1 // script.cc -- handle linker scripts for gold.
3 // Copyright 2006, 2007, 2008, 2009, 2010 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
151 gold_assert(this->classification_ == TOKEN_INTEGER);
153 std::string s(this->value_, this->value_length_);
154 return strtoull(s.c_str(), NULL, 0);
158 // The token classification.
159 Classification classification_;
160 // The token value, for TOKEN_STRING or TOKEN_QUOTED_STRING or
163 // The length of the token value.
164 size_t value_length_;
165 // The token value, for TOKEN_OPERATOR.
167 // The line number where this token started (one based).
169 // The character position within the line where this token started
174 // This class handles lexing a file into a sequence of tokens.
179 // We unfortunately have to support different lexing modes, because
180 // when reading different parts of a linker script we need to parse
181 // things differently.
184 // Reading an ordinary linker script.
186 // Reading an expression in a linker script.
188 // Reading a version script.
190 // Reading a --dynamic-list file.
194 Lex(const char* input_string, size_t input_length, int parsing_token)
195 : input_string_(input_string), input_length_(input_length),
196 current_(input_string), mode_(LINKER_SCRIPT),
197 first_token_(parsing_token), token_(),
198 lineno_(1), linestart_(input_string)
201 // Read a file into a string.
203 read_file(Input_file*, std::string*);
205 // Return the next token.
209 // Return the current lexing mode.
212 { return this->mode_; }
214 // Set the lexing mode.
217 { this->mode_ = mode; }
221 Lex& operator=(const Lex&);
223 // Make a general token with no value at the current location.
225 make_token(Token::Classification c, const char* start) const
226 { return Token(c, this->lineno_, start - this->linestart_ + 1); }
228 // Make a general token with a value at the current location.
230 make_token(Token::Classification c, const char* v, size_t len,
233 { return Token(c, v, len, this->lineno_, start - this->linestart_ + 1); }
235 // Make an operator token at the current location.
237 make_token(int opcode, const char* start) const
238 { return Token(opcode, this->lineno_, start - this->linestart_ + 1); }
240 // Make an invalid token at the current location.
242 make_invalid_token(const char* start)
243 { return this->make_token(Token::TOKEN_INVALID, start); }
245 // Make an EOF token at the current location.
247 make_eof_token(const char* start)
248 { return this->make_token(Token::TOKEN_EOF, start); }
250 // Return whether C can be the first character in a name. C2 is the
251 // next character, since we sometimes need that.
253 can_start_name(char c, char c2);
255 // If C can appear in a name which has already started, return a
256 // pointer to a character later in the token or just past
257 // it. Otherwise, return NULL.
259 can_continue_name(const char* c);
261 // Return whether C, C2, C3 can start a hex number.
263 can_start_hex(char c, char c2, char c3);
265 // If C can appear in a hex number which has already started, return
266 // a pointer to a character later in the token or just past
267 // it. Otherwise, return NULL.
269 can_continue_hex(const char* c);
271 // Return whether C can start a non-hex number.
273 can_start_number(char c);
275 // If C can appear in a decimal number which has already started,
276 // return a pointer to a character later in the token or just past
277 // it. Otherwise, return NULL.
279 can_continue_number(const char* c)
280 { return Lex::can_start_number(*c) ? c + 1 : NULL; }
282 // If C1 C2 C3 form a valid three character operator, return the
283 // opcode. Otherwise return 0.
285 three_char_operator(char c1, char c2, char c3);
287 // If C1 C2 form a valid two character operator, return the opcode.
288 // Otherwise return 0.
290 two_char_operator(char c1, char c2);
292 // If C1 is a valid one character operator, return the opcode.
293 // Otherwise return 0.
295 one_char_operator(char c1);
297 // Read the next token.
299 get_token(const char**);
301 // Skip a C style /* */ comment. Return false if the comment did
304 skip_c_comment(const char**);
306 // Skip a line # comment. Return false if there was no newline.
308 skip_line_comment(const char**);
310 // Build a token CLASSIFICATION from all characters that match
311 // CAN_CONTINUE_FN. The token starts at START. Start matching from
312 // MATCH. Set *PP to the character following the token.
314 gather_token(Token::Classification,
315 const char* (Lex::*can_continue_fn)(const char*),
316 const char* start, const char* match, const char** pp);
318 // Build a token from a quoted string.
320 gather_quoted_string(const char** pp);
322 // The string we are tokenizing.
323 const char* input_string_;
324 // The length of the string.
325 size_t input_length_;
326 // The current offset into the string.
327 const char* current_;
328 // The current lexing mode.
330 // The code to use for the first token. This is set to 0 after it
333 // The current token.
335 // The current line number.
337 // The start of the current line in the string.
338 const char* linestart_;
341 // Read the whole file into memory. We don't expect linker scripts to
342 // be large, so we just use a std::string as a buffer. We ignore the
343 // data we've already read, so that we read aligned buffers.
346 Lex::read_file(Input_file* input_file, std::string* contents)
348 off_t filesize = input_file->file().filesize();
350 contents->reserve(filesize);
353 unsigned char buf[BUFSIZ];
354 while (off < filesize)
357 if (get > filesize - off)
358 get = filesize - off;
359 input_file->file().read(off, get, buf);
360 contents->append(reinterpret_cast<char*>(&buf[0]), get);
365 // Return whether C can be the start of a name, if the next character
366 // is C2. A name can being with a letter, underscore, period, or
367 // dollar sign. Because a name can be a file name, we also permit
368 // forward slash, backslash, and tilde. Tilde is the tricky case
369 // here; GNU ld also uses it as a bitwise not operator. It is only
370 // recognized as the operator if it is not immediately followed by
371 // some character which can appear in a symbol. That is, when we
372 // don't know that we are looking at an expression, "~0" is a file
373 // name, and "~ 0" is an expression using bitwise not. We are
377 Lex::can_start_name(char c, char c2)
381 case 'A': case 'B': case 'C': case 'D': case 'E': case 'F':
382 case 'G': case 'H': case 'I': case 'J': case 'K': case 'L':
383 case 'M': case 'N': case 'O': case 'Q': case 'P': case 'R':
384 case 'S': case 'T': case 'U': case 'V': case 'W': case 'X':
386 case 'a': case 'b': case 'c': case 'd': case 'e': case 'f':
387 case 'g': case 'h': case 'i': case 'j': case 'k': case 'l':
388 case 'm': case 'n': case 'o': case 'q': case 'p': case 'r':
389 case 's': case 't': case 'u': case 'v': case 'w': case 'x':
391 case '_': case '.': case '$':
395 return this->mode_ == LINKER_SCRIPT;
398 return this->mode_ == LINKER_SCRIPT && can_continue_name(&c2);
401 return (this->mode_ == VERSION_SCRIPT
402 || this->mode_ == DYNAMIC_LIST
403 || (this->mode_ == LINKER_SCRIPT
404 && can_continue_name(&c2)));
411 // Return whether C can continue a name which has already started.
412 // Subsequent characters in a name are the same as the leading
413 // characters, plus digits and "=+-:[],?*". So in general the linker
414 // script language requires spaces around operators, unless we know
415 // that we are parsing an expression.
418 Lex::can_continue_name(const char* c)
422 case 'A': case 'B': case 'C': case 'D': case 'E': case 'F':
423 case 'G': case 'H': case 'I': case 'J': case 'K': case 'L':
424 case 'M': case 'N': case 'O': case 'Q': case 'P': case 'R':
425 case 'S': case 'T': case 'U': case 'V': case 'W': case 'X':
427 case 'a': case 'b': case 'c': case 'd': case 'e': case 'f':
428 case 'g': case 'h': case 'i': case 'j': case 'k': case 'l':
429 case 'm': case 'n': case 'o': case 'q': case 'p': case 'r':
430 case 's': case 't': case 'u': case 'v': case 'w': case 'x':
432 case '_': case '.': case '$':
433 case '0': case '1': case '2': case '3': case '4':
434 case '5': case '6': case '7': case '8': case '9':
437 // TODO(csilvers): why not allow ~ in names for version-scripts?
438 case '/': case '\\': case '~':
441 if (this->mode_ == LINKER_SCRIPT)
445 case '[': case ']': case '*': case '?': case '-':
446 if (this->mode_ == LINKER_SCRIPT || this->mode_ == VERSION_SCRIPT
447 || this->mode_ == DYNAMIC_LIST)
451 // TODO(csilvers): why allow this? ^ is meaningless in version scripts.
453 if (this->mode_ == VERSION_SCRIPT || this->mode_ == DYNAMIC_LIST)
458 if (this->mode_ == LINKER_SCRIPT)
460 else if ((this->mode_ == VERSION_SCRIPT || this->mode_ == DYNAMIC_LIST)
463 // A name can have '::' in it, as that's a c++ namespace
464 // separator. But a single colon is not part of a name.
474 // For a number we accept 0x followed by hex digits, or any sequence
475 // of digits. The old linker accepts leading '$' for hex, and
476 // trailing HXBOD. Those are for MRI compatibility and we don't
477 // accept them. The old linker also accepts trailing MK for mega or
478 // kilo. FIXME: Those are mentioned in the documentation, and we
479 // should accept them.
481 // Return whether C1 C2 C3 can start a hex number.
484 Lex::can_start_hex(char c1, char c2, char c3)
486 if (c1 == '0' && (c2 == 'x' || c2 == 'X'))
487 return this->can_continue_hex(&c3);
491 // Return whether C can appear in a hex number.
494 Lex::can_continue_hex(const char* c)
498 case '0': case '1': case '2': case '3': case '4':
499 case '5': case '6': case '7': case '8': case '9':
500 case 'A': case 'B': case 'C': case 'D': case 'E': case 'F':
501 case 'a': case 'b': case 'c': case 'd': case 'e': case 'f':
509 // Return whether C can start a non-hex number.
512 Lex::can_start_number(char c)
516 case '0': case '1': case '2': case '3': case '4':
517 case '5': case '6': case '7': case '8': case '9':
525 // If C1 C2 C3 form a valid three character operator, return the
526 // opcode (defined in the yyscript.h file generated from yyscript.y).
527 // Otherwise return 0.
530 Lex::three_char_operator(char c1, char c2, char c3)
535 if (c2 == '<' && c3 == '=')
539 if (c2 == '>' && c3 == '=')
548 // If C1 C2 form a valid two character operator, return the opcode
549 // (defined in the yyscript.h file generated from yyscript.y).
550 // Otherwise return 0.
553 Lex::two_char_operator(char c1, char c2)
611 // If C1 is a valid operator, return the opcode. Otherwise return 0.
614 Lex::one_char_operator(char c1)
647 // Skip a C style comment. *PP points to just after the "/*". Return
648 // false if the comment did not end.
651 Lex::skip_c_comment(const char** pp)
654 while (p[0] != '*' || p[1] != '/')
665 this->linestart_ = p + 1;
674 // Skip a line # comment. Return false if there was no newline.
677 Lex::skip_line_comment(const char** pp)
680 size_t skip = strcspn(p, "\n");
689 this->linestart_ = p;
695 // Build a token CLASSIFICATION from all characters that match
696 // CAN_CONTINUE_FN. Update *PP.
699 Lex::gather_token(Token::Classification classification,
700 const char* (Lex::*can_continue_fn)(const char*),
705 const char* new_match = NULL;
706 while ((new_match = (this->*can_continue_fn)(match)))
709 return this->make_token(classification, start, match - start, start);
712 // Build a token from a quoted string.
715 Lex::gather_quoted_string(const char** pp)
717 const char* start = *pp;
718 const char* p = start;
720 size_t skip = strcspn(p, "\"\n");
722 return this->make_invalid_token(start);
724 return this->make_token(Token::TOKEN_QUOTED_STRING, p, skip, start);
727 // Return the next token at *PP. Update *PP. General guideline: we
728 // require linker scripts to be simple ASCII. No unicode linker
729 // scripts. In particular we can assume that any '\0' is the end of
733 Lex::get_token(const char** pp)
742 return this->make_eof_token(p);
745 // Skip whitespace quickly.
746 while (*p == ' ' || *p == '\t' || *p == '\r')
753 this->linestart_ = p;
757 // Skip C style comments.
758 if (p[0] == '/' && p[1] == '*')
760 int lineno = this->lineno_;
761 int charpos = p - this->linestart_ + 1;
764 if (!this->skip_c_comment(pp))
765 return Token(Token::TOKEN_INVALID, lineno, charpos);
771 // Skip line comments.
775 if (!this->skip_line_comment(pp))
776 return this->make_eof_token(p);
782 if (this->can_start_name(p[0], p[1]))
783 return this->gather_token(Token::TOKEN_STRING,
784 &Lex::can_continue_name,
787 // We accept any arbitrary name in double quotes, as long as it
788 // does not cross a line boundary.
792 return this->gather_quoted_string(pp);
795 // Check for a number.
797 if (this->can_start_hex(p[0], p[1], p[2]))
798 return this->gather_token(Token::TOKEN_INTEGER,
799 &Lex::can_continue_hex,
802 if (Lex::can_start_number(p[0]))
803 return this->gather_token(Token::TOKEN_INTEGER,
804 &Lex::can_continue_number,
807 // Check for operators.
809 int opcode = Lex::three_char_operator(p[0], p[1], p[2]);
813 return this->make_token(opcode, p);
816 opcode = Lex::two_char_operator(p[0], p[1]);
820 return this->make_token(opcode, p);
823 opcode = Lex::one_char_operator(p[0]);
827 return this->make_token(opcode, p);
830 return this->make_token(Token::TOKEN_INVALID, p);
834 // Return the next token.
839 // The first token is special.
840 if (this->first_token_ != 0)
842 this->token_ = Token(this->first_token_, 0, 0);
843 this->first_token_ = 0;
844 return &this->token_;
847 this->token_ = this->get_token(&this->current_);
849 // Don't let an early null byte fool us into thinking that we've
850 // reached the end of the file.
851 if (this->token_.is_eof()
852 && (static_cast<size_t>(this->current_ - this->input_string_)
853 < this->input_length_))
854 this->token_ = this->make_invalid_token(this->current_);
856 return &this->token_;
859 // class Symbol_assignment.
861 // Add the symbol to the symbol table. This makes sure the symbol is
862 // there and defined. The actual value is stored later. We can't
863 // determine the actual value at this point, because we can't
864 // necessarily evaluate the expression until all ordinary symbols have
867 // The GNU linker lets symbol assignments in the linker script
868 // silently override defined symbols in object files. We are
869 // compatible. FIXME: Should we issue a warning?
872 Symbol_assignment::add_to_table(Symbol_table* symtab)
874 elfcpp::STV vis = this->hidden_ ? elfcpp::STV_HIDDEN : elfcpp::STV_DEFAULT;
875 this->sym_ = symtab->define_as_constant(this->name_.c_str(),
878 ? Symbol_table::DEFSYM
879 : Symbol_table::SCRIPT),
887 true); // force_override
890 // Finalize a symbol value.
893 Symbol_assignment::finalize(Symbol_table* symtab, const Layout* layout)
895 this->finalize_maybe_dot(symtab, layout, false, 0, NULL);
898 // Finalize a symbol value which can refer to the dot symbol.
901 Symbol_assignment::finalize_with_dot(Symbol_table* symtab,
902 const Layout* layout,
904 Output_section* dot_section)
906 this->finalize_maybe_dot(symtab, layout, true, dot_value, dot_section);
909 // Finalize a symbol value, internal version.
912 Symbol_assignment::finalize_maybe_dot(Symbol_table* symtab,
913 const Layout* layout,
914 bool is_dot_available,
916 Output_section* dot_section)
918 // If we were only supposed to provide this symbol, the sym_ field
919 // will be NULL if the symbol was not referenced.
920 if (this->sym_ == NULL)
922 gold_assert(this->provide_);
926 if (parameters->target().get_size() == 32)
928 #if defined(HAVE_TARGET_32_LITTLE) || defined(HAVE_TARGET_32_BIG)
929 this->sized_finalize<32>(symtab, layout, is_dot_available, dot_value,
935 else if (parameters->target().get_size() == 64)
937 #if defined(HAVE_TARGET_64_LITTLE) || defined(HAVE_TARGET_64_BIG)
938 this->sized_finalize<64>(symtab, layout, is_dot_available, dot_value,
950 Symbol_assignment::sized_finalize(Symbol_table* symtab, const Layout* layout,
951 bool is_dot_available, uint64_t dot_value,
952 Output_section* dot_section)
954 Output_section* section;
955 uint64_t final_val = this->val_->eval_maybe_dot(symtab, layout, true,
957 dot_value, dot_section,
959 Sized_symbol<size>* ssym = symtab->get_sized_symbol<size>(this->sym_);
960 ssym->set_value(final_val);
962 ssym->set_output_section(section);
965 // Set the symbol value if the expression yields an absolute value.
968 Symbol_assignment::set_if_absolute(Symbol_table* symtab, const Layout* layout,
969 bool is_dot_available, uint64_t dot_value)
971 if (this->sym_ == NULL)
974 Output_section* val_section;
975 uint64_t val = this->val_->eval_maybe_dot(symtab, layout, false,
976 is_dot_available, dot_value,
977 NULL, &val_section, NULL);
978 if (val_section != NULL)
981 if (parameters->target().get_size() == 32)
983 #if defined(HAVE_TARGET_32_LITTLE) || defined(HAVE_TARGET_32_BIG)
984 Sized_symbol<32>* ssym = symtab->get_sized_symbol<32>(this->sym_);
985 ssym->set_value(val);
990 else if (parameters->target().get_size() == 64)
992 #if defined(HAVE_TARGET_64_LITTLE) || defined(HAVE_TARGET_64_BIG)
993 Sized_symbol<64>* ssym = symtab->get_sized_symbol<64>(this->sym_);
994 ssym->set_value(val);
1003 // Print for debugging.
1006 Symbol_assignment::print(FILE* f) const
1008 if (this->provide_ && this->hidden_)
1009 fprintf(f, "PROVIDE_HIDDEN(");
1010 else if (this->provide_)
1011 fprintf(f, "PROVIDE(");
1012 else if (this->hidden_)
1015 fprintf(f, "%s = ", this->name_.c_str());
1016 this->val_->print(f);
1018 if (this->provide_ || this->hidden_)
1024 // Class Script_assertion.
1026 // Check the assertion.
1029 Script_assertion::check(const Symbol_table* symtab, const Layout* layout)
1031 if (!this->check_->eval(symtab, layout, true))
1032 gold_error("%s", this->message_.c_str());
1035 // Print for debugging.
1038 Script_assertion::print(FILE* f) const
1040 fprintf(f, "ASSERT(");
1041 this->check_->print(f);
1042 fprintf(f, ", \"%s\")\n", this->message_.c_str());
1045 // Class Script_options.
1047 Script_options::Script_options()
1048 : entry_(), symbol_assignments_(), symbol_definitions_(),
1049 symbol_references_(), version_script_info_(), script_sections_()
1053 // Returns true if NAME is on the list of symbol assignments waiting
1057 Script_options::is_pending_assignment(const char* name)
1059 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin();
1060 p != this->symbol_assignments_.end();
1062 if ((*p)->name() == name)
1067 // Add a symbol to be defined.
1070 Script_options::add_symbol_assignment(const char* name, size_t length,
1071 bool is_defsym, Expression* value,
1072 bool provide, bool hidden)
1074 if (length != 1 || name[0] != '.')
1076 if (this->script_sections_.in_sections_clause())
1078 gold_assert(!is_defsym);
1079 this->script_sections_.add_symbol_assignment(name, length, value,
1084 Symbol_assignment* p = new Symbol_assignment(name, length, is_defsym,
1085 value, provide, hidden);
1086 this->symbol_assignments_.push_back(p);
1091 std::string n(name, length);
1092 this->symbol_definitions_.insert(n);
1093 this->symbol_references_.erase(n);
1098 if (provide || hidden)
1099 gold_error(_("invalid use of PROVIDE for dot symbol"));
1101 // The GNU linker permits assignments to dot outside of SECTIONS
1102 // clauses and treats them as occurring inside, so we don't
1103 // check in_sections_clause here.
1104 this->script_sections_.add_dot_assignment(value);
1108 // Add a reference to a symbol.
1111 Script_options::add_symbol_reference(const char* name, size_t length)
1113 if (length != 1 || name[0] != '.')
1115 std::string n(name, length);
1116 if (this->symbol_definitions_.find(n) == this->symbol_definitions_.end())
1117 this->symbol_references_.insert(n);
1121 // Add an assertion.
1124 Script_options::add_assertion(Expression* check, const char* message,
1127 if (this->script_sections_.in_sections_clause())
1128 this->script_sections_.add_assertion(check, message, messagelen);
1131 Script_assertion* p = new Script_assertion(check, message, messagelen);
1132 this->assertions_.push_back(p);
1136 // Create sections required by any linker scripts.
1139 Script_options::create_script_sections(Layout* layout)
1141 if (this->saw_sections_clause())
1142 this->script_sections_.create_sections(layout);
1145 // Add any symbols we are defining to the symbol table.
1148 Script_options::add_symbols_to_table(Symbol_table* symtab)
1150 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin();
1151 p != this->symbol_assignments_.end();
1153 (*p)->add_to_table(symtab);
1154 this->script_sections_.add_symbols_to_table(symtab);
1157 // Finalize symbol values. Also check assertions.
1160 Script_options::finalize_symbols(Symbol_table* symtab, const Layout* layout)
1162 // We finalize the symbols defined in SECTIONS first, because they
1163 // are the ones which may have changed. This way if symbol outside
1164 // SECTIONS are defined in terms of symbols inside SECTIONS, they
1165 // will get the right value.
1166 this->script_sections_.finalize_symbols(symtab, layout);
1168 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin();
1169 p != this->symbol_assignments_.end();
1171 (*p)->finalize(symtab, layout);
1173 for (Assertions::iterator p = this->assertions_.begin();
1174 p != this->assertions_.end();
1176 (*p)->check(symtab, layout);
1179 // Set section addresses. We set all the symbols which have absolute
1180 // values. Then we let the SECTIONS clause do its thing. This
1181 // returns the segment which holds the file header and segment
1185 Script_options::set_section_addresses(Symbol_table* symtab, Layout* layout)
1187 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin();
1188 p != this->symbol_assignments_.end();
1190 (*p)->set_if_absolute(symtab, layout, false, 0);
1192 return this->script_sections_.set_section_addresses(symtab, layout);
1195 // This class holds data passed through the parser to the lexer and to
1196 // the parser support functions. This avoids global variables. We
1197 // can't use global variables because we need not be called by a
1198 // singleton thread.
1200 class Parser_closure
1203 Parser_closure(const char* filename,
1204 const Position_dependent_options& posdep_options,
1205 bool parsing_defsym, bool in_group, bool is_in_sysroot,
1206 Command_line* command_line,
1207 Script_options* script_options,
1209 bool skip_on_incompatible_target)
1210 : filename_(filename), posdep_options_(posdep_options),
1211 parsing_defsym_(parsing_defsym), in_group_(in_group),
1212 is_in_sysroot_(is_in_sysroot),
1213 skip_on_incompatible_target_(skip_on_incompatible_target),
1214 found_incompatible_target_(false),
1215 command_line_(command_line), script_options_(script_options),
1216 version_script_info_(script_options->version_script_info()),
1217 lex_(lex), lineno_(0), charpos_(0), lex_mode_stack_(), inputs_(NULL)
1219 // We start out processing C symbols in the default lex mode.
1220 this->language_stack_.push_back(Version_script_info::LANGUAGE_C);
1221 this->lex_mode_stack_.push_back(lex->mode());
1224 // Return the file name.
1227 { return this->filename_; }
1229 // Return the position dependent options. The caller may modify
1231 Position_dependent_options&
1232 position_dependent_options()
1233 { return this->posdep_options_; }
1235 // Whether we are parsing a --defsym.
1237 parsing_defsym() const
1238 { return this->parsing_defsym_; }
1240 // Return whether this script is being run in a group.
1243 { return this->in_group_; }
1245 // Return whether this script was found using a directory in the
1248 is_in_sysroot() const
1249 { return this->is_in_sysroot_; }
1251 // Whether to skip to the next file with the same name if we find an
1252 // incompatible target in an OUTPUT_FORMAT statement.
1254 skip_on_incompatible_target() const
1255 { return this->skip_on_incompatible_target_; }
1257 // Stop skipping to the next file on an incompatible target. This
1258 // is called when we make some unrevocable change to the data
1261 clear_skip_on_incompatible_target()
1262 { this->skip_on_incompatible_target_ = false; }
1264 // Whether we found an incompatible target in an OUTPUT_FORMAT
1267 found_incompatible_target() const
1268 { return this->found_incompatible_target_; }
1270 // Note that we found an incompatible target.
1272 set_found_incompatible_target()
1273 { this->found_incompatible_target_ = true; }
1275 // Returns the Command_line structure passed in at constructor time.
1276 // This value may be NULL. The caller may modify this, which modifies
1277 // the passed-in Command_line object (not a copy).
1280 { return this->command_line_; }
1282 // Return the options which may be set by a script.
1285 { return this->script_options_; }
1287 // Return the object in which version script information should be stored.
1288 Version_script_info*
1290 { return this->version_script_info_; }
1292 // Return the next token, and advance.
1296 const Token* token = this->lex_->next_token();
1297 this->lineno_ = token->lineno();
1298 this->charpos_ = token->charpos();
1302 // Set a new lexer mode, pushing the current one.
1304 push_lex_mode(Lex::Mode mode)
1306 this->lex_mode_stack_.push_back(this->lex_->mode());
1307 this->lex_->set_mode(mode);
1310 // Pop the lexer mode.
1314 gold_assert(!this->lex_mode_stack_.empty());
1315 this->lex_->set_mode(this->lex_mode_stack_.back());
1316 this->lex_mode_stack_.pop_back();
1319 // Return the current lexer mode.
1322 { return this->lex_mode_stack_.back(); }
1324 // Return the line number of the last token.
1327 { return this->lineno_; }
1329 // Return the character position in the line of the last token.
1332 { return this->charpos_; }
1334 // Return the list of input files, creating it if necessary. This
1335 // is a space leak--we never free the INPUTS_ pointer.
1339 if (this->inputs_ == NULL)
1340 this->inputs_ = new Input_arguments();
1341 return this->inputs_;
1344 // Return whether we saw any input files.
1347 { return this->inputs_ != NULL && !this->inputs_->empty(); }
1349 // Return the current language being processed in a version script
1350 // (eg, "C++"). The empty string represents unmangled C names.
1351 Version_script_info::Language
1352 get_current_language() const
1353 { return this->language_stack_.back(); }
1355 // Push a language onto the stack when entering an extern block.
1357 push_language(Version_script_info::Language lang)
1358 { this->language_stack_.push_back(lang); }
1360 // Pop a language off of the stack when exiting an extern block.
1364 gold_assert(!this->language_stack_.empty());
1365 this->language_stack_.pop_back();
1369 // The name of the file we are reading.
1370 const char* filename_;
1371 // The position dependent options.
1372 Position_dependent_options posdep_options_;
1373 // True if we are parsing a --defsym.
1374 bool parsing_defsym_;
1375 // Whether we are currently in a --start-group/--end-group.
1377 // Whether the script was found in a sysrooted directory.
1378 bool is_in_sysroot_;
1379 // If this is true, then if we find an OUTPUT_FORMAT with an
1380 // incompatible target, then we tell the parser to abort so that we
1381 // can search for the next file with the same name.
1382 bool skip_on_incompatible_target_;
1383 // True if we found an OUTPUT_FORMAT with an incompatible target.
1384 bool found_incompatible_target_;
1385 // May be NULL if the user chooses not to pass one in.
1386 Command_line* command_line_;
1387 // Options which may be set from any linker script.
1388 Script_options* script_options_;
1389 // Information parsed from a version script.
1390 Version_script_info* version_script_info_;
1393 // The line number of the last token returned by next_token.
1395 // The column number of the last token returned by next_token.
1397 // A stack of lexer modes.
1398 std::vector<Lex::Mode> lex_mode_stack_;
1399 // A stack of which extern/language block we're inside. Can be C++,
1400 // java, or empty for C.
1401 std::vector<Version_script_info::Language> language_stack_;
1402 // New input files found to add to the link.
1403 Input_arguments* inputs_;
1406 // FILE was found as an argument on the command line. Try to read it
1407 // as a script. Return true if the file was handled.
1410 read_input_script(Workqueue* workqueue, Symbol_table* symtab, Layout* layout,
1411 Dirsearch* dirsearch, int dirindex,
1412 Input_objects* input_objects, Mapfile* mapfile,
1413 Input_group* input_group,
1414 const Input_argument* input_argument,
1415 Input_file* input_file, Task_token* next_blocker,
1416 bool* used_next_blocker)
1418 *used_next_blocker = false;
1420 std::string input_string;
1421 Lex::read_file(input_file, &input_string);
1423 Lex lex(input_string.c_str(), input_string.length(), PARSING_LINKER_SCRIPT);
1425 Parser_closure closure(input_file->filename().c_str(),
1426 input_argument->file().options(),
1428 input_group != NULL,
1429 input_file->is_in_sysroot(),
1431 layout->script_options(),
1433 input_file->will_search_for());
1435 bool old_saw_sections_clause =
1436 layout->script_options()->saw_sections_clause();
1438 if (yyparse(&closure) != 0)
1440 if (closure.found_incompatible_target())
1442 Read_symbols::incompatible_warning(input_argument, input_file);
1443 Read_symbols::requeue(workqueue, input_objects, symtab, layout,
1444 dirsearch, dirindex, mapfile, input_argument,
1445 input_group, next_blocker);
1451 if (!old_saw_sections_clause
1452 && layout->script_options()->saw_sections_clause()
1453 && layout->have_added_input_section())
1454 gold_error(_("%s: SECTIONS seen after other input files; try -T/--script"),
1455 input_file->filename().c_str());
1457 if (!closure.saw_inputs())
1460 Task_token* this_blocker = NULL;
1461 for (Input_arguments::const_iterator p = closure.inputs()->begin();
1462 p != closure.inputs()->end();
1466 if (p + 1 == closure.inputs()->end())
1470 nb = new Task_token(true);
1473 workqueue->queue_soon(new Read_symbols(input_objects, symtab,
1474 layout, dirsearch, 0, mapfile, &*p,
1475 input_group, NULL, this_blocker, nb));
1479 if (layout->incremental_inputs() != NULL)
1481 // Like new Read_symbols(...) above, we rely on closure.inputs()
1482 // getting leaked by closure.
1483 const std::string& filename = input_file->filename();
1484 Script_info* info = new Script_info(closure.inputs());
1485 Timespec mtime = input_file->file().get_mtime();
1486 layout->incremental_inputs()->report_script(filename, info, mtime);
1489 *used_next_blocker = true;
1494 // Helper function for read_version_script() and
1495 // read_commandline_script(). Processes the given file in the mode
1496 // indicated by first_token and lex_mode.
1499 read_script_file(const char* filename, Command_line* cmdline,
1500 Script_options* script_options,
1501 int first_token, Lex::Mode lex_mode)
1503 // TODO: if filename is a relative filename, search for it manually
1504 // using "." + cmdline->options()->search_path() -- not dirsearch.
1505 Dirsearch dirsearch;
1507 // The file locking code wants to record a Task, but we haven't
1508 // started the workqueue yet. This is only for debugging purposes,
1509 // so we invent a fake value.
1510 const Task* task = reinterpret_cast<const Task*>(-1);
1512 // We don't want this file to be opened in binary mode.
1513 Position_dependent_options posdep = cmdline->position_dependent_options();
1514 if (posdep.format_enum() == General_options::OBJECT_FORMAT_BINARY)
1515 posdep.set_format_enum(General_options::OBJECT_FORMAT_ELF);
1516 Input_file_argument input_argument(filename,
1517 Input_file_argument::INPUT_FILE_TYPE_FILE,
1519 Input_file input_file(&input_argument);
1521 if (!input_file.open(dirsearch, task, &dummy))
1524 std::string input_string;
1525 Lex::read_file(&input_file, &input_string);
1527 Lex lex(input_string.c_str(), input_string.length(), first_token);
1528 lex.set_mode(lex_mode);
1530 Parser_closure closure(filename,
1531 cmdline->position_dependent_options(),
1532 first_token == Lex::DYNAMIC_LIST,
1534 input_file.is_in_sysroot(),
1539 if (yyparse(&closure) != 0)
1541 input_file.file().unlock(task);
1545 input_file.file().unlock(task);
1547 gold_assert(!closure.saw_inputs());
1552 // FILENAME was found as an argument to --script (-T).
1553 // Read it as a script, and execute its contents immediately.
1556 read_commandline_script(const char* filename, Command_line* cmdline)
1558 return read_script_file(filename, cmdline, &cmdline->script_options(),
1559 PARSING_LINKER_SCRIPT, Lex::LINKER_SCRIPT);
1562 // FILENAME was found as an argument to --version-script. Read it as
1563 // a version script, and store its contents in
1564 // cmdline->script_options()->version_script_info().
1567 read_version_script(const char* filename, Command_line* cmdline)
1569 return read_script_file(filename, cmdline, &cmdline->script_options(),
1570 PARSING_VERSION_SCRIPT, Lex::VERSION_SCRIPT);
1573 // FILENAME was found as an argument to --dynamic-list. Read it as a
1574 // list of symbols, and store its contents in DYNAMIC_LIST.
1577 read_dynamic_list(const char* filename, Command_line* cmdline,
1578 Script_options* dynamic_list)
1580 return read_script_file(filename, cmdline, dynamic_list,
1581 PARSING_DYNAMIC_LIST, Lex::DYNAMIC_LIST);
1584 // Implement the --defsym option on the command line. Return true if
1588 Script_options::define_symbol(const char* definition)
1590 Lex lex(definition, strlen(definition), PARSING_DEFSYM);
1591 lex.set_mode(Lex::EXPRESSION);
1594 Position_dependent_options posdep_options;
1596 Parser_closure closure("command line", posdep_options, true,
1597 false, false, NULL, this, &lex, false);
1599 if (yyparse(&closure) != 0)
1602 gold_assert(!closure.saw_inputs());
1607 // Print the script to F for debugging.
1610 Script_options::print(FILE* f) const
1612 fprintf(f, "%s: Dumping linker script\n", program_name);
1614 if (!this->entry_.empty())
1615 fprintf(f, "ENTRY(%s)\n", this->entry_.c_str());
1617 for (Symbol_assignments::const_iterator p =
1618 this->symbol_assignments_.begin();
1619 p != this->symbol_assignments_.end();
1623 for (Assertions::const_iterator p = this->assertions_.begin();
1624 p != this->assertions_.end();
1628 this->script_sections_.print(f);
1630 this->version_script_info_.print(f);
1633 // Manage mapping from keywords to the codes expected by the bison
1634 // parser. We construct one global object for each lex mode with
1637 class Keyword_to_parsecode
1640 // The structure which maps keywords to parsecodes.
1641 struct Keyword_parsecode
1644 const char* keyword;
1645 // Corresponding parsecode.
1649 Keyword_to_parsecode(const Keyword_parsecode* keywords,
1651 : keyword_parsecodes_(keywords), keyword_count_(keyword_count)
1654 // Return the parsecode corresponding KEYWORD, or 0 if it is not a
1657 keyword_to_parsecode(const char* keyword, size_t len) const;
1660 const Keyword_parsecode* keyword_parsecodes_;
1661 const int keyword_count_;
1664 // Mapping from keyword string to keyword parsecode. This array must
1665 // be kept in sorted order. Parsecodes are looked up using bsearch.
1666 // This array must correspond to the list of parsecodes in yyscript.y.
1668 static const Keyword_to_parsecode::Keyword_parsecode
1669 script_keyword_parsecodes[] =
1671 { "ABSOLUTE", ABSOLUTE },
1673 { "ALIGN", ALIGN_K },
1674 { "ALIGNOF", ALIGNOF },
1675 { "ASSERT", ASSERT_K },
1676 { "AS_NEEDED", AS_NEEDED },
1681 { "CONSTANT", CONSTANT },
1682 { "CONSTRUCTORS", CONSTRUCTORS },
1684 { "CREATE_OBJECT_SYMBOLS", CREATE_OBJECT_SYMBOLS },
1685 { "DATA_SEGMENT_ALIGN", DATA_SEGMENT_ALIGN },
1686 { "DATA_SEGMENT_END", DATA_SEGMENT_END },
1687 { "DATA_SEGMENT_RELRO_END", DATA_SEGMENT_RELRO_END },
1688 { "DEFINED", DEFINED },
1691 { "EXCLUDE_FILE", EXCLUDE_FILE },
1692 { "EXTERN", EXTERN },
1695 { "FORCE_COMMON_ALLOCATION", FORCE_COMMON_ALLOCATION },
1698 { "INCLUDE", INCLUDE },
1700 { "INHIBIT_COMMON_ALLOCATION", INHIBIT_COMMON_ALLOCATION },
1703 { "LENGTH", LENGTH },
1704 { "LOADADDR", LOADADDR },
1708 { "MEMORY", MEMORY },
1711 { "NOCROSSREFS", NOCROSSREFS },
1712 { "NOFLOAT", NOFLOAT },
1713 { "NOLOAD", NOLOAD },
1714 { "ONLY_IF_RO", ONLY_IF_RO },
1715 { "ONLY_IF_RW", ONLY_IF_RW },
1716 { "OPTION", OPTION },
1717 { "ORIGIN", ORIGIN },
1718 { "OUTPUT", OUTPUT },
1719 { "OUTPUT_ARCH", OUTPUT_ARCH },
1720 { "OUTPUT_FORMAT", OUTPUT_FORMAT },
1721 { "OVERLAY", OVERLAY },
1723 { "PROVIDE", PROVIDE },
1724 { "PROVIDE_HIDDEN", PROVIDE_HIDDEN },
1726 { "SEARCH_DIR", SEARCH_DIR },
1727 { "SECTIONS", SECTIONS },
1728 { "SEGMENT_START", SEGMENT_START },
1730 { "SIZEOF", SIZEOF },
1731 { "SIZEOF_HEADERS", SIZEOF_HEADERS },
1732 { "SORT", SORT_BY_NAME },
1733 { "SORT_BY_ALIGNMENT", SORT_BY_ALIGNMENT },
1734 { "SORT_BY_NAME", SORT_BY_NAME },
1735 { "SPECIAL", SPECIAL },
1737 { "STARTUP", STARTUP },
1738 { "SUBALIGN", SUBALIGN },
1739 { "SYSLIB", SYSLIB },
1740 { "TARGET", TARGET_K },
1741 { "TRUNCATE", TRUNCATE },
1742 { "VERSION", VERSIONK },
1743 { "global", GLOBAL },
1749 { "sizeof_headers", SIZEOF_HEADERS },
1752 static const Keyword_to_parsecode
1753 script_keywords(&script_keyword_parsecodes[0],
1754 (sizeof(script_keyword_parsecodes)
1755 / sizeof(script_keyword_parsecodes[0])));
1757 static const Keyword_to_parsecode::Keyword_parsecode
1758 version_script_keyword_parsecodes[] =
1760 { "extern", EXTERN },
1761 { "global", GLOBAL },
1765 static const Keyword_to_parsecode
1766 version_script_keywords(&version_script_keyword_parsecodes[0],
1767 (sizeof(version_script_keyword_parsecodes)
1768 / sizeof(version_script_keyword_parsecodes[0])));
1770 static const Keyword_to_parsecode::Keyword_parsecode
1771 dynamic_list_keyword_parsecodes[] =
1773 { "extern", EXTERN },
1776 static const Keyword_to_parsecode
1777 dynamic_list_keywords(&dynamic_list_keyword_parsecodes[0],
1778 (sizeof(dynamic_list_keyword_parsecodes)
1779 / sizeof(dynamic_list_keyword_parsecodes[0])));
1783 // Comparison function passed to bsearch.
1795 ktt_compare(const void* keyv, const void* kttv)
1797 const Ktt_key* key = static_cast<const Ktt_key*>(keyv);
1798 const Keyword_to_parsecode::Keyword_parsecode* ktt =
1799 static_cast<const Keyword_to_parsecode::Keyword_parsecode*>(kttv);
1800 int i = strncmp(key->str, ktt->keyword, key->len);
1803 if (ktt->keyword[key->len] != '\0')
1808 } // End extern "C".
1811 Keyword_to_parsecode::keyword_to_parsecode(const char* keyword,
1817 void* kttv = bsearch(&key,
1818 this->keyword_parsecodes_,
1819 this->keyword_count_,
1820 sizeof(this->keyword_parsecodes_[0]),
1824 Keyword_parsecode* ktt = static_cast<Keyword_parsecode*>(kttv);
1825 return ktt->parsecode;
1828 // The following structs are used within the VersionInfo class as well
1829 // as in the bison helper functions. They store the information
1830 // parsed from the version script.
1832 // A single version expression.
1833 // For example, pattern="std::map*" and language="C++".
1834 struct Version_expression
1836 Version_expression(const std::string& a_pattern,
1837 Version_script_info::Language a_language,
1839 : pattern(a_pattern), language(a_language), exact_match(a_exact_match),
1840 was_matched_by_symbol(false)
1843 std::string pattern;
1844 Version_script_info::Language language;
1845 // If false, we use glob() to match pattern. If true, we use strcmp().
1847 // True if --no-undefined-version is in effect and we found this
1848 // version in get_symbol_version. We use mutable because this
1849 // struct is generally not modifiable after it has been created.
1850 mutable bool was_matched_by_symbol;
1853 // A list of expressions.
1854 struct Version_expression_list
1856 std::vector<struct Version_expression> expressions;
1859 // A list of which versions upon which another version depends.
1860 // Strings should be from the Stringpool.
1861 struct Version_dependency_list
1863 std::vector<std::string> dependencies;
1866 // The total definition of a version. It includes the tag for the
1867 // version, its global and local expressions, and any dependencies.
1871 : tag(), global(NULL), local(NULL), dependencies(NULL)
1875 const struct Version_expression_list* global;
1876 const struct Version_expression_list* local;
1877 const struct Version_dependency_list* dependencies;
1880 // Helper class that calls cplus_demangle when needed and takes care of freeing
1883 class Lazy_demangler
1886 Lazy_demangler(const char* symbol, int options)
1887 : symbol_(symbol), options_(options), demangled_(NULL), did_demangle_(false)
1891 { free(this->demangled_); }
1893 // Return the demangled name. The actual demangling happens on the first call,
1894 // and the result is later cached.
1899 // The symbol to demangle.
1900 const char* symbol_;
1901 // Option flags to pass to cplus_demagle.
1903 // The cached demangled value, or NULL if demangling didn't happen yet or
1906 // Whether we already called cplus_demangle
1910 // Return the demangled name. The actual demangling happens on the first call,
1911 // and the result is later cached. Returns NULL if the symbol cannot be
1915 Lazy_demangler::get()
1917 if (!this->did_demangle_)
1919 this->demangled_ = cplus_demangle(this->symbol_, this->options_);
1920 this->did_demangle_ = true;
1922 return this->demangled_;
1925 // Class Version_script_info.
1927 Version_script_info::Version_script_info()
1928 : dependency_lists_(), expression_lists_(), version_trees_(), globs_(),
1929 default_version_(NULL), default_is_global_(false), is_finalized_(false)
1931 for (int i = 0; i < LANGUAGE_COUNT; ++i)
1932 this->exact_[i] = NULL;
1935 Version_script_info::~Version_script_info()
1939 // Forget all the known version script information.
1942 Version_script_info::clear()
1944 for (size_t k = 0; k < this->dependency_lists_.size(); ++k)
1945 delete this->dependency_lists_[k];
1946 this->dependency_lists_.clear();
1947 for (size_t k = 0; k < this->version_trees_.size(); ++k)
1948 delete this->version_trees_[k];
1949 this->version_trees_.clear();
1950 for (size_t k = 0; k < this->expression_lists_.size(); ++k)
1951 delete this->expression_lists_[k];
1952 this->expression_lists_.clear();
1955 // Finalize the version script information.
1958 Version_script_info::finalize()
1960 if (!this->is_finalized_)
1962 this->build_lookup_tables();
1963 this->is_finalized_ = true;
1967 // Return all the versions.
1969 std::vector<std::string>
1970 Version_script_info::get_versions() const
1972 std::vector<std::string> ret;
1973 for (size_t j = 0; j < this->version_trees_.size(); ++j)
1974 if (!this->version_trees_[j]->tag.empty())
1975 ret.push_back(this->version_trees_[j]->tag);
1979 // Return the dependencies of VERSION.
1981 std::vector<std::string>
1982 Version_script_info::get_dependencies(const char* version) const
1984 std::vector<std::string> ret;
1985 for (size_t j = 0; j < this->version_trees_.size(); ++j)
1986 if (this->version_trees_[j]->tag == version)
1988 const struct Version_dependency_list* deps =
1989 this->version_trees_[j]->dependencies;
1991 for (size_t k = 0; k < deps->dependencies.size(); ++k)
1992 ret.push_back(deps->dependencies[k]);
1998 // A version script essentially maps a symbol name to a version tag
1999 // and an indication of whether symbol is global or local within that
2000 // version tag. Each symbol maps to at most one version tag.
2001 // Unfortunately, in practice, version scripts are ambiguous, and list
2002 // symbols multiple times. Thus, we have to document the matching
2005 // This is a description of what the GNU linker does as of 2010-01-11.
2006 // It walks through the version tags in the order in which they appear
2007 // in the version script. For each tag, it first walks through the
2008 // global patterns for that tag, then the local patterns. When
2009 // looking at a single pattern, it first applies any language specific
2010 // demangling as specified for the pattern, and then matches the
2011 // resulting symbol name to the pattern. If it finds an exact match
2012 // for a literal pattern (a pattern enclosed in quotes or with no
2013 // wildcard characters), then that is the match that it uses. If
2014 // finds a match with a wildcard pattern, then it saves it and
2015 // continues searching. Wildcard patterns that are exactly "*" are
2016 // saved separately.
2018 // If no exact match with a literal pattern is ever found, then if a
2019 // wildcard match with a global pattern was found it is used,
2020 // otherwise if a wildcard match with a local pattern was found it is
2023 // This is the result:
2024 // * If there is an exact match, then we use the first tag in the
2025 // version script where it matches.
2026 // + If the exact match in that tag is global, it is used.
2027 // + Otherwise the exact match in that tag is local, and is used.
2028 // * Otherwise, if there is any match with a global wildcard pattern:
2029 // + If there is any match with a wildcard pattern which is not
2030 // "*", then we use the tag in which the *last* such pattern
2032 // + Otherwise, we matched "*". If there is no match with a local
2033 // wildcard pattern which is not "*", then we use the *last*
2034 // match with a global "*". Otherwise, continue.
2035 // * Otherwise, if there is any match with a local wildcard pattern:
2036 // + If there is any match with a wildcard pattern which is not
2037 // "*", then we use the tag in which the *last* such pattern
2039 // + Otherwise, we matched "*", and we use the tag in which the
2040 // *last* such match occurred.
2042 // There is an additional wrinkle. When the GNU linker finds a symbol
2043 // with a version defined in an object file due to a .symver
2044 // directive, it looks up that symbol name in that version tag. If it
2045 // finds it, it matches the symbol name against the patterns for that
2046 // version. If there is no match with a global pattern, but there is
2047 // a match with a local pattern, then the GNU linker marks the symbol
2050 // We want gold to be generally compatible, but we also want gold to
2051 // be fast. These are the rules that gold implements:
2052 // * If there is an exact match for the mangled name, we use it.
2053 // + If there is more than one exact match, we give a warning, and
2054 // we use the first tag in the script which matches.
2055 // + If a symbol has an exact match as both global and local for
2056 // the same version tag, we give an error.
2057 // * Otherwise, we look for an extern C++ or an extern Java exact
2058 // match. If we find an exact match, we use it.
2059 // + If there is more than one exact match, we give a warning, and
2060 // we use the first tag in the script which matches.
2061 // + If a symbol has an exact match as both global and local for
2062 // the same version tag, we give an error.
2063 // * Otherwise, we look through the wildcard patterns, ignoring "*"
2064 // patterns. We look through the version tags in reverse order.
2065 // For each version tag, we look through the global patterns and
2066 // then the local patterns. We use the first match we find (i.e.,
2067 // the last matching version tag in the file).
2068 // * Otherwise, we use the "*" pattern if there is one. We give an
2069 // error if there are multiple "*" patterns.
2071 // At least for now, gold does not look up the version tag for a
2072 // symbol version found in an object file to see if it should be
2073 // forced local. There are other ways to force a symbol to be local,
2074 // and I don't understand why this one is useful.
2076 // Build a set of fast lookup tables for a version script.
2079 Version_script_info::build_lookup_tables()
2081 size_t size = this->version_trees_.size();
2082 for (size_t j = 0; j < size; ++j)
2084 const Version_tree* v = this->version_trees_[j];
2085 this->build_expression_list_lookup(v->local, v, false);
2086 this->build_expression_list_lookup(v->global, v, true);
2090 // If a pattern has backlashes but no unquoted wildcard characters,
2091 // then we apply backslash unquoting and look for an exact match.
2092 // Otherwise we treat it as a wildcard pattern. This function returns
2093 // true for a wildcard pattern. Otherwise, it does backslash
2094 // unquoting on *PATTERN and returns false. If this returns true,
2095 // *PATTERN may have been partially unquoted.
2098 Version_script_info::unquote(std::string* pattern) const
2100 bool saw_backslash = false;
2101 size_t len = pattern->length();
2103 for (size_t i = 0; i < len; ++i)
2106 saw_backslash = false;
2109 switch ((*pattern)[i])
2111 case '?': case '[': case '*':
2114 saw_backslash = true;
2122 (*pattern)[j] = (*pattern)[i];
2128 // Add an exact match for MATCH to *PE. The result of the match is
2132 Version_script_info::add_exact_match(const std::string& match,
2133 const Version_tree* v, bool is_global,
2134 const Version_expression* ve,
2137 std::pair<Exact::iterator, bool> ins =
2138 pe->insert(std::make_pair(match, Version_tree_match(v, is_global, ve)));
2141 // This is the first time we have seen this match.
2145 Version_tree_match& vtm(ins.first->second);
2146 if (vtm.real->tag != v->tag)
2148 // This is an ambiguous match. We still return the
2149 // first version that we found in the script, but we
2150 // record the new version to issue a warning if we
2151 // wind up looking up this symbol.
2152 if (vtm.ambiguous == NULL)
2155 else if (is_global != vtm.is_global)
2157 // We have a match for both the global and local entries for a
2158 // version tag. That's got to be wrong.
2159 gold_error(_("'%s' appears as both a global and a local symbol "
2160 "for version '%s' in script"),
2161 match.c_str(), v->tag.c_str());
2165 // Build fast lookup information for EXPLIST and store it in LOOKUP.
2166 // All matches go to V, and IS_GLOBAL is true if they are global
2170 Version_script_info::build_expression_list_lookup(
2171 const Version_expression_list* explist,
2172 const Version_tree* v,
2175 if (explist == NULL)
2177 size_t size = explist->expressions.size();
2178 for (size_t i = 0; i < size; ++i)
2180 const Version_expression& exp(explist->expressions[i]);
2182 if (exp.pattern.length() == 1 && exp.pattern[0] == '*')
2184 if (this->default_version_ != NULL
2185 && this->default_version_->tag != v->tag)
2186 gold_warning(_("wildcard match appears in both version '%s' "
2187 "and '%s' in script"),
2188 this->default_version_->tag.c_str(), v->tag.c_str());
2189 else if (this->default_version_ != NULL
2190 && this->default_is_global_ != is_global)
2191 gold_error(_("wildcard match appears as both global and local "
2192 "in version '%s' in script"),
2194 this->default_version_ = v;
2195 this->default_is_global_ = is_global;
2199 std::string pattern = exp.pattern;
2200 if (!exp.exact_match)
2202 if (this->unquote(&pattern))
2204 this->globs_.push_back(Glob(&exp, v, is_global));
2209 if (this->exact_[exp.language] == NULL)
2210 this->exact_[exp.language] = new Exact();
2211 this->add_exact_match(pattern, v, is_global, &exp,
2212 this->exact_[exp.language]);
2216 // Return the name to match given a name, a language code, and two
2220 Version_script_info::get_name_to_match(const char* name,
2222 Lazy_demangler* cpp_demangler,
2223 Lazy_demangler* java_demangler) const
2230 return cpp_demangler->get();
2232 return java_demangler->get();
2238 // Look up SYMBOL_NAME in the list of versions. Return true if the
2239 // symbol is found, false if not. If the symbol is found, then if
2240 // PVERSION is not NULL, set *PVERSION to the version tag, and if
2241 // P_IS_GLOBAL is not NULL, set *P_IS_GLOBAL according to whether the
2242 // symbol is global or not.
2245 Version_script_info::get_symbol_version(const char* symbol_name,
2246 std::string* pversion,
2247 bool* p_is_global) const
2249 Lazy_demangler cpp_demangled_name(symbol_name, DMGL_ANSI | DMGL_PARAMS);
2250 Lazy_demangler java_demangled_name(symbol_name,
2251 DMGL_ANSI | DMGL_PARAMS | DMGL_JAVA);
2253 gold_assert(this->is_finalized_);
2254 for (int i = 0; i < LANGUAGE_COUNT; ++i)
2256 Exact* exact = this->exact_[i];
2260 const char* name_to_match = this->get_name_to_match(symbol_name, i,
2261 &cpp_demangled_name,
2262 &java_demangled_name);
2263 if (name_to_match == NULL)
2265 // If the name can not be demangled, the GNU linker goes
2266 // ahead and tries to match it anyhow. That does not
2267 // make sense to me and I have not implemented it.
2271 Exact::const_iterator pe = exact->find(name_to_match);
2272 if (pe != exact->end())
2274 const Version_tree_match& vtm(pe->second);
2275 if (vtm.ambiguous != NULL)
2276 gold_warning(_("using '%s' as version for '%s' which is also "
2277 "named in version '%s' in script"),
2278 vtm.real->tag.c_str(), name_to_match,
2279 vtm.ambiguous->tag.c_str());
2281 if (pversion != NULL)
2282 *pversion = vtm.real->tag;
2283 if (p_is_global != NULL)
2284 *p_is_global = vtm.is_global;
2286 // If we are using --no-undefined-version, and this is a
2287 // global symbol, we have to record that we have found this
2288 // symbol, so that we don't warn about it. We have to do
2289 // this now, because otherwise we have no way to get from a
2290 // non-C language back to the demangled name that we
2292 if (p_is_global != NULL && vtm.is_global)
2293 vtm.expression->was_matched_by_symbol = true;
2299 // Look through the glob patterns in reverse order.
2301 for (Globs::const_reverse_iterator p = this->globs_.rbegin();
2302 p != this->globs_.rend();
2305 int language = p->expression->language;
2306 const char* name_to_match = this->get_name_to_match(symbol_name,
2308 &cpp_demangled_name,
2309 &java_demangled_name);
2310 if (name_to_match == NULL)
2313 if (fnmatch(p->expression->pattern.c_str(), name_to_match,
2316 if (pversion != NULL)
2317 *pversion = p->version->tag;
2318 if (p_is_global != NULL)
2319 *p_is_global = p->is_global;
2324 // Finally, there may be a wildcard.
2325 if (this->default_version_ != NULL)
2327 if (pversion != NULL)
2328 *pversion = this->default_version_->tag;
2329 if (p_is_global != NULL)
2330 *p_is_global = this->default_is_global_;
2337 // Give an error if any exact symbol names (not wildcards) appear in a
2338 // version script, but there is no such symbol.
2341 Version_script_info::check_unmatched_names(const Symbol_table* symtab) const
2343 for (size_t i = 0; i < this->version_trees_.size(); ++i)
2345 const Version_tree* vt = this->version_trees_[i];
2346 if (vt->global == NULL)
2348 for (size_t j = 0; j < vt->global->expressions.size(); ++j)
2350 const Version_expression& expression(vt->global->expressions[j]);
2352 // Ignore cases where we used the version because we saw a
2353 // symbol that we looked up. Note that
2354 // WAS_MATCHED_BY_SYMBOL will be true even if the symbol was
2355 // not a definition. That's OK as in that case we most
2356 // likely gave an undefined symbol error anyhow.
2357 if (expression.was_matched_by_symbol)
2360 // Just ignore names which are in languages other than C.
2361 // We have no way to look them up in the symbol table.
2362 if (expression.language != LANGUAGE_C)
2365 // Remove backslash quoting, and ignore wildcard patterns.
2366 std::string pattern = expression.pattern;
2367 if (!expression.exact_match)
2369 if (this->unquote(&pattern))
2373 if (symtab->lookup(pattern.c_str(), vt->tag.c_str()) == NULL)
2374 gold_error(_("version script assignment of %s to symbol %s "
2375 "failed: symbol not defined"),
2376 vt->tag.c_str(), pattern.c_str());
2381 struct Version_dependency_list*
2382 Version_script_info::allocate_dependency_list()
2384 dependency_lists_.push_back(new Version_dependency_list);
2385 return dependency_lists_.back();
2388 struct Version_expression_list*
2389 Version_script_info::allocate_expression_list()
2391 expression_lists_.push_back(new Version_expression_list);
2392 return expression_lists_.back();
2395 struct Version_tree*
2396 Version_script_info::allocate_version_tree()
2398 version_trees_.push_back(new Version_tree);
2399 return version_trees_.back();
2402 // Print for debugging.
2405 Version_script_info::print(FILE* f) const
2410 fprintf(f, "VERSION {");
2412 for (size_t i = 0; i < this->version_trees_.size(); ++i)
2414 const Version_tree* vt = this->version_trees_[i];
2416 if (vt->tag.empty())
2419 fprintf(f, " %s {\n", vt->tag.c_str());
2421 if (vt->global != NULL)
2423 fprintf(f, " global :\n");
2424 this->print_expression_list(f, vt->global);
2427 if (vt->local != NULL)
2429 fprintf(f, " local :\n");
2430 this->print_expression_list(f, vt->local);
2434 if (vt->dependencies != NULL)
2436 const Version_dependency_list* deps = vt->dependencies;
2437 for (size_t j = 0; j < deps->dependencies.size(); ++j)
2439 if (j < deps->dependencies.size() - 1)
2441 fprintf(f, " %s", deps->dependencies[j].c_str());
2451 Version_script_info::print_expression_list(
2453 const Version_expression_list* vel) const
2455 Version_script_info::Language current_language = LANGUAGE_C;
2456 for (size_t i = 0; i < vel->expressions.size(); ++i)
2458 const Version_expression& ve(vel->expressions[i]);
2460 if (ve.language != current_language)
2462 if (current_language != LANGUAGE_C)
2464 switch (ve.language)
2469 fprintf(f, " extern \"C++\" {\n");
2472 fprintf(f, " extern \"Java\" {\n");
2477 current_language = ve.language;
2481 if (current_language != LANGUAGE_C)
2486 fprintf(f, "%s", ve.pattern.c_str());
2493 if (current_language != LANGUAGE_C)
2497 } // End namespace gold.
2499 // The remaining functions are extern "C", so it's clearer to not put
2500 // them in namespace gold.
2502 using namespace gold;
2504 // This function is called by the bison parser to return the next
2508 yylex(YYSTYPE* lvalp, void* closurev)
2510 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2511 const Token* token = closure->next_token();
2512 switch (token->classification())
2517 case Token::TOKEN_INVALID:
2518 yyerror(closurev, "invalid character");
2521 case Token::TOKEN_EOF:
2524 case Token::TOKEN_STRING:
2526 // This is either a keyword or a STRING.
2528 const char* str = token->string_value(&len);
2530 switch (closure->lex_mode())
2532 case Lex::LINKER_SCRIPT:
2533 parsecode = script_keywords.keyword_to_parsecode(str, len);
2535 case Lex::VERSION_SCRIPT:
2536 parsecode = version_script_keywords.keyword_to_parsecode(str, len);
2538 case Lex::DYNAMIC_LIST:
2539 parsecode = dynamic_list_keywords.keyword_to_parsecode(str, len);
2546 lvalp->string.value = str;
2547 lvalp->string.length = len;
2551 case Token::TOKEN_QUOTED_STRING:
2552 lvalp->string.value = token->string_value(&lvalp->string.length);
2553 return QUOTED_STRING;
2555 case Token::TOKEN_OPERATOR:
2556 return token->operator_value();
2558 case Token::TOKEN_INTEGER:
2559 lvalp->integer = token->integer_value();
2564 // This function is called by the bison parser to report an error.
2567 yyerror(void* closurev, const char* message)
2569 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2570 gold_error(_("%s:%d:%d: %s"), closure->filename(), closure->lineno(),
2571 closure->charpos(), message);
2574 // Called by the bison parser to add an external symbol to the link.
2577 script_add_extern(void* closurev, const char* name, size_t length)
2579 // We treat exactly like -u NAME. FIXME: If it seems useful, we
2580 // could handle this after the command line has been read, by adding
2581 // entries to the symbol table directly.
2582 std::string arg("--undefined=");
2583 arg.append(name, length);
2584 script_parse_option(closurev, arg.c_str(), arg.size());
2587 // Called by the bison parser to add a file to the link.
2590 script_add_file(void* closurev, const char* name, size_t length)
2592 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2594 // If this is an absolute path, and we found the script in the
2595 // sysroot, then we want to prepend the sysroot to the file name.
2596 // For example, this is how we handle a cross link to the x86_64
2597 // libc.so, which refers to /lib/libc.so.6.
2598 std::string name_string(name, length);
2599 const char* extra_search_path = ".";
2600 std::string script_directory;
2601 if (IS_ABSOLUTE_PATH(name_string.c_str()))
2603 if (closure->is_in_sysroot())
2605 const std::string& sysroot(parameters->options().sysroot());
2606 gold_assert(!sysroot.empty());
2607 name_string = sysroot + name_string;
2612 // In addition to checking the normal library search path, we
2613 // also want to check in the script-directory.
2614 const char* slash = strrchr(closure->filename(), '/');
2617 script_directory.assign(closure->filename(),
2618 slash - closure->filename() + 1);
2619 extra_search_path = script_directory.c_str();
2623 Input_file_argument file(name_string.c_str(),
2624 Input_file_argument::INPUT_FILE_TYPE_FILE,
2625 extra_search_path, false,
2626 closure->position_dependent_options());
2627 closure->inputs()->add_file(file);
2630 // Called by the bison parser to add a library to the link.
2633 script_add_library(void* closurev, const char* name, size_t length)
2635 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2636 std::string name_string(name, length);
2638 if (name_string[0] != 'l')
2639 gold_error(_("library name must be prefixed with -l"));
2641 Input_file_argument file(name_string.c_str() + 1,
2642 Input_file_argument::INPUT_FILE_TYPE_LIBRARY,
2644 closure->position_dependent_options());
2645 closure->inputs()->add_file(file);
2648 // Called by the bison parser to start a group. If we are already in
2649 // a group, that means that this script was invoked within a
2650 // --start-group --end-group sequence on the command line, or that
2651 // this script was found in a GROUP of another script. In that case,
2652 // we simply continue the existing group, rather than starting a new
2653 // one. It is possible to construct a case in which this will do
2654 // something other than what would happen if we did a recursive group,
2655 // but it's hard to imagine why the different behaviour would be
2656 // useful for a real program. Avoiding recursive groups is simpler
2657 // and more efficient.
2660 script_start_group(void* closurev)
2662 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2663 if (!closure->in_group())
2664 closure->inputs()->start_group();
2667 // Called by the bison parser at the end of a group.
2670 script_end_group(void* closurev)
2672 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2673 if (!closure->in_group())
2674 closure->inputs()->end_group();
2677 // Called by the bison parser to start an AS_NEEDED list.
2680 script_start_as_needed(void* closurev)
2682 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2683 closure->position_dependent_options().set_as_needed(true);
2686 // Called by the bison parser at the end of an AS_NEEDED list.
2689 script_end_as_needed(void* closurev)
2691 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2692 closure->position_dependent_options().set_as_needed(false);
2695 // Called by the bison parser to set the entry symbol.
2698 script_set_entry(void* closurev, const char* entry, size_t length)
2700 // We'll parse this exactly the same as --entry=ENTRY on the commandline
2701 // TODO(csilvers): FIXME -- call set_entry directly.
2702 std::string arg("--entry=");
2703 arg.append(entry, length);
2704 script_parse_option(closurev, arg.c_str(), arg.size());
2707 // Called by the bison parser to set whether to define common symbols.
2710 script_set_common_allocation(void* closurev, int set)
2712 const char* arg = set != 0 ? "--define-common" : "--no-define-common";
2713 script_parse_option(closurev, arg, strlen(arg));
2716 // Called by the bison parser to refer to a symbol.
2718 extern "C" Expression*
2719 script_symbol(void* closurev, const char* name, size_t length)
2721 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2722 if (length != 1 || name[0] != '.')
2723 closure->script_options()->add_symbol_reference(name, length);
2724 return script_exp_string(name, length);
2727 // Called by the bison parser to define a symbol.
2730 script_set_symbol(void* closurev, const char* name, size_t length,
2731 Expression* value, int providei, int hiddeni)
2733 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2734 const bool provide = providei != 0;
2735 const bool hidden = hiddeni != 0;
2736 closure->script_options()->add_symbol_assignment(name, length,
2737 closure->parsing_defsym(),
2738 value, provide, hidden);
2739 closure->clear_skip_on_incompatible_target();
2742 // Called by the bison parser to add an assertion.
2745 script_add_assertion(void* closurev, Expression* check, const char* message,
2748 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2749 closure->script_options()->add_assertion(check, message, messagelen);
2750 closure->clear_skip_on_incompatible_target();
2753 // Called by the bison parser to parse an OPTION.
2756 script_parse_option(void* closurev, const char* option, size_t length)
2758 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2759 // We treat the option as a single command-line option, even if
2760 // it has internal whitespace.
2761 if (closure->command_line() == NULL)
2763 // There are some options that we could handle here--e.g.,
2764 // -lLIBRARY. Should we bother?
2765 gold_warning(_("%s:%d:%d: ignoring command OPTION; OPTION is only valid"
2766 " for scripts specified via -T/--script"),
2767 closure->filename(), closure->lineno(), closure->charpos());
2771 bool past_a_double_dash_option = false;
2772 const char* mutable_option = strndup(option, length);
2773 gold_assert(mutable_option != NULL);
2774 closure->command_line()->process_one_option(1, &mutable_option, 0,
2775 &past_a_double_dash_option);
2776 // The General_options class will quite possibly store a pointer
2777 // into mutable_option, so we can't free it. In cases the class
2778 // does not store such a pointer, this is a memory leak. Alas. :(
2780 closure->clear_skip_on_incompatible_target();
2783 // Called by the bison parser to handle OUTPUT_FORMAT. OUTPUT_FORMAT
2784 // takes either one or three arguments. In the three argument case,
2785 // the format depends on the endianness option, which we don't
2786 // currently support (FIXME). If we see an OUTPUT_FORMAT for the
2787 // wrong format, then we want to search for a new file. Returning 0
2788 // here will cause the parser to immediately abort.
2791 script_check_output_format(void* closurev,
2792 const char* default_name, size_t default_length,
2793 const char*, size_t, const char*, size_t)
2795 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2796 std::string name(default_name, default_length);
2797 Target* target = select_target_by_name(name.c_str());
2798 if (target == NULL || !parameters->is_compatible_target(target))
2800 if (closure->skip_on_incompatible_target())
2802 closure->set_found_incompatible_target();
2805 // FIXME: Should we warn about the unknown target?
2810 // Called by the bison parser to handle TARGET.
2813 script_set_target(void* closurev, const char* target, size_t len)
2815 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2816 std::string s(target, len);
2817 General_options::Object_format format_enum;
2818 format_enum = General_options::string_to_object_format(s.c_str());
2819 closure->position_dependent_options().set_format_enum(format_enum);
2822 // Called by the bison parser to handle SEARCH_DIR. This is handled
2823 // exactly like a -L option.
2826 script_add_search_dir(void* closurev, const char* option, size_t length)
2828 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2829 if (closure->command_line() == NULL)
2830 gold_warning(_("%s:%d:%d: ignoring SEARCH_DIR; SEARCH_DIR is only valid"
2831 " for scripts specified via -T/--script"),
2832 closure->filename(), closure->lineno(), closure->charpos());
2833 else if (!closure->command_line()->options().nostdlib())
2835 std::string s = "-L" + std::string(option, length);
2836 script_parse_option(closurev, s.c_str(), s.size());
2840 /* Called by the bison parser to push the lexer into expression
2844 script_push_lex_into_expression_mode(void* closurev)
2846 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2847 closure->push_lex_mode(Lex::EXPRESSION);
2850 /* Called by the bison parser to push the lexer into version
2854 script_push_lex_into_version_mode(void* closurev)
2856 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2857 if (closure->version_script()->is_finalized())
2858 gold_error(_("%s:%d:%d: invalid use of VERSION in input file"),
2859 closure->filename(), closure->lineno(), closure->charpos());
2860 closure->push_lex_mode(Lex::VERSION_SCRIPT);
2863 /* Called by the bison parser to pop the lexer mode. */
2866 script_pop_lex_mode(void* closurev)
2868 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2869 closure->pop_lex_mode();
2872 // Register an entire version node. For example:
2878 // - tag is "GLIBC_2.1"
2879 // - tree contains the information "global: foo"
2880 // - deps contains "GLIBC_2.0"
2883 script_register_vers_node(void*,
2886 struct Version_tree* tree,
2887 struct Version_dependency_list* deps)
2889 gold_assert(tree != NULL);
2890 tree->dependencies = deps;
2892 tree->tag = std::string(tag, taglen);
2895 // Add a dependencies to the list of existing dependencies, if any,
2896 // and return the expanded list.
2898 extern "C" struct Version_dependency_list*
2899 script_add_vers_depend(void* closurev,
2900 struct Version_dependency_list* all_deps,
2901 const char* depend_to_add, int deplen)
2903 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2904 if (all_deps == NULL)
2905 all_deps = closure->version_script()->allocate_dependency_list();
2906 all_deps->dependencies.push_back(std::string(depend_to_add, deplen));
2910 // Add a pattern expression to an existing list of expressions, if any.
2912 extern "C" struct Version_expression_list*
2913 script_new_vers_pattern(void* closurev,
2914 struct Version_expression_list* expressions,
2915 const char* pattern, int patlen, int exact_match)
2917 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2918 if (expressions == NULL)
2919 expressions = closure->version_script()->allocate_expression_list();
2920 expressions->expressions.push_back(
2921 Version_expression(std::string(pattern, patlen),
2922 closure->get_current_language(),
2923 static_cast<bool>(exact_match)));
2927 // Attaches b to the end of a, and clears b. So a = a + b and b = {}.
2929 extern "C" struct Version_expression_list*
2930 script_merge_expressions(struct Version_expression_list* a,
2931 struct Version_expression_list* b)
2933 a->expressions.insert(a->expressions.end(),
2934 b->expressions.begin(), b->expressions.end());
2935 // We could delete b and remove it from expressions_lists_, but
2936 // that's a lot of work. This works just as well.
2937 b->expressions.clear();
2941 // Combine the global and local expressions into a a Version_tree.
2943 extern "C" struct Version_tree*
2944 script_new_vers_node(void* closurev,
2945 struct Version_expression_list* global,
2946 struct Version_expression_list* local)
2948 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2949 Version_tree* tree = closure->version_script()->allocate_version_tree();
2950 tree->global = global;
2951 tree->local = local;
2955 // Handle a transition in language, such as at the
2956 // start or end of 'extern "C++"'
2959 version_script_push_lang(void* closurev, const char* lang, int langlen)
2961 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2962 std::string language(lang, langlen);
2963 Version_script_info::Language code;
2964 if (language.empty() || language == "C")
2965 code = Version_script_info::LANGUAGE_C;
2966 else if (language == "C++")
2967 code = Version_script_info::LANGUAGE_CXX;
2968 else if (language == "Java")
2969 code = Version_script_info::LANGUAGE_JAVA;
2972 char* buf = new char[langlen + 100];
2973 snprintf(buf, langlen + 100,
2974 _("unrecognized version script language '%s'"),
2976 yyerror(closurev, buf);
2978 code = Version_script_info::LANGUAGE_C;
2980 closure->push_language(code);
2984 version_script_pop_lang(void* closurev)
2986 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2987 closure->pop_language();
2990 // Called by the bison parser to start a SECTIONS clause.
2993 script_start_sections(void* closurev)
2995 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2996 closure->script_options()->script_sections()->start_sections();
2997 closure->clear_skip_on_incompatible_target();
3000 // Called by the bison parser to finish a SECTIONS clause.
3003 script_finish_sections(void* closurev)
3005 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3006 closure->script_options()->script_sections()->finish_sections();
3009 // Start processing entries for an output section.
3012 script_start_output_section(void* closurev, const char* name, size_t namelen,
3013 const struct Parser_output_section_header* header)
3015 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3016 closure->script_options()->script_sections()->start_output_section(name,
3021 // Finish processing entries for an output section.
3024 script_finish_output_section(void* closurev,
3025 const struct Parser_output_section_trailer* trail)
3027 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3028 closure->script_options()->script_sections()->finish_output_section(trail);
3031 // Add a data item (e.g., "WORD (0)") to the current output section.
3034 script_add_data(void* closurev, int data_token, Expression* val)
3036 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3038 bool is_signed = true;
3060 closure->script_options()->script_sections()->add_data(size, is_signed, val);
3063 // Add a clause setting the fill value to the current output section.
3066 script_add_fill(void* closurev, Expression* val)
3068 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3069 closure->script_options()->script_sections()->add_fill(val);
3072 // Add a new input section specification to the current output
3076 script_add_input_section(void* closurev,
3077 const struct Input_section_spec* spec,
3080 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3081 bool keep = keepi != 0;
3082 closure->script_options()->script_sections()->add_input_section(spec, keep);
3085 // When we see DATA_SEGMENT_ALIGN we record that following output
3086 // sections may be relro.
3089 script_data_segment_align(void* closurev)
3091 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3092 if (!closure->script_options()->saw_sections_clause())
3093 gold_error(_("%s:%d:%d: DATA_SEGMENT_ALIGN not in SECTIONS clause"),
3094 closure->filename(), closure->lineno(), closure->charpos());
3096 closure->script_options()->script_sections()->data_segment_align();
3099 // When we see DATA_SEGMENT_RELRO_END we know that all output sections
3100 // since DATA_SEGMENT_ALIGN should be relro.
3103 script_data_segment_relro_end(void* closurev)
3105 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3106 if (!closure->script_options()->saw_sections_clause())
3107 gold_error(_("%s:%d:%d: DATA_SEGMENT_ALIGN not in SECTIONS clause"),
3108 closure->filename(), closure->lineno(), closure->charpos());
3110 closure->script_options()->script_sections()->data_segment_relro_end();
3113 // Create a new list of string/sort pairs.
3115 extern "C" String_sort_list_ptr
3116 script_new_string_sort_list(const struct Wildcard_section* string_sort)
3118 return new String_sort_list(1, *string_sort);
3121 // Add an entry to a list of string/sort pairs. The way the parser
3122 // works permits us to simply modify the first parameter, rather than
3125 extern "C" String_sort_list_ptr
3126 script_string_sort_list_add(String_sort_list_ptr pv,
3127 const struct Wildcard_section* string_sort)
3130 return script_new_string_sort_list(string_sort);
3133 pv->push_back(*string_sort);
3138 // Create a new list of strings.
3140 extern "C" String_list_ptr
3141 script_new_string_list(const char* str, size_t len)
3143 return new String_list(1, std::string(str, len));
3146 // Add an element to a list of strings. The way the parser works
3147 // permits us to simply modify the first parameter, rather than copy
3150 extern "C" String_list_ptr
3151 script_string_list_push_back(String_list_ptr pv, const char* str, size_t len)
3154 return script_new_string_list(str, len);
3157 pv->push_back(std::string(str, len));
3162 // Concatenate two string lists. Either or both may be NULL. The way
3163 // the parser works permits us to modify the parameters, rather than
3166 extern "C" String_list_ptr
3167 script_string_list_append(String_list_ptr pv1, String_list_ptr pv2)
3173 pv1->insert(pv1->end(), pv2->begin(), pv2->end());
3177 // Add a new program header.
3180 script_add_phdr(void* closurev, const char* name, size_t namelen,
3181 unsigned int type, const Phdr_info* info)
3183 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3184 bool includes_filehdr = info->includes_filehdr != 0;
3185 bool includes_phdrs = info->includes_phdrs != 0;
3186 bool is_flags_valid = info->is_flags_valid != 0;
3187 Script_sections* ss = closure->script_options()->script_sections();
3188 ss->add_phdr(name, namelen, type, includes_filehdr, includes_phdrs,
3189 is_flags_valid, info->flags, info->load_address);
3190 closure->clear_skip_on_incompatible_target();
3193 // Convert a program header string to a type.
3195 #define PHDR_TYPE(NAME) { #NAME, sizeof(#NAME) - 1, elfcpp::NAME }
3202 } phdr_type_names[] =
3206 PHDR_TYPE(PT_DYNAMIC),
3207 PHDR_TYPE(PT_INTERP),
3209 PHDR_TYPE(PT_SHLIB),
3212 PHDR_TYPE(PT_GNU_EH_FRAME),
3213 PHDR_TYPE(PT_GNU_STACK),
3214 PHDR_TYPE(PT_GNU_RELRO)
3217 extern "C" unsigned int
3218 script_phdr_string_to_type(void* closurev, const char* name, size_t namelen)
3220 for (unsigned int i = 0;
3221 i < sizeof(phdr_type_names) / sizeof(phdr_type_names[0]);
3223 if (namelen == phdr_type_names[i].namelen
3224 && strncmp(name, phdr_type_names[i].name, namelen) == 0)
3225 return phdr_type_names[i].val;
3226 yyerror(closurev, _("unknown PHDR type (try integer)"));
3227 return elfcpp::PT_NULL;
3231 script_saw_segment_start_expression(void* closurev)
3233 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3234 Script_sections* ss = closure->script_options()->script_sections();
3235 ss->set_saw_segment_start_expression(true);
3239 script_set_section_region(void* closurev, const char* name, size_t namelen,
3242 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3243 if (!closure->script_options()->saw_sections_clause())
3245 gold_error(_("%s:%d:%d: MEMORY region '%.*s' referred to outside of "
3247 closure->filename(), closure->lineno(), closure->charpos(),
3248 static_cast<int>(namelen), name);
3252 Script_sections* ss = closure->script_options()->script_sections();
3253 Memory_region* mr = ss->find_memory_region(name, namelen);
3256 gold_error(_("%s:%d:%d: MEMORY region '%.*s' not declared"),
3257 closure->filename(), closure->lineno(), closure->charpos(),
3258 static_cast<int>(namelen), name);
3262 ss->set_memory_region(mr, set_vma);
3266 script_add_memory(void* closurev, const char* name, size_t namelen,
3267 unsigned int attrs, Expression* origin, Expression* length)
3269 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3270 Script_sections* ss = closure->script_options()->script_sections();
3271 ss->add_memory_region(name, namelen, attrs, origin, length);
3274 extern "C" unsigned int
3275 script_parse_memory_attr(void* closurev, const char* attrs, size_t attrlen,
3285 attributes |= MEM_READABLE; break;
3288 attributes |= MEM_READABLE | MEM_WRITEABLE; break;
3291 attributes |= MEM_EXECUTABLE; break;
3294 attributes |= MEM_ALLOCATABLE; break;
3299 attributes |= MEM_INITIALIZED; break;
3301 yyerror(closurev, _("unknown MEMORY attribute"));
3305 attributes = (~ attributes) & MEM_ATTR_MASK;
3311 script_include_directive(void* closurev, const char*, size_t)
3313 // FIXME: Implement ?
3314 yyerror (closurev, _("GOLD does not currently support INCLUDE directives"));
3317 // Functions for memory regions.
3319 extern "C" Expression*
3320 script_exp_function_origin(void* closurev, const char* name, size_t namelen)
3322 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3323 Script_sections* ss = closure->script_options()->script_sections();
3324 Expression* origin = ss->find_memory_region_origin(name, namelen);
3328 gold_error(_("undefined memory region '%s' referenced "
3329 "in ORIGIN expression"),
3331 // Create a dummy expression to prevent crashes later on.
3332 origin = script_exp_integer(0);
3338 extern "C" Expression*
3339 script_exp_function_length(void* closurev, const char* name, size_t namelen)
3341 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3342 Script_sections* ss = closure->script_options()->script_sections();
3343 Expression* length = ss->find_memory_region_length(name, namelen);
3347 gold_error(_("undefined memory region '%s' referenced "
3348 "in LENGTH expression"),
3350 // Create a dummy expression to prevent crashes later on.
3351 length = script_exp_integer(0);