1 // script.cc -- handle linker scripts for gold.
3 // Copyright 2006, 2007, 2008, 2009, 2010, 2011 Free Software Foundation, Inc.
4 // Written by Ian Lance Taylor <iant@google.com>.
6 // This file is part of gold.
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.
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.
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.
31 #include "filenames.h"
35 #include "dirsearch.h"
38 #include "workqueue.h"
40 #include "parameters.h"
43 #include "target-select.h"
46 #include "incremental.h"
51 // A token read from a script file. We don't implement keywords here;
52 // all keywords are simply represented as a string.
57 // Token classification.
62 // Token indicates end of input.
64 // Token is a string of characters.
66 // Token is a quoted string of characters.
68 // Token is an operator.
70 // Token is a number (an integer).
74 // We need an empty constructor so that we can put this STL objects.
76 : classification_(TOKEN_INVALID), value_(NULL), value_length_(0),
77 opcode_(0), lineno_(0), charpos_(0)
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)
85 gold_assert(classification == TOKEN_INVALID
86 || classification == TOKEN_EOF);
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)
95 gold_assert(classification != TOKEN_INVALID
96 && classification != TOKEN_EOF);
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)
105 // Return whether the token is invalid.
108 { return this->classification_ == TOKEN_INVALID; }
110 // Return whether this is an EOF token.
113 { return this->classification_ == TOKEN_EOF; }
115 // Return the token classification.
117 classification() const
118 { return this->classification_; }
120 // Return the line number at which the token starts.
123 { return this->lineno_; }
125 // Return the character position at this the token starts.
128 { return this->charpos_; }
130 // Get the value of a token.
133 string_value(size_t* length) const
135 gold_assert(this->classification_ == TOKEN_STRING
136 || this->classification_ == TOKEN_QUOTED_STRING);
137 *length = this->value_length_;
142 operator_value() const
144 gold_assert(this->classification_ == TOKEN_OPERATOR);
145 return this->opcode_;
149 integer_value() const;
152 // The token classification.
153 Classification classification_;
154 // The token value, for TOKEN_STRING or TOKEN_QUOTED_STRING or
157 // The length of the token value.
158 size_t value_length_;
159 // The token value, for TOKEN_OPERATOR.
161 // The line number where this token started (one based).
163 // The character position within the line where this token started
168 // Return the value of a TOKEN_INTEGER.
171 Token::integer_value() const
173 gold_assert(this->classification_ == TOKEN_INTEGER);
175 size_t len = this->value_length_;
177 uint64_t multiplier = 1;
178 char last = this->value_[len - 1];
179 if (last == 'm' || last == 'M')
181 multiplier = 1024 * 1024;
184 else if (last == 'k' || last == 'K')
191 uint64_t ret = strtoull(this->value_, &end, 0);
192 gold_assert(static_cast<size_t>(end - this->value_) == len);
194 return ret * multiplier;
197 // This class handles lexing a file into a sequence of tokens.
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.
207 // Reading an ordinary linker script.
209 // Reading an expression in a linker script.
211 // Reading a version script.
213 // Reading a --dynamic-list file.
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)
224 // Read a file into a string.
226 read_file(Input_file*, std::string*);
228 // Return the next token.
232 // Return the current lexing mode.
235 { return this->mode_; }
237 // Set the lexing mode.
240 { this->mode_ = mode; }
244 Lex& operator=(const Lex&);
246 // Make a general token with no value at the current location.
248 make_token(Token::Classification c, const char* start) const
249 { return Token(c, this->lineno_, start - this->linestart_ + 1); }
251 // Make a general token with a value at the current location.
253 make_token(Token::Classification c, const char* v, size_t len,
256 { return Token(c, v, len, this->lineno_, start - this->linestart_ + 1); }
258 // Make an operator token at the current location.
260 make_token(int opcode, const char* start) const
261 { return Token(opcode, this->lineno_, start - this->linestart_ + 1); }
263 // Make an invalid token at the current location.
265 make_invalid_token(const char* start)
266 { return this->make_token(Token::TOKEN_INVALID, start); }
268 // Make an EOF token at the current location.
270 make_eof_token(const char* start)
271 { return this->make_token(Token::TOKEN_EOF, start); }
273 // Return whether C can be the first character in a name. C2 is the
274 // next character, since we sometimes need that.
276 can_start_name(char c, char c2);
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.
282 can_continue_name(const char* c);
284 // Return whether C, C2, C3 can start a hex number.
286 can_start_hex(char c, char c2, char c3);
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.
292 can_continue_hex(const char* c);
294 // Return whether C can start a non-hex number.
296 can_start_number(char c);
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.
302 can_continue_number(const char* c)
303 { return Lex::can_start_number(*c) ? c + 1 : NULL; }
305 // If C1 C2 C3 form a valid three character operator, return the
306 // opcode. Otherwise return 0.
308 three_char_operator(char c1, char c2, char c3);
310 // If C1 C2 form a valid two character operator, return the opcode.
311 // Otherwise return 0.
313 two_char_operator(char c1, char c2);
315 // If C1 is a valid one character operator, return the opcode.
316 // Otherwise return 0.
318 one_char_operator(char c1);
320 // Read the next token.
322 get_token(const char**);
324 // Skip a C style /* */ comment. Return false if the comment did
327 skip_c_comment(const char**);
329 // Skip a line # comment. Return false if there was no newline.
331 skip_line_comment(const char**);
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.
337 gather_token(Token::Classification,
338 const char* (Lex::*can_continue_fn)(const char*),
339 const char* start, const char* match, const char** pp);
341 // Build a token from a quoted string.
343 gather_quoted_string(const char** pp);
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.
353 // The code to use for the first token. This is set to 0 after it
356 // The current token.
358 // The current line number.
360 // The start of the current line in the string.
361 const char* linestart_;
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.
369 Lex::read_file(Input_file* input_file, std::string* contents)
371 off_t filesize = input_file->file().filesize();
373 contents->reserve(filesize);
376 unsigned char buf[BUFSIZ];
377 while (off < filesize)
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);
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
400 Lex::can_start_name(char c, char c2)
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':
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':
414 case '_': case '.': case '$':
418 return this->mode_ == LINKER_SCRIPT;
421 return this->mode_ == LINKER_SCRIPT && can_continue_name(&c2);
424 return (this->mode_ == VERSION_SCRIPT
425 || this->mode_ == DYNAMIC_LIST
426 || (this->mode_ == LINKER_SCRIPT
427 && can_continue_name(&c2)));
434 // Return whether C can continue a name which has already started.
435 // Subsequent characters in a name are the same as the leading
436 // characters, plus digits and "=+-:[],?*". So in general the linker
437 // script language requires spaces around operators, unless we know
438 // that we are parsing an expression.
441 Lex::can_continue_name(const char* c)
445 case 'A': case 'B': case 'C': case 'D': case 'E': case 'F':
446 case 'G': case 'H': case 'I': case 'J': case 'K': case 'L':
447 case 'M': case 'N': case 'O': case 'Q': case 'P': case 'R':
448 case 'S': case 'T': case 'U': case 'V': case 'W': case 'X':
450 case 'a': case 'b': case 'c': case 'd': case 'e': case 'f':
451 case 'g': case 'h': case 'i': case 'j': case 'k': case 'l':
452 case 'm': case 'n': case 'o': case 'q': case 'p': case 'r':
453 case 's': case 't': case 'u': case 'v': case 'w': case 'x':
455 case '_': case '.': case '$':
456 case '0': case '1': case '2': case '3': case '4':
457 case '5': case '6': case '7': case '8': case '9':
460 // TODO(csilvers): why not allow ~ in names for version-scripts?
461 case '/': case '\\': case '~':
464 if (this->mode_ == LINKER_SCRIPT)
468 case '[': case ']': case '*': case '?': case '-':
469 if (this->mode_ == LINKER_SCRIPT || this->mode_ == VERSION_SCRIPT
470 || this->mode_ == DYNAMIC_LIST)
474 // TODO(csilvers): why allow this? ^ is meaningless in version scripts.
476 if (this->mode_ == VERSION_SCRIPT || this->mode_ == DYNAMIC_LIST)
481 if (this->mode_ == LINKER_SCRIPT)
483 else if ((this->mode_ == VERSION_SCRIPT || this->mode_ == DYNAMIC_LIST)
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.
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
502 // Return whether C1 C2 C3 can start a hex number.
505 Lex::can_start_hex(char c1, char c2, char c3)
507 if (c1 == '0' && (c2 == 'x' || c2 == 'X'))
508 return this->can_continue_hex(&c3);
512 // Return whether C can appear in a hex number.
515 Lex::can_continue_hex(const char* c)
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':
530 // Return whether C can start a non-hex number.
533 Lex::can_start_number(char c)
537 case '0': case '1': case '2': case '3': case '4':
538 case '5': case '6': case '7': case '8': case '9':
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.
551 Lex::three_char_operator(char c1, char c2, char c3)
556 if (c2 == '<' && c3 == '=')
560 if (c2 == '>' && c3 == '=')
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.
574 Lex::two_char_operator(char c1, char c2)
632 // If C1 is a valid operator, return the opcode. Otherwise return 0.
635 Lex::one_char_operator(char c1)
668 // Skip a C style comment. *PP points to just after the "/*". Return
669 // false if the comment did not end.
672 Lex::skip_c_comment(const char** pp)
675 while (p[0] != '*' || p[1] != '/')
686 this->linestart_ = p + 1;
695 // Skip a line # comment. Return false if there was no newline.
698 Lex::skip_line_comment(const char** pp)
701 size_t skip = strcspn(p, "\n");
710 this->linestart_ = p;
716 // Build a token CLASSIFICATION from all characters that match
717 // CAN_CONTINUE_FN. Update *PP.
720 Lex::gather_token(Token::Classification classification,
721 const char* (Lex::*can_continue_fn)(const char*),
726 const char* new_match = NULL;
727 while ((new_match = (this->*can_continue_fn)(match)) != NULL)
730 // A special case: integers may be followed by a single M or K,
732 if (classification == Token::TOKEN_INTEGER
733 && (*match == 'm' || *match == 'M' || *match == 'k' || *match == 'K'))
737 return this->make_token(classification, start, match - start, start);
740 // Build a token from a quoted string.
743 Lex::gather_quoted_string(const char** pp)
745 const char* start = *pp;
746 const char* p = start;
748 size_t skip = strcspn(p, "\"\n");
750 return this->make_invalid_token(start);
752 return this->make_token(Token::TOKEN_QUOTED_STRING, p, skip, start);
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
761 Lex::get_token(const char** pp)
770 return this->make_eof_token(p);
773 // Skip whitespace quickly.
774 while (*p == ' ' || *p == '\t' || *p == '\r')
781 this->linestart_ = p;
785 // Skip C style comments.
786 if (p[0] == '/' && p[1] == '*')
788 int lineno = this->lineno_;
789 int charpos = p - this->linestart_ + 1;
792 if (!this->skip_c_comment(pp))
793 return Token(Token::TOKEN_INVALID, lineno, charpos);
799 // Skip line comments.
803 if (!this->skip_line_comment(pp))
804 return this->make_eof_token(p);
810 if (this->can_start_name(p[0], p[1]))
811 return this->gather_token(Token::TOKEN_STRING,
812 &Lex::can_continue_name,
815 // We accept any arbitrary name in double quotes, as long as it
816 // does not cross a line boundary.
820 return this->gather_quoted_string(pp);
823 // Check for a number.
825 if (this->can_start_hex(p[0], p[1], p[2]))
826 return this->gather_token(Token::TOKEN_INTEGER,
827 &Lex::can_continue_hex,
830 if (Lex::can_start_number(p[0]))
831 return this->gather_token(Token::TOKEN_INTEGER,
832 &Lex::can_continue_number,
835 // Check for operators.
837 int opcode = Lex::three_char_operator(p[0], p[1], p[2]);
841 return this->make_token(opcode, p);
844 opcode = Lex::two_char_operator(p[0], p[1]);
848 return this->make_token(opcode, p);
851 opcode = Lex::one_char_operator(p[0]);
855 return this->make_token(opcode, p);
858 return this->make_token(Token::TOKEN_INVALID, p);
862 // Return the next token.
867 // The first token is special.
868 if (this->first_token_ != 0)
870 this->token_ = Token(this->first_token_, 0, 0);
871 this->first_token_ = 0;
872 return &this->token_;
875 this->token_ = this->get_token(&this->current_);
877 // Don't let an early null byte fool us into thinking that we've
878 // reached the end of the file.
879 if (this->token_.is_eof()
880 && (static_cast<size_t>(this->current_ - this->input_string_)
881 < this->input_length_))
882 this->token_ = this->make_invalid_token(this->current_);
884 return &this->token_;
887 // class Symbol_assignment.
889 // Add the symbol to the symbol table. This makes sure the symbol is
890 // there and defined. The actual value is stored later. We can't
891 // determine the actual value at this point, because we can't
892 // necessarily evaluate the expression until all ordinary symbols have
895 // The GNU linker lets symbol assignments in the linker script
896 // silently override defined symbols in object files. We are
897 // compatible. FIXME: Should we issue a warning?
900 Symbol_assignment::add_to_table(Symbol_table* symtab)
902 elfcpp::STV vis = this->hidden_ ? elfcpp::STV_HIDDEN : elfcpp::STV_DEFAULT;
903 this->sym_ = symtab->define_as_constant(this->name_.c_str(),
906 ? Symbol_table::DEFSYM
907 : Symbol_table::SCRIPT),
915 true); // force_override
918 // Finalize a symbol value.
921 Symbol_assignment::finalize(Symbol_table* symtab, const Layout* layout)
923 this->finalize_maybe_dot(symtab, layout, false, 0, NULL);
926 // Finalize a symbol value which can refer to the dot symbol.
929 Symbol_assignment::finalize_with_dot(Symbol_table* symtab,
930 const Layout* layout,
932 Output_section* dot_section)
934 this->finalize_maybe_dot(symtab, layout, true, dot_value, dot_section);
937 // Finalize a symbol value, internal version.
940 Symbol_assignment::finalize_maybe_dot(Symbol_table* symtab,
941 const Layout* layout,
942 bool is_dot_available,
944 Output_section* dot_section)
946 // If we were only supposed to provide this symbol, the sym_ field
947 // will be NULL if the symbol was not referenced.
948 if (this->sym_ == NULL)
950 gold_assert(this->provide_);
954 if (parameters->target().get_size() == 32)
956 #if defined(HAVE_TARGET_32_LITTLE) || defined(HAVE_TARGET_32_BIG)
957 this->sized_finalize<32>(symtab, layout, is_dot_available, dot_value,
963 else if (parameters->target().get_size() == 64)
965 #if defined(HAVE_TARGET_64_LITTLE) || defined(HAVE_TARGET_64_BIG)
966 this->sized_finalize<64>(symtab, layout, is_dot_available, dot_value,
978 Symbol_assignment::sized_finalize(Symbol_table* symtab, const Layout* layout,
979 bool is_dot_available, uint64_t dot_value,
980 Output_section* dot_section)
982 Output_section* section;
983 uint64_t final_val = this->val_->eval_maybe_dot(symtab, layout, true,
985 dot_value, dot_section,
987 Sized_symbol<size>* ssym = symtab->get_sized_symbol<size>(this->sym_);
988 ssym->set_value(final_val);
990 ssym->set_output_section(section);
993 // Set the symbol value if the expression yields an absolute value.
996 Symbol_assignment::set_if_absolute(Symbol_table* symtab, const Layout* layout,
997 bool is_dot_available, uint64_t dot_value)
999 if (this->sym_ == NULL)
1002 Output_section* val_section;
1003 uint64_t val = this->val_->eval_maybe_dot(symtab, layout, false,
1004 is_dot_available, dot_value,
1005 NULL, &val_section, NULL);
1006 if (val_section != NULL)
1009 if (parameters->target().get_size() == 32)
1011 #if defined(HAVE_TARGET_32_LITTLE) || defined(HAVE_TARGET_32_BIG)
1012 Sized_symbol<32>* ssym = symtab->get_sized_symbol<32>(this->sym_);
1013 ssym->set_value(val);
1018 else if (parameters->target().get_size() == 64)
1020 #if defined(HAVE_TARGET_64_LITTLE) || defined(HAVE_TARGET_64_BIG)
1021 Sized_symbol<64>* ssym = symtab->get_sized_symbol<64>(this->sym_);
1022 ssym->set_value(val);
1031 // Print for debugging.
1034 Symbol_assignment::print(FILE* f) const
1036 if (this->provide_ && this->hidden_)
1037 fprintf(f, "PROVIDE_HIDDEN(");
1038 else if (this->provide_)
1039 fprintf(f, "PROVIDE(");
1040 else if (this->hidden_)
1043 fprintf(f, "%s = ", this->name_.c_str());
1044 this->val_->print(f);
1046 if (this->provide_ || this->hidden_)
1052 // Class Script_assertion.
1054 // Check the assertion.
1057 Script_assertion::check(const Symbol_table* symtab, const Layout* layout)
1059 if (!this->check_->eval(symtab, layout, true))
1060 gold_error("%s", this->message_.c_str());
1063 // Print for debugging.
1066 Script_assertion::print(FILE* f) const
1068 fprintf(f, "ASSERT(");
1069 this->check_->print(f);
1070 fprintf(f, ", \"%s\")\n", this->message_.c_str());
1073 // Class Script_options.
1075 Script_options::Script_options()
1076 : entry_(), symbol_assignments_(), symbol_definitions_(),
1077 symbol_references_(), version_script_info_(), script_sections_()
1081 // Returns true if NAME is on the list of symbol assignments waiting
1085 Script_options::is_pending_assignment(const char* name)
1087 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin();
1088 p != this->symbol_assignments_.end();
1090 if ((*p)->name() == name)
1095 // Add a symbol to be defined.
1098 Script_options::add_symbol_assignment(const char* name, size_t length,
1099 bool is_defsym, Expression* value,
1100 bool provide, bool hidden)
1102 if (length != 1 || name[0] != '.')
1104 if (this->script_sections_.in_sections_clause())
1106 gold_assert(!is_defsym);
1107 this->script_sections_.add_symbol_assignment(name, length, value,
1112 Symbol_assignment* p = new Symbol_assignment(name, length, is_defsym,
1113 value, provide, hidden);
1114 this->symbol_assignments_.push_back(p);
1119 std::string n(name, length);
1120 this->symbol_definitions_.insert(n);
1121 this->symbol_references_.erase(n);
1126 if (provide || hidden)
1127 gold_error(_("invalid use of PROVIDE for dot symbol"));
1129 // The GNU linker permits assignments to dot outside of SECTIONS
1130 // clauses and treats them as occurring inside, so we don't
1131 // check in_sections_clause here.
1132 this->script_sections_.add_dot_assignment(value);
1136 // Add a reference to a symbol.
1139 Script_options::add_symbol_reference(const char* name, size_t length)
1141 if (length != 1 || name[0] != '.')
1143 std::string n(name, length);
1144 if (this->symbol_definitions_.find(n) == this->symbol_definitions_.end())
1145 this->symbol_references_.insert(n);
1149 // Add an assertion.
1152 Script_options::add_assertion(Expression* check, const char* message,
1155 if (this->script_sections_.in_sections_clause())
1156 this->script_sections_.add_assertion(check, message, messagelen);
1159 Script_assertion* p = new Script_assertion(check, message, messagelen);
1160 this->assertions_.push_back(p);
1164 // Create sections required by any linker scripts.
1167 Script_options::create_script_sections(Layout* layout)
1169 if (this->saw_sections_clause())
1170 this->script_sections_.create_sections(layout);
1173 // Add any symbols we are defining to the symbol table.
1176 Script_options::add_symbols_to_table(Symbol_table* symtab)
1178 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin();
1179 p != this->symbol_assignments_.end();
1181 (*p)->add_to_table(symtab);
1182 this->script_sections_.add_symbols_to_table(symtab);
1185 // Finalize symbol values. Also check assertions.
1188 Script_options::finalize_symbols(Symbol_table* symtab, const Layout* layout)
1190 // We finalize the symbols defined in SECTIONS first, because they
1191 // are the ones which may have changed. This way if symbol outside
1192 // SECTIONS are defined in terms of symbols inside SECTIONS, they
1193 // will get the right value.
1194 this->script_sections_.finalize_symbols(symtab, layout);
1196 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin();
1197 p != this->symbol_assignments_.end();
1199 (*p)->finalize(symtab, layout);
1201 for (Assertions::iterator p = this->assertions_.begin();
1202 p != this->assertions_.end();
1204 (*p)->check(symtab, layout);
1207 // Set section addresses. We set all the symbols which have absolute
1208 // values. Then we let the SECTIONS clause do its thing. This
1209 // returns the segment which holds the file header and segment
1213 Script_options::set_section_addresses(Symbol_table* symtab, Layout* layout)
1215 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin();
1216 p != this->symbol_assignments_.end();
1218 (*p)->set_if_absolute(symtab, layout, false, 0);
1220 return this->script_sections_.set_section_addresses(symtab, layout);
1223 // This class holds data passed through the parser to the lexer and to
1224 // the parser support functions. This avoids global variables. We
1225 // can't use global variables because we need not be called by a
1226 // singleton thread.
1228 class Parser_closure
1231 Parser_closure(const char* filename,
1232 const Position_dependent_options& posdep_options,
1233 bool parsing_defsym, bool in_group, bool is_in_sysroot,
1234 Command_line* command_line,
1235 Script_options* script_options,
1237 bool skip_on_incompatible_target,
1238 Script_info* script_info)
1239 : filename_(filename), posdep_options_(posdep_options),
1240 parsing_defsym_(parsing_defsym), in_group_(in_group),
1241 is_in_sysroot_(is_in_sysroot),
1242 skip_on_incompatible_target_(skip_on_incompatible_target),
1243 found_incompatible_target_(false),
1244 command_line_(command_line), script_options_(script_options),
1245 version_script_info_(script_options->version_script_info()),
1246 lex_(lex), lineno_(0), charpos_(0), lex_mode_stack_(), inputs_(NULL),
1247 script_info_(script_info)
1249 // We start out processing C symbols in the default lex mode.
1250 this->language_stack_.push_back(Version_script_info::LANGUAGE_C);
1251 this->lex_mode_stack_.push_back(lex->mode());
1254 // Return the file name.
1257 { return this->filename_; }
1259 // Return the position dependent options. The caller may modify
1261 Position_dependent_options&
1262 position_dependent_options()
1263 { return this->posdep_options_; }
1265 // Whether we are parsing a --defsym.
1267 parsing_defsym() const
1268 { return this->parsing_defsym_; }
1270 // Return whether this script is being run in a group.
1273 { return this->in_group_; }
1275 // Return whether this script was found using a directory in the
1278 is_in_sysroot() const
1279 { return this->is_in_sysroot_; }
1281 // Whether to skip to the next file with the same name if we find an
1282 // incompatible target in an OUTPUT_FORMAT statement.
1284 skip_on_incompatible_target() const
1285 { return this->skip_on_incompatible_target_; }
1287 // Stop skipping to the next file on an incompatible target. This
1288 // is called when we make some unrevocable change to the data
1291 clear_skip_on_incompatible_target()
1292 { this->skip_on_incompatible_target_ = false; }
1294 // Whether we found an incompatible target in an OUTPUT_FORMAT
1297 found_incompatible_target() const
1298 { return this->found_incompatible_target_; }
1300 // Note that we found an incompatible target.
1302 set_found_incompatible_target()
1303 { this->found_incompatible_target_ = true; }
1305 // Returns the Command_line structure passed in at constructor time.
1306 // This value may be NULL. The caller may modify this, which modifies
1307 // the passed-in Command_line object (not a copy).
1310 { return this->command_line_; }
1312 // Return the options which may be set by a script.
1315 { return this->script_options_; }
1317 // Return the object in which version script information should be stored.
1318 Version_script_info*
1320 { return this->version_script_info_; }
1322 // Return the next token, and advance.
1326 const Token* token = this->lex_->next_token();
1327 this->lineno_ = token->lineno();
1328 this->charpos_ = token->charpos();
1332 // Set a new lexer mode, pushing the current one.
1334 push_lex_mode(Lex::Mode mode)
1336 this->lex_mode_stack_.push_back(this->lex_->mode());
1337 this->lex_->set_mode(mode);
1340 // Pop the lexer mode.
1344 gold_assert(!this->lex_mode_stack_.empty());
1345 this->lex_->set_mode(this->lex_mode_stack_.back());
1346 this->lex_mode_stack_.pop_back();
1349 // Return the current lexer mode.
1352 { return this->lex_mode_stack_.back(); }
1354 // Return the line number of the last token.
1357 { return this->lineno_; }
1359 // Return the character position in the line of the last token.
1362 { return this->charpos_; }
1364 // Return the list of input files, creating it if necessary. This
1365 // is a space leak--we never free the INPUTS_ pointer.
1369 if (this->inputs_ == NULL)
1370 this->inputs_ = new Input_arguments();
1371 return this->inputs_;
1374 // Return whether we saw any input files.
1377 { return this->inputs_ != NULL && !this->inputs_->empty(); }
1379 // Return the current language being processed in a version script
1380 // (eg, "C++"). The empty string represents unmangled C names.
1381 Version_script_info::Language
1382 get_current_language() const
1383 { return this->language_stack_.back(); }
1385 // Push a language onto the stack when entering an extern block.
1387 push_language(Version_script_info::Language lang)
1388 { this->language_stack_.push_back(lang); }
1390 // Pop a language off of the stack when exiting an extern block.
1394 gold_assert(!this->language_stack_.empty());
1395 this->language_stack_.pop_back();
1398 // Return a pointer to the incremental info.
1401 { return this->script_info_; }
1404 // The name of the file we are reading.
1405 const char* filename_;
1406 // The position dependent options.
1407 Position_dependent_options posdep_options_;
1408 // True if we are parsing a --defsym.
1409 bool parsing_defsym_;
1410 // Whether we are currently in a --start-group/--end-group.
1412 // Whether the script was found in a sysrooted directory.
1413 bool is_in_sysroot_;
1414 // If this is true, then if we find an OUTPUT_FORMAT with an
1415 // incompatible target, then we tell the parser to abort so that we
1416 // can search for the next file with the same name.
1417 bool skip_on_incompatible_target_;
1418 // True if we found an OUTPUT_FORMAT with an incompatible target.
1419 bool found_incompatible_target_;
1420 // May be NULL if the user chooses not to pass one in.
1421 Command_line* command_line_;
1422 // Options which may be set from any linker script.
1423 Script_options* script_options_;
1424 // Information parsed from a version script.
1425 Version_script_info* version_script_info_;
1428 // The line number of the last token returned by next_token.
1430 // The column number of the last token returned by next_token.
1432 // A stack of lexer modes.
1433 std::vector<Lex::Mode> lex_mode_stack_;
1434 // A stack of which extern/language block we're inside. Can be C++,
1435 // java, or empty for C.
1436 std::vector<Version_script_info::Language> language_stack_;
1437 // New input files found to add to the link.
1438 Input_arguments* inputs_;
1439 // Pointer to incremental linking info.
1440 Script_info* script_info_;
1443 // FILE was found as an argument on the command line. Try to read it
1444 // as a script. Return true if the file was handled.
1447 read_input_script(Workqueue* workqueue, Symbol_table* symtab, Layout* layout,
1448 Dirsearch* dirsearch, int dirindex,
1449 Input_objects* input_objects, Mapfile* mapfile,
1450 Input_group* input_group,
1451 const Input_argument* input_argument,
1452 Input_file* input_file, Task_token* next_blocker,
1453 bool* used_next_blocker)
1455 *used_next_blocker = false;
1457 std::string input_string;
1458 Lex::read_file(input_file, &input_string);
1460 Lex lex(input_string.c_str(), input_string.length(), PARSING_LINKER_SCRIPT);
1462 Script_info* script_info = NULL;
1463 if (layout->incremental_inputs() != NULL)
1465 const std::string& filename = input_file->filename();
1466 Timespec mtime = input_file->file().get_mtime();
1467 unsigned int arg_serial = input_argument->file().arg_serial();
1468 script_info = new Script_info(filename);
1469 layout->incremental_inputs()->report_script(script_info, arg_serial,
1473 Parser_closure closure(input_file->filename().c_str(),
1474 input_argument->file().options(),
1476 input_group != NULL,
1477 input_file->is_in_sysroot(),
1479 layout->script_options(),
1481 input_file->will_search_for(),
1484 bool old_saw_sections_clause =
1485 layout->script_options()->saw_sections_clause();
1487 if (yyparse(&closure) != 0)
1489 if (closure.found_incompatible_target())
1491 Read_symbols::incompatible_warning(input_argument, input_file);
1492 Read_symbols::requeue(workqueue, input_objects, symtab, layout,
1493 dirsearch, dirindex, mapfile, input_argument,
1494 input_group, next_blocker);
1500 if (!old_saw_sections_clause
1501 && layout->script_options()->saw_sections_clause()
1502 && layout->have_added_input_section())
1503 gold_error(_("%s: SECTIONS seen after other input files; try -T/--script"),
1504 input_file->filename().c_str());
1506 if (!closure.saw_inputs())
1509 Task_token* this_blocker = NULL;
1510 for (Input_arguments::const_iterator p = closure.inputs()->begin();
1511 p != closure.inputs()->end();
1515 if (p + 1 == closure.inputs()->end())
1519 nb = new Task_token(true);
1522 workqueue->queue_soon(new Read_symbols(input_objects, symtab,
1523 layout, dirsearch, 0, mapfile, &*p,
1524 input_group, NULL, this_blocker, nb));
1528 *used_next_blocker = true;
1533 // Helper function for read_version_script() and
1534 // read_commandline_script(). Processes the given file in the mode
1535 // indicated by first_token and lex_mode.
1538 read_script_file(const char* filename, Command_line* cmdline,
1539 Script_options* script_options,
1540 int first_token, Lex::Mode lex_mode)
1542 // TODO: if filename is a relative filename, search for it manually
1543 // using "." + cmdline->options()->search_path() -- not dirsearch.
1544 Dirsearch dirsearch;
1546 // The file locking code wants to record a Task, but we haven't
1547 // started the workqueue yet. This is only for debugging purposes,
1548 // so we invent a fake value.
1549 const Task* task = reinterpret_cast<const Task*>(-1);
1551 // We don't want this file to be opened in binary mode.
1552 Position_dependent_options posdep = cmdline->position_dependent_options();
1553 if (posdep.format_enum() == General_options::OBJECT_FORMAT_BINARY)
1554 posdep.set_format_enum(General_options::OBJECT_FORMAT_ELF);
1555 Input_file_argument input_argument(filename,
1556 Input_file_argument::INPUT_FILE_TYPE_FILE,
1558 Input_file input_file(&input_argument);
1560 if (!input_file.open(dirsearch, task, &dummy))
1563 std::string input_string;
1564 Lex::read_file(&input_file, &input_string);
1566 Lex lex(input_string.c_str(), input_string.length(), first_token);
1567 lex.set_mode(lex_mode);
1569 Parser_closure closure(filename,
1570 cmdline->position_dependent_options(),
1571 first_token == Lex::DYNAMIC_LIST,
1573 input_file.is_in_sysroot(),
1579 if (yyparse(&closure) != 0)
1581 input_file.file().unlock(task);
1585 input_file.file().unlock(task);
1587 gold_assert(!closure.saw_inputs());
1592 // FILENAME was found as an argument to --script (-T).
1593 // Read it as a script, and execute its contents immediately.
1596 read_commandline_script(const char* filename, Command_line* cmdline)
1598 return read_script_file(filename, cmdline, &cmdline->script_options(),
1599 PARSING_LINKER_SCRIPT, Lex::LINKER_SCRIPT);
1602 // FILENAME was found as an argument to --version-script. Read it as
1603 // a version script, and store its contents in
1604 // cmdline->script_options()->version_script_info().
1607 read_version_script(const char* filename, Command_line* cmdline)
1609 return read_script_file(filename, cmdline, &cmdline->script_options(),
1610 PARSING_VERSION_SCRIPT, Lex::VERSION_SCRIPT);
1613 // FILENAME was found as an argument to --dynamic-list. Read it as a
1614 // list of symbols, and store its contents in DYNAMIC_LIST.
1617 read_dynamic_list(const char* filename, Command_line* cmdline,
1618 Script_options* dynamic_list)
1620 return read_script_file(filename, cmdline, dynamic_list,
1621 PARSING_DYNAMIC_LIST, Lex::DYNAMIC_LIST);
1624 // Implement the --defsym option on the command line. Return true if
1628 Script_options::define_symbol(const char* definition)
1630 Lex lex(definition, strlen(definition), PARSING_DEFSYM);
1631 lex.set_mode(Lex::EXPRESSION);
1634 Position_dependent_options posdep_options;
1636 Parser_closure closure("command line", posdep_options, true,
1637 false, false, NULL, this, &lex, false, NULL);
1639 if (yyparse(&closure) != 0)
1642 gold_assert(!closure.saw_inputs());
1647 // Print the script to F for debugging.
1650 Script_options::print(FILE* f) const
1652 fprintf(f, "%s: Dumping linker script\n", program_name);
1654 if (!this->entry_.empty())
1655 fprintf(f, "ENTRY(%s)\n", this->entry_.c_str());
1657 for (Symbol_assignments::const_iterator p =
1658 this->symbol_assignments_.begin();
1659 p != this->symbol_assignments_.end();
1663 for (Assertions::const_iterator p = this->assertions_.begin();
1664 p != this->assertions_.end();
1668 this->script_sections_.print(f);
1670 this->version_script_info_.print(f);
1673 // Manage mapping from keywords to the codes expected by the bison
1674 // parser. We construct one global object for each lex mode with
1677 class Keyword_to_parsecode
1680 // The structure which maps keywords to parsecodes.
1681 struct Keyword_parsecode
1684 const char* keyword;
1685 // Corresponding parsecode.
1689 Keyword_to_parsecode(const Keyword_parsecode* keywords,
1691 : keyword_parsecodes_(keywords), keyword_count_(keyword_count)
1694 // Return the parsecode corresponding KEYWORD, or 0 if it is not a
1697 keyword_to_parsecode(const char* keyword, size_t len) const;
1700 const Keyword_parsecode* keyword_parsecodes_;
1701 const int keyword_count_;
1704 // Mapping from keyword string to keyword parsecode. This array must
1705 // be kept in sorted order. Parsecodes are looked up using bsearch.
1706 // This array must correspond to the list of parsecodes in yyscript.y.
1708 static const Keyword_to_parsecode::Keyword_parsecode
1709 script_keyword_parsecodes[] =
1711 { "ABSOLUTE", ABSOLUTE },
1713 { "ALIGN", ALIGN_K },
1714 { "ALIGNOF", ALIGNOF },
1715 { "ASSERT", ASSERT_K },
1716 { "AS_NEEDED", AS_NEEDED },
1721 { "CONSTANT", CONSTANT },
1722 { "CONSTRUCTORS", CONSTRUCTORS },
1724 { "CREATE_OBJECT_SYMBOLS", CREATE_OBJECT_SYMBOLS },
1725 { "DATA_SEGMENT_ALIGN", DATA_SEGMENT_ALIGN },
1726 { "DATA_SEGMENT_END", DATA_SEGMENT_END },
1727 { "DATA_SEGMENT_RELRO_END", DATA_SEGMENT_RELRO_END },
1728 { "DEFINED", DEFINED },
1731 { "EXCLUDE_FILE", EXCLUDE_FILE },
1732 { "EXTERN", EXTERN },
1735 { "FORCE_COMMON_ALLOCATION", FORCE_COMMON_ALLOCATION },
1738 { "INCLUDE", INCLUDE },
1740 { "INHIBIT_COMMON_ALLOCATION", INHIBIT_COMMON_ALLOCATION },
1743 { "LENGTH", LENGTH },
1744 { "LOADADDR", LOADADDR },
1748 { "MEMORY", MEMORY },
1751 { "NOCROSSREFS", NOCROSSREFS },
1752 { "NOFLOAT", NOFLOAT },
1753 { "NOLOAD", NOLOAD },
1754 { "ONLY_IF_RO", ONLY_IF_RO },
1755 { "ONLY_IF_RW", ONLY_IF_RW },
1756 { "OPTION", OPTION },
1757 { "ORIGIN", ORIGIN },
1758 { "OUTPUT", OUTPUT },
1759 { "OUTPUT_ARCH", OUTPUT_ARCH },
1760 { "OUTPUT_FORMAT", OUTPUT_FORMAT },
1761 { "OVERLAY", OVERLAY },
1763 { "PROVIDE", PROVIDE },
1764 { "PROVIDE_HIDDEN", PROVIDE_HIDDEN },
1766 { "SEARCH_DIR", SEARCH_DIR },
1767 { "SECTIONS", SECTIONS },
1768 { "SEGMENT_START", SEGMENT_START },
1770 { "SIZEOF", SIZEOF },
1771 { "SIZEOF_HEADERS", SIZEOF_HEADERS },
1772 { "SORT", SORT_BY_NAME },
1773 { "SORT_BY_ALIGNMENT", SORT_BY_ALIGNMENT },
1774 { "SORT_BY_NAME", SORT_BY_NAME },
1775 { "SPECIAL", SPECIAL },
1777 { "STARTUP", STARTUP },
1778 { "SUBALIGN", SUBALIGN },
1779 { "SYSLIB", SYSLIB },
1780 { "TARGET", TARGET_K },
1781 { "TRUNCATE", TRUNCATE },
1782 { "VERSION", VERSIONK },
1783 { "global", GLOBAL },
1789 { "sizeof_headers", SIZEOF_HEADERS },
1792 static const Keyword_to_parsecode
1793 script_keywords(&script_keyword_parsecodes[0],
1794 (sizeof(script_keyword_parsecodes)
1795 / sizeof(script_keyword_parsecodes[0])));
1797 static const Keyword_to_parsecode::Keyword_parsecode
1798 version_script_keyword_parsecodes[] =
1800 { "extern", EXTERN },
1801 { "global", GLOBAL },
1805 static const Keyword_to_parsecode
1806 version_script_keywords(&version_script_keyword_parsecodes[0],
1807 (sizeof(version_script_keyword_parsecodes)
1808 / sizeof(version_script_keyword_parsecodes[0])));
1810 static const Keyword_to_parsecode::Keyword_parsecode
1811 dynamic_list_keyword_parsecodes[] =
1813 { "extern", EXTERN },
1816 static const Keyword_to_parsecode
1817 dynamic_list_keywords(&dynamic_list_keyword_parsecodes[0],
1818 (sizeof(dynamic_list_keyword_parsecodes)
1819 / sizeof(dynamic_list_keyword_parsecodes[0])));
1823 // Comparison function passed to bsearch.
1835 ktt_compare(const void* keyv, const void* kttv)
1837 const Ktt_key* key = static_cast<const Ktt_key*>(keyv);
1838 const Keyword_to_parsecode::Keyword_parsecode* ktt =
1839 static_cast<const Keyword_to_parsecode::Keyword_parsecode*>(kttv);
1840 int i = strncmp(key->str, ktt->keyword, key->len);
1843 if (ktt->keyword[key->len] != '\0')
1848 } // End extern "C".
1851 Keyword_to_parsecode::keyword_to_parsecode(const char* keyword,
1857 void* kttv = bsearch(&key,
1858 this->keyword_parsecodes_,
1859 this->keyword_count_,
1860 sizeof(this->keyword_parsecodes_[0]),
1864 Keyword_parsecode* ktt = static_cast<Keyword_parsecode*>(kttv);
1865 return ktt->parsecode;
1868 // The following structs are used within the VersionInfo class as well
1869 // as in the bison helper functions. They store the information
1870 // parsed from the version script.
1872 // A single version expression.
1873 // For example, pattern="std::map*" and language="C++".
1874 struct Version_expression
1876 Version_expression(const std::string& a_pattern,
1877 Version_script_info::Language a_language,
1879 : pattern(a_pattern), language(a_language), exact_match(a_exact_match),
1880 was_matched_by_symbol(false)
1883 std::string pattern;
1884 Version_script_info::Language language;
1885 // If false, we use glob() to match pattern. If true, we use strcmp().
1887 // True if --no-undefined-version is in effect and we found this
1888 // version in get_symbol_version. We use mutable because this
1889 // struct is generally not modifiable after it has been created.
1890 mutable bool was_matched_by_symbol;
1893 // A list of expressions.
1894 struct Version_expression_list
1896 std::vector<struct Version_expression> expressions;
1899 // A list of which versions upon which another version depends.
1900 // Strings should be from the Stringpool.
1901 struct Version_dependency_list
1903 std::vector<std::string> dependencies;
1906 // The total definition of a version. It includes the tag for the
1907 // version, its global and local expressions, and any dependencies.
1911 : tag(), global(NULL), local(NULL), dependencies(NULL)
1915 const struct Version_expression_list* global;
1916 const struct Version_expression_list* local;
1917 const struct Version_dependency_list* dependencies;
1920 // Helper class that calls cplus_demangle when needed and takes care of freeing
1923 class Lazy_demangler
1926 Lazy_demangler(const char* symbol, int options)
1927 : symbol_(symbol), options_(options), demangled_(NULL), did_demangle_(false)
1931 { free(this->demangled_); }
1933 // Return the demangled name. The actual demangling happens on the first call,
1934 // and the result is later cached.
1939 // The symbol to demangle.
1940 const char* symbol_;
1941 // Option flags to pass to cplus_demagle.
1943 // The cached demangled value, or NULL if demangling didn't happen yet or
1946 // Whether we already called cplus_demangle
1950 // Return the demangled name. The actual demangling happens on the first call,
1951 // and the result is later cached. Returns NULL if the symbol cannot be
1955 Lazy_demangler::get()
1957 if (!this->did_demangle_)
1959 this->demangled_ = cplus_demangle(this->symbol_, this->options_);
1960 this->did_demangle_ = true;
1962 return this->demangled_;
1965 // Class Version_script_info.
1967 Version_script_info::Version_script_info()
1968 : dependency_lists_(), expression_lists_(), version_trees_(), globs_(),
1969 default_version_(NULL), default_is_global_(false), is_finalized_(false)
1971 for (int i = 0; i < LANGUAGE_COUNT; ++i)
1972 this->exact_[i] = NULL;
1975 Version_script_info::~Version_script_info()
1979 // Forget all the known version script information.
1982 Version_script_info::clear()
1984 for (size_t k = 0; k < this->dependency_lists_.size(); ++k)
1985 delete this->dependency_lists_[k];
1986 this->dependency_lists_.clear();
1987 for (size_t k = 0; k < this->version_trees_.size(); ++k)
1988 delete this->version_trees_[k];
1989 this->version_trees_.clear();
1990 for (size_t k = 0; k < this->expression_lists_.size(); ++k)
1991 delete this->expression_lists_[k];
1992 this->expression_lists_.clear();
1995 // Finalize the version script information.
1998 Version_script_info::finalize()
2000 if (!this->is_finalized_)
2002 this->build_lookup_tables();
2003 this->is_finalized_ = true;
2007 // Return all the versions.
2009 std::vector<std::string>
2010 Version_script_info::get_versions() const
2012 std::vector<std::string> ret;
2013 for (size_t j = 0; j < this->version_trees_.size(); ++j)
2014 if (!this->version_trees_[j]->tag.empty())
2015 ret.push_back(this->version_trees_[j]->tag);
2019 // Return the dependencies of VERSION.
2021 std::vector<std::string>
2022 Version_script_info::get_dependencies(const char* version) const
2024 std::vector<std::string> ret;
2025 for (size_t j = 0; j < this->version_trees_.size(); ++j)
2026 if (this->version_trees_[j]->tag == version)
2028 const struct Version_dependency_list* deps =
2029 this->version_trees_[j]->dependencies;
2031 for (size_t k = 0; k < deps->dependencies.size(); ++k)
2032 ret.push_back(deps->dependencies[k]);
2038 // A version script essentially maps a symbol name to a version tag
2039 // and an indication of whether symbol is global or local within that
2040 // version tag. Each symbol maps to at most one version tag.
2041 // Unfortunately, in practice, version scripts are ambiguous, and list
2042 // symbols multiple times. Thus, we have to document the matching
2045 // This is a description of what the GNU linker does as of 2010-01-11.
2046 // It walks through the version tags in the order in which they appear
2047 // in the version script. For each tag, it first walks through the
2048 // global patterns for that tag, then the local patterns. When
2049 // looking at a single pattern, it first applies any language specific
2050 // demangling as specified for the pattern, and then matches the
2051 // resulting symbol name to the pattern. If it finds an exact match
2052 // for a literal pattern (a pattern enclosed in quotes or with no
2053 // wildcard characters), then that is the match that it uses. If
2054 // finds a match with a wildcard pattern, then it saves it and
2055 // continues searching. Wildcard patterns that are exactly "*" are
2056 // saved separately.
2058 // If no exact match with a literal pattern is ever found, then if a
2059 // wildcard match with a global pattern was found it is used,
2060 // otherwise if a wildcard match with a local pattern was found it is
2063 // This is the result:
2064 // * If there is an exact match, then we use the first tag in the
2065 // version script where it matches.
2066 // + If the exact match in that tag is global, it is used.
2067 // + Otherwise the exact match in that tag is local, and is used.
2068 // * Otherwise, if there is any match with a global wildcard pattern:
2069 // + If there is any match with a wildcard pattern which is not
2070 // "*", then we use the tag in which the *last* such pattern
2072 // + Otherwise, we matched "*". If there is no match with a local
2073 // wildcard pattern which is not "*", then we use the *last*
2074 // match with a global "*". Otherwise, continue.
2075 // * Otherwise, if there is any match with a local wildcard pattern:
2076 // + If there is any match with a wildcard pattern which is not
2077 // "*", then we use the tag in which the *last* such pattern
2079 // + Otherwise, we matched "*", and we use the tag in which the
2080 // *last* such match occurred.
2082 // There is an additional wrinkle. When the GNU linker finds a symbol
2083 // with a version defined in an object file due to a .symver
2084 // directive, it looks up that symbol name in that version tag. If it
2085 // finds it, it matches the symbol name against the patterns for that
2086 // version. If there is no match with a global pattern, but there is
2087 // a match with a local pattern, then the GNU linker marks the symbol
2090 // We want gold to be generally compatible, but we also want gold to
2091 // be fast. These are the rules that gold implements:
2092 // * If there is an exact match for the mangled name, we use it.
2093 // + If there is more than one exact match, we give a warning, and
2094 // we use the first tag in the script which matches.
2095 // + If a symbol has an exact match as both global and local for
2096 // the same version tag, we give an error.
2097 // * Otherwise, we look for an extern C++ or an extern Java exact
2098 // match. If we find an exact match, we use it.
2099 // + If there is more than one exact match, we give a warning, and
2100 // we use the first tag in the script which matches.
2101 // + If a symbol has an exact match as both global and local for
2102 // the same version tag, we give an error.
2103 // * Otherwise, we look through the wildcard patterns, ignoring "*"
2104 // patterns. We look through the version tags in reverse order.
2105 // For each version tag, we look through the global patterns and
2106 // then the local patterns. We use the first match we find (i.e.,
2107 // the last matching version tag in the file).
2108 // * Otherwise, we use the "*" pattern if there is one. We give an
2109 // error if there are multiple "*" patterns.
2111 // At least for now, gold does not look up the version tag for a
2112 // symbol version found in an object file to see if it should be
2113 // forced local. There are other ways to force a symbol to be local,
2114 // and I don't understand why this one is useful.
2116 // Build a set of fast lookup tables for a version script.
2119 Version_script_info::build_lookup_tables()
2121 size_t size = this->version_trees_.size();
2122 for (size_t j = 0; j < size; ++j)
2124 const Version_tree* v = this->version_trees_[j];
2125 this->build_expression_list_lookup(v->local, v, false);
2126 this->build_expression_list_lookup(v->global, v, true);
2130 // If a pattern has backlashes but no unquoted wildcard characters,
2131 // then we apply backslash unquoting and look for an exact match.
2132 // Otherwise we treat it as a wildcard pattern. This function returns
2133 // true for a wildcard pattern. Otherwise, it does backslash
2134 // unquoting on *PATTERN and returns false. If this returns true,
2135 // *PATTERN may have been partially unquoted.
2138 Version_script_info::unquote(std::string* pattern) const
2140 bool saw_backslash = false;
2141 size_t len = pattern->length();
2143 for (size_t i = 0; i < len; ++i)
2146 saw_backslash = false;
2149 switch ((*pattern)[i])
2151 case '?': case '[': case '*':
2154 saw_backslash = true;
2162 (*pattern)[j] = (*pattern)[i];
2168 // Add an exact match for MATCH to *PE. The result of the match is
2172 Version_script_info::add_exact_match(const std::string& match,
2173 const Version_tree* v, bool is_global,
2174 const Version_expression* ve,
2177 std::pair<Exact::iterator, bool> ins =
2178 pe->insert(std::make_pair(match, Version_tree_match(v, is_global, ve)));
2181 // This is the first time we have seen this match.
2185 Version_tree_match& vtm(ins.first->second);
2186 if (vtm.real->tag != v->tag)
2188 // This is an ambiguous match. We still return the
2189 // first version that we found in the script, but we
2190 // record the new version to issue a warning if we
2191 // wind up looking up this symbol.
2192 if (vtm.ambiguous == NULL)
2195 else if (is_global != vtm.is_global)
2197 // We have a match for both the global and local entries for a
2198 // version tag. That's got to be wrong.
2199 gold_error(_("'%s' appears as both a global and a local symbol "
2200 "for version '%s' in script"),
2201 match.c_str(), v->tag.c_str());
2205 // Build fast lookup information for EXPLIST and store it in LOOKUP.
2206 // All matches go to V, and IS_GLOBAL is true if they are global
2210 Version_script_info::build_expression_list_lookup(
2211 const Version_expression_list* explist,
2212 const Version_tree* v,
2215 if (explist == NULL)
2217 size_t size = explist->expressions.size();
2218 for (size_t i = 0; i < size; ++i)
2220 const Version_expression& exp(explist->expressions[i]);
2222 if (exp.pattern.length() == 1 && exp.pattern[0] == '*')
2224 if (this->default_version_ != NULL
2225 && this->default_version_->tag != v->tag)
2226 gold_warning(_("wildcard match appears in both version '%s' "
2227 "and '%s' in script"),
2228 this->default_version_->tag.c_str(), v->tag.c_str());
2229 else if (this->default_version_ != NULL
2230 && this->default_is_global_ != is_global)
2231 gold_error(_("wildcard match appears as both global and local "
2232 "in version '%s' in script"),
2234 this->default_version_ = v;
2235 this->default_is_global_ = is_global;
2239 std::string pattern = exp.pattern;
2240 if (!exp.exact_match)
2242 if (this->unquote(&pattern))
2244 this->globs_.push_back(Glob(&exp, v, is_global));
2249 if (this->exact_[exp.language] == NULL)
2250 this->exact_[exp.language] = new Exact();
2251 this->add_exact_match(pattern, v, is_global, &exp,
2252 this->exact_[exp.language]);
2256 // Return the name to match given a name, a language code, and two
2260 Version_script_info::get_name_to_match(const char* name,
2262 Lazy_demangler* cpp_demangler,
2263 Lazy_demangler* java_demangler) const
2270 return cpp_demangler->get();
2272 return java_demangler->get();
2278 // Look up SYMBOL_NAME in the list of versions. Return true if the
2279 // symbol is found, false if not. If the symbol is found, then if
2280 // PVERSION is not NULL, set *PVERSION to the version tag, and if
2281 // P_IS_GLOBAL is not NULL, set *P_IS_GLOBAL according to whether the
2282 // symbol is global or not.
2285 Version_script_info::get_symbol_version(const char* symbol_name,
2286 std::string* pversion,
2287 bool* p_is_global) const
2289 Lazy_demangler cpp_demangled_name(symbol_name, DMGL_ANSI | DMGL_PARAMS);
2290 Lazy_demangler java_demangled_name(symbol_name,
2291 DMGL_ANSI | DMGL_PARAMS | DMGL_JAVA);
2293 gold_assert(this->is_finalized_);
2294 for (int i = 0; i < LANGUAGE_COUNT; ++i)
2296 Exact* exact = this->exact_[i];
2300 const char* name_to_match = this->get_name_to_match(symbol_name, i,
2301 &cpp_demangled_name,
2302 &java_demangled_name);
2303 if (name_to_match == NULL)
2305 // If the name can not be demangled, the GNU linker goes
2306 // ahead and tries to match it anyhow. That does not
2307 // make sense to me and I have not implemented it.
2311 Exact::const_iterator pe = exact->find(name_to_match);
2312 if (pe != exact->end())
2314 const Version_tree_match& vtm(pe->second);
2315 if (vtm.ambiguous != NULL)
2316 gold_warning(_("using '%s' as version for '%s' which is also "
2317 "named in version '%s' in script"),
2318 vtm.real->tag.c_str(), name_to_match,
2319 vtm.ambiguous->tag.c_str());
2321 if (pversion != NULL)
2322 *pversion = vtm.real->tag;
2323 if (p_is_global != NULL)
2324 *p_is_global = vtm.is_global;
2326 // If we are using --no-undefined-version, and this is a
2327 // global symbol, we have to record that we have found this
2328 // symbol, so that we don't warn about it. We have to do
2329 // this now, because otherwise we have no way to get from a
2330 // non-C language back to the demangled name that we
2332 if (p_is_global != NULL && vtm.is_global)
2333 vtm.expression->was_matched_by_symbol = true;
2339 // Look through the glob patterns in reverse order.
2341 for (Globs::const_reverse_iterator p = this->globs_.rbegin();
2342 p != this->globs_.rend();
2345 int language = p->expression->language;
2346 const char* name_to_match = this->get_name_to_match(symbol_name,
2348 &cpp_demangled_name,
2349 &java_demangled_name);
2350 if (name_to_match == NULL)
2353 if (fnmatch(p->expression->pattern.c_str(), name_to_match,
2356 if (pversion != NULL)
2357 *pversion = p->version->tag;
2358 if (p_is_global != NULL)
2359 *p_is_global = p->is_global;
2364 // Finally, there may be a wildcard.
2365 if (this->default_version_ != NULL)
2367 if (pversion != NULL)
2368 *pversion = this->default_version_->tag;
2369 if (p_is_global != NULL)
2370 *p_is_global = this->default_is_global_;
2377 // Give an error if any exact symbol names (not wildcards) appear in a
2378 // version script, but there is no such symbol.
2381 Version_script_info::check_unmatched_names(const Symbol_table* symtab) const
2383 for (size_t i = 0; i < this->version_trees_.size(); ++i)
2385 const Version_tree* vt = this->version_trees_[i];
2386 if (vt->global == NULL)
2388 for (size_t j = 0; j < vt->global->expressions.size(); ++j)
2390 const Version_expression& expression(vt->global->expressions[j]);
2392 // Ignore cases where we used the version because we saw a
2393 // symbol that we looked up. Note that
2394 // WAS_MATCHED_BY_SYMBOL will be true even if the symbol was
2395 // not a definition. That's OK as in that case we most
2396 // likely gave an undefined symbol error anyhow.
2397 if (expression.was_matched_by_symbol)
2400 // Just ignore names which are in languages other than C.
2401 // We have no way to look them up in the symbol table.
2402 if (expression.language != LANGUAGE_C)
2405 // Remove backslash quoting, and ignore wildcard patterns.
2406 std::string pattern = expression.pattern;
2407 if (!expression.exact_match)
2409 if (this->unquote(&pattern))
2413 if (symtab->lookup(pattern.c_str(), vt->tag.c_str()) == NULL)
2414 gold_error(_("version script assignment of %s to symbol %s "
2415 "failed: symbol not defined"),
2416 vt->tag.c_str(), pattern.c_str());
2421 struct Version_dependency_list*
2422 Version_script_info::allocate_dependency_list()
2424 dependency_lists_.push_back(new Version_dependency_list);
2425 return dependency_lists_.back();
2428 struct Version_expression_list*
2429 Version_script_info::allocate_expression_list()
2431 expression_lists_.push_back(new Version_expression_list);
2432 return expression_lists_.back();
2435 struct Version_tree*
2436 Version_script_info::allocate_version_tree()
2438 version_trees_.push_back(new Version_tree);
2439 return version_trees_.back();
2442 // Print for debugging.
2445 Version_script_info::print(FILE* f) const
2450 fprintf(f, "VERSION {");
2452 for (size_t i = 0; i < this->version_trees_.size(); ++i)
2454 const Version_tree* vt = this->version_trees_[i];
2456 if (vt->tag.empty())
2459 fprintf(f, " %s {\n", vt->tag.c_str());
2461 if (vt->global != NULL)
2463 fprintf(f, " global :\n");
2464 this->print_expression_list(f, vt->global);
2467 if (vt->local != NULL)
2469 fprintf(f, " local :\n");
2470 this->print_expression_list(f, vt->local);
2474 if (vt->dependencies != NULL)
2476 const Version_dependency_list* deps = vt->dependencies;
2477 for (size_t j = 0; j < deps->dependencies.size(); ++j)
2479 if (j < deps->dependencies.size() - 1)
2481 fprintf(f, " %s", deps->dependencies[j].c_str());
2491 Version_script_info::print_expression_list(
2493 const Version_expression_list* vel) const
2495 Version_script_info::Language current_language = LANGUAGE_C;
2496 for (size_t i = 0; i < vel->expressions.size(); ++i)
2498 const Version_expression& ve(vel->expressions[i]);
2500 if (ve.language != current_language)
2502 if (current_language != LANGUAGE_C)
2504 switch (ve.language)
2509 fprintf(f, " extern \"C++\" {\n");
2512 fprintf(f, " extern \"Java\" {\n");
2517 current_language = ve.language;
2521 if (current_language != LANGUAGE_C)
2526 fprintf(f, "%s", ve.pattern.c_str());
2533 if (current_language != LANGUAGE_C)
2537 } // End namespace gold.
2539 // The remaining functions are extern "C", so it's clearer to not put
2540 // them in namespace gold.
2542 using namespace gold;
2544 // This function is called by the bison parser to return the next
2548 yylex(YYSTYPE* lvalp, void* closurev)
2550 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2551 const Token* token = closure->next_token();
2552 switch (token->classification())
2557 case Token::TOKEN_INVALID:
2558 yyerror(closurev, "invalid character");
2561 case Token::TOKEN_EOF:
2564 case Token::TOKEN_STRING:
2566 // This is either a keyword or a STRING.
2568 const char* str = token->string_value(&len);
2570 switch (closure->lex_mode())
2572 case Lex::LINKER_SCRIPT:
2573 parsecode = script_keywords.keyword_to_parsecode(str, len);
2575 case Lex::VERSION_SCRIPT:
2576 parsecode = version_script_keywords.keyword_to_parsecode(str, len);
2578 case Lex::DYNAMIC_LIST:
2579 parsecode = dynamic_list_keywords.keyword_to_parsecode(str, len);
2586 lvalp->string.value = str;
2587 lvalp->string.length = len;
2591 case Token::TOKEN_QUOTED_STRING:
2592 lvalp->string.value = token->string_value(&lvalp->string.length);
2593 return QUOTED_STRING;
2595 case Token::TOKEN_OPERATOR:
2596 return token->operator_value();
2598 case Token::TOKEN_INTEGER:
2599 lvalp->integer = token->integer_value();
2604 // This function is called by the bison parser to report an error.
2607 yyerror(void* closurev, const char* message)
2609 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2610 gold_error(_("%s:%d:%d: %s"), closure->filename(), closure->lineno(),
2611 closure->charpos(), message);
2614 // Called by the bison parser to add an external symbol to the link.
2617 script_add_extern(void* closurev, const char* name, size_t length)
2619 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2620 closure->script_options()->add_symbol_reference(name, length);
2623 // Called by the bison parser to add a file to the link.
2626 script_add_file(void* closurev, const char* name, size_t length)
2628 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2630 // If this is an absolute path, and we found the script in the
2631 // sysroot, then we want to prepend the sysroot to the file name.
2632 // For example, this is how we handle a cross link to the x86_64
2633 // libc.so, which refers to /lib/libc.so.6.
2634 std::string name_string(name, length);
2635 const char* extra_search_path = ".";
2636 std::string script_directory;
2637 if (IS_ABSOLUTE_PATH(name_string.c_str()))
2639 if (closure->is_in_sysroot())
2641 const std::string& sysroot(parameters->options().sysroot());
2642 gold_assert(!sysroot.empty());
2643 name_string = sysroot + name_string;
2648 // In addition to checking the normal library search path, we
2649 // also want to check in the script-directory.
2650 const char* slash = strrchr(closure->filename(), '/');
2653 script_directory.assign(closure->filename(),
2654 slash - closure->filename() + 1);
2655 extra_search_path = script_directory.c_str();
2659 Input_file_argument file(name_string.c_str(),
2660 Input_file_argument::INPUT_FILE_TYPE_FILE,
2661 extra_search_path, false,
2662 closure->position_dependent_options());
2663 Input_argument& arg = closure->inputs()->add_file(file);
2664 arg.set_script_info(closure->script_info());
2667 // Called by the bison parser to add a library to the link.
2670 script_add_library(void* closurev, const char* name, size_t length)
2672 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2673 std::string name_string(name, length);
2675 if (name_string[0] != 'l')
2676 gold_error(_("library name must be prefixed with -l"));
2678 Input_file_argument file(name_string.c_str() + 1,
2679 Input_file_argument::INPUT_FILE_TYPE_LIBRARY,
2681 closure->position_dependent_options());
2682 Input_argument& arg = closure->inputs()->add_file(file);
2683 arg.set_script_info(closure->script_info());
2686 // Called by the bison parser to start a group. If we are already in
2687 // a group, that means that this script was invoked within a
2688 // --start-group --end-group sequence on the command line, or that
2689 // this script was found in a GROUP of another script. In that case,
2690 // we simply continue the existing group, rather than starting a new
2691 // one. It is possible to construct a case in which this will do
2692 // something other than what would happen if we did a recursive group,
2693 // but it's hard to imagine why the different behaviour would be
2694 // useful for a real program. Avoiding recursive groups is simpler
2695 // and more efficient.
2698 script_start_group(void* closurev)
2700 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2701 if (!closure->in_group())
2702 closure->inputs()->start_group();
2705 // Called by the bison parser at the end of a group.
2708 script_end_group(void* closurev)
2710 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2711 if (!closure->in_group())
2712 closure->inputs()->end_group();
2715 // Called by the bison parser to start an AS_NEEDED list.
2718 script_start_as_needed(void* closurev)
2720 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2721 closure->position_dependent_options().set_as_needed(true);
2724 // Called by the bison parser at the end of an AS_NEEDED list.
2727 script_end_as_needed(void* closurev)
2729 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2730 closure->position_dependent_options().set_as_needed(false);
2733 // Called by the bison parser to set the entry symbol.
2736 script_set_entry(void* closurev, const char* entry, size_t length)
2738 // We'll parse this exactly the same as --entry=ENTRY on the commandline
2739 // TODO(csilvers): FIXME -- call set_entry directly.
2740 std::string arg("--entry=");
2741 arg.append(entry, length);
2742 script_parse_option(closurev, arg.c_str(), arg.size());
2745 // Called by the bison parser to set whether to define common symbols.
2748 script_set_common_allocation(void* closurev, int set)
2750 const char* arg = set != 0 ? "--define-common" : "--no-define-common";
2751 script_parse_option(closurev, arg, strlen(arg));
2754 // Called by the bison parser to refer to a symbol.
2756 extern "C" Expression*
2757 script_symbol(void* closurev, const char* name, size_t length)
2759 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2760 if (length != 1 || name[0] != '.')
2761 closure->script_options()->add_symbol_reference(name, length);
2762 return script_exp_string(name, length);
2765 // Called by the bison parser to define a symbol.
2768 script_set_symbol(void* closurev, const char* name, size_t length,
2769 Expression* value, int providei, int hiddeni)
2771 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2772 const bool provide = providei != 0;
2773 const bool hidden = hiddeni != 0;
2774 closure->script_options()->add_symbol_assignment(name, length,
2775 closure->parsing_defsym(),
2776 value, provide, hidden);
2777 closure->clear_skip_on_incompatible_target();
2780 // Called by the bison parser to add an assertion.
2783 script_add_assertion(void* closurev, Expression* check, const char* message,
2786 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2787 closure->script_options()->add_assertion(check, message, messagelen);
2788 closure->clear_skip_on_incompatible_target();
2791 // Called by the bison parser to parse an OPTION.
2794 script_parse_option(void* closurev, const char* option, size_t length)
2796 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2797 // We treat the option as a single command-line option, even if
2798 // it has internal whitespace.
2799 if (closure->command_line() == NULL)
2801 // There are some options that we could handle here--e.g.,
2802 // -lLIBRARY. Should we bother?
2803 gold_warning(_("%s:%d:%d: ignoring command OPTION; OPTION is only valid"
2804 " for scripts specified via -T/--script"),
2805 closure->filename(), closure->lineno(), closure->charpos());
2809 bool past_a_double_dash_option = false;
2810 const char* mutable_option = strndup(option, length);
2811 gold_assert(mutable_option != NULL);
2812 closure->command_line()->process_one_option(1, &mutable_option, 0,
2813 &past_a_double_dash_option);
2814 // The General_options class will quite possibly store a pointer
2815 // into mutable_option, so we can't free it. In cases the class
2816 // does not store such a pointer, this is a memory leak. Alas. :(
2818 closure->clear_skip_on_incompatible_target();
2821 // Called by the bison parser to handle OUTPUT_FORMAT. OUTPUT_FORMAT
2822 // takes either one or three arguments. In the three argument case,
2823 // the format depends on the endianness option, which we don't
2824 // currently support (FIXME). If we see an OUTPUT_FORMAT for the
2825 // wrong format, then we want to search for a new file. Returning 0
2826 // here will cause the parser to immediately abort.
2829 script_check_output_format(void* closurev,
2830 const char* default_name, size_t default_length,
2831 const char*, size_t, const char*, size_t)
2833 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2834 std::string name(default_name, default_length);
2835 Target* target = select_target_by_bfd_name(name.c_str());
2836 if (target == NULL || !parameters->is_compatible_target(target))
2838 if (closure->skip_on_incompatible_target())
2840 closure->set_found_incompatible_target();
2843 // FIXME: Should we warn about the unknown target?
2848 // Called by the bison parser to handle TARGET.
2851 script_set_target(void* closurev, const char* target, size_t len)
2853 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2854 std::string s(target, len);
2855 General_options::Object_format format_enum;
2856 format_enum = General_options::string_to_object_format(s.c_str());
2857 closure->position_dependent_options().set_format_enum(format_enum);
2860 // Called by the bison parser to handle SEARCH_DIR. This is handled
2861 // exactly like a -L option.
2864 script_add_search_dir(void* closurev, const char* option, size_t length)
2866 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2867 if (closure->command_line() == NULL)
2868 gold_warning(_("%s:%d:%d: ignoring SEARCH_DIR; SEARCH_DIR is only valid"
2869 " for scripts specified via -T/--script"),
2870 closure->filename(), closure->lineno(), closure->charpos());
2871 else if (!closure->command_line()->options().nostdlib())
2873 std::string s = "-L" + std::string(option, length);
2874 script_parse_option(closurev, s.c_str(), s.size());
2878 /* Called by the bison parser to push the lexer into expression
2882 script_push_lex_into_expression_mode(void* closurev)
2884 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2885 closure->push_lex_mode(Lex::EXPRESSION);
2888 /* Called by the bison parser to push the lexer into version
2892 script_push_lex_into_version_mode(void* closurev)
2894 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2895 if (closure->version_script()->is_finalized())
2896 gold_error(_("%s:%d:%d: invalid use of VERSION in input file"),
2897 closure->filename(), closure->lineno(), closure->charpos());
2898 closure->push_lex_mode(Lex::VERSION_SCRIPT);
2901 /* Called by the bison parser to pop the lexer mode. */
2904 script_pop_lex_mode(void* closurev)
2906 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2907 closure->pop_lex_mode();
2910 // Register an entire version node. For example:
2916 // - tag is "GLIBC_2.1"
2917 // - tree contains the information "global: foo"
2918 // - deps contains "GLIBC_2.0"
2921 script_register_vers_node(void*,
2924 struct Version_tree* tree,
2925 struct Version_dependency_list* deps)
2927 gold_assert(tree != NULL);
2928 tree->dependencies = deps;
2930 tree->tag = std::string(tag, taglen);
2933 // Add a dependencies to the list of existing dependencies, if any,
2934 // and return the expanded list.
2936 extern "C" struct Version_dependency_list*
2937 script_add_vers_depend(void* closurev,
2938 struct Version_dependency_list* all_deps,
2939 const char* depend_to_add, int deplen)
2941 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2942 if (all_deps == NULL)
2943 all_deps = closure->version_script()->allocate_dependency_list();
2944 all_deps->dependencies.push_back(std::string(depend_to_add, deplen));
2948 // Add a pattern expression to an existing list of expressions, if any.
2950 extern "C" struct Version_expression_list*
2951 script_new_vers_pattern(void* closurev,
2952 struct Version_expression_list* expressions,
2953 const char* pattern, int patlen, int exact_match)
2955 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2956 if (expressions == NULL)
2957 expressions = closure->version_script()->allocate_expression_list();
2958 expressions->expressions.push_back(
2959 Version_expression(std::string(pattern, patlen),
2960 closure->get_current_language(),
2961 static_cast<bool>(exact_match)));
2965 // Attaches b to the end of a, and clears b. So a = a + b and b = {}.
2967 extern "C" struct Version_expression_list*
2968 script_merge_expressions(struct Version_expression_list* a,
2969 struct Version_expression_list* b)
2971 a->expressions.insert(a->expressions.end(),
2972 b->expressions.begin(), b->expressions.end());
2973 // We could delete b and remove it from expressions_lists_, but
2974 // that's a lot of work. This works just as well.
2975 b->expressions.clear();
2979 // Combine the global and local expressions into a a Version_tree.
2981 extern "C" struct Version_tree*
2982 script_new_vers_node(void* closurev,
2983 struct Version_expression_list* global,
2984 struct Version_expression_list* local)
2986 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2987 Version_tree* tree = closure->version_script()->allocate_version_tree();
2988 tree->global = global;
2989 tree->local = local;
2993 // Handle a transition in language, such as at the
2994 // start or end of 'extern "C++"'
2997 version_script_push_lang(void* closurev, const char* lang, int langlen)
2999 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3000 std::string language(lang, langlen);
3001 Version_script_info::Language code;
3002 if (language.empty() || language == "C")
3003 code = Version_script_info::LANGUAGE_C;
3004 else if (language == "C++")
3005 code = Version_script_info::LANGUAGE_CXX;
3006 else if (language == "Java")
3007 code = Version_script_info::LANGUAGE_JAVA;
3010 char* buf = new char[langlen + 100];
3011 snprintf(buf, langlen + 100,
3012 _("unrecognized version script language '%s'"),
3014 yyerror(closurev, buf);
3016 code = Version_script_info::LANGUAGE_C;
3018 closure->push_language(code);
3022 version_script_pop_lang(void* closurev)
3024 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3025 closure->pop_language();
3028 // Called by the bison parser to start a SECTIONS clause.
3031 script_start_sections(void* closurev)
3033 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3034 closure->script_options()->script_sections()->start_sections();
3035 closure->clear_skip_on_incompatible_target();
3038 // Called by the bison parser to finish a SECTIONS clause.
3041 script_finish_sections(void* closurev)
3043 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3044 closure->script_options()->script_sections()->finish_sections();
3047 // Start processing entries for an output section.
3050 script_start_output_section(void* closurev, const char* name, size_t namelen,
3051 const struct Parser_output_section_header* header)
3053 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3054 closure->script_options()->script_sections()->start_output_section(name,
3059 // Finish processing entries for an output section.
3062 script_finish_output_section(void* closurev,
3063 const struct Parser_output_section_trailer* trail)
3065 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3066 closure->script_options()->script_sections()->finish_output_section(trail);
3069 // Add a data item (e.g., "WORD (0)") to the current output section.
3072 script_add_data(void* closurev, int data_token, Expression* val)
3074 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3076 bool is_signed = true;
3098 closure->script_options()->script_sections()->add_data(size, is_signed, val);
3101 // Add a clause setting the fill value to the current output section.
3104 script_add_fill(void* closurev, Expression* val)
3106 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3107 closure->script_options()->script_sections()->add_fill(val);
3110 // Add a new input section specification to the current output
3114 script_add_input_section(void* closurev,
3115 const struct Input_section_spec* spec,
3118 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3119 bool keep = keepi != 0;
3120 closure->script_options()->script_sections()->add_input_section(spec, keep);
3123 // When we see DATA_SEGMENT_ALIGN we record that following output
3124 // sections may be relro.
3127 script_data_segment_align(void* closurev)
3129 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3130 if (!closure->script_options()->saw_sections_clause())
3131 gold_error(_("%s:%d:%d: DATA_SEGMENT_ALIGN not in SECTIONS clause"),
3132 closure->filename(), closure->lineno(), closure->charpos());
3134 closure->script_options()->script_sections()->data_segment_align();
3137 // When we see DATA_SEGMENT_RELRO_END we know that all output sections
3138 // since DATA_SEGMENT_ALIGN should be relro.
3141 script_data_segment_relro_end(void* closurev)
3143 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3144 if (!closure->script_options()->saw_sections_clause())
3145 gold_error(_("%s:%d:%d: DATA_SEGMENT_ALIGN not in SECTIONS clause"),
3146 closure->filename(), closure->lineno(), closure->charpos());
3148 closure->script_options()->script_sections()->data_segment_relro_end();
3151 // Create a new list of string/sort pairs.
3153 extern "C" String_sort_list_ptr
3154 script_new_string_sort_list(const struct Wildcard_section* string_sort)
3156 return new String_sort_list(1, *string_sort);
3159 // Add an entry to a list of string/sort pairs. The way the parser
3160 // works permits us to simply modify the first parameter, rather than
3163 extern "C" String_sort_list_ptr
3164 script_string_sort_list_add(String_sort_list_ptr pv,
3165 const struct Wildcard_section* string_sort)
3168 return script_new_string_sort_list(string_sort);
3171 pv->push_back(*string_sort);
3176 // Create a new list of strings.
3178 extern "C" String_list_ptr
3179 script_new_string_list(const char* str, size_t len)
3181 return new String_list(1, std::string(str, len));
3184 // Add an element to a list of strings. The way the parser works
3185 // permits us to simply modify the first parameter, rather than copy
3188 extern "C" String_list_ptr
3189 script_string_list_push_back(String_list_ptr pv, const char* str, size_t len)
3192 return script_new_string_list(str, len);
3195 pv->push_back(std::string(str, len));
3200 // Concatenate two string lists. Either or both may be NULL. The way
3201 // the parser works permits us to modify the parameters, rather than
3204 extern "C" String_list_ptr
3205 script_string_list_append(String_list_ptr pv1, String_list_ptr pv2)
3211 pv1->insert(pv1->end(), pv2->begin(), pv2->end());
3215 // Add a new program header.
3218 script_add_phdr(void* closurev, const char* name, size_t namelen,
3219 unsigned int type, const Phdr_info* info)
3221 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3222 bool includes_filehdr = info->includes_filehdr != 0;
3223 bool includes_phdrs = info->includes_phdrs != 0;
3224 bool is_flags_valid = info->is_flags_valid != 0;
3225 Script_sections* ss = closure->script_options()->script_sections();
3226 ss->add_phdr(name, namelen, type, includes_filehdr, includes_phdrs,
3227 is_flags_valid, info->flags, info->load_address);
3228 closure->clear_skip_on_incompatible_target();
3231 // Convert a program header string to a type.
3233 #define PHDR_TYPE(NAME) { #NAME, sizeof(#NAME) - 1, elfcpp::NAME }
3240 } phdr_type_names[] =
3244 PHDR_TYPE(PT_DYNAMIC),
3245 PHDR_TYPE(PT_INTERP),
3247 PHDR_TYPE(PT_SHLIB),
3250 PHDR_TYPE(PT_GNU_EH_FRAME),
3251 PHDR_TYPE(PT_GNU_STACK),
3252 PHDR_TYPE(PT_GNU_RELRO)
3255 extern "C" unsigned int
3256 script_phdr_string_to_type(void* closurev, const char* name, size_t namelen)
3258 for (unsigned int i = 0;
3259 i < sizeof(phdr_type_names) / sizeof(phdr_type_names[0]);
3261 if (namelen == phdr_type_names[i].namelen
3262 && strncmp(name, phdr_type_names[i].name, namelen) == 0)
3263 return phdr_type_names[i].val;
3264 yyerror(closurev, _("unknown PHDR type (try integer)"));
3265 return elfcpp::PT_NULL;
3269 script_saw_segment_start_expression(void* closurev)
3271 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3272 Script_sections* ss = closure->script_options()->script_sections();
3273 ss->set_saw_segment_start_expression(true);
3277 script_set_section_region(void* closurev, const char* name, size_t namelen,
3280 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3281 if (!closure->script_options()->saw_sections_clause())
3283 gold_error(_("%s:%d:%d: MEMORY region '%.*s' referred to outside of "
3285 closure->filename(), closure->lineno(), closure->charpos(),
3286 static_cast<int>(namelen), name);
3290 Script_sections* ss = closure->script_options()->script_sections();
3291 Memory_region* mr = ss->find_memory_region(name, namelen);
3294 gold_error(_("%s:%d:%d: MEMORY region '%.*s' not declared"),
3295 closure->filename(), closure->lineno(), closure->charpos(),
3296 static_cast<int>(namelen), name);
3300 ss->set_memory_region(mr, set_vma);
3304 script_add_memory(void* closurev, const char* name, size_t namelen,
3305 unsigned int attrs, Expression* origin, Expression* length)
3307 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3308 Script_sections* ss = closure->script_options()->script_sections();
3309 ss->add_memory_region(name, namelen, attrs, origin, length);
3312 extern "C" unsigned int
3313 script_parse_memory_attr(void* closurev, const char* attrs, size_t attrlen,
3323 attributes |= MEM_READABLE; break;
3326 attributes |= MEM_READABLE | MEM_WRITEABLE; break;
3329 attributes |= MEM_EXECUTABLE; break;
3332 attributes |= MEM_ALLOCATABLE; break;
3337 attributes |= MEM_INITIALIZED; break;
3339 yyerror(closurev, _("unknown MEMORY attribute"));
3343 attributes = (~ attributes) & MEM_ATTR_MASK;
3349 script_include_directive(void* closurev, const char*, size_t)
3351 // FIXME: Implement ?
3352 yyerror (closurev, _("GOLD does not currently support INCLUDE directives"));
3355 // Functions for memory regions.
3357 extern "C" Expression*
3358 script_exp_function_origin(void* closurev, const char* name, size_t namelen)
3360 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3361 Script_sections* ss = closure->script_options()->script_sections();
3362 Expression* origin = ss->find_memory_region_origin(name, namelen);
3366 gold_error(_("undefined memory region '%s' referenced "
3367 "in ORIGIN expression"),
3369 // Create a dummy expression to prevent crashes later on.
3370 origin = script_exp_integer(0);
3376 extern "C" Expression*
3377 script_exp_function_length(void* closurev, const char* name, size_t namelen)
3379 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3380 Script_sections* ss = closure->script_options()->script_sections();
3381 Expression* length = ss->find_memory_region_length(name, namelen);
3385 gold_error(_("undefined memory region '%s' referenced "
3386 "in LENGTH expression"),
3388 // Create a dummy expression to prevent crashes later on.
3389 length = script_exp_integer(0);