1 // script.cc -- handle linker scripts for gold. 2 3 // Copyright (C) 2006-2016 Free Software Foundation, Inc. 4 // Written by Ian Lance Taylor <iant@google.com>. 5 6 // This file is part of gold. 7 8 // This program is free software; you can redistribute it and/or modify 9 // it under the terms of the GNU General Public License as published by 10 // the Free Software Foundation; either version 3 of the License, or 11 // (at your option) any later version. 12 13 // This program is distributed in the hope that it will be useful, 14 // but WITHOUT ANY WARRANTY; without even the implied warranty of 15 // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 16 // GNU General Public License for more details. 17 18 // You should have received a copy of the GNU General Public License 19 // along with this program; if not, write to the Free Software 20 // Foundation, Inc., 51 Franklin Street - Fifth Floor, Boston, 21 // MA 02110-1301, USA. 22 23 #include "gold.h" 24 25 #include <cstdio> 26 #include <cstdlib> 27 #include <cstring> 28 #include <fnmatch.h> 29 #include <string> 30 #include <vector> 31 #include "filenames.h" 32 33 #include "elfcpp.h" 34 #include "demangle.h" 35 #include "dirsearch.h" 36 #include "options.h" 37 #include "fileread.h" 38 #include "workqueue.h" 39 #include "readsyms.h" 40 #include "parameters.h" 41 #include "layout.h" 42 #include "symtab.h" 43 #include "target-select.h" 44 #include "script.h" 45 #include "script-c.h" 46 #include "incremental.h" 47 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, NULL); 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 bool is_valid; 1013 uint64_t val = this->val_->eval_maybe_dot(symtab, layout, false, 1014 is_dot_available, dot_value, 1015 dot_section, &val_section, NULL, 1016 NULL, NULL, NULL, false, &is_valid); 1017 if (!is_valid || (val_section != NULL && val_section != dot_section)) 1018 return; 1019 1020 if (parameters->target().get_size() == 32) 1021 { 1022 #if defined(HAVE_TARGET_32_LITTLE) || defined(HAVE_TARGET_32_BIG) 1023 Sized_symbol<32>* ssym = symtab->get_sized_symbol<32>(this->sym_); 1024 ssym->set_value(val); 1025 #else 1026 gold_unreachable(); 1027 #endif 1028 } 1029 else if (parameters->target().get_size() == 64) 1030 { 1031 #if defined(HAVE_TARGET_64_LITTLE) || defined(HAVE_TARGET_64_BIG) 1032 Sized_symbol<64>* ssym = symtab->get_sized_symbol<64>(this->sym_); 1033 ssym->set_value(val); 1034 #else 1035 gold_unreachable(); 1036 #endif 1037 } 1038 else 1039 gold_unreachable(); 1040 if (val_section != NULL) 1041 this->sym_->set_output_section(val_section); 1042 } 1043 1044 // Print for debugging. 1045 1046 void 1047 Symbol_assignment::print(FILE* f) const 1048 { 1049 if (this->provide_ && this->hidden_) 1050 fprintf(f, "PROVIDE_HIDDEN("); 1051 else if (this->provide_) 1052 fprintf(f, "PROVIDE("); 1053 else if (this->hidden_) 1054 gold_unreachable(); 1055 1056 fprintf(f, "%s = ", this->name_.c_str()); 1057 this->val_->print(f); 1058 1059 if (this->provide_ || this->hidden_) 1060 fprintf(f, ")"); 1061 1062 fprintf(f, "\n"); 1063 } 1064 1065 // Class Script_assertion. 1066 1067 // Check the assertion. 1068 1069 void 1070 Script_assertion::check(const Symbol_table* symtab, const Layout* layout) 1071 { 1072 if (!this->check_->eval(symtab, layout, true)) 1073 gold_error("%s", this->message_.c_str()); 1074 } 1075 1076 // Print for debugging. 1077 1078 void 1079 Script_assertion::print(FILE* f) const 1080 { 1081 fprintf(f, "ASSERT("); 1082 this->check_->print(f); 1083 fprintf(f, ", \"%s\")\n", this->message_.c_str()); 1084 } 1085 1086 // Class Script_options. 1087 1088 Script_options::Script_options() 1089 : entry_(), symbol_assignments_(), symbol_definitions_(), 1090 symbol_references_(), version_script_info_(), script_sections_() 1091 { 1092 } 1093 1094 // Returns true if NAME is on the list of symbol assignments waiting 1095 // to be processed. 1096 1097 bool 1098 Script_options::is_pending_assignment(const char* name) 1099 { 1100 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin(); 1101 p != this->symbol_assignments_.end(); 1102 ++p) 1103 if ((*p)->name() == name) 1104 return true; 1105 return false; 1106 } 1107 1108 // Add a symbol to be defined. 1109 1110 void 1111 Script_options::add_symbol_assignment(const char* name, size_t length, 1112 bool is_defsym, Expression* value, 1113 bool provide, bool hidden) 1114 { 1115 if (length != 1 || name[0] != '.') 1116 { 1117 if (this->script_sections_.in_sections_clause()) 1118 { 1119 gold_assert(!is_defsym); 1120 this->script_sections_.add_symbol_assignment(name, length, value, 1121 provide, hidden); 1122 } 1123 else 1124 { 1125 Symbol_assignment* p = new Symbol_assignment(name, length, is_defsym, 1126 value, provide, hidden); 1127 this->symbol_assignments_.push_back(p); 1128 } 1129 1130 if (!provide) 1131 { 1132 std::string n(name, length); 1133 this->symbol_definitions_.insert(n); 1134 this->symbol_references_.erase(n); 1135 } 1136 } 1137 else 1138 { 1139 if (provide || hidden) 1140 gold_error(_("invalid use of PROVIDE for dot symbol")); 1141 1142 // The GNU linker permits assignments to dot outside of SECTIONS 1143 // clauses and treats them as occurring inside, so we don't 1144 // check in_sections_clause here. 1145 this->script_sections_.add_dot_assignment(value); 1146 } 1147 } 1148 1149 // Add a reference to a symbol. 1150 1151 void 1152 Script_options::add_symbol_reference(const char* name, size_t length) 1153 { 1154 if (length != 1 || name[0] != '.') 1155 { 1156 std::string n(name, length); 1157 if (this->symbol_definitions_.find(n) == this->symbol_definitions_.end()) 1158 this->symbol_references_.insert(n); 1159 } 1160 } 1161 1162 // Add an assertion. 1163 1164 void 1165 Script_options::add_assertion(Expression* check, const char* message, 1166 size_t messagelen) 1167 { 1168 if (this->script_sections_.in_sections_clause()) 1169 this->script_sections_.add_assertion(check, message, messagelen); 1170 else 1171 { 1172 Script_assertion* p = new Script_assertion(check, message, messagelen); 1173 this->assertions_.push_back(p); 1174 } 1175 } 1176 1177 // Create sections required by any linker scripts. 1178 1179 void 1180 Script_options::create_script_sections(Layout* layout) 1181 { 1182 if (this->saw_sections_clause()) 1183 this->script_sections_.create_sections(layout); 1184 } 1185 1186 // Add any symbols we are defining to the symbol table. 1187 1188 void 1189 Script_options::add_symbols_to_table(Symbol_table* symtab) 1190 { 1191 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin(); 1192 p != this->symbol_assignments_.end(); 1193 ++p) 1194 (*p)->add_to_table(symtab); 1195 this->script_sections_.add_symbols_to_table(symtab); 1196 } 1197 1198 // Finalize symbol values. Also check assertions. 1199 1200 void 1201 Script_options::finalize_symbols(Symbol_table* symtab, const Layout* layout) 1202 { 1203 // We finalize the symbols defined in SECTIONS first, because they 1204 // are the ones which may have changed. This way if symbol outside 1205 // SECTIONS are defined in terms of symbols inside SECTIONS, they 1206 // will get the right value. 1207 this->script_sections_.finalize_symbols(symtab, layout); 1208 1209 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin(); 1210 p != this->symbol_assignments_.end(); 1211 ++p) 1212 (*p)->finalize(symtab, layout); 1213 1214 for (Assertions::iterator p = this->assertions_.begin(); 1215 p != this->assertions_.end(); 1216 ++p) 1217 (*p)->check(symtab, layout); 1218 } 1219 1220 // Set section addresses. We set all the symbols which have absolute 1221 // values. Then we let the SECTIONS clause do its thing. This 1222 // returns the segment which holds the file header and segment 1223 // headers, if any. 1224 1225 Output_segment* 1226 Script_options::set_section_addresses(Symbol_table* symtab, Layout* layout) 1227 { 1228 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin(); 1229 p != this->symbol_assignments_.end(); 1230 ++p) 1231 (*p)->set_if_absolute(symtab, layout, false, 0, NULL); 1232 1233 return this->script_sections_.set_section_addresses(symtab, layout); 1234 } 1235 1236 // This class holds data passed through the parser to the lexer and to 1237 // the parser support functions. This avoids global variables. We 1238 // can't use global variables because we need not be called by a 1239 // singleton thread. 1240 1241 class Parser_closure 1242 { 1243 public: 1244 Parser_closure(const char* filename, 1245 const Position_dependent_options& posdep_options, 1246 bool parsing_defsym, bool in_group, bool is_in_sysroot, 1247 Command_line* command_line, 1248 Script_options* script_options, 1249 Lex* lex, 1250 bool skip_on_incompatible_target, 1251 Script_info* script_info) 1252 : filename_(filename), posdep_options_(posdep_options), 1253 parsing_defsym_(parsing_defsym), in_group_(in_group), 1254 is_in_sysroot_(is_in_sysroot), 1255 skip_on_incompatible_target_(skip_on_incompatible_target), 1256 found_incompatible_target_(false), 1257 command_line_(command_line), script_options_(script_options), 1258 version_script_info_(script_options->version_script_info()), 1259 lex_(lex), lineno_(0), charpos_(0), lex_mode_stack_(), inputs_(NULL), 1260 script_info_(script_info) 1261 { 1262 // We start out processing C symbols in the default lex mode. 1263 this->language_stack_.push_back(Version_script_info::LANGUAGE_C); 1264 this->lex_mode_stack_.push_back(lex->mode()); 1265 } 1266 1267 // Return the file name. 1268 const char* 1269 filename() const 1270 { return this->filename_; } 1271 1272 // Return the position dependent options. The caller may modify 1273 // this. 1274 Position_dependent_options& 1275 position_dependent_options() 1276 { return this->posdep_options_; } 1277 1278 // Whether we are parsing a --defsym. 1279 bool 1280 parsing_defsym() const 1281 { return this->parsing_defsym_; } 1282 1283 // Return whether this script is being run in a group. 1284 bool 1285 in_group() const 1286 { return this->in_group_; } 1287 1288 // Return whether this script was found using a directory in the 1289 // sysroot. 1290 bool 1291 is_in_sysroot() const 1292 { return this->is_in_sysroot_; } 1293 1294 // Whether to skip to the next file with the same name if we find an 1295 // incompatible target in an OUTPUT_FORMAT statement. 1296 bool 1297 skip_on_incompatible_target() const 1298 { return this->skip_on_incompatible_target_; } 1299 1300 // Stop skipping to the next file on an incompatible target. This 1301 // is called when we make some unrevocable change to the data 1302 // structures. 1303 void 1304 clear_skip_on_incompatible_target() 1305 { this->skip_on_incompatible_target_ = false; } 1306 1307 // Whether we found an incompatible target in an OUTPUT_FORMAT 1308 // statement. 1309 bool 1310 found_incompatible_target() const 1311 { return this->found_incompatible_target_; } 1312 1313 // Note that we found an incompatible target. 1314 void 1315 set_found_incompatible_target() 1316 { this->found_incompatible_target_ = true; } 1317 1318 // Returns the Command_line structure passed in at constructor time. 1319 // This value may be NULL. The caller may modify this, which modifies 1320 // the passed-in Command_line object (not a copy). 1321 Command_line* 1322 command_line() 1323 { return this->command_line_; } 1324 1325 // Return the options which may be set by a script. 1326 Script_options* 1327 script_options() 1328 { return this->script_options_; } 1329 1330 // Return the object in which version script information should be stored. 1331 Version_script_info* 1332 version_script() 1333 { return this->version_script_info_; } 1334 1335 // Return the next token, and advance. 1336 const Token* 1337 next_token() 1338 { 1339 const Token* token = this->lex_->next_token(); 1340 this->lineno_ = token->lineno(); 1341 this->charpos_ = token->charpos(); 1342 return token; 1343 } 1344 1345 // Set a new lexer mode, pushing the current one. 1346 void 1347 push_lex_mode(Lex::Mode mode) 1348 { 1349 this->lex_mode_stack_.push_back(this->lex_->mode()); 1350 this->lex_->set_mode(mode); 1351 } 1352 1353 // Pop the lexer mode. 1354 void 1355 pop_lex_mode() 1356 { 1357 gold_assert(!this->lex_mode_stack_.empty()); 1358 this->lex_->set_mode(this->lex_mode_stack_.back()); 1359 this->lex_mode_stack_.pop_back(); 1360 } 1361 1362 // Return the current lexer mode. 1363 Lex::Mode 1364 lex_mode() const 1365 { return this->lex_mode_stack_.back(); } 1366 1367 // Return the line number of the last token. 1368 int 1369 lineno() const 1370 { return this->lineno_; } 1371 1372 // Return the character position in the line of the last token. 1373 int 1374 charpos() const 1375 { return this->charpos_; } 1376 1377 // Return the list of input files, creating it if necessary. This 1378 // is a space leak--we never free the INPUTS_ pointer. 1379 Input_arguments* 1380 inputs() 1381 { 1382 if (this->inputs_ == NULL) 1383 this->inputs_ = new Input_arguments(); 1384 return this->inputs_; 1385 } 1386 1387 // Return whether we saw any input files. 1388 bool 1389 saw_inputs() const 1390 { return this->inputs_ != NULL && !this->inputs_->empty(); } 1391 1392 // Return the current language being processed in a version script 1393 // (eg, "C++"). The empty string represents unmangled C names. 1394 Version_script_info::Language 1395 get_current_language() const 1396 { return this->language_stack_.back(); } 1397 1398 // Push a language onto the stack when entering an extern block. 1399 void 1400 push_language(Version_script_info::Language lang) 1401 { this->language_stack_.push_back(lang); } 1402 1403 // Pop a language off of the stack when exiting an extern block. 1404 void 1405 pop_language() 1406 { 1407 gold_assert(!this->language_stack_.empty()); 1408 this->language_stack_.pop_back(); 1409 } 1410 1411 // Return a pointer to the incremental info. 1412 Script_info* 1413 script_info() 1414 { return this->script_info_; } 1415 1416 private: 1417 // The name of the file we are reading. 1418 const char* filename_; 1419 // The position dependent options. 1420 Position_dependent_options posdep_options_; 1421 // True if we are parsing a --defsym. 1422 bool parsing_defsym_; 1423 // Whether we are currently in a --start-group/--end-group. 1424 bool in_group_; 1425 // Whether the script was found in a sysrooted directory. 1426 bool is_in_sysroot_; 1427 // If this is true, then if we find an OUTPUT_FORMAT with an 1428 // incompatible target, then we tell the parser to abort so that we 1429 // can search for the next file with the same name. 1430 bool skip_on_incompatible_target_; 1431 // True if we found an OUTPUT_FORMAT with an incompatible target. 1432 bool found_incompatible_target_; 1433 // May be NULL if the user chooses not to pass one in. 1434 Command_line* command_line_; 1435 // Options which may be set from any linker script. 1436 Script_options* script_options_; 1437 // Information parsed from a version script. 1438 Version_script_info* version_script_info_; 1439 // The lexer. 1440 Lex* lex_; 1441 // The line number of the last token returned by next_token. 1442 int lineno_; 1443 // The column number of the last token returned by next_token. 1444 int charpos_; 1445 // A stack of lexer modes. 1446 std::vector<Lex::Mode> lex_mode_stack_; 1447 // A stack of which extern/language block we're inside. Can be C++, 1448 // java, or empty for C. 1449 std::vector<Version_script_info::Language> language_stack_; 1450 // New input files found to add to the link. 1451 Input_arguments* inputs_; 1452 // Pointer to incremental linking info. 1453 Script_info* script_info_; 1454 }; 1455 1456 // FILE was found as an argument on the command line. Try to read it 1457 // as a script. Return true if the file was handled. 1458 1459 bool 1460 read_input_script(Workqueue* workqueue, Symbol_table* symtab, Layout* layout, 1461 Dirsearch* dirsearch, int dirindex, 1462 Input_objects* input_objects, Mapfile* mapfile, 1463 Input_group* input_group, 1464 const Input_argument* input_argument, 1465 Input_file* input_file, Task_token* next_blocker, 1466 bool* used_next_blocker) 1467 { 1468 *used_next_blocker = false; 1469 1470 std::string input_string; 1471 Lex::read_file(input_file, &input_string); 1472 1473 Lex lex(input_string.c_str(), input_string.length(), PARSING_LINKER_SCRIPT); 1474 1475 Script_info* script_info = NULL; 1476 if (layout->incremental_inputs() != NULL) 1477 { 1478 const std::string& filename = input_file->filename(); 1479 Timespec mtime = input_file->file().get_mtime(); 1480 unsigned int arg_serial = input_argument->file().arg_serial(); 1481 script_info = new Script_info(filename); 1482 layout->incremental_inputs()->report_script(script_info, arg_serial, 1483 mtime); 1484 } 1485 1486 Parser_closure closure(input_file->filename().c_str(), 1487 input_argument->file().options(), 1488 false, 1489 input_group != NULL, 1490 input_file->is_in_sysroot(), 1491 NULL, 1492 layout->script_options(), 1493 &lex, 1494 input_file->will_search_for(), 1495 script_info); 1496 1497 bool old_saw_sections_clause = 1498 layout->script_options()->saw_sections_clause(); 1499 1500 if (yyparse(&closure) != 0) 1501 { 1502 if (closure.found_incompatible_target()) 1503 { 1504 Read_symbols::incompatible_warning(input_argument, input_file); 1505 Read_symbols::requeue(workqueue, input_objects, symtab, layout, 1506 dirsearch, dirindex, mapfile, input_argument, 1507 input_group, next_blocker); 1508 return true; 1509 } 1510 return false; 1511 } 1512 1513 if (!old_saw_sections_clause 1514 && layout->script_options()->saw_sections_clause() 1515 && layout->have_added_input_section()) 1516 gold_error(_("%s: SECTIONS seen after other input files; try -T/--script"), 1517 input_file->filename().c_str()); 1518 1519 if (!closure.saw_inputs()) 1520 return true; 1521 1522 Task_token* this_blocker = NULL; 1523 for (Input_arguments::const_iterator p = closure.inputs()->begin(); 1524 p != closure.inputs()->end(); 1525 ++p) 1526 { 1527 Task_token* nb; 1528 if (p + 1 == closure.inputs()->end()) 1529 nb = next_blocker; 1530 else 1531 { 1532 nb = new Task_token(true); 1533 nb->add_blocker(); 1534 } 1535 workqueue->queue_soon(new Read_symbols(input_objects, symtab, 1536 layout, dirsearch, 0, mapfile, &*p, 1537 input_group, NULL, this_blocker, nb)); 1538 this_blocker = nb; 1539 } 1540 1541 *used_next_blocker = true; 1542 1543 return true; 1544 } 1545 1546 // Helper function for read_version_script(), read_commandline_script() and 1547 // script_include_directive(). Processes the given file in the mode indicated 1548 // by first_token and lex_mode. 1549 1550 static bool 1551 read_script_file(const char* filename, Command_line* cmdline, 1552 Script_options* script_options, 1553 int first_token, Lex::Mode lex_mode) 1554 { 1555 Dirsearch dirsearch; 1556 std::string name = filename; 1557 1558 // If filename is a relative filename, search for it manually using "." + 1559 // cmdline->options()->library_path() -- not dirsearch. 1560 if (!IS_ABSOLUTE_PATH(filename)) 1561 { 1562 const General_options::Dir_list& search_path = 1563 cmdline->options().library_path(); 1564 name = Dirsearch::find_file_in_dir_list(name, search_path, "."); 1565 } 1566 1567 // The file locking code wants to record a Task, but we haven't 1568 // started the workqueue yet. This is only for debugging purposes, 1569 // so we invent a fake value. 1570 const Task* task = reinterpret_cast<const Task*>(-1); 1571 1572 // We don't want this file to be opened in binary mode. 1573 Position_dependent_options posdep = cmdline->position_dependent_options(); 1574 if (posdep.format_enum() == General_options::OBJECT_FORMAT_BINARY) 1575 posdep.set_format_enum(General_options::OBJECT_FORMAT_ELF); 1576 Input_file_argument input_argument(name.c_str(), 1577 Input_file_argument::INPUT_FILE_TYPE_FILE, 1578 "", false, posdep); 1579 Input_file input_file(&input_argument); 1580 int dummy = 0; 1581 if (!input_file.open(dirsearch, task, &dummy)) 1582 return false; 1583 1584 std::string input_string; 1585 Lex::read_file(&input_file, &input_string); 1586 1587 Lex lex(input_string.c_str(), input_string.length(), first_token); 1588 lex.set_mode(lex_mode); 1589 1590 Parser_closure closure(filename, 1591 cmdline->position_dependent_options(), 1592 first_token == Lex::DYNAMIC_LIST, 1593 false, 1594 input_file.is_in_sysroot(), 1595 cmdline, 1596 script_options, 1597 &lex, 1598 false, 1599 NULL); 1600 if (yyparse(&closure) != 0) 1601 { 1602 input_file.file().unlock(task); 1603 return false; 1604 } 1605 1606 input_file.file().unlock(task); 1607 1608 gold_assert(!closure.saw_inputs()); 1609 1610 return true; 1611 } 1612 1613 // FILENAME was found as an argument to --script (-T). 1614 // Read it as a script, and execute its contents immediately. 1615 1616 bool 1617 read_commandline_script(const char* filename, Command_line* cmdline) 1618 { 1619 return read_script_file(filename, cmdline, &cmdline->script_options(), 1620 PARSING_LINKER_SCRIPT, Lex::LINKER_SCRIPT); 1621 } 1622 1623 // FILENAME was found as an argument to --version-script. Read it as 1624 // a version script, and store its contents in 1625 // cmdline->script_options()->version_script_info(). 1626 1627 bool 1628 read_version_script(const char* filename, Command_line* cmdline) 1629 { 1630 return read_script_file(filename, cmdline, &cmdline->script_options(), 1631 PARSING_VERSION_SCRIPT, Lex::VERSION_SCRIPT); 1632 } 1633 1634 // FILENAME was found as an argument to --dynamic-list. Read it as a 1635 // list of symbols, and store its contents in DYNAMIC_LIST. 1636 1637 bool 1638 read_dynamic_list(const char* filename, Command_line* cmdline, 1639 Script_options* dynamic_list) 1640 { 1641 return read_script_file(filename, cmdline, dynamic_list, 1642 PARSING_DYNAMIC_LIST, Lex::DYNAMIC_LIST); 1643 } 1644 1645 // Implement the --defsym option on the command line. Return true if 1646 // all is well. 1647 1648 bool 1649 Script_options::define_symbol(const char* definition) 1650 { 1651 Lex lex(definition, strlen(definition), PARSING_DEFSYM); 1652 lex.set_mode(Lex::EXPRESSION); 1653 1654 // Dummy value. 1655 Position_dependent_options posdep_options; 1656 1657 Parser_closure closure("command line", posdep_options, true, 1658 false, false, NULL, this, &lex, false, NULL); 1659 1660 if (yyparse(&closure) != 0) 1661 return false; 1662 1663 gold_assert(!closure.saw_inputs()); 1664 1665 return true; 1666 } 1667 1668 // Print the script to F for debugging. 1669 1670 void 1671 Script_options::print(FILE* f) const 1672 { 1673 fprintf(f, "%s: Dumping linker script\n", program_name); 1674 1675 if (!this->entry_.empty()) 1676 fprintf(f, "ENTRY(%s)\n", this->entry_.c_str()); 1677 1678 for (Symbol_assignments::const_iterator p = 1679 this->symbol_assignments_.begin(); 1680 p != this->symbol_assignments_.end(); 1681 ++p) 1682 (*p)->print(f); 1683 1684 for (Assertions::const_iterator p = this->assertions_.begin(); 1685 p != this->assertions_.end(); 1686 ++p) 1687 (*p)->print(f); 1688 1689 this->script_sections_.print(f); 1690 1691 this->version_script_info_.print(f); 1692 } 1693 1694 // Manage mapping from keywords to the codes expected by the bison 1695 // parser. We construct one global object for each lex mode with 1696 // keywords. 1697 1698 class Keyword_to_parsecode 1699 { 1700 public: 1701 // The structure which maps keywords to parsecodes. 1702 struct Keyword_parsecode 1703 { 1704 // Keyword. 1705 const char* keyword; 1706 // Corresponding parsecode. 1707 int parsecode; 1708 }; 1709 1710 Keyword_to_parsecode(const Keyword_parsecode* keywords, 1711 int keyword_count) 1712 : keyword_parsecodes_(keywords), keyword_count_(keyword_count) 1713 { } 1714 1715 // Return the parsecode corresponding KEYWORD, or 0 if it is not a 1716 // keyword. 1717 int 1718 keyword_to_parsecode(const char* keyword, size_t len) const; 1719 1720 private: 1721 const Keyword_parsecode* keyword_parsecodes_; 1722 const int keyword_count_; 1723 }; 1724 1725 // Mapping from keyword string to keyword parsecode. This array must 1726 // be kept in sorted order. Parsecodes are looked up using bsearch. 1727 // This array must correspond to the list of parsecodes in yyscript.y. 1728 1729 static const Keyword_to_parsecode::Keyword_parsecode 1730 script_keyword_parsecodes[] = 1731 { 1732 { "ABSOLUTE", ABSOLUTE }, 1733 { "ADDR", ADDR }, 1734 { "ALIGN", ALIGN_K }, 1735 { "ALIGNOF", ALIGNOF }, 1736 { "ASSERT", ASSERT_K }, 1737 { "AS_NEEDED", AS_NEEDED }, 1738 { "AT", AT }, 1739 { "BIND", BIND }, 1740 { "BLOCK", BLOCK }, 1741 { "BYTE", BYTE }, 1742 { "CONSTANT", CONSTANT }, 1743 { "CONSTRUCTORS", CONSTRUCTORS }, 1744 { "COPY", COPY }, 1745 { "CREATE_OBJECT_SYMBOLS", CREATE_OBJECT_SYMBOLS }, 1746 { "DATA_SEGMENT_ALIGN", DATA_SEGMENT_ALIGN }, 1747 { "DATA_SEGMENT_END", DATA_SEGMENT_END }, 1748 { "DATA_SEGMENT_RELRO_END", DATA_SEGMENT_RELRO_END }, 1749 { "DEFINED", DEFINED }, 1750 { "DSECT", DSECT }, 1751 { "ENTRY", ENTRY }, 1752 { "EXCLUDE_FILE", EXCLUDE_FILE }, 1753 { "EXTERN", EXTERN }, 1754 { "FILL", FILL }, 1755 { "FLOAT", FLOAT }, 1756 { "FORCE_COMMON_ALLOCATION", FORCE_COMMON_ALLOCATION }, 1757 { "GROUP", GROUP }, 1758 { "HLL", HLL }, 1759 { "INCLUDE", INCLUDE }, 1760 { "INFO", INFO }, 1761 { "INHIBIT_COMMON_ALLOCATION", INHIBIT_COMMON_ALLOCATION }, 1762 { "INPUT", INPUT }, 1763 { "KEEP", KEEP }, 1764 { "LENGTH", LENGTH }, 1765 { "LOADADDR", LOADADDR }, 1766 { "LONG", LONG }, 1767 { "MAP", MAP }, 1768 { "MAX", MAX_K }, 1769 { "MEMORY", MEMORY }, 1770 { "MIN", MIN_K }, 1771 { "NEXT", NEXT }, 1772 { "NOCROSSREFS", NOCROSSREFS }, 1773 { "NOFLOAT", NOFLOAT }, 1774 { "NOLOAD", NOLOAD }, 1775 { "ONLY_IF_RO", ONLY_IF_RO }, 1776 { "ONLY_IF_RW", ONLY_IF_RW }, 1777 { "OPTION", OPTION }, 1778 { "ORIGIN", ORIGIN }, 1779 { "OUTPUT", OUTPUT }, 1780 { "OUTPUT_ARCH", OUTPUT_ARCH }, 1781 { "OUTPUT_FORMAT", OUTPUT_FORMAT }, 1782 { "OVERLAY", OVERLAY }, 1783 { "PHDRS", PHDRS }, 1784 { "PROVIDE", PROVIDE }, 1785 { "PROVIDE_HIDDEN", PROVIDE_HIDDEN }, 1786 { "QUAD", QUAD }, 1787 { "SEARCH_DIR", SEARCH_DIR }, 1788 { "SECTIONS", SECTIONS }, 1789 { "SEGMENT_START", SEGMENT_START }, 1790 { "SHORT", SHORT }, 1791 { "SIZEOF", SIZEOF }, 1792 { "SIZEOF_HEADERS", SIZEOF_HEADERS }, 1793 { "SORT", SORT_BY_NAME }, 1794 { "SORT_BY_ALIGNMENT", SORT_BY_ALIGNMENT }, 1795 { "SORT_BY_INIT_PRIORITY", SORT_BY_INIT_PRIORITY }, 1796 { "SORT_BY_NAME", SORT_BY_NAME }, 1797 { "SPECIAL", SPECIAL }, 1798 { "SQUAD", SQUAD }, 1799 { "STARTUP", STARTUP }, 1800 { "SUBALIGN", SUBALIGN }, 1801 { "SYSLIB", SYSLIB }, 1802 { "TARGET", TARGET_K }, 1803 { "TRUNCATE", TRUNCATE }, 1804 { "VERSION", VERSIONK }, 1805 { "global", GLOBAL }, 1806 { "l", LENGTH }, 1807 { "len", LENGTH }, 1808 { "local", LOCAL }, 1809 { "o", ORIGIN }, 1810 { "org", ORIGIN }, 1811 { "sizeof_headers", SIZEOF_HEADERS }, 1812 }; 1813 1814 static const Keyword_to_parsecode 1815 script_keywords(&script_keyword_parsecodes[0], 1816 (sizeof(script_keyword_parsecodes) 1817 / sizeof(script_keyword_parsecodes[0]))); 1818 1819 static const Keyword_to_parsecode::Keyword_parsecode 1820 version_script_keyword_parsecodes[] = 1821 { 1822 { "extern", EXTERN }, 1823 { "global", GLOBAL }, 1824 { "local", LOCAL }, 1825 }; 1826 1827 static const Keyword_to_parsecode 1828 version_script_keywords(&version_script_keyword_parsecodes[0], 1829 (sizeof(version_script_keyword_parsecodes) 1830 / sizeof(version_script_keyword_parsecodes[0]))); 1831 1832 static const Keyword_to_parsecode::Keyword_parsecode 1833 dynamic_list_keyword_parsecodes[] = 1834 { 1835 { "extern", EXTERN }, 1836 }; 1837 1838 static const Keyword_to_parsecode 1839 dynamic_list_keywords(&dynamic_list_keyword_parsecodes[0], 1840 (sizeof(dynamic_list_keyword_parsecodes) 1841 / sizeof(dynamic_list_keyword_parsecodes[0]))); 1842 1843 1844 1845 // Comparison function passed to bsearch. 1846 1847 extern "C" 1848 { 1849 1850 struct Ktt_key 1851 { 1852 const char* str; 1853 size_t len; 1854 }; 1855 1856 static int 1857 ktt_compare(const void* keyv, const void* kttv) 1858 { 1859 const Ktt_key* key = static_cast<const Ktt_key*>(keyv); 1860 const Keyword_to_parsecode::Keyword_parsecode* ktt = 1861 static_cast<const Keyword_to_parsecode::Keyword_parsecode*>(kttv); 1862 int i = strncmp(key->str, ktt->keyword, key->len); 1863 if (i != 0) 1864 return i; 1865 if (ktt->keyword[key->len] != '\0') 1866 return -1; 1867 return 0; 1868 } 1869 1870 } // End extern "C". 1871 1872 int 1873 Keyword_to_parsecode::keyword_to_parsecode(const char* keyword, 1874 size_t len) const 1875 { 1876 Ktt_key key; 1877 key.str = keyword; 1878 key.len = len; 1879 void* kttv = bsearch(&key, 1880 this->keyword_parsecodes_, 1881 this->keyword_count_, 1882 sizeof(this->keyword_parsecodes_[0]), 1883 ktt_compare); 1884 if (kttv == NULL) 1885 return 0; 1886 Keyword_parsecode* ktt = static_cast<Keyword_parsecode*>(kttv); 1887 return ktt->parsecode; 1888 } 1889 1890 // The following structs are used within the VersionInfo class as well 1891 // as in the bison helper functions. They store the information 1892 // parsed from the version script. 1893 1894 // A single version expression. 1895 // For example, pattern="std::map*" and language="C++". 1896 struct Version_expression 1897 { 1898 Version_expression(const std::string& a_pattern, 1899 Version_script_info::Language a_language, 1900 bool a_exact_match) 1901 : pattern(a_pattern), language(a_language), exact_match(a_exact_match), 1902 was_matched_by_symbol(false) 1903 { } 1904 1905 std::string pattern; 1906 Version_script_info::Language language; 1907 // If false, we use glob() to match pattern. If true, we use strcmp(). 1908 bool exact_match; 1909 // True if --no-undefined-version is in effect and we found this 1910 // version in get_symbol_version. We use mutable because this 1911 // struct is generally not modifiable after it has been created. 1912 mutable bool was_matched_by_symbol; 1913 }; 1914 1915 // A list of expressions. 1916 struct Version_expression_list 1917 { 1918 std::vector<struct Version_expression> expressions; 1919 }; 1920 1921 // A list of which versions upon which another version depends. 1922 // Strings should be from the Stringpool. 1923 struct Version_dependency_list 1924 { 1925 std::vector<std::string> dependencies; 1926 }; 1927 1928 // The total definition of a version. It includes the tag for the 1929 // version, its global and local expressions, and any dependencies. 1930 struct Version_tree 1931 { 1932 Version_tree() 1933 : tag(), global(NULL), local(NULL), dependencies(NULL) 1934 { } 1935 1936 std::string tag; 1937 const struct Version_expression_list* global; 1938 const struct Version_expression_list* local; 1939 const struct Version_dependency_list* dependencies; 1940 }; 1941 1942 // Helper class that calls cplus_demangle when needed and takes care of freeing 1943 // the result. 1944 1945 class Lazy_demangler 1946 { 1947 public: 1948 Lazy_demangler(const char* symbol, int options) 1949 : symbol_(symbol), options_(options), demangled_(NULL), did_demangle_(false) 1950 { } 1951 1952 ~Lazy_demangler() 1953 { free(this->demangled_); } 1954 1955 // Return the demangled name. The actual demangling happens on the first call, 1956 // and the result is later cached. 1957 inline char* 1958 get(); 1959 1960 private: 1961 // The symbol to demangle. 1962 const char* symbol_; 1963 // Option flags to pass to cplus_demagle. 1964 const int options_; 1965 // The cached demangled value, or NULL if demangling didn't happen yet or 1966 // failed. 1967 char* demangled_; 1968 // Whether we already called cplus_demangle 1969 bool did_demangle_; 1970 }; 1971 1972 // Return the demangled name. The actual demangling happens on the first call, 1973 // and the result is later cached. Returns NULL if the symbol cannot be 1974 // demangled. 1975 1976 inline char* 1977 Lazy_demangler::get() 1978 { 1979 if (!this->did_demangle_) 1980 { 1981 this->demangled_ = cplus_demangle(this->symbol_, this->options_); 1982 this->did_demangle_ = true; 1983 } 1984 return this->demangled_; 1985 } 1986 1987 // Class Version_script_info. 1988 1989 Version_script_info::Version_script_info() 1990 : dependency_lists_(), expression_lists_(), version_trees_(), globs_(), 1991 default_version_(NULL), default_is_global_(false), is_finalized_(false) 1992 { 1993 for (int i = 0; i < LANGUAGE_COUNT; ++i) 1994 this->exact_[i] = NULL; 1995 } 1996 1997 Version_script_info::~Version_script_info() 1998 { 1999 } 2000 2001 // Forget all the known version script information. 2002 2003 void 2004 Version_script_info::clear() 2005 { 2006 for (size_t k = 0; k < this->dependency_lists_.size(); ++k) 2007 delete this->dependency_lists_[k]; 2008 this->dependency_lists_.clear(); 2009 for (size_t k = 0; k < this->version_trees_.size(); ++k) 2010 delete this->version_trees_[k]; 2011 this->version_trees_.clear(); 2012 for (size_t k = 0; k < this->expression_lists_.size(); ++k) 2013 delete this->expression_lists_[k]; 2014 this->expression_lists_.clear(); 2015 } 2016 2017 // Finalize the version script information. 2018 2019 void 2020 Version_script_info::finalize() 2021 { 2022 if (!this->is_finalized_) 2023 { 2024 this->build_lookup_tables(); 2025 this->is_finalized_ = true; 2026 } 2027 } 2028 2029 // Return all the versions. 2030 2031 std::vector<std::string> 2032 Version_script_info::get_versions() const 2033 { 2034 std::vector<std::string> ret; 2035 for (size_t j = 0; j < this->version_trees_.size(); ++j) 2036 if (!this->version_trees_[j]->tag.empty()) 2037 ret.push_back(this->version_trees_[j]->tag); 2038 return ret; 2039 } 2040 2041 // Return the dependencies of VERSION. 2042 2043 std::vector<std::string> 2044 Version_script_info::get_dependencies(const char* version) const 2045 { 2046 std::vector<std::string> ret; 2047 for (size_t j = 0; j < this->version_trees_.size(); ++j) 2048 if (this->version_trees_[j]->tag == version) 2049 { 2050 const struct Version_dependency_list* deps = 2051 this->version_trees_[j]->dependencies; 2052 if (deps != NULL) 2053 for (size_t k = 0; k < deps->dependencies.size(); ++k) 2054 ret.push_back(deps->dependencies[k]); 2055 return ret; 2056 } 2057 return ret; 2058 } 2059 2060 // A version script essentially maps a symbol name to a version tag 2061 // and an indication of whether symbol is global or local within that 2062 // version tag. Each symbol maps to at most one version tag. 2063 // Unfortunately, in practice, version scripts are ambiguous, and list 2064 // symbols multiple times. Thus, we have to document the matching 2065 // process. 2066 2067 // This is a description of what the GNU linker does as of 2010-01-11. 2068 // It walks through the version tags in the order in which they appear 2069 // in the version script. For each tag, it first walks through the 2070 // global patterns for that tag, then the local patterns. When 2071 // looking at a single pattern, it first applies any language specific 2072 // demangling as specified for the pattern, and then matches the 2073 // resulting symbol name to the pattern. If it finds an exact match 2074 // for a literal pattern (a pattern enclosed in quotes or with no 2075 // wildcard characters), then that is the match that it uses. If 2076 // finds a match with a wildcard pattern, then it saves it and 2077 // continues searching. Wildcard patterns that are exactly "*" are 2078 // saved separately. 2079 2080 // If no exact match with a literal pattern is ever found, then if a 2081 // wildcard match with a global pattern was found it is used, 2082 // otherwise if a wildcard match with a local pattern was found it is 2083 // used. 2084 2085 // This is the result: 2086 // * If there is an exact match, then we use the first tag in the 2087 // version script where it matches. 2088 // + If the exact match in that tag is global, it is used. 2089 // + Otherwise the exact match in that tag is local, and is used. 2090 // * Otherwise, if there is any match with a global wildcard pattern: 2091 // + If there is any match with a wildcard pattern which is not 2092 // "*", then we use the tag in which the *last* such pattern 2093 // appears. 2094 // + Otherwise, we matched "*". If there is no match with a local 2095 // wildcard pattern which is not "*", then we use the *last* 2096 // match with a global "*". Otherwise, continue. 2097 // * Otherwise, if there is any match with a local wildcard pattern: 2098 // + If there is any match with a wildcard pattern which is not 2099 // "*", then we use the tag in which the *last* such pattern 2100 // appears. 2101 // + Otherwise, we matched "*", and we use the tag in which the 2102 // *last* such match occurred. 2103 2104 // There is an additional wrinkle. When the GNU linker finds a symbol 2105 // with a version defined in an object file due to a .symver 2106 // directive, it looks up that symbol name in that version tag. If it 2107 // finds it, it matches the symbol name against the patterns for that 2108 // version. If there is no match with a global pattern, but there is 2109 // a match with a local pattern, then the GNU linker marks the symbol 2110 // as local. 2111 2112 // We want gold to be generally compatible, but we also want gold to 2113 // be fast. These are the rules that gold implements: 2114 // * If there is an exact match for the mangled name, we use it. 2115 // + If there is more than one exact match, we give a warning, and 2116 // we use the first tag in the script which matches. 2117 // + If a symbol has an exact match as both global and local for 2118 // the same version tag, we give an error. 2119 // * Otherwise, we look for an extern C++ or an extern Java exact 2120 // match. If we find an exact match, we use it. 2121 // + If there is more than one exact match, we give a warning, and 2122 // we use the first tag in the script which matches. 2123 // + If a symbol has an exact match as both global and local for 2124 // the same version tag, we give an error. 2125 // * Otherwise, we look through the wildcard patterns, ignoring "*" 2126 // patterns. We look through the version tags in reverse order. 2127 // For each version tag, we look through the global patterns and 2128 // then the local patterns. We use the first match we find (i.e., 2129 // the last matching version tag in the file). 2130 // * Otherwise, we use the "*" pattern if there is one. We give an 2131 // error if there are multiple "*" patterns. 2132 2133 // At least for now, gold does not look up the version tag for a 2134 // symbol version found in an object file to see if it should be 2135 // forced local. There are other ways to force a symbol to be local, 2136 // and I don't understand why this one is useful. 2137 2138 // Build a set of fast lookup tables for a version script. 2139 2140 void 2141 Version_script_info::build_lookup_tables() 2142 { 2143 size_t size = this->version_trees_.size(); 2144 for (size_t j = 0; j < size; ++j) 2145 { 2146 const Version_tree* v = this->version_trees_[j]; 2147 this->build_expression_list_lookup(v->local, v, false); 2148 this->build_expression_list_lookup(v->global, v, true); 2149 } 2150 } 2151 2152 // If a pattern has backlashes but no unquoted wildcard characters, 2153 // then we apply backslash unquoting and look for an exact match. 2154 // Otherwise we treat it as a wildcard pattern. This function returns 2155 // true for a wildcard pattern. Otherwise, it does backslash 2156 // unquoting on *PATTERN and returns false. If this returns true, 2157 // *PATTERN may have been partially unquoted. 2158 2159 bool 2160 Version_script_info::unquote(std::string* pattern) const 2161 { 2162 bool saw_backslash = false; 2163 size_t len = pattern->length(); 2164 size_t j = 0; 2165 for (size_t i = 0; i < len; ++i) 2166 { 2167 if (saw_backslash) 2168 saw_backslash = false; 2169 else 2170 { 2171 switch ((*pattern)[i]) 2172 { 2173 case '?': case '[': case '*': 2174 return true; 2175 case '\\': 2176 saw_backslash = true; 2177 continue; 2178 default: 2179 break; 2180 } 2181 } 2182 2183 if (i != j) 2184 (*pattern)[j] = (*pattern)[i]; 2185 ++j; 2186 } 2187 return false; 2188 } 2189 2190 // Add an exact match for MATCH to *PE. The result of the match is 2191 // V/IS_GLOBAL. 2192 2193 void 2194 Version_script_info::add_exact_match(const std::string& match, 2195 const Version_tree* v, bool is_global, 2196 const Version_expression* ve, 2197 Exact* pe) 2198 { 2199 std::pair<Exact::iterator, bool> ins = 2200 pe->insert(std::make_pair(match, Version_tree_match(v, is_global, ve))); 2201 if (ins.second) 2202 { 2203 // This is the first time we have seen this match. 2204 return; 2205 } 2206 2207 Version_tree_match& vtm(ins.first->second); 2208 if (vtm.real->tag != v->tag) 2209 { 2210 // This is an ambiguous match. We still return the 2211 // first version that we found in the script, but we 2212 // record the new version to issue a warning if we 2213 // wind up looking up this symbol. 2214 if (vtm.ambiguous == NULL) 2215 vtm.ambiguous = v; 2216 } 2217 else if (is_global != vtm.is_global) 2218 { 2219 // We have a match for both the global and local entries for a 2220 // version tag. That's got to be wrong. 2221 gold_error(_("'%s' appears as both a global and a local symbol " 2222 "for version '%s' in script"), 2223 match.c_str(), v->tag.c_str()); 2224 } 2225 } 2226 2227 // Build fast lookup information for EXPLIST and store it in LOOKUP. 2228 // All matches go to V, and IS_GLOBAL is true if they are global 2229 // matches. 2230 2231 void 2232 Version_script_info::build_expression_list_lookup( 2233 const Version_expression_list* explist, 2234 const Version_tree* v, 2235 bool is_global) 2236 { 2237 if (explist == NULL) 2238 return; 2239 size_t size = explist->expressions.size(); 2240 for (size_t i = 0; i < size; ++i) 2241 { 2242 const Version_expression& exp(explist->expressions[i]); 2243 2244 if (exp.pattern.length() == 1 && exp.pattern[0] == '*') 2245 { 2246 if (this->default_version_ != NULL 2247 && this->default_version_->tag != v->tag) 2248 gold_warning(_("wildcard match appears in both version '%s' " 2249 "and '%s' in script"), 2250 this->default_version_->tag.c_str(), v->tag.c_str()); 2251 else if (this->default_version_ != NULL 2252 && this->default_is_global_ != is_global) 2253 gold_error(_("wildcard match appears as both global and local " 2254 "in version '%s' in script"), 2255 v->tag.c_str()); 2256 this->default_version_ = v; 2257 this->default_is_global_ = is_global; 2258 continue; 2259 } 2260 2261 std::string pattern = exp.pattern; 2262 if (!exp.exact_match) 2263 { 2264 if (this->unquote(&pattern)) 2265 { 2266 this->globs_.push_back(Glob(&exp, v, is_global)); 2267 continue; 2268 } 2269 } 2270 2271 if (this->exact_[exp.language] == NULL) 2272 this->exact_[exp.language] = new Exact(); 2273 this->add_exact_match(pattern, v, is_global, &exp, 2274 this->exact_[exp.language]); 2275 } 2276 } 2277 2278 // Return the name to match given a name, a language code, and two 2279 // lazy demanglers. 2280 2281 const char* 2282 Version_script_info::get_name_to_match(const char* name, 2283 int language, 2284 Lazy_demangler* cpp_demangler, 2285 Lazy_demangler* java_demangler) const 2286 { 2287 switch (language) 2288 { 2289 case LANGUAGE_C: 2290 return name; 2291 case LANGUAGE_CXX: 2292 return cpp_demangler->get(); 2293 case LANGUAGE_JAVA: 2294 return java_demangler->get(); 2295 default: 2296 gold_unreachable(); 2297 } 2298 } 2299 2300 // Look up SYMBOL_NAME in the list of versions. Return true if the 2301 // symbol is found, false if not. If the symbol is found, then if 2302 // PVERSION is not NULL, set *PVERSION to the version tag, and if 2303 // P_IS_GLOBAL is not NULL, set *P_IS_GLOBAL according to whether the 2304 // symbol is global or not. 2305 2306 bool 2307 Version_script_info::get_symbol_version(const char* symbol_name, 2308 std::string* pversion, 2309 bool* p_is_global) const 2310 { 2311 Lazy_demangler cpp_demangled_name(symbol_name, DMGL_ANSI | DMGL_PARAMS); 2312 Lazy_demangler java_demangled_name(symbol_name, 2313 DMGL_ANSI | DMGL_PARAMS | DMGL_JAVA); 2314 2315 gold_assert(this->is_finalized_); 2316 for (int i = 0; i < LANGUAGE_COUNT; ++i) 2317 { 2318 Exact* exact = this->exact_[i]; 2319 if (exact == NULL) 2320 continue; 2321 2322 const char* name_to_match = this->get_name_to_match(symbol_name, i, 2323 &cpp_demangled_name, 2324 &java_demangled_name); 2325 if (name_to_match == NULL) 2326 { 2327 // If the name can not be demangled, the GNU linker goes 2328 // ahead and tries to match it anyhow. That does not 2329 // make sense to me and I have not implemented it. 2330 continue; 2331 } 2332 2333 Exact::const_iterator pe = exact->find(name_to_match); 2334 if (pe != exact->end()) 2335 { 2336 const Version_tree_match& vtm(pe->second); 2337 if (vtm.ambiguous != NULL) 2338 gold_warning(_("using '%s' as version for '%s' which is also " 2339 "named in version '%s' in script"), 2340 vtm.real->tag.c_str(), name_to_match, 2341 vtm.ambiguous->tag.c_str()); 2342 2343 if (pversion != NULL) 2344 *pversion = vtm.real->tag; 2345 if (p_is_global != NULL) 2346 *p_is_global = vtm.is_global; 2347 2348 // If we are using --no-undefined-version, and this is a 2349 // global symbol, we have to record that we have found this 2350 // symbol, so that we don't warn about it. We have to do 2351 // this now, because otherwise we have no way to get from a 2352 // non-C language back to the demangled name that we 2353 // matched. 2354 if (p_is_global != NULL && vtm.is_global) 2355 vtm.expression->was_matched_by_symbol = true; 2356 2357 return true; 2358 } 2359 } 2360 2361 // Look through the glob patterns in reverse order. 2362 2363 for (Globs::const_reverse_iterator p = this->globs_.rbegin(); 2364 p != this->globs_.rend(); 2365 ++p) 2366 { 2367 int language = p->expression->language; 2368 const char* name_to_match = this->get_name_to_match(symbol_name, 2369 language, 2370 &cpp_demangled_name, 2371 &java_demangled_name); 2372 if (name_to_match == NULL) 2373 continue; 2374 2375 if (fnmatch(p->expression->pattern.c_str(), name_to_match, 2376 FNM_NOESCAPE) == 0) 2377 { 2378 if (pversion != NULL) 2379 *pversion = p->version->tag; 2380 if (p_is_global != NULL) 2381 *p_is_global = p->is_global; 2382 return true; 2383 } 2384 } 2385 2386 // Finally, there may be a wildcard. 2387 if (this->default_version_ != NULL) 2388 { 2389 if (pversion != NULL) 2390 *pversion = this->default_version_->tag; 2391 if (p_is_global != NULL) 2392 *p_is_global = this->default_is_global_; 2393 return true; 2394 } 2395 2396 return false; 2397 } 2398 2399 // Give an error if any exact symbol names (not wildcards) appear in a 2400 // version script, but there is no such symbol. 2401 2402 void 2403 Version_script_info::check_unmatched_names(const Symbol_table* symtab) const 2404 { 2405 for (size_t i = 0; i < this->version_trees_.size(); ++i) 2406 { 2407 const Version_tree* vt = this->version_trees_[i]; 2408 if (vt->global == NULL) 2409 continue; 2410 for (size_t j = 0; j < vt->global->expressions.size(); ++j) 2411 { 2412 const Version_expression& expression(vt->global->expressions[j]); 2413 2414 // Ignore cases where we used the version because we saw a 2415 // symbol that we looked up. Note that 2416 // WAS_MATCHED_BY_SYMBOL will be true even if the symbol was 2417 // not a definition. That's OK as in that case we most 2418 // likely gave an undefined symbol error anyhow. 2419 if (expression.was_matched_by_symbol) 2420 continue; 2421 2422 // Just ignore names which are in languages other than C. 2423 // We have no way to look them up in the symbol table. 2424 if (expression.language != LANGUAGE_C) 2425 continue; 2426 2427 // Remove backslash quoting, and ignore wildcard patterns. 2428 std::string pattern = expression.pattern; 2429 if (!expression.exact_match) 2430 { 2431 if (this->unquote(&pattern)) 2432 continue; 2433 } 2434 2435 if (symtab->lookup(pattern.c_str(), vt->tag.c_str()) == NULL) 2436 gold_error(_("version script assignment of %s to symbol %s " 2437 "failed: symbol not defined"), 2438 vt->tag.c_str(), pattern.c_str()); 2439 } 2440 } 2441 } 2442 2443 struct Version_dependency_list* 2444 Version_script_info::allocate_dependency_list() 2445 { 2446 dependency_lists_.push_back(new Version_dependency_list); 2447 return dependency_lists_.back(); 2448 } 2449 2450 struct Version_expression_list* 2451 Version_script_info::allocate_expression_list() 2452 { 2453 expression_lists_.push_back(new Version_expression_list); 2454 return expression_lists_.back(); 2455 } 2456 2457 struct Version_tree* 2458 Version_script_info::allocate_version_tree() 2459 { 2460 version_trees_.push_back(new Version_tree); 2461 return version_trees_.back(); 2462 } 2463 2464 // Print for debugging. 2465 2466 void 2467 Version_script_info::print(FILE* f) const 2468 { 2469 if (this->empty()) 2470 return; 2471 2472 fprintf(f, "VERSION {"); 2473 2474 for (size_t i = 0; i < this->version_trees_.size(); ++i) 2475 { 2476 const Version_tree* vt = this->version_trees_[i]; 2477 2478 if (vt->tag.empty()) 2479 fprintf(f, " {\n"); 2480 else 2481 fprintf(f, " %s {\n", vt->tag.c_str()); 2482 2483 if (vt->global != NULL) 2484 { 2485 fprintf(f, " global :\n"); 2486 this->print_expression_list(f, vt->global); 2487 } 2488 2489 if (vt->local != NULL) 2490 { 2491 fprintf(f, " local :\n"); 2492 this->print_expression_list(f, vt->local); 2493 } 2494 2495 fprintf(f, " }"); 2496 if (vt->dependencies != NULL) 2497 { 2498 const Version_dependency_list* deps = vt->dependencies; 2499 for (size_t j = 0; j < deps->dependencies.size(); ++j) 2500 { 2501 if (j < deps->dependencies.size() - 1) 2502 fprintf(f, "\n"); 2503 fprintf(f, " %s", deps->dependencies[j].c_str()); 2504 } 2505 } 2506 fprintf(f, ";\n"); 2507 } 2508 2509 fprintf(f, "}\n"); 2510 } 2511 2512 void 2513 Version_script_info::print_expression_list( 2514 FILE* f, 2515 const Version_expression_list* vel) const 2516 { 2517 Version_script_info::Language current_language = LANGUAGE_C; 2518 for (size_t i = 0; i < vel->expressions.size(); ++i) 2519 { 2520 const Version_expression& ve(vel->expressions[i]); 2521 2522 if (ve.language != current_language) 2523 { 2524 if (current_language != LANGUAGE_C) 2525 fprintf(f, " }\n"); 2526 switch (ve.language) 2527 { 2528 case LANGUAGE_C: 2529 break; 2530 case LANGUAGE_CXX: 2531 fprintf(f, " extern \"C++\" {\n"); 2532 break; 2533 case LANGUAGE_JAVA: 2534 fprintf(f, " extern \"Java\" {\n"); 2535 break; 2536 default: 2537 gold_unreachable(); 2538 } 2539 current_language = ve.language; 2540 } 2541 2542 fprintf(f, " "); 2543 if (current_language != LANGUAGE_C) 2544 fprintf(f, " "); 2545 2546 if (ve.exact_match) 2547 fprintf(f, "\""); 2548 fprintf(f, "%s", ve.pattern.c_str()); 2549 if (ve.exact_match) 2550 fprintf(f, "\""); 2551 2552 fprintf(f, "\n"); 2553 } 2554 2555 if (current_language != LANGUAGE_C) 2556 fprintf(f, " }\n"); 2557 } 2558 2559 } // End namespace gold. 2560 2561 // The remaining functions are extern "C", so it's clearer to not put 2562 // them in namespace gold. 2563 2564 using namespace gold; 2565 2566 // This function is called by the bison parser to return the next 2567 // token. 2568 2569 extern "C" int 2570 yylex(YYSTYPE* lvalp, void* closurev) 2571 { 2572 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2573 const Token* token = closure->next_token(); 2574 switch (token->classification()) 2575 { 2576 default: 2577 gold_unreachable(); 2578 2579 case Token::TOKEN_INVALID: 2580 yyerror(closurev, "invalid character"); 2581 return 0; 2582 2583 case Token::TOKEN_EOF: 2584 return 0; 2585 2586 case Token::TOKEN_STRING: 2587 { 2588 // This is either a keyword or a STRING. 2589 size_t len; 2590 const char* str = token->string_value(&len); 2591 int parsecode = 0; 2592 switch (closure->lex_mode()) 2593 { 2594 case Lex::LINKER_SCRIPT: 2595 parsecode = script_keywords.keyword_to_parsecode(str, len); 2596 break; 2597 case Lex::VERSION_SCRIPT: 2598 parsecode = version_script_keywords.keyword_to_parsecode(str, len); 2599 break; 2600 case Lex::DYNAMIC_LIST: 2601 parsecode = dynamic_list_keywords.keyword_to_parsecode(str, len); 2602 break; 2603 default: 2604 break; 2605 } 2606 if (parsecode != 0) 2607 return parsecode; 2608 lvalp->string.value = str; 2609 lvalp->string.length = len; 2610 return STRING; 2611 } 2612 2613 case Token::TOKEN_QUOTED_STRING: 2614 lvalp->string.value = token->string_value(&lvalp->string.length); 2615 return QUOTED_STRING; 2616 2617 case Token::TOKEN_OPERATOR: 2618 return token->operator_value(); 2619 2620 case Token::TOKEN_INTEGER: 2621 lvalp->integer = token->integer_value(); 2622 return INTEGER; 2623 } 2624 } 2625 2626 // This function is called by the bison parser to report an error. 2627 2628 extern "C" void 2629 yyerror(void* closurev, const char* message) 2630 { 2631 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2632 gold_error(_("%s:%d:%d: %s"), closure->filename(), closure->lineno(), 2633 closure->charpos(), message); 2634 } 2635 2636 // Called by the bison parser to add an external symbol to the link. 2637 2638 extern "C" void 2639 script_add_extern(void* closurev, const char* name, size_t length) 2640 { 2641 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2642 closure->script_options()->add_symbol_reference(name, length); 2643 } 2644 2645 // Called by the bison parser to add a file to the link. 2646 2647 extern "C" void 2648 script_add_file(void* closurev, const char* name, size_t length) 2649 { 2650 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2651 2652 // If this is an absolute path, and we found the script in the 2653 // sysroot, then we want to prepend the sysroot to the file name. 2654 // For example, this is how we handle a cross link to the x86_64 2655 // libc.so, which refers to /lib/libc.so.6. 2656 std::string name_string(name, length); 2657 const char* extra_search_path = "."; 2658 std::string script_directory; 2659 if (IS_ABSOLUTE_PATH(name_string.c_str())) 2660 { 2661 if (closure->is_in_sysroot()) 2662 { 2663 const std::string& sysroot(parameters->options().sysroot()); 2664 gold_assert(!sysroot.empty()); 2665 name_string = sysroot + name_string; 2666 } 2667 } 2668 else 2669 { 2670 // In addition to checking the normal library search path, we 2671 // also want to check in the script-directory. 2672 const char* slash = strrchr(closure->filename(), '/'); 2673 if (slash != NULL) 2674 { 2675 script_directory.assign(closure->filename(), 2676 slash - closure->filename() + 1); 2677 extra_search_path = script_directory.c_str(); 2678 } 2679 } 2680 2681 Input_file_argument file(name_string.c_str(), 2682 Input_file_argument::INPUT_FILE_TYPE_FILE, 2683 extra_search_path, false, 2684 closure->position_dependent_options()); 2685 Input_argument& arg = closure->inputs()->add_file(file); 2686 arg.set_script_info(closure->script_info()); 2687 } 2688 2689 // Called by the bison parser to add a library to the link. 2690 2691 extern "C" void 2692 script_add_library(void* closurev, const char* name, size_t length) 2693 { 2694 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2695 std::string name_string(name, length); 2696 2697 if (name_string[0] != 'l') 2698 gold_error(_("library name must be prefixed with -l")); 2699 2700 Input_file_argument file(name_string.c_str() + 1, 2701 Input_file_argument::INPUT_FILE_TYPE_LIBRARY, 2702 "", false, 2703 closure->position_dependent_options()); 2704 Input_argument& arg = closure->inputs()->add_file(file); 2705 arg.set_script_info(closure->script_info()); 2706 } 2707 2708 // Called by the bison parser to start a group. If we are already in 2709 // a group, that means that this script was invoked within a 2710 // --start-group --end-group sequence on the command line, or that 2711 // this script was found in a GROUP of another script. In that case, 2712 // we simply continue the existing group, rather than starting a new 2713 // one. It is possible to construct a case in which this will do 2714 // something other than what would happen if we did a recursive group, 2715 // but it's hard to imagine why the different behaviour would be 2716 // useful for a real program. Avoiding recursive groups is simpler 2717 // and more efficient. 2718 2719 extern "C" void 2720 script_start_group(void* closurev) 2721 { 2722 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2723 if (!closure->in_group()) 2724 closure->inputs()->start_group(); 2725 } 2726 2727 // Called by the bison parser at the end of a group. 2728 2729 extern "C" void 2730 script_end_group(void* closurev) 2731 { 2732 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2733 if (!closure->in_group()) 2734 closure->inputs()->end_group(); 2735 } 2736 2737 // Called by the bison parser to start an AS_NEEDED list. 2738 2739 extern "C" void 2740 script_start_as_needed(void* closurev) 2741 { 2742 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2743 closure->position_dependent_options().set_as_needed(true); 2744 } 2745 2746 // Called by the bison parser at the end of an AS_NEEDED list. 2747 2748 extern "C" void 2749 script_end_as_needed(void* closurev) 2750 { 2751 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2752 closure->position_dependent_options().set_as_needed(false); 2753 } 2754 2755 // Called by the bison parser to set the entry symbol. 2756 2757 extern "C" void 2758 script_set_entry(void* closurev, const char* entry, size_t length) 2759 { 2760 // We'll parse this exactly the same as --entry=ENTRY on the commandline 2761 // TODO(csilvers): FIXME -- call set_entry directly. 2762 std::string arg("--entry="); 2763 arg.append(entry, length); 2764 script_parse_option(closurev, arg.c_str(), arg.size()); 2765 } 2766 2767 // Called by the bison parser to set whether to define common symbols. 2768 2769 extern "C" void 2770 script_set_common_allocation(void* closurev, int set) 2771 { 2772 const char* arg = set != 0 ? "--define-common" : "--no-define-common"; 2773 script_parse_option(closurev, arg, strlen(arg)); 2774 } 2775 2776 // Called by the bison parser to refer to a symbol. 2777 2778 extern "C" Expression* 2779 script_symbol(void* closurev, const char* name, size_t length) 2780 { 2781 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2782 if (length != 1 || name[0] != '.') 2783 closure->script_options()->add_symbol_reference(name, length); 2784 return script_exp_string(name, length); 2785 } 2786 2787 // Called by the bison parser to define a symbol. 2788 2789 extern "C" void 2790 script_set_symbol(void* closurev, const char* name, size_t length, 2791 Expression* value, int providei, int hiddeni) 2792 { 2793 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2794 const bool provide = providei != 0; 2795 const bool hidden = hiddeni != 0; 2796 closure->script_options()->add_symbol_assignment(name, length, 2797 closure->parsing_defsym(), 2798 value, provide, hidden); 2799 closure->clear_skip_on_incompatible_target(); 2800 } 2801 2802 // Called by the bison parser to add an assertion. 2803 2804 extern "C" void 2805 script_add_assertion(void* closurev, Expression* check, const char* message, 2806 size_t messagelen) 2807 { 2808 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2809 closure->script_options()->add_assertion(check, message, messagelen); 2810 closure->clear_skip_on_incompatible_target(); 2811 } 2812 2813 // Called by the bison parser to parse an OPTION. 2814 2815 extern "C" void 2816 script_parse_option(void* closurev, const char* option, size_t length) 2817 { 2818 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2819 // We treat the option as a single command-line option, even if 2820 // it has internal whitespace. 2821 if (closure->command_line() == NULL) 2822 { 2823 // There are some options that we could handle here--e.g., 2824 // -lLIBRARY. Should we bother? 2825 gold_warning(_("%s:%d:%d: ignoring command OPTION; OPTION is only valid" 2826 " for scripts specified via -T/--script"), 2827 closure->filename(), closure->lineno(), closure->charpos()); 2828 } 2829 else 2830 { 2831 bool past_a_double_dash_option = false; 2832 const char* mutable_option = strndup(option, length); 2833 gold_assert(mutable_option != NULL); 2834 closure->command_line()->process_one_option(1, &mutable_option, 0, 2835 &past_a_double_dash_option); 2836 // The General_options class will quite possibly store a pointer 2837 // into mutable_option, so we can't free it. In cases the class 2838 // does not store such a pointer, this is a memory leak. Alas. :( 2839 } 2840 closure->clear_skip_on_incompatible_target(); 2841 } 2842 2843 // Called by the bison parser to handle OUTPUT_FORMAT. OUTPUT_FORMAT 2844 // takes either one or three arguments. In the three argument case, 2845 // the format depends on the endianness option, which we don't 2846 // currently support (FIXME). If we see an OUTPUT_FORMAT for the 2847 // wrong format, then we want to search for a new file. Returning 0 2848 // here will cause the parser to immediately abort. 2849 2850 extern "C" int 2851 script_check_output_format(void* closurev, 2852 const char* default_name, size_t default_length, 2853 const char*, size_t, const char*, size_t) 2854 { 2855 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2856 std::string name(default_name, default_length); 2857 Target* target = select_target_by_bfd_name(name.c_str()); 2858 if (target == NULL || !parameters->is_compatible_target(target)) 2859 { 2860 if (closure->skip_on_incompatible_target()) 2861 { 2862 closure->set_found_incompatible_target(); 2863 return 0; 2864 } 2865 // FIXME: Should we warn about the unknown target? 2866 } 2867 return 1; 2868 } 2869 2870 // Called by the bison parser to handle TARGET. 2871 2872 extern "C" void 2873 script_set_target(void* closurev, const char* target, size_t len) 2874 { 2875 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2876 std::string s(target, len); 2877 General_options::Object_format format_enum; 2878 format_enum = General_options::string_to_object_format(s.c_str()); 2879 closure->position_dependent_options().set_format_enum(format_enum); 2880 } 2881 2882 // Called by the bison parser to handle SEARCH_DIR. This is handled 2883 // exactly like a -L option. 2884 2885 extern "C" void 2886 script_add_search_dir(void* closurev, const char* option, size_t length) 2887 { 2888 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2889 if (closure->command_line() == NULL) 2890 gold_warning(_("%s:%d:%d: ignoring SEARCH_DIR; SEARCH_DIR is only valid" 2891 " for scripts specified via -T/--script"), 2892 closure->filename(), closure->lineno(), closure->charpos()); 2893 else if (!closure->command_line()->options().nostdlib()) 2894 { 2895 std::string s = "-L" + std::string(option, length); 2896 script_parse_option(closurev, s.c_str(), s.size()); 2897 } 2898 } 2899 2900 /* Called by the bison parser to push the lexer into expression 2901 mode. */ 2902 2903 extern "C" void 2904 script_push_lex_into_expression_mode(void* closurev) 2905 { 2906 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2907 closure->push_lex_mode(Lex::EXPRESSION); 2908 } 2909 2910 /* Called by the bison parser to push the lexer into version 2911 mode. */ 2912 2913 extern "C" void 2914 script_push_lex_into_version_mode(void* closurev) 2915 { 2916 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2917 if (closure->version_script()->is_finalized()) 2918 gold_error(_("%s:%d:%d: invalid use of VERSION in input file"), 2919 closure->filename(), closure->lineno(), closure->charpos()); 2920 closure->push_lex_mode(Lex::VERSION_SCRIPT); 2921 } 2922 2923 /* Called by the bison parser to pop the lexer mode. */ 2924 2925 extern "C" void 2926 script_pop_lex_mode(void* closurev) 2927 { 2928 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2929 closure->pop_lex_mode(); 2930 } 2931 2932 // Register an entire version node. For example: 2933 // 2934 // GLIBC_2.1 { 2935 // global: foo; 2936 // } GLIBC_2.0; 2937 // 2938 // - tag is "GLIBC_2.1" 2939 // - tree contains the information "global: foo" 2940 // - deps contains "GLIBC_2.0" 2941 2942 extern "C" void 2943 script_register_vers_node(void*, 2944 const char* tag, 2945 int taglen, 2946 struct Version_tree* tree, 2947 struct Version_dependency_list* deps) 2948 { 2949 gold_assert(tree != NULL); 2950 tree->dependencies = deps; 2951 if (tag != NULL) 2952 tree->tag = std::string(tag, taglen); 2953 } 2954 2955 // Add a dependencies to the list of existing dependencies, if any, 2956 // and return the expanded list. 2957 2958 extern "C" struct Version_dependency_list* 2959 script_add_vers_depend(void* closurev, 2960 struct Version_dependency_list* all_deps, 2961 const char* depend_to_add, int deplen) 2962 { 2963 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2964 if (all_deps == NULL) 2965 all_deps = closure->version_script()->allocate_dependency_list(); 2966 all_deps->dependencies.push_back(std::string(depend_to_add, deplen)); 2967 return all_deps; 2968 } 2969 2970 // Add a pattern expression to an existing list of expressions, if any. 2971 2972 extern "C" struct Version_expression_list* 2973 script_new_vers_pattern(void* closurev, 2974 struct Version_expression_list* expressions, 2975 const char* pattern, int patlen, int exact_match) 2976 { 2977 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2978 if (expressions == NULL) 2979 expressions = closure->version_script()->allocate_expression_list(); 2980 expressions->expressions.push_back( 2981 Version_expression(std::string(pattern, patlen), 2982 closure->get_current_language(), 2983 static_cast<bool>(exact_match))); 2984 return expressions; 2985 } 2986 2987 // Attaches b to the end of a, and clears b. So a = a + b and b = {}. 2988 2989 extern "C" struct Version_expression_list* 2990 script_merge_expressions(struct Version_expression_list* a, 2991 struct Version_expression_list* b) 2992 { 2993 a->expressions.insert(a->expressions.end(), 2994 b->expressions.begin(), b->expressions.end()); 2995 // We could delete b and remove it from expressions_lists_, but 2996 // that's a lot of work. This works just as well. 2997 b->expressions.clear(); 2998 return a; 2999 } 3000 3001 // Combine the global and local expressions into a a Version_tree. 3002 3003 extern "C" struct Version_tree* 3004 script_new_vers_node(void* closurev, 3005 struct Version_expression_list* global, 3006 struct Version_expression_list* local) 3007 { 3008 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3009 Version_tree* tree = closure->version_script()->allocate_version_tree(); 3010 tree->global = global; 3011 tree->local = local; 3012 return tree; 3013 } 3014 3015 // Handle a transition in language, such as at the 3016 // start or end of 'extern "C++"' 3017 3018 extern "C" void 3019 version_script_push_lang(void* closurev, const char* lang, int langlen) 3020 { 3021 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3022 std::string language(lang, langlen); 3023 Version_script_info::Language code; 3024 if (language.empty() || language == "C") 3025 code = Version_script_info::LANGUAGE_C; 3026 else if (language == "C++") 3027 code = Version_script_info::LANGUAGE_CXX; 3028 else if (language == "Java") 3029 code = Version_script_info::LANGUAGE_JAVA; 3030 else 3031 { 3032 char* buf = new char[langlen + 100]; 3033 snprintf(buf, langlen + 100, 3034 _("unrecognized version script language '%s'"), 3035 language.c_str()); 3036 yyerror(closurev, buf); 3037 delete[] buf; 3038 code = Version_script_info::LANGUAGE_C; 3039 } 3040 closure->push_language(code); 3041 } 3042 3043 extern "C" void 3044 version_script_pop_lang(void* closurev) 3045 { 3046 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3047 closure->pop_language(); 3048 } 3049 3050 // Called by the bison parser to start a SECTIONS clause. 3051 3052 extern "C" void 3053 script_start_sections(void* closurev) 3054 { 3055 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3056 closure->script_options()->script_sections()->start_sections(); 3057 closure->clear_skip_on_incompatible_target(); 3058 } 3059 3060 // Called by the bison parser to finish a SECTIONS clause. 3061 3062 extern "C" void 3063 script_finish_sections(void* closurev) 3064 { 3065 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3066 closure->script_options()->script_sections()->finish_sections(); 3067 } 3068 3069 // Start processing entries for an output section. 3070 3071 extern "C" void 3072 script_start_output_section(void* closurev, const char* name, size_t namelen, 3073 const struct Parser_output_section_header* header) 3074 { 3075 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3076 closure->script_options()->script_sections()->start_output_section(name, 3077 namelen, 3078 header); 3079 } 3080 3081 // Finish processing entries for an output section. 3082 3083 extern "C" void 3084 script_finish_output_section(void* closurev, 3085 const struct Parser_output_section_trailer* trail) 3086 { 3087 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3088 closure->script_options()->script_sections()->finish_output_section(trail); 3089 } 3090 3091 // Add a data item (e.g., "WORD (0)") to the current output section. 3092 3093 extern "C" void 3094 script_add_data(void* closurev, int data_token, Expression* val) 3095 { 3096 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3097 int size; 3098 bool is_signed = true; 3099 switch (data_token) 3100 { 3101 case QUAD: 3102 size = 8; 3103 is_signed = false; 3104 break; 3105 case SQUAD: 3106 size = 8; 3107 break; 3108 case LONG: 3109 size = 4; 3110 break; 3111 case SHORT: 3112 size = 2; 3113 break; 3114 case BYTE: 3115 size = 1; 3116 break; 3117 default: 3118 gold_unreachable(); 3119 } 3120 closure->script_options()->script_sections()->add_data(size, is_signed, val); 3121 } 3122 3123 // Add a clause setting the fill value to the current output section. 3124 3125 extern "C" void 3126 script_add_fill(void* closurev, Expression* val) 3127 { 3128 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3129 closure->script_options()->script_sections()->add_fill(val); 3130 } 3131 3132 // Add a new input section specification to the current output 3133 // section. 3134 3135 extern "C" void 3136 script_add_input_section(void* closurev, 3137 const struct Input_section_spec* spec, 3138 int keepi) 3139 { 3140 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3141 bool keep = keepi != 0; 3142 closure->script_options()->script_sections()->add_input_section(spec, keep); 3143 } 3144 3145 // When we see DATA_SEGMENT_ALIGN we record that following output 3146 // sections may be relro. 3147 3148 extern "C" void 3149 script_data_segment_align(void* closurev) 3150 { 3151 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3152 if (!closure->script_options()->saw_sections_clause()) 3153 gold_error(_("%s:%d:%d: DATA_SEGMENT_ALIGN not in SECTIONS clause"), 3154 closure->filename(), closure->lineno(), closure->charpos()); 3155 else 3156 closure->script_options()->script_sections()->data_segment_align(); 3157 } 3158 3159 // When we see DATA_SEGMENT_RELRO_END we know that all output sections 3160 // since DATA_SEGMENT_ALIGN should be relro. 3161 3162 extern "C" void 3163 script_data_segment_relro_end(void* closurev) 3164 { 3165 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3166 if (!closure->script_options()->saw_sections_clause()) 3167 gold_error(_("%s:%d:%d: DATA_SEGMENT_ALIGN not in SECTIONS clause"), 3168 closure->filename(), closure->lineno(), closure->charpos()); 3169 else 3170 closure->script_options()->script_sections()->data_segment_relro_end(); 3171 } 3172 3173 // Create a new list of string/sort pairs. 3174 3175 extern "C" String_sort_list_ptr 3176 script_new_string_sort_list(const struct Wildcard_section* string_sort) 3177 { 3178 return new String_sort_list(1, *string_sort); 3179 } 3180 3181 // Add an entry to a list of string/sort pairs. The way the parser 3182 // works permits us to simply modify the first parameter, rather than 3183 // copy the vector. 3184 3185 extern "C" String_sort_list_ptr 3186 script_string_sort_list_add(String_sort_list_ptr pv, 3187 const struct Wildcard_section* string_sort) 3188 { 3189 if (pv == NULL) 3190 return script_new_string_sort_list(string_sort); 3191 else 3192 { 3193 pv->push_back(*string_sort); 3194 return pv; 3195 } 3196 } 3197 3198 // Create a new list of strings. 3199 3200 extern "C" String_list_ptr 3201 script_new_string_list(const char* str, size_t len) 3202 { 3203 return new String_list(1, std::string(str, len)); 3204 } 3205 3206 // Add an element to a list of strings. The way the parser works 3207 // permits us to simply modify the first parameter, rather than copy 3208 // the vector. 3209 3210 extern "C" String_list_ptr 3211 script_string_list_push_back(String_list_ptr pv, const char* str, size_t len) 3212 { 3213 if (pv == NULL) 3214 return script_new_string_list(str, len); 3215 else 3216 { 3217 pv->push_back(std::string(str, len)); 3218 return pv; 3219 } 3220 } 3221 3222 // Concatenate two string lists. Either or both may be NULL. The way 3223 // the parser works permits us to modify the parameters, rather than 3224 // copy the vector. 3225 3226 extern "C" String_list_ptr 3227 script_string_list_append(String_list_ptr pv1, String_list_ptr pv2) 3228 { 3229 if (pv1 == NULL) 3230 return pv2; 3231 if (pv2 == NULL) 3232 return pv1; 3233 pv1->insert(pv1->end(), pv2->begin(), pv2->end()); 3234 return pv1; 3235 } 3236 3237 // Add a new program header. 3238 3239 extern "C" void 3240 script_add_phdr(void* closurev, const char* name, size_t namelen, 3241 unsigned int type, const Phdr_info* info) 3242 { 3243 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3244 bool includes_filehdr = info->includes_filehdr != 0; 3245 bool includes_phdrs = info->includes_phdrs != 0; 3246 bool is_flags_valid = info->is_flags_valid != 0; 3247 Script_sections* ss = closure->script_options()->script_sections(); 3248 ss->add_phdr(name, namelen, type, includes_filehdr, includes_phdrs, 3249 is_flags_valid, info->flags, info->load_address); 3250 closure->clear_skip_on_incompatible_target(); 3251 } 3252 3253 // Convert a program header string to a type. 3254 3255 #define PHDR_TYPE(NAME) { #NAME, sizeof(#NAME) - 1, elfcpp::NAME } 3256 3257 static struct 3258 { 3259 const char* name; 3260 size_t namelen; 3261 unsigned int val; 3262 } phdr_type_names[] = 3263 { 3264 PHDR_TYPE(PT_NULL), 3265 PHDR_TYPE(PT_LOAD), 3266 PHDR_TYPE(PT_DYNAMIC), 3267 PHDR_TYPE(PT_INTERP), 3268 PHDR_TYPE(PT_NOTE), 3269 PHDR_TYPE(PT_SHLIB), 3270 PHDR_TYPE(PT_PHDR), 3271 PHDR_TYPE(PT_TLS), 3272 PHDR_TYPE(PT_GNU_EH_FRAME), 3273 PHDR_TYPE(PT_GNU_STACK), 3274 PHDR_TYPE(PT_GNU_RELRO) 3275 }; 3276 3277 extern "C" unsigned int 3278 script_phdr_string_to_type(void* closurev, const char* name, size_t namelen) 3279 { 3280 for (unsigned int i = 0; 3281 i < sizeof(phdr_type_names) / sizeof(phdr_type_names[0]); 3282 ++i) 3283 if (namelen == phdr_type_names[i].namelen 3284 && strncmp(name, phdr_type_names[i].name, namelen) == 0) 3285 return phdr_type_names[i].val; 3286 yyerror(closurev, _("unknown PHDR type (try integer)")); 3287 return elfcpp::PT_NULL; 3288 } 3289 3290 extern "C" void 3291 script_saw_segment_start_expression(void* closurev) 3292 { 3293 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3294 Script_sections* ss = closure->script_options()->script_sections(); 3295 ss->set_saw_segment_start_expression(true); 3296 } 3297 3298 extern "C" void 3299 script_set_section_region(void* closurev, const char* name, size_t namelen, 3300 int set_vma) 3301 { 3302 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3303 if (!closure->script_options()->saw_sections_clause()) 3304 { 3305 gold_error(_("%s:%d:%d: MEMORY region '%.*s' referred to outside of " 3306 "SECTIONS clause"), 3307 closure->filename(), closure->lineno(), closure->charpos(), 3308 static_cast<int>(namelen), name); 3309 return; 3310 } 3311 3312 Script_sections* ss = closure->script_options()->script_sections(); 3313 Memory_region* mr = ss->find_memory_region(name, namelen); 3314 if (mr == NULL) 3315 { 3316 gold_error(_("%s:%d:%d: MEMORY region '%.*s' not declared"), 3317 closure->filename(), closure->lineno(), closure->charpos(), 3318 static_cast<int>(namelen), name); 3319 return; 3320 } 3321 3322 ss->set_memory_region(mr, set_vma); 3323 } 3324 3325 extern "C" void 3326 script_add_memory(void* closurev, const char* name, size_t namelen, 3327 unsigned int attrs, Expression* origin, Expression* length) 3328 { 3329 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3330 Script_sections* ss = closure->script_options()->script_sections(); 3331 ss->add_memory_region(name, namelen, attrs, origin, length); 3332 } 3333 3334 extern "C" unsigned int 3335 script_parse_memory_attr(void* closurev, const char* attrs, size_t attrlen, 3336 int invert) 3337 { 3338 int attributes = 0; 3339 3340 while (attrlen--) 3341 switch (*attrs++) 3342 { 3343 case 'R': 3344 case 'r': 3345 attributes |= MEM_READABLE; break; 3346 case 'W': 3347 case 'w': 3348 attributes |= MEM_READABLE | MEM_WRITEABLE; break; 3349 case 'X': 3350 case 'x': 3351 attributes |= MEM_EXECUTABLE; break; 3352 case 'A': 3353 case 'a': 3354 attributes |= MEM_ALLOCATABLE; break; 3355 case 'I': 3356 case 'i': 3357 case 'L': 3358 case 'l': 3359 attributes |= MEM_INITIALIZED; break; 3360 default: 3361 yyerror(closurev, _("unknown MEMORY attribute")); 3362 } 3363 3364 if (invert) 3365 attributes = (~ attributes) & MEM_ATTR_MASK; 3366 3367 return attributes; 3368 } 3369 3370 extern "C" void 3371 script_include_directive(int first_token, void* closurev, 3372 const char* filename, size_t length) 3373 { 3374 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3375 std::string name(filename, length); 3376 Command_line* cmdline = closure->command_line(); 3377 read_script_file(name.c_str(), cmdline, &cmdline->script_options(), 3378 first_token, Lex::LINKER_SCRIPT); 3379 } 3380 3381 // Functions for memory regions. 3382 3383 extern "C" Expression* 3384 script_exp_function_origin(void* closurev, const char* name, size_t namelen) 3385 { 3386 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3387 Script_sections* ss = closure->script_options()->script_sections(); 3388 Expression* origin = ss->find_memory_region_origin(name, namelen); 3389 3390 if (origin == NULL) 3391 { 3392 gold_error(_("undefined memory region '%s' referenced " 3393 "in ORIGIN expression"), 3394 name); 3395 // Create a dummy expression to prevent crashes later on. 3396 origin = script_exp_integer(0); 3397 } 3398 3399 return origin; 3400 } 3401 3402 extern "C" Expression* 3403 script_exp_function_length(void* closurev, const char* name, size_t namelen) 3404 { 3405 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3406 Script_sections* ss = closure->script_options()->script_sections(); 3407 Expression* length = ss->find_memory_region_length(name, namelen); 3408 3409 if (length == NULL) 3410 { 3411 gold_error(_("undefined memory region '%s' referenced " 3412 "in LENGTH expression"), 3413 name); 3414 // Create a dummy expression to prevent crashes later on. 3415 length = script_exp_integer(0); 3416 } 3417 3418 return length; 3419 } 3420