1 // script.cc -- handle linker scripts for gold.
3 // Copyright (C) 2006-2014 Free Software Foundation, Inc.
4 // Written by Ian Lance Taylor <iant@google.com>.
6 // This file is part of gold.
8 // This program is free software; you can redistribute it and/or modify
9 // it under the terms of the GNU General Public License as published by
10 // the Free Software Foundation; either version 3 of the License, or
11 // (at your option) any later version.
13 // This program is distributed in the hope that it will be useful,
14 // but WITHOUT ANY WARRANTY; without even the implied warranty of
15 // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 // GNU General Public License for more details.
18 // You should have received a copy of the GNU General Public License
19 // along with this program; if not, write to the Free Software
20 // Foundation, Inc., 51 Franklin Street - Fifth Floor, Boston,
21 // MA 02110-1301, USA.
31 #include "filenames.h"
35 #include "dirsearch.h"
38 #include "workqueue.h"
40 #include "parameters.h"
43 #include "target-select.h"
46 #include "incremental.h"
51 // A token read from a script file. We don't implement keywords here;
52 // all keywords are simply represented as a string.
57 // Token classification.
62 // Token indicates end of input.
64 // Token is a string of characters.
66 // Token is a quoted string of characters.
68 // Token is an operator.
70 // Token is a number (an integer).
74 // We need an empty constructor so that we can put this STL objects.
76 : classification_(TOKEN_INVALID), value_(NULL), value_length_(0),
77 opcode_(0), lineno_(0), charpos_(0)
80 // A general token with no value.
81 Token(Classification classification, int lineno, int charpos)
82 : classification_(classification), value_(NULL), value_length_(0),
83 opcode_(0), lineno_(lineno), charpos_(charpos)
85 gold_assert(classification == TOKEN_INVALID
86 || classification == TOKEN_EOF);
89 // A general token with a value.
90 Token(Classification classification, const char* value, size_t length,
91 int lineno, int charpos)
92 : classification_(classification), value_(value), value_length_(length),
93 opcode_(0), lineno_(lineno), charpos_(charpos)
95 gold_assert(classification != TOKEN_INVALID
96 && classification != TOKEN_EOF);
99 // A token representing an operator.
100 Token(int opcode, int lineno, int charpos)
101 : classification_(TOKEN_OPERATOR), value_(NULL), value_length_(0),
102 opcode_(opcode), lineno_(lineno), charpos_(charpos)
105 // Return whether the token is invalid.
108 { return this->classification_ == TOKEN_INVALID; }
110 // Return whether this is an EOF token.
113 { return this->classification_ == TOKEN_EOF; }
115 // Return the token classification.
117 classification() const
118 { return this->classification_; }
120 // Return the line number at which the token starts.
123 { return this->lineno_; }
125 // Return the character position at this the token starts.
128 { return this->charpos_; }
130 // Get the value of a token.
133 string_value(size_t* length) const
135 gold_assert(this->classification_ == TOKEN_STRING
136 || this->classification_ == TOKEN_QUOTED_STRING);
137 *length = this->value_length_;
142 operator_value() const
144 gold_assert(this->classification_ == TOKEN_OPERATOR);
145 return this->opcode_;
149 integer_value() const;
152 // The token classification.
153 Classification classification_;
154 // The token value, for TOKEN_STRING or TOKEN_QUOTED_STRING or
157 // The length of the token value.
158 size_t value_length_;
159 // The token value, for TOKEN_OPERATOR.
161 // The line number where this token started (one based).
163 // The character position within the line where this token started
168 // Return the value of a TOKEN_INTEGER.
171 Token::integer_value() const
173 gold_assert(this->classification_ == TOKEN_INTEGER);
175 size_t len = this->value_length_;
177 uint64_t multiplier = 1;
178 char last = this->value_[len - 1];
179 if (last == 'm' || last == 'M')
181 multiplier = 1024 * 1024;
184 else if (last == 'k' || last == 'K')
191 uint64_t ret = strtoull(this->value_, &end, 0);
192 gold_assert(static_cast<size_t>(end - this->value_) == len);
194 return ret * multiplier;
197 // This class handles lexing a file into a sequence of tokens.
202 // We unfortunately have to support different lexing modes, because
203 // when reading different parts of a linker script we need to parse
204 // things differently.
207 // Reading an ordinary linker script.
209 // Reading an expression in a linker script.
211 // Reading a version script.
213 // Reading a --dynamic-list file.
217 Lex(const char* input_string, size_t input_length, int parsing_token)
218 : input_string_(input_string), input_length_(input_length),
219 current_(input_string), mode_(LINKER_SCRIPT),
220 first_token_(parsing_token), token_(),
221 lineno_(1), linestart_(input_string)
224 // Read a file into a string.
226 read_file(Input_file*, std::string*);
228 // Return the next token.
232 // Return the current lexing mode.
235 { return this->mode_; }
237 // Set the lexing mode.
240 { this->mode_ = mode; }
244 Lex& operator=(const Lex&);
246 // Make a general token with no value at the current location.
248 make_token(Token::Classification c, const char* start) const
249 { return Token(c, this->lineno_, start - this->linestart_ + 1); }
251 // Make a general token with a value at the current location.
253 make_token(Token::Classification c, const char* v, size_t len,
256 { return Token(c, v, len, this->lineno_, start - this->linestart_ + 1); }
258 // Make an operator token at the current location.
260 make_token(int opcode, const char* start) const
261 { return Token(opcode, this->lineno_, start - this->linestart_ + 1); }
263 // Make an invalid token at the current location.
265 make_invalid_token(const char* start)
266 { return this->make_token(Token::TOKEN_INVALID, start); }
268 // Make an EOF token at the current location.
270 make_eof_token(const char* start)
271 { return this->make_token(Token::TOKEN_EOF, start); }
273 // Return whether C can be the first character in a name. C2 is the
274 // next character, since we sometimes need that.
276 can_start_name(char c, char c2);
278 // If C can appear in a name which has already started, return a
279 // pointer to a character later in the token or just past
280 // it. Otherwise, return NULL.
282 can_continue_name(const char* c);
284 // Return whether C, C2, C3 can start a hex number.
286 can_start_hex(char c, char c2, char c3);
288 // If C can appear in a hex number which has already started, return
289 // a pointer to a character later in the token or just past
290 // it. Otherwise, return NULL.
292 can_continue_hex(const char* c);
294 // Return whether C can start a non-hex number.
296 can_start_number(char c);
298 // If C can appear in a decimal number which has already started,
299 // return a pointer to a character later in the token or just past
300 // it. Otherwise, return NULL.
302 can_continue_number(const char* c)
303 { return Lex::can_start_number(*c) ? c + 1 : NULL; }
305 // If C1 C2 C3 form a valid three character operator, return the
306 // opcode. Otherwise return 0.
308 three_char_operator(char c1, char c2, char c3);
310 // If C1 C2 form a valid two character operator, return the opcode.
311 // Otherwise return 0.
313 two_char_operator(char c1, char c2);
315 // If C1 is a valid one character operator, return the opcode.
316 // Otherwise return 0.
318 one_char_operator(char c1);
320 // Read the next token.
322 get_token(const char**);
324 // Skip a C style /* */ comment. Return false if the comment did
327 skip_c_comment(const char**);
329 // Skip a line # comment. Return false if there was no newline.
331 skip_line_comment(const char**);
333 // Build a token CLASSIFICATION from all characters that match
334 // CAN_CONTINUE_FN. The token starts at START. Start matching from
335 // MATCH. Set *PP to the character following the token.
337 gather_token(Token::Classification,
338 const char* (Lex::*can_continue_fn)(const char*),
339 const char* start, const char* match, const char** pp);
341 // Build a token from a quoted string.
343 gather_quoted_string(const char** pp);
345 // The string we are tokenizing.
346 const char* input_string_;
347 // The length of the string.
348 size_t input_length_;
349 // The current offset into the string.
350 const char* current_;
351 // The current lexing mode.
353 // The code to use for the first token. This is set to 0 after it
356 // The current token.
358 // The current line number.
360 // The start of the current line in the string.
361 const char* linestart_;
364 // Read the whole file into memory. We don't expect linker scripts to
365 // be large, so we just use a std::string as a buffer. We ignore the
366 // data we've already read, so that we read aligned buffers.
369 Lex::read_file(Input_file* input_file, std::string* contents)
371 off_t filesize = input_file->file().filesize();
373 contents->reserve(filesize);
376 unsigned char buf[BUFSIZ];
377 while (off < filesize)
380 if (get > filesize - off)
381 get = filesize - off;
382 input_file->file().read(off, get, buf);
383 contents->append(reinterpret_cast<char*>(&buf[0]), get);
388 // Return whether C can be the start of a name, if the next character
389 // is C2. A name can being with a letter, underscore, period, or
390 // dollar sign. Because a name can be a file name, we also permit
391 // forward slash, backslash, and tilde. Tilde is the tricky case
392 // here; GNU ld also uses it as a bitwise not operator. It is only
393 // recognized as the operator if it is not immediately followed by
394 // some character which can appear in a symbol. That is, when we
395 // don't know that we are looking at an expression, "~0" is a file
396 // name, and "~ 0" is an expression using bitwise not. We are
400 Lex::can_start_name(char c, char c2)
404 case 'A': case 'B': case 'C': case 'D': case 'E': case 'F':
405 case 'G': case 'H': case 'I': case 'J': case 'K': case 'L':
406 case 'M': case 'N': case 'O': case 'Q': case 'P': case 'R':
407 case 'S': case 'T': case 'U': case 'V': case 'W': case 'X':
409 case 'a': case 'b': case 'c': case 'd': case 'e': case 'f':
410 case 'g': case 'h': case 'i': case 'j': case 'k': case 'l':
411 case 'm': case 'n': case 'o': case 'q': case 'p': case 'r':
412 case 's': case 't': case 'u': case 'v': case 'w': case 'x':
414 case '_': case '.': case '$':
418 return this->mode_ == LINKER_SCRIPT;
421 return this->mode_ == LINKER_SCRIPT && can_continue_name(&c2);
424 return (this->mode_ == VERSION_SCRIPT
425 || this->mode_ == DYNAMIC_LIST
426 || (this->mode_ == LINKER_SCRIPT
427 && can_continue_name(&c2)));
434 // Return whether C can continue a name which has already started.
435 // Subsequent characters in a name are the same as the leading
436 // characters, plus digits and "=+-:[],?*". So in general the linker
437 // script language requires spaces around operators, unless we know
438 // that we are parsing an expression.
441 Lex::can_continue_name(const char* c)
445 case 'A': case 'B': case 'C': case 'D': case 'E': case 'F':
446 case 'G': case 'H': case 'I': case 'J': case 'K': case 'L':
447 case 'M': case 'N': case 'O': case 'Q': case 'P': case 'R':
448 case 'S': case 'T': case 'U': case 'V': case 'W': case 'X':
450 case 'a': case 'b': case 'c': case 'd': case 'e': case 'f':
451 case 'g': case 'h': case 'i': case 'j': case 'k': case 'l':
452 case 'm': case 'n': case 'o': case 'q': case 'p': case 'r':
453 case 's': case 't': case 'u': case 'v': case 'w': case 'x':
455 case '_': case '.': case '$':
456 case '0': case '1': case '2': case '3': case '4':
457 case '5': case '6': case '7': case '8': case '9':
460 // TODO(csilvers): why not allow ~ in names for version-scripts?
461 case '/': case '\\': case '~':
464 if (this->mode_ == LINKER_SCRIPT)
468 case '[': case ']': case '*': case '?': case '-':
469 if (this->mode_ == LINKER_SCRIPT || this->mode_ == VERSION_SCRIPT
470 || this->mode_ == DYNAMIC_LIST)
474 // TODO(csilvers): why allow this? ^ is meaningless in version scripts.
476 if (this->mode_ == VERSION_SCRIPT || this->mode_ == DYNAMIC_LIST)
481 if (this->mode_ == LINKER_SCRIPT)
483 else if ((this->mode_ == VERSION_SCRIPT || this->mode_ == DYNAMIC_LIST)
486 // A name can have '::' in it, as that's a c++ namespace
487 // separator. But a single colon is not part of a name.
497 // For a number we accept 0x followed by hex digits, or any sequence
498 // of digits. The old linker accepts leading '$' for hex, and
499 // trailing HXBOD. Those are for MRI compatibility and we don't
502 // Return whether C1 C2 C3 can start a hex number.
505 Lex::can_start_hex(char c1, char c2, char c3)
507 if (c1 == '0' && (c2 == 'x' || c2 == 'X'))
508 return this->can_continue_hex(&c3);
512 // Return whether C can appear in a hex number.
515 Lex::can_continue_hex(const char* c)
519 case '0': case '1': case '2': case '3': case '4':
520 case '5': case '6': case '7': case '8': case '9':
521 case 'A': case 'B': case 'C': case 'D': case 'E': case 'F':
522 case 'a': case 'b': case 'c': case 'd': case 'e': case 'f':
530 // Return whether C can start a non-hex number.
533 Lex::can_start_number(char c)
537 case '0': case '1': case '2': case '3': case '4':
538 case '5': case '6': case '7': case '8': case '9':
546 // If C1 C2 C3 form a valid three character operator, return the
547 // opcode (defined in the yyscript.h file generated from yyscript.y).
548 // Otherwise return 0.
551 Lex::three_char_operator(char c1, char c2, char c3)
556 if (c2 == '<' && c3 == '=')
560 if (c2 == '>' && c3 == '=')
569 // If C1 C2 form a valid two character operator, return the opcode
570 // (defined in the yyscript.h file generated from yyscript.y).
571 // Otherwise return 0.
574 Lex::two_char_operator(char c1, char c2)
632 // If C1 is a valid operator, return the opcode. Otherwise return 0.
635 Lex::one_char_operator(char c1)
668 // Skip a C style comment. *PP points to just after the "/*". Return
669 // false if the comment did not end.
672 Lex::skip_c_comment(const char** pp)
675 while (p[0] != '*' || p[1] != '/')
686 this->linestart_ = p + 1;
695 // Skip a line # comment. Return false if there was no newline.
698 Lex::skip_line_comment(const char** pp)
701 size_t skip = strcspn(p, "\n");
710 this->linestart_ = p;
716 // Build a token CLASSIFICATION from all characters that match
717 // CAN_CONTINUE_FN. Update *PP.
720 Lex::gather_token(Token::Classification classification,
721 const char* (Lex::*can_continue_fn)(const char*),
726 const char* new_match = NULL;
727 while ((new_match = (this->*can_continue_fn)(match)) != NULL)
730 // A special case: integers may be followed by a single M or K,
732 if (classification == Token::TOKEN_INTEGER
733 && (*match == 'm' || *match == 'M' || *match == 'k' || *match == 'K'))
737 return this->make_token(classification, start, match - start, start);
740 // Build a token from a quoted string.
743 Lex::gather_quoted_string(const char** pp)
745 const char* start = *pp;
746 const char* p = start;
748 size_t skip = strcspn(p, "\"\n");
750 return this->make_invalid_token(start);
752 return this->make_token(Token::TOKEN_QUOTED_STRING, p, skip, start);
755 // Return the next token at *PP. Update *PP. General guideline: we
756 // require linker scripts to be simple ASCII. No unicode linker
757 // scripts. In particular we can assume that any '\0' is the end of
761 Lex::get_token(const char** pp)
770 return this->make_eof_token(p);
773 // Skip whitespace quickly.
774 while (*p == ' ' || *p == '\t' || *p == '\r')
781 this->linestart_ = p;
785 // Skip C style comments.
786 if (p[0] == '/' && p[1] == '*')
788 int lineno = this->lineno_;
789 int charpos = p - this->linestart_ + 1;
792 if (!this->skip_c_comment(pp))
793 return Token(Token::TOKEN_INVALID, lineno, charpos);
799 // Skip line comments.
803 if (!this->skip_line_comment(pp))
804 return this->make_eof_token(p);
810 if (this->can_start_name(p[0], p[1]))
811 return this->gather_token(Token::TOKEN_STRING,
812 &Lex::can_continue_name,
815 // We accept any arbitrary name in double quotes, as long as it
816 // does not cross a line boundary.
820 return this->gather_quoted_string(pp);
823 // Check for a number.
825 if (this->can_start_hex(p[0], p[1], p[2]))
826 return this->gather_token(Token::TOKEN_INTEGER,
827 &Lex::can_continue_hex,
830 if (Lex::can_start_number(p[0]))
831 return this->gather_token(Token::TOKEN_INTEGER,
832 &Lex::can_continue_number,
835 // Check for operators.
837 int opcode = Lex::three_char_operator(p[0], p[1], p[2]);
841 return this->make_token(opcode, p);
844 opcode = Lex::two_char_operator(p[0], p[1]);
848 return this->make_token(opcode, p);
851 opcode = Lex::one_char_operator(p[0]);
855 return this->make_token(opcode, p);
858 return this->make_token(Token::TOKEN_INVALID, p);
862 // Return the next token.
867 // The first token is special.
868 if (this->first_token_ != 0)
870 this->token_ = Token(this->first_token_, 0, 0);
871 this->first_token_ = 0;
872 return &this->token_;
875 this->token_ = this->get_token(&this->current_);
877 // Don't let an early null byte fool us into thinking that we've
878 // reached the end of the file.
879 if (this->token_.is_eof()
880 && (static_cast<size_t>(this->current_ - this->input_string_)
881 < this->input_length_))
882 this->token_ = this->make_invalid_token(this->current_);
884 return &this->token_;
887 // class Symbol_assignment.
889 // Add the symbol to the symbol table. This makes sure the symbol is
890 // there and defined. The actual value is stored later. We can't
891 // determine the actual value at this point, because we can't
892 // necessarily evaluate the expression until all ordinary symbols have
895 // The GNU linker lets symbol assignments in the linker script
896 // silently override defined symbols in object files. We are
897 // compatible. FIXME: Should we issue a warning?
900 Symbol_assignment::add_to_table(Symbol_table* symtab)
902 elfcpp::STV vis = this->hidden_ ? elfcpp::STV_HIDDEN : elfcpp::STV_DEFAULT;
903 this->sym_ = symtab->define_as_constant(this->name_.c_str(),
906 ? Symbol_table::DEFSYM
907 : Symbol_table::SCRIPT),
915 true); // force_override
918 // Finalize a symbol value.
921 Symbol_assignment::finalize(Symbol_table* symtab, const Layout* layout)
923 this->finalize_maybe_dot(symtab, layout, false, 0, NULL);
926 // Finalize a symbol value which can refer to the dot symbol.
929 Symbol_assignment::finalize_with_dot(Symbol_table* symtab,
930 const Layout* layout,
932 Output_section* dot_section)
934 this->finalize_maybe_dot(symtab, layout, true, dot_value, dot_section);
937 // Finalize a symbol value, internal version.
940 Symbol_assignment::finalize_maybe_dot(Symbol_table* symtab,
941 const Layout* layout,
942 bool is_dot_available,
944 Output_section* dot_section)
946 // If we were only supposed to provide this symbol, the sym_ field
947 // will be NULL if the symbol was not referenced.
948 if (this->sym_ == NULL)
950 gold_assert(this->provide_);
954 if (parameters->target().get_size() == 32)
956 #if defined(HAVE_TARGET_32_LITTLE) || defined(HAVE_TARGET_32_BIG)
957 this->sized_finalize<32>(symtab, layout, is_dot_available, dot_value,
963 else if (parameters->target().get_size() == 64)
965 #if defined(HAVE_TARGET_64_LITTLE) || defined(HAVE_TARGET_64_BIG)
966 this->sized_finalize<64>(symtab, layout, is_dot_available, dot_value,
978 Symbol_assignment::sized_finalize(Symbol_table* symtab, const Layout* layout,
979 bool is_dot_available, uint64_t dot_value,
980 Output_section* dot_section)
982 Output_section* section;
983 uint64_t final_val = this->val_->eval_maybe_dot(symtab, layout, true,
985 dot_value, dot_section,
986 §ion, NULL, false);
987 Sized_symbol<size>* ssym = symtab->get_sized_symbol<size>(this->sym_);
988 ssym->set_value(final_val);
990 ssym->set_output_section(section);
993 // Set the symbol value if the expression yields an absolute value or
994 // a value relative to DOT_SECTION.
997 Symbol_assignment::set_if_absolute(Symbol_table* symtab, const Layout* layout,
998 bool is_dot_available, uint64_t dot_value,
999 Output_section* dot_section)
1001 if (this->sym_ == NULL)
1004 Output_section* val_section;
1005 uint64_t val = this->val_->eval_maybe_dot(symtab, layout, false,
1006 is_dot_available, dot_value,
1007 dot_section, &val_section, NULL,
1009 if (val_section != NULL && val_section != dot_section)
1012 if (parameters->target().get_size() == 32)
1014 #if defined(HAVE_TARGET_32_LITTLE) || defined(HAVE_TARGET_32_BIG)
1015 Sized_symbol<32>* ssym = symtab->get_sized_symbol<32>(this->sym_);
1016 ssym->set_value(val);
1021 else if (parameters->target().get_size() == 64)
1023 #if defined(HAVE_TARGET_64_LITTLE) || defined(HAVE_TARGET_64_BIG)
1024 Sized_symbol<64>* ssym = symtab->get_sized_symbol<64>(this->sym_);
1025 ssym->set_value(val);
1032 if (val_section != NULL)
1033 this->sym_->set_output_section(val_section);
1036 // Print for debugging.
1039 Symbol_assignment::print(FILE* f) const
1041 if (this->provide_ && this->hidden_)
1042 fprintf(f, "PROVIDE_HIDDEN(");
1043 else if (this->provide_)
1044 fprintf(f, "PROVIDE(");
1045 else if (this->hidden_)
1048 fprintf(f, "%s = ", this->name_.c_str());
1049 this->val_->print(f);
1051 if (this->provide_ || this->hidden_)
1057 // Class Script_assertion.
1059 // Check the assertion.
1062 Script_assertion::check(const Symbol_table* symtab, const Layout* layout)
1064 if (!this->check_->eval(symtab, layout, true))
1065 gold_error("%s", this->message_.c_str());
1068 // Print for debugging.
1071 Script_assertion::print(FILE* f) const
1073 fprintf(f, "ASSERT(");
1074 this->check_->print(f);
1075 fprintf(f, ", \"%s\")\n", this->message_.c_str());
1078 // Class Script_options.
1080 Script_options::Script_options()
1081 : entry_(), symbol_assignments_(), symbol_definitions_(),
1082 symbol_references_(), version_script_info_(), script_sections_()
1086 // Returns true if NAME is on the list of symbol assignments waiting
1090 Script_options::is_pending_assignment(const char* name)
1092 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin();
1093 p != this->symbol_assignments_.end();
1095 if ((*p)->name() == name)
1100 // Add a symbol to be defined.
1103 Script_options::add_symbol_assignment(const char* name, size_t length,
1104 bool is_defsym, Expression* value,
1105 bool provide, bool hidden)
1107 if (length != 1 || name[0] != '.')
1109 if (this->script_sections_.in_sections_clause())
1111 gold_assert(!is_defsym);
1112 this->script_sections_.add_symbol_assignment(name, length, value,
1117 Symbol_assignment* p = new Symbol_assignment(name, length, is_defsym,
1118 value, provide, hidden);
1119 this->symbol_assignments_.push_back(p);
1124 std::string n(name, length);
1125 this->symbol_definitions_.insert(n);
1126 this->symbol_references_.erase(n);
1131 if (provide || hidden)
1132 gold_error(_("invalid use of PROVIDE for dot symbol"));
1134 // The GNU linker permits assignments to dot outside of SECTIONS
1135 // clauses and treats them as occurring inside, so we don't
1136 // check in_sections_clause here.
1137 this->script_sections_.add_dot_assignment(value);
1141 // Add a reference to a symbol.
1144 Script_options::add_symbol_reference(const char* name, size_t length)
1146 if (length != 1 || name[0] != '.')
1148 std::string n(name, length);
1149 if (this->symbol_definitions_.find(n) == this->symbol_definitions_.end())
1150 this->symbol_references_.insert(n);
1154 // Add an assertion.
1157 Script_options::add_assertion(Expression* check, const char* message,
1160 if (this->script_sections_.in_sections_clause())
1161 this->script_sections_.add_assertion(check, message, messagelen);
1164 Script_assertion* p = new Script_assertion(check, message, messagelen);
1165 this->assertions_.push_back(p);
1169 // Create sections required by any linker scripts.
1172 Script_options::create_script_sections(Layout* layout)
1174 if (this->saw_sections_clause())
1175 this->script_sections_.create_sections(layout);
1178 // Add any symbols we are defining to the symbol table.
1181 Script_options::add_symbols_to_table(Symbol_table* symtab)
1183 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin();
1184 p != this->symbol_assignments_.end();
1186 (*p)->add_to_table(symtab);
1187 this->script_sections_.add_symbols_to_table(symtab);
1190 // Finalize symbol values. Also check assertions.
1193 Script_options::finalize_symbols(Symbol_table* symtab, const Layout* layout)
1195 // We finalize the symbols defined in SECTIONS first, because they
1196 // are the ones which may have changed. This way if symbol outside
1197 // SECTIONS are defined in terms of symbols inside SECTIONS, they
1198 // will get the right value.
1199 this->script_sections_.finalize_symbols(symtab, layout);
1201 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin();
1202 p != this->symbol_assignments_.end();
1204 (*p)->finalize(symtab, layout);
1206 for (Assertions::iterator p = this->assertions_.begin();
1207 p != this->assertions_.end();
1209 (*p)->check(symtab, layout);
1212 // Set section addresses. We set all the symbols which have absolute
1213 // values. Then we let the SECTIONS clause do its thing. This
1214 // returns the segment which holds the file header and segment
1218 Script_options::set_section_addresses(Symbol_table* symtab, Layout* layout)
1220 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin();
1221 p != this->symbol_assignments_.end();
1223 (*p)->set_if_absolute(symtab, layout, false, 0, NULL);
1225 return this->script_sections_.set_section_addresses(symtab, layout);
1228 // This class holds data passed through the parser to the lexer and to
1229 // the parser support functions. This avoids global variables. We
1230 // can't use global variables because we need not be called by a
1231 // singleton thread.
1233 class Parser_closure
1236 Parser_closure(const char* filename,
1237 const Position_dependent_options& posdep_options,
1238 bool parsing_defsym, bool in_group, bool is_in_sysroot,
1239 Command_line* command_line,
1240 Script_options* script_options,
1242 bool skip_on_incompatible_target,
1243 Script_info* script_info)
1244 : filename_(filename), posdep_options_(posdep_options),
1245 parsing_defsym_(parsing_defsym), in_group_(in_group),
1246 is_in_sysroot_(is_in_sysroot),
1247 skip_on_incompatible_target_(skip_on_incompatible_target),
1248 found_incompatible_target_(false),
1249 command_line_(command_line), script_options_(script_options),
1250 version_script_info_(script_options->version_script_info()),
1251 lex_(lex), lineno_(0), charpos_(0), lex_mode_stack_(), inputs_(NULL),
1252 script_info_(script_info)
1254 // We start out processing C symbols in the default lex mode.
1255 this->language_stack_.push_back(Version_script_info::LANGUAGE_C);
1256 this->lex_mode_stack_.push_back(lex->mode());
1259 // Return the file name.
1262 { return this->filename_; }
1264 // Return the position dependent options. The caller may modify
1266 Position_dependent_options&
1267 position_dependent_options()
1268 { return this->posdep_options_; }
1270 // Whether we are parsing a --defsym.
1272 parsing_defsym() const
1273 { return this->parsing_defsym_; }
1275 // Return whether this script is being run in a group.
1278 { return this->in_group_; }
1280 // Return whether this script was found using a directory in the
1283 is_in_sysroot() const
1284 { return this->is_in_sysroot_; }
1286 // Whether to skip to the next file with the same name if we find an
1287 // incompatible target in an OUTPUT_FORMAT statement.
1289 skip_on_incompatible_target() const
1290 { return this->skip_on_incompatible_target_; }
1292 // Stop skipping to the next file on an incompatible target. This
1293 // is called when we make some unrevocable change to the data
1296 clear_skip_on_incompatible_target()
1297 { this->skip_on_incompatible_target_ = false; }
1299 // Whether we found an incompatible target in an OUTPUT_FORMAT
1302 found_incompatible_target() const
1303 { return this->found_incompatible_target_; }
1305 // Note that we found an incompatible target.
1307 set_found_incompatible_target()
1308 { this->found_incompatible_target_ = true; }
1310 // Returns the Command_line structure passed in at constructor time.
1311 // This value may be NULL. The caller may modify this, which modifies
1312 // the passed-in Command_line object (not a copy).
1315 { return this->command_line_; }
1317 // Return the options which may be set by a script.
1320 { return this->script_options_; }
1322 // Return the object in which version script information should be stored.
1323 Version_script_info*
1325 { return this->version_script_info_; }
1327 // Return the next token, and advance.
1331 const Token* token = this->lex_->next_token();
1332 this->lineno_ = token->lineno();
1333 this->charpos_ = token->charpos();
1337 // Set a new lexer mode, pushing the current one.
1339 push_lex_mode(Lex::Mode mode)
1341 this->lex_mode_stack_.push_back(this->lex_->mode());
1342 this->lex_->set_mode(mode);
1345 // Pop the lexer mode.
1349 gold_assert(!this->lex_mode_stack_.empty());
1350 this->lex_->set_mode(this->lex_mode_stack_.back());
1351 this->lex_mode_stack_.pop_back();
1354 // Return the current lexer mode.
1357 { return this->lex_mode_stack_.back(); }
1359 // Return the line number of the last token.
1362 { return this->lineno_; }
1364 // Return the character position in the line of the last token.
1367 { return this->charpos_; }
1369 // Return the list of input files, creating it if necessary. This
1370 // is a space leak--we never free the INPUTS_ pointer.
1374 if (this->inputs_ == NULL)
1375 this->inputs_ = new Input_arguments();
1376 return this->inputs_;
1379 // Return whether we saw any input files.
1382 { return this->inputs_ != NULL && !this->inputs_->empty(); }
1384 // Return the current language being processed in a version script
1385 // (eg, "C++"). The empty string represents unmangled C names.
1386 Version_script_info::Language
1387 get_current_language() const
1388 { return this->language_stack_.back(); }
1390 // Push a language onto the stack when entering an extern block.
1392 push_language(Version_script_info::Language lang)
1393 { this->language_stack_.push_back(lang); }
1395 // Pop a language off of the stack when exiting an extern block.
1399 gold_assert(!this->language_stack_.empty());
1400 this->language_stack_.pop_back();
1403 // Return a pointer to the incremental info.
1406 { return this->script_info_; }
1409 // The name of the file we are reading.
1410 const char* filename_;
1411 // The position dependent options.
1412 Position_dependent_options posdep_options_;
1413 // True if we are parsing a --defsym.
1414 bool parsing_defsym_;
1415 // Whether we are currently in a --start-group/--end-group.
1417 // Whether the script was found in a sysrooted directory.
1418 bool is_in_sysroot_;
1419 // If this is true, then if we find an OUTPUT_FORMAT with an
1420 // incompatible target, then we tell the parser to abort so that we
1421 // can search for the next file with the same name.
1422 bool skip_on_incompatible_target_;
1423 // True if we found an OUTPUT_FORMAT with an incompatible target.
1424 bool found_incompatible_target_;
1425 // May be NULL if the user chooses not to pass one in.
1426 Command_line* command_line_;
1427 // Options which may be set from any linker script.
1428 Script_options* script_options_;
1429 // Information parsed from a version script.
1430 Version_script_info* version_script_info_;
1433 // The line number of the last token returned by next_token.
1435 // The column number of the last token returned by next_token.
1437 // A stack of lexer modes.
1438 std::vector<Lex::Mode> lex_mode_stack_;
1439 // A stack of which extern/language block we're inside. Can be C++,
1440 // java, or empty for C.
1441 std::vector<Version_script_info::Language> language_stack_;
1442 // New input files found to add to the link.
1443 Input_arguments* inputs_;
1444 // Pointer to incremental linking info.
1445 Script_info* script_info_;
1448 // FILE was found as an argument on the command line. Try to read it
1449 // as a script. Return true if the file was handled.
1452 read_input_script(Workqueue* workqueue, Symbol_table* symtab, Layout* layout,
1453 Dirsearch* dirsearch, int dirindex,
1454 Input_objects* input_objects, Mapfile* mapfile,
1455 Input_group* input_group,
1456 const Input_argument* input_argument,
1457 Input_file* input_file, Task_token* next_blocker,
1458 bool* used_next_blocker)
1460 *used_next_blocker = false;
1462 std::string input_string;
1463 Lex::read_file(input_file, &input_string);
1465 Lex lex(input_string.c_str(), input_string.length(), PARSING_LINKER_SCRIPT);
1467 Script_info* script_info = NULL;
1468 if (layout->incremental_inputs() != NULL)
1470 const std::string& filename = input_file->filename();
1471 Timespec mtime = input_file->file().get_mtime();
1472 unsigned int arg_serial = input_argument->file().arg_serial();
1473 script_info = new Script_info(filename);
1474 layout->incremental_inputs()->report_script(script_info, arg_serial,
1478 Parser_closure closure(input_file->filename().c_str(),
1479 input_argument->file().options(),
1481 input_group != NULL,
1482 input_file->is_in_sysroot(),
1484 layout->script_options(),
1486 input_file->will_search_for(),
1489 bool old_saw_sections_clause =
1490 layout->script_options()->saw_sections_clause();
1492 if (yyparse(&closure) != 0)
1494 if (closure.found_incompatible_target())
1496 Read_symbols::incompatible_warning(input_argument, input_file);
1497 Read_symbols::requeue(workqueue, input_objects, symtab, layout,
1498 dirsearch, dirindex, mapfile, input_argument,
1499 input_group, next_blocker);
1505 if (!old_saw_sections_clause
1506 && layout->script_options()->saw_sections_clause()
1507 && layout->have_added_input_section())
1508 gold_error(_("%s: SECTIONS seen after other input files; try -T/--script"),
1509 input_file->filename().c_str());
1511 if (!closure.saw_inputs())
1514 Task_token* this_blocker = NULL;
1515 for (Input_arguments::const_iterator p = closure.inputs()->begin();
1516 p != closure.inputs()->end();
1520 if (p + 1 == closure.inputs()->end())
1524 nb = new Task_token(true);
1527 workqueue->queue_soon(new Read_symbols(input_objects, symtab,
1528 layout, dirsearch, 0, mapfile, &*p,
1529 input_group, NULL, this_blocker, nb));
1533 *used_next_blocker = true;
1538 // Helper function for read_version_script(), read_commandline_script() and
1539 // script_include_directive(). Processes the given file in the mode indicated
1540 // by first_token and lex_mode.
1543 read_script_file(const char* filename, Command_line* cmdline,
1544 Script_options* script_options,
1545 int first_token, Lex::Mode lex_mode)
1547 Dirsearch dirsearch;
1548 std::string name = filename;
1550 // If filename is a relative filename, search for it manually using "." +
1551 // cmdline->options()->library_path() -- not dirsearch.
1552 if (!IS_ABSOLUTE_PATH(filename))
1554 const General_options::Dir_list& search_path =
1555 cmdline->options().library_path();
1556 name = Dirsearch::find_file_in_dir_list(name, search_path, ".");
1559 // The file locking code wants to record a Task, but we haven't
1560 // started the workqueue yet. This is only for debugging purposes,
1561 // so we invent a fake value.
1562 const Task* task = reinterpret_cast<const Task*>(-1);
1564 // We don't want this file to be opened in binary mode.
1565 Position_dependent_options posdep = cmdline->position_dependent_options();
1566 if (posdep.format_enum() == General_options::OBJECT_FORMAT_BINARY)
1567 posdep.set_format_enum(General_options::OBJECT_FORMAT_ELF);
1568 Input_file_argument input_argument(name.c_str(),
1569 Input_file_argument::INPUT_FILE_TYPE_FILE,
1571 Input_file input_file(&input_argument);
1573 if (!input_file.open(dirsearch, task, &dummy))
1576 std::string input_string;
1577 Lex::read_file(&input_file, &input_string);
1579 Lex lex(input_string.c_str(), input_string.length(), first_token);
1580 lex.set_mode(lex_mode);
1582 Parser_closure closure(filename,
1583 cmdline->position_dependent_options(),
1584 first_token == Lex::DYNAMIC_LIST,
1586 input_file.is_in_sysroot(),
1592 if (yyparse(&closure) != 0)
1594 input_file.file().unlock(task);
1598 input_file.file().unlock(task);
1600 gold_assert(!closure.saw_inputs());
1605 // FILENAME was found as an argument to --script (-T).
1606 // Read it as a script, and execute its contents immediately.
1609 read_commandline_script(const char* filename, Command_line* cmdline)
1611 return read_script_file(filename, cmdline, &cmdline->script_options(),
1612 PARSING_LINKER_SCRIPT, Lex::LINKER_SCRIPT);
1615 // FILENAME was found as an argument to --version-script. Read it as
1616 // a version script, and store its contents in
1617 // cmdline->script_options()->version_script_info().
1620 read_version_script(const char* filename, Command_line* cmdline)
1622 return read_script_file(filename, cmdline, &cmdline->script_options(),
1623 PARSING_VERSION_SCRIPT, Lex::VERSION_SCRIPT);
1626 // FILENAME was found as an argument to --dynamic-list. Read it as a
1627 // list of symbols, and store its contents in DYNAMIC_LIST.
1630 read_dynamic_list(const char* filename, Command_line* cmdline,
1631 Script_options* dynamic_list)
1633 return read_script_file(filename, cmdline, dynamic_list,
1634 PARSING_DYNAMIC_LIST, Lex::DYNAMIC_LIST);
1637 // Implement the --defsym option on the command line. Return true if
1641 Script_options::define_symbol(const char* definition)
1643 Lex lex(definition, strlen(definition), PARSING_DEFSYM);
1644 lex.set_mode(Lex::EXPRESSION);
1647 Position_dependent_options posdep_options;
1649 Parser_closure closure("command line", posdep_options, true,
1650 false, false, NULL, this, &lex, false, NULL);
1652 if (yyparse(&closure) != 0)
1655 gold_assert(!closure.saw_inputs());
1660 // Print the script to F for debugging.
1663 Script_options::print(FILE* f) const
1665 fprintf(f, "%s: Dumping linker script\n", program_name);
1667 if (!this->entry_.empty())
1668 fprintf(f, "ENTRY(%s)\n", this->entry_.c_str());
1670 for (Symbol_assignments::const_iterator p =
1671 this->symbol_assignments_.begin();
1672 p != this->symbol_assignments_.end();
1676 for (Assertions::const_iterator p = this->assertions_.begin();
1677 p != this->assertions_.end();
1681 this->script_sections_.print(f);
1683 this->version_script_info_.print(f);
1686 // Manage mapping from keywords to the codes expected by the bison
1687 // parser. We construct one global object for each lex mode with
1690 class Keyword_to_parsecode
1693 // The structure which maps keywords to parsecodes.
1694 struct Keyword_parsecode
1697 const char* keyword;
1698 // Corresponding parsecode.
1702 Keyword_to_parsecode(const Keyword_parsecode* keywords,
1704 : keyword_parsecodes_(keywords), keyword_count_(keyword_count)
1707 // Return the parsecode corresponding KEYWORD, or 0 if it is not a
1710 keyword_to_parsecode(const char* keyword, size_t len) const;
1713 const Keyword_parsecode* keyword_parsecodes_;
1714 const int keyword_count_;
1717 // Mapping from keyword string to keyword parsecode. This array must
1718 // be kept in sorted order. Parsecodes are looked up using bsearch.
1719 // This array must correspond to the list of parsecodes in yyscript.y.
1721 static const Keyword_to_parsecode::Keyword_parsecode
1722 script_keyword_parsecodes[] =
1724 { "ABSOLUTE", ABSOLUTE },
1726 { "ALIGN", ALIGN_K },
1727 { "ALIGNOF", ALIGNOF },
1728 { "ASSERT", ASSERT_K },
1729 { "AS_NEEDED", AS_NEEDED },
1734 { "CONSTANT", CONSTANT },
1735 { "CONSTRUCTORS", CONSTRUCTORS },
1737 { "CREATE_OBJECT_SYMBOLS", CREATE_OBJECT_SYMBOLS },
1738 { "DATA_SEGMENT_ALIGN", DATA_SEGMENT_ALIGN },
1739 { "DATA_SEGMENT_END", DATA_SEGMENT_END },
1740 { "DATA_SEGMENT_RELRO_END", DATA_SEGMENT_RELRO_END },
1741 { "DEFINED", DEFINED },
1744 { "EXCLUDE_FILE", EXCLUDE_FILE },
1745 { "EXTERN", EXTERN },
1748 { "FORCE_COMMON_ALLOCATION", FORCE_COMMON_ALLOCATION },
1751 { "INCLUDE", INCLUDE },
1753 { "INHIBIT_COMMON_ALLOCATION", INHIBIT_COMMON_ALLOCATION },
1756 { "LENGTH", LENGTH },
1757 { "LOADADDR", LOADADDR },
1761 { "MEMORY", MEMORY },
1764 { "NOCROSSREFS", NOCROSSREFS },
1765 { "NOFLOAT", NOFLOAT },
1766 { "NOLOAD", NOLOAD },
1767 { "ONLY_IF_RO", ONLY_IF_RO },
1768 { "ONLY_IF_RW", ONLY_IF_RW },
1769 { "OPTION", OPTION },
1770 { "ORIGIN", ORIGIN },
1771 { "OUTPUT", OUTPUT },
1772 { "OUTPUT_ARCH", OUTPUT_ARCH },
1773 { "OUTPUT_FORMAT", OUTPUT_FORMAT },
1774 { "OVERLAY", OVERLAY },
1776 { "PROVIDE", PROVIDE },
1777 { "PROVIDE_HIDDEN", PROVIDE_HIDDEN },
1779 { "SEARCH_DIR", SEARCH_DIR },
1780 { "SECTIONS", SECTIONS },
1781 { "SEGMENT_START", SEGMENT_START },
1783 { "SIZEOF", SIZEOF },
1784 { "SIZEOF_HEADERS", SIZEOF_HEADERS },
1785 { "SORT", SORT_BY_NAME },
1786 { "SORT_BY_ALIGNMENT", SORT_BY_ALIGNMENT },
1787 { "SORT_BY_NAME", SORT_BY_NAME },
1788 { "SPECIAL", SPECIAL },
1790 { "STARTUP", STARTUP },
1791 { "SUBALIGN", SUBALIGN },
1792 { "SYSLIB", SYSLIB },
1793 { "TARGET", TARGET_K },
1794 { "TRUNCATE", TRUNCATE },
1795 { "VERSION", VERSIONK },
1796 { "global", GLOBAL },
1802 { "sizeof_headers", SIZEOF_HEADERS },
1805 static const Keyword_to_parsecode
1806 script_keywords(&script_keyword_parsecodes[0],
1807 (sizeof(script_keyword_parsecodes)
1808 / sizeof(script_keyword_parsecodes[0])));
1810 static const Keyword_to_parsecode::Keyword_parsecode
1811 version_script_keyword_parsecodes[] =
1813 { "extern", EXTERN },
1814 { "global", GLOBAL },
1818 static const Keyword_to_parsecode
1819 version_script_keywords(&version_script_keyword_parsecodes[0],
1820 (sizeof(version_script_keyword_parsecodes)
1821 / sizeof(version_script_keyword_parsecodes[0])));
1823 static const Keyword_to_parsecode::Keyword_parsecode
1824 dynamic_list_keyword_parsecodes[] =
1826 { "extern", EXTERN },
1829 static const Keyword_to_parsecode
1830 dynamic_list_keywords(&dynamic_list_keyword_parsecodes[0],
1831 (sizeof(dynamic_list_keyword_parsecodes)
1832 / sizeof(dynamic_list_keyword_parsecodes[0])));
1836 // Comparison function passed to bsearch.
1848 ktt_compare(const void* keyv, const void* kttv)
1850 const Ktt_key* key = static_cast<const Ktt_key*>(keyv);
1851 const Keyword_to_parsecode::Keyword_parsecode* ktt =
1852 static_cast<const Keyword_to_parsecode::Keyword_parsecode*>(kttv);
1853 int i = strncmp(key->str, ktt->keyword, key->len);
1856 if (ktt->keyword[key->len] != '\0')
1861 } // End extern "C".
1864 Keyword_to_parsecode::keyword_to_parsecode(const char* keyword,
1870 void* kttv = bsearch(&key,
1871 this->keyword_parsecodes_,
1872 this->keyword_count_,
1873 sizeof(this->keyword_parsecodes_[0]),
1877 Keyword_parsecode* ktt = static_cast<Keyword_parsecode*>(kttv);
1878 return ktt->parsecode;
1881 // The following structs are used within the VersionInfo class as well
1882 // as in the bison helper functions. They store the information
1883 // parsed from the version script.
1885 // A single version expression.
1886 // For example, pattern="std::map*" and language="C++".
1887 struct Version_expression
1889 Version_expression(const std::string& a_pattern,
1890 Version_script_info::Language a_language,
1892 : pattern(a_pattern), language(a_language), exact_match(a_exact_match),
1893 was_matched_by_symbol(false)
1896 std::string pattern;
1897 Version_script_info::Language language;
1898 // If false, we use glob() to match pattern. If true, we use strcmp().
1900 // True if --no-undefined-version is in effect and we found this
1901 // version in get_symbol_version. We use mutable because this
1902 // struct is generally not modifiable after it has been created.
1903 mutable bool was_matched_by_symbol;
1906 // A list of expressions.
1907 struct Version_expression_list
1909 std::vector<struct Version_expression> expressions;
1912 // A list of which versions upon which another version depends.
1913 // Strings should be from the Stringpool.
1914 struct Version_dependency_list
1916 std::vector<std::string> dependencies;
1919 // The total definition of a version. It includes the tag for the
1920 // version, its global and local expressions, and any dependencies.
1924 : tag(), global(NULL), local(NULL), dependencies(NULL)
1928 const struct Version_expression_list* global;
1929 const struct Version_expression_list* local;
1930 const struct Version_dependency_list* dependencies;
1933 // Helper class that calls cplus_demangle when needed and takes care of freeing
1936 class Lazy_demangler
1939 Lazy_demangler(const char* symbol, int options)
1940 : symbol_(symbol), options_(options), demangled_(NULL), did_demangle_(false)
1944 { free(this->demangled_); }
1946 // Return the demangled name. The actual demangling happens on the first call,
1947 // and the result is later cached.
1952 // The symbol to demangle.
1953 const char* symbol_;
1954 // Option flags to pass to cplus_demagle.
1956 // The cached demangled value, or NULL if demangling didn't happen yet or
1959 // Whether we already called cplus_demangle
1963 // Return the demangled name. The actual demangling happens on the first call,
1964 // and the result is later cached. Returns NULL if the symbol cannot be
1968 Lazy_demangler::get()
1970 if (!this->did_demangle_)
1972 this->demangled_ = cplus_demangle(this->symbol_, this->options_);
1973 this->did_demangle_ = true;
1975 return this->demangled_;
1978 // Class Version_script_info.
1980 Version_script_info::Version_script_info()
1981 : dependency_lists_(), expression_lists_(), version_trees_(), globs_(),
1982 default_version_(NULL), default_is_global_(false), is_finalized_(false)
1984 for (int i = 0; i < LANGUAGE_COUNT; ++i)
1985 this->exact_[i] = NULL;
1988 Version_script_info::~Version_script_info()
1992 // Forget all the known version script information.
1995 Version_script_info::clear()
1997 for (size_t k = 0; k < this->dependency_lists_.size(); ++k)
1998 delete this->dependency_lists_[k];
1999 this->dependency_lists_.clear();
2000 for (size_t k = 0; k < this->version_trees_.size(); ++k)
2001 delete this->version_trees_[k];
2002 this->version_trees_.clear();
2003 for (size_t k = 0; k < this->expression_lists_.size(); ++k)
2004 delete this->expression_lists_[k];
2005 this->expression_lists_.clear();
2008 // Finalize the version script information.
2011 Version_script_info::finalize()
2013 if (!this->is_finalized_)
2015 this->build_lookup_tables();
2016 this->is_finalized_ = true;
2020 // Return all the versions.
2022 std::vector<std::string>
2023 Version_script_info::get_versions() const
2025 std::vector<std::string> ret;
2026 for (size_t j = 0; j < this->version_trees_.size(); ++j)
2027 if (!this->version_trees_[j]->tag.empty())
2028 ret.push_back(this->version_trees_[j]->tag);
2032 // Return the dependencies of VERSION.
2034 std::vector<std::string>
2035 Version_script_info::get_dependencies(const char* version) const
2037 std::vector<std::string> ret;
2038 for (size_t j = 0; j < this->version_trees_.size(); ++j)
2039 if (this->version_trees_[j]->tag == version)
2041 const struct Version_dependency_list* deps =
2042 this->version_trees_[j]->dependencies;
2044 for (size_t k = 0; k < deps->dependencies.size(); ++k)
2045 ret.push_back(deps->dependencies[k]);
2051 // A version script essentially maps a symbol name to a version tag
2052 // and an indication of whether symbol is global or local within that
2053 // version tag. Each symbol maps to at most one version tag.
2054 // Unfortunately, in practice, version scripts are ambiguous, and list
2055 // symbols multiple times. Thus, we have to document the matching
2058 // This is a description of what the GNU linker does as of 2010-01-11.
2059 // It walks through the version tags in the order in which they appear
2060 // in the version script. For each tag, it first walks through the
2061 // global patterns for that tag, then the local patterns. When
2062 // looking at a single pattern, it first applies any language specific
2063 // demangling as specified for the pattern, and then matches the
2064 // resulting symbol name to the pattern. If it finds an exact match
2065 // for a literal pattern (a pattern enclosed in quotes or with no
2066 // wildcard characters), then that is the match that it uses. If
2067 // finds a match with a wildcard pattern, then it saves it and
2068 // continues searching. Wildcard patterns that are exactly "*" are
2069 // saved separately.
2071 // If no exact match with a literal pattern is ever found, then if a
2072 // wildcard match with a global pattern was found it is used,
2073 // otherwise if a wildcard match with a local pattern was found it is
2076 // This is the result:
2077 // * If there is an exact match, then we use the first tag in the
2078 // version script where it matches.
2079 // + If the exact match in that tag is global, it is used.
2080 // + Otherwise the exact match in that tag is local, and is used.
2081 // * Otherwise, if there is any match with a global wildcard pattern:
2082 // + If there is any match with a wildcard pattern which is not
2083 // "*", then we use the tag in which the *last* such pattern
2085 // + Otherwise, we matched "*". If there is no match with a local
2086 // wildcard pattern which is not "*", then we use the *last*
2087 // match with a global "*". Otherwise, continue.
2088 // * Otherwise, if there is any match with a local wildcard pattern:
2089 // + If there is any match with a wildcard pattern which is not
2090 // "*", then we use the tag in which the *last* such pattern
2092 // + Otherwise, we matched "*", and we use the tag in which the
2093 // *last* such match occurred.
2095 // There is an additional wrinkle. When the GNU linker finds a symbol
2096 // with a version defined in an object file due to a .symver
2097 // directive, it looks up that symbol name in that version tag. If it
2098 // finds it, it matches the symbol name against the patterns for that
2099 // version. If there is no match with a global pattern, but there is
2100 // a match with a local pattern, then the GNU linker marks the symbol
2103 // We want gold to be generally compatible, but we also want gold to
2104 // be fast. These are the rules that gold implements:
2105 // * If there is an exact match for the mangled name, we use it.
2106 // + If there is more than one exact match, we give a warning, and
2107 // we use the first tag in the script which matches.
2108 // + If a symbol has an exact match as both global and local for
2109 // the same version tag, we give an error.
2110 // * Otherwise, we look for an extern C++ or an extern Java exact
2111 // match. If we find an exact match, we use it.
2112 // + If there is more than one exact match, we give a warning, and
2113 // we use the first tag in the script which matches.
2114 // + If a symbol has an exact match as both global and local for
2115 // the same version tag, we give an error.
2116 // * Otherwise, we look through the wildcard patterns, ignoring "*"
2117 // patterns. We look through the version tags in reverse order.
2118 // For each version tag, we look through the global patterns and
2119 // then the local patterns. We use the first match we find (i.e.,
2120 // the last matching version tag in the file).
2121 // * Otherwise, we use the "*" pattern if there is one. We give an
2122 // error if there are multiple "*" patterns.
2124 // At least for now, gold does not look up the version tag for a
2125 // symbol version found in an object file to see if it should be
2126 // forced local. There are other ways to force a symbol to be local,
2127 // and I don't understand why this one is useful.
2129 // Build a set of fast lookup tables for a version script.
2132 Version_script_info::build_lookup_tables()
2134 size_t size = this->version_trees_.size();
2135 for (size_t j = 0; j < size; ++j)
2137 const Version_tree* v = this->version_trees_[j];
2138 this->build_expression_list_lookup(v->local, v, false);
2139 this->build_expression_list_lookup(v->global, v, true);
2143 // If a pattern has backlashes but no unquoted wildcard characters,
2144 // then we apply backslash unquoting and look for an exact match.
2145 // Otherwise we treat it as a wildcard pattern. This function returns
2146 // true for a wildcard pattern. Otherwise, it does backslash
2147 // unquoting on *PATTERN and returns false. If this returns true,
2148 // *PATTERN may have been partially unquoted.
2151 Version_script_info::unquote(std::string* pattern) const
2153 bool saw_backslash = false;
2154 size_t len = pattern->length();
2156 for (size_t i = 0; i < len; ++i)
2159 saw_backslash = false;
2162 switch ((*pattern)[i])
2164 case '?': case '[': case '*':
2167 saw_backslash = true;
2175 (*pattern)[j] = (*pattern)[i];
2181 // Add an exact match for MATCH to *PE. The result of the match is
2185 Version_script_info::add_exact_match(const std::string& match,
2186 const Version_tree* v, bool is_global,
2187 const Version_expression* ve,
2190 std::pair<Exact::iterator, bool> ins =
2191 pe->insert(std::make_pair(match, Version_tree_match(v, is_global, ve)));
2194 // This is the first time we have seen this match.
2198 Version_tree_match& vtm(ins.first->second);
2199 if (vtm.real->tag != v->tag)
2201 // This is an ambiguous match. We still return the
2202 // first version that we found in the script, but we
2203 // record the new version to issue a warning if we
2204 // wind up looking up this symbol.
2205 if (vtm.ambiguous == NULL)
2208 else if (is_global != vtm.is_global)
2210 // We have a match for both the global and local entries for a
2211 // version tag. That's got to be wrong.
2212 gold_error(_("'%s' appears as both a global and a local symbol "
2213 "for version '%s' in script"),
2214 match.c_str(), v->tag.c_str());
2218 // Build fast lookup information for EXPLIST and store it in LOOKUP.
2219 // All matches go to V, and IS_GLOBAL is true if they are global
2223 Version_script_info::build_expression_list_lookup(
2224 const Version_expression_list* explist,
2225 const Version_tree* v,
2228 if (explist == NULL)
2230 size_t size = explist->expressions.size();
2231 for (size_t i = 0; i < size; ++i)
2233 const Version_expression& exp(explist->expressions[i]);
2235 if (exp.pattern.length() == 1 && exp.pattern[0] == '*')
2237 if (this->default_version_ != NULL
2238 && this->default_version_->tag != v->tag)
2239 gold_warning(_("wildcard match appears in both version '%s' "
2240 "and '%s' in script"),
2241 this->default_version_->tag.c_str(), v->tag.c_str());
2242 else if (this->default_version_ != NULL
2243 && this->default_is_global_ != is_global)
2244 gold_error(_("wildcard match appears as both global and local "
2245 "in version '%s' in script"),
2247 this->default_version_ = v;
2248 this->default_is_global_ = is_global;
2252 std::string pattern = exp.pattern;
2253 if (!exp.exact_match)
2255 if (this->unquote(&pattern))
2257 this->globs_.push_back(Glob(&exp, v, is_global));
2262 if (this->exact_[exp.language] == NULL)
2263 this->exact_[exp.language] = new Exact();
2264 this->add_exact_match(pattern, v, is_global, &exp,
2265 this->exact_[exp.language]);
2269 // Return the name to match given a name, a language code, and two
2273 Version_script_info::get_name_to_match(const char* name,
2275 Lazy_demangler* cpp_demangler,
2276 Lazy_demangler* java_demangler) const
2283 return cpp_demangler->get();
2285 return java_demangler->get();
2291 // Look up SYMBOL_NAME in the list of versions. Return true if the
2292 // symbol is found, false if not. If the symbol is found, then if
2293 // PVERSION is not NULL, set *PVERSION to the version tag, and if
2294 // P_IS_GLOBAL is not NULL, set *P_IS_GLOBAL according to whether the
2295 // symbol is global or not.
2298 Version_script_info::get_symbol_version(const char* symbol_name,
2299 std::string* pversion,
2300 bool* p_is_global) const
2302 Lazy_demangler cpp_demangled_name(symbol_name, DMGL_ANSI | DMGL_PARAMS);
2303 Lazy_demangler java_demangled_name(symbol_name,
2304 DMGL_ANSI | DMGL_PARAMS | DMGL_JAVA);
2306 gold_assert(this->is_finalized_);
2307 for (int i = 0; i < LANGUAGE_COUNT; ++i)
2309 Exact* exact = this->exact_[i];
2313 const char* name_to_match = this->get_name_to_match(symbol_name, i,
2314 &cpp_demangled_name,
2315 &java_demangled_name);
2316 if (name_to_match == NULL)
2318 // If the name can not be demangled, the GNU linker goes
2319 // ahead and tries to match it anyhow. That does not
2320 // make sense to me and I have not implemented it.
2324 Exact::const_iterator pe = exact->find(name_to_match);
2325 if (pe != exact->end())
2327 const Version_tree_match& vtm(pe->second);
2328 if (vtm.ambiguous != NULL)
2329 gold_warning(_("using '%s' as version for '%s' which is also "
2330 "named in version '%s' in script"),
2331 vtm.real->tag.c_str(), name_to_match,
2332 vtm.ambiguous->tag.c_str());
2334 if (pversion != NULL)
2335 *pversion = vtm.real->tag;
2336 if (p_is_global != NULL)
2337 *p_is_global = vtm.is_global;
2339 // If we are using --no-undefined-version, and this is a
2340 // global symbol, we have to record that we have found this
2341 // symbol, so that we don't warn about it. We have to do
2342 // this now, because otherwise we have no way to get from a
2343 // non-C language back to the demangled name that we
2345 if (p_is_global != NULL && vtm.is_global)
2346 vtm.expression->was_matched_by_symbol = true;
2352 // Look through the glob patterns in reverse order.
2354 for (Globs::const_reverse_iterator p = this->globs_.rbegin();
2355 p != this->globs_.rend();
2358 int language = p->expression->language;
2359 const char* name_to_match = this->get_name_to_match(symbol_name,
2361 &cpp_demangled_name,
2362 &java_demangled_name);
2363 if (name_to_match == NULL)
2366 if (fnmatch(p->expression->pattern.c_str(), name_to_match,
2369 if (pversion != NULL)
2370 *pversion = p->version->tag;
2371 if (p_is_global != NULL)
2372 *p_is_global = p->is_global;
2377 // Finally, there may be a wildcard.
2378 if (this->default_version_ != NULL)
2380 if (pversion != NULL)
2381 *pversion = this->default_version_->tag;
2382 if (p_is_global != NULL)
2383 *p_is_global = this->default_is_global_;
2390 // Give an error if any exact symbol names (not wildcards) appear in a
2391 // version script, but there is no such symbol.
2394 Version_script_info::check_unmatched_names(const Symbol_table* symtab) const
2396 for (size_t i = 0; i < this->version_trees_.size(); ++i)
2398 const Version_tree* vt = this->version_trees_[i];
2399 if (vt->global == NULL)
2401 for (size_t j = 0; j < vt->global->expressions.size(); ++j)
2403 const Version_expression& expression(vt->global->expressions[j]);
2405 // Ignore cases where we used the version because we saw a
2406 // symbol that we looked up. Note that
2407 // WAS_MATCHED_BY_SYMBOL will be true even if the symbol was
2408 // not a definition. That's OK as in that case we most
2409 // likely gave an undefined symbol error anyhow.
2410 if (expression.was_matched_by_symbol)
2413 // Just ignore names which are in languages other than C.
2414 // We have no way to look them up in the symbol table.
2415 if (expression.language != LANGUAGE_C)
2418 // Remove backslash quoting, and ignore wildcard patterns.
2419 std::string pattern = expression.pattern;
2420 if (!expression.exact_match)
2422 if (this->unquote(&pattern))
2426 if (symtab->lookup(pattern.c_str(), vt->tag.c_str()) == NULL)
2427 gold_error(_("version script assignment of %s to symbol %s "
2428 "failed: symbol not defined"),
2429 vt->tag.c_str(), pattern.c_str());
2434 struct Version_dependency_list*
2435 Version_script_info::allocate_dependency_list()
2437 dependency_lists_.push_back(new Version_dependency_list);
2438 return dependency_lists_.back();
2441 struct Version_expression_list*
2442 Version_script_info::allocate_expression_list()
2444 expression_lists_.push_back(new Version_expression_list);
2445 return expression_lists_.back();
2448 struct Version_tree*
2449 Version_script_info::allocate_version_tree()
2451 version_trees_.push_back(new Version_tree);
2452 return version_trees_.back();
2455 // Print for debugging.
2458 Version_script_info::print(FILE* f) const
2463 fprintf(f, "VERSION {");
2465 for (size_t i = 0; i < this->version_trees_.size(); ++i)
2467 const Version_tree* vt = this->version_trees_[i];
2469 if (vt->tag.empty())
2472 fprintf(f, " %s {\n", vt->tag.c_str());
2474 if (vt->global != NULL)
2476 fprintf(f, " global :\n");
2477 this->print_expression_list(f, vt->global);
2480 if (vt->local != NULL)
2482 fprintf(f, " local :\n");
2483 this->print_expression_list(f, vt->local);
2487 if (vt->dependencies != NULL)
2489 const Version_dependency_list* deps = vt->dependencies;
2490 for (size_t j = 0; j < deps->dependencies.size(); ++j)
2492 if (j < deps->dependencies.size() - 1)
2494 fprintf(f, " %s", deps->dependencies[j].c_str());
2504 Version_script_info::print_expression_list(
2506 const Version_expression_list* vel) const
2508 Version_script_info::Language current_language = LANGUAGE_C;
2509 for (size_t i = 0; i < vel->expressions.size(); ++i)
2511 const Version_expression& ve(vel->expressions[i]);
2513 if (ve.language != current_language)
2515 if (current_language != LANGUAGE_C)
2517 switch (ve.language)
2522 fprintf(f, " extern \"C++\" {\n");
2525 fprintf(f, " extern \"Java\" {\n");
2530 current_language = ve.language;
2534 if (current_language != LANGUAGE_C)
2539 fprintf(f, "%s", ve.pattern.c_str());
2546 if (current_language != LANGUAGE_C)
2550 } // End namespace gold.
2552 // The remaining functions are extern "C", so it's clearer to not put
2553 // them in namespace gold.
2555 using namespace gold;
2557 // This function is called by the bison parser to return the next
2561 yylex(YYSTYPE* lvalp, void* closurev)
2563 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2564 const Token* token = closure->next_token();
2565 switch (token->classification())
2570 case Token::TOKEN_INVALID:
2571 yyerror(closurev, "invalid character");
2574 case Token::TOKEN_EOF:
2577 case Token::TOKEN_STRING:
2579 // This is either a keyword or a STRING.
2581 const char* str = token->string_value(&len);
2583 switch (closure->lex_mode())
2585 case Lex::LINKER_SCRIPT:
2586 parsecode = script_keywords.keyword_to_parsecode(str, len);
2588 case Lex::VERSION_SCRIPT:
2589 parsecode = version_script_keywords.keyword_to_parsecode(str, len);
2591 case Lex::DYNAMIC_LIST:
2592 parsecode = dynamic_list_keywords.keyword_to_parsecode(str, len);
2599 lvalp->string.value = str;
2600 lvalp->string.length = len;
2604 case Token::TOKEN_QUOTED_STRING:
2605 lvalp->string.value = token->string_value(&lvalp->string.length);
2606 return QUOTED_STRING;
2608 case Token::TOKEN_OPERATOR:
2609 return token->operator_value();
2611 case Token::TOKEN_INTEGER:
2612 lvalp->integer = token->integer_value();
2617 // This function is called by the bison parser to report an error.
2620 yyerror(void* closurev, const char* message)
2622 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2623 gold_error(_("%s:%d:%d: %s"), closure->filename(), closure->lineno(),
2624 closure->charpos(), message);
2627 // Called by the bison parser to add an external symbol to the link.
2630 script_add_extern(void* closurev, const char* name, size_t length)
2632 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2633 closure->script_options()->add_symbol_reference(name, length);
2636 // Called by the bison parser to add a file to the link.
2639 script_add_file(void* closurev, const char* name, size_t length)
2641 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2643 // If this is an absolute path, and we found the script in the
2644 // sysroot, then we want to prepend the sysroot to the file name.
2645 // For example, this is how we handle a cross link to the x86_64
2646 // libc.so, which refers to /lib/libc.so.6.
2647 std::string name_string(name, length);
2648 const char* extra_search_path = ".";
2649 std::string script_directory;
2650 if (IS_ABSOLUTE_PATH(name_string.c_str()))
2652 if (closure->is_in_sysroot())
2654 const std::string& sysroot(parameters->options().sysroot());
2655 gold_assert(!sysroot.empty());
2656 name_string = sysroot + name_string;
2661 // In addition to checking the normal library search path, we
2662 // also want to check in the script-directory.
2663 const char* slash = strrchr(closure->filename(), '/');
2666 script_directory.assign(closure->filename(),
2667 slash - closure->filename() + 1);
2668 extra_search_path = script_directory.c_str();
2672 Input_file_argument file(name_string.c_str(),
2673 Input_file_argument::INPUT_FILE_TYPE_FILE,
2674 extra_search_path, false,
2675 closure->position_dependent_options());
2676 Input_argument& arg = closure->inputs()->add_file(file);
2677 arg.set_script_info(closure->script_info());
2680 // Called by the bison parser to add a library to the link.
2683 script_add_library(void* closurev, const char* name, size_t length)
2685 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2686 std::string name_string(name, length);
2688 if (name_string[0] != 'l')
2689 gold_error(_("library name must be prefixed with -l"));
2691 Input_file_argument file(name_string.c_str() + 1,
2692 Input_file_argument::INPUT_FILE_TYPE_LIBRARY,
2694 closure->position_dependent_options());
2695 Input_argument& arg = closure->inputs()->add_file(file);
2696 arg.set_script_info(closure->script_info());
2699 // Called by the bison parser to start a group. If we are already in
2700 // a group, that means that this script was invoked within a
2701 // --start-group --end-group sequence on the command line, or that
2702 // this script was found in a GROUP of another script. In that case,
2703 // we simply continue the existing group, rather than starting a new
2704 // one. It is possible to construct a case in which this will do
2705 // something other than what would happen if we did a recursive group,
2706 // but it's hard to imagine why the different behaviour would be
2707 // useful for a real program. Avoiding recursive groups is simpler
2708 // and more efficient.
2711 script_start_group(void* closurev)
2713 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2714 if (!closure->in_group())
2715 closure->inputs()->start_group();
2718 // Called by the bison parser at the end of a group.
2721 script_end_group(void* closurev)
2723 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2724 if (!closure->in_group())
2725 closure->inputs()->end_group();
2728 // Called by the bison parser to start an AS_NEEDED list.
2731 script_start_as_needed(void* closurev)
2733 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2734 closure->position_dependent_options().set_as_needed(true);
2737 // Called by the bison parser at the end of an AS_NEEDED list.
2740 script_end_as_needed(void* closurev)
2742 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2743 closure->position_dependent_options().set_as_needed(false);
2746 // Called by the bison parser to set the entry symbol.
2749 script_set_entry(void* closurev, const char* entry, size_t length)
2751 // We'll parse this exactly the same as --entry=ENTRY on the commandline
2752 // TODO(csilvers): FIXME -- call set_entry directly.
2753 std::string arg("--entry=");
2754 arg.append(entry, length);
2755 script_parse_option(closurev, arg.c_str(), arg.size());
2758 // Called by the bison parser to set whether to define common symbols.
2761 script_set_common_allocation(void* closurev, int set)
2763 const char* arg = set != 0 ? "--define-common" : "--no-define-common";
2764 script_parse_option(closurev, arg, strlen(arg));
2767 // Called by the bison parser to refer to a symbol.
2769 extern "C" Expression*
2770 script_symbol(void* closurev, const char* name, size_t length)
2772 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2773 if (length != 1 || name[0] != '.')
2774 closure->script_options()->add_symbol_reference(name, length);
2775 return script_exp_string(name, length);
2778 // Called by the bison parser to define a symbol.
2781 script_set_symbol(void* closurev, const char* name, size_t length,
2782 Expression* value, int providei, int hiddeni)
2784 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2785 const bool provide = providei != 0;
2786 const bool hidden = hiddeni != 0;
2787 closure->script_options()->add_symbol_assignment(name, length,
2788 closure->parsing_defsym(),
2789 value, provide, hidden);
2790 closure->clear_skip_on_incompatible_target();
2793 // Called by the bison parser to add an assertion.
2796 script_add_assertion(void* closurev, Expression* check, const char* message,
2799 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2800 closure->script_options()->add_assertion(check, message, messagelen);
2801 closure->clear_skip_on_incompatible_target();
2804 // Called by the bison parser to parse an OPTION.
2807 script_parse_option(void* closurev, const char* option, size_t length)
2809 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2810 // We treat the option as a single command-line option, even if
2811 // it has internal whitespace.
2812 if (closure->command_line() == NULL)
2814 // There are some options that we could handle here--e.g.,
2815 // -lLIBRARY. Should we bother?
2816 gold_warning(_("%s:%d:%d: ignoring command OPTION; OPTION is only valid"
2817 " for scripts specified via -T/--script"),
2818 closure->filename(), closure->lineno(), closure->charpos());
2822 bool past_a_double_dash_option = false;
2823 const char* mutable_option = strndup(option, length);
2824 gold_assert(mutable_option != NULL);
2825 closure->command_line()->process_one_option(1, &mutable_option, 0,
2826 &past_a_double_dash_option);
2827 // The General_options class will quite possibly store a pointer
2828 // into mutable_option, so we can't free it. In cases the class
2829 // does not store such a pointer, this is a memory leak. Alas. :(
2831 closure->clear_skip_on_incompatible_target();
2834 // Called by the bison parser to handle OUTPUT_FORMAT. OUTPUT_FORMAT
2835 // takes either one or three arguments. In the three argument case,
2836 // the format depends on the endianness option, which we don't
2837 // currently support (FIXME). If we see an OUTPUT_FORMAT for the
2838 // wrong format, then we want to search for a new file. Returning 0
2839 // here will cause the parser to immediately abort.
2842 script_check_output_format(void* closurev,
2843 const char* default_name, size_t default_length,
2844 const char*, size_t, const char*, size_t)
2846 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2847 std::string name(default_name, default_length);
2848 Target* target = select_target_by_bfd_name(name.c_str());
2849 if (target == NULL || !parameters->is_compatible_target(target))
2851 if (closure->skip_on_incompatible_target())
2853 closure->set_found_incompatible_target();
2856 // FIXME: Should we warn about the unknown target?
2861 // Called by the bison parser to handle TARGET.
2864 script_set_target(void* closurev, const char* target, size_t len)
2866 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2867 std::string s(target, len);
2868 General_options::Object_format format_enum;
2869 format_enum = General_options::string_to_object_format(s.c_str());
2870 closure->position_dependent_options().set_format_enum(format_enum);
2873 // Called by the bison parser to handle SEARCH_DIR. This is handled
2874 // exactly like a -L option.
2877 script_add_search_dir(void* closurev, const char* option, size_t length)
2879 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2880 if (closure->command_line() == NULL)
2881 gold_warning(_("%s:%d:%d: ignoring SEARCH_DIR; SEARCH_DIR is only valid"
2882 " for scripts specified via -T/--script"),
2883 closure->filename(), closure->lineno(), closure->charpos());
2884 else if (!closure->command_line()->options().nostdlib())
2886 std::string s = "-L" + std::string(option, length);
2887 script_parse_option(closurev, s.c_str(), s.size());
2891 /* Called by the bison parser to push the lexer into expression
2895 script_push_lex_into_expression_mode(void* closurev)
2897 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2898 closure->push_lex_mode(Lex::EXPRESSION);
2901 /* Called by the bison parser to push the lexer into version
2905 script_push_lex_into_version_mode(void* closurev)
2907 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2908 if (closure->version_script()->is_finalized())
2909 gold_error(_("%s:%d:%d: invalid use of VERSION in input file"),
2910 closure->filename(), closure->lineno(), closure->charpos());
2911 closure->push_lex_mode(Lex::VERSION_SCRIPT);
2914 /* Called by the bison parser to pop the lexer mode. */
2917 script_pop_lex_mode(void* closurev)
2919 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2920 closure->pop_lex_mode();
2923 // Register an entire version node. For example:
2929 // - tag is "GLIBC_2.1"
2930 // - tree contains the information "global: foo"
2931 // - deps contains "GLIBC_2.0"
2934 script_register_vers_node(void*,
2937 struct Version_tree* tree,
2938 struct Version_dependency_list* deps)
2940 gold_assert(tree != NULL);
2941 tree->dependencies = deps;
2943 tree->tag = std::string(tag, taglen);
2946 // Add a dependencies to the list of existing dependencies, if any,
2947 // and return the expanded list.
2949 extern "C" struct Version_dependency_list*
2950 script_add_vers_depend(void* closurev,
2951 struct Version_dependency_list* all_deps,
2952 const char* depend_to_add, int deplen)
2954 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2955 if (all_deps == NULL)
2956 all_deps = closure->version_script()->allocate_dependency_list();
2957 all_deps->dependencies.push_back(std::string(depend_to_add, deplen));
2961 // Add a pattern expression to an existing list of expressions, if any.
2963 extern "C" struct Version_expression_list*
2964 script_new_vers_pattern(void* closurev,
2965 struct Version_expression_list* expressions,
2966 const char* pattern, int patlen, int exact_match)
2968 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2969 if (expressions == NULL)
2970 expressions = closure->version_script()->allocate_expression_list();
2971 expressions->expressions.push_back(
2972 Version_expression(std::string(pattern, patlen),
2973 closure->get_current_language(),
2974 static_cast<bool>(exact_match)));
2978 // Attaches b to the end of a, and clears b. So a = a + b and b = {}.
2980 extern "C" struct Version_expression_list*
2981 script_merge_expressions(struct Version_expression_list* a,
2982 struct Version_expression_list* b)
2984 a->expressions.insert(a->expressions.end(),
2985 b->expressions.begin(), b->expressions.end());
2986 // We could delete b and remove it from expressions_lists_, but
2987 // that's a lot of work. This works just as well.
2988 b->expressions.clear();
2992 // Combine the global and local expressions into a a Version_tree.
2994 extern "C" struct Version_tree*
2995 script_new_vers_node(void* closurev,
2996 struct Version_expression_list* global,
2997 struct Version_expression_list* local)
2999 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3000 Version_tree* tree = closure->version_script()->allocate_version_tree();
3001 tree->global = global;
3002 tree->local = local;
3006 // Handle a transition in language, such as at the
3007 // start or end of 'extern "C++"'
3010 version_script_push_lang(void* closurev, const char* lang, int langlen)
3012 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3013 std::string language(lang, langlen);
3014 Version_script_info::Language code;
3015 if (language.empty() || language == "C")
3016 code = Version_script_info::LANGUAGE_C;
3017 else if (language == "C++")
3018 code = Version_script_info::LANGUAGE_CXX;
3019 else if (language == "Java")
3020 code = Version_script_info::LANGUAGE_JAVA;
3023 char* buf = new char[langlen + 100];
3024 snprintf(buf, langlen + 100,
3025 _("unrecognized version script language '%s'"),
3027 yyerror(closurev, buf);
3029 code = Version_script_info::LANGUAGE_C;
3031 closure->push_language(code);
3035 version_script_pop_lang(void* closurev)
3037 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3038 closure->pop_language();
3041 // Called by the bison parser to start a SECTIONS clause.
3044 script_start_sections(void* closurev)
3046 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3047 closure->script_options()->script_sections()->start_sections();
3048 closure->clear_skip_on_incompatible_target();
3051 // Called by the bison parser to finish a SECTIONS clause.
3054 script_finish_sections(void* closurev)
3056 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3057 closure->script_options()->script_sections()->finish_sections();
3060 // Start processing entries for an output section.
3063 script_start_output_section(void* closurev, const char* name, size_t namelen,
3064 const struct Parser_output_section_header* header)
3066 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3067 closure->script_options()->script_sections()->start_output_section(name,
3072 // Finish processing entries for an output section.
3075 script_finish_output_section(void* closurev,
3076 const struct Parser_output_section_trailer* trail)
3078 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3079 closure->script_options()->script_sections()->finish_output_section(trail);
3082 // Add a data item (e.g., "WORD (0)") to the current output section.
3085 script_add_data(void* closurev, int data_token, Expression* val)
3087 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3089 bool is_signed = true;
3111 closure->script_options()->script_sections()->add_data(size, is_signed, val);
3114 // Add a clause setting the fill value to the current output section.
3117 script_add_fill(void* closurev, Expression* val)
3119 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3120 closure->script_options()->script_sections()->add_fill(val);
3123 // Add a new input section specification to the current output
3127 script_add_input_section(void* closurev,
3128 const struct Input_section_spec* spec,
3131 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3132 bool keep = keepi != 0;
3133 closure->script_options()->script_sections()->add_input_section(spec, keep);
3136 // When we see DATA_SEGMENT_ALIGN we record that following output
3137 // sections may be relro.
3140 script_data_segment_align(void* closurev)
3142 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3143 if (!closure->script_options()->saw_sections_clause())
3144 gold_error(_("%s:%d:%d: DATA_SEGMENT_ALIGN not in SECTIONS clause"),
3145 closure->filename(), closure->lineno(), closure->charpos());
3147 closure->script_options()->script_sections()->data_segment_align();
3150 // When we see DATA_SEGMENT_RELRO_END we know that all output sections
3151 // since DATA_SEGMENT_ALIGN should be relro.
3154 script_data_segment_relro_end(void* closurev)
3156 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3157 if (!closure->script_options()->saw_sections_clause())
3158 gold_error(_("%s:%d:%d: DATA_SEGMENT_ALIGN not in SECTIONS clause"),
3159 closure->filename(), closure->lineno(), closure->charpos());
3161 closure->script_options()->script_sections()->data_segment_relro_end();
3164 // Create a new list of string/sort pairs.
3166 extern "C" String_sort_list_ptr
3167 script_new_string_sort_list(const struct Wildcard_section* string_sort)
3169 return new String_sort_list(1, *string_sort);
3172 // Add an entry to a list of string/sort pairs. The way the parser
3173 // works permits us to simply modify the first parameter, rather than
3176 extern "C" String_sort_list_ptr
3177 script_string_sort_list_add(String_sort_list_ptr pv,
3178 const struct Wildcard_section* string_sort)
3181 return script_new_string_sort_list(string_sort);
3184 pv->push_back(*string_sort);
3189 // Create a new list of strings.
3191 extern "C" String_list_ptr
3192 script_new_string_list(const char* str, size_t len)
3194 return new String_list(1, std::string(str, len));
3197 // Add an element to a list of strings. The way the parser works
3198 // permits us to simply modify the first parameter, rather than copy
3201 extern "C" String_list_ptr
3202 script_string_list_push_back(String_list_ptr pv, const char* str, size_t len)
3205 return script_new_string_list(str, len);
3208 pv->push_back(std::string(str, len));
3213 // Concatenate two string lists. Either or both may be NULL. The way
3214 // the parser works permits us to modify the parameters, rather than
3217 extern "C" String_list_ptr
3218 script_string_list_append(String_list_ptr pv1, String_list_ptr pv2)
3224 pv1->insert(pv1->end(), pv2->begin(), pv2->end());
3228 // Add a new program header.
3231 script_add_phdr(void* closurev, const char* name, size_t namelen,
3232 unsigned int type, const Phdr_info* info)
3234 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3235 bool includes_filehdr = info->includes_filehdr != 0;
3236 bool includes_phdrs = info->includes_phdrs != 0;
3237 bool is_flags_valid = info->is_flags_valid != 0;
3238 Script_sections* ss = closure->script_options()->script_sections();
3239 ss->add_phdr(name, namelen, type, includes_filehdr, includes_phdrs,
3240 is_flags_valid, info->flags, info->load_address);
3241 closure->clear_skip_on_incompatible_target();
3244 // Convert a program header string to a type.
3246 #define PHDR_TYPE(NAME) { #NAME, sizeof(#NAME) - 1, elfcpp::NAME }
3253 } phdr_type_names[] =
3257 PHDR_TYPE(PT_DYNAMIC),
3258 PHDR_TYPE(PT_INTERP),
3260 PHDR_TYPE(PT_SHLIB),
3263 PHDR_TYPE(PT_GNU_EH_FRAME),
3264 PHDR_TYPE(PT_GNU_STACK),
3265 PHDR_TYPE(PT_GNU_RELRO)
3268 extern "C" unsigned int
3269 script_phdr_string_to_type(void* closurev, const char* name, size_t namelen)
3271 for (unsigned int i = 0;
3272 i < sizeof(phdr_type_names) / sizeof(phdr_type_names[0]);
3274 if (namelen == phdr_type_names[i].namelen
3275 && strncmp(name, phdr_type_names[i].name, namelen) == 0)
3276 return phdr_type_names[i].val;
3277 yyerror(closurev, _("unknown PHDR type (try integer)"));
3278 return elfcpp::PT_NULL;
3282 script_saw_segment_start_expression(void* closurev)
3284 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3285 Script_sections* ss = closure->script_options()->script_sections();
3286 ss->set_saw_segment_start_expression(true);
3290 script_set_section_region(void* closurev, const char* name, size_t namelen,
3293 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3294 if (!closure->script_options()->saw_sections_clause())
3296 gold_error(_("%s:%d:%d: MEMORY region '%.*s' referred to outside of "
3298 closure->filename(), closure->lineno(), closure->charpos(),
3299 static_cast<int>(namelen), name);
3303 Script_sections* ss = closure->script_options()->script_sections();
3304 Memory_region* mr = ss->find_memory_region(name, namelen);
3307 gold_error(_("%s:%d:%d: MEMORY region '%.*s' not declared"),
3308 closure->filename(), closure->lineno(), closure->charpos(),
3309 static_cast<int>(namelen), name);
3313 ss->set_memory_region(mr, set_vma);
3317 script_add_memory(void* closurev, const char* name, size_t namelen,
3318 unsigned int attrs, Expression* origin, Expression* length)
3320 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3321 Script_sections* ss = closure->script_options()->script_sections();
3322 ss->add_memory_region(name, namelen, attrs, origin, length);
3325 extern "C" unsigned int
3326 script_parse_memory_attr(void* closurev, const char* attrs, size_t attrlen,
3336 attributes |= MEM_READABLE; break;
3339 attributes |= MEM_READABLE | MEM_WRITEABLE; break;
3342 attributes |= MEM_EXECUTABLE; break;
3345 attributes |= MEM_ALLOCATABLE; break;
3350 attributes |= MEM_INITIALIZED; break;
3352 yyerror(closurev, _("unknown MEMORY attribute"));
3356 attributes = (~ attributes) & MEM_ATTR_MASK;
3362 script_include_directive(void* closurev, const char* filename, size_t length)
3364 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3365 std::string name(filename, length);
3366 Command_line* cmdline = closure->command_line();
3367 read_script_file(name.c_str(), cmdline, &cmdline->script_options(),
3368 PARSING_LINKER_SCRIPT, Lex::LINKER_SCRIPT);
3371 // Functions for memory regions.
3373 extern "C" Expression*
3374 script_exp_function_origin(void* closurev, const char* name, size_t namelen)
3376 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3377 Script_sections* ss = closure->script_options()->script_sections();
3378 Expression* origin = ss->find_memory_region_origin(name, namelen);
3382 gold_error(_("undefined memory region '%s' referenced "
3383 "in ORIGIN expression"),
3385 // Create a dummy expression to prevent crashes later on.
3386 origin = script_exp_integer(0);
3392 extern "C" Expression*
3393 script_exp_function_length(void* closurev, const char* name, size_t namelen)
3395 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3396 Script_sections* ss = closure->script_options()->script_sections();
3397 Expression* length = ss->find_memory_region_length(name, namelen);
3401 gold_error(_("undefined memory region '%s' referenced "
3402 "in LENGTH expression"),
3404 // Create a dummy expression to prevent crashes later on.
3405 length = script_exp_integer(0);