exact match. 2152 // Otherwise we treat it as a wildcard pattern. This function returns 2153 // true for a wildcard pattern. Otherwise, it does backslash 2154 // unquoting on *PATTERN and returns false. If this returns true, 2155 // *PATTERN may have been partially unquoted. 2156 2157 bool 2158 Version_script_info::unquote(std::string* pattern) const 2159 { 2160 bool saw_backslash = false; 2161 size_t len = pattern->length(); 2162 size_t j = 0; 2163 for (size_t i = 0; i < len; ++i) 2164 { 2165 if (saw_backslash) 2166 saw_backslash = false; 2167 else 2168 { 2169 switch ((*pattern)[i]) 2170 { 2171 case '?': case '[': case '*': 2172 return true; 2173 case '\\': 2174 saw_backslash = true; 2175 continue; 2176 default: 2177 break; 2178 } 2179 } 2180 2181 if (i != j) 2182 (*pattern)[j] = (*pattern)[i]; 2183 ++j; 2184 } 2185 return false; 2186 } 2187 2188 // Add an exact match for MATCH to *PE. The result of the match is 2189 // V/IS_GLOBAL. 2190 2191 void 2192 Version_script_info::add_exact_match(const std::string& match, 2193 const Version_tree* v, bool is_global, 2194 const Version_expression* ve, 2195 Exact* pe) 2196 { 2197 std::pair<Exact::iterator, bool> ins = 2198 pe->insert(std::make_pair(match, Version_tree_match(v, is_global, ve))); 2199 if (ins.second) 2200 { 2201 // This is the first time we have seen this match. 2202 return; 2203 } 2204 2205 Version_tree_match& vtm(ins.first->second); 2206 if (vtm.real->tag != v->tag) 2207 { 2208 // This is an ambiguous match. We still return the 2209 // first version that we found in the script, but we 2210 // record the new version to issue a warning if we 2211 // wind up looking up this symbol. 2212 if (vtm.ambiguous == NULL) 2213 vtm.ambiguous = v; 2214 } 2215 else if (is_global != vtm.is_global) 2216 { 2217 // We have a match for both the global and local entries for a 2218 // version tag. That's got to be wrong. 2219 gold_error(_("'%s' appears as both a global and a local symbol " 2220 "for version '%s' in script"), 2221 match.c_str(), v->tag.c_str()); 2222 } 2223 } 2224 2225 // Build fast lookup information for EXPLIST and store it in LOOKUP. 2226 // All matches go to V, and IS_GLOBAL is true if they are global 2227 // matches. 2228 2229 void 2230 Version_script_info::build_expression_list_lookup( 2231 const Version_expression_list* explist, 2232 const Version_tree* v, 2233 bool is_global) 2234 { 2235 if (explist == NULL) 2236 return; 2237 size_t size = explist->expressions.size(); 2238 for (size_t i = 0; i < size; ++i) 2239 { 2240 const Version_expression& exp(explist->expressions[i]); 2241 2242 if (exp.pattern.length() == 1 && exp.pattern[0] == '*') 2243 { 2244 if (this->default_version_ != NULL 2245 && this->default_version_->tag != v->tag) 2246 gold_warning(_("wildcard match appears in both version '%s' " 2247 "and '%s' in script"), 2248 this->default_version_->tag.c_str(), v->tag.c_str()); 2249 else if (this->default_version_ != NULL 2250 && this->default_is_global_ != is_global) 2251 gold_error(_("wildcard match appears as both global and local " 2252 "in version '%s' in script"), 2253 v->tag.c_str()); 2254 this->default_version_ = v; 2255 this->default_is_global_ = is_global; 2256 continue; 2257 } 2258 2259 std::string pattern = exp.pattern; 2260 if (!exp.exact_match) 2261 { 2262 if (this->unquote(&pattern)) 2263 { 2264 this->globs_.push_back(Glob(&exp, v, is_global)); 2265 continue; 2266 } 2267 } 2268 2269 if (this->exact_[exp.language] == NULL) 2270 this->exact_[exp.language] = new Exact(); 2271 this->add_exact_match(pattern, v, is_global, &exp, 2272 this->exact_[exp.language]); 2273 } 2274 } 2275 2276 // Return the name to match given a name, a language code, and two 2277 // lazy demanglers. 2278 2279 const char* 2280 Version_script_info::get_name_to_match(const char* name, 2281 int language, 2282 Lazy_demangler* cpp_demangler, 2283 Lazy_demangler* java_demangler) const 2284 { 2285 switch (language) 2286 { 2287 case LANGUAGE_C: 2288 return name; 2289 case LANGUAGE_CXX: 2290 return cpp_demangler->get(); 2291 case LANGUAGE_JAVA: 2292 return java_demangler->get(); 2293 default: 2294 gold_unreachable(); 2295 } 2296 } 2297 2298 // Look up SYMBOL_NAME in the list of versions. Return true if the 2299 // symbol is found, false if not. If the symbol is found, then if 2300 // PVERSION is not NULL, set *PVERSION to the version tag, and if 2301 // P_IS_GLOBAL is not NULL, set *P_IS_GLOBAL according to whether the 2302 // symbol is global or not. 2303 2304 bool 2305 Version_script_info::get_symbol_version(const char* symbol_name, 2306 std::string* pversion, 2307 bool* p_is_global) const 2308 { 2309 Lazy_demangler cpp_demangled_name(symbol_name, DMGL_ANSI | DMGL_PARAMS); 2310 Lazy_demangler java_demangled_name(symbol_name, 2311 DMGL_ANSI | DMGL_PARAMS | DMGL_JAVA); 2312 2313 gold_assert(this->is_finalized_); 2314 for (int i = 0; i < LANGUAGE_COUNT; ++i) 2315 { 2316 Exact* exact = this->exact_[i]; 2317 if (exact == NULL) 2318 continue; 2319 2320 const char* name_to_match = this->get_name_to_match(symbol_name, i, 2321 &cpp_demangled_name, 2322 &java_demangled_name); 2323 if (name_to_match == NULL) 2324 { 2325 // If the name can not be demangled, the GNU linker goes 2326 // ahead and tries to match it anyhow. That does not 2327 // make sense to me and I have not implemented it. 2328 continue; 2329 } 2330 2331 Exact::const_iterator pe = exact->find(name_to_match); 2332 if (pe != exact->end()) 2333 { 2334 const Version_tree_match& vtm(pe->second); 2335 if (vtm.ambiguous != NULL) 2336 gold_warning(_("using '%s' as version for '%s' which is also " 2337 "named in version '%s' in script"), 2338 vtm.real->tag.c_str(), name_to_match, 2339 vtm.ambiguous->tag.c_str()); 2340 2341 if (pversion != NULL) 2342 *pversion = vtm.real->tag; 2343 if (p_is_global != NULL) 2344 *p_is_global = vtm.is_global; 2345 2346 // If we are using --no-undefined-version, and this is a 2347 // global symbol, we have to record that we have found this 2348 // symbol, so that we don't warn about it. We have to do 2349 // this now, because otherwise we have no way to get from a 2350 // non-C language back to the demangled name that we 2351 // matched. 2352 if (p_is_global != NULL && vtm.is_global) 2353 vtm.expression->was_matched_by_symbol = true; 2354 2355 return true; 2356 } 2357 } 2358 2359 // Look through the glob patterns in reverse order. 2360 2361 for (Globs::const_reverse_iterator p = this->globs_.rbegin(); 2362 p != this->globs_.rend(); 2363 ++p) 2364 { 2365 int language = p->expression->language; 2366 const char* name_to_match = this->get_name_to_match(symbol_name, 2367 language, 2368 &cpp_demangled_name, 2369 &java_demangled_name); 2370 if (name_to_match == NULL) 2371 continue; 2372 2373 if (fnmatch(p->expression->pattern.c_str(), name_to_match, 2374 FNM_NOESCAPE) == 0) 2375 { 2376 if (pversion != NULL) 2377 *pversion = p->version->tag; 2378 if (p_is_global != NULL) 2379 *p_is_global = p->is_global; 2380 return true; 2381 } 2382 } 2383 2384 // Finally, there may be a wildcard. 2385 if (this->default_version_ != NULL) 2386 { 2387 if (pversion != NULL) 2388 *pversion = this->default_version_->tag; 2389 if (p_is_global != NULL) 2390 *p_is_global = this->default_is_global_; 2391 return true; 2392 } 2393 2394 return false; 2395 } 2396 2397 // Give an error if any exact symbol names (not wildcards) appear in a 2398 // version script, but there is no such symbol. 2399 2400 void 2401 Version_script_info::check_unmatched_names(const Symbol_table* symtab) const 2402 { 2403 for (size_t i = 0; i < this->version_trees_.size(); ++i) 2404 { 2405 const Version_tree* vt = this->version_trees_[i]; 2406 if (vt->global == NULL) 2407 continue; 2408 for (size_t j = 0; j < vt->global->expressions.size(); ++j) 2409 { 2410 const Version_expression& expression(vt->global->expressions[j]); 2411 2412 // Ignore cases where we used the version because we saw a 2413 // symbol that we looked up. Note that 2414 // WAS_MATCHED_BY_SYMBOL will be true even if the symbol was 2415 // not a definition. That's OK as in that case we most 2416 // likely gave an undefined symbol error anyhow. 2417 if (expression.was_matched_by_symbol) 2418 continue; 2419 2420 // Just ignore names which are in languages other than C. 2421 // We have no way to look them up in the symbol table. 2422 if (expression.language != LANGUAGE_C) 2423 continue; 2424 2425 // Remove backslash quoting, and ignore wildcard patterns. 2426 std::string pattern = expression.pattern; 2427 if (!expression.exact_match) 2428 { 2429 if (this->unquote(&pattern)) 2430 continue; 2431 } 2432 2433 if (symtab->lookup(pattern.c_str(), vt->tag.c_str()) == NULL) 2434 gold_error(_("version script assignment of %s to symbol %s " 2435 "failed: symbol not defined"), 2436 vt->tag.c_str(), pattern.c_str()); 2437 } 2438 } 2439 } 2440 2441 struct Version_dependency_list* 2442 Version_script_info::allocate_dependency_list() 2443 { 2444 dependency_lists_.push_back(new Version_dependency_list); 2445 return dependency_lists_.back(); 2446 } 2447 2448 struct Version_expression_list* 2449 Version_script_info::allocate_expression_list() 2450 { 2451 expression_lists_.push_back(new Version_expression_list); 2452 return expression_lists_.back(); 2453 } 2454 2455 struct Version_tree* 2456 Version_script_info::allocate_version_tree() 2457 { 2458 version_trees_.push_back(new Version_tree); 2459 return version_trees_.back(); 2460 } 2461 2462 // Print for debugging. 2463 2464 void 2465 Version_script_info::print(FILE* f) const 2466 { 2467 if (this->empty()) 2468 return; 2469 2470 fprintf(f, "VERSION {"); 2471 2472 for (size_t i = 0; i < this->version_trees_.size(); ++i) 2473 { 2474 const Version_tree* vt = this->version_trees_[i]; 2475 2476 if (vt->tag.empty()) 2477 fprintf(f, " {\n"); 2478 else 2479 fprintf(f, " %s {\n", vt->tag.c_str()); 2480 2481 if (vt->global != NULL) 2482 { 2483 fprintf(f, " global :\n"); 2484 this->print_expression_list(f, vt->global); 2485 } 2486 2487 if (vt->local != NULL) 2488 { 2489 fprintf(f, " local :\n"); 2490 this->print_expression_list(f, vt->local); 2491 } 2492 2493 fprintf(f, " }"); 2494 if (vt->dependencies != NULL) 2495 { 2496 const Version_dependency_list* deps = vt->dependencies; 2497 for (size_t j = 0; j < deps->dependencies.size(); ++j) 2498 { 2499 if (j < deps->dependencies.size() - 1) 2500 fprintf(f, "\n"); 2501 fprintf(f, " %s", deps->dependencies[j].c_str()); 2502 } 2503 } 2504 fprintf(f, ";\n"); 2505 } 2506 2507 fprintf(f, "}\n"); 2508 } 2509 2510 void 2511 Version_script_info::print_expression_list( 2512 FILE* f, 2513 const Version_expression_list* vel) const 2514 { 2515 Version_script_info::Language current_language = LANGUAGE_C; 2516 for (size_t i = 0; i < vel->expressions.size(); ++i) 2517 { 2518 const Version_expression& ve(vel->expressions[i]); 2519 2520 if (ve.language != current_language) 2521 { 2522 if (current_language != LANGUAGE_C) 2523 fprintf(f, " }\n"); 2524 switch (ve.language) 2525 { 2526 case LANGUAGE_C: 2527 break; 2528 case LANGUAGE_CXX: 2529 fprintf(f, " extern \"C++\" {\n"); 2530 break; 2531 case LANGUAGE_JAVA: 2532 fprintf(f, " extern \"Java\" {\n"); 2533 break; 2534 default: 2535 gold_unreachable(); 2536 } 2537 current_language = ve.language; 2538 } 2539 2540 fprintf(f, " "); 2541 if (current_language != LANGUAGE_C) 2542 fprintf(f, " "); 2543 2544 if (ve.exact_match) 2545 fprintf(f, "\""); 2546 fprintf(f, "%s", ve.pattern.c_str()); 2547 if (ve.exact_match) 2548 fprintf(f, "\""); 2549 2550 fprintf(f, "\n"); 2551 } 2552 2553 if (current_language != LANGUAGE_C) 2554 fprintf(f, " }\n"); 2555 } 2556 2557 } // End namespace gold. 2558 2559 // The remaining functions are extern "C", so it's clearer to not put 2560 // them in namespace gold. 2561 2562 using namespace gold; 2563 2564 // This function is called by the bison parser to return the next 2565 // token. 2566 2567 extern "C" int 2568 yylex(YYSTYPE* lvalp, void* closurev) 2569 { 2570 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2571 const Token* token = closure->next_token(); 2572 switch (token->classification()) 2573 { 2574 default: 2575 gold_unreachable(); 2576 2577 case Token::TOKEN_INVALID: 2578 yyerror(closurev, "invalid character"); 2579 return 0; 2580 2581 case Token::TOKEN_EOF: 2582 return 0; 2583 2584 case Token::TOKEN_STRING: 2585 { 2586 // This is either a keyword or a STRING. 2587 size_t len; 2588 const char* str = token->string_value(&len); 2589 int parsecode = 0; 2590 switch (closure->lex_mode()) 2591 { 2592 case Lex::LINKER_SCRIPT: 2593 parsecode = script_keywords.keyword_to_parsecode(str, len); 2594 break; 2595 case Lex::VERSION_SCRIPT: 2596 parsecode = version_script_keywords.keyword_to_parsecode(str, len); 2597 break; 2598 case Lex::DYNAMIC_LIST: 2599 parsecode = dynamic_list_keywords.keyword_to_parsecode(str, len); 2600 break; 2601 default: 2602 break; 2603 } 2604 if (parsecode != 0) 2605 return parsecode; 2606 lvalp->string.value = str; 2607 lvalp->string.length = len; 2608 return STRING; 2609 } 2610 2611 case Token::TOKEN_QUOTED_STRING: 2612 lvalp->string.value = token->string_value(&lvalp->string.length); 2613 return QUOTED_STRING; 2614 2615 case Token::TOKEN_OPERATOR: 2616 return token->operator_value(); 2617 2618 case Token::TOKEN_INTEGER: 2619 lvalp->integer = token->integer_value(); 2620 return INTEGER; 2621 } 2622 } 2623 2624 // This function is called by the bison parser to report an error. 2625 2626 extern "C" void 2627 yyerror(void* closurev, const char* message) 2628 { 2629 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2630 gold_error(_("%s:%d:%d: %s"), closure->filename(), closure->lineno(), 2631 closure->charpos(), message); 2632 } 2633 2634 // Called by the bison parser to add an external symbol to the link. 2635 2636 extern "C" void 2637 script_add_extern(void* closurev, const char* name, size_t length) 2638 { 2639 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2640 closure->script_options()->add_symbol_reference(name, length); 2641 } 2642 2643 // Called by the bison parser to add a file to the link. 2644 2645 extern "C" void 2646 script_add_file(void* closurev, const char* name, size_t length) 2647 { 2648 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2649 2650 // If this is an absolute path, and we found the script in the 2651 // sysroot, then we want to prepend the sysroot to the file name. 2652 // For example, this is how we handle a cross link to the x86_64 2653 // libc.so, which refers to /lib/libc.so.6. 2654 std::string name_string(name, length); 2655 const char* extra_search_path = "."; 2656 std::string script_directory; 2657 if (IS_ABSOLUTE_PATH(name_string.c_str())) 2658 { 2659 if (closure->is_in_sysroot()) 2660 { 2661 const std::string& sysroot(parameters->options().sysroot()); 2662 gold_assert(!sysroot.empty()); 2663 name_string = sysroot + name_string; 2664 } 2665 } 2666 else 2667 { 2668 // In addition to checking the normal library search path, we 2669 // also want to check in the script-directory. 2670 const char* slash = strrchr(closure->filename(), '/'); 2671 if (slash != NULL) 2672 { 2673 script_directory.assign(closure->filename(), 2674 slash - closure->filename() + 1); 2675 extra_search_path = script_directory.c_str(); 2676 } 2677 } 2678 2679 Input_file_argument file(name_string.c_str(), 2680 Input_file_argument::INPUT_FILE_TYPE_FILE, 2681 extra_search_path, false, 2682 closure->position_dependent_options()); 2683 Input_argument& arg = closure->inputs()->add_file(file); 2684 arg.set_script_info(closure->script_info()); 2685 } 2686 2687 // Called by the bison parser to add a library to the link. 2688 2689 extern "C" void 2690 script_add_library(void* closurev, const char* name, size_t length) 2691 { 2692 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2693 std::string name_string(name, length); 2694 2695 if (name_string[0] != 'l') 2696 gold_error(_("library name must be prefixed with -l")); 2697 2698 Input_file_argument file(name_string.c_str() + 1, 2699 Input_file_argument::INPUT_FILE_TYPE_LIBRARY, 2700 "", false, 2701 closure->position_dependent_options()); 2702 Input_argument& arg = closure->inputs()->add_file(file); 2703 arg.set_script_info(closure->script_info()); 2704 } 2705 2706 // Called by the bison parser to start a group. If we are already in 2707 // a group, that means that this script was invoked within a 2708 // --start-group --end-group sequence on the command line, or that 2709 // this script was found in a GROUP of another script. In that case, 2710 // we simply continue the existing group, rather than starting a new 2711 // one. It is possible to construct a case in which this will do 2712 // something other than what would happen if we did a recursive group, 2713 // but it's hard to imagine why the different behaviour would be 2714 // useful for a real program. Avoiding recursive groups is simpler 2715 // and more efficient. 2716 2717 extern "C" void 2718 script_start_group(void* closurev) 2719 { 2720 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2721 if (!closure->in_group()) 2722 closure->inputs()->start_group(); 2723 } 2724 2725 // Called by the bison parser at the end of a group. 2726 2727 extern "C" void 2728 script_end_group(void* closurev) 2729 { 2730 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2731 if (!closure->in_group()) 2732 closure->inputs()->end_group(); 2733 } 2734 2735 // Called by the bison parser to start an AS_NEEDED list. 2736 2737 extern "C" void 2738 script_start_as_needed(void* closurev) 2739 { 2740 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2741 closure->position_dependent_options().set_as_needed(true); 2742 } 2743 2744 // Called by the bison parser at the end of an AS_NEEDED list. 2745 2746 extern "C" void 2747 script_end_as_needed(void* closurev) 2748 { 2749 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2750 closure->position_dependent_options().set_as_needed(false); 2751 } 2752 2753 // Called by the bison parser to set the entry symbol. 2754 2755 extern "C" void 2756 script_set_entry(void* closurev, const char* entry, size_t length) 2757 { 2758 // We'll parse this exactly the same as --entry=ENTRY on the commandline 2759 // TODO(csilvers): FIXME -- call set_entry directly. 2760 std::string arg("--entry="); 2761 arg.append(entry, length); 2762 script_parse_option(closurev, arg.c_str(), arg.size()); 2763 } 2764 2765 // Called by the bison parser to set whether to define common symbols. 2766 2767 extern "C" void 2768 script_set_common_allocation(void* closurev, int set) 2769 { 2770 const char* arg = set != 0 ? "--define-common" : "--no-define-common"; 2771 script_parse_option(closurev, arg, strlen(arg)); 2772 } 2773 2774 // Called by the bison parser to refer to a symbol. 2775 2776 extern "C" Expression* 2777 script_symbol(void* closurev, const char* name, size_t length) 2778 { 2779 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2780 if (length != 1 || name[0] != '.') 2781 closure->script_options()->add_symbol_reference(name, length); 2782 return script_exp_string(name, length); 2783 } 2784 2785 // Called by the bison parser to define a symbol. 2786 2787 extern "C" void 2788 script_set_symbol(void* closurev, const char* name, size_t length, 2789 Expression* value, int providei, int hiddeni) 2790 { 2791 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2792 const bool provide = providei != 0; 2793 const bool hidden = hiddeni != 0; 2794 closure->script_options()->add_symbol_assignment(name, length, 2795 closure->parsing_defsym(), 2796 value, provide, hidden); 2797 closure->clear_skip_on_incompatible_target(); 2798 } 2799 2800 // Called by the bison parser to add an assertion. 2801 2802 extern "C" void 2803 script_add_assertion(void* closurev, Expression* check, const char* message, 2804 size_t messagelen) 2805 { 2806 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2807 closure->script_options()->add_assertion(check, message, messagelen); 2808 closure->clear_skip_on_incompatible_target(); 2809 } 2810 2811 // Called by the bison parser to parse an OPTION. 2812 2813 extern "C" void 2814 script_parse_option(void* closurev, const char* option, size_t length) 2815 { 2816 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2817 // We treat the option as a single command-line option, even if 2818 // it has internal whitespace. 2819 if (closure->command_line() == NULL) 2820 { 2821 // There are some options that we could handle here--e.g., 2822 // -lLIBRARY. Should we bother? 2823 gold_warning(_("%s:%d:%d: ignoring command OPTION; OPTION is only valid" 2824 " for scripts specified via -T/--script"), 2825 closure->filename(), closure->lineno(), closure->charpos()); 2826 } 2827 else 2828 { 2829 bool past_a_double_dash_option = false; 2830 const char* mutable_option = strndup(option, length); 2831 gold_assert(mutable_option != NULL); 2832 closure->command_line()->process_one_option(1, &mutable_option, 0, 2833 &past_a_double_dash_option); 2834 // The General_options class will quite possibly store a pointer 2835 // into mutable_option, so we can't free it. In cases the class 2836 // does not store such a pointer, this is a memory leak. Alas. :( 2837 } 2838 closure->clear_skip_on_incompatible_target(); 2839 } 2840 2841 // Called by the bison parser to handle OUTPUT_FORMAT. OUTPUT_FORMAT 2842 // takes either one or three arguments. In the three argument case, 2843 // the format depends on the endianness option, which we don't 2844 // currently support (FIXME). If we see an OUTPUT_FORMAT for the 2845 // wrong format, then we want to search for a new file. Returning 0 2846 // here will cause the parser to immediately abort. 2847 2848 extern "C" int 2849 script_check_output_format(void* closurev, 2850 const char* default_name, size_t default_length, 2851 const char*, size_t, const char*, size_t) 2852 { 2853 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2854 std::string name(default_name, default_length); 2855 Target* target = select_target_by_bfd_name(name.c_str()); 2856 if (target == NULL || !parameters->is_compatible_target(target)) 2857 { 2858 if (closure->skip_on_incompatible_target()) 2859 { 2860 closure->set_found_incompatible_target(); 2861 return 0; 2862 } 2863 // FIXME: Should we warn about the unknown target? 2864 } 2865 return 1; 2866 } 2867 2868 // Called by the bison parser to handle TARGET. 2869 2870 extern "C" void 2871 script_set_target(void* closurev, const char* target, size_t len) 2872 { 2873 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2874 std::string s(target, len); 2875 General_options::Object_format format_enum; 2876 format_enum = General_options::string_to_object_format(s.c_str()); 2877 closure->position_dependent_options().set_format_enum(format_enum); 2878 } 2879 2880 // Called by the bison parser to handle SEARCH_DIR. This is handled 2881 // exactly like a -L option. 2882 2883 extern "C" void 2884 script_add_search_dir(void* closurev, const char* option, size_t length) 2885 { 2886 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2887 if (closure->command_line() == NULL) 2888 gold_warning(_("%s:%d:%d: ignoring SEARCH_DIR; SEARCH_DIR is only valid" 2889 " for scripts specified via -T/--script"), 2890 closure->filename(), closure->lineno(), closure->charpos()); 2891 else if (!closure->command_line()->options().nostdlib()) 2892 { 2893 std::string s = "-L" + std::string(option, length); 2894 script_parse_option(closurev, s.c_str(), s.size()); 2895 } 2896 } 2897 2898 /* Called by the bison parser to push the lexer into expression 2899 mode. */ 2900 2901 extern "C" void 2902 script_push_lex_into_expression_mode(void* closurev) 2903 { 2904 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2905 closure->push_lex_mode(Lex::EXPRESSION); 2906 } 2907 2908 /* Called by the bison parser to push the lexer into version 2909 mode. */ 2910 2911 extern "C" void 2912 script_push_lex_into_version_mode(void* closurev) 2913 { 2914 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2915 if (closure->version_script()->is_finalized()) 2916 gold_error(_("%s:%d:%d: invalid use of VERSION in input file"), 2917 closure->filename(), closure->lineno(), closure->charpos()); 2918 closure->push_lex_mode(Lex::VERSION_SCRIPT); 2919 } 2920 2921 /* Called by the bison parser to pop the lexer mode. */ 2922 2923 extern "C" void 2924 script_pop_lex_mode(void* closurev) 2925 { 2926 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2927 closure->pop_lex_mode(); 2928 } 2929 2930 // Register an entire version node. For example: 2931 // 2932 // GLIBC_2.1 { 2933 // global: foo; 2934 // } GLIBC_2.0; 2935 // 2936 // - tag is "GLIBC_2.1" 2937 // - tree contains the information "global: foo" 2938 // - deps contains "GLIBC_2.0" 2939 2940 extern "C" void 2941 script_register_vers_node(void*, 2942 const char* tag, 2943 int taglen, 2944 struct Version_tree* tree, 2945 struct Version_dependency_list* deps) 2946 { 2947 gold_assert(tree != NULL); 2948 tree->dependencies = deps; 2949 if (tag != NULL) 2950 tree->tag = std::string(tag, taglen); 2951 } 2952 2953 // Add a dependencies to the list of existing dependencies, if any, 2954 // and return the expanded list. 2955 2956 extern "C" struct Version_dependency_list* 2957 script_add_vers_depend(void* closurev, 2958 struct Version_dependency_list* all_deps, 2959 const char* depend_to_add, int deplen) 2960 { 2961 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2962 if (all_deps == NULL) 2963 all_deps = closure->version_script()->allocate_dependency_list(); 2964 all_deps->dependencies.push_back(std::string(depend_to_add, deplen)); 2965 return all_deps; 2966 } 2967 2968 // Add a pattern expression to an existing list of expressions, if any. 2969 2970 extern "C" struct Version_expression_list* 2971 script_new_vers_pattern(void* closurev, 2972 struct Version_expression_list* expressions, 2973 const char* pattern, int patlen, int exact_match) 2974 { 2975 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2976 if (expressions == NULL) 2977 expressions = closure->version_script()->allocate_expression_list(); 2978 expressions->expressions.push_back( 2979 Version_expression(std::string(pattern, patlen), 2980 closure->get_current_language(), 2981 static_cast<bool>(exact_match))); 2982 return expressions; 2983 } 2984 2985 // Attaches b to the end of a, and clears b. So a = a + b and b = {}. 2986 2987 extern "C" struct Version_expression_list* 2988 script_merge_expressions(struct Version_expression_list* a, 2989 struct Version_expression_list* b) 2990 { 2991 a->expressions.insert(a->expressions.end(), 2992 b->expressions.begin(), b->expressions.end()); 2993 // We could delete b and remove it from expressions_lists_, but 2994 // that's a lot of work. This works just as well. 2995 b->expressions.clear(); 2996 return a; 2997 } 2998 2999 // Combine the global and local expressions into a a Version_tree. 3000 3001 extern "C" struct Version_tree* 3002 script_new_vers_node(void* closurev, 3003 struct Version_expression_list* global, 3004 struct Version_expression_list* local) 3005 { 3006 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3007 Version_tree* tree = closure->version_script()->allocate_version_tree(); 3008 tree->global = global; 3009 tree->local = local; 3010 return tree; 3011 } 3012 3013 // Handle a transition in language, such as at the 3014 // start or end of 'extern "C++"' 3015 3016 extern "C" void 3017 version_script_push_lang(void* closurev, const char* lang, int langlen) 3018 { 3019 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3020 std::string language(lang, langlen); 3021 Version_script_info::Language code; 3022 if (language.empty() || language == "C") 3023 code = Version_script_info::LANGUAGE_C; 3024 else if (language == "C++") 3025 code = Version_script_info::LANGUAGE_CXX; 3026 else if (language == "Java") 3027 code = Version_script_info::LANGUAGE_JAVA; 3028 else 3029 { 3030 char* buf = new char[langlen + 100]; 3031 snprintf(buf, langlen + 100, 3032 _("unrecognized version script language '%s'"), 3033 language.c_str()); 3034 yyerror(closurev, buf); 3035 delete[] buf; 3036 code = Version_script_info::LANGUAGE_C; 3037 } 3038 closure->push_language(code); 3039 } 3040 3041 extern "C" void 3042 version_script_pop_lang(void* closurev) 3043 { 3044 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3045 closure->pop_language(); 3046 } 3047 3048 // Called by the bison parser to start a SECTIONS clause. 3049 3050 extern "C" void 3051 script_start_sections(void* closurev) 3052 { 3053 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3054 closure->script_options()->script_sections()->start_sections(); 3055 closure->clear_skip_on_incompatible_target(); 3056 } 3057 3058 // Called by the bison parser to finish a SECTIONS clause. 3059 3060 extern "C" void 3061 script_finish_sections(void* closurev) 3062 { 3063 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3064 closure->script_options()->script_sections()->finish_sections(); 3065 } 3066 3067 // Start processing entries for an output section. 3068 3069 extern "C" void 3070 script_start_output_section(void* closurev, const char* name, size_t namelen, 3071 const struct Parser_output_section_header* header) 3072 { 3073 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3074 closure->script_options()->script_sections()->start_output_section(name, 3075 namelen, 3076 header); 3077 } 3078 3079 // Finish processing entries for an output section. 3080 3081 extern "C" void 3082 script_finish_output_section(void* closurev, 3083 const struct Parser_output_section_trailer* trail) 3084 { 3085 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3086 closure->script_options()->script_sections()->finish_output_section(trail); 3087 } 3088 3089 // Add a data item (e.g., "WORD (0)") to the current output section. 3090 3091 extern "C" void 3092 script_add_data(void* closurev, int data_token, Expression* val) 3093 { 3094 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3095 int size; 3096 bool is_signed = true; 3097 switch (data_token) 3098 { 3099 case QUAD: 3100 size = 8; 3101 is_signed = false; 3102 break; 3103 case SQUAD: 3104 size = 8; 3105 break; 3106 case LONG: 3107 size = 4; 3108 break; 3109 case SHORT: 3110 size = 2; 3111 break; 3112 case BYTE: 3113 size = 1; 3114 break; 3115 default: 3116 gold_unreachable(); 3117 } 3118 closure->script_options()->script_sections()->add_data(size, is_signed, val); 3119 } 3120 3121 // Add a clause setting the fill value to the current output section. 3122 3123 extern "C" void 3124 script_add_fill(void* closurev, Expression* val) 3125 { 3126 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3127 closure->script_options()->script_sections()->add_fill(val); 3128 } 3129 3130 // Add a new input section specification to the current output 3131 // section. 3132 3133 extern "C" void 3134 script_add_input_section(void* closurev, 3135 const struct Input_section_spec* spec, 3136 int keepi) 3137 { 3138 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3139 bool keep = keepi != 0; 3140 closure->script_options()->script_sections()->add_input_section(spec, keep); 3141 } 3142 3143 // When we see DATA_SEGMENT_ALIGN we record that following output 3144 // sections may be relro. 3145 3146 extern "C" void 3147 script_data_segment_align(void* closurev) 3148 { 3149 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3150 if (!closure->script_options()->saw_sections_clause()) 3151 gold_error(_("%s:%d:%d: DATA_SEGMENT_ALIGN not in SECTIONS clause"), 3152 closure->filename(), closure->lineno(), closure->charpos()); 3153 else 3154 closure->script_options()->script_sections()->data_segment_align(); 3155 } 3156 3157 // When we see DATA_SEGMENT_RELRO_END we know that all output sections 3158 // since DATA_SEGMENT_ALIGN should be relro. 3159 3160 extern "C" void 3161 script_data_segment_relro_end(void* closurev) 3162 { 3163 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3164 if (!closure->script_options()->saw_sections_clause()) 3165 gold_error(_("%s:%d:%d: DATA_SEGMENT_ALIGN not in SECTIONS clause"), 3166 closure->filename(), closure->lineno(), closure->charpos()); 3167 else 3168 closure->script_options()->script_sections()->data_segment_relro_end(); 3169 } 3170 3171 // Create a new list of string/sort pairs. 3172 3173 extern "C" String_sort_list_ptr 3174 script_new_string_sort_list(const struct Wildcard_section* string_sort) 3175 { 3176 return new String_sort_list(1, *string_sort); 3177 } 3178 3179 // Add an entry to a list of string/sort pairs. The way the parser 3180 // works permits us to simply modify the first parameter, rather than 3181 // copy the vector. 3182 3183 extern "C" String_sort_list_ptr 3184 script_string_sort_list_add(String_sort_list_ptr pv, 3185 const struct Wildcard_section* string_sort) 3186 { 3187 if (pv == NULL) 3188 return script_new_string_sort_list(string_sort); 3189 else 3190 { 3191 pv->push_back(*string_sort); 3192 return pv; 3193 } 3194 } 3195 3196 // Create a new list of strings. 3197 3198 extern "C" String_list_ptr 3199 script_new_string_list(const char* str, size_t len) 3200 { 3201 return new String_list(1, std::string(str, len)); 3202 } 3203 3204 // Add an element to a list of strings. The way the parser works 3205 // permits us to simply modify the first parameter, rather than copy 3206 // the vector. 3207 3208 extern "C" String_list_ptr 3209 script_string_list_push_back(String_list_ptr pv, const char* str, size_t len) 3210 { 3211 if (pv == NULL) 3212 return script_new_string_list(str, len); 3213 else 3214 { 3215 pv->push_back(std::string(str, len)); 3216 return pv; 3217 } 3218 } 3219 3220 // Concatenate two string lists. Either or both may be NULL. The way 3221 // the parser works permits us to modify the parameters, rather than 3222 // copy the vector. 3223 3224 extern "C" String_list_ptr 3225 script_string_list_append(String_list_ptr pv1, String_list_ptr pv2) 3226 { 3227 if (pv1 == NULL) 3228 return pv2; 3229 if (pv2 == NULL) 3230 return pv1; 3231 pv1->insert(pv1->end(), pv2->begin(), pv2->end()); 3232 return pv1; 3233 } 3234 3235 // Add a new program header. 3236 3237 extern "C" void 3238 script_add_phdr(void* closurev, const char* name, size_t namelen, 3239 unsigned int type, const Phdr_info* info) 3240 { 3241 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3242 bool includes_filehdr = info->includes_filehdr != 0; 3243 bool includes_phdrs = info->includes_phdrs != 0; 3244 bool is_flags_valid = info->is_flags_valid != 0; 3245 Script_sections* ss = closure->script_options()->script_sections(); 3246 ss->add_phdr(name, namelen, type, includes_filehdr, includes_phdrs, 3247 is_flags_valid, info->flags, info->load_address); 3248 closure->clear_skip_on_incompatible_target(); 3249 } 3250 3251 // Convert a program header string to a type. 3252 3253 #define PHDR_TYPE(NAME) { #NAME, sizeof(#NAME) - 1, elfcpp::NAME } 3254 3255 static struct 3256 { 3257 const char* name; 3258 size_t namelen; 3259 unsigned int val; 3260 } phdr_type_names[] = 3261 { 3262 PHDR_TYPE(PT_NULL), 3263 PHDR_TYPE(PT_LOAD), 3264 PHDR_TYPE(PT_DYNAMIC), 3265 PHDR_TYPE(PT_INTERP), 3266 PHDR_TYPE(PT_NOTE), 3267 PHDR_TYPE(PT_SHLIB), 3268 PHDR_TYPE(PT_PHDR), 3269 PHDR_TYPE(PT_TLS), 3270 PHDR_TYPE(PT_GNU_EH_FRAME), 3271 PHDR_TYPE(PT_GNU_STACK), 3272 PHDR_TYPE(PT_GNU_RELRO) 3273 }; 3274 3275 extern "C" unsigned int 3276 script_phdr_string_to_type(void* closurev, const char* name, size_t namelen) 3277 { 3278 for (unsigned int i = 0; 3279 i < sizeof(phdr_type_names) / sizeof(phdr_type_names[0]); 3280 ++i) 3281 if (namelen == phdr_type_names[i].namelen 3282 && strncmp(name, phdr_type_names[i].name, namelen) == 0) 3283 return phdr_type_names[i].val; 3284 yyerror(closurev, _("unknown PHDR type (try integer)")); 3285 return elfcpp::PT_NULL; 3286 } 3287 3288 extern "C" void 3289 script_saw_segment_start_expression(void* closurev) 3290 { 3291 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3292 Script_sections* ss = closure->script_options()->script_sections(); 3293 ss->set_saw_segment_start_expression(true); 3294 } 3295 3296 extern "C" void 3297 script_set_section_region(void* closurev, const char* name, size_t namelen, 3298 int set_vma) 3299 { 3300 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3301 if (!closure->script_options()->saw_sections_clause()) 3302 { 3303 gold_error(_("%s:%d:%d: MEMORY region '%.*s' referred to outside of " 3304 "SECTIONS clause"), 3305 closure->filename(), closure->lineno(), closure->charpos(), 3306 static_cast<int>(namelen), name); 3307 return; 3308 } 3309 3310 Script_sections* ss = closure->script_options()->script_sections(); 3311 Memory_region* mr = ss->find_memory_region(name, namelen); 3312 if (mr == NULL) 3313 { 3314 gold_error(_("%s:%d:%d: MEMORY region '%.*s' not declared"), 3315 closure->filename(), closure->lineno(), closure->charpos(), 3316 static_cast<int>(namelen), name); 3317 return; 3318 } 3319 3320 ss->set_memory_region(mr, set_vma); 3321 } 3322 3323 extern "C" void 3324 script_add_memory(void* closurev, const char* name, size_t namelen, 3325 unsigned int attrs, Expression* origin, Expression* length) 3326 { 3327 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3328 Script_sections* ss = closure->script_options()->script_sections(); 3329 ss->add_memory_region(name, namelen, attrs, origin, length); 3330 } 3331 3332 extern "C" unsigned int 3333 script_parse_memory_attr(void* closurev, const char* attrs, size_t attrlen, 3334 int invert) 3335 { 3336 int attributes = 0; 3337 3338 while (attrlen--) 3339 switch (*attrs++) 3340 { 3341 case 'R': 3342 case 'r': 3343 attributes |= MEM_READABLE; break; 3344 case 'W': 3345 case 'w': 3346 attributes |= MEM_READABLE | MEM_WRITEABLE; break; 3347 case 'X': 3348 case 'x': 3349 attributes |= MEM_EXECUTABLE; break; 3350 case 'A': 3351 case 'a': 3352 attributes |= MEM_ALLOCATABLE; break; 3353 case 'I': 3354 case 'i': 3355 case 'L': 3356 case 'l': 3357 attributes |= MEM_INITIALIZED; break; 3358 default: 3359 yyerror(closurev, _("unknown MEMORY attribute")); 3360 } 3361 3362 if (invert) 3363 attributes = (~ attributes) & MEM_ATTR_MASK; 3364 3365 return attributes; 3366 } 3367 3368 extern "C" void 3369 script_include_directive(int first_token, void* closurev, 3370 const char* filename, size_t length) 3371 { 3372 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3373 std::string name(filename, length); 3374 Command_line* cmdline = closure->command_line(); 3375 read_script_file(name.c_str(), cmdline, &cmdline->script_options(), 3376 first_token, Lex::LINKER_SCRIPT); 3377 } 3378 3379 // Functions for memory regions. 3380 3381 extern "C" Expression* 3382 script_exp_function_origin(void* closurev, const char* name, size_t namelen) 3383 { 3384 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3385 Script_sections* ss = closure->script_options()->script_sections(); 3386 Expression* origin = ss->find_memory_region_origin(name, namelen); 3387 3388 if (origin == NULL) 3389 { 3390 gold_error(_("undefined memory region '%s' referenced " 3391 "in ORIGIN expression"), 3392 name); 3393 // Create a dummy expression to prevent crashes later on. 3394 origin = script_exp_integer(0); 3395 } 3396 3397 return origin; 3398 } 3399 3400 extern "C" Expression* 3401 script_exp_function_length(void* closurev, const char* name, size_t namelen) 3402 { 3403 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3404 Script_sections* ss = closure->script_options()->script_sections(); 3405 Expression* length = ss->find_memory_region_length(name, namelen); 3406 3407 if (length == NULL) 3408 { 3409 gold_error(_("undefined memory region '%s' referenced " 3410 "in LENGTH expression"), 3411 name); 3412 // Create a dummy expression to prevent crashes later on. 3413 length = script_exp_integer(0); 3414 } 3415 3416 return length; 3417 } 3418
