1 // script.cc -- handle linker scripts for gold.
3 // Copyright (C) 2006-2015 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 elfcpp::STT type = elfcpp::STT_NOTYPE;
984 elfcpp::STV vis = elfcpp::STV_DEFAULT;
985 unsigned char nonvis = 0;
986 uint64_t final_val = this->val_->eval_maybe_dot(symtab, layout, true,
988 dot_value, dot_section,
989 §ion, NULL, &type,
990 &vis, &nonvis, false);
991 Sized_symbol<size>* ssym = symtab->get_sized_symbol<size>(this->sym_);
992 ssym->set_value(final_val);
993 ssym->set_type(type);
994 ssym->set_visibility(vis);
995 ssym->set_nonvis(nonvis);
997 ssym->set_output_section(section);
1000 // Set the symbol value if the expression yields an absolute value or
1001 // a value relative to DOT_SECTION.
1004 Symbol_assignment::set_if_absolute(Symbol_table* symtab, const Layout* layout,
1005 bool is_dot_available, uint64_t dot_value,
1006 Output_section* dot_section)
1008 if (this->sym_ == NULL)
1011 Output_section* val_section;
1012 uint64_t val = this->val_->eval_maybe_dot(symtab, layout, false,
1013 is_dot_available, dot_value,
1014 dot_section, &val_section, NULL,
1015 NULL, NULL, NULL, false);
1016 if (val_section != NULL && val_section != dot_section)
1019 if (parameters->target().get_size() == 32)
1021 #if defined(HAVE_TARGET_32_LITTLE) || defined(HAVE_TARGET_32_BIG)
1022 Sized_symbol<32>* ssym = symtab->get_sized_symbol<32>(this->sym_);
1023 ssym->set_value(val);
1028 else if (parameters->target().get_size() == 64)
1030 #if defined(HAVE_TARGET_64_LITTLE) || defined(HAVE_TARGET_64_BIG)
1031 Sized_symbol<64>* ssym = symtab->get_sized_symbol<64>(this->sym_);
1032 ssym->set_value(val);
1039 if (val_section != NULL)
1040 this->sym_->set_output_section(val_section);
1043 // Print for debugging.
1046 Symbol_assignment::print(FILE* f) const
1048 if (this->provide_ && this->hidden_)
1049 fprintf(f, "PROVIDE_HIDDEN(");
1050 else if (this->provide_)
1051 fprintf(f, "PROVIDE(");
1052 else if (this->hidden_)
1055 fprintf(f, "%s = ", this->name_.c_str());
1056 this->val_->print(f);
1058 if (this->provide_ || this->hidden_)
1064 // Class Script_assertion.
1066 // Check the assertion.
1069 Script_assertion::check(const Symbol_table* symtab, const Layout* layout)
1071 if (!this->check_->eval(symtab, layout, true))
1072 gold_error("%s", this->message_.c_str());
1075 // Print for debugging.
1078 Script_assertion::print(FILE* f) const
1080 fprintf(f, "ASSERT(");
1081 this->check_->print(f);
1082 fprintf(f, ", \"%s\")\n", this->message_.c_str());
1085 // Class Script_options.
1087 Script_options::Script_options()
1088 : entry_(), symbol_assignments_(), symbol_definitions_(),
1089 symbol_references_(), version_script_info_(), script_sections_()
1093 // Returns true if NAME is on the list of symbol assignments waiting
1097 Script_options::is_pending_assignment(const char* name)
1099 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin();
1100 p != this->symbol_assignments_.end();
1102 if ((*p)->name() == name)
1107 // Add a symbol to be defined.
1110 Script_options::add_symbol_assignment(const char* name, size_t length,
1111 bool is_defsym, Expression* value,
1112 bool provide, bool hidden)
1114 if (length != 1 || name[0] != '.')
1116 if (this->script_sections_.in_sections_clause())
1118 gold_assert(!is_defsym);
1119 this->script_sections_.add_symbol_assignment(name, length, value,
1124 Symbol_assignment* p = new Symbol_assignment(name, length, is_defsym,
1125 value, provide, hidden);
1126 this->symbol_assignments_.push_back(p);
1131 std::string n(name, length);
1132 this->symbol_definitions_.insert(n);
1133 this->symbol_references_.erase(n);
1138 if (provide || hidden)
1139 gold_error(_("invalid use of PROVIDE for dot symbol"));
1141 // The GNU linker permits assignments to dot outside of SECTIONS
1142 // clauses and treats them as occurring inside, so we don't
1143 // check in_sections_clause here.
1144 this->script_sections_.add_dot_assignment(value);
1148 // Add a reference to a symbol.
1151 Script_options::add_symbol_reference(const char* name, size_t length)
1153 if (length != 1 || name[0] != '.')
1155 std::string n(name, length);
1156 if (this->symbol_definitions_.find(n) == this->symbol_definitions_.end())
1157 this->symbol_references_.insert(n);
1161 // Add an assertion.
1164 Script_options::add_assertion(Expression* check, const char* message,
1167 if (this->script_sections_.in_sections_clause())
1168 this->script_sections_.add_assertion(check, message, messagelen);
1171 Script_assertion* p = new Script_assertion(check, message, messagelen);
1172 this->assertions_.push_back(p);
1176 // Create sections required by any linker scripts.
1179 Script_options::create_script_sections(Layout* layout)
1181 if (this->saw_sections_clause())
1182 this->script_sections_.create_sections(layout);
1185 // Add any symbols we are defining to the symbol table.
1188 Script_options::add_symbols_to_table(Symbol_table* symtab)
1190 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin();
1191 p != this->symbol_assignments_.end();
1193 (*p)->add_to_table(symtab);
1194 this->script_sections_.add_symbols_to_table(symtab);
1197 // Finalize symbol values. Also check assertions.
1200 Script_options::finalize_symbols(Symbol_table* symtab, const Layout* layout)
1202 // We finalize the symbols defined in SECTIONS first, because they
1203 // are the ones which may have changed. This way if symbol outside
1204 // SECTIONS are defined in terms of symbols inside SECTIONS, they
1205 // will get the right value.
1206 this->script_sections_.finalize_symbols(symtab, layout);
1208 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin();
1209 p != this->symbol_assignments_.end();
1211 (*p)->finalize(symtab, layout);
1213 for (Assertions::iterator p = this->assertions_.begin();
1214 p != this->assertions_.end();
1216 (*p)->check(symtab, layout);
1219 // Set section addresses. We set all the symbols which have absolute
1220 // values. Then we let the SECTIONS clause do its thing. This
1221 // returns the segment which holds the file header and segment
1225 Script_options::set_section_addresses(Symbol_table* symtab, Layout* layout)
1227 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin();
1228 p != this->symbol_assignments_.end();
1230 (*p)->set_if_absolute(symtab, layout, false, 0, NULL);
1232 return this->script_sections_.set_section_addresses(symtab, layout);
1235 // This class holds data passed through the parser to the lexer and to
1236 // the parser support functions. This avoids global variables. We
1237 // can't use global variables because we need not be called by a
1238 // singleton thread.
1240 class Parser_closure
1243 Parser_closure(const char* filename,
1244 const Position_dependent_options& posdep_options,
1245 bool parsing_defsym, bool in_group, bool is_in_sysroot,
1246 Command_line* command_line,
1247 Script_options* script_options,
1249 bool skip_on_incompatible_target,
1250 Script_info* script_info)
1251 : filename_(filename), posdep_options_(posdep_options),
1252 parsing_defsym_(parsing_defsym), in_group_(in_group),
1253 is_in_sysroot_(is_in_sysroot),
1254 skip_on_incompatible_target_(skip_on_incompatible_target),
1255 found_incompatible_target_(false),
1256 command_line_(command_line), script_options_(script_options),
1257 version_script_info_(script_options->version_script_info()),
1258 lex_(lex), lineno_(0), charpos_(0), lex_mode_stack_(), inputs_(NULL),
1259 script_info_(script_info)
1261 // We start out processing C symbols in the default lex mode.
1262 this->language_stack_.push_back(Version_script_info::LANGUAGE_C);
1263 this->lex_mode_stack_.push_back(lex->mode());
1266 // Return the file name.
1269 { return this->filename_; }
1271 // Return the position dependent options. The caller may modify
1273 Position_dependent_options&
1274 position_dependent_options()
1275 { return this->posdep_options_; }
1277 // Whether we are parsing a --defsym.
1279 parsing_defsym() const
1280 { return this->parsing_defsym_; }
1282 // Return whether this script is being run in a group.
1285 { return this->in_group_; }
1287 // Return whether this script was found using a directory in the
1290 is_in_sysroot() const
1291 { return this->is_in_sysroot_; }
1293 // Whether to skip to the next file with the same name if we find an
1294 // incompatible target in an OUTPUT_FORMAT statement.
1296 skip_on_incompatible_target() const
1297 { return this->skip_on_incompatible_target_; }
1299 // Stop skipping to the next file on an incompatible target. This
1300 // is called when we make some unrevocable change to the data
1303 clear_skip_on_incompatible_target()
1304 { this->skip_on_incompatible_target_ = false; }
1306 // Whether we found an incompatible target in an OUTPUT_FORMAT
1309 found_incompatible_target() const
1310 { return this->found_incompatible_target_; }
1312 // Note that we found an incompatible target.
1314 set_found_incompatible_target()
1315 { this->found_incompatible_target_ = true; }
1317 // Returns the Command_line structure passed in at constructor time.
1318 // This value may be NULL. The caller may modify this, which modifies
1319 // the passed-in Command_line object (not a copy).
1322 { return this->command_line_; }
1324 // Return the options which may be set by a script.
1327 { return this->script_options_; }
1329 // Return the object in which version script information should be stored.
1330 Version_script_info*
1332 { return this->version_script_info_; }
1334 // Return the next token, and advance.
1338 const Token* token = this->lex_->next_token();
1339 this->lineno_ = token->lineno();
1340 this->charpos_ = token->charpos();
1344 // Set a new lexer mode, pushing the current one.
1346 push_lex_mode(Lex::Mode mode)
1348 this->lex_mode_stack_.push_back(this->lex_->mode());
1349 this->lex_->set_mode(mode);
1352 // Pop the lexer mode.
1356 gold_assert(!this->lex_mode_stack_.empty());
1357 this->lex_->set_mode(this->lex_mode_stack_.back());
1358 this->lex_mode_stack_.pop_back();
1361 // Return the current lexer mode.
1364 { return this->lex_mode_stack_.back(); }
1366 // Return the line number of the last token.
1369 { return this->lineno_; }
1371 // Return the character position in the line of the last token.
1374 { return this->charpos_; }
1376 // Return the list of input files, creating it if necessary. This
1377 // is a space leak--we never free the INPUTS_ pointer.
1381 if (this->inputs_ == NULL)
1382 this->inputs_ = new Input_arguments();
1383 return this->inputs_;
1386 // Return whether we saw any input files.
1389 { return this->inputs_ != NULL && !this->inputs_->empty(); }
1391 // Return the current language being processed in a version script
1392 // (eg, "C++"). The empty string represents unmangled C names.
1393 Version_script_info::Language
1394 get_current_language() const
1395 { return this->language_stack_.back(); }
1397 // Push a language onto the stack when entering an extern block.
1399 push_language(Version_script_info::Language lang)
1400 { this->language_stack_.push_back(lang); }
1402 // Pop a language off of the stack when exiting an extern block.
1406 gold_assert(!this->language_stack_.empty());
1407 this->language_stack_.pop_back();
1410 // Return a pointer to the incremental info.
1413 { return this->script_info_; }
1416 // The name of the file we are reading.
1417 const char* filename_;
1418 // The position dependent options.
1419 Position_dependent_options posdep_options_;
1420 // True if we are parsing a --defsym.
1421 bool parsing_defsym_;
1422 // Whether we are currently in a --start-group/--end-group.
1424 // Whether the script was found in a sysrooted directory.
1425 bool is_in_sysroot_;
1426 // If this is true, then if we find an OUTPUT_FORMAT with an
1427 // incompatible target, then we tell the parser to abort so that we
1428 // can search for the next file with the same name.
1429 bool skip_on_incompatible_target_;
1430 // True if we found an OUTPUT_FORMAT with an incompatible target.
1431 bool found_incompatible_target_;
1432 // May be NULL if the user chooses not to pass one in.
1433 Command_line* command_line_;
1434 // Options which may be set from any linker script.
1435 Script_options* script_options_;
1436 // Information parsed from a version script.
1437 Version_script_info* version_script_info_;
1440 // The line number of the last token returned by next_token.
1442 // The column number of the last token returned by next_token.
1444 // A stack of lexer modes.
1445 std::vector<Lex::Mode> lex_mode_stack_;
1446 // A stack of which extern/language block we're inside. Can be C++,
1447 // java, or empty for C.
1448 std::vector<Version_script_info::Language> language_stack_;
1449 // New input files found to add to the link.
1450 Input_arguments* inputs_;
1451 // Pointer to incremental linking info.
1452 Script_info* script_info_;
1455 // FILE was found as an argument on the command line. Try to read it
1456 // as a script. Return true if the file was handled.
1459 read_input_script(Workqueue* workqueue, Symbol_table* symtab, Layout* layout,
1460 Dirsearch* dirsearch, int dirindex,
1461 Input_objects* input_objects, Mapfile* mapfile,
1462 Input_group* input_group,
1463 const Input_argument* input_argument,
1464 Input_file* input_file, Task_token* next_blocker,
1465 bool* used_next_blocker)
1467 *used_next_blocker = false;
1469 std::string input_string;
1470 Lex::read_file(input_file, &input_string);
1472 Lex lex(input_string.c_str(), input_string.length(), PARSING_LINKER_SCRIPT);
1474 Script_info* script_info = NULL;
1475 if (layout->incremental_inputs() != NULL)
1477 const std::string& filename = input_file->filename();
1478 Timespec mtime = input_file->file().get_mtime();
1479 unsigned int arg_serial = input_argument->file().arg_serial();
1480 script_info = new Script_info(filename);
1481 layout->incremental_inputs()->report_script(script_info, arg_serial,
1485 Parser_closure closure(input_file->filename().c_str(),
1486 input_argument->file().options(),
1488 input_group != NULL,
1489 input_file->is_in_sysroot(),
1491 layout->script_options(),
1493 input_file->will_search_for(),
1496 bool old_saw_sections_clause =
1497 layout->script_options()->saw_sections_clause();
1499 if (yyparse(&closure) != 0)
1501 if (closure.found_incompatible_target())
1503 Read_symbols::incompatible_warning(input_argument, input_file);
1504 Read_symbols::requeue(workqueue, input_objects, symtab, layout,
1505 dirsearch, dirindex, mapfile, input_argument,
1506 input_group, next_blocker);
1512 if (!old_saw_sections_clause
1513 && layout->script_options()->saw_sections_clause()
1514 && layout->have_added_input_section())
1515 gold_error(_("%s: SECTIONS seen after other input files; try -T/--script"),
1516 input_file->filename().c_str());
1518 if (!closure.saw_inputs())
1521 Task_token* this_blocker = NULL;
1522 for (Input_arguments::const_iterator p = closure.inputs()->begin();
1523 p != closure.inputs()->end();
1527 if (p + 1 == closure.inputs()->end())
1531 nb = new Task_token(true);
1534 workqueue->queue_soon(new Read_symbols(input_objects, symtab,
1535 layout, dirsearch, 0, mapfile, &*p,
1536 input_group, NULL, this_blocker, nb));
1540 *used_next_blocker = true;
1545 // Helper function for read_version_script(), read_commandline_script() and
1546 // script_include_directive(). Processes the given file in the mode indicated
1547 // by first_token and lex_mode.
1550 read_script_file(const char* filename, Command_line* cmdline,
1551 Script_options* script_options,
1552 int first_token, Lex::Mode lex_mode)
1554 Dirsearch dirsearch;
1555 std::string name = filename;
1557 // If filename is a relative filename, search for it manually using "." +
1558 // cmdline->options()->library_path() -- not dirsearch.
1559 if (!IS_ABSOLUTE_PATH(filename))
1561 const General_options::Dir_list& search_path =
1562 cmdline->options().library_path();
1563 name = Dirsearch::find_file_in_dir_list(name, search_path, ".");
1566 // The file locking code wants to record a Task, but we haven't
1567 // started the workqueue yet. This is only for debugging purposes,
1568 // so we invent a fake value.
1569 const Task* task = reinterpret_cast<const Task*>(-1);
1571 // We don't want this file to be opened in binary mode.
1572 Position_dependent_options posdep = cmdline->position_dependent_options();
1573 if (posdep.format_enum() == General_options::OBJECT_FORMAT_BINARY)
1574 posdep.set_format_enum(General_options::OBJECT_FORMAT_ELF);
1575 Input_file_argument input_argument(name.c_str(),
1576 Input_file_argument::INPUT_FILE_TYPE_FILE,
1578 Input_file input_file(&input_argument);
1580 if (!input_file.open(dirsearch, task, &dummy))
1583 std::string input_string;
1584 Lex::read_file(&input_file, &input_string);
1586 Lex lex(input_string.c_str(), input_string.length(), first_token);
1587 lex.set_mode(lex_mode);
1589 Parser_closure closure(filename,
1590 cmdline->position_dependent_options(),
1591 first_token == Lex::DYNAMIC_LIST,
1593 input_file.is_in_sysroot(),
1599 if (yyparse(&closure) != 0)
1601 input_file.file().unlock(task);
1605 input_file.file().unlock(task);
1607 gold_assert(!closure.saw_inputs());
1612 // FILENAME was found as an argument to --script (-T).
1613 // Read it as a script, and execute its contents immediately.
1616 read_commandline_script(const char* filename, Command_line* cmdline)
1618 return read_script_file(filename, cmdline, &cmdline->script_options(),
1619 PARSING_LINKER_SCRIPT, Lex::LINKER_SCRIPT);
1622 // FILENAME was found as an argument to --version-script. Read it as
1623 // a version script, and store its contents in
1624 // cmdline->script_options()->version_script_info().
1627 read_version_script(const char* filename, Command_line* cmdline)
1629 return read_script_file(filename, cmdline, &cmdline->script_options(),
1630 PARSING_VERSION_SCRIPT, Lex::VERSION_SCRIPT);
1633 // FILENAME was found as an argument to --dynamic-list. Read it as a
1634 // list of symbols, and store its contents in DYNAMIC_LIST.
1637 read_dynamic_list(const char* filename, Command_line* cmdline,
1638 Script_options* dynamic_list)
1640 return read_script_file(filename, cmdline, dynamic_list,
1641 PARSING_DYNAMIC_LIST, Lex::DYNAMIC_LIST);
1644 // Implement the --defsym option on the command line. Return true if
1648 Script_options::define_symbol(const char* definition)
1650 Lex lex(definition, strlen(definition), PARSING_DEFSYM);
1651 lex.set_mode(Lex::EXPRESSION);
1654 Position_dependent_options posdep_options;
1656 Parser_closure closure("command line", posdep_options, true,
1657 false, false, NULL, this, &lex, false, NULL);
1659 if (yyparse(&closure) != 0)
1662 gold_assert(!closure.saw_inputs());
1667 // Print the script to F for debugging.
1670 Script_options::print(FILE* f) const
1672 fprintf(f, "%s: Dumping linker script\n", program_name);
1674 if (!this->entry_.empty())
1675 fprintf(f, "ENTRY(%s)\n", this->entry_.c_str());
1677 for (Symbol_assignments::const_iterator p =
1678 this->symbol_assignments_.begin();
1679 p != this->symbol_assignments_.end();
1683 for (Assertions::const_iterator p = this->assertions_.begin();
1684 p != this->assertions_.end();
1688 this->script_sections_.print(f);
1690 this->version_script_info_.print(f);
1693 // Manage mapping from keywords to the codes expected by the bison
1694 // parser. We construct one global object for each lex mode with
1697 class Keyword_to_parsecode
1700 // The structure which maps keywords to parsecodes.
1701 struct Keyword_parsecode
1704 const char* keyword;
1705 // Corresponding parsecode.
1709 Keyword_to_parsecode(const Keyword_parsecode* keywords,
1711 : keyword_parsecodes_(keywords), keyword_count_(keyword_count)
1714 // Return the parsecode corresponding KEYWORD, or 0 if it is not a
1717 keyword_to_parsecode(const char* keyword, size_t len) const;
1720 const Keyword_parsecode* keyword_parsecodes_;
1721 const int keyword_count_;
1724 // Mapping from keyword string to keyword parsecode. This array must
1725 // be kept in sorted order. Parsecodes are looked up using bsearch.
1726 // This array must correspond to the list of parsecodes in yyscript.y.
1728 static const Keyword_to_parsecode::Keyword_parsecode
1729 script_keyword_parsecodes[] =
1731 { "ABSOLUTE", ABSOLUTE },
1733 { "ALIGN", ALIGN_K },
1734 { "ALIGNOF", ALIGNOF },
1735 { "ASSERT", ASSERT_K },
1736 { "AS_NEEDED", AS_NEEDED },
1741 { "CONSTANT", CONSTANT },
1742 { "CONSTRUCTORS", CONSTRUCTORS },
1744 { "CREATE_OBJECT_SYMBOLS", CREATE_OBJECT_SYMBOLS },
1745 { "DATA_SEGMENT_ALIGN", DATA_SEGMENT_ALIGN },
1746 { "DATA_SEGMENT_END", DATA_SEGMENT_END },
1747 { "DATA_SEGMENT_RELRO_END", DATA_SEGMENT_RELRO_END },
1748 { "DEFINED", DEFINED },
1751 { "EXCLUDE_FILE", EXCLUDE_FILE },
1752 { "EXTERN", EXTERN },
1755 { "FORCE_COMMON_ALLOCATION", FORCE_COMMON_ALLOCATION },
1758 { "INCLUDE", INCLUDE },
1760 { "INHIBIT_COMMON_ALLOCATION", INHIBIT_COMMON_ALLOCATION },
1763 { "LENGTH", LENGTH },
1764 { "LOADADDR", LOADADDR },
1768 { "MEMORY", MEMORY },
1771 { "NOCROSSREFS", NOCROSSREFS },
1772 { "NOFLOAT", NOFLOAT },
1773 { "NOLOAD", NOLOAD },
1774 { "ONLY_IF_RO", ONLY_IF_RO },
1775 { "ONLY_IF_RW", ONLY_IF_RW },
1776 { "OPTION", OPTION },
1777 { "ORIGIN", ORIGIN },
1778 { "OUTPUT", OUTPUT },
1779 { "OUTPUT_ARCH", OUTPUT_ARCH },
1780 { "OUTPUT_FORMAT", OUTPUT_FORMAT },
1781 { "OVERLAY", OVERLAY },
1783 { "PROVIDE", PROVIDE },
1784 { "PROVIDE_HIDDEN", PROVIDE_HIDDEN },
1786 { "SEARCH_DIR", SEARCH_DIR },
1787 { "SECTIONS", SECTIONS },
1788 { "SEGMENT_START", SEGMENT_START },
1790 { "SIZEOF", SIZEOF },
1791 { "SIZEOF_HEADERS", SIZEOF_HEADERS },
1792 { "SORT", SORT_BY_NAME },
1793 { "SORT_BY_ALIGNMENT", SORT_BY_ALIGNMENT },
1794 { "SORT_BY_NAME", SORT_BY_NAME },
1795 { "SPECIAL", SPECIAL },
1797 { "STARTUP", STARTUP },
1798 { "SUBALIGN", SUBALIGN },
1799 { "SYSLIB", SYSLIB },
1800 { "TARGET", TARGET_K },
1801 { "TRUNCATE", TRUNCATE },
1802 { "VERSION", VERSIONK },
1803 { "global", GLOBAL },
1809 { "sizeof_headers", SIZEOF_HEADERS },
1812 static const Keyword_to_parsecode
1813 script_keywords(&script_keyword_parsecodes[0],
1814 (sizeof(script_keyword_parsecodes)
1815 / sizeof(script_keyword_parsecodes[0])));
1817 static const Keyword_to_parsecode::Keyword_parsecode
1818 version_script_keyword_parsecodes[] =
1820 { "extern", EXTERN },
1821 { "global", GLOBAL },
1825 static const Keyword_to_parsecode
1826 version_script_keywords(&version_script_keyword_parsecodes[0],
1827 (sizeof(version_script_keyword_parsecodes)
1828 / sizeof(version_script_keyword_parsecodes[0])));
1830 static const Keyword_to_parsecode::Keyword_parsecode
1831 dynamic_list_keyword_parsecodes[] =
1833 { "extern", EXTERN },
1836 static const Keyword_to_parsecode
1837 dynamic_list_keywords(&dynamic_list_keyword_parsecodes[0],
1838 (sizeof(dynamic_list_keyword_parsecodes)
1839 / sizeof(dynamic_list_keyword_parsecodes[0])));
1843 // Comparison function passed to bsearch.
1855 ktt_compare(const void* keyv, const void* kttv)
1857 const Ktt_key* key = static_cast<const Ktt_key*>(keyv);
1858 const Keyword_to_parsecode::Keyword_parsecode* ktt =
1859 static_cast<const Keyword_to_parsecode::Keyword_parsecode*>(kttv);
1860 int i = strncmp(key->str, ktt->keyword, key->len);
1863 if (ktt->keyword[key->len] != '\0')
1868 } // End extern "C".
1871 Keyword_to_parsecode::keyword_to_parsecode(const char* keyword,
1877 void* kttv = bsearch(&key,
1878 this->keyword_parsecodes_,
1879 this->keyword_count_,
1880 sizeof(this->keyword_parsecodes_[0]),
1884 Keyword_parsecode* ktt = static_cast<Keyword_parsecode*>(kttv);
1885 return ktt->parsecode;
1888 // The following structs are used within the VersionInfo class as well
1889 // as in the bison helper functions. They store the information
1890 // parsed from the version script.
1892 // A single version expression.
1893 // For example, pattern="std::map*" and language="C++".
1894 struct Version_expression
1896 Version_expression(const std::string& a_pattern,
1897 Version_script_info::Language a_language,
1899 : pattern(a_pattern), language(a_language), exact_match(a_exact_match),
1900 was_matched_by_symbol(false)
1903 std::string pattern;
1904 Version_script_info::Language language;
1905 // If false, we use glob() to match pattern. If true, we use strcmp().
1907 // True if --no-undefined-version is in effect and we found this
1908 // version in get_symbol_version. We use mutable because this
1909 // struct is generally not modifiable after it has been created.
1910 mutable bool was_matched_by_symbol;
1913 // A list of expressions.
1914 struct Version_expression_list
1916 std::vector<struct Version_expression> expressions;
1919 // A list of which versions upon which another version depends.
1920 // Strings should be from the Stringpool.
1921 struct Version_dependency_list
1923 std::vector<std::string> dependencies;
1926 // The total definition of a version. It includes the tag for the
1927 // version, its global and local expressions, and any dependencies.
1931 : tag(), global(NULL), local(NULL), dependencies(NULL)
1935 const struct Version_expression_list* global;
1936 const struct Version_expression_list* local;
1937 const struct Version_dependency_list* dependencies;
1940 // Helper class that calls cplus_demangle when needed and takes care of freeing
1943 class Lazy_demangler
1946 Lazy_demangler(const char* symbol, int options)
1947 : symbol_(symbol), options_(options), demangled_(NULL), did_demangle_(false)
1951 { free(this->demangled_); }
1953 // Return the demangled name. The actual demangling happens on the first call,
1954 // and the result is later cached.
1959 // The symbol to demangle.
1960 const char* symbol_;
1961 // Option flags to pass to cplus_demagle.
1963 // The cached demangled value, or NULL if demangling didn't happen yet or
1966 // Whether we already called cplus_demangle
1970 // Return the demangled name. The actual demangling happens on the first call,
1971 // and the result is later cached. Returns NULL if the symbol cannot be
1975 Lazy_demangler::get()
1977 if (!this->did_demangle_)
1979 this->demangled_ = cplus_demangle(this->symbol_, this->options_);
1980 this->did_demangle_ = true;
1982 return this->demangled_;
1985 // Class Version_script_info.
1987 Version_script_info::Version_script_info()
1988 : dependency_lists_(), expression_lists_(), version_trees_(), globs_(),
1989 default_version_(NULL), default_is_global_(false), is_finalized_(false)
1991 for (int i = 0; i < LANGUAGE_COUNT; ++i)
1992 this->exact_[i] = NULL;
1995 Version_script_info::~Version_script_info()
1999 // Forget all the known version script information.
2002 Version_script_info::clear()
2004 for (size_t k = 0; k < this->dependency_lists_.size(); ++k)
2005 delete this->dependency_lists_[k];
2006 this->dependency_lists_.clear();
2007 for (size_t k = 0; k < this->version_trees_.size(); ++k)
2008 delete this->version_trees_[k];
2009 this->version_trees_.clear();
2010 for (size_t k = 0; k < this->expression_lists_.size(); ++k)
2011 delete this->expression_lists_[k];
2012 this->expression_lists_.clear();
2015 // Finalize the version script information.
2018 Version_script_info::finalize()
2020 if (!this->is_finalized_)
2022 this->build_lookup_tables();
2023 this->is_finalized_ = true;
2027 // Return all the versions.
2029 std::vector<std::string>
2030 Version_script_info::get_versions() const
2032 std::vector<std::string> ret;
2033 for (size_t j = 0; j < this->version_trees_.size(); ++j)
2034 if (!this->version_trees_[j]->tag.empty())
2035 ret.push_back(this->version_trees_[j]->tag);
2039 // Return the dependencies of VERSION.
2041 std::vector<std::string>
2042 Version_script_info::get_dependencies(const char* version) const
2044 std::vector<std::string> ret;
2045 for (size_t j = 0; j < this->version_trees_.size(); ++j)
2046 if (this->version_trees_[j]->tag == version)
2048 const struct Version_dependency_list* deps =
2049 this->version_trees_[j]->dependencies;
2051 for (size_t k = 0; k < deps->dependencies.size(); ++k)
2052 ret.push_back(deps->dependencies[k]);
2058 // A version script essentially maps a symbol name to a version tag
2059 // and an indication of whether symbol is global or local within that
2060 // version tag. Each symbol maps to at most one version tag.
2061 // Unfortunately, in practice, version scripts are ambiguous, and list
2062 // symbols multiple times. Thus, we have to document the matching
2065 // This is a description of what the GNU linker does as of 2010-01-11.
2066 // It walks through the version tags in the order in which they appear
2067 // in the version script. For each tag, it first walks through the
2068 // global patterns for that tag, then the local patterns. When
2069 // looking at a single pattern, it first applies any language specific
2070 // demangling as specified for the pattern, and then matches the
2071 // resulting symbol name to the pattern. If it finds an exact match
2072 // for a literal pattern (a pattern enclosed in quotes or with no
2073 // wildcard characters), then that is the match that it uses. If
2074 // finds a match with a wildcard pattern, then it saves it and
2075 // continues searching. Wildcard patterns that are exactly "*" are
2076 // saved separately.
2078 // If no exact match with a literal pattern is ever found, then if a
2079 // wildcard match with a global pattern was found it is used,
2080 // otherwise if a wildcard match with a local pattern was found it is
2083 // This is the result:
2084 // * If there is an exact match, then we use the first tag in the
2085 // version script where it matches.
2086 // + If the exact match in that tag is global, it is used.
2087 // + Otherwise the exact match in that tag is local, and is used.
2088 // * Otherwise, if there is any match with a global 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 "*". If there is no match with a local
2093 // wildcard pattern which is not "*", then we use the *last*
2094 // match with a global "*". Otherwise, continue.
2095 // * Otherwise, if there is any match with a local wildcard pattern:
2096 // + If there is any match with a wildcard pattern which is not
2097 // "*", then we use the tag in which the *last* such pattern
2099 // + Otherwise, we matched "*", and we use the tag in which the
2100 // *last* such match occurred.
2102 // There is an additional wrinkle. When the GNU linker finds a symbol
2103 // with a version defined in an object file due to a .symver
2104 // directive, it looks up that symbol name in that version tag. If it
2105 // finds it, it matches the symbol name against the patterns for that
2106 // version. If there is no match with a global pattern, but there is
2107 // a match with a local pattern, then the GNU linker marks the symbol
2110 // We want gold to be generally compatible, but we also want gold to
2111 // be fast. These are the rules that gold implements:
2112 // * If there is an exact match for the mangled name, we use it.
2113 // + If there is more than one exact match, we give a warning, and
2114 // we use the first tag in the script which matches.
2115 // + If a symbol has an exact match as both global and local for
2116 // the same version tag, we give an error.
2117 // * Otherwise, we look for an extern C++ or an extern Java exact
2118 // match. If we find an exact match, we use it.
2119 // + If there is more than one exact match, we give a warning, and
2120 // we use the first tag in the script which matches.
2121 // + If a symbol has an exact match as both global and local for
2122 // the same version tag, we give an error.
2123 // * Otherwise, we look through the wildcard patterns, ignoring "*"
2124 // patterns. We look through the version tags in reverse order.
2125 // For each version tag, we look through the global patterns and
2126 // then the local patterns. We use the first match we find (i.e.,
2127 // the last matching version tag in the file).
2128 // * Otherwise, we use the "*" pattern if there is one. We give an
2129 // error if there are multiple "*" patterns.
2131 // At least for now, gold does not look up the version tag for a
2132 // symbol version found in an object file to see if it should be
2133 // forced local. There are other ways to force a symbol to be local,
2134 // and I don't understand why this one is useful.
2136 // Build a set of fast lookup tables for a version script.
2139 Version_script_info::build_lookup_tables()
2141 size_t size = this->version_trees_.size();
2142 for (size_t j = 0; j < size; ++j)
2144 const Version_tree* v = this->version_trees_[j];
2145 this->build_expression_list_lookup(v->local, v, false);
2146 this->build_expression_list_lookup(v->global, v, true);
2150 // If a pattern has backlashes but no unquoted wildcard characters,
2151 // then we apply backslash unquoting and look for an exact match.
2152 // Otherwise we treat it as a wildcard pattern. This function returns
2153 // true for a wildcard pattern. Otherwise, it does backslash
2154 // unquoting on *PATTERN and returns false. If this returns true,
2155 // *PATTERN may have been partially unquoted.
2158 Version_script_info::unquote(std::string* pattern) const
2160 bool saw_backslash = false;
2161 size_t len = pattern->length();
2163 for (size_t i = 0; i < len; ++i)
2166 saw_backslash = false;
2169 switch ((*pattern)[i])
2171 case '?': case '[': case '*':
2174 saw_backslash = true;
2182 (*pattern)[j] = (*pattern)[i];
2188 // Add an exact match for MATCH to *PE. The result of the match is
2192 Version_script_info::add_exact_match(const std::string& match,
2193 const Version_tree* v, bool is_global,
2194 const Version_expression* ve,
2197 std::pair<Exact::iterator, bool> ins =
2198 pe->insert(std::make_pair(match, Version_tree_match(v, is_global, ve)));
2201 // This is the first time we have seen this match.
2205 Version_tree_match& vtm(ins.first->second);
2206 if (vtm.real->tag != v->tag)
2208 // This is an ambiguous match. We still return the
2209 // first version that we found in the script, but we
2210 // record the new version to issue a warning if we
2211 // wind up looking up this symbol.
2212 if (vtm.ambiguous == NULL)
2215 else if (is_global != vtm.is_global)
2217 // We have a match for both the global and local entries for a
2218 // version tag. That's got to be wrong.
2219 gold_error(_("'%s' appears as both a global and a local symbol "
2220 "for version '%s' in script"),
2221 match.c_str(), v->tag.c_str());
2225 // Build fast lookup information for EXPLIST and store it in LOOKUP.
2226 // All matches go to V, and IS_GLOBAL is true if they are global
2230 Version_script_info::build_expression_list_lookup(
2231 const Version_expression_list* explist,
2232 const Version_tree* v,
2235 if (explist == NULL)
2237 size_t size = explist->expressions.size();
2238 for (size_t i = 0; i < size; ++i)
2240 const Version_expression& exp(explist->expressions[i]);
2242 if (exp.pattern.length() == 1 && exp.pattern[0] == '*')
2244 if (this->default_version_ != NULL
2245 && this->default_version_->tag != v->tag)
2246 gold_warning(_("wildcard match appears in both version '%s' "
2247 "and '%s' in script"),
2248 this->default_version_->tag.c_str(), v->tag.c_str());
2249 else if (this->default_version_ != NULL
2250 && this->default_is_global_ != is_global)
2251 gold_error(_("wildcard match appears as both global and local "
2252 "in version '%s' in script"),
2254 this->default_version_ = v;
2255 this->default_is_global_ = is_global;
2259 std::string pattern = exp.pattern;
2260 if (!exp.exact_match)
2262 if (this->unquote(&pattern))
2264 this->globs_.push_back(Glob(&exp, v, is_global));
2269 if (this->exact_[exp.language] == NULL)
2270 this->exact_[exp.language] = new Exact();
2271 this->add_exact_match(pattern, v, is_global, &exp,
2272 this->exact_[exp.language]);
2276 // Return the name to match given a name, a language code, and two
2280 Version_script_info::get_name_to_match(const char* name,
2282 Lazy_demangler* cpp_demangler,
2283 Lazy_demangler* java_demangler) const
2290 return cpp_demangler->get();
2292 return java_demangler->get();
2298 // Look up SYMBOL_NAME in the list of versions. Return true if the
2299 // symbol is found, false if not. If the symbol is found, then if
2300 // PVERSION is not NULL, set *PVERSION to the version tag, and if
2301 // P_IS_GLOBAL is not NULL, set *P_IS_GLOBAL according to whether the
2302 // symbol is global or not.
2305 Version_script_info::get_symbol_version(const char* symbol_name,
2306 std::string* pversion,
2307 bool* p_is_global) const
2309 Lazy_demangler cpp_demangled_name(symbol_name, DMGL_ANSI | DMGL_PARAMS);
2310 Lazy_demangler java_demangled_name(symbol_name,
2311 DMGL_ANSI | DMGL_PARAMS | DMGL_JAVA);
2313 gold_assert(this->is_finalized_);
2314 for (int i = 0; i < LANGUAGE_COUNT; ++i)
2316 Exact* exact = this->exact_[i];
2320 const char* name_to_match = this->get_name_to_match(symbol_name, i,
2321 &cpp_demangled_name,
2322 &java_demangled_name);
2323 if (name_to_match == NULL)
2325 // If the name can not be demangled, the GNU linker goes
2326 // ahead and tries to match it anyhow. That does not
2327 // make sense to me and I have not implemented it.
2331 Exact::const_iterator pe = exact->find(name_to_match);
2332 if (pe != exact->end())
2334 const Version_tree_match& vtm(pe->second);
2335 if (vtm.ambiguous != NULL)
2336 gold_warning(_("using '%s' as version for '%s' which is also "
2337 "named in version '%s' in script"),
2338 vtm.real->tag.c_str(), name_to_match,
2339 vtm.ambiguous->tag.c_str());
2341 if (pversion != NULL)
2342 *pversion = vtm.real->tag;
2343 if (p_is_global != NULL)
2344 *p_is_global = vtm.is_global;
2346 // If we are using --no-undefined-version, and this is a
2347 // global symbol, we have to record that we have found this
2348 // symbol, so that we don't warn about it. We have to do
2349 // this now, because otherwise we have no way to get from a
2350 // non-C language back to the demangled name that we
2352 if (p_is_global != NULL && vtm.is_global)
2353 vtm.expression->was_matched_by_symbol = true;
2359 // Look through the glob patterns in reverse order.
2361 for (Globs::const_reverse_iterator p = this->globs_.rbegin();
2362 p != this->globs_.rend();
2365 int language = p->expression->language;
2366 const char* name_to_match = this->get_name_to_match(symbol_name,
2368 &cpp_demangled_name,
2369 &java_demangled_name);
2370 if (name_to_match == NULL)
2373 if (fnmatch(p->expression->pattern.c_str(), name_to_match,
2376 if (pversion != NULL)
2377 *pversion = p->version->tag;
2378 if (p_is_global != NULL)
2379 *p_is_global = p->is_global;
2384 // Finally, there may be a wildcard.
2385 if (this->default_version_ != NULL)
2387 if (pversion != NULL)
2388 *pversion = this->default_version_->tag;
2389 if (p_is_global != NULL)
2390 *p_is_global = this->default_is_global_;
2397 // Give an error if any exact symbol names (not wildcards) appear in a
2398 // version script, but there is no such symbol.
2401 Version_script_info::check_unmatched_names(const Symbol_table* symtab) const
2403 for (size_t i = 0; i < this->version_trees_.size(); ++i)
2405 const Version_tree* vt = this->version_trees_[i];
2406 if (vt->global == NULL)
2408 for (size_t j = 0; j < vt->global->expressions.size(); ++j)
2410 const Version_expression& expression(vt->global->expressions[j]);
2412 // Ignore cases where we used the version because we saw a
2413 // symbol that we looked up. Note that
2414 // WAS_MATCHED_BY_SYMBOL will be true even if the symbol was
2415 // not a definition. That's OK as in that case we most
2416 // likely gave an undefined symbol error anyhow.
2417 if (expression.was_matched_by_symbol)
2420 // Just ignore names which are in languages other than C.
2421 // We have no way to look them up in the symbol table.
2422 if (expression.language != LANGUAGE_C)
2425 // Remove backslash quoting, and ignore wildcard patterns.
2426 std::string pattern = expression.pattern;
2427 if (!expression.exact_match)
2429 if (this->unquote(&pattern))
2433 if (symtab->lookup(pattern.c_str(), vt->tag.c_str()) == NULL)
2434 gold_error(_("version script assignment of %s to symbol %s "
2435 "failed: symbol not defined"),
2436 vt->tag.c_str(), pattern.c_str());
2441 struct Version_dependency_list*
2442 Version_script_info::allocate_dependency_list()
2444 dependency_lists_.push_back(new Version_dependency_list);
2445 return dependency_lists_.back();
2448 struct Version_expression_list*
2449 Version_script_info::allocate_expression_list()
2451 expression_lists_.push_back(new Version_expression_list);
2452 return expression_lists_.back();
2455 struct Version_tree*
2456 Version_script_info::allocate_version_tree()
2458 version_trees_.push_back(new Version_tree);
2459 return version_trees_.back();
2462 // Print for debugging.
2465 Version_script_info::print(FILE* f) const
2470 fprintf(f, "VERSION {");
2472 for (size_t i = 0; i < this->version_trees_.size(); ++i)
2474 const Version_tree* vt = this->version_trees_[i];
2476 if (vt->tag.empty())
2479 fprintf(f, " %s {\n", vt->tag.c_str());
2481 if (vt->global != NULL)
2483 fprintf(f, " global :\n");
2484 this->print_expression_list(f, vt->global);
2487 if (vt->local != NULL)
2489 fprintf(f, " local :\n");
2490 this->print_expression_list(f, vt->local);
2494 if (vt->dependencies != NULL)
2496 const Version_dependency_list* deps = vt->dependencies;
2497 for (size_t j = 0; j < deps->dependencies.size(); ++j)
2499 if (j < deps->dependencies.size() - 1)
2501 fprintf(f, " %s", deps->dependencies[j].c_str());
2511 Version_script_info::print_expression_list(
2513 const Version_expression_list* vel) const
2515 Version_script_info::Language current_language = LANGUAGE_C;
2516 for (size_t i = 0; i < vel->expressions.size(); ++i)
2518 const Version_expression& ve(vel->expressions[i]);
2520 if (ve.language != current_language)
2522 if (current_language != LANGUAGE_C)
2524 switch (ve.language)
2529 fprintf(f, " extern \"C++\" {\n");
2532 fprintf(f, " extern \"Java\" {\n");
2537 current_language = ve.language;
2541 if (current_language != LANGUAGE_C)
2546 fprintf(f, "%s", ve.pattern.c_str());
2553 if (current_language != LANGUAGE_C)
2557 } // End namespace gold.
2559 // The remaining functions are extern "C", so it's clearer to not put
2560 // them in namespace gold.
2562 using namespace gold;
2564 // This function is called by the bison parser to return the next
2568 yylex(YYSTYPE* lvalp, void* closurev)
2570 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2571 const Token* token = closure->next_token();
2572 switch (token->classification())
2577 case Token::TOKEN_INVALID:
2578 yyerror(closurev, "invalid character");
2581 case Token::TOKEN_EOF:
2584 case Token::TOKEN_STRING:
2586 // This is either a keyword or a STRING.
2588 const char* str = token->string_value(&len);
2590 switch (closure->lex_mode())
2592 case Lex::LINKER_SCRIPT:
2593 parsecode = script_keywords.keyword_to_parsecode(str, len);
2595 case Lex::VERSION_SCRIPT:
2596 parsecode = version_script_keywords.keyword_to_parsecode(str, len);
2598 case Lex::DYNAMIC_LIST:
2599 parsecode = dynamic_list_keywords.keyword_to_parsecode(str, len);
2606 lvalp->string.value = str;
2607 lvalp->string.length = len;
2611 case Token::TOKEN_QUOTED_STRING:
2612 lvalp->string.value = token->string_value(&lvalp->string.length);
2613 return QUOTED_STRING;
2615 case Token::TOKEN_OPERATOR:
2616 return token->operator_value();
2618 case Token::TOKEN_INTEGER:
2619 lvalp->integer = token->integer_value();
2624 // This function is called by the bison parser to report an error.
2627 yyerror(void* closurev, const char* message)
2629 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2630 gold_error(_("%s:%d:%d: %s"), closure->filename(), closure->lineno(),
2631 closure->charpos(), message);
2634 // Called by the bison parser to add an external symbol to the link.
2637 script_add_extern(void* closurev, const char* name, size_t length)
2639 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2640 closure->script_options()->add_symbol_reference(name, length);
2643 // Called by the bison parser to add a file to the link.
2646 script_add_file(void* closurev, const char* name, size_t length)
2648 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2650 // If this is an absolute path, and we found the script in the
2651 // sysroot, then we want to prepend the sysroot to the file name.
2652 // For example, this is how we handle a cross link to the x86_64
2653 // libc.so, which refers to /lib/libc.so.6.
2654 std::string name_string(name, length);
2655 const char* extra_search_path = ".";
2656 std::string script_directory;
2657 if (IS_ABSOLUTE_PATH(name_string.c_str()))
2659 if (closure->is_in_sysroot())
2661 const std::string& sysroot(parameters->options().sysroot());
2662 gold_assert(!sysroot.empty());
2663 name_string = sysroot + name_string;
2668 // In addition to checking the normal library search path, we
2669 // also want to check in the script-directory.
2670 const char* slash = strrchr(closure->filename(), '/');
2673 script_directory.assign(closure->filename(),
2674 slash - closure->filename() + 1);
2675 extra_search_path = script_directory.c_str();
2679 Input_file_argument file(name_string.c_str(),
2680 Input_file_argument::INPUT_FILE_TYPE_FILE,
2681 extra_search_path, false,
2682 closure->position_dependent_options());
2683 Input_argument& arg = closure->inputs()->add_file(file);
2684 arg.set_script_info(closure->script_info());
2687 // Called by the bison parser to add a library to the link.
2690 script_add_library(void* closurev, const char* name, size_t length)
2692 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2693 std::string name_string(name, length);
2695 if (name_string[0] != 'l')
2696 gold_error(_("library name must be prefixed with -l"));
2698 Input_file_argument file(name_string.c_str() + 1,
2699 Input_file_argument::INPUT_FILE_TYPE_LIBRARY,
2701 closure->position_dependent_options());
2702 Input_argument& arg = closure->inputs()->add_file(file);
2703 arg.set_script_info(closure->script_info());
2706 // Called by the bison parser to start a group. If we are already in
2707 // a group, that means that this script was invoked within a
2708 // --start-group --end-group sequence on the command line, or that
2709 // this script was found in a GROUP of another script. In that case,
2710 // we simply continue the existing group, rather than starting a new
2711 // one. It is possible to construct a case in which this will do
2712 // something other than what would happen if we did a recursive group,
2713 // but it's hard to imagine why the different behaviour would be
2714 // useful for a real program. Avoiding recursive groups is simpler
2715 // and more efficient.
2718 script_start_group(void* closurev)
2720 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2721 if (!closure->in_group())
2722 closure->inputs()->start_group();
2725 // Called by the bison parser at the end of a group.
2728 script_end_group(void* closurev)
2730 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2731 if (!closure->in_group())
2732 closure->inputs()->end_group();
2735 // Called by the bison parser to start an AS_NEEDED list.
2738 script_start_as_needed(void* closurev)
2740 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2741 closure->position_dependent_options().set_as_needed(true);
2744 // Called by the bison parser at the end of an AS_NEEDED list.
2747 script_end_as_needed(void* closurev)
2749 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2750 closure->position_dependent_options().set_as_needed(false);
2753 // Called by the bison parser to set the entry symbol.
2756 script_set_entry(void* closurev, const char* entry, size_t length)
2758 // We'll parse this exactly the same as --entry=ENTRY on the commandline
2759 // TODO(csilvers): FIXME -- call set_entry directly.
2760 std::string arg("--entry=");
2761 arg.append(entry, length);
2762 script_parse_option(closurev, arg.c_str(), arg.size());
2765 // Called by the bison parser to set whether to define common symbols.
2768 script_set_common_allocation(void* closurev, int set)
2770 const char* arg = set != 0 ? "--define-common" : "--no-define-common";
2771 script_parse_option(closurev, arg, strlen(arg));
2774 // Called by the bison parser to refer to a symbol.
2776 extern "C" Expression*
2777 script_symbol(void* closurev, const char* name, size_t length)
2779 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2780 if (length != 1 || name[0] != '.')
2781 closure->script_options()->add_symbol_reference(name, length);
2782 return script_exp_string(name, length);
2785 // Called by the bison parser to define a symbol.
2788 script_set_symbol(void* closurev, const char* name, size_t length,
2789 Expression* value, int providei, int hiddeni)
2791 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2792 const bool provide = providei != 0;
2793 const bool hidden = hiddeni != 0;
2794 closure->script_options()->add_symbol_assignment(name, length,
2795 closure->parsing_defsym(),
2796 value, provide, hidden);
2797 closure->clear_skip_on_incompatible_target();
2800 // Called by the bison parser to add an assertion.
2803 script_add_assertion(void* closurev, Expression* check, const char* message,
2806 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2807 closure->script_options()->add_assertion(check, message, messagelen);
2808 closure->clear_skip_on_incompatible_target();
2811 // Called by the bison parser to parse an OPTION.
2814 script_parse_option(void* closurev, const char* option, size_t length)
2816 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2817 // We treat the option as a single command-line option, even if
2818 // it has internal whitespace.
2819 if (closure->command_line() == NULL)
2821 // There are some options that we could handle here--e.g.,
2822 // -lLIBRARY. Should we bother?
2823 gold_warning(_("%s:%d:%d: ignoring command OPTION; OPTION is only valid"
2824 " for scripts specified via -T/--script"),
2825 closure->filename(), closure->lineno(), closure->charpos());
2829 bool past_a_double_dash_option = false;
2830 const char* mutable_option = strndup(option, length);
2831 gold_assert(mutable_option != NULL);
2832 closure->command_line()->process_one_option(1, &mutable_option, 0,
2833 &past_a_double_dash_option);
2834 // The General_options class will quite possibly store a pointer
2835 // into mutable_option, so we can't free it. In cases the class
2836 // does not store such a pointer, this is a memory leak. Alas. :(
2838 closure->clear_skip_on_incompatible_target();
2841 // Called by the bison parser to handle OUTPUT_FORMAT. OUTPUT_FORMAT
2842 // takes either one or three arguments. In the three argument case,
2843 // the format depends on the endianness option, which we don't
2844 // currently support (FIXME). If we see an OUTPUT_FORMAT for the
2845 // wrong format, then we want to search for a new file. Returning 0
2846 // here will cause the parser to immediately abort.
2849 script_check_output_format(void* closurev,
2850 const char* default_name, size_t default_length,
2851 const char*, size_t, const char*, size_t)
2853 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2854 std::string name(default_name, default_length);
2855 Target* target = select_target_by_bfd_name(name.c_str());
2856 if (target == NULL || !parameters->is_compatible_target(target))
2858 if (closure->skip_on_incompatible_target())
2860 closure->set_found_incompatible_target();
2863 // FIXME: Should we warn about the unknown target?
2868 // Called by the bison parser to handle TARGET.
2871 script_set_target(void* closurev, const char* target, size_t len)
2873 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2874 std::string s(target, len);
2875 General_options::Object_format format_enum;
2876 format_enum = General_options::string_to_object_format(s.c_str());
2877 closure->position_dependent_options().set_format_enum(format_enum);
2880 // Called by the bison parser to handle SEARCH_DIR. This is handled
2881 // exactly like a -L option.
2884 script_add_search_dir(void* closurev, const char* option, size_t length)
2886 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2887 if (closure->command_line() == NULL)
2888 gold_warning(_("%s:%d:%d: ignoring SEARCH_DIR; SEARCH_DIR is only valid"
2889 " for scripts specified via -T/--script"),
2890 closure->filename(), closure->lineno(), closure->charpos());
2891 else if (!closure->command_line()->options().nostdlib())
2893 std::string s = "-L" + std::string(option, length);
2894 script_parse_option(closurev, s.c_str(), s.size());
2898 /* Called by the bison parser to push the lexer into expression
2902 script_push_lex_into_expression_mode(void* closurev)
2904 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2905 closure->push_lex_mode(Lex::EXPRESSION);
2908 /* Called by the bison parser to push the lexer into version
2912 script_push_lex_into_version_mode(void* closurev)
2914 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2915 if (closure->version_script()->is_finalized())
2916 gold_error(_("%s:%d:%d: invalid use of VERSION in input file"),
2917 closure->filename(), closure->lineno(), closure->charpos());
2918 closure->push_lex_mode(Lex::VERSION_SCRIPT);
2921 /* Called by the bison parser to pop the lexer mode. */
2924 script_pop_lex_mode(void* closurev)
2926 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2927 closure->pop_lex_mode();
2930 // Register an entire version node. For example:
2936 // - tag is "GLIBC_2.1"
2937 // - tree contains the information "global: foo"
2938 // - deps contains "GLIBC_2.0"
2941 script_register_vers_node(void*,
2944 struct Version_tree* tree,
2945 struct Version_dependency_list* deps)
2947 gold_assert(tree != NULL);
2948 tree->dependencies = deps;
2950 tree->tag = std::string(tag, taglen);
2953 // Add a dependencies to the list of existing dependencies, if any,
2954 // and return the expanded list.
2956 extern "C" struct Version_dependency_list*
2957 script_add_vers_depend(void* closurev,
2958 struct Version_dependency_list* all_deps,
2959 const char* depend_to_add, int deplen)
2961 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2962 if (all_deps == NULL)
2963 all_deps = closure->version_script()->allocate_dependency_list();
2964 all_deps->dependencies.push_back(std::string(depend_to_add, deplen));
2968 // Add a pattern expression to an existing list of expressions, if any.
2970 extern "C" struct Version_expression_list*
2971 script_new_vers_pattern(void* closurev,
2972 struct Version_expression_list* expressions,
2973 const char* pattern, int patlen, int exact_match)
2975 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2976 if (expressions == NULL)
2977 expressions = closure->version_script()->allocate_expression_list();
2978 expressions->expressions.push_back(
2979 Version_expression(std::string(pattern, patlen),
2980 closure->get_current_language(),
2981 static_cast<bool>(exact_match)));
2985 // Attaches b to the end of a, and clears b. So a = a + b and b = {}.
2987 extern "C" struct Version_expression_list*
2988 script_merge_expressions(struct Version_expression_list* a,
2989 struct Version_expression_list* b)
2991 a->expressions.insert(a->expressions.end(),
2992 b->expressions.begin(), b->expressions.end());
2993 // We could delete b and remove it from expressions_lists_, but
2994 // that's a lot of work. This works just as well.
2995 b->expressions.clear();
2999 // Combine the global and local expressions into a a Version_tree.
3001 extern "C" struct Version_tree*
3002 script_new_vers_node(void* closurev,
3003 struct Version_expression_list* global,
3004 struct Version_expression_list* local)
3006 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3007 Version_tree* tree = closure->version_script()->allocate_version_tree();
3008 tree->global = global;
3009 tree->local = local;
3013 // Handle a transition in language, such as at the
3014 // start or end of 'extern "C++"'
3017 version_script_push_lang(void* closurev, const char* lang, int langlen)
3019 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3020 std::string language(lang, langlen);
3021 Version_script_info::Language code;
3022 if (language.empty() || language == "C")
3023 code = Version_script_info::LANGUAGE_C;
3024 else if (language == "C++")
3025 code = Version_script_info::LANGUAGE_CXX;
3026 else if (language == "Java")
3027 code = Version_script_info::LANGUAGE_JAVA;
3030 char* buf = new char[langlen + 100];
3031 snprintf(buf, langlen + 100,
3032 _("unrecognized version script language '%s'"),
3034 yyerror(closurev, buf);
3036 code = Version_script_info::LANGUAGE_C;
3038 closure->push_language(code);
3042 version_script_pop_lang(void* closurev)
3044 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3045 closure->pop_language();
3048 // Called by the bison parser to start a SECTIONS clause.
3051 script_start_sections(void* closurev)
3053 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3054 closure->script_options()->script_sections()->start_sections();
3055 closure->clear_skip_on_incompatible_target();
3058 // Called by the bison parser to finish a SECTIONS clause.
3061 script_finish_sections(void* closurev)
3063 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3064 closure->script_options()->script_sections()->finish_sections();
3067 // Start processing entries for an output section.
3070 script_start_output_section(void* closurev, const char* name, size_t namelen,
3071 const struct Parser_output_section_header* header)
3073 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3074 closure->script_options()->script_sections()->start_output_section(name,
3079 // Finish processing entries for an output section.
3082 script_finish_output_section(void* closurev,
3083 const struct Parser_output_section_trailer* trail)
3085 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3086 closure->script_options()->script_sections()->finish_output_section(trail);
3089 // Add a data item (e.g., "WORD (0)") to the current output section.
3092 script_add_data(void* closurev, int data_token, Expression* val)
3094 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3096 bool is_signed = true;
3118 closure->script_options()->script_sections()->add_data(size, is_signed, val);
3121 // Add a clause setting the fill value to the current output section.
3124 script_add_fill(void* closurev, Expression* val)
3126 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3127 closure->script_options()->script_sections()->add_fill(val);
3130 // Add a new input section specification to the current output
3134 script_add_input_section(void* closurev,
3135 const struct Input_section_spec* spec,
3138 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3139 bool keep = keepi != 0;
3140 closure->script_options()->script_sections()->add_input_section(spec, keep);
3143 // When we see DATA_SEGMENT_ALIGN we record that following output
3144 // sections may be relro.
3147 script_data_segment_align(void* closurev)
3149 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3150 if (!closure->script_options()->saw_sections_clause())
3151 gold_error(_("%s:%d:%d: DATA_SEGMENT_ALIGN not in SECTIONS clause"),
3152 closure->filename(), closure->lineno(), closure->charpos());
3154 closure->script_options()->script_sections()->data_segment_align();
3157 // When we see DATA_SEGMENT_RELRO_END we know that all output sections
3158 // since DATA_SEGMENT_ALIGN should be relro.
3161 script_data_segment_relro_end(void* closurev)
3163 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3164 if (!closure->script_options()->saw_sections_clause())
3165 gold_error(_("%s:%d:%d: DATA_SEGMENT_ALIGN not in SECTIONS clause"),
3166 closure->filename(), closure->lineno(), closure->charpos());
3168 closure->script_options()->script_sections()->data_segment_relro_end();
3171 // Create a new list of string/sort pairs.
3173 extern "C" String_sort_list_ptr
3174 script_new_string_sort_list(const struct Wildcard_section* string_sort)
3176 return new String_sort_list(1, *string_sort);
3179 // Add an entry to a list of string/sort pairs. The way the parser
3180 // works permits us to simply modify the first parameter, rather than
3183 extern "C" String_sort_list_ptr
3184 script_string_sort_list_add(String_sort_list_ptr pv,
3185 const struct Wildcard_section* string_sort)
3188 return script_new_string_sort_list(string_sort);
3191 pv->push_back(*string_sort);
3196 // Create a new list of strings.
3198 extern "C" String_list_ptr
3199 script_new_string_list(const char* str, size_t len)
3201 return new String_list(1, std::string(str, len));
3204 // Add an element to a list of strings. The way the parser works
3205 // permits us to simply modify the first parameter, rather than copy
3208 extern "C" String_list_ptr
3209 script_string_list_push_back(String_list_ptr pv, const char* str, size_t len)
3212 return script_new_string_list(str, len);
3215 pv->push_back(std::string(str, len));
3220 // Concatenate two string lists. Either or both may be NULL. The way
3221 // the parser works permits us to modify the parameters, rather than
3224 extern "C" String_list_ptr
3225 script_string_list_append(String_list_ptr pv1, String_list_ptr pv2)
3231 pv1->insert(pv1->end(), pv2->begin(), pv2->end());
3235 // Add a new program header.
3238 script_add_phdr(void* closurev, const char* name, size_t namelen,
3239 unsigned int type, const Phdr_info* info)
3241 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3242 bool includes_filehdr = info->includes_filehdr != 0;
3243 bool includes_phdrs = info->includes_phdrs != 0;
3244 bool is_flags_valid = info->is_flags_valid != 0;
3245 Script_sections* ss = closure->script_options()->script_sections();
3246 ss->add_phdr(name, namelen, type, includes_filehdr, includes_phdrs,
3247 is_flags_valid, info->flags, info->load_address);
3248 closure->clear_skip_on_incompatible_target();
3251 // Convert a program header string to a type.
3253 #define PHDR_TYPE(NAME) { #NAME, sizeof(#NAME) - 1, elfcpp::NAME }
3260 } phdr_type_names[] =
3264 PHDR_TYPE(PT_DYNAMIC),
3265 PHDR_TYPE(PT_INTERP),
3267 PHDR_TYPE(PT_SHLIB),
3270 PHDR_TYPE(PT_GNU_EH_FRAME),
3271 PHDR_TYPE(PT_GNU_STACK),
3272 PHDR_TYPE(PT_GNU_RELRO)
3275 extern "C" unsigned int
3276 script_phdr_string_to_type(void* closurev, const char* name, size_t namelen)
3278 for (unsigned int i = 0;
3279 i < sizeof(phdr_type_names) / sizeof(phdr_type_names[0]);
3281 if (namelen == phdr_type_names[i].namelen
3282 && strncmp(name, phdr_type_names[i].name, namelen) == 0)
3283 return phdr_type_names[i].val;
3284 yyerror(closurev, _("unknown PHDR type (try integer)"));
3285 return elfcpp::PT_NULL;
3289 script_saw_segment_start_expression(void* closurev)
3291 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3292 Script_sections* ss = closure->script_options()->script_sections();
3293 ss->set_saw_segment_start_expression(true);
3297 script_set_section_region(void* closurev, const char* name, size_t namelen,
3300 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3301 if (!closure->script_options()->saw_sections_clause())
3303 gold_error(_("%s:%d:%d: MEMORY region '%.*s' referred to outside of "
3305 closure->filename(), closure->lineno(), closure->charpos(),
3306 static_cast<int>(namelen), name);
3310 Script_sections* ss = closure->script_options()->script_sections();
3311 Memory_region* mr = ss->find_memory_region(name, namelen);
3314 gold_error(_("%s:%d:%d: MEMORY region '%.*s' not declared"),
3315 closure->filename(), closure->lineno(), closure->charpos(),
3316 static_cast<int>(namelen), name);
3320 ss->set_memory_region(mr, set_vma);
3324 script_add_memory(void* closurev, const char* name, size_t namelen,
3325 unsigned int attrs, Expression* origin, Expression* length)
3327 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3328 Script_sections* ss = closure->script_options()->script_sections();
3329 ss->add_memory_region(name, namelen, attrs, origin, length);
3332 extern "C" unsigned int
3333 script_parse_memory_attr(void* closurev, const char* attrs, size_t attrlen,
3343 attributes |= MEM_READABLE; break;
3346 attributes |= MEM_READABLE | MEM_WRITEABLE; break;
3349 attributes |= MEM_EXECUTABLE; break;
3352 attributes |= MEM_ALLOCATABLE; break;
3357 attributes |= MEM_INITIALIZED; break;
3359 yyerror(closurev, _("unknown MEMORY attribute"));
3363 attributes = (~ attributes) & MEM_ATTR_MASK;
3369 script_include_directive(int first_token, void* closurev,
3370 const char* filename, size_t length)
3372 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3373 std::string name(filename, length);
3374 Command_line* cmdline = closure->command_line();
3375 read_script_file(name.c_str(), cmdline, &cmdline->script_options(),
3376 first_token, Lex::LINKER_SCRIPT);
3379 // Functions for memory regions.
3381 extern "C" Expression*
3382 script_exp_function_origin(void* closurev, const char* name, size_t namelen)
3384 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3385 Script_sections* ss = closure->script_options()->script_sections();
3386 Expression* origin = ss->find_memory_region_origin(name, namelen);
3390 gold_error(_("undefined memory region '%s' referenced "
3391 "in ORIGIN expression"),
3393 // Create a dummy expression to prevent crashes later on.
3394 origin = script_exp_integer(0);
3400 extern "C" Expression*
3401 script_exp_function_length(void* closurev, const char* name, size_t namelen)
3403 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3404 Script_sections* ss = closure->script_options()->script_sections();
3405 Expression* length = ss->find_memory_region_length(name, namelen);
3409 gold_error(_("undefined memory region '%s' referenced "
3410 "in LENGTH expression"),
3412 // Create a dummy expression to prevent crashes later on.
3413 length = script_exp_integer(0);