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 Script_info* script_info)
1211 : filename_(filename), posdep_options_(posdep_options),
1212 parsing_defsym_(parsing_defsym), in_group_(in_group),
1213 is_in_sysroot_(is_in_sysroot),
1214 skip_on_incompatible_target_(skip_on_incompatible_target),
1215 found_incompatible_target_(false),
1216 command_line_(command_line), script_options_(script_options),
1217 version_script_info_(script_options->version_script_info()),
1218 lex_(lex), lineno_(0), charpos_(0), lex_mode_stack_(), inputs_(NULL),
1219 script_info_(script_info)
1221 // We start out processing C symbols in the default lex mode.
1222 this->language_stack_.push_back(Version_script_info::LANGUAGE_C);
1223 this->lex_mode_stack_.push_back(lex->mode());
1226 // Return the file name.
1229 { return this->filename_; }
1231 // Return the position dependent options. The caller may modify
1233 Position_dependent_options&
1234 position_dependent_options()
1235 { return this->posdep_options_; }
1237 // Whether we are parsing a --defsym.
1239 parsing_defsym() const
1240 { return this->parsing_defsym_; }
1242 // Return whether this script is being run in a group.
1245 { return this->in_group_; }
1247 // Return whether this script was found using a directory in the
1250 is_in_sysroot() const
1251 { return this->is_in_sysroot_; }
1253 // Whether to skip to the next file with the same name if we find an
1254 // incompatible target in an OUTPUT_FORMAT statement.
1256 skip_on_incompatible_target() const
1257 { return this->skip_on_incompatible_target_; }
1259 // Stop skipping to the next file on an incompatible target. This
1260 // is called when we make some unrevocable change to the data
1263 clear_skip_on_incompatible_target()
1264 { this->skip_on_incompatible_target_ = false; }
1266 // Whether we found an incompatible target in an OUTPUT_FORMAT
1269 found_incompatible_target() const
1270 { return this->found_incompatible_target_; }
1272 // Note that we found an incompatible target.
1274 set_found_incompatible_target()
1275 { this->found_incompatible_target_ = true; }
1277 // Returns the Command_line structure passed in at constructor time.
1278 // This value may be NULL. The caller may modify this, which modifies
1279 // the passed-in Command_line object (not a copy).
1282 { return this->command_line_; }
1284 // Return the options which may be set by a script.
1287 { return this->script_options_; }
1289 // Return the object in which version script information should be stored.
1290 Version_script_info*
1292 { return this->version_script_info_; }
1294 // Return the next token, and advance.
1298 const Token* token = this->lex_->next_token();
1299 this->lineno_ = token->lineno();
1300 this->charpos_ = token->charpos();
1304 // Set a new lexer mode, pushing the current one.
1306 push_lex_mode(Lex::Mode mode)
1308 this->lex_mode_stack_.push_back(this->lex_->mode());
1309 this->lex_->set_mode(mode);
1312 // Pop the lexer mode.
1316 gold_assert(!this->lex_mode_stack_.empty());
1317 this->lex_->set_mode(this->lex_mode_stack_.back());
1318 this->lex_mode_stack_.pop_back();
1321 // Return the current lexer mode.
1324 { return this->lex_mode_stack_.back(); }
1326 // Return the line number of the last token.
1329 { return this->lineno_; }
1331 // Return the character position in the line of the last token.
1334 { return this->charpos_; }
1336 // Return the list of input files, creating it if necessary. This
1337 // is a space leak--we never free the INPUTS_ pointer.
1341 if (this->inputs_ == NULL)
1342 this->inputs_ = new Input_arguments();
1343 return this->inputs_;
1346 // Return whether we saw any input files.
1349 { return this->inputs_ != NULL && !this->inputs_->empty(); }
1351 // Return the current language being processed in a version script
1352 // (eg, "C++"). The empty string represents unmangled C names.
1353 Version_script_info::Language
1354 get_current_language() const
1355 { return this->language_stack_.back(); }
1357 // Push a language onto the stack when entering an extern block.
1359 push_language(Version_script_info::Language lang)
1360 { this->language_stack_.push_back(lang); }
1362 // Pop a language off of the stack when exiting an extern block.
1366 gold_assert(!this->language_stack_.empty());
1367 this->language_stack_.pop_back();
1370 // Return a pointer to the incremental info.
1373 { return this->script_info_; }
1376 // The name of the file we are reading.
1377 const char* filename_;
1378 // The position dependent options.
1379 Position_dependent_options posdep_options_;
1380 // True if we are parsing a --defsym.
1381 bool parsing_defsym_;
1382 // Whether we are currently in a --start-group/--end-group.
1384 // Whether the script was found in a sysrooted directory.
1385 bool is_in_sysroot_;
1386 // If this is true, then if we find an OUTPUT_FORMAT with an
1387 // incompatible target, then we tell the parser to abort so that we
1388 // can search for the next file with the same name.
1389 bool skip_on_incompatible_target_;
1390 // True if we found an OUTPUT_FORMAT with an incompatible target.
1391 bool found_incompatible_target_;
1392 // May be NULL if the user chooses not to pass one in.
1393 Command_line* command_line_;
1394 // Options which may be set from any linker script.
1395 Script_options* script_options_;
1396 // Information parsed from a version script.
1397 Version_script_info* version_script_info_;
1400 // The line number of the last token returned by next_token.
1402 // The column number of the last token returned by next_token.
1404 // A stack of lexer modes.
1405 std::vector<Lex::Mode> lex_mode_stack_;
1406 // A stack of which extern/language block we're inside. Can be C++,
1407 // java, or empty for C.
1408 std::vector<Version_script_info::Language> language_stack_;
1409 // New input files found to add to the link.
1410 Input_arguments* inputs_;
1411 // Pointer to incremental linking info.
1412 Script_info* script_info_;
1415 // FILE was found as an argument on the command line. Try to read it
1416 // as a script. Return true if the file was handled.
1419 read_input_script(Workqueue* workqueue, Symbol_table* symtab, Layout* layout,
1420 Dirsearch* dirsearch, int dirindex,
1421 Input_objects* input_objects, Mapfile* mapfile,
1422 Input_group* input_group,
1423 const Input_argument* input_argument,
1424 Input_file* input_file, Task_token* next_blocker,
1425 bool* used_next_blocker)
1427 *used_next_blocker = false;
1429 std::string input_string;
1430 Lex::read_file(input_file, &input_string);
1432 Lex lex(input_string.c_str(), input_string.length(), PARSING_LINKER_SCRIPT);
1434 Script_info* script_info = NULL;
1435 if (layout->incremental_inputs() != NULL)
1437 const std::string& filename = input_file->filename();
1438 Timespec mtime = input_file->file().get_mtime();
1439 script_info = new Script_info();
1440 layout->incremental_inputs()->report_script(filename, script_info, mtime);
1443 Parser_closure closure(input_file->filename().c_str(),
1444 input_argument->file().options(),
1446 input_group != NULL,
1447 input_file->is_in_sysroot(),
1449 layout->script_options(),
1451 input_file->will_search_for(),
1454 bool old_saw_sections_clause =
1455 layout->script_options()->saw_sections_clause();
1457 if (yyparse(&closure) != 0)
1459 if (closure.found_incompatible_target())
1461 Read_symbols::incompatible_warning(input_argument, input_file);
1462 Read_symbols::requeue(workqueue, input_objects, symtab, layout,
1463 dirsearch, dirindex, mapfile, input_argument,
1464 input_group, next_blocker);
1470 if (!old_saw_sections_clause
1471 && layout->script_options()->saw_sections_clause()
1472 && layout->have_added_input_section())
1473 gold_error(_("%s: SECTIONS seen after other input files; try -T/--script"),
1474 input_file->filename().c_str());
1476 if (!closure.saw_inputs())
1479 Task_token* this_blocker = NULL;
1480 for (Input_arguments::const_iterator p = closure.inputs()->begin();
1481 p != closure.inputs()->end();
1485 if (p + 1 == closure.inputs()->end())
1489 nb = new Task_token(true);
1492 workqueue->queue_soon(new Read_symbols(input_objects, symtab,
1493 layout, dirsearch, 0, mapfile, &*p,
1494 input_group, NULL, this_blocker, nb));
1498 *used_next_blocker = true;
1503 // Helper function for read_version_script() and
1504 // read_commandline_script(). Processes the given file in the mode
1505 // indicated by first_token and lex_mode.
1508 read_script_file(const char* filename, Command_line* cmdline,
1509 Script_options* script_options,
1510 int first_token, Lex::Mode lex_mode)
1512 // TODO: if filename is a relative filename, search for it manually
1513 // using "." + cmdline->options()->search_path() -- not dirsearch.
1514 Dirsearch dirsearch;
1516 // The file locking code wants to record a Task, but we haven't
1517 // started the workqueue yet. This is only for debugging purposes,
1518 // so we invent a fake value.
1519 const Task* task = reinterpret_cast<const Task*>(-1);
1521 // We don't want this file to be opened in binary mode.
1522 Position_dependent_options posdep = cmdline->position_dependent_options();
1523 if (posdep.format_enum() == General_options::OBJECT_FORMAT_BINARY)
1524 posdep.set_format_enum(General_options::OBJECT_FORMAT_ELF);
1525 Input_file_argument input_argument(filename,
1526 Input_file_argument::INPUT_FILE_TYPE_FILE,
1528 Input_file input_file(&input_argument);
1530 if (!input_file.open(dirsearch, task, &dummy))
1533 std::string input_string;
1534 Lex::read_file(&input_file, &input_string);
1536 Lex lex(input_string.c_str(), input_string.length(), first_token);
1537 lex.set_mode(lex_mode);
1539 Parser_closure closure(filename,
1540 cmdline->position_dependent_options(),
1541 first_token == Lex::DYNAMIC_LIST,
1543 input_file.is_in_sysroot(),
1549 if (yyparse(&closure) != 0)
1551 input_file.file().unlock(task);
1555 input_file.file().unlock(task);
1557 gold_assert(!closure.saw_inputs());
1562 // FILENAME was found as an argument to --script (-T).
1563 // Read it as a script, and execute its contents immediately.
1566 read_commandline_script(const char* filename, Command_line* cmdline)
1568 return read_script_file(filename, cmdline, &cmdline->script_options(),
1569 PARSING_LINKER_SCRIPT, Lex::LINKER_SCRIPT);
1572 // FILENAME was found as an argument to --version-script. Read it as
1573 // a version script, and store its contents in
1574 // cmdline->script_options()->version_script_info().
1577 read_version_script(const char* filename, Command_line* cmdline)
1579 return read_script_file(filename, cmdline, &cmdline->script_options(),
1580 PARSING_VERSION_SCRIPT, Lex::VERSION_SCRIPT);
1583 // FILENAME was found as an argument to --dynamic-list. Read it as a
1584 // list of symbols, and store its contents in DYNAMIC_LIST.
1587 read_dynamic_list(const char* filename, Command_line* cmdline,
1588 Script_options* dynamic_list)
1590 return read_script_file(filename, cmdline, dynamic_list,
1591 PARSING_DYNAMIC_LIST, Lex::DYNAMIC_LIST);
1594 // Implement the --defsym option on the command line. Return true if
1598 Script_options::define_symbol(const char* definition)
1600 Lex lex(definition, strlen(definition), PARSING_DEFSYM);
1601 lex.set_mode(Lex::EXPRESSION);
1604 Position_dependent_options posdep_options;
1606 Parser_closure closure("command line", posdep_options, true,
1607 false, false, NULL, this, &lex, false, NULL);
1609 if (yyparse(&closure) != 0)
1612 gold_assert(!closure.saw_inputs());
1617 // Print the script to F for debugging.
1620 Script_options::print(FILE* f) const
1622 fprintf(f, "%s: Dumping linker script\n", program_name);
1624 if (!this->entry_.empty())
1625 fprintf(f, "ENTRY(%s)\n", this->entry_.c_str());
1627 for (Symbol_assignments::const_iterator p =
1628 this->symbol_assignments_.begin();
1629 p != this->symbol_assignments_.end();
1633 for (Assertions::const_iterator p = this->assertions_.begin();
1634 p != this->assertions_.end();
1638 this->script_sections_.print(f);
1640 this->version_script_info_.print(f);
1643 // Manage mapping from keywords to the codes expected by the bison
1644 // parser. We construct one global object for each lex mode with
1647 class Keyword_to_parsecode
1650 // The structure which maps keywords to parsecodes.
1651 struct Keyword_parsecode
1654 const char* keyword;
1655 // Corresponding parsecode.
1659 Keyword_to_parsecode(const Keyword_parsecode* keywords,
1661 : keyword_parsecodes_(keywords), keyword_count_(keyword_count)
1664 // Return the parsecode corresponding KEYWORD, or 0 if it is not a
1667 keyword_to_parsecode(const char* keyword, size_t len) const;
1670 const Keyword_parsecode* keyword_parsecodes_;
1671 const int keyword_count_;
1674 // Mapping from keyword string to keyword parsecode. This array must
1675 // be kept in sorted order. Parsecodes are looked up using bsearch.
1676 // This array must correspond to the list of parsecodes in yyscript.y.
1678 static const Keyword_to_parsecode::Keyword_parsecode
1679 script_keyword_parsecodes[] =
1681 { "ABSOLUTE", ABSOLUTE },
1683 { "ALIGN", ALIGN_K },
1684 { "ALIGNOF", ALIGNOF },
1685 { "ASSERT", ASSERT_K },
1686 { "AS_NEEDED", AS_NEEDED },
1691 { "CONSTANT", CONSTANT },
1692 { "CONSTRUCTORS", CONSTRUCTORS },
1694 { "CREATE_OBJECT_SYMBOLS", CREATE_OBJECT_SYMBOLS },
1695 { "DATA_SEGMENT_ALIGN", DATA_SEGMENT_ALIGN },
1696 { "DATA_SEGMENT_END", DATA_SEGMENT_END },
1697 { "DATA_SEGMENT_RELRO_END", DATA_SEGMENT_RELRO_END },
1698 { "DEFINED", DEFINED },
1701 { "EXCLUDE_FILE", EXCLUDE_FILE },
1702 { "EXTERN", EXTERN },
1705 { "FORCE_COMMON_ALLOCATION", FORCE_COMMON_ALLOCATION },
1708 { "INCLUDE", INCLUDE },
1710 { "INHIBIT_COMMON_ALLOCATION", INHIBIT_COMMON_ALLOCATION },
1713 { "LENGTH", LENGTH },
1714 { "LOADADDR", LOADADDR },
1718 { "MEMORY", MEMORY },
1721 { "NOCROSSREFS", NOCROSSREFS },
1722 { "NOFLOAT", NOFLOAT },
1723 { "NOLOAD", NOLOAD },
1724 { "ONLY_IF_RO", ONLY_IF_RO },
1725 { "ONLY_IF_RW", ONLY_IF_RW },
1726 { "OPTION", OPTION },
1727 { "ORIGIN", ORIGIN },
1728 { "OUTPUT", OUTPUT },
1729 { "OUTPUT_ARCH", OUTPUT_ARCH },
1730 { "OUTPUT_FORMAT", OUTPUT_FORMAT },
1731 { "OVERLAY", OVERLAY },
1733 { "PROVIDE", PROVIDE },
1734 { "PROVIDE_HIDDEN", PROVIDE_HIDDEN },
1736 { "SEARCH_DIR", SEARCH_DIR },
1737 { "SECTIONS", SECTIONS },
1738 { "SEGMENT_START", SEGMENT_START },
1740 { "SIZEOF", SIZEOF },
1741 { "SIZEOF_HEADERS", SIZEOF_HEADERS },
1742 { "SORT", SORT_BY_NAME },
1743 { "SORT_BY_ALIGNMENT", SORT_BY_ALIGNMENT },
1744 { "SORT_BY_NAME", SORT_BY_NAME },
1745 { "SPECIAL", SPECIAL },
1747 { "STARTUP", STARTUP },
1748 { "SUBALIGN", SUBALIGN },
1749 { "SYSLIB", SYSLIB },
1750 { "TARGET", TARGET_K },
1751 { "TRUNCATE", TRUNCATE },
1752 { "VERSION", VERSIONK },
1753 { "global", GLOBAL },
1759 { "sizeof_headers", SIZEOF_HEADERS },
1762 static const Keyword_to_parsecode
1763 script_keywords(&script_keyword_parsecodes[0],
1764 (sizeof(script_keyword_parsecodes)
1765 / sizeof(script_keyword_parsecodes[0])));
1767 static const Keyword_to_parsecode::Keyword_parsecode
1768 version_script_keyword_parsecodes[] =
1770 { "extern", EXTERN },
1771 { "global", GLOBAL },
1775 static const Keyword_to_parsecode
1776 version_script_keywords(&version_script_keyword_parsecodes[0],
1777 (sizeof(version_script_keyword_parsecodes)
1778 / sizeof(version_script_keyword_parsecodes[0])));
1780 static const Keyword_to_parsecode::Keyword_parsecode
1781 dynamic_list_keyword_parsecodes[] =
1783 { "extern", EXTERN },
1786 static const Keyword_to_parsecode
1787 dynamic_list_keywords(&dynamic_list_keyword_parsecodes[0],
1788 (sizeof(dynamic_list_keyword_parsecodes)
1789 / sizeof(dynamic_list_keyword_parsecodes[0])));
1793 // Comparison function passed to bsearch.
1805 ktt_compare(const void* keyv, const void* kttv)
1807 const Ktt_key* key = static_cast<const Ktt_key*>(keyv);
1808 const Keyword_to_parsecode::Keyword_parsecode* ktt =
1809 static_cast<const Keyword_to_parsecode::Keyword_parsecode*>(kttv);
1810 int i = strncmp(key->str, ktt->keyword, key->len);
1813 if (ktt->keyword[key->len] != '\0')
1818 } // End extern "C".
1821 Keyword_to_parsecode::keyword_to_parsecode(const char* keyword,
1827 void* kttv = bsearch(&key,
1828 this->keyword_parsecodes_,
1829 this->keyword_count_,
1830 sizeof(this->keyword_parsecodes_[0]),
1834 Keyword_parsecode* ktt = static_cast<Keyword_parsecode*>(kttv);
1835 return ktt->parsecode;
1838 // The following structs are used within the VersionInfo class as well
1839 // as in the bison helper functions. They store the information
1840 // parsed from the version script.
1842 // A single version expression.
1843 // For example, pattern="std::map*" and language="C++".
1844 struct Version_expression
1846 Version_expression(const std::string& a_pattern,
1847 Version_script_info::Language a_language,
1849 : pattern(a_pattern), language(a_language), exact_match(a_exact_match),
1850 was_matched_by_symbol(false)
1853 std::string pattern;
1854 Version_script_info::Language language;
1855 // If false, we use glob() to match pattern. If true, we use strcmp().
1857 // True if --no-undefined-version is in effect and we found this
1858 // version in get_symbol_version. We use mutable because this
1859 // struct is generally not modifiable after it has been created.
1860 mutable bool was_matched_by_symbol;
1863 // A list of expressions.
1864 struct Version_expression_list
1866 std::vector<struct Version_expression> expressions;
1869 // A list of which versions upon which another version depends.
1870 // Strings should be from the Stringpool.
1871 struct Version_dependency_list
1873 std::vector<std::string> dependencies;
1876 // The total definition of a version. It includes the tag for the
1877 // version, its global and local expressions, and any dependencies.
1881 : tag(), global(NULL), local(NULL), dependencies(NULL)
1885 const struct Version_expression_list* global;
1886 const struct Version_expression_list* local;
1887 const struct Version_dependency_list* dependencies;
1890 // Helper class that calls cplus_demangle when needed and takes care of freeing
1893 class Lazy_demangler
1896 Lazy_demangler(const char* symbol, int options)
1897 : symbol_(symbol), options_(options), demangled_(NULL), did_demangle_(false)
1901 { free(this->demangled_); }
1903 // Return the demangled name. The actual demangling happens on the first call,
1904 // and the result is later cached.
1909 // The symbol to demangle.
1910 const char* symbol_;
1911 // Option flags to pass to cplus_demagle.
1913 // The cached demangled value, or NULL if demangling didn't happen yet or
1916 // Whether we already called cplus_demangle
1920 // Return the demangled name. The actual demangling happens on the first call,
1921 // and the result is later cached. Returns NULL if the symbol cannot be
1925 Lazy_demangler::get()
1927 if (!this->did_demangle_)
1929 this->demangled_ = cplus_demangle(this->symbol_, this->options_);
1930 this->did_demangle_ = true;
1932 return this->demangled_;
1935 // Class Version_script_info.
1937 Version_script_info::Version_script_info()
1938 : dependency_lists_(), expression_lists_(), version_trees_(), globs_(),
1939 default_version_(NULL), default_is_global_(false), is_finalized_(false)
1941 for (int i = 0; i < LANGUAGE_COUNT; ++i)
1942 this->exact_[i] = NULL;
1945 Version_script_info::~Version_script_info()
1949 // Forget all the known version script information.
1952 Version_script_info::clear()
1954 for (size_t k = 0; k < this->dependency_lists_.size(); ++k)
1955 delete this->dependency_lists_[k];
1956 this->dependency_lists_.clear();
1957 for (size_t k = 0; k < this->version_trees_.size(); ++k)
1958 delete this->version_trees_[k];
1959 this->version_trees_.clear();
1960 for (size_t k = 0; k < this->expression_lists_.size(); ++k)
1961 delete this->expression_lists_[k];
1962 this->expression_lists_.clear();
1965 // Finalize the version script information.
1968 Version_script_info::finalize()
1970 if (!this->is_finalized_)
1972 this->build_lookup_tables();
1973 this->is_finalized_ = true;
1977 // Return all the versions.
1979 std::vector<std::string>
1980 Version_script_info::get_versions() const
1982 std::vector<std::string> ret;
1983 for (size_t j = 0; j < this->version_trees_.size(); ++j)
1984 if (!this->version_trees_[j]->tag.empty())
1985 ret.push_back(this->version_trees_[j]->tag);
1989 // Return the dependencies of VERSION.
1991 std::vector<std::string>
1992 Version_script_info::get_dependencies(const char* version) const
1994 std::vector<std::string> ret;
1995 for (size_t j = 0; j < this->version_trees_.size(); ++j)
1996 if (this->version_trees_[j]->tag == version)
1998 const struct Version_dependency_list* deps =
1999 this->version_trees_[j]->dependencies;
2001 for (size_t k = 0; k < deps->dependencies.size(); ++k)
2002 ret.push_back(deps->dependencies[k]);
2008 // A version script essentially maps a symbol name to a version tag
2009 // and an indication of whether symbol is global or local within that
2010 // version tag. Each symbol maps to at most one version tag.
2011 // Unfortunately, in practice, version scripts are ambiguous, and list
2012 // symbols multiple times. Thus, we have to document the matching
2015 // This is a description of what the GNU linker does as of 2010-01-11.
2016 // It walks through the version tags in the order in which they appear
2017 // in the version script. For each tag, it first walks through the
2018 // global patterns for that tag, then the local patterns. When
2019 // looking at a single pattern, it first applies any language specific
2020 // demangling as specified for the pattern, and then matches the
2021 // resulting symbol name to the pattern. If it finds an exact match
2022 // for a literal pattern (a pattern enclosed in quotes or with no
2023 // wildcard characters), then that is the match that it uses. If
2024 // finds a match with a wildcard pattern, then it saves it and
2025 // continues searching. Wildcard patterns that are exactly "*" are
2026 // saved separately.
2028 // If no exact match with a literal pattern is ever found, then if a
2029 // wildcard match with a global pattern was found it is used,
2030 // otherwise if a wildcard match with a local pattern was found it is
2033 // This is the result:
2034 // * If there is an exact match, then we use the first tag in the
2035 // version script where it matches.
2036 // + If the exact match in that tag is global, it is used.
2037 // + Otherwise the exact match in that tag is local, and is used.
2038 // * Otherwise, if there is any match with a global wildcard pattern:
2039 // + If there is any match with a wildcard pattern which is not
2040 // "*", then we use the tag in which the *last* such pattern
2042 // + Otherwise, we matched "*". If there is no match with a local
2043 // wildcard pattern which is not "*", then we use the *last*
2044 // match with a global "*". Otherwise, continue.
2045 // * Otherwise, if there is any match with a local wildcard pattern:
2046 // + If there is any match with a wildcard pattern which is not
2047 // "*", then we use the tag in which the *last* such pattern
2049 // + Otherwise, we matched "*", and we use the tag in which the
2050 // *last* such match occurred.
2052 // There is an additional wrinkle. When the GNU linker finds a symbol
2053 // with a version defined in an object file due to a .symver
2054 // directive, it looks up that symbol name in that version tag. If it
2055 // finds it, it matches the symbol name against the patterns for that
2056 // version. If there is no match with a global pattern, but there is
2057 // a match with a local pattern, then the GNU linker marks the symbol
2060 // We want gold to be generally compatible, but we also want gold to
2061 // be fast. These are the rules that gold implements:
2062 // * If there is an exact match for the mangled name, we use it.
2063 // + If there is more than one exact match, we give a warning, and
2064 // we use the first tag in the script which matches.
2065 // + If a symbol has an exact match as both global and local for
2066 // the same version tag, we give an error.
2067 // * Otherwise, we look for an extern C++ or an extern Java exact
2068 // match. If we find an exact match, we use it.
2069 // + If there is more than one exact match, we give a warning, and
2070 // we use the first tag in the script which matches.
2071 // + If a symbol has an exact match as both global and local for
2072 // the same version tag, we give an error.
2073 // * Otherwise, we look through the wildcard patterns, ignoring "*"
2074 // patterns. We look through the version tags in reverse order.
2075 // For each version tag, we look through the global patterns and
2076 // then the local patterns. We use the first match we find (i.e.,
2077 // the last matching version tag in the file).
2078 // * Otherwise, we use the "*" pattern if there is one. We give an
2079 // error if there are multiple "*" patterns.
2081 // At least for now, gold does not look up the version tag for a
2082 // symbol version found in an object file to see if it should be
2083 // forced local. There are other ways to force a symbol to be local,
2084 // and I don't understand why this one is useful.
2086 // Build a set of fast lookup tables for a version script.
2089 Version_script_info::build_lookup_tables()
2091 size_t size = this->version_trees_.size();
2092 for (size_t j = 0; j < size; ++j)
2094 const Version_tree* v = this->version_trees_[j];
2095 this->build_expression_list_lookup(v->local, v, false);
2096 this->build_expression_list_lookup(v->global, v, true);
2100 // If a pattern has backlashes but no unquoted wildcard characters,
2101 // then we apply backslash unquoting and look for an exact match.
2102 // Otherwise we treat it as a wildcard pattern. This function returns
2103 // true for a wildcard pattern. Otherwise, it does backslash
2104 // unquoting on *PATTERN and returns false. If this returns true,
2105 // *PATTERN may have been partially unquoted.
2108 Version_script_info::unquote(std::string* pattern) const
2110 bool saw_backslash = false;
2111 size_t len = pattern->length();
2113 for (size_t i = 0; i < len; ++i)
2116 saw_backslash = false;
2119 switch ((*pattern)[i])
2121 case '?': case '[': case '*':
2124 saw_backslash = true;
2132 (*pattern)[j] = (*pattern)[i];
2138 // Add an exact match for MATCH to *PE. The result of the match is
2142 Version_script_info::add_exact_match(const std::string& match,
2143 const Version_tree* v, bool is_global,
2144 const Version_expression* ve,
2147 std::pair<Exact::iterator, bool> ins =
2148 pe->insert(std::make_pair(match, Version_tree_match(v, is_global, ve)));
2151 // This is the first time we have seen this match.
2155 Version_tree_match& vtm(ins.first->second);
2156 if (vtm.real->tag != v->tag)
2158 // This is an ambiguous match. We still return the
2159 // first version that we found in the script, but we
2160 // record the new version to issue a warning if we
2161 // wind up looking up this symbol.
2162 if (vtm.ambiguous == NULL)
2165 else if (is_global != vtm.is_global)
2167 // We have a match for both the global and local entries for a
2168 // version tag. That's got to be wrong.
2169 gold_error(_("'%s' appears as both a global and a local symbol "
2170 "for version '%s' in script"),
2171 match.c_str(), v->tag.c_str());
2175 // Build fast lookup information for EXPLIST and store it in LOOKUP.
2176 // All matches go to V, and IS_GLOBAL is true if they are global
2180 Version_script_info::build_expression_list_lookup(
2181 const Version_expression_list* explist,
2182 const Version_tree* v,
2185 if (explist == NULL)
2187 size_t size = explist->expressions.size();
2188 for (size_t i = 0; i < size; ++i)
2190 const Version_expression& exp(explist->expressions[i]);
2192 if (exp.pattern.length() == 1 && exp.pattern[0] == '*')
2194 if (this->default_version_ != NULL
2195 && this->default_version_->tag != v->tag)
2196 gold_warning(_("wildcard match appears in both version '%s' "
2197 "and '%s' in script"),
2198 this->default_version_->tag.c_str(), v->tag.c_str());
2199 else if (this->default_version_ != NULL
2200 && this->default_is_global_ != is_global)
2201 gold_error(_("wildcard match appears as both global and local "
2202 "in version '%s' in script"),
2204 this->default_version_ = v;
2205 this->default_is_global_ = is_global;
2209 std::string pattern = exp.pattern;
2210 if (!exp.exact_match)
2212 if (this->unquote(&pattern))
2214 this->globs_.push_back(Glob(&exp, v, is_global));
2219 if (this->exact_[exp.language] == NULL)
2220 this->exact_[exp.language] = new Exact();
2221 this->add_exact_match(pattern, v, is_global, &exp,
2222 this->exact_[exp.language]);
2226 // Return the name to match given a name, a language code, and two
2230 Version_script_info::get_name_to_match(const char* name,
2232 Lazy_demangler* cpp_demangler,
2233 Lazy_demangler* java_demangler) const
2240 return cpp_demangler->get();
2242 return java_demangler->get();
2248 // Look up SYMBOL_NAME in the list of versions. Return true if the
2249 // symbol is found, false if not. If the symbol is found, then if
2250 // PVERSION is not NULL, set *PVERSION to the version tag, and if
2251 // P_IS_GLOBAL is not NULL, set *P_IS_GLOBAL according to whether the
2252 // symbol is global or not.
2255 Version_script_info::get_symbol_version(const char* symbol_name,
2256 std::string* pversion,
2257 bool* p_is_global) const
2259 Lazy_demangler cpp_demangled_name(symbol_name, DMGL_ANSI | DMGL_PARAMS);
2260 Lazy_demangler java_demangled_name(symbol_name,
2261 DMGL_ANSI | DMGL_PARAMS | DMGL_JAVA);
2263 gold_assert(this->is_finalized_);
2264 for (int i = 0; i < LANGUAGE_COUNT; ++i)
2266 Exact* exact = this->exact_[i];
2270 const char* name_to_match = this->get_name_to_match(symbol_name, i,
2271 &cpp_demangled_name,
2272 &java_demangled_name);
2273 if (name_to_match == NULL)
2275 // If the name can not be demangled, the GNU linker goes
2276 // ahead and tries to match it anyhow. That does not
2277 // make sense to me and I have not implemented it.
2281 Exact::const_iterator pe = exact->find(name_to_match);
2282 if (pe != exact->end())
2284 const Version_tree_match& vtm(pe->second);
2285 if (vtm.ambiguous != NULL)
2286 gold_warning(_("using '%s' as version for '%s' which is also "
2287 "named in version '%s' in script"),
2288 vtm.real->tag.c_str(), name_to_match,
2289 vtm.ambiguous->tag.c_str());
2291 if (pversion != NULL)
2292 *pversion = vtm.real->tag;
2293 if (p_is_global != NULL)
2294 *p_is_global = vtm.is_global;
2296 // If we are using --no-undefined-version, and this is a
2297 // global symbol, we have to record that we have found this
2298 // symbol, so that we don't warn about it. We have to do
2299 // this now, because otherwise we have no way to get from a
2300 // non-C language back to the demangled name that we
2302 if (p_is_global != NULL && vtm.is_global)
2303 vtm.expression->was_matched_by_symbol = true;
2309 // Look through the glob patterns in reverse order.
2311 for (Globs::const_reverse_iterator p = this->globs_.rbegin();
2312 p != this->globs_.rend();
2315 int language = p->expression->language;
2316 const char* name_to_match = this->get_name_to_match(symbol_name,
2318 &cpp_demangled_name,
2319 &java_demangled_name);
2320 if (name_to_match == NULL)
2323 if (fnmatch(p->expression->pattern.c_str(), name_to_match,
2326 if (pversion != NULL)
2327 *pversion = p->version->tag;
2328 if (p_is_global != NULL)
2329 *p_is_global = p->is_global;
2334 // Finally, there may be a wildcard.
2335 if (this->default_version_ != NULL)
2337 if (pversion != NULL)
2338 *pversion = this->default_version_->tag;
2339 if (p_is_global != NULL)
2340 *p_is_global = this->default_is_global_;
2347 // Give an error if any exact symbol names (not wildcards) appear in a
2348 // version script, but there is no such symbol.
2351 Version_script_info::check_unmatched_names(const Symbol_table* symtab) const
2353 for (size_t i = 0; i < this->version_trees_.size(); ++i)
2355 const Version_tree* vt = this->version_trees_[i];
2356 if (vt->global == NULL)
2358 for (size_t j = 0; j < vt->global->expressions.size(); ++j)
2360 const Version_expression& expression(vt->global->expressions[j]);
2362 // Ignore cases where we used the version because we saw a
2363 // symbol that we looked up. Note that
2364 // WAS_MATCHED_BY_SYMBOL will be true even if the symbol was
2365 // not a definition. That's OK as in that case we most
2366 // likely gave an undefined symbol error anyhow.
2367 if (expression.was_matched_by_symbol)
2370 // Just ignore names which are in languages other than C.
2371 // We have no way to look them up in the symbol table.
2372 if (expression.language != LANGUAGE_C)
2375 // Remove backslash quoting, and ignore wildcard patterns.
2376 std::string pattern = expression.pattern;
2377 if (!expression.exact_match)
2379 if (this->unquote(&pattern))
2383 if (symtab->lookup(pattern.c_str(), vt->tag.c_str()) == NULL)
2384 gold_error(_("version script assignment of %s to symbol %s "
2385 "failed: symbol not defined"),
2386 vt->tag.c_str(), pattern.c_str());
2391 struct Version_dependency_list*
2392 Version_script_info::allocate_dependency_list()
2394 dependency_lists_.push_back(new Version_dependency_list);
2395 return dependency_lists_.back();
2398 struct Version_expression_list*
2399 Version_script_info::allocate_expression_list()
2401 expression_lists_.push_back(new Version_expression_list);
2402 return expression_lists_.back();
2405 struct Version_tree*
2406 Version_script_info::allocate_version_tree()
2408 version_trees_.push_back(new Version_tree);
2409 return version_trees_.back();
2412 // Print for debugging.
2415 Version_script_info::print(FILE* f) const
2420 fprintf(f, "VERSION {");
2422 for (size_t i = 0; i < this->version_trees_.size(); ++i)
2424 const Version_tree* vt = this->version_trees_[i];
2426 if (vt->tag.empty())
2429 fprintf(f, " %s {\n", vt->tag.c_str());
2431 if (vt->global != NULL)
2433 fprintf(f, " global :\n");
2434 this->print_expression_list(f, vt->global);
2437 if (vt->local != NULL)
2439 fprintf(f, " local :\n");
2440 this->print_expression_list(f, vt->local);
2444 if (vt->dependencies != NULL)
2446 const Version_dependency_list* deps = vt->dependencies;
2447 for (size_t j = 0; j < deps->dependencies.size(); ++j)
2449 if (j < deps->dependencies.size() - 1)
2451 fprintf(f, " %s", deps->dependencies[j].c_str());
2461 Version_script_info::print_expression_list(
2463 const Version_expression_list* vel) const
2465 Version_script_info::Language current_language = LANGUAGE_C;
2466 for (size_t i = 0; i < vel->expressions.size(); ++i)
2468 const Version_expression& ve(vel->expressions[i]);
2470 if (ve.language != current_language)
2472 if (current_language != LANGUAGE_C)
2474 switch (ve.language)
2479 fprintf(f, " extern \"C++\" {\n");
2482 fprintf(f, " extern \"Java\" {\n");
2487 current_language = ve.language;
2491 if (current_language != LANGUAGE_C)
2496 fprintf(f, "%s", ve.pattern.c_str());
2503 if (current_language != LANGUAGE_C)
2507 } // End namespace gold.
2509 // The remaining functions are extern "C", so it's clearer to not put
2510 // them in namespace gold.
2512 using namespace gold;
2514 // This function is called by the bison parser to return the next
2518 yylex(YYSTYPE* lvalp, void* closurev)
2520 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2521 const Token* token = closure->next_token();
2522 switch (token->classification())
2527 case Token::TOKEN_INVALID:
2528 yyerror(closurev, "invalid character");
2531 case Token::TOKEN_EOF:
2534 case Token::TOKEN_STRING:
2536 // This is either a keyword or a STRING.
2538 const char* str = token->string_value(&len);
2540 switch (closure->lex_mode())
2542 case Lex::LINKER_SCRIPT:
2543 parsecode = script_keywords.keyword_to_parsecode(str, len);
2545 case Lex::VERSION_SCRIPT:
2546 parsecode = version_script_keywords.keyword_to_parsecode(str, len);
2548 case Lex::DYNAMIC_LIST:
2549 parsecode = dynamic_list_keywords.keyword_to_parsecode(str, len);
2556 lvalp->string.value = str;
2557 lvalp->string.length = len;
2561 case Token::TOKEN_QUOTED_STRING:
2562 lvalp->string.value = token->string_value(&lvalp->string.length);
2563 return QUOTED_STRING;
2565 case Token::TOKEN_OPERATOR:
2566 return token->operator_value();
2568 case Token::TOKEN_INTEGER:
2569 lvalp->integer = token->integer_value();
2574 // This function is called by the bison parser to report an error.
2577 yyerror(void* closurev, const char* message)
2579 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2580 gold_error(_("%s:%d:%d: %s"), closure->filename(), closure->lineno(),
2581 closure->charpos(), message);
2584 // Called by the bison parser to add an external symbol to the link.
2587 script_add_extern(void* closurev, const char* name, size_t length)
2589 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2590 closure->script_options()->add_symbol_reference(name, length);
2593 // Called by the bison parser to add a file to the link.
2596 script_add_file(void* closurev, const char* name, size_t length)
2598 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2600 // If this is an absolute path, and we found the script in the
2601 // sysroot, then we want to prepend the sysroot to the file name.
2602 // For example, this is how we handle a cross link to the x86_64
2603 // libc.so, which refers to /lib/libc.so.6.
2604 std::string name_string(name, length);
2605 const char* extra_search_path = ".";
2606 std::string script_directory;
2607 if (IS_ABSOLUTE_PATH(name_string.c_str()))
2609 if (closure->is_in_sysroot())
2611 const std::string& sysroot(parameters->options().sysroot());
2612 gold_assert(!sysroot.empty());
2613 name_string = sysroot + name_string;
2618 // In addition to checking the normal library search path, we
2619 // also want to check in the script-directory.
2620 const char* slash = strrchr(closure->filename(), '/');
2623 script_directory.assign(closure->filename(),
2624 slash - closure->filename() + 1);
2625 extra_search_path = script_directory.c_str();
2629 Input_file_argument file(name_string.c_str(),
2630 Input_file_argument::INPUT_FILE_TYPE_FILE,
2631 extra_search_path, false,
2632 closure->position_dependent_options());
2633 Input_argument& arg = closure->inputs()->add_file(file);
2634 arg.set_script_info(closure->script_info());
2637 // Called by the bison parser to add a library to the link.
2640 script_add_library(void* closurev, const char* name, size_t length)
2642 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2643 std::string name_string(name, length);
2645 if (name_string[0] != 'l')
2646 gold_error(_("library name must be prefixed with -l"));
2648 Input_file_argument file(name_string.c_str() + 1,
2649 Input_file_argument::INPUT_FILE_TYPE_LIBRARY,
2651 closure->position_dependent_options());
2652 Input_argument& arg = closure->inputs()->add_file(file);
2653 arg.set_script_info(closure->script_info());
2656 // Called by the bison parser to start a group. If we are already in
2657 // a group, that means that this script was invoked within a
2658 // --start-group --end-group sequence on the command line, or that
2659 // this script was found in a GROUP of another script. In that case,
2660 // we simply continue the existing group, rather than starting a new
2661 // one. It is possible to construct a case in which this will do
2662 // something other than what would happen if we did a recursive group,
2663 // but it's hard to imagine why the different behaviour would be
2664 // useful for a real program. Avoiding recursive groups is simpler
2665 // and more efficient.
2668 script_start_group(void* closurev)
2670 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2671 if (!closure->in_group())
2672 closure->inputs()->start_group();
2675 // Called by the bison parser at the end of a group.
2678 script_end_group(void* closurev)
2680 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2681 if (!closure->in_group())
2682 closure->inputs()->end_group();
2685 // Called by the bison parser to start an AS_NEEDED list.
2688 script_start_as_needed(void* closurev)
2690 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2691 closure->position_dependent_options().set_as_needed(true);
2694 // Called by the bison parser at the end of an AS_NEEDED list.
2697 script_end_as_needed(void* closurev)
2699 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2700 closure->position_dependent_options().set_as_needed(false);
2703 // Called by the bison parser to set the entry symbol.
2706 script_set_entry(void* closurev, const char* entry, size_t length)
2708 // We'll parse this exactly the same as --entry=ENTRY on the commandline
2709 // TODO(csilvers): FIXME -- call set_entry directly.
2710 std::string arg("--entry=");
2711 arg.append(entry, length);
2712 script_parse_option(closurev, arg.c_str(), arg.size());
2715 // Called by the bison parser to set whether to define common symbols.
2718 script_set_common_allocation(void* closurev, int set)
2720 const char* arg = set != 0 ? "--define-common" : "--no-define-common";
2721 script_parse_option(closurev, arg, strlen(arg));
2724 // Called by the bison parser to refer to a symbol.
2726 extern "C" Expression*
2727 script_symbol(void* closurev, const char* name, size_t length)
2729 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2730 if (length != 1 || name[0] != '.')
2731 closure->script_options()->add_symbol_reference(name, length);
2732 return script_exp_string(name, length);
2735 // Called by the bison parser to define a symbol.
2738 script_set_symbol(void* closurev, const char* name, size_t length,
2739 Expression* value, int providei, int hiddeni)
2741 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2742 const bool provide = providei != 0;
2743 const bool hidden = hiddeni != 0;
2744 closure->script_options()->add_symbol_assignment(name, length,
2745 closure->parsing_defsym(),
2746 value, provide, hidden);
2747 closure->clear_skip_on_incompatible_target();
2750 // Called by the bison parser to add an assertion.
2753 script_add_assertion(void* closurev, Expression* check, const char* message,
2756 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2757 closure->script_options()->add_assertion(check, message, messagelen);
2758 closure->clear_skip_on_incompatible_target();
2761 // Called by the bison parser to parse an OPTION.
2764 script_parse_option(void* closurev, const char* option, size_t length)
2766 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2767 // We treat the option as a single command-line option, even if
2768 // it has internal whitespace.
2769 if (closure->command_line() == NULL)
2771 // There are some options that we could handle here--e.g.,
2772 // -lLIBRARY. Should we bother?
2773 gold_warning(_("%s:%d:%d: ignoring command OPTION; OPTION is only valid"
2774 " for scripts specified via -T/--script"),
2775 closure->filename(), closure->lineno(), closure->charpos());
2779 bool past_a_double_dash_option = false;
2780 const char* mutable_option = strndup(option, length);
2781 gold_assert(mutable_option != NULL);
2782 closure->command_line()->process_one_option(1, &mutable_option, 0,
2783 &past_a_double_dash_option);
2784 // The General_options class will quite possibly store a pointer
2785 // into mutable_option, so we can't free it. In cases the class
2786 // does not store such a pointer, this is a memory leak. Alas. :(
2788 closure->clear_skip_on_incompatible_target();
2791 // Called by the bison parser to handle OUTPUT_FORMAT. OUTPUT_FORMAT
2792 // takes either one or three arguments. In the three argument case,
2793 // the format depends on the endianness option, which we don't
2794 // currently support (FIXME). If we see an OUTPUT_FORMAT for the
2795 // wrong format, then we want to search for a new file. Returning 0
2796 // here will cause the parser to immediately abort.
2799 script_check_output_format(void* closurev,
2800 const char* default_name, size_t default_length,
2801 const char*, size_t, const char*, size_t)
2803 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2804 std::string name(default_name, default_length);
2805 Target* target = select_target_by_name(name.c_str());
2806 if (target == NULL || !parameters->is_compatible_target(target))
2808 if (closure->skip_on_incompatible_target())
2810 closure->set_found_incompatible_target();
2813 // FIXME: Should we warn about the unknown target?
2818 // Called by the bison parser to handle TARGET.
2821 script_set_target(void* closurev, const char* target, size_t len)
2823 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2824 std::string s(target, len);
2825 General_options::Object_format format_enum;
2826 format_enum = General_options::string_to_object_format(s.c_str());
2827 closure->position_dependent_options().set_format_enum(format_enum);
2830 // Called by the bison parser to handle SEARCH_DIR. This is handled
2831 // exactly like a -L option.
2834 script_add_search_dir(void* closurev, const char* option, size_t length)
2836 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2837 if (closure->command_line() == NULL)
2838 gold_warning(_("%s:%d:%d: ignoring SEARCH_DIR; SEARCH_DIR is only valid"
2839 " for scripts specified via -T/--script"),
2840 closure->filename(), closure->lineno(), closure->charpos());
2841 else if (!closure->command_line()->options().nostdlib())
2843 std::string s = "-L" + std::string(option, length);
2844 script_parse_option(closurev, s.c_str(), s.size());
2848 /* Called by the bison parser to push the lexer into expression
2852 script_push_lex_into_expression_mode(void* closurev)
2854 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2855 closure->push_lex_mode(Lex::EXPRESSION);
2858 /* Called by the bison parser to push the lexer into version
2862 script_push_lex_into_version_mode(void* closurev)
2864 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2865 if (closure->version_script()->is_finalized())
2866 gold_error(_("%s:%d:%d: invalid use of VERSION in input file"),
2867 closure->filename(), closure->lineno(), closure->charpos());
2868 closure->push_lex_mode(Lex::VERSION_SCRIPT);
2871 /* Called by the bison parser to pop the lexer mode. */
2874 script_pop_lex_mode(void* closurev)
2876 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2877 closure->pop_lex_mode();
2880 // Register an entire version node. For example:
2886 // - tag is "GLIBC_2.1"
2887 // - tree contains the information "global: foo"
2888 // - deps contains "GLIBC_2.0"
2891 script_register_vers_node(void*,
2894 struct Version_tree* tree,
2895 struct Version_dependency_list* deps)
2897 gold_assert(tree != NULL);
2898 tree->dependencies = deps;
2900 tree->tag = std::string(tag, taglen);
2903 // Add a dependencies to the list of existing dependencies, if any,
2904 // and return the expanded list.
2906 extern "C" struct Version_dependency_list*
2907 script_add_vers_depend(void* closurev,
2908 struct Version_dependency_list* all_deps,
2909 const char* depend_to_add, int deplen)
2911 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2912 if (all_deps == NULL)
2913 all_deps = closure->version_script()->allocate_dependency_list();
2914 all_deps->dependencies.push_back(std::string(depend_to_add, deplen));
2918 // Add a pattern expression to an existing list of expressions, if any.
2920 extern "C" struct Version_expression_list*
2921 script_new_vers_pattern(void* closurev,
2922 struct Version_expression_list* expressions,
2923 const char* pattern, int patlen, int exact_match)
2925 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2926 if (expressions == NULL)
2927 expressions = closure->version_script()->allocate_expression_list();
2928 expressions->expressions.push_back(
2929 Version_expression(std::string(pattern, patlen),
2930 closure->get_current_language(),
2931 static_cast<bool>(exact_match)));
2935 // Attaches b to the end of a, and clears b. So a = a + b and b = {}.
2937 extern "C" struct Version_expression_list*
2938 script_merge_expressions(struct Version_expression_list* a,
2939 struct Version_expression_list* b)
2941 a->expressions.insert(a->expressions.end(),
2942 b->expressions.begin(), b->expressions.end());
2943 // We could delete b and remove it from expressions_lists_, but
2944 // that's a lot of work. This works just as well.
2945 b->expressions.clear();
2949 // Combine the global and local expressions into a a Version_tree.
2951 extern "C" struct Version_tree*
2952 script_new_vers_node(void* closurev,
2953 struct Version_expression_list* global,
2954 struct Version_expression_list* local)
2956 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2957 Version_tree* tree = closure->version_script()->allocate_version_tree();
2958 tree->global = global;
2959 tree->local = local;
2963 // Handle a transition in language, such as at the
2964 // start or end of 'extern "C++"'
2967 version_script_push_lang(void* closurev, const char* lang, int langlen)
2969 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2970 std::string language(lang, langlen);
2971 Version_script_info::Language code;
2972 if (language.empty() || language == "C")
2973 code = Version_script_info::LANGUAGE_C;
2974 else if (language == "C++")
2975 code = Version_script_info::LANGUAGE_CXX;
2976 else if (language == "Java")
2977 code = Version_script_info::LANGUAGE_JAVA;
2980 char* buf = new char[langlen + 100];
2981 snprintf(buf, langlen + 100,
2982 _("unrecognized version script language '%s'"),
2984 yyerror(closurev, buf);
2986 code = Version_script_info::LANGUAGE_C;
2988 closure->push_language(code);
2992 version_script_pop_lang(void* closurev)
2994 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2995 closure->pop_language();
2998 // Called by the bison parser to start a SECTIONS clause.
3001 script_start_sections(void* closurev)
3003 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3004 closure->script_options()->script_sections()->start_sections();
3005 closure->clear_skip_on_incompatible_target();
3008 // Called by the bison parser to finish a SECTIONS clause.
3011 script_finish_sections(void* closurev)
3013 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3014 closure->script_options()->script_sections()->finish_sections();
3017 // Start processing entries for an output section.
3020 script_start_output_section(void* closurev, const char* name, size_t namelen,
3021 const struct Parser_output_section_header* header)
3023 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3024 closure->script_options()->script_sections()->start_output_section(name,
3029 // Finish processing entries for an output section.
3032 script_finish_output_section(void* closurev,
3033 const struct Parser_output_section_trailer* trail)
3035 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3036 closure->script_options()->script_sections()->finish_output_section(trail);
3039 // Add a data item (e.g., "WORD (0)") to the current output section.
3042 script_add_data(void* closurev, int data_token, Expression* val)
3044 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3046 bool is_signed = true;
3068 closure->script_options()->script_sections()->add_data(size, is_signed, val);
3071 // Add a clause setting the fill value to the current output section.
3074 script_add_fill(void* closurev, Expression* val)
3076 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3077 closure->script_options()->script_sections()->add_fill(val);
3080 // Add a new input section specification to the current output
3084 script_add_input_section(void* closurev,
3085 const struct Input_section_spec* spec,
3088 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3089 bool keep = keepi != 0;
3090 closure->script_options()->script_sections()->add_input_section(spec, keep);
3093 // When we see DATA_SEGMENT_ALIGN we record that following output
3094 // sections may be relro.
3097 script_data_segment_align(void* closurev)
3099 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3100 if (!closure->script_options()->saw_sections_clause())
3101 gold_error(_("%s:%d:%d: DATA_SEGMENT_ALIGN not in SECTIONS clause"),
3102 closure->filename(), closure->lineno(), closure->charpos());
3104 closure->script_options()->script_sections()->data_segment_align();
3107 // When we see DATA_SEGMENT_RELRO_END we know that all output sections
3108 // since DATA_SEGMENT_ALIGN should be relro.
3111 script_data_segment_relro_end(void* closurev)
3113 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3114 if (!closure->script_options()->saw_sections_clause())
3115 gold_error(_("%s:%d:%d: DATA_SEGMENT_ALIGN not in SECTIONS clause"),
3116 closure->filename(), closure->lineno(), closure->charpos());
3118 closure->script_options()->script_sections()->data_segment_relro_end();
3121 // Create a new list of string/sort pairs.
3123 extern "C" String_sort_list_ptr
3124 script_new_string_sort_list(const struct Wildcard_section* string_sort)
3126 return new String_sort_list(1, *string_sort);
3129 // Add an entry to a list of string/sort pairs. The way the parser
3130 // works permits us to simply modify the first parameter, rather than
3133 extern "C" String_sort_list_ptr
3134 script_string_sort_list_add(String_sort_list_ptr pv,
3135 const struct Wildcard_section* string_sort)
3138 return script_new_string_sort_list(string_sort);
3141 pv->push_back(*string_sort);
3146 // Create a new list of strings.
3148 extern "C" String_list_ptr
3149 script_new_string_list(const char* str, size_t len)
3151 return new String_list(1, std::string(str, len));
3154 // Add an element to a list of strings. The way the parser works
3155 // permits us to simply modify the first parameter, rather than copy
3158 extern "C" String_list_ptr
3159 script_string_list_push_back(String_list_ptr pv, const char* str, size_t len)
3162 return script_new_string_list(str, len);
3165 pv->push_back(std::string(str, len));
3170 // Concatenate two string lists. Either or both may be NULL. The way
3171 // the parser works permits us to modify the parameters, rather than
3174 extern "C" String_list_ptr
3175 script_string_list_append(String_list_ptr pv1, String_list_ptr pv2)
3181 pv1->insert(pv1->end(), pv2->begin(), pv2->end());
3185 // Add a new program header.
3188 script_add_phdr(void* closurev, const char* name, size_t namelen,
3189 unsigned int type, const Phdr_info* info)
3191 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3192 bool includes_filehdr = info->includes_filehdr != 0;
3193 bool includes_phdrs = info->includes_phdrs != 0;
3194 bool is_flags_valid = info->is_flags_valid != 0;
3195 Script_sections* ss = closure->script_options()->script_sections();
3196 ss->add_phdr(name, namelen, type, includes_filehdr, includes_phdrs,
3197 is_flags_valid, info->flags, info->load_address);
3198 closure->clear_skip_on_incompatible_target();
3201 // Convert a program header string to a type.
3203 #define PHDR_TYPE(NAME) { #NAME, sizeof(#NAME) - 1, elfcpp::NAME }
3210 } phdr_type_names[] =
3214 PHDR_TYPE(PT_DYNAMIC),
3215 PHDR_TYPE(PT_INTERP),
3217 PHDR_TYPE(PT_SHLIB),
3220 PHDR_TYPE(PT_GNU_EH_FRAME),
3221 PHDR_TYPE(PT_GNU_STACK),
3222 PHDR_TYPE(PT_GNU_RELRO)
3225 extern "C" unsigned int
3226 script_phdr_string_to_type(void* closurev, const char* name, size_t namelen)
3228 for (unsigned int i = 0;
3229 i < sizeof(phdr_type_names) / sizeof(phdr_type_names[0]);
3231 if (namelen == phdr_type_names[i].namelen
3232 && strncmp(name, phdr_type_names[i].name, namelen) == 0)
3233 return phdr_type_names[i].val;
3234 yyerror(closurev, _("unknown PHDR type (try integer)"));
3235 return elfcpp::PT_NULL;
3239 script_saw_segment_start_expression(void* closurev)
3241 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3242 Script_sections* ss = closure->script_options()->script_sections();
3243 ss->set_saw_segment_start_expression(true);
3247 script_set_section_region(void* closurev, const char* name, size_t namelen,
3250 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3251 if (!closure->script_options()->saw_sections_clause())
3253 gold_error(_("%s:%d:%d: MEMORY region '%.*s' referred to outside of "
3255 closure->filename(), closure->lineno(), closure->charpos(),
3256 static_cast<int>(namelen), name);
3260 Script_sections* ss = closure->script_options()->script_sections();
3261 Memory_region* mr = ss->find_memory_region(name, namelen);
3264 gold_error(_("%s:%d:%d: MEMORY region '%.*s' not declared"),
3265 closure->filename(), closure->lineno(), closure->charpos(),
3266 static_cast<int>(namelen), name);
3270 ss->set_memory_region(mr, set_vma);
3274 script_add_memory(void* closurev, const char* name, size_t namelen,
3275 unsigned int attrs, Expression* origin, Expression* length)
3277 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3278 Script_sections* ss = closure->script_options()->script_sections();
3279 ss->add_memory_region(name, namelen, attrs, origin, length);
3282 extern "C" unsigned int
3283 script_parse_memory_attr(void* closurev, const char* attrs, size_t attrlen,
3293 attributes |= MEM_READABLE; break;
3296 attributes |= MEM_READABLE | MEM_WRITEABLE; break;
3299 attributes |= MEM_EXECUTABLE; break;
3302 attributes |= MEM_ALLOCATABLE; break;
3307 attributes |= MEM_INITIALIZED; break;
3309 yyerror(closurev, _("unknown MEMORY attribute"));
3313 attributes = (~ attributes) & MEM_ATTR_MASK;
3319 script_include_directive(void* closurev, const char*, size_t)
3321 // FIXME: Implement ?
3322 yyerror (closurev, _("GOLD does not currently support INCLUDE directives"));
3325 // Functions for memory regions.
3327 extern "C" Expression*
3328 script_exp_function_origin(void* closurev, const char* name, size_t namelen)
3330 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3331 Script_sections* ss = closure->script_options()->script_sections();
3332 Expression* origin = ss->find_memory_region_origin(name, namelen);
3336 gold_error(_("undefined memory region '%s' referenced "
3337 "in ORIGIN expression"),
3339 // Create a dummy expression to prevent crashes later on.
3340 origin = script_exp_integer(0);
3346 extern "C" Expression*
3347 script_exp_function_length(void* closurev, const char* name, size_t namelen)
3349 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3350 Script_sections* ss = closure->script_options()->script_sections();
3351 Expression* length = ss->find_memory_region_length(name, namelen);
3355 gold_error(_("undefined memory region '%s' referenced "
3356 "in LENGTH expression"),
3358 // Create a dummy expression to prevent crashes later on.
3359 length = script_exp_integer(0);