1// script.cc -- handle linker scripts for gold. 2 3// Copyright (C) 2006-2017 Free Software Foundation, Inc. 4// Written by Ian Lance Taylor <iant@google.com>. 5 6// This file is part of gold. 7 8// This program is free software; you can redistribute it and/or modify 9// it under the terms of the GNU General Public License as published by 10// the Free Software Foundation; either version 3 of the License, or 11// (at your option) any later version. 12 13// This program is distributed in the hope that it will be useful, 14// but WITHOUT ANY WARRANTY; without even the implied warranty of 15// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 16// GNU General Public License for more details. 17 18// You should have received a copy of the GNU General Public License 19// along with this program; if not, write to the Free Software 20// Foundation, Inc., 51 Franklin Street - Fifth Floor, Boston, 21// MA 02110-1301, USA. 22 23#include "gold.h" 24 25#include <cstdio> 26#include <cstdlib> 27#include <cstring> 28#include <fnmatch.h> 29#include <string> 30#include <vector> 31#include "filenames.h" 32 33#include "elfcpp.h" 34#include "demangle.h" 35#include "dirsearch.h" 36#include "options.h" 37#include "fileread.h" 38#include "workqueue.h" 39#include "readsyms.h" 40#include "parameters.h" 41#include "layout.h" 42#include "symtab.h" 43#include "target-select.h" 44#include "script.h" 45#include "script-c.h" 46#include "incremental.h" 47 48namespace gold 49{ 50 51// A token read from a script file. We don't implement keywords here; 52// all keywords are simply represented as a string. 53 54class Token 55{ 56 public: 57 // Token classification. 58 enum Classification 59 { 60 // Token is invalid. 61 TOKEN_INVALID, 62 // Token indicates end of input. 63 TOKEN_EOF, 64 // Token is a string of characters. 65 TOKEN_STRING, 66 // Token is a quoted string of characters. 67 TOKEN_QUOTED_STRING, 68 // Token is an operator. 69 TOKEN_OPERATOR, 70 // Token is a number (an integer). 71 TOKEN_INTEGER 72 }; 73 74 // We need an empty constructor so that we can put this STL objects. 75 Token() 76 : classification_(TOKEN_INVALID), value_(NULL), value_length_(0), 77 opcode_(0), lineno_(0), charpos_(0) 78 { } 79 80 // A general token with no value. 81 Token(Classification classification, int lineno, int charpos) 82 : classification_(classification), value_(NULL), value_length_(0), 83 opcode_(0), lineno_(lineno), charpos_(charpos) 84 { 85 gold_assert(classification == TOKEN_INVALID 86 || classification == TOKEN_EOF); 87 } 88 89 // A general token with a value. 90 Token(Classification classification, const char* value, size_t length, 91 int lineno, int charpos) 92 : classification_(classification), value_(value), value_length_(length), 93 opcode_(0), lineno_(lineno), charpos_(charpos) 94 { 95 gold_assert(classification != TOKEN_INVALID 96 && classification != TOKEN_EOF); 97 } 98 99 // A token representing an operator. 100 Token(int opcode, int lineno, int charpos) 101 : classification_(TOKEN_OPERATOR), value_(NULL), value_length_(0), 102 opcode_(opcode), lineno_(lineno), charpos_(charpos) 103 { } 104 105 // Return whether the token is invalid. 106 bool 107 is_invalid() const 108 { return this->classification_ == TOKEN_INVALID; } 109 110 // Return whether this is an EOF token. 111 bool 112 is_eof() const 113 { return this->classification_ == TOKEN_EOF; } 114 115 // Return the token classification. 116 Classification 117 classification() const 118 { return this->classification_; } 119 120 // Return the line number at which the token starts. 121 int 122 lineno() const 123 { return this->lineno_; } 124 125 // Return the character position at this the token starts. 126 int 127 charpos() const 128 { return this->charpos_; } 129 130 // Get the value of a token. 131 132 const char* 133 string_value(size_t* length) const 134 { 135 gold_assert(this->classification_ == TOKEN_STRING 136 || this->classification_ == TOKEN_QUOTED_STRING); 137 *length = this->value_length_; 138 return this->value_; 139 } 140 141 int 142 operator_value() const 143 { 144 gold_assert(this->classification_ == TOKEN_OPERATOR); 145 return this->opcode_; 146 } 147 148 uint64_t 149 integer_value() const; 150 151 private: 152 // The token classification. 153 Classification classification_; 154 // The token value, for TOKEN_STRING or TOKEN_QUOTED_STRING or 155 // TOKEN_INTEGER. 156 const char* value_; 157 // The length of the token value. 158 size_t value_length_; 159 // The token value, for TOKEN_OPERATOR. 160 int opcode_; 161 // The line number where this token started (one based). 162 int lineno_; 163 // The character position within the line where this token started 164 // (one based). 165 int charpos_; 166}; 167 168// Return the value of a TOKEN_INTEGER. 169 170uint64_t 171Token::integer_value() const 172{ 173 gold_assert(this->classification_ == TOKEN_INTEGER); 174 175 size_t len = this->value_length_; 176 177 uint64_t multiplier = 1; 178 char last = this->value_[len - 1]; 179 if (last == 'm' || last == 'M') 180 { 181 multiplier = 1024 * 1024; 182 --len; 183 } 184 else if (last == 'k' || last == 'K') 185 { 186 multiplier = 1024; 187 --len; 188 } 189 190 char *end; 191 uint64_t ret = strtoull(this->value_, &end, 0); 192 gold_assert(static_cast<size_t>(end - this->value_) == len); 193 194 return ret * multiplier; 195} 196 197// This class handles lexing a file into a sequence of tokens. 198 199class Lex 200{ 201 public: 202 // We unfortunately have to support different lexing modes, because 203 // when reading different parts of a linker script we need to parse 204 // things differently. 205 enum Mode 206 { 207 // Reading an ordinary linker script. 208 LINKER_SCRIPT, 209 // Reading an expression in a linker script. 210 EXPRESSION, 211 // Reading a version script. 212 VERSION_SCRIPT, 213 // Reading a --dynamic-list file. 214 DYNAMIC_LIST 215 }; 216 217 Lex(const char* input_string, size_t input_length, int parsing_token) 218 : input_string_(input_string), input_length_(input_length), 219 current_(input_string), mode_(LINKER_SCRIPT), 220 first_token_(parsing_token), token_(), 221 lineno_(1), linestart_(input_string) 222 { } 223 224 // Read a file into a string. 225 static void 226 read_file(Input_file*, std::string*); 227 228 // Return the next token. 229 const Token* 230 next_token(); 231 232 // Return the current lexing mode. 233 Lex::Mode 234 mode() const 235 { return this->mode_; } 236 237 // Set the lexing mode. 238 void 239 set_mode(Mode mode) 240 { this->mode_ = mode; } 241 242 private: 243 Lex(const Lex&); 244 Lex& operator=(const Lex&); 245 246 // Make a general token with no value at the current location. 247 Token 248 make_token(Token::Classification c, const char* start) const 249 { return Token(c, this->lineno_, start - this->linestart_ + 1); } 250 251 // Make a general token with a value at the current location. 252 Token 253 make_token(Token::Classification c, const char* v, size_t len, 254 const char* start) 255 const 256 { return Token(c, v, len, this->lineno_, start - this->linestart_ + 1); } 257 258 // Make an operator token at the current location. 259 Token 260 make_token(int opcode, const char* start) const 261 { return Token(opcode, this->lineno_, start - this->linestart_ + 1); } 262 263 // Make an invalid token at the current location. 264 Token 265 make_invalid_token(const char* start) 266 { return this->make_token(Token::TOKEN_INVALID, start); } 267 268 // Make an EOF token at the current location. 269 Token 270 make_eof_token(const char* start) 271 { return this->make_token(Token::TOKEN_EOF, start); } 272 273 // Return whether C can be the first character in a name. C2 is the 274 // next character, since we sometimes need that. 275 inline bool 276 can_start_name(char c, char c2); 277 278 // If C can appear in a name which has already started, return a 279 // pointer to a character later in the token or just past 280 // it. Otherwise, return NULL. 281 inline const char* 282 can_continue_name(const char* c); 283 284 // Return whether C, C2, C3 can start a hex number. 285 inline bool 286 can_start_hex(char c, char c2, char c3); 287 288 // If C can appear in a hex number which has already started, return 289 // a pointer to a character later in the token or just past 290 // it. Otherwise, return NULL. 291 inline const char* 292 can_continue_hex(const char* c); 293 294 // Return whether C can start a non-hex number. 295 static inline bool 296 can_start_number(char c); 297 298 // If C can appear in a decimal number which has already started, 299 // return a pointer to a character later in the token or just past 300 // it. Otherwise, return NULL. 301 inline const char* 302 can_continue_number(const char* c) 303 { return Lex::can_start_number(*c) ? c + 1 : NULL; } 304 305 // If C1 C2 C3 form a valid three character operator, return the 306 // opcode. Otherwise return 0. 307 static inline int 308 three_char_operator(char c1, char c2, char c3); 309 310 // If C1 C2 form a valid two character operator, return the opcode. 311 // Otherwise return 0. 312 static inline int 313 two_char_operator(char c1, char c2); 314 315 // If C1 is a valid one character operator, return the opcode. 316 // Otherwise return 0. 317 static inline int 318 one_char_operator(char c1); 319 320 // Read the next token. 321 Token 322 get_token(const char**); 323 324 // Skip a C style /* */ comment. Return false if the comment did 325 // not end. 326 bool 327 skip_c_comment(const char**); 328 329 // Skip a line # comment. Return false if there was no newline. 330 bool 331 skip_line_comment(const char**); 332 333 // Build a token CLASSIFICATION from all characters that match 334 // CAN_CONTINUE_FN. The token starts at START. Start matching from 335 // MATCH. Set *PP to the character following the token. 336 inline Token 337 gather_token(Token::Classification, 338 const char* (Lex::*can_continue_fn)(const char*), 339 const char* start, const char* match, const char** pp); 340 341 // Build a token from a quoted string. 342 Token 343 gather_quoted_string(const char** pp); 344 345 // The string we are tokenizing. 346 const char* input_string_; 347 // The length of the string. 348 size_t input_length_; 349 // The current offset into the string. 350 const char* current_; 351 // The current lexing mode. 352 Mode mode_; 353 // The code to use for the first token. This is set to 0 after it 354 // is used. 355 int first_token_; 356 // The current token. 357 Token token_; 358 // The current line number. 359 int lineno_; 360 // The start of the current line in the string. 361 const char* linestart_; 362}; 363 364// Read the whole file into memory. We don't expect linker scripts to 365// be large, so we just use a std::string as a buffer. We ignore the 366// data we've already read, so that we read aligned buffers. 367 368void 369Lex::read_file(Input_file* input_file, std::string* contents) 370{ 371 off_t filesize = input_file->file().filesize(); 372 contents->clear(); 373 contents->reserve(filesize); 374 375 off_t off = 0; 376 unsigned char buf[BUFSIZ]; 377 while (off < filesize) 378 { 379 off_t get = BUFSIZ; 380 if (get > filesize - off) 381 get = filesize - off; 382 input_file->file().read(off, get, buf); 383 contents->append(reinterpret_cast<char*>(&buf[0]), get); 384 off += get; 385 } 386} 387 388// Return whether C can be the start of a name, if the next character 389// is C2. A name can being with a letter, underscore, period, or 390// dollar sign. Because a name can be a file name, we also permit 391// forward slash, backslash, and tilde. Tilde is the tricky case 392// here; GNU ld also uses it as a bitwise not operator. It is only 393// recognized as the operator if it is not immediately followed by 394// some character which can appear in a symbol. That is, when we 395// don't know that we are looking at an expression, "~0" is a file 396// name, and "~ 0" is an expression using bitwise not. We are 397// compatible. 398 399inline bool 400Lex::can_start_name(char c, char c2) 401{ 402 switch (c) 403 { 404 case 'A': case 'B': case 'C': case 'D': case 'E': case 'F': 405 case 'G': case 'H': case 'I': case 'J': case 'K': case 'L': 406 case 'M': case 'N': case 'O': case 'Q': case 'P': case 'R': 407 case 'S': case 'T': case 'U': case 'V': case 'W': case 'X': 408 case 'Y': case 'Z': 409 case 'a': case 'b': case 'c': case 'd': case 'e': case 'f': 410 case 'g': case 'h': case 'i': case 'j': case 'k': case 'l': 411 case 'm': case 'n': case 'o': case 'q': case 'p': case 'r': 412 case 's': case 't': case 'u': case 'v': case 'w': case 'x': 413 case 'y': case 'z': 414 case '_': case '.': case '$': 415 return true; 416 417 case '/': case '\\': 418 return this->mode_ == LINKER_SCRIPT; 419 420 case '~': 421 return this->mode_ == LINKER_SCRIPT && can_continue_name(&c2); 422 423 case '*': case '[': 424 return (this->mode_ == VERSION_SCRIPT 425 || this->mode_ == DYNAMIC_LIST 426 || (this->mode_ == LINKER_SCRIPT 427 && can_continue_name(&c2))); 428 429 default: 430 return false; 431 } 432} 433 434// Return whether C can continue a name which has already started. 435// Subsequent characters in a name are the same as the leading 436// characters, plus digits and "=+-:[],?*". So in general the linker 437// script language requires spaces around operators, unless we know 438// that we are parsing an expression. 439 440inline const char* 441Lex::can_continue_name(const char* c) 442{ 443 switch (*c) 444 { 445 case 'A': case 'B': case 'C': case 'D': case 'E': case 'F': 446 case 'G': case 'H': case 'I': case 'J': case 'K': case 'L': 447 case 'M': case 'N': case 'O': case 'Q': case 'P': case 'R': 448 case 'S': case 'T': case 'U': case 'V': case 'W': case 'X': 449 case 'Y': case 'Z': 450 case 'a': case 'b': case 'c': case 'd': case 'e': case 'f': 451 case 'g': case 'h': case 'i': case 'j': case 'k': case 'l': 452 case 'm': case 'n': case 'o': case 'q': case 'p': case 'r': 453 case 's': case 't': case 'u': case 'v': case 'w': case 'x': 454 case 'y': case 'z': 455 case '_': case '.': case '$': 456 case '0': case '1': case '2': case '3': case '4': 457 case '5': case '6': case '7': case '8': case '9': 458 return c + 1; 459 460 // TODO(csilvers): why not allow ~ in names for version-scripts? 461 case '/': case '\\': case '~': 462 case '=': case '+': 463 case ',': 464 if (this->mode_ == LINKER_SCRIPT) 465 return c + 1; 466 return NULL; 467 468 case '[': case ']': case '*': case '?': case '-': 469 if (this->mode_ == LINKER_SCRIPT || this->mode_ == VERSION_SCRIPT 470 || this->mode_ == DYNAMIC_LIST) 471 return c + 1; 472 return NULL; 473 474 // TODO(csilvers): why allow this? ^ is meaningless in version scripts. 475 case '^': 476 if (this->mode_ == VERSION_SCRIPT || this->mode_ == DYNAMIC_LIST) 477 return c + 1; 478 return NULL; 479 480 case ':': 481 if (this->mode_ == LINKER_SCRIPT) 482 return c + 1; 483 else if ((this->mode_ == VERSION_SCRIPT || this->mode_ == DYNAMIC_LIST) 484 && (c[1] == ':')) 485 { 486 // A name can have '::' in it, as that's a c++ namespace 487 // separator. But a single colon is not part of a name. 488 return c + 2; 489 } 490 return NULL; 491 492 default: 493 return NULL; 494 } 495} 496 497// For a number we accept 0x followed by hex digits, or any sequence 498// of digits. The old linker accepts leading '$' for hex, and 499// trailing HXBOD. Those are for MRI compatibility and we don't 500// accept them. 501 502// Return whether C1 C2 C3 can start a hex number. 503 504inline bool 505Lex::can_start_hex(char c1, char c2, char c3) 506{ 507 if (c1 == '0' && (c2 == 'x' || c2 == 'X')) 508 return this->can_continue_hex(&c3); 509 return false; 510} 511 512// Return whether C can appear in a hex number. 513 514inline const char* 515Lex::can_continue_hex(const char* c) 516{ 517 switch (*c) 518 { 519 case '0': case '1': case '2': case '3': case '4': 520 case '5': case '6': case '7': case '8': case '9': 521 case 'A': case 'B': case 'C': case 'D': case 'E': case 'F': 522 case 'a': case 'b': case 'c': case 'd': case 'e': case 'f': 523 return c + 1; 524 525 default: 526 return NULL; 527 } 528} 529 530// Return whether C can start a non-hex number. 531 532inline bool 533Lex::can_start_number(char c) 534{ 535 switch (c) 536 { 537 case '0': case '1': case '2': case '3': case '4': 538 case '5': case '6': case '7': case '8': case '9': 539 return true; 540 541 default: 542 return false; 543 } 544} 545 546// If C1 C2 C3 form a valid three character operator, return the 547// opcode (defined in the yyscript.h file generated from yyscript.y). 548// Otherwise return 0. 549 550inline int 551Lex::three_char_operator(char c1, char c2, char c3) 552{ 553 switch (c1) 554 { 555 case '<': 556 if (c2 == '<' && c3 == '=') 557 return LSHIFTEQ; 558 break; 559 case '>': 560 if (c2 == '>' && c3 == '=') 561 return RSHIFTEQ; 562 break; 563 default: 564 break; 565 } 566 return 0; 567} 568 569// If C1 C2 form a valid two character operator, return the opcode 570// (defined in the yyscript.h file generated from yyscript.y). 571// Otherwise return 0. 572 573inline int 574Lex::two_char_operator(char c1, char c2) 575{ 576 switch (c1) 577 { 578 case '=': 579 if (c2 == '=') 580 return EQ; 581 break; 582 case '!': 583 if (c2 == '=') 584 return NE; 585 break; 586 case '+': 587 if (c2 == '=') 588 return PLUSEQ; 589 break; 590 case '-': 591 if (c2 == '=') 592 return MINUSEQ; 593 break; 594 case '*': 595 if (c2 == '=') 596 return MULTEQ; 597 break; 598 case '/': 599 if (c2 == '=') 600 return DIVEQ; 601 break; 602 case '|': 603 if (c2 == '=') 604 return OREQ; 605 if (c2 == '|') 606 return OROR; 607 break; 608 case '&': 609 if (c2 == '=') 610 return ANDEQ; 611 if (c2 == '&') 612 return ANDAND; 613 break; 614 case '>': 615 if (c2 == '=') 616 return GE; 617 if (c2 == '>') 618 return RSHIFT; 619 break; 620 case '<': 621 if (c2 == '=') 622 return LE; 623 if (c2 == '<') 624 return LSHIFT; 625 break; 626 default: 627 break; 628 } 629 return 0; 630} 631 632// If C1 is a valid operator, return the opcode. Otherwise return 0. 633 634inline int 635Lex::one_char_operator(char c1) 636{ 637 switch (c1) 638 { 639 case '+': 640 case '-': 641 case '*': 642 case '/': 643 case '%': 644 case '!': 645 case '&': 646 case '|': 647 case '^': 648 case '~': 649 case '<': 650 case '>': 651 case '=': 652 case '?': 653 case ',': 654 case '(': 655 case ')': 656 case '{': 657 case '}': 658 case '[': 659 case ']': 660 case ':': 661 case ';': 662 return c1; 663 default: 664 return 0; 665 } 666} 667 668// Skip a C style comment. *PP points to just after the "/*". Return 669// false if the comment did not end. 670 671bool 672Lex::skip_c_comment(const char** pp) 673{ 674 const char* p = *pp; 675 while (p[0] != '*' || p[1] != '/') 676 { 677 if (*p == '\0') 678 { 679 *pp = p; 680 return false; 681 } 682 683 if (*p == '\n') 684 { 685 ++this->lineno_; 686 this->linestart_ = p + 1; 687 } 688 ++p; 689 } 690 691 *pp = p + 2; 692 return true; 693} 694 695// Skip a line # comment. Return false if there was no newline. 696 697bool 698Lex::skip_line_comment(const char** pp) 699{ 700 const char* p = *pp; 701 size_t skip = strcspn(p, "\n"); 702 if (p[skip] == '\0') 703 { 704 *pp = p + skip; 705 return false; 706 } 707 708 p += skip + 1; 709 ++this->lineno_; 710 this->linestart_ = p; 711 *pp = p; 712 713 return true; 714} 715 716// Build a token CLASSIFICATION from all characters that match 717// CAN_CONTINUE_FN. Update *PP. 718 719inline Token 720Lex::gather_token(Token::Classification classification, 721 const char* (Lex::*can_continue_fn)(const char*), 722 const char* start, 723 const char* match, 724 const char** pp) 725{ 726 const char* new_match = NULL; 727 while ((new_match = (this->*can_continue_fn)(match)) != NULL) 728 match = new_match; 729 730 // A special case: integers may be followed by a single M or K, 731 // case-insensitive. 732 if (classification == Token::TOKEN_INTEGER 733 && (*match == 'm' || *match == 'M' || *match == 'k' || *match == 'K')) 734 ++match; 735 736 *pp = match; 737 return this->make_token(classification, start, match - start, start); 738} 739 740// Build a token from a quoted string. 741 742Token 743Lex::gather_quoted_string(const char** pp) 744{ 745 const char* start = *pp; 746 const char* p = start; 747 ++p; 748 size_t skip = strcspn(p, "\"\n"); 749 if (p[skip] != '"') 750 return this->make_invalid_token(start); 751 *pp = p + skip + 1; 752 return this->make_token(Token::TOKEN_QUOTED_STRING, p, skip, start); 753} 754 755// Return the next token at *PP. Update *PP. General guideline: we 756// require linker scripts to be simple ASCII. No unicode linker 757// scripts. In particular we can assume that any '\0' is the end of 758// the input. 759 760Token 761Lex::get_token(const char** pp) 762{ 763 const char* p = *pp; 764 765 while (true) 766 { 767 // Skip whitespace quickly. 768 while (*p == ' ' || *p == '\t' || *p == '\r') 769 ++p; 770 771 if (*p == '\n') 772 { 773 ++p; 774 ++this->lineno_; 775 this->linestart_ = p; 776 continue; 777 } 778 779 char c0 = *p; 780 781 if (c0 == '\0') 782 { 783 *pp = p; 784 return this->make_eof_token(p); 785 } 786 787 char c1 = p[1]; 788 789 // Skip C style comments. 790 if (c0 == '/' && c1 == '*') 791 { 792 int lineno = this->lineno_; 793 int charpos = p - this->linestart_ + 1; 794 795 *pp = p + 2; 796 if (!this->skip_c_comment(pp)) 797 return Token(Token::TOKEN_INVALID, lineno, charpos); 798 p = *pp; 799 800 continue; 801 } 802 803 // Skip line comments. 804 if (c0 == '#') 805 { 806 *pp = p + 1; 807 if (!this->skip_line_comment(pp)) 808 return this->make_eof_token(p); 809 p = *pp; 810 continue; 811 } 812 813 // Check for a name. 814 if (this->can_start_name(c0, c1)) 815 return this->gather_token(Token::TOKEN_STRING, 816 &Lex::can_continue_name, 817 p, p + 1, pp); 818 819 // We accept any arbitrary name in double quotes, as long as it 820 // does not cross a line boundary. 821 if (*p == '"') 822 { 823 *pp = p; 824 return this->gather_quoted_string(pp); 825 } 826 827 // Be careful not to lookahead past the end of the buffer. 828 char c2 = (c1 == '\0' ? '\0' : p[2]); 829 830 // Check for a number. 831 832 if (this->can_start_hex(c0, c1, c2)) 833 return this->gather_token(Token::TOKEN_INTEGER, 834 &Lex::can_continue_hex, 835 p, p + 3, pp); 836 837 if (Lex::can_start_number(c0)) 838 return this->gather_token(Token::TOKEN_INTEGER, 839 &Lex::can_continue_number, 840 p, p + 1, pp); 841 842 // Check for operators. 843 844 int opcode = Lex::three_char_operator(c0, c1, c2); 845 if (opcode != 0) 846 { 847 *pp = p + 3; 848 return this->make_token(opcode, p); 849 } 850 851 opcode = Lex::two_char_operator(c0, c1); 852 if (opcode != 0) 853 { 854 *pp = p + 2; 855 return this->make_token(opcode, p); 856 } 857 858 opcode = Lex::one_char_operator(c0); 859 if (opcode != 0) 860 { 861 *pp = p + 1; 862 return this->make_token(opcode, p); 863 } 864 865 return this->make_token(Token::TOKEN_INVALID, p); 866 } 867} 868 869// Return the next token. 870 871const Token* 872Lex::next_token() 873{ 874 // The first token is special. 875 if (this->first_token_ != 0) 876 { 877 this->token_ = Token(this->first_token_, 0, 0); 878 this->first_token_ = 0; 879 return &this->token_; 880 } 881 882 this->token_ = this->get_token(&this->current_); 883 884 // Don't let an early null byte fool us into thinking that we've 885 // reached the end of the file. 886 if (this->token_.is_eof() 887 && (static_cast<size_t>(this->current_ - this->input_string_) 888 < this->input_length_)) 889 this->token_ = this->make_invalid_token(this->current_); 890 891 return &this->token_; 892} 893 894// class Symbol_assignment. 895 896// Add the symbol to the symbol table. This makes sure the symbol is 897// there and defined. The actual value is stored later. We can't 898// determine the actual value at this point, because we can't 899// necessarily evaluate the expression until all ordinary symbols have 900// been finalized. 901 902// The GNU linker lets symbol assignments in the linker script 903// silently override defined symbols in object files. We are 904// compatible. FIXME: Should we issue a warning? 905 906void 907Symbol_assignment::add_to_table(Symbol_table* symtab) 908{ 909 elfcpp::STV vis = this->hidden_ ? elfcpp::STV_HIDDEN : elfcpp::STV_DEFAULT; 910 this->sym_ = symtab->define_as_constant(this->name_.c_str(), 911 NULL, // version 912 (this->is_defsym_ 913 ? Symbol_table::DEFSYM 914 : Symbol_table::SCRIPT), 915 0, // value 916 0, // size 917 elfcpp::STT_NOTYPE, 918 elfcpp::STB_GLOBAL, 919 vis, 920 0, // nonvis 921 this->provide_, 922 true); // force_override 923} 924 925// Finalize a symbol value. 926 927void 928Symbol_assignment::finalize(Symbol_table* symtab, const Layout* layout) 929{ 930 this->finalize_maybe_dot(symtab, layout, false, 0, NULL); 931} 932 933// Finalize a symbol value which can refer to the dot symbol. 934 935void 936Symbol_assignment::finalize_with_dot(Symbol_table* symtab, 937 const Layout* layout, 938 uint64_t dot_value, 939 Output_section* dot_section) 940{ 941 this->finalize_maybe_dot(symtab, layout, true, dot_value, dot_section); 942} 943 944// Finalize a symbol value, internal version. 945 946void 947Symbol_assignment::finalize_maybe_dot(Symbol_table* symtab, 948 const Layout* layout, 949 bool is_dot_available, 950 uint64_t dot_value, 951 Output_section* dot_section) 952{ 953 // If we were only supposed to provide this symbol, the sym_ field 954 // will be NULL if the symbol was not referenced. 955 if (this->sym_ == NULL) 956 { 957 gold_assert(this->provide_); 958 return; 959 } 960 961 if (parameters->target().get_size() == 32) 962 { 963#if defined(HAVE_TARGET_32_LITTLE) || defined(HAVE_TARGET_32_BIG) 964 this->sized_finalize<32>(symtab, layout, is_dot_available, dot_value, 965 dot_section); 966#else 967 gold_unreachable(); 968#endif 969 } 970 else if (parameters->target().get_size() == 64) 971 { 972#if defined(HAVE_TARGET_64_LITTLE) || defined(HAVE_TARGET_64_BIG) 973 this->sized_finalize<64>(symtab, layout, is_dot_available, dot_value, 974 dot_section); 975#else 976 gold_unreachable(); 977#endif 978 } 979 else 980 gold_unreachable(); 981} 982 983template<int size> 984void 985Symbol_assignment::sized_finalize(Symbol_table* symtab, const Layout* layout, 986 bool is_dot_available, uint64_t dot_value, 987 Output_section* dot_section) 988{ 989 Output_section* section; 990 elfcpp::STT type = elfcpp::STT_NOTYPE; 991 elfcpp::STV vis = elfcpp::STV_DEFAULT; 992 unsigned char nonvis = 0; 993 uint64_t final_val = this->val_->eval_maybe_dot(symtab, layout, true, 994 is_dot_available, 995 dot_value, dot_section, 996 §ion, NULL, &type, 997 &vis, &nonvis, false, NULL); 998 Sized_symbol<size>* ssym = symtab->get_sized_symbol<size>(this->sym_); 999 ssym->set_value(final_val); 1000 ssym->set_type(type); 1001 ssym->set_visibility(vis); 1002 ssym->set_nonvis(nonvis); 1003 if (section != NULL) 1004 ssym->set_output_section(section); 1005} 1006 1007// Set the symbol value if the expression yields an absolute value or 1008// a value relative to DOT_SECTION. 1009 1010void 1011Symbol_assignment::set_if_absolute(Symbol_table* symtab, const Layout* layout, 1012 bool is_dot_available, uint64_t dot_value, 1013 Output_section* dot_section) 1014{ 1015 if (this->sym_ == NULL) 1016 return; 1017 1018 Output_section* val_section; 1019 bool is_valid; 1020 uint64_t val = this->val_->eval_maybe_dot(symtab, layout, false, 1021 is_dot_available, dot_value, 1022 dot_section, &val_section, NULL, 1023 NULL, NULL, NULL, false, &is_valid); 1024 if (!is_valid || (val_section != NULL && val_section != dot_section)) 1025 return; 1026 1027 if (parameters->target().get_size() == 32) 1028 { 1029#if defined(HAVE_TARGET_32_LITTLE) || defined(HAVE_TARGET_32_BIG) 1030 Sized_symbol<32>* ssym = symtab->get_sized_symbol<32>(this->sym_); 1031 ssym->set_value(val); 1032#else 1033 gold_unreachable(); 1034#endif 1035 } 1036 else if (parameters->target().get_size() == 64) 1037 { 1038#if defined(HAVE_TARGET_64_LITTLE) || defined(HAVE_TARGET_64_BIG) 1039 Sized_symbol<64>* ssym = symtab->get_sized_symbol<64>(this->sym_); 1040 ssym->set_value(val); 1041#else 1042 gold_unreachable(); 1043#endif 1044 } 1045 else 1046 gold_unreachable(); 1047 if (val_section != NULL) 1048 this->sym_->set_output_section(val_section); 1049} 1050 1051// Print for debugging. 1052 1053void 1054Symbol_assignment::print(FILE* f) const 1055{ 1056 if (this->provide_ && this->hidden_) 1057 fprintf(f, "PROVIDE_HIDDEN("); 1058 else if (this->provide_) 1059 fprintf(f, "PROVIDE("); 1060 else if (this->hidden_) 1061 gold_unreachable(); 1062 1063 fprintf(f, "%s = ", this->name_.c_str()); 1064 this->val_->print(f); 1065 1066 if (this->provide_ || this->hidden_) 1067 fprintf(f, ")"); 1068 1069 fprintf(f, "\n"); 1070} 1071 1072// Class Script_assertion. 1073 1074// Check the assertion. 1075 1076void 1077Script_assertion::check(const Symbol_table* symtab, const Layout* layout) 1078{ 1079 if (!this->check_->eval(symtab, layout, true)) 1080 gold_error("%s", this->message_.c_str()); 1081} 1082 1083// Print for debugging. 1084 1085void 1086Script_assertion::print(FILE* f) const 1087{ 1088 fprintf(f, "ASSERT("); 1089 this->check_->print(f); 1090 fprintf(f, ", \"%s\")\n", this->message_.c_str()); 1091} 1092 1093// Class Script_options. 1094 1095Script_options::Script_options() 1096 : entry_(), symbol_assignments_(), symbol_definitions_(), 1097 symbol_references_(), version_script_info_(), script_sections_() 1098{ 1099} 1100 1101// Returns true if NAME is on the list of symbol assignments waiting 1102// to be processed. 1103 1104bool 1105Script_options::is_pending_assignment(const char* name) 1106{ 1107 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin(); 1108 p != this->symbol_assignments_.end(); 1109 ++p) 1110 if ((*p)->name() == name) 1111 return true; 1112 return false; 1113} 1114 1115// Add a symbol to be defined. 1116 1117void 1118Script_options::add_symbol_assignment(const char* name, size_t length, 1119 bool is_defsym, Expression* value, 1120 bool provide, bool hidden) 1121{ 1122 if (length != 1 || name[0] != '.') 1123 { 1124 if (this->script_sections_.in_sections_clause()) 1125 { 1126 gold_assert(!is_defsym); 1127 this->script_sections_.add_symbol_assignment(name, length, value, 1128 provide, hidden); 1129 } 1130 else 1131 { 1132 Symbol_assignment* p = new Symbol_assignment(name, length, is_defsym, 1133 value, provide, hidden); 1134 this->symbol_assignments_.push_back(p); 1135 } 1136 1137 if (!provide) 1138 { 1139 std::string n(name, length); 1140 this->symbol_definitions_.insert(n); 1141 this->symbol_references_.erase(n); 1142 } 1143 } 1144 else 1145 { 1146 if (provide || hidden) 1147 gold_error(_("invalid use of PROVIDE for dot symbol")); 1148 1149 // The GNU linker permits assignments to dot outside of SECTIONS 1150 // clauses and treats them as occurring inside, so we don't 1151 // check in_sections_clause here. 1152 this->script_sections_.add_dot_assignment(value); 1153 } 1154} 1155 1156// Add a reference to a symbol. 1157 1158void 1159Script_options::add_symbol_reference(const char* name, size_t length) 1160{ 1161 if (length != 1 || name[0] != '.') 1162 { 1163 std::string n(name, length); 1164 if (this->symbol_definitions_.find(n) == this->symbol_definitions_.end()) 1165 this->symbol_references_.insert(n); 1166 } 1167} 1168 1169// Add an assertion. 1170 1171void 1172Script_options::add_assertion(Expression* check, const char* message, 1173 size_t messagelen) 1174{ 1175 if (this->script_sections_.in_sections_clause()) 1176 this->script_sections_.add_assertion(check, message, messagelen); 1177 else 1178 { 1179 Script_assertion* p = new Script_assertion(check, message, messagelen); 1180 this->assertions_.push_back(p); 1181 } 1182} 1183 1184// Create sections required by any linker scripts. 1185 1186void 1187Script_options::create_script_sections(Layout* layout) 1188{ 1189 if (this->saw_sections_clause()) 1190 this->script_sections_.create_sections(layout); 1191} 1192 1193// Add any symbols we are defining to the symbol table. 1194 1195void 1196Script_options::add_symbols_to_table(Symbol_table* symtab) 1197{ 1198 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin(); 1199 p != this->symbol_assignments_.end(); 1200 ++p) 1201 (*p)->add_to_table(symtab); 1202 this->script_sections_.add_symbols_to_table(symtab); 1203} 1204 1205// Finalize symbol values. Also check assertions. 1206 1207void 1208Script_options::finalize_symbols(Symbol_table* symtab, const Layout* layout) 1209{ 1210 // We finalize the symbols defined in SECTIONS first, because they 1211 // are the ones which may have changed. This way if symbol outside 1212 // SECTIONS are defined in terms of symbols inside SECTIONS, they 1213 // will get the right value. 1214 this->script_sections_.finalize_symbols(symtab, layout); 1215 1216 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin(); 1217 p != this->symbol_assignments_.end(); 1218 ++p) 1219 (*p)->finalize(symtab, layout); 1220 1221 for (Assertions::iterator p = this->assertions_.begin(); 1222 p != this->assertions_.end(); 1223 ++p) 1224 (*p)->check(symtab, layout); 1225} 1226 1227// Set section addresses. We set all the symbols which have absolute 1228// values. Then we let the SECTIONS clause do its thing. This 1229// returns the segment which holds the file header and segment 1230// headers, if any. 1231 1232Output_segment* 1233Script_options::set_section_addresses(Symbol_table* symtab, Layout* layout) 1234{ 1235 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin(); 1236 p != this->symbol_assignments_.end(); 1237 ++p) 1238 (*p)->set_if_absolute(symtab, layout, false, 0, NULL); 1239 1240 return this->script_sections_.set_section_addresses(symtab, layout); 1241} 1242 1243// This class holds data passed through the parser to the lexer and to 1244// the parser support functions. This avoids global variables. We 1245// can't use global variables because we need not be called by a 1246// singleton thread. 1247 1248class Parser_closure 1249{ 1250 public: 1251 Parser_closure(const char* filename, 1252 const Position_dependent_options& posdep_options, 1253 bool parsing_defsym, bool in_group, bool is_in_sysroot, 1254 Command_line* command_line, 1255 Script_options* script_options, 1256 Lex* lex, 1257 bool skip_on_incompatible_target, 1258 Script_info* script_info) 1259 : filename_(filename), posdep_options_(posdep_options), 1260 parsing_defsym_(parsing_defsym), in_group_(in_group), 1261 is_in_sysroot_(is_in_sysroot), 1262 skip_on_incompatible_target_(skip_on_incompatible_target), 1263 found_incompatible_target_(false), 1264 command_line_(command_line), script_options_(script_options), 1265 version_script_info_(script_options->version_script_info()), 1266 lex_(lex), lineno_(0), charpos_(0), lex_mode_stack_(), inputs_(NULL), 1267 script_info_(script_info) 1268 { 1269 // We start out processing C symbols in the default lex mode. 1270 this->language_stack_.push_back(Version_script_info::LANGUAGE_C); 1271 this->lex_mode_stack_.push_back(lex->mode()); 1272 } 1273 1274 // Return the file name. 1275 const char* 1276 filename() const 1277 { return this->filename_; } 1278 1279 // Return the position dependent options. The caller may modify 1280 // this. 1281 Position_dependent_options& 1282 position_dependent_options() 1283 { return this->posdep_options_; } 1284 1285 // Whether we are parsing a --defsym. 1286 bool 1287 parsing_defsym() const 1288 { return this->parsing_defsym_; } 1289 1290 // Return whether this script is being run in a group. 1291 bool 1292 in_group() const 1293 { return this->in_group_; } 1294 1295 // Return whether this script was found using a directory in the 1296 // sysroot. 1297 bool 1298 is_in_sysroot() const 1299 { return this->is_in_sysroot_; } 1300 1301 // Whether to skip to the next file with the same name if we find an 1302 // incompatible target in an OUTPUT_FORMAT statement. 1303 bool 1304 skip_on_incompatible_target() const 1305 { return this->skip_on_incompatible_target_; } 1306 1307 // Stop skipping to the next file on an incompatible target. This 1308 // is called when we make some unrevocable change to the data 1309 // structures. 1310 void 1311 clear_skip_on_incompatible_target() 1312 { this->skip_on_incompatible_target_ = false; } 1313 1314 // Whether we found an incompatible target in an OUTPUT_FORMAT 1315 // statement. 1316 bool 1317 found_incompatible_target() const 1318 { return this->found_incompatible_target_; } 1319 1320 // Note that we found an incompatible target. 1321 void 1322 set_found_incompatible_target() 1323 { this->found_incompatible_target_ = true; } 1324 1325 // Returns the Command_line structure passed in at constructor time. 1326 // This value may be NULL. The caller may modify this, which modifies 1327 // the passed-in Command_line object (not a copy). 1328 Command_line* 1329 command_line() 1330 { return this->command_line_; } 1331 1332 // Return the options which may be set by a script. 1333 Script_options* 1334 script_options() 1335 { return this->script_options_; } 1336 1337 // Return the object in which version script information should be stored. 1338 Version_script_info* 1339 version_script() 1340 { return this->version_script_info_; } 1341 1342 // Return the next token, and advance. 1343 const Token* 1344 next_token() 1345 { 1346 const Token* token = this->lex_->next_token(); 1347 this->lineno_ = token->lineno(); 1348 this->charpos_ = token->charpos(); 1349 return token; 1350 } 1351 1352 // Set a new lexer mode, pushing the current one. 1353 void 1354 push_lex_mode(Lex::Mode mode) 1355 { 1356 this->lex_mode_stack_.push_back(this->lex_->mode()); 1357 this->lex_->set_mode(mode); 1358 } 1359 1360 // Pop the lexer mode. 1361 void 1362 pop_lex_mode() 1363 { 1364 gold_assert(!this->lex_mode_stack_.empty()); 1365 this->lex_->set_mode(this->lex_mode_stack_.back()); 1366 this->lex_mode_stack_.pop_back(); 1367 } 1368 1369 // Return the current lexer mode. 1370 Lex::Mode 1371 lex_mode() const 1372 { return this->lex_mode_stack_.back(); } 1373 1374 // Return the line number of the last token. 1375 int 1376 lineno() const 1377 { return this->lineno_; } 1378 1379 // Return the character position in the line of the last token. 1380 int 1381 charpos() const 1382 { return this->charpos_; } 1383 1384 // Return the list of input files, creating it if necessary. This 1385 // is a space leak--we never free the INPUTS_ pointer. 1386 Input_arguments* 1387 inputs() 1388 { 1389 if (this->inputs_ == NULL) 1390 this->inputs_ = new Input_arguments(); 1391 return this->inputs_; 1392 } 1393 1394 // Return whether we saw any input files. 1395 bool 1396 saw_inputs() const 1397 { return this->inputs_ != NULL && !this->inputs_->empty(); } 1398 1399 // Return the current language being processed in a version script 1400 // (eg, "C++"). The empty string represents unmangled C names. 1401 Version_script_info::Language 1402 get_current_language() const 1403 { return this->language_stack_.back(); } 1404 1405 // Push a language onto the stack when entering an extern block. 1406 void 1407 push_language(Version_script_info::Language lang) 1408 { this->language_stack_.push_back(lang); } 1409 1410 // Pop a language off of the stack when exiting an extern block. 1411 void 1412 pop_language() 1413 { 1414 gold_assert(!this->language_stack_.empty()); 1415 this->language_stack_.pop_back(); 1416 } 1417 1418 // Return a pointer to the incremental info. 1419 Script_info* 1420 script_info() 1421 { return this->script_info_; } 1422 1423 private: 1424 // The name of the file we are reading. 1425 const char* filename_; 1426 // The position dependent options. 1427 Position_dependent_options posdep_options_; 1428 // True if we are parsing a --defsym. 1429 bool parsing_defsym_; 1430 // Whether we are currently in a --start-group/--end-group. 1431 bool in_group_; 1432 // Whether the script was found in a sysrooted directory. 1433 bool is_in_sysroot_; 1434 // If this is true, then if we find an OUTPUT_FORMAT with an 1435 // incompatible target, then we tell the parser to abort so that we 1436 // can search for the next file with the same name. 1437 bool skip_on_incompatible_target_; 1438 // True if we found an OUTPUT_FORMAT with an incompatible target. 1439 bool found_incompatible_target_; 1440 // May be NULL if the user chooses not to pass one in. 1441 Command_line* command_line_; 1442 // Options which may be set from any linker script. 1443 Script_options* script_options_; 1444 // Information parsed from a version script. 1445 Version_script_info* version_script_info_; 1446 // The lexer. 1447 Lex* lex_; 1448 // The line number of the last token returned by next_token. 1449 int lineno_; 1450 // The column number of the last token returned by next_token. 1451 int charpos_; 1452 // A stack of lexer modes. 1453 std::vector<Lex::Mode> lex_mode_stack_; 1454 // A stack of which extern/language block we're inside. Can be C++, 1455 // java, or empty for C. 1456 std::vector<Version_script_info::Language> language_stack_; 1457 // New input files found to add to the link. 1458 Input_arguments* inputs_; 1459 // Pointer to incremental linking info. 1460 Script_info* script_info_; 1461}; 1462 1463// FILE was found as an argument on the command line. Try to read it 1464// as a script. Return true if the file was handled. 1465 1466bool 1467read_input_script(Workqueue* workqueue, Symbol_table* symtab, Layout* layout, 1468 Dirsearch* dirsearch, int dirindex, 1469 Input_objects* input_objects, Mapfile* mapfile, 1470 Input_group* input_group, 1471 const Input_argument* input_argument, 1472 Input_file* input_file, Task_token* next_blocker, 1473 bool* used_next_blocker) 1474{ 1475 *used_next_blocker = false; 1476 1477 std::string input_string; 1478 Lex::read_file(input_file, &input_string); 1479 1480 Lex lex(input_string.c_str(), input_string.length(), PARSING_LINKER_SCRIPT); 1481 1482 Script_info* script_info = NULL; 1483 if (layout->incremental_inputs() != NULL) 1484 { 1485 const std::string& filename = input_file->filename(); 1486 Timespec mtime = input_file->file().get_mtime(); 1487 unsigned int arg_serial = input_argument->file().arg_serial(); 1488 script_info = new Script_info(filename); 1489 layout->incremental_inputs()->report_script(script_info, arg_serial, 1490 mtime); 1491 } 1492 1493 Parser_closure closure(input_file->filename().c_str(), 1494 input_argument->file().options(), 1495 false, 1496 input_group != NULL, 1497 input_file->is_in_sysroot(), 1498 NULL, 1499 layout->script_options(), 1500 &lex, 1501 input_file->will_search_for(), 1502 script_info); 1503 1504 bool old_saw_sections_clause = 1505 layout->script_options()->saw_sections_clause(); 1506 1507 if (yyparse(&closure) != 0) 1508 { 1509 if (closure.found_incompatible_target()) 1510 { 1511 Read_symbols::incompatible_warning(input_argument, input_file); 1512 Read_symbols::requeue(workqueue, input_objects, symtab, layout, 1513 dirsearch, dirindex, mapfile, input_argument, 1514 input_group, next_blocker); 1515 return true; 1516 } 1517 return false; 1518 } 1519 1520 if (!old_saw_sections_clause 1521 && layout->script_options()->saw_sections_clause() 1522 && layout->have_added_input_section()) 1523 gold_error(_("%s: SECTIONS seen after other input files; try -T/--script"), 1524 input_file->filename().c_str()); 1525 1526 if (!closure.saw_inputs()) 1527 return true; 1528 1529 Task_token* this_blocker = NULL; 1530 for (Input_arguments::const_iterator p = closure.inputs()->begin(); 1531 p != closure.inputs()->end(); 1532 ++p) 1533 { 1534 Task_token* nb; 1535 if (p + 1 == closure.inputs()->end()) 1536 nb = next_blocker; 1537 else 1538 { 1539 nb = new Task_token(true); 1540 nb->add_blocker(); 1541 } 1542 workqueue->queue_soon(new Read_symbols(input_objects, symtab, 1543 layout, dirsearch, 0, mapfile, &*p, 1544 input_group, NULL, this_blocker, nb)); 1545 this_blocker = nb; 1546 } 1547 1548 *used_next_blocker = true; 1549 1550 return true; 1551} 1552 1553// Helper function for read_version_script(), read_commandline_script() and 1554// script_include_directive(). Processes the given file in the mode indicated 1555// by first_token and lex_mode. 1556 1557static bool 1558read_script_file(const char* filename, Command_line* cmdline, 1559 Script_options* script_options, 1560 int first_token, Lex::Mode lex_mode) 1561{ 1562 Dirsearch dirsearch; 1563 std::string name = filename; 1564 1565 // If filename is a relative filename, search for it manually using "." + 1566 // cmdline->options()->library_path() -- not dirsearch. 1567 if (!IS_ABSOLUTE_PATH(filename)) 1568 { 1569 const General_options::Dir_list& search_path = 1570 cmdline->options().library_path(); 1571 name = Dirsearch::find_file_in_dir_list(name, search_path, "."); 1572 } 1573 1574 // The file locking code wants to record a Task, but we haven't 1575 // started the workqueue yet. This is only for debugging purposes, 1576 // so we invent a fake value. 1577 const Task* task = reinterpret_cast<const Task*>(-1); 1578 1579 // We don't want this file to be opened in binary mode. 1580 Position_dependent_options posdep = cmdline->position_dependent_options(); 1581 if (posdep.format_enum() == General_options::OBJECT_FORMAT_BINARY) 1582 posdep.set_format_enum(General_options::OBJECT_FORMAT_ELF); 1583 Input_file_argument input_argument(name.c_str(), 1584 Input_file_argument::INPUT_FILE_TYPE_FILE, 1585 "", false, posdep); 1586 Input_file input_file(&input_argument); 1587 int dummy = 0; 1588 if (!input_file.open(dirsearch, task, &dummy)) 1589 return false; 1590 1591 std::string input_string; 1592 Lex::read_file(&input_file, &input_string); 1593 1594 Lex lex(input_string.c_str(), input_string.length(), first_token); 1595 lex.set_mode(lex_mode); 1596 1597 Parser_closure closure(filename, 1598 cmdline->position_dependent_options(), 1599 first_token == Lex::DYNAMIC_LIST, 1600 false, 1601 input_file.is_in_sysroot(), 1602 cmdline, 1603 script_options, 1604 &lex, 1605 false, 1606 NULL); 1607 if (yyparse(&closure) != 0) 1608 { 1609 input_file.file().unlock(task); 1610 return false; 1611 } 1612 1613 input_file.file().unlock(task); 1614 1615 gold_assert(!closure.saw_inputs()); 1616 1617 return true; 1618} 1619 1620// FILENAME was found as an argument to --script (-T). 1621// Read it as a script, and execute its contents immediately. 1622 1623bool 1624read_commandline_script(const char* filename, Command_line* cmdline) 1625{ 1626 return read_script_file(filename, cmdline, &cmdline->script_options(), 1627 PARSING_LINKER_SCRIPT, Lex::LINKER_SCRIPT); 1628} 1629 1630// FILENAME was found as an argument to --version-script. Read it as 1631// a version script, and store its contents in 1632// cmdline->script_options()->version_script_info(). 1633 1634bool 1635read_version_script(const char* filename, Command_line* cmdline) 1636{ 1637 return read_script_file(filename, cmdline, &cmdline->script_options(), 1638 PARSING_VERSION_SCRIPT, Lex::VERSION_SCRIPT); 1639} 1640 1641// FILENAME was found as an argument to --dynamic-list. Read it as a 1642// list of symbols, and store its contents in DYNAMIC_LIST. 1643 1644bool 1645read_dynamic_list(const char* filename, Command_line* cmdline, 1646 Script_options* dynamic_list) 1647{ 1648 return read_script_file(filename, cmdline, dynamic_list, 1649 PARSING_DYNAMIC_LIST, Lex::DYNAMIC_LIST); 1650} 1651 1652// Implement the --defsym option on the command line. Return true if 1653// all is well. 1654 1655bool 1656Script_options::define_symbol(const char* definition) 1657{ 1658 Lex lex(definition, strlen(definition), PARSING_DEFSYM); 1659 lex.set_mode(Lex::EXPRESSION); 1660 1661 // Dummy value. 1662 Position_dependent_options posdep_options; 1663 1664 Parser_closure closure("command line", posdep_options, true, 1665 false, false, NULL, this, &lex, false, NULL); 1666 1667 if (yyparse(&closure) != 0) 1668 return false; 1669 1670 gold_assert(!closure.saw_inputs()); 1671 1672 return true; 1673} 1674 1675// Print the script to F for debugging. 1676 1677void 1678Script_options::print(FILE* f) const 1679{ 1680 fprintf(f, "%s: Dumping linker script\n", program_name); 1681 1682 if (!this->entry_.empty()) 1683 fprintf(f, "ENTRY(%s)\n", this->entry_.c_str()); 1684 1685 for (Symbol_assignments::const_iterator p = 1686 this->symbol_assignments_.begin(); 1687 p != this->symbol_assignments_.end(); 1688 ++p) 1689 (*p)->print(f); 1690 1691 for (Assertions::const_iterator p = this->assertions_.begin(); 1692 p != this->assertions_.end(); 1693 ++p) 1694 (*p)->print(f); 1695 1696 this->script_sections_.print(f); 1697 1698 this->version_script_info_.print(f); 1699} 1700 1701// Manage mapping from keywords to the codes expected by the bison 1702// parser. We construct one global object for each lex mode with 1703// keywords. 1704 1705class Keyword_to_parsecode 1706{ 1707 public: 1708 // The structure which maps keywords to parsecodes. 1709 struct Keyword_parsecode 1710 { 1711 // Keyword. 1712 const char* keyword; 1713 // Corresponding parsecode. 1714 int parsecode; 1715 }; 1716 1717 Keyword_to_parsecode(const Keyword_parsecode* keywords, 1718 int keyword_count) 1719 : keyword_parsecodes_(keywords), keyword_count_(keyword_count) 1720 { } 1721 1722 // Return the parsecode corresponding KEYWORD, or 0 if it is not a 1723 // keyword. 1724 int 1725 keyword_to_parsecode(const char* keyword, size_t len) const; 1726 1727 private: 1728 const Keyword_parsecode* keyword_parsecodes_; 1729 const int keyword_count_; 1730}; 1731 1732// Mapping from keyword string to keyword parsecode. This array must 1733// be kept in sorted order. Parsecodes are looked up using bsearch. 1734// This array must correspond to the list of parsecodes in yyscript.y. 1735 1736static const Keyword_to_parsecode::Keyword_parsecode 1737script_keyword_parsecodes[] = 1738{ 1739 { "ABSOLUTE", ABSOLUTE }, 1740 { "ADDR", ADDR }, 1741 { "ALIGN", ALIGN_K }, 1742 { "ALIGNOF", ALIGNOF }, 1743 { "ASSERT", ASSERT_K }, 1744 { "AS_NEEDED", AS_NEEDED }, 1745 { "AT", AT }, 1746 { "BIND", BIND }, 1747 { "BLOCK", BLOCK }, 1748 { "BYTE", BYTE }, 1749 { "CONSTANT", CONSTANT }, 1750 { "CONSTRUCTORS", CONSTRUCTORS }, 1751 { "COPY", COPY }, 1752 { "CREATE_OBJECT_SYMBOLS", CREATE_OBJECT_SYMBOLS }, 1753 { "DATA_SEGMENT_ALIGN", DATA_SEGMENT_ALIGN }, 1754 { "DATA_SEGMENT_END", DATA_SEGMENT_END }, 1755 { "DATA_SEGMENT_RELRO_END", DATA_SEGMENT_RELRO_END }, 1756 { "DEFINED", DEFINED }, 1757 { "DSECT", DSECT }, 1758 { "ENTRY", ENTRY }, 1759 { "EXCLUDE_FILE", EXCLUDE_FILE }, 1760 { "EXTERN", EXTERN }, 1761 { "FILL", FILL }, 1762 { "FLOAT", FLOAT }, 1763 { "FORCE_COMMON_ALLOCATION", FORCE_COMMON_ALLOCATION }, 1764 { "GROUP", GROUP }, 1765 { "HIDDEN", HIDDEN }, 1766 { "HLL", HLL }, 1767 { "INCLUDE", INCLUDE }, 1768 { "INFO", INFO }, 1769 { "INHIBIT_COMMON_ALLOCATION", INHIBIT_COMMON_ALLOCATION }, 1770 { "INPUT", INPUT }, 1771 { "KEEP", KEEP }, 1772 { "LENGTH", LENGTH }, 1773 { "LOADADDR", LOADADDR }, 1774 { "LONG", LONG }, 1775 { "MAP", MAP }, 1776 { "MAX", MAX_K }, 1777 { "MEMORY", MEMORY }, 1778 { "MIN", MIN_K }, 1779 { "NEXT", NEXT }, 1780 { "NOCROSSREFS", NOCROSSREFS }, 1781 { "NOFLOAT", NOFLOAT }, 1782 { "NOLOAD", NOLOAD }, 1783 { "ONLY_IF_RO", ONLY_IF_RO }, 1784 { "ONLY_IF_RW", ONLY_IF_RW }, 1785 { "OPTION", OPTION }, 1786 { "ORIGIN", ORIGIN }, 1787 { "OUTPUT", OUTPUT }, 1788 { "OUTPUT_ARCH", OUTPUT_ARCH }, 1789 { "OUTPUT_FORMAT", OUTPUT_FORMAT }, 1790 { "OVERLAY", OVERLAY }, 1791 { "PHDRS", PHDRS }, 1792 { "PROVIDE", PROVIDE }, 1793 { "PROVIDE_HIDDEN", PROVIDE_HIDDEN }, 1794 { "QUAD", QUAD }, 1795 { "SEARCH_DIR", SEARCH_DIR }, 1796 { "SECTIONS", SECTIONS }, 1797 { "SEGMENT_START", SEGMENT_START }, 1798 { "SHORT", SHORT }, 1799 { "SIZEOF", SIZEOF }, 1800 { "SIZEOF_HEADERS", SIZEOF_HEADERS }, 1801 { "SORT", SORT_BY_NAME }, 1802 { "SORT_BY_ALIGNMENT", SORT_BY_ALIGNMENT }, 1803 { "SORT_BY_INIT_PRIORITY", SORT_BY_INIT_PRIORITY }, 1804 { "SORT_BY_NAME", SORT_BY_NAME }, 1805 { "SPECIAL", SPECIAL }, 1806 { "SQUAD", SQUAD }, 1807 { "STARTUP", STARTUP }, 1808 { "SUBALIGN", SUBALIGN }, 1809 { "SYSLIB", SYSLIB }, 1810 { "TARGET", TARGET_K }, 1811 { "TRUNCATE", TRUNCATE }, 1812 { "VERSION", VERSIONK }, 1813 { "global", GLOBAL }, 1814 { "l", LENGTH }, 1815 { "len", LENGTH }, 1816 { "local", LOCAL }, 1817 { "o", ORIGIN }, 1818 { "org", ORIGIN }, 1819 { "sizeof_headers", SIZEOF_HEADERS }, 1820}; 1821 1822static const Keyword_to_parsecode 1823script_keywords(&script_keyword_parsecodes[0], 1824 (sizeof(script_keyword_parsecodes) 1825 / sizeof(script_keyword_parsecodes[0]))); 1826 1827static const Keyword_to_parsecode::Keyword_parsecode 1828version_script_keyword_parsecodes[] = 1829{ 1830 { "extern", EXTERN }, 1831 { "global", GLOBAL }, 1832 { "local", LOCAL }, 1833}; 1834 1835static const Keyword_to_parsecode 1836version_script_keywords(&version_script_keyword_parsecodes[0], 1837 (sizeof(version_script_keyword_parsecodes) 1838 / sizeof(version_script_keyword_parsecodes[0]))); 1839 1840static const Keyword_to_parsecode::Keyword_parsecode 1841dynamic_list_keyword_parsecodes[] = 1842{ 1843 { "extern", EXTERN }, 1844}; 1845 1846static const Keyword_to_parsecode 1847dynamic_list_keywords(&dynamic_list_keyword_parsecodes[0], 1848 (sizeof(dynamic_list_keyword_parsecodes) 1849 / sizeof(dynamic_list_keyword_parsecodes[0]))); 1850 1851 1852 1853// Comparison function passed to bsearch. 1854 1855extern "C" 1856{ 1857 1858struct Ktt_key 1859{ 1860 const char* str; 1861 size_t len; 1862}; 1863 1864static int 1865ktt_compare(const void* keyv, const void* kttv) 1866{ 1867 const Ktt_key* key = static_cast<const Ktt_key*>(keyv); 1868 const Keyword_to_parsecode::Keyword_parsecode* ktt = 1869 static_cast<const Keyword_to_parsecode::Keyword_parsecode*>(kttv); 1870 int i = strncmp(key->str, ktt->keyword, key->len); 1871 if (i != 0) 1872 return i; 1873 if (ktt->keyword[key->len] != '\0') 1874 return -1; 1875 return 0; 1876} 1877 1878} // End extern "C". 1879 1880int 1881Keyword_to_parsecode::keyword_to_parsecode(const char* keyword, 1882 size_t len) const 1883{ 1884 Ktt_key key; 1885 key.str = keyword; 1886 key.len = len; 1887 void* kttv = bsearch(&key, 1888 this->keyword_parsecodes_, 1889 this->keyword_count_, 1890 sizeof(this->keyword_parsecodes_[0]), 1891 ktt_compare); 1892 if (kttv == NULL) 1893 return 0; 1894 Keyword_parsecode* ktt = static_cast<Keyword_parsecode*>(kttv); 1895 return ktt->parsecode; 1896} 1897 1898// The following structs are used within the VersionInfo class as well 1899// as in the bison helper functions. They store the information 1900// parsed from the version script. 1901 1902// A single version expression. 1903// For example, pattern="std::map*" and language="C++". 1904struct Version_expression 1905{ 1906 Version_expression(const std::string& a_pattern, 1907 Version_script_info::Language a_language, 1908 bool a_exact_match) 1909 : pattern(a_pattern), language(a_language), exact_match(a_exact_match), 1910 was_matched_by_symbol(false) 1911 { } 1912 1913 std::string pattern; 1914 Version_script_info::Language language; 1915 // If false, we use glob() to match pattern. If true, we use strcmp(). 1916 bool exact_match; 1917 // True if --no-undefined-version is in effect and we found this 1918 // version in get_symbol_version. We use mutable because this 1919 // struct is generally not modifiable after it has been created. 1920 mutable bool was_matched_by_symbol; 1921}; 1922 1923// A list of expressions. 1924struct Version_expression_list 1925{ 1926 std::vector<struct Version_expression> expressions; 1927}; 1928 1929// A list of which versions upon which another version depends. 1930// Strings should be from the Stringpool. 1931struct Version_dependency_list 1932{ 1933 std::vector<std::string> dependencies; 1934}; 1935 1936// The total definition of a version. It includes the tag for the 1937// version, its global and local expressions, and any dependencies. 1938struct Version_tree 1939{ 1940 Version_tree() 1941 : tag(), global(NULL), local(NULL), dependencies(NULL) 1942 { } 1943 1944 std::string tag; 1945 const struct Version_expression_list* global; 1946 const struct Version_expression_list* local; 1947 const struct Version_dependency_list* dependencies; 1948}; 1949 1950// Helper class that calls cplus_demangle when needed and takes care of freeing 1951// the result. 1952 1953class Lazy_demangler 1954{ 1955 public: 1956 Lazy_demangler(const char* symbol, int options) 1957 : symbol_(symbol), options_(options), demangled_(NULL), did_demangle_(false) 1958 { } 1959 1960 ~Lazy_demangler() 1961 { free(this->demangled_); } 1962 1963 // Return the demangled name. The actual demangling happens on the first call, 1964 // and the result is later cached. 1965 inline char* 1966 get(); 1967 1968 private: 1969 // The symbol to demangle. 1970 const char* symbol_; 1971 // Option flags to pass to cplus_demagle. 1972 const int options_; 1973 // The cached demangled value, or NULL if demangling didn't happen yet or 1974 // failed. 1975 char* demangled_; 1976 // Whether we already called cplus_demangle 1977 bool did_demangle_; 1978}; 1979 1980// Return the demangled name. The actual demangling happens on the first call, 1981// and the result is later cached. Returns NULL if the symbol cannot be 1982// demangled. 1983 1984inline char* 1985Lazy_demangler::get() 1986{ 1987 if (!this->did_demangle_) 1988 { 1989 this->demangled_ = cplus_demangle(this->symbol_, this->options_); 1990 this->did_demangle_ = true; 1991 } 1992 return this->demangled_; 1993} 1994 1995// Class Version_script_info. 1996 1997Version_script_info::Version_script_info() 1998 : dependency_lists_(), expression_lists_(), version_trees_(), globs_(), 1999 default_version_(NULL), default_is_global_(false), is_finalized_(false) 2000{ 2001 for (int i = 0; i < LANGUAGE_COUNT; ++i) 2002 this->exact_[i] = NULL; 2003} 2004 2005Version_script_info::~Version_script_info() 2006{ 2007} 2008 2009// Forget all the known version script information. 2010 2011void 2012Version_script_info::clear() 2013{ 2014 for (size_t k = 0; k < this->dependency_lists_.size(); ++k) 2015 delete this->dependency_lists_[k]; 2016 this->dependency_lists_.clear(); 2017 for (size_t k = 0; k < this->version_trees_.size(); ++k) 2018 delete this->version_trees_[k]; 2019 this->version_trees_.clear(); 2020 for (size_t k = 0; k < this->expression_lists_.size(); ++k) 2021 delete this->expression_lists_[k]; 2022 this->expression_lists_.clear(); 2023} 2024 2025// Finalize the version script information. 2026 2027void 2028Version_script_info::finalize() 2029{ 2030 if (!this->is_finalized_) 2031 { 2032 this->build_lookup_tables(); 2033 this->is_finalized_ = true; 2034 } 2035} 2036 2037// Return all the versions. 2038 2039std::vector<std::string> 2040Version_script_info::get_versions() const 2041{ 2042 std::vector<std::string> ret; 2043 for (size_t j = 0; j < this->version_trees_.size(); ++j) 2044 if (!this->version_trees_[j]->tag.empty()) 2045 ret.push_back(this->version_trees_[j]->tag); 2046 return ret; 2047} 2048 2049// Return the dependencies of VERSION. 2050 2051std::vector<std::string> 2052Version_script_info::get_dependencies(const char* version) const 2053{ 2054 std::vector<std::string> ret; 2055 for (size_t j = 0; j < this->version_trees_.size(); ++j) 2056 if (this->version_trees_[j]->tag == version) 2057 { 2058 const struct Version_dependency_list* deps = 2059 this->version_trees_[j]->dependencies; 2060 if (deps != NULL) 2061 for (size_t k = 0; k < deps->dependencies.size(); ++k) 2062 ret.push_back(deps->dependencies[k]); 2063 return ret; 2064 } 2065 return ret; 2066} 2067 2068// A version script essentially maps a symbol name to a version tag 2069// and an indication of whether symbol is global or local within that 2070// version tag. Each symbol maps to at most one version tag. 2071// Unfortunately, in practice, version scripts are ambiguous, and list 2072// symbols multiple times. Thus, we have to document the matching 2073// process. 2074 2075// This is a description of what the GNU linker does as of 2010-01-11. 2076// It walks through the version tags in the order in which they appear 2077// in the version script. For each tag, it first walks through the 2078// global patterns for that tag, then the local patterns. When 2079// looking at a single pattern, it first applies any language specific 2080// demangling as specified for the pattern, and then matches the 2081// resulting symbol name to the pattern. If it finds an exact match 2082// for a literal pattern (a pattern enclosed in quotes or with no 2083// wildcard characters), then that is the match that it uses. If 2084// finds a match with a wildcard pattern, then it saves it and 2085// continues searching. Wildcard patterns that are exactly "*" are 2086// saved separately. 2087 2088// If no exact match with a literal pattern is ever found, then if a 2089// wildcard match with a global pattern was found it is used, 2090// otherwise if a wildcard match with a local pattern was found it is 2091// used. 2092 2093// This is the result: 2094// * If there is an exact match, then we use the first tag in the 2095// version script where it matches. 2096// + If the exact match in that tag is global, it is used. 2097// + Otherwise the exact match in that tag is local, and is used. 2098// * Otherwise, if there is any match with a global wildcard pattern: 2099// + If there is any match with a wildcard pattern which is not 2100// "*", then we use the tag in which the *last* such pattern 2101// appears. 2102// + Otherwise, we matched "*". If there is no match with a local 2103// wildcard pattern which is not "*", then we use the *last* 2104// match with a global "*". Otherwise, continue. 2105// * Otherwise, if there is any match with a local wildcard pattern: 2106// + If there is any match with a wildcard pattern which is not 2107// "*", then we use the tag in which the *last* such pattern 2108// appears. 2109// + Otherwise, we matched "*", and we use the tag in which the 2110// *last* such match occurred. 2111 2112// There is an additional wrinkle. When the GNU linker finds a symbol 2113// with a version defined in an object file due to a .symver 2114// directive, it looks up that symbol name in that version tag. If it 2115// finds it, it matches the symbol name against the patterns for that 2116// version. If there is no match with a global pattern, but there is 2117// a match with a local pattern, then the GNU linker marks the symbol 2118// as local. 2119 2120// We want gold to be generally compatible, but we also want gold to 2121// be fast. These are the rules that gold implements: 2122// * If there is an exact match for the mangled name, we use it. 2123// + If there is more than one exact match, we give a warning, and 2124// we use the first tag in the script which matches. 2125// + If a symbol has an exact match as both global and local for 2126// the same version tag, we give an error. 2127// * Otherwise, we look for an extern C++ or an extern Java exact 2128// match. If we find an exact match, we use it. 2129// + If there is more than one exact match, we give a warning, and 2130// we use the first tag in the script which matches. 2131// + If a symbol has an exact match as both global and local for 2132// the same version tag, we give an error. 2133// * Otherwise, we look through the wildcard patterns, ignoring "*" 2134// patterns. We look through the version tags in reverse order. 2135// For each version tag, we look through the global patterns and 2136// then the local patterns. We use the first match we find (i.e., 2137// the last matching version tag in the file). 2138// * Otherwise, we use the "*" pattern if there is one. We give an 2139// error if there are multiple "*" patterns. 2140 2141// At least for now, gold does not look up the version tag for a 2142// symbol version found in an object file to see if it should be 2143// forced local. There are other ways to force a symbol to be local, 2144// and I don't understand why this one is useful. 2145 2146// Build a set of fast lookup tables for a version script. 2147 2148void 2149Version_script_info::build_lookup_tables() 2150{ 2151 size_t size = this->version_trees_.size(); 2152 for (size_t j = 0; j < size; ++j) 2153 { 2154 const Version_tree* v = this->version_trees_[j]; 2155 this->build_expression_list_lookup(v->local, v, false); 2156 this->build_expression_list_lookup(v->global, v, true); 2157 } 2158} 2159 2160// If a pattern has backlashes but no unquoted wildcard characters, 2161// then we apply backslash unquoting and look for an exact match. 2162// Otherwise we treat it as a wildcard pattern. This function returns 2163// true for a wildcard pattern. Otherwise, it does backslash 2164// unquoting on *PATTERN and returns false. If this returns true, 2165// *PATTERN may have been partially unquoted. 2166 2167bool 2168Version_script_info::unquote(std::string* pattern) const 2169{ 2170 bool saw_backslash = false; 2171 size_t len = pattern->length(); 2172 size_t j = 0; 2173 for (size_t i = 0; i < len; ++i) 2174 { 2175 if (saw_backslash) 2176 saw_backslash = false; 2177 else 2178 { 2179 switch ((*pattern)[i]) 2180 { 2181 case '?': case '[': case '*': 2182 return true; 2183 case '\\': 2184 saw_backslash = true; 2185 continue; 2186 default: 2187 break; 2188 } 2189 } 2190 2191 if (i != j) 2192 (*pattern)[j] = (*pattern)[i]; 2193 ++j; 2194 } 2195 return false; 2196} 2197 2198// Add an exact match for MATCH to *PE. The result of the match is 2199// V/IS_GLOBAL. 2200 2201void 2202Version_script_info::add_exact_match(const std::string& match, 2203 const Version_tree* v, bool is_global, 2204 const Version_expression* ve, 2205 Exact* pe) 2206{ 2207 std::pair<Exact::iterator, bool> ins = 2208 pe->insert(std::make_pair(match, Version_tree_match(v, is_global, ve))); 2209 if (ins.second) 2210 { 2211 // This is the first time we have seen this match. 2212 return; 2213 } 2214 2215 Version_tree_match& vtm(ins.first->second); 2216 if (vtm.real->tag != v->tag) 2217 { 2218 // This is an ambiguous match. We still return the 2219 // first version that we found in the script, but we 2220 // record the new version to issue a warning if we 2221 // wind up looking up this symbol. 2222 if (vtm.ambiguous == NULL) 2223 vtm.ambiguous = v; 2224 } 2225 else if (is_global != vtm.is_global) 2226 { 2227 // We have a match for both the global and local entries for a 2228 // version tag. That's got to be wrong. 2229 gold_error(_("'%s' appears as both a global and a local symbol " 2230 "for version '%s' in script"), 2231 match.c_str(), v->tag.c_str()); 2232 } 2233} 2234 2235// Build fast lookup information for EXPLIST and store it in LOOKUP. 2236// All matches go to V, and IS_GLOBAL is true if they are global 2237// matches. 2238 2239void 2240Version_script_info::build_expression_list_lookup( 2241 const Version_expression_list* explist, 2242 const Version_tree* v, 2243 bool is_global) 2244{ 2245 if (explist == NULL) 2246 return; 2247 size_t size = explist->expressions.size(); 2248 for (size_t i = 0; i < size; ++i) 2249 { 2250 const Version_expression& exp(explist->expressions[i]); 2251 2252 if (exp.pattern.length() == 1 && exp.pattern[0] == '*') 2253 { 2254 if (this->default_version_ != NULL 2255 && this->default_version_->tag != v->tag) 2256 gold_warning(_("wildcard match appears in both version '%s' " 2257 "and '%s' in script"), 2258 this->default_version_->tag.c_str(), v->tag.c_str()); 2259 else if (this->default_version_ != NULL 2260 && this->default_is_global_ != is_global) 2261 gold_error(_("wildcard match appears as both global and local " 2262 "in version '%s' in script"), 2263 v->tag.c_str()); 2264 this->default_version_ = v; 2265 this->default_is_global_ = is_global; 2266 continue; 2267 } 2268 2269 std::string pattern = exp.pattern; 2270 if (!exp.exact_match) 2271 { 2272 if (this->unquote(&pattern)) 2273 { 2274 this->globs_.push_back(Glob(&exp, v, is_global)); 2275 continue; 2276 } 2277 } 2278 2279 if (this->exact_[exp.language] == NULL) 2280 this->exact_[exp.language] = new Exact(); 2281 this->add_exact_match(pattern, v, is_global, &exp, 2282 this->exact_[exp.language]); 2283 } 2284} 2285 2286// Return the name to match given a name, a language code, and two 2287// lazy demanglers. 2288 2289const char* 2290Version_script_info::get_name_to_match(const char* name, 2291 int language, 2292 Lazy_demangler* cpp_demangler, 2293 Lazy_demangler* java_demangler) const 2294{ 2295 switch (language) 2296 { 2297 case LANGUAGE_C: 2298 return name; 2299 case LANGUAGE_CXX: 2300 return cpp_demangler->get(); 2301 case LANGUAGE_JAVA: 2302 return java_demangler->get(); 2303 default: 2304 gold_unreachable(); 2305 } 2306} 2307 2308// Look up SYMBOL_NAME in the list of versions. Return true if the 2309// symbol is found, false if not. If the symbol is found, then if 2310// PVERSION is not NULL, set *PVERSION to the version tag, and if 2311// P_IS_GLOBAL is not NULL, set *P_IS_GLOBAL according to whether the 2312// symbol is global or not. 2313 2314bool 2315Version_script_info::get_symbol_version(const char* symbol_name, 2316 std::string* pversion, 2317 bool* p_is_global) const 2318{ 2319 Lazy_demangler cpp_demangled_name(symbol_name, DMGL_ANSI | DMGL_PARAMS); 2320 Lazy_demangler java_demangled_name(symbol_name, 2321 DMGL_ANSI | DMGL_PARAMS | DMGL_JAVA); 2322 2323 gold_assert(this->is_finalized_); 2324 for (int i = 0; i < LANGUAGE_COUNT; ++i) 2325 { 2326 Exact* exact = this->exact_[i]; 2327 if (exact == NULL) 2328 continue; 2329 2330 const char* name_to_match = this->get_name_to_match(symbol_name, i, 2331 &cpp_demangled_name, 2332 &java_demangled_name); 2333 if (name_to_match == NULL) 2334 { 2335 // If the name can not be demangled, the GNU linker goes 2336 // ahead and tries to match it anyhow. That does not 2337 // make sense to me and I have not implemented it. 2338 continue; 2339 } 2340 2341 Exact::const_iterator pe = exact->find(name_to_match); 2342 if (pe != exact->end()) 2343 { 2344 const Version_tree_match& vtm(pe->second); 2345 if (vtm.ambiguous != NULL) 2346 gold_warning(_("using '%s' as version for '%s' which is also " 2347 "named in version '%s' in script"), 2348 vtm.real->tag.c_str(), name_to_match, 2349 vtm.ambiguous->tag.c_str()); 2350 2351 if (pversion != NULL) 2352 *pversion = vtm.real->tag; 2353 if (p_is_global != NULL) 2354 *p_is_global = vtm.is_global; 2355 2356 // If we are using --no-undefined-version, and this is a 2357 // global symbol, we have to record that we have found this 2358 // symbol, so that we don't warn about it. We have to do 2359 // this now, because otherwise we have no way to get from a 2360 // non-C language back to the demangled name that we 2361 // matched. 2362 if (p_is_global != NULL && vtm.is_global) 2363 vtm.expression->was_matched_by_symbol = true; 2364 2365 return true; 2366 } 2367 } 2368 2369 // Look through the glob patterns in reverse order. 2370 2371 for (Globs::const_reverse_iterator p = this->globs_.rbegin(); 2372 p != this->globs_.rend(); 2373 ++p) 2374 { 2375 int language = p->expression->language; 2376 const char* name_to_match = this->get_name_to_match(symbol_name, 2377 language, 2378 &cpp_demangled_name, 2379 &java_demangled_name); 2380 if (name_to_match == NULL) 2381 continue; 2382 2383 if (fnmatch(p->expression->pattern.c_str(), name_to_match, 2384 FNM_NOESCAPE) == 0) 2385 { 2386 if (pversion != NULL) 2387 *pversion = p->version->tag; 2388 if (p_is_global != NULL) 2389 *p_is_global = p->is_global; 2390 return true; 2391 } 2392 } 2393 2394 // Finally, there may be a wildcard. 2395 if (this->default_version_ != NULL) 2396 { 2397 if (pversion != NULL) 2398 *pversion = this->default_version_->tag; 2399 if (p_is_global != NULL) 2400 *p_is_global = this->default_is_global_; 2401 return true; 2402 } 2403 2404 return false; 2405} 2406 2407// Give an error if any exact symbol names (not wildcards) appear in a 2408// version script, but there is no such symbol. 2409 2410void 2411Version_script_info::check_unmatched_names(const Symbol_table* symtab) const 2412{ 2413 for (size_t i = 0; i < this->version_trees_.size(); ++i) 2414 { 2415 const Version_tree* vt = this->version_trees_[i]; 2416 if (vt->global == NULL) 2417 continue; 2418 for (size_t j = 0; j < vt->global->expressions.size(); ++j) 2419 { 2420 const Version_expression& expression(vt->global->expressions[j]); 2421 2422 // Ignore cases where we used the version because we saw a 2423 // symbol that we looked up. Note that 2424 // WAS_MATCHED_BY_SYMBOL will be true even if the symbol was 2425 // not a definition. That's OK as in that case we most 2426 // likely gave an undefined symbol error anyhow. 2427 if (expression.was_matched_by_symbol) 2428 continue; 2429 2430 // Just ignore names which are in languages other than C. 2431 // We have no way to look them up in the symbol table. 2432 if (expression.language != LANGUAGE_C) 2433 continue; 2434 2435 // Remove backslash quoting, and ignore wildcard patterns. 2436 std::string pattern = expression.pattern; 2437 if (!expression.exact_match) 2438 { 2439 if (this->unquote(&pattern)) 2440 continue; 2441 } 2442 2443 if (symtab->lookup(pattern.c_str(), vt->tag.c_str()) == NULL) 2444 gold_error(_("version script assignment of %s to symbol %s " 2445 "failed: symbol not defined"), 2446 vt->tag.c_str(), pattern.c_str()); 2447 } 2448 } 2449} 2450 2451struct Version_dependency_list* 2452Version_script_info::allocate_dependency_list() 2453{ 2454 dependency_lists_.push_back(new Version_dependency_list); 2455 return dependency_lists_.back(); 2456} 2457 2458struct Version_expression_list* 2459Version_script_info::allocate_expression_list() 2460{ 2461 expression_lists_.push_back(new Version_expression_list); 2462 return expression_lists_.back(); 2463} 2464 2465struct Version_tree* 2466Version_script_info::allocate_version_tree() 2467{ 2468 version_trees_.push_back(new Version_tree); 2469 return version_trees_.back(); 2470} 2471 2472// Print for debugging. 2473 2474void 2475Version_script_info::print(FILE* f) const 2476{ 2477 if (this->empty()) 2478 return; 2479 2480 fprintf(f, "VERSION {"); 2481 2482 for (size_t i = 0; i < this->version_trees_.size(); ++i) 2483 { 2484 const Version_tree* vt = this->version_trees_[i]; 2485 2486 if (vt->tag.empty()) 2487 fprintf(f, " {\n"); 2488 else 2489 fprintf(f, " %s {\n", vt->tag.c_str()); 2490 2491 if (vt->global != NULL) 2492 { 2493 fprintf(f, " global :\n"); 2494 this->print_expression_list(f, vt->global); 2495 } 2496 2497 if (vt->local != NULL) 2498 { 2499 fprintf(f, " local :\n"); 2500 this->print_expression_list(f, vt->local); 2501 } 2502 2503 fprintf(f, " }"); 2504 if (vt->dependencies != NULL) 2505 { 2506 const Version_dependency_list* deps = vt->dependencies; 2507 for (size_t j = 0; j < deps->dependencies.size(); ++j) 2508 { 2509 if (j < deps->dependencies.size() - 1) 2510 fprintf(f, "\n"); 2511 fprintf(f, " %s", deps->dependencies[j].c_str()); 2512 } 2513 } 2514 fprintf(f, ";\n"); 2515 } 2516 2517 fprintf(f, "}\n"); 2518} 2519 2520void 2521Version_script_info::print_expression_list( 2522 FILE* f, 2523 const Version_expression_list* vel) const 2524{ 2525 Version_script_info::Language current_language = LANGUAGE_C; 2526 for (size_t i = 0; i < vel->expressions.size(); ++i) 2527 { 2528 const Version_expression& ve(vel->expressions[i]); 2529 2530 if (ve.language != current_language) 2531 { 2532 if (current_language != LANGUAGE_C) 2533 fprintf(f, " }\n"); 2534 switch (ve.language) 2535 { 2536 case LANGUAGE_C: 2537 break; 2538 case LANGUAGE_CXX: 2539 fprintf(f, " extern \"C++\" {\n"); 2540 break; 2541 case LANGUAGE_JAVA: 2542 fprintf(f, " extern \"Java\" {\n"); 2543 break; 2544 default: 2545 gold_unreachable(); 2546 } 2547 current_language = ve.language; 2548 } 2549 2550 fprintf(f, " "); 2551 if (current_language != LANGUAGE_C) 2552 fprintf(f, " "); 2553 2554 if (ve.exact_match) 2555 fprintf(f, "\""); 2556 fprintf(f, "%s", ve.pattern.c_str()); 2557 if (ve.exact_match) 2558 fprintf(f, "\""); 2559 2560 fprintf(f, "\n"); 2561 } 2562 2563 if (current_language != LANGUAGE_C) 2564 fprintf(f, " }\n"); 2565} 2566 2567} // End namespace gold. 2568 2569// The remaining functions are extern "C", so it's clearer to not put 2570// them in namespace gold. 2571 2572using namespace gold; 2573 2574// This function is called by the bison parser to return the next 2575// token. 2576 2577extern "C" int 2578yylex(YYSTYPE* lvalp, void* closurev) 2579{ 2580 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2581 const Token* token = closure->next_token(); 2582 switch (token->classification()) 2583 { 2584 default: 2585 gold_unreachable(); 2586 2587 case Token::TOKEN_INVALID: 2588 yyerror(closurev, "invalid character"); 2589 return 0; 2590 2591 case Token::TOKEN_EOF: 2592 return 0; 2593 2594 case Token::TOKEN_STRING: 2595 { 2596 // This is either a keyword or a STRING. 2597 size_t len; 2598 const char* str = token->string_value(&len); 2599 int parsecode = 0; 2600 switch (closure->lex_mode()) 2601 { 2602 case Lex::LINKER_SCRIPT: 2603 parsecode = script_keywords.keyword_to_parsecode(str, len); 2604 break; 2605 case Lex::VERSION_SCRIPT: 2606 parsecode = version_script_keywords.keyword_to_parsecode(str, len); 2607 break; 2608 case Lex::DYNAMIC_LIST: 2609 parsecode = dynamic_list_keywords.keyword_to_parsecode(str, len); 2610 break; 2611 default: 2612 break; 2613 } 2614 if (parsecode != 0) 2615 return parsecode; 2616 lvalp->string.value = str; 2617 lvalp->string.length = len; 2618 return STRING; 2619 } 2620 2621 case Token::TOKEN_QUOTED_STRING: 2622 lvalp->string.value = token->string_value(&lvalp->string.length); 2623 return QUOTED_STRING; 2624 2625 case Token::TOKEN_OPERATOR: 2626 return token->operator_value(); 2627 2628 case Token::TOKEN_INTEGER: 2629 lvalp->integer = token->integer_value(); 2630 return INTEGER; 2631 } 2632} 2633 2634// This function is called by the bison parser to report an error. 2635 2636extern "C" void 2637yyerror(void* closurev, const char* message) 2638{ 2639 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2640 gold_error(_("%s:%d:%d: %s"), closure->filename(), closure->lineno(), 2641 closure->charpos(), message); 2642} 2643 2644// Called by the bison parser to add an external symbol to the link. 2645 2646extern "C" void 2647script_add_extern(void* closurev, const char* name, size_t length) 2648{ 2649 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2650 closure->script_options()->add_symbol_reference(name, length); 2651} 2652 2653// Called by the bison parser to add a file to the link. 2654 2655extern "C" void 2656script_add_file(void* closurev, const char* name, size_t length) 2657{ 2658 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2659 2660 // If this is an absolute path, and we found the script in the 2661 // sysroot, then we want to prepend the sysroot to the file name. 2662 // For example, this is how we handle a cross link to the x86_64 2663 // libc.so, which refers to /lib/libc.so.6. 2664 std::string name_string(name, length); 2665 const char* extra_search_path = "."; 2666 std::string script_directory; 2667 if (IS_ABSOLUTE_PATH(name_string.c_str())) 2668 { 2669 if (closure->is_in_sysroot()) 2670 { 2671 const std::string& sysroot(parameters->options().sysroot()); 2672 gold_assert(!sysroot.empty()); 2673 name_string = sysroot + name_string; 2674 } 2675 } 2676 else 2677 { 2678 // In addition to checking the normal library search path, we 2679 // also want to check in the script-directory. 2680 const char* slash = strrchr(closure->filename(), '/'); 2681 if (slash != NULL) 2682 { 2683 script_directory.assign(closure->filename(), 2684 slash - closure->filename() + 1); 2685 extra_search_path = script_directory.c_str(); 2686 } 2687 } 2688 2689 Input_file_argument file(name_string.c_str(), 2690 Input_file_argument::INPUT_FILE_TYPE_FILE, 2691 extra_search_path, false, 2692 closure->position_dependent_options()); 2693 Input_argument& arg = closure->inputs()->add_file(file); 2694 arg.set_script_info(closure->script_info()); 2695} 2696 2697// Called by the bison parser to add a library to the link. 2698 2699extern "C" void 2700script_add_library(void* closurev, const char* name, size_t length) 2701{ 2702 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2703 std::string name_string(name, length); 2704 2705 if (name_string[0] != 'l') 2706 gold_error(_("library name must be prefixed with -l")); 2707 2708 Input_file_argument file(name_string.c_str() + 1, 2709 Input_file_argument::INPUT_FILE_TYPE_LIBRARY, 2710 "", false, 2711 closure->position_dependent_options()); 2712 Input_argument& arg = closure->inputs()->add_file(file); 2713 arg.set_script_info(closure->script_info()); 2714} 2715 2716// Called by the bison parser to start a group. If we are already in 2717// a group, that means that this script was invoked within a 2718// --start-group --end-group sequence on the command line, or that 2719// this script was found in a GROUP of another script. In that case, 2720// we simply continue the existing group, rather than starting a new 2721// one. It is possible to construct a case in which this will do 2722// something other than what would happen if we did a recursive group, 2723// but it's hard to imagine why the different behaviour would be 2724// useful for a real program. Avoiding recursive groups is simpler 2725// and more efficient. 2726 2727extern "C" void 2728script_start_group(void* closurev) 2729{ 2730 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2731 if (!closure->in_group()) 2732 closure->inputs()->start_group(); 2733} 2734 2735// Called by the bison parser at the end of a group. 2736 2737extern "C" void 2738script_end_group(void* closurev) 2739{ 2740 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2741 if (!closure->in_group()) 2742 closure->inputs()->end_group(); 2743} 2744 2745// Called by the bison parser to start an AS_NEEDED list. 2746 2747extern "C" void 2748script_start_as_needed(void* closurev) 2749{ 2750 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2751 closure->position_dependent_options().set_as_needed(true); 2752} 2753 2754// Called by the bison parser at the end of an AS_NEEDED list. 2755 2756extern "C" void 2757script_end_as_needed(void* closurev) 2758{ 2759 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2760 closure->position_dependent_options().set_as_needed(false); 2761} 2762 2763// Called by the bison parser to set the entry symbol. 2764 2765extern "C" void 2766script_set_entry(void* closurev, const char* entry, size_t length) 2767{ 2768 // We'll parse this exactly the same as --entry=ENTRY on the commandline 2769 // TODO(csilvers): FIXME -- call set_entry directly. 2770 std::string arg("--entry="); 2771 arg.append(entry, length); 2772 script_parse_option(closurev, arg.c_str(), arg.size()); 2773} 2774 2775// Called by the bison parser to set whether to define common symbols. 2776 2777extern "C" void 2778script_set_common_allocation(void* closurev, int set) 2779{ 2780 const char* arg = set != 0 ? "--define-common" : "--no-define-common"; 2781 script_parse_option(closurev, arg, strlen(arg)); 2782} 2783 2784// Called by the bison parser to refer to a symbol. 2785 2786extern "C" Expression* 2787script_symbol(void* closurev, const char* name, size_t length) 2788{ 2789 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2790 if (length != 1 || name[0] != '.') 2791 closure->script_options()->add_symbol_reference(name, length); 2792 return script_exp_string(name, length); 2793} 2794 2795// Called by the bison parser to define a symbol. 2796 2797extern "C" void 2798script_set_symbol(void* closurev, const char* name, size_t length, 2799 Expression* value, int providei, int hiddeni) 2800{ 2801 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2802 const bool provide = providei != 0; 2803 const bool hidden = hiddeni != 0; 2804 closure->script_options()->add_symbol_assignment(name, length, 2805 closure->parsing_defsym(), 2806 value, provide, hidden); 2807 closure->clear_skip_on_incompatible_target(); 2808} 2809 2810// Called by the bison parser to add an assertion. 2811 2812extern "C" void 2813script_add_assertion(void* closurev, Expression* check, const char* message, 2814 size_t messagelen) 2815{ 2816 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2817 closure->script_options()->add_assertion(check, message, messagelen); 2818 closure->clear_skip_on_incompatible_target(); 2819} 2820 2821// Called by the bison parser to parse an OPTION. 2822 2823extern "C" void 2824script_parse_option(void* closurev, const char* option, size_t length) 2825{ 2826 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2827 // We treat the option as a single command-line option, even if 2828 // it has internal whitespace. 2829 if (closure->command_line() == NULL) 2830 { 2831 // There are some options that we could handle here--e.g., 2832 // -lLIBRARY. Should we bother? 2833 gold_warning(_("%s:%d:%d: ignoring command OPTION; OPTION is only valid" 2834 " for scripts specified via -T/--script"), 2835 closure->filename(), closure->lineno(), closure->charpos()); 2836 } 2837 else 2838 { 2839 bool past_a_double_dash_option = false; 2840 const char* mutable_option = strndup(option, length); 2841 gold_assert(mutable_option != NULL); 2842 closure->command_line()->process_one_option(1, &mutable_option, 0, 2843 &past_a_double_dash_option); 2844 // The General_options class will quite possibly store a pointer 2845 // into mutable_option, so we can't free it. In cases the class 2846 // does not store such a pointer, this is a memory leak. Alas. :( 2847 } 2848 closure->clear_skip_on_incompatible_target(); 2849} 2850 2851// Called by the bison parser to handle OUTPUT_FORMAT. OUTPUT_FORMAT 2852// takes either one or three arguments. In the three argument case, 2853// the format depends on the endianness option, which we don't 2854// currently support (FIXME). If we see an OUTPUT_FORMAT for the 2855// wrong format, then we want to search for a new file. Returning 0 2856// here will cause the parser to immediately abort. 2857 2858extern "C" int 2859script_check_output_format(void* closurev, 2860 const char* default_name, size_t default_length, 2861 const char*, size_t, const char*, size_t) 2862{ 2863 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2864 std::string name(default_name, default_length); 2865 Target* target = select_target_by_bfd_name(name.c_str()); 2866 if (target == NULL || !parameters->is_compatible_target(target)) 2867 { 2868 if (closure->skip_on_incompatible_target()) 2869 { 2870 closure->set_found_incompatible_target(); 2871 return 0; 2872 } 2873 // FIXME: Should we warn about the unknown target? 2874 } 2875 return 1; 2876} 2877 2878// Called by the bison parser to handle TARGET. 2879 2880extern "C" void 2881script_set_target(void* closurev, const char* target, size_t len) 2882{ 2883 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2884 std::string s(target, len); 2885 General_options::Object_format format_enum; 2886 format_enum = General_options::string_to_object_format(s.c_str()); 2887 closure->position_dependent_options().set_format_enum(format_enum); 2888} 2889 2890// Called by the bison parser to handle SEARCH_DIR. This is handled 2891// exactly like a -L option. 2892 2893extern "C" void 2894script_add_search_dir(void* closurev, const char* option, size_t length) 2895{ 2896 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2897 if (closure->command_line() == NULL) 2898 gold_warning(_("%s:%d:%d: ignoring SEARCH_DIR; SEARCH_DIR is only valid" 2899 " for scripts specified via -T/--script"), 2900 closure->filename(), closure->lineno(), closure->charpos()); 2901 else if (!closure->command_line()->options().nostdlib()) 2902 { 2903 std::string s = "-L" + std::string(option, length); 2904 script_parse_option(closurev, s.c_str(), s.size()); 2905 } 2906} 2907 2908/* Called by the bison parser to push the lexer into expression 2909 mode. */ 2910 2911extern "C" void 2912script_push_lex_into_expression_mode(void* closurev) 2913{ 2914 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2915 closure->push_lex_mode(Lex::EXPRESSION); 2916} 2917 2918/* Called by the bison parser to push the lexer into version 2919 mode. */ 2920 2921extern "C" void 2922script_push_lex_into_version_mode(void* closurev) 2923{ 2924 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2925 if (closure->version_script()->is_finalized()) 2926 gold_error(_("%s:%d:%d: invalid use of VERSION in input file"), 2927 closure->filename(), closure->lineno(), closure->charpos()); 2928 closure->push_lex_mode(Lex::VERSION_SCRIPT); 2929} 2930 2931/* Called by the bison parser to pop the lexer mode. */ 2932 2933extern "C" void 2934script_pop_lex_mode(void* closurev) 2935{ 2936 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2937 closure->pop_lex_mode(); 2938} 2939 2940// Register an entire version node. For example: 2941// 2942// GLIBC_2.1 { 2943// global: foo; 2944// } GLIBC_2.0; 2945// 2946// - tag is "GLIBC_2.1" 2947// - tree contains the information "global: foo" 2948// - deps contains "GLIBC_2.0" 2949 2950extern "C" void 2951script_register_vers_node(void*, 2952 const char* tag, 2953 int taglen, 2954 struct Version_tree* tree, 2955 struct Version_dependency_list* deps) 2956{ 2957 gold_assert(tree != NULL); 2958 tree->dependencies = deps; 2959 if (tag != NULL) 2960 tree->tag = std::string(tag, taglen); 2961} 2962 2963// Add a dependencies to the list of existing dependencies, if any, 2964// and return the expanded list. 2965 2966extern "C" struct Version_dependency_list* 2967script_add_vers_depend(void* closurev, 2968 struct Version_dependency_list* all_deps, 2969 const char* depend_to_add, int deplen) 2970{ 2971 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2972 if (all_deps == NULL) 2973 all_deps = closure->version_script()->allocate_dependency_list(); 2974 all_deps->dependencies.push_back(std::string(depend_to_add, deplen)); 2975 return all_deps; 2976} 2977 2978// Add a pattern expression to an existing list of expressions, if any. 2979 2980extern "C" struct Version_expression_list* 2981script_new_vers_pattern(void* closurev, 2982 struct Version_expression_list* expressions, 2983 const char* pattern, int patlen, int exact_match) 2984{ 2985 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 2986 if (expressions == NULL) 2987 expressions = closure->version_script()->allocate_expression_list(); 2988 expressions->expressions.push_back( 2989 Version_expression(std::string(pattern, patlen), 2990 closure->get_current_language(), 2991 static_cast<bool>(exact_match))); 2992 return expressions; 2993} 2994 2995// Attaches b to the end of a, and clears b. So a = a + b and b = {}. 2996 2997extern "C" struct Version_expression_list* 2998script_merge_expressions(struct Version_expression_list* a, 2999 struct Version_expression_list* b) 3000{ 3001 a->expressions.insert(a->expressions.end(), 3002 b->expressions.begin(), b->expressions.end()); 3003 // We could delete b and remove it from expressions_lists_, but 3004 // that's a lot of work. This works just as well. 3005 b->expressions.clear(); 3006 return a; 3007} 3008 3009// Combine the global and local expressions into a a Version_tree. 3010 3011extern "C" struct Version_tree* 3012script_new_vers_node(void* closurev, 3013 struct Version_expression_list* global, 3014 struct Version_expression_list* local) 3015{ 3016 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3017 Version_tree* tree = closure->version_script()->allocate_version_tree(); 3018 tree->global = global; 3019 tree->local = local; 3020 return tree; 3021} 3022 3023// Handle a transition in language, such as at the 3024// start or end of 'extern "C++"' 3025 3026extern "C" void 3027version_script_push_lang(void* closurev, const char* lang, int langlen) 3028{ 3029 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3030 std::string language(lang, langlen); 3031 Version_script_info::Language code; 3032 if (language.empty() || language == "C") 3033 code = Version_script_info::LANGUAGE_C; 3034 else if (language == "C++") 3035 code = Version_script_info::LANGUAGE_CXX; 3036 else if (language == "Java") 3037 code = Version_script_info::LANGUAGE_JAVA; 3038 else 3039 { 3040 char* buf = new char[langlen + 100]; 3041 snprintf(buf, langlen + 100, 3042 _("unrecognized version script language '%s'"), 3043 language.c_str()); 3044 yyerror(closurev, buf); 3045 delete[] buf; 3046 code = Version_script_info::LANGUAGE_C; 3047 } 3048 closure->push_language(code); 3049} 3050 3051extern "C" void 3052version_script_pop_lang(void* closurev) 3053{ 3054 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3055 closure->pop_language(); 3056} 3057 3058// Called by the bison parser to start a SECTIONS clause. 3059 3060extern "C" void 3061script_start_sections(void* closurev) 3062{ 3063 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3064 closure->script_options()->script_sections()->start_sections(); 3065 closure->clear_skip_on_incompatible_target(); 3066} 3067 3068// Called by the bison parser to finish a SECTIONS clause. 3069 3070extern "C" void 3071script_finish_sections(void* closurev) 3072{ 3073 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3074 closure->script_options()->script_sections()->finish_sections(); 3075} 3076 3077// Start processing entries for an output section. 3078 3079extern "C" void 3080script_start_output_section(void* closurev, const char* name, size_t namelen, 3081 const struct Parser_output_section_header* header) 3082{ 3083 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3084 closure->script_options()->script_sections()->start_output_section(name, 3085 namelen, 3086 header); 3087} 3088 3089// Finish processing entries for an output section. 3090 3091extern "C" void 3092script_finish_output_section(void* closurev, 3093 const struct Parser_output_section_trailer* trail) 3094{ 3095 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3096 closure->script_options()->script_sections()->finish_output_section(trail); 3097} 3098 3099// Add a data item (e.g., "WORD (0)") to the current output section. 3100 3101extern "C" void 3102script_add_data(void* closurev, int data_token, Expression* val) 3103{ 3104 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3105 int size; 3106 bool is_signed = true; 3107 switch (data_token) 3108 { 3109 case QUAD: 3110 size = 8; 3111 is_signed = false; 3112 break; 3113 case SQUAD: 3114 size = 8; 3115 break; 3116 case LONG: 3117 size = 4; 3118 break; 3119 case SHORT: 3120 size = 2; 3121 break; 3122 case BYTE: 3123 size = 1; 3124 break; 3125 default: 3126 gold_unreachable(); 3127 } 3128 closure->script_options()->script_sections()->add_data(size, is_signed, val); 3129} 3130 3131// Add a clause setting the fill value to the current output section. 3132 3133extern "C" void 3134script_add_fill(void* closurev, Expression* val) 3135{ 3136 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3137 closure->script_options()->script_sections()->add_fill(val); 3138} 3139 3140// Add a new input section specification to the current output 3141// section. 3142 3143extern "C" void 3144script_add_input_section(void* closurev, 3145 const struct Input_section_spec* spec, 3146 int keepi) 3147{ 3148 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3149 bool keep = keepi != 0; 3150 closure->script_options()->script_sections()->add_input_section(spec, keep); 3151} 3152 3153// When we see DATA_SEGMENT_ALIGN we record that following output 3154// sections may be relro. 3155 3156extern "C" void 3157script_data_segment_align(void* closurev) 3158{ 3159 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3160 if (!closure->script_options()->saw_sections_clause()) 3161 gold_error(_("%s:%d:%d: DATA_SEGMENT_ALIGN not in SECTIONS clause"), 3162 closure->filename(), closure->lineno(), closure->charpos()); 3163 else 3164 closure->script_options()->script_sections()->data_segment_align(); 3165} 3166 3167// When we see DATA_SEGMENT_RELRO_END we know that all output sections 3168// since DATA_SEGMENT_ALIGN should be relro. 3169 3170extern "C" void 3171script_data_segment_relro_end(void* closurev) 3172{ 3173 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3174 if (!closure->script_options()->saw_sections_clause()) 3175 gold_error(_("%s:%d:%d: DATA_SEGMENT_ALIGN not in SECTIONS clause"), 3176 closure->filename(), closure->lineno(), closure->charpos()); 3177 else 3178 closure->script_options()->script_sections()->data_segment_relro_end(); 3179} 3180 3181// Create a new list of string/sort pairs. 3182 3183extern "C" String_sort_list_ptr 3184script_new_string_sort_list(const struct Wildcard_section* string_sort) 3185{ 3186 return new String_sort_list(1, *string_sort); 3187} 3188 3189// Add an entry to a list of string/sort pairs. The way the parser 3190// works permits us to simply modify the first parameter, rather than 3191// copy the vector. 3192 3193extern "C" String_sort_list_ptr 3194script_string_sort_list_add(String_sort_list_ptr pv, 3195 const struct Wildcard_section* string_sort) 3196{ 3197 if (pv == NULL) 3198 return script_new_string_sort_list(string_sort); 3199 else 3200 { 3201 pv->push_back(*string_sort); 3202 return pv; 3203 } 3204} 3205 3206// Create a new list of strings. 3207 3208extern "C" String_list_ptr 3209script_new_string_list(const char* str, size_t len) 3210{ 3211 return new String_list(1, std::string(str, len)); 3212} 3213 3214// Add an element to a list of strings. The way the parser works 3215// permits us to simply modify the first parameter, rather than copy 3216// the vector. 3217 3218extern "C" String_list_ptr 3219script_string_list_push_back(String_list_ptr pv, const char* str, size_t len) 3220{ 3221 if (pv == NULL) 3222 return script_new_string_list(str, len); 3223 else 3224 { 3225 pv->push_back(std::string(str, len)); 3226 return pv; 3227 } 3228} 3229 3230// Concatenate two string lists. Either or both may be NULL. The way 3231// the parser works permits us to modify the parameters, rather than 3232// copy the vector. 3233 3234extern "C" String_list_ptr 3235script_string_list_append(String_list_ptr pv1, String_list_ptr pv2) 3236{ 3237 if (pv1 == NULL) 3238 return pv2; 3239 if (pv2 == NULL) 3240 return pv1; 3241 pv1->insert(pv1->end(), pv2->begin(), pv2->end()); 3242 return pv1; 3243} 3244 3245// Add a new program header. 3246 3247extern "C" void 3248script_add_phdr(void* closurev, const char* name, size_t namelen, 3249 unsigned int type, const Phdr_info* info) 3250{ 3251 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3252 bool includes_filehdr = info->includes_filehdr != 0; 3253 bool includes_phdrs = info->includes_phdrs != 0; 3254 bool is_flags_valid = info->is_flags_valid != 0; 3255 Script_sections* ss = closure->script_options()->script_sections(); 3256 ss->add_phdr(name, namelen, type, includes_filehdr, includes_phdrs, 3257 is_flags_valid, info->flags, info->load_address); 3258 closure->clear_skip_on_incompatible_target(); 3259} 3260 3261// Convert a program header string to a type. 3262 3263#define PHDR_TYPE(NAME) { #NAME, sizeof(#NAME) - 1, elfcpp::NAME } 3264 3265static struct 3266{ 3267 const char* name; 3268 size_t namelen; 3269 unsigned int val; 3270} phdr_type_names[] = 3271{ 3272 PHDR_TYPE(PT_NULL), 3273 PHDR_TYPE(PT_LOAD), 3274 PHDR_TYPE(PT_DYNAMIC), 3275 PHDR_TYPE(PT_INTERP), 3276 PHDR_TYPE(PT_NOTE), 3277 PHDR_TYPE(PT_SHLIB), 3278 PHDR_TYPE(PT_PHDR), 3279 PHDR_TYPE(PT_TLS), 3280 PHDR_TYPE(PT_GNU_EH_FRAME), 3281 PHDR_TYPE(PT_GNU_STACK), 3282 PHDR_TYPE(PT_GNU_RELRO) 3283}; 3284 3285extern "C" unsigned int 3286script_phdr_string_to_type(void* closurev, const char* name, size_t namelen) 3287{ 3288 for (unsigned int i = 0; 3289 i < sizeof(phdr_type_names) / sizeof(phdr_type_names[0]); 3290 ++i) 3291 if (namelen == phdr_type_names[i].namelen 3292 && strncmp(name, phdr_type_names[i].name, namelen) == 0) 3293 return phdr_type_names[i].val; 3294 yyerror(closurev, _("unknown PHDR type (try integer)")); 3295 return elfcpp::PT_NULL; 3296} 3297 3298extern "C" void 3299script_saw_segment_start_expression(void* closurev) 3300{ 3301 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3302 Script_sections* ss = closure->script_options()->script_sections(); 3303 ss->set_saw_segment_start_expression(true); 3304} 3305 3306extern "C" void 3307script_set_section_region(void* closurev, const char* name, size_t namelen, 3308 int set_vma) 3309{ 3310 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3311 if (!closure->script_options()->saw_sections_clause()) 3312 { 3313 gold_error(_("%s:%d:%d: MEMORY region '%.*s' referred to outside of " 3314 "SECTIONS clause"), 3315 closure->filename(), closure->lineno(), closure->charpos(), 3316 static_cast<int>(namelen), name); 3317 return; 3318 } 3319 3320 Script_sections* ss = closure->script_options()->script_sections(); 3321 Memory_region* mr = ss->find_memory_region(name, namelen); 3322 if (mr == NULL) 3323 { 3324 gold_error(_("%s:%d:%d: MEMORY region '%.*s' not declared"), 3325 closure->filename(), closure->lineno(), closure->charpos(), 3326 static_cast<int>(namelen), name); 3327 return; 3328 } 3329 3330 ss->set_memory_region(mr, set_vma); 3331} 3332 3333extern "C" void 3334script_add_memory(void* closurev, const char* name, size_t namelen, 3335 unsigned int attrs, Expression* origin, Expression* length) 3336{ 3337 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3338 Script_sections* ss = closure->script_options()->script_sections(); 3339 ss->add_memory_region(name, namelen, attrs, origin, length); 3340} 3341 3342extern "C" unsigned int 3343script_parse_memory_attr(void* closurev, const char* attrs, size_t attrlen, 3344 int invert) 3345{ 3346 int attributes = 0; 3347 3348 while (attrlen--) 3349 switch (*attrs++) 3350 { 3351 case 'R': 3352 case 'r': 3353 attributes |= MEM_READABLE; break; 3354 case 'W': 3355 case 'w': 3356 attributes |= MEM_READABLE | MEM_WRITEABLE; break; 3357 case 'X': 3358 case 'x': 3359 attributes |= MEM_EXECUTABLE; break; 3360 case 'A': 3361 case 'a': 3362 attributes |= MEM_ALLOCATABLE; break; 3363 case 'I': 3364 case 'i': 3365 case 'L': 3366 case 'l': 3367 attributes |= MEM_INITIALIZED; break; 3368 default: 3369 yyerror(closurev, _("unknown MEMORY attribute")); 3370 } 3371 3372 if (invert) 3373 attributes = (~ attributes) & MEM_ATTR_MASK; 3374 3375 return attributes; 3376} 3377 3378extern "C" void 3379script_include_directive(int first_token, void* closurev, 3380 const char* filename, size_t length) 3381{ 3382 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3383 std::string name(filename, length); 3384 Command_line* cmdline = closure->command_line(); 3385 read_script_file(name.c_str(), cmdline, &cmdline->script_options(), 3386 first_token, Lex::LINKER_SCRIPT); 3387} 3388 3389// Functions for memory regions. 3390 3391extern "C" Expression* 3392script_exp_function_origin(void* closurev, const char* name, size_t namelen) 3393{ 3394 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3395 Script_sections* ss = closure->script_options()->script_sections(); 3396 Expression* origin = ss->find_memory_region_origin(name, namelen); 3397 3398 if (origin == NULL) 3399 { 3400 gold_error(_("undefined memory region '%s' referenced " 3401 "in ORIGIN expression"), 3402 name); 3403 // Create a dummy expression to prevent crashes later on. 3404 origin = script_exp_integer(0); 3405 } 3406 3407 return origin; 3408} 3409 3410extern "C" Expression* 3411script_exp_function_length(void* closurev, const char* name, size_t namelen) 3412{ 3413 Parser_closure* closure = static_cast<Parser_closure*>(closurev); 3414 Script_sections* ss = closure->script_options()->script_sections(); 3415 Expression* length = ss->find_memory_region_length(name, namelen); 3416 3417 if (length == NULL) 3418 { 3419 gold_error(_("undefined memory region '%s' referenced " 3420 "in LENGTH expression"), 3421 name); 3422 // Create a dummy expression to prevent crashes later on. 3423 length = script_exp_integer(0); 3424 } 3425 3426 return length; 3427} 3428