1 // script.cc -- handle linker scripts for gold.
3 // Copyright 2006, 2007, 2008, 2009 Free Software Foundation, Inc.
4 // Written by Ian Lance Taylor <iant@google.com>.
6 // This file is part of gold.
8 // This program is free software; you can redistribute it and/or modify
9 // it under the terms of the GNU General Public License as published by
10 // the Free Software Foundation; either version 3 of the License, or
11 // (at your option) any later version.
13 // This program is distributed in the hope that it will be useful,
14 // but WITHOUT ANY WARRANTY; without even the implied warranty of
15 // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 // GNU General Public License for more details.
18 // You should have received a copy of the GNU General Public License
19 // along with this program; if not, write to the Free Software
20 // Foundation, Inc., 51 Franklin Street - Fifth Floor, Boston,
21 // MA 02110-1301, USA.
31 #include "filenames.h"
35 #include "dirsearch.h"
38 #include "workqueue.h"
40 #include "parameters.h"
43 #include "target-select.h"
46 #include "incremental.h"
51 // A token read from a script file. We don't implement keywords here;
52 // all keywords are simply represented as a string.
57 // Token classification.
62 // Token indicates end of input.
64 // Token is a string of characters.
66 // Token is a quoted string of characters.
68 // Token is an operator.
70 // Token is a number (an integer).
74 // We need an empty constructor so that we can put this STL objects.
76 : classification_(TOKEN_INVALID), value_(NULL), value_length_(0),
77 opcode_(0), lineno_(0), charpos_(0)
80 // A general token with no value.
81 Token(Classification classification, int lineno, int charpos)
82 : classification_(classification), value_(NULL), value_length_(0),
83 opcode_(0), lineno_(lineno), charpos_(charpos)
85 gold_assert(classification == TOKEN_INVALID
86 || classification == TOKEN_EOF);
89 // A general token with a value.
90 Token(Classification classification, const char* value, size_t length,
91 int lineno, int charpos)
92 : classification_(classification), value_(value), value_length_(length),
93 opcode_(0), lineno_(lineno), charpos_(charpos)
95 gold_assert(classification != TOKEN_INVALID
96 && classification != TOKEN_EOF);
99 // A token representing an operator.
100 Token(int opcode, int lineno, int charpos)
101 : classification_(TOKEN_OPERATOR), value_(NULL), value_length_(0),
102 opcode_(opcode), lineno_(lineno), charpos_(charpos)
105 // Return whether the token is invalid.
108 { return this->classification_ == TOKEN_INVALID; }
110 // Return whether this is an EOF token.
113 { return this->classification_ == TOKEN_EOF; }
115 // Return the token classification.
117 classification() const
118 { return this->classification_; }
120 // Return the line number at which the token starts.
123 { return this->lineno_; }
125 // Return the character position at this the token starts.
128 { return this->charpos_; }
130 // Get the value of a token.
133 string_value(size_t* length) const
135 gold_assert(this->classification_ == TOKEN_STRING
136 || this->classification_ == TOKEN_QUOTED_STRING);
137 *length = this->value_length_;
142 operator_value() const
144 gold_assert(this->classification_ == TOKEN_OPERATOR);
145 return this->opcode_;
149 integer_value() const
151 gold_assert(this->classification_ == TOKEN_INTEGER);
153 std::string s(this->value_, this->value_length_);
154 return strtoull(s.c_str(), NULL, 0);
158 // The token classification.
159 Classification classification_;
160 // The token value, for TOKEN_STRING or TOKEN_QUOTED_STRING or
163 // The length of the token value.
164 size_t value_length_;
165 // The token value, for TOKEN_OPERATOR.
167 // The line number where this token started (one based).
169 // The character position within the line where this token started
174 // This class handles lexing a file into a sequence of tokens.
179 // We unfortunately have to support different lexing modes, because
180 // when reading different parts of a linker script we need to parse
181 // things differently.
184 // Reading an ordinary linker script.
186 // Reading an expression in a linker script.
188 // Reading a version script.
190 // Reading a --dynamic-list file.
194 Lex(const char* input_string, size_t input_length, int parsing_token)
195 : input_string_(input_string), input_length_(input_length),
196 current_(input_string), mode_(LINKER_SCRIPT),
197 first_token_(parsing_token), token_(),
198 lineno_(1), linestart_(input_string)
201 // Read a file into a string.
203 read_file(Input_file*, std::string*);
205 // Return the next token.
209 // Return the current lexing mode.
212 { return this->mode_; }
214 // Set the lexing mode.
217 { this->mode_ = mode; }
221 Lex& operator=(const Lex&);
223 // Make a general token with no value at the current location.
225 make_token(Token::Classification c, const char* start) const
226 { return Token(c, this->lineno_, start - this->linestart_ + 1); }
228 // Make a general token with a value at the current location.
230 make_token(Token::Classification c, const char* v, size_t len,
233 { return Token(c, v, len, this->lineno_, start - this->linestart_ + 1); }
235 // Make an operator token at the current location.
237 make_token(int opcode, const char* start) const
238 { return Token(opcode, this->lineno_, start - this->linestart_ + 1); }
240 // Make an invalid token at the current location.
242 make_invalid_token(const char* start)
243 { return this->make_token(Token::TOKEN_INVALID, start); }
245 // Make an EOF token at the current location.
247 make_eof_token(const char* start)
248 { return this->make_token(Token::TOKEN_EOF, start); }
250 // Return whether C can be the first character in a name. C2 is the
251 // next character, since we sometimes need that.
253 can_start_name(char c, char c2);
255 // If C can appear in a name which has already started, return a
256 // pointer to a character later in the token or just past
257 // it. Otherwise, return NULL.
259 can_continue_name(const char* c);
261 // Return whether C, C2, C3 can start a hex number.
263 can_start_hex(char c, char c2, char c3);
265 // If C can appear in a hex number which has already started, return
266 // a pointer to a character later in the token or just past
267 // it. Otherwise, return NULL.
269 can_continue_hex(const char* c);
271 // Return whether C can start a non-hex number.
273 can_start_number(char c);
275 // If C can appear in a decimal number which has already started,
276 // return a pointer to a character later in the token or just past
277 // it. Otherwise, return NULL.
279 can_continue_number(const char* c)
280 { return Lex::can_start_number(*c) ? c + 1 : NULL; }
282 // If C1 C2 C3 form a valid three character operator, return the
283 // opcode. Otherwise return 0.
285 three_char_operator(char c1, char c2, char c3);
287 // If C1 C2 form a valid two character operator, return the opcode.
288 // Otherwise return 0.
290 two_char_operator(char c1, char c2);
292 // If C1 is a valid one character operator, return the opcode.
293 // Otherwise return 0.
295 one_char_operator(char c1);
297 // Read the next token.
299 get_token(const char**);
301 // Skip a C style /* */ comment. Return false if the comment did
304 skip_c_comment(const char**);
306 // Skip a line # comment. Return false if there was no newline.
308 skip_line_comment(const char**);
310 // Build a token CLASSIFICATION from all characters that match
311 // CAN_CONTINUE_FN. The token starts at START. Start matching from
312 // MATCH. Set *PP to the character following the token.
314 gather_token(Token::Classification,
315 const char* (Lex::*can_continue_fn)(const char*),
316 const char* start, const char* match, const char** pp);
318 // Build a token from a quoted string.
320 gather_quoted_string(const char** pp);
322 // The string we are tokenizing.
323 const char* input_string_;
324 // The length of the string.
325 size_t input_length_;
326 // The current offset into the string.
327 const char* current_;
328 // The current lexing mode.
330 // The code to use for the first token. This is set to 0 after it
333 // The current token.
335 // The current line number.
337 // The start of the current line in the string.
338 const char* linestart_;
341 // Read the whole file into memory. We don't expect linker scripts to
342 // be large, so we just use a std::string as a buffer. We ignore the
343 // data we've already read, so that we read aligned buffers.
346 Lex::read_file(Input_file* input_file, std::string* contents)
348 off_t filesize = input_file->file().filesize();
350 contents->reserve(filesize);
353 unsigned char buf[BUFSIZ];
354 while (off < filesize)
357 if (get > filesize - off)
358 get = filesize - off;
359 input_file->file().read(off, get, buf);
360 contents->append(reinterpret_cast<char*>(&buf[0]), get);
365 // Return whether C can be the start of a name, if the next character
366 // is C2. A name can being with a letter, underscore, period, or
367 // dollar sign. Because a name can be a file name, we also permit
368 // forward slash, backslash, and tilde. Tilde is the tricky case
369 // here; GNU ld also uses it as a bitwise not operator. It is only
370 // recognized as the operator if it is not immediately followed by
371 // some character which can appear in a symbol. That is, when we
372 // don't know that we are looking at an expression, "~0" is a file
373 // name, and "~ 0" is an expression using bitwise not. We are
377 Lex::can_start_name(char c, char c2)
381 case 'A': case 'B': case 'C': case 'D': case 'E': case 'F':
382 case 'G': case 'H': case 'I': case 'J': case 'K': case 'L':
383 case 'M': case 'N': case 'O': case 'Q': case 'P': case 'R':
384 case 'S': case 'T': case 'U': case 'V': case 'W': case 'X':
386 case 'a': case 'b': case 'c': case 'd': case 'e': case 'f':
387 case 'g': case 'h': case 'i': case 'j': case 'k': case 'l':
388 case 'm': case 'n': case 'o': case 'q': case 'p': case 'r':
389 case 's': case 't': case 'u': case 'v': case 'w': case 'x':
391 case '_': case '.': case '$':
395 return this->mode_ == LINKER_SCRIPT;
398 return this->mode_ == LINKER_SCRIPT && can_continue_name(&c2);
401 return (this->mode_ == VERSION_SCRIPT
402 || this->mode_ == DYNAMIC_LIST
403 || (this->mode_ == LINKER_SCRIPT
404 && can_continue_name(&c2)));
411 // Return whether C can continue a name which has already started.
412 // Subsequent characters in a name are the same as the leading
413 // characters, plus digits and "=+-:[],?*". So in general the linker
414 // script language requires spaces around operators, unless we know
415 // that we are parsing an expression.
418 Lex::can_continue_name(const char* c)
422 case 'A': case 'B': case 'C': case 'D': case 'E': case 'F':
423 case 'G': case 'H': case 'I': case 'J': case 'K': case 'L':
424 case 'M': case 'N': case 'O': case 'Q': case 'P': case 'R':
425 case 'S': case 'T': case 'U': case 'V': case 'W': case 'X':
427 case 'a': case 'b': case 'c': case 'd': case 'e': case 'f':
428 case 'g': case 'h': case 'i': case 'j': case 'k': case 'l':
429 case 'm': case 'n': case 'o': case 'q': case 'p': case 'r':
430 case 's': case 't': case 'u': case 'v': case 'w': case 'x':
432 case '_': case '.': case '$':
433 case '0': case '1': case '2': case '3': case '4':
434 case '5': case '6': case '7': case '8': case '9':
437 // TODO(csilvers): why not allow ~ in names for version-scripts?
438 case '/': case '\\': case '~':
441 if (this->mode_ == LINKER_SCRIPT)
445 case '[': case ']': case '*': case '?': case '-':
446 if (this->mode_ == LINKER_SCRIPT || this->mode_ == VERSION_SCRIPT
447 || this->mode_ == DYNAMIC_LIST)
451 // TODO(csilvers): why allow this? ^ is meaningless in version scripts.
453 if (this->mode_ == VERSION_SCRIPT || this->mode_ == DYNAMIC_LIST)
458 if (this->mode_ == LINKER_SCRIPT)
460 else if ((this->mode_ == VERSION_SCRIPT || this->mode_ == DYNAMIC_LIST)
463 // A name can have '::' in it, as that's a c++ namespace
464 // separator. But a single colon is not part of a name.
474 // For a number we accept 0x followed by hex digits, or any sequence
475 // of digits. The old linker accepts leading '$' for hex, and
476 // trailing HXBOD. Those are for MRI compatibility and we don't
477 // accept them. The old linker also accepts trailing MK for mega or
478 // kilo. FIXME: Those are mentioned in the documentation, and we
479 // should accept them.
481 // Return whether C1 C2 C3 can start a hex number.
484 Lex::can_start_hex(char c1, char c2, char c3)
486 if (c1 == '0' && (c2 == 'x' || c2 == 'X'))
487 return this->can_continue_hex(&c3);
491 // Return whether C can appear in a hex number.
494 Lex::can_continue_hex(const char* c)
498 case '0': case '1': case '2': case '3': case '4':
499 case '5': case '6': case '7': case '8': case '9':
500 case 'A': case 'B': case 'C': case 'D': case 'E': case 'F':
501 case 'a': case 'b': case 'c': case 'd': case 'e': case 'f':
509 // Return whether C can start a non-hex number.
512 Lex::can_start_number(char c)
516 case '0': case '1': case '2': case '3': case '4':
517 case '5': case '6': case '7': case '8': case '9':
525 // If C1 C2 C3 form a valid three character operator, return the
526 // opcode (defined in the yyscript.h file generated from yyscript.y).
527 // Otherwise return 0.
530 Lex::three_char_operator(char c1, char c2, char c3)
535 if (c2 == '<' && c3 == '=')
539 if (c2 == '>' && c3 == '=')
548 // If C1 C2 form a valid two character operator, return the opcode
549 // (defined in the yyscript.h file generated from yyscript.y).
550 // Otherwise return 0.
553 Lex::two_char_operator(char c1, char c2)
611 // If C1 is a valid operator, return the opcode. Otherwise return 0.
614 Lex::one_char_operator(char c1)
647 // Skip a C style comment. *PP points to just after the "/*". Return
648 // false if the comment did not end.
651 Lex::skip_c_comment(const char** pp)
654 while (p[0] != '*' || p[1] != '/')
665 this->linestart_ = p + 1;
674 // Skip a line # comment. Return false if there was no newline.
677 Lex::skip_line_comment(const char** pp)
680 size_t skip = strcspn(p, "\n");
689 this->linestart_ = p;
695 // Build a token CLASSIFICATION from all characters that match
696 // CAN_CONTINUE_FN. Update *PP.
699 Lex::gather_token(Token::Classification classification,
700 const char* (Lex::*can_continue_fn)(const char*),
705 const char* new_match = NULL;
706 while ((new_match = (this->*can_continue_fn)(match)))
709 return this->make_token(classification, start, match - start, start);
712 // Build a token from a quoted string.
715 Lex::gather_quoted_string(const char** pp)
717 const char* start = *pp;
718 const char* p = start;
720 size_t skip = strcspn(p, "\"\n");
722 return this->make_invalid_token(start);
724 return this->make_token(Token::TOKEN_QUOTED_STRING, p, skip, start);
727 // Return the next token at *PP. Update *PP. General guideline: we
728 // require linker scripts to be simple ASCII. No unicode linker
729 // scripts. In particular we can assume that any '\0' is the end of
733 Lex::get_token(const char** pp)
742 return this->make_eof_token(p);
745 // Skip whitespace quickly.
746 while (*p == ' ' || *p == '\t' || *p == '\r')
753 this->linestart_ = p;
757 // Skip C style comments.
758 if (p[0] == '/' && p[1] == '*')
760 int lineno = this->lineno_;
761 int charpos = p - this->linestart_ + 1;
764 if (!this->skip_c_comment(pp))
765 return Token(Token::TOKEN_INVALID, lineno, charpos);
771 // Skip line comments.
775 if (!this->skip_line_comment(pp))
776 return this->make_eof_token(p);
782 if (this->can_start_name(p[0], p[1]))
783 return this->gather_token(Token::TOKEN_STRING,
784 &Lex::can_continue_name,
787 // We accept any arbitrary name in double quotes, as long as it
788 // does not cross a line boundary.
792 return this->gather_quoted_string(pp);
795 // Check for a number.
797 if (this->can_start_hex(p[0], p[1], p[2]))
798 return this->gather_token(Token::TOKEN_INTEGER,
799 &Lex::can_continue_hex,
802 if (Lex::can_start_number(p[0]))
803 return this->gather_token(Token::TOKEN_INTEGER,
804 &Lex::can_continue_number,
807 // Check for operators.
809 int opcode = Lex::three_char_operator(p[0], p[1], p[2]);
813 return this->make_token(opcode, p);
816 opcode = Lex::two_char_operator(p[0], p[1]);
820 return this->make_token(opcode, p);
823 opcode = Lex::one_char_operator(p[0]);
827 return this->make_token(opcode, p);
830 return this->make_token(Token::TOKEN_INVALID, p);
834 // Return the next token.
839 // The first token is special.
840 if (this->first_token_ != 0)
842 this->token_ = Token(this->first_token_, 0, 0);
843 this->first_token_ = 0;
844 return &this->token_;
847 this->token_ = this->get_token(&this->current_);
849 // Don't let an early null byte fool us into thinking that we've
850 // reached the end of the file.
851 if (this->token_.is_eof()
852 && (static_cast<size_t>(this->current_ - this->input_string_)
853 < this->input_length_))
854 this->token_ = this->make_invalid_token(this->current_);
856 return &this->token_;
859 // class Symbol_assignment.
861 // Add the symbol to the symbol table. This makes sure the symbol is
862 // there and defined. The actual value is stored later. We can't
863 // determine the actual value at this point, because we can't
864 // necessarily evaluate the expression until all ordinary symbols have
867 // The GNU linker lets symbol assignments in the linker script
868 // silently override defined symbols in object files. We are
869 // compatible. FIXME: Should we issue a warning?
872 Symbol_assignment::add_to_table(Symbol_table* symtab)
874 elfcpp::STV vis = this->hidden_ ? elfcpp::STV_HIDDEN : elfcpp::STV_DEFAULT;
875 this->sym_ = symtab->define_as_constant(this->name_.c_str(),
878 ? Symbol_table::DEFSYM
879 : Symbol_table::SCRIPT),
887 true); // force_override
890 // Finalize a symbol value.
893 Symbol_assignment::finalize(Symbol_table* symtab, const Layout* layout)
895 this->finalize_maybe_dot(symtab, layout, false, 0, NULL);
898 // Finalize a symbol value which can refer to the dot symbol.
901 Symbol_assignment::finalize_with_dot(Symbol_table* symtab,
902 const Layout* layout,
904 Output_section* dot_section)
906 this->finalize_maybe_dot(symtab, layout, true, dot_value, dot_section);
909 // Finalize a symbol value, internal version.
912 Symbol_assignment::finalize_maybe_dot(Symbol_table* symtab,
913 const Layout* layout,
914 bool is_dot_available,
916 Output_section* dot_section)
918 // If we were only supposed to provide this symbol, the sym_ field
919 // will be NULL if the symbol was not referenced.
920 if (this->sym_ == NULL)
922 gold_assert(this->provide_);
926 if (parameters->target().get_size() == 32)
928 #if defined(HAVE_TARGET_32_LITTLE) || defined(HAVE_TARGET_32_BIG)
929 this->sized_finalize<32>(symtab, layout, is_dot_available, dot_value,
935 else if (parameters->target().get_size() == 64)
937 #if defined(HAVE_TARGET_64_LITTLE) || defined(HAVE_TARGET_64_BIG)
938 this->sized_finalize<64>(symtab, layout, is_dot_available, dot_value,
950 Symbol_assignment::sized_finalize(Symbol_table* symtab, const Layout* layout,
951 bool is_dot_available, uint64_t dot_value,
952 Output_section* dot_section)
954 Output_section* section;
955 uint64_t final_val = this->val_->eval_maybe_dot(symtab, layout, true,
957 dot_value, dot_section,
959 Sized_symbol<size>* ssym = symtab->get_sized_symbol<size>(this->sym_);
960 ssym->set_value(final_val);
962 ssym->set_output_section(section);
965 // Set the symbol value if the expression yields an absolute value.
968 Symbol_assignment::set_if_absolute(Symbol_table* symtab, const Layout* layout,
969 bool is_dot_available, uint64_t dot_value)
971 if (this->sym_ == NULL)
974 Output_section* val_section;
975 uint64_t val = this->val_->eval_maybe_dot(symtab, layout, false,
976 is_dot_available, dot_value,
978 if (val_section != NULL)
981 if (parameters->target().get_size() == 32)
983 #if defined(HAVE_TARGET_32_LITTLE) || defined(HAVE_TARGET_32_BIG)
984 Sized_symbol<32>* ssym = symtab->get_sized_symbol<32>(this->sym_);
985 ssym->set_value(val);
990 else if (parameters->target().get_size() == 64)
992 #if defined(HAVE_TARGET_64_LITTLE) || defined(HAVE_TARGET_64_BIG)
993 Sized_symbol<64>* ssym = symtab->get_sized_symbol<64>(this->sym_);
994 ssym->set_value(val);
1003 // Print for debugging.
1006 Symbol_assignment::print(FILE* f) const
1008 if (this->provide_ && this->hidden_)
1009 fprintf(f, "PROVIDE_HIDDEN(");
1010 else if (this->provide_)
1011 fprintf(f, "PROVIDE(");
1012 else if (this->hidden_)
1015 fprintf(f, "%s = ", this->name_.c_str());
1016 this->val_->print(f);
1018 if (this->provide_ || this->hidden_)
1024 // Class Script_assertion.
1026 // Check the assertion.
1029 Script_assertion::check(const Symbol_table* symtab, const Layout* layout)
1031 if (!this->check_->eval(symtab, layout, true))
1032 gold_error("%s", this->message_.c_str());
1035 // Print for debugging.
1038 Script_assertion::print(FILE* f) const
1040 fprintf(f, "ASSERT(");
1041 this->check_->print(f);
1042 fprintf(f, ", \"%s\")\n", this->message_.c_str());
1045 // Class Script_options.
1047 Script_options::Script_options()
1048 : entry_(), symbol_assignments_(), version_script_info_(),
1053 // Add a symbol to be defined.
1056 Script_options::add_symbol_assignment(const char* name, size_t length,
1057 bool is_defsym, Expression* value,
1058 bool provide, bool hidden)
1060 if (length != 1 || name[0] != '.')
1062 if (this->script_sections_.in_sections_clause())
1064 gold_assert(!is_defsym);
1065 this->script_sections_.add_symbol_assignment(name, length, value,
1070 Symbol_assignment* p = new Symbol_assignment(name, length, is_defsym,
1071 value, provide, hidden);
1072 this->symbol_assignments_.push_back(p);
1077 if (provide || hidden)
1078 gold_error(_("invalid use of PROVIDE for dot symbol"));
1080 // The GNU linker permits assignments to dot outside of SECTIONS
1081 // clauses and treats them as occurring inside, so we don't
1082 // check in_sections_clause here.
1083 this->script_sections_.add_dot_assignment(value);
1087 // Add an assertion.
1090 Script_options::add_assertion(Expression* check, const char* message,
1093 if (this->script_sections_.in_sections_clause())
1094 this->script_sections_.add_assertion(check, message, messagelen);
1097 Script_assertion* p = new Script_assertion(check, message, messagelen);
1098 this->assertions_.push_back(p);
1102 // Create sections required by any linker scripts.
1105 Script_options::create_script_sections(Layout* layout)
1107 if (this->saw_sections_clause())
1108 this->script_sections_.create_sections(layout);
1111 // Add any symbols we are defining to the symbol table.
1114 Script_options::add_symbols_to_table(Symbol_table* symtab)
1116 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin();
1117 p != this->symbol_assignments_.end();
1119 (*p)->add_to_table(symtab);
1120 this->script_sections_.add_symbols_to_table(symtab);
1123 // Finalize symbol values. Also check assertions.
1126 Script_options::finalize_symbols(Symbol_table* symtab, const Layout* layout)
1128 // We finalize the symbols defined in SECTIONS first, because they
1129 // are the ones which may have changed. This way if symbol outside
1130 // SECTIONS are defined in terms of symbols inside SECTIONS, they
1131 // will get the right value.
1132 this->script_sections_.finalize_symbols(symtab, layout);
1134 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin();
1135 p != this->symbol_assignments_.end();
1137 (*p)->finalize(symtab, layout);
1139 for (Assertions::iterator p = this->assertions_.begin();
1140 p != this->assertions_.end();
1142 (*p)->check(symtab, layout);
1145 // Set section addresses. We set all the symbols which have absolute
1146 // values. Then we let the SECTIONS clause do its thing. This
1147 // returns the segment which holds the file header and segment
1151 Script_options::set_section_addresses(Symbol_table* symtab, Layout* layout)
1153 for (Symbol_assignments::iterator p = this->symbol_assignments_.begin();
1154 p != this->symbol_assignments_.end();
1156 (*p)->set_if_absolute(symtab, layout, false, 0);
1158 return this->script_sections_.set_section_addresses(symtab, layout);
1161 // This class holds data passed through the parser to the lexer and to
1162 // the parser support functions. This avoids global variables. We
1163 // can't use global variables because we need not be called by a
1164 // singleton thread.
1166 class Parser_closure
1169 Parser_closure(const char* filename,
1170 const Position_dependent_options& posdep_options,
1171 bool parsing_defsym, bool in_group, bool is_in_sysroot,
1172 Command_line* command_line,
1173 Script_options* script_options,
1175 bool skip_on_incompatible_target)
1176 : filename_(filename), posdep_options_(posdep_options),
1177 parsing_defsym_(parsing_defsym), in_group_(in_group),
1178 is_in_sysroot_(is_in_sysroot),
1179 skip_on_incompatible_target_(skip_on_incompatible_target),
1180 found_incompatible_target_(false),
1181 command_line_(command_line), script_options_(script_options),
1182 version_script_info_(script_options->version_script_info()),
1183 lex_(lex), lineno_(0), charpos_(0), lex_mode_stack_(), inputs_(NULL)
1185 // We start out processing C symbols in the default lex mode.
1186 language_stack_.push_back("");
1187 lex_mode_stack_.push_back(lex->mode());
1190 // Return the file name.
1193 { return this->filename_; }
1195 // Return the position dependent options. The caller may modify
1197 Position_dependent_options&
1198 position_dependent_options()
1199 { return this->posdep_options_; }
1201 // Whether we are parsing a --defsym.
1203 parsing_defsym() const
1204 { return this->parsing_defsym_; }
1206 // Return whether this script is being run in a group.
1209 { return this->in_group_; }
1211 // Return whether this script was found using a directory in the
1214 is_in_sysroot() const
1215 { return this->is_in_sysroot_; }
1217 // Whether to skip to the next file with the same name if we find an
1218 // incompatible target in an OUTPUT_FORMAT statement.
1220 skip_on_incompatible_target() const
1221 { return this->skip_on_incompatible_target_; }
1223 // Stop skipping to the next file on an incompatible target. This
1224 // is called when we make some unrevocable change to the data
1227 clear_skip_on_incompatible_target()
1228 { this->skip_on_incompatible_target_ = false; }
1230 // Whether we found an incompatible target in an OUTPUT_FORMAT
1233 found_incompatible_target() const
1234 { return this->found_incompatible_target_; }
1236 // Note that we found an incompatible target.
1238 set_found_incompatible_target()
1239 { this->found_incompatible_target_ = true; }
1241 // Returns the Command_line structure passed in at constructor time.
1242 // This value may be NULL. The caller may modify this, which modifies
1243 // the passed-in Command_line object (not a copy).
1246 { return this->command_line_; }
1248 // Return the options which may be set by a script.
1251 { return this->script_options_; }
1253 // Return the object in which version script information should be stored.
1254 Version_script_info*
1256 { return this->version_script_info_; }
1258 // Return the next token, and advance.
1262 const Token* token = this->lex_->next_token();
1263 this->lineno_ = token->lineno();
1264 this->charpos_ = token->charpos();
1268 // Set a new lexer mode, pushing the current one.
1270 push_lex_mode(Lex::Mode mode)
1272 this->lex_mode_stack_.push_back(this->lex_->mode());
1273 this->lex_->set_mode(mode);
1276 // Pop the lexer mode.
1280 gold_assert(!this->lex_mode_stack_.empty());
1281 this->lex_->set_mode(this->lex_mode_stack_.back());
1282 this->lex_mode_stack_.pop_back();
1285 // Return the current lexer mode.
1288 { return this->lex_mode_stack_.back(); }
1290 // Return the line number of the last token.
1293 { return this->lineno_; }
1295 // Return the character position in the line of the last token.
1298 { return this->charpos_; }
1300 // Return the list of input files, creating it if necessary. This
1301 // is a space leak--we never free the INPUTS_ pointer.
1305 if (this->inputs_ == NULL)
1306 this->inputs_ = new Input_arguments();
1307 return this->inputs_;
1310 // Return whether we saw any input files.
1313 { return this->inputs_ != NULL && !this->inputs_->empty(); }
1315 // Return the current language being processed in a version script
1316 // (eg, "C++"). The empty string represents unmangled C names.
1318 get_current_language() const
1319 { return this->language_stack_.back(); }
1321 // Push a language onto the stack when entering an extern block.
1322 void push_language(const std::string& lang)
1323 { this->language_stack_.push_back(lang); }
1325 // Pop a language off of the stack when exiting an extern block.
1328 gold_assert(!this->language_stack_.empty());
1329 this->language_stack_.pop_back();
1333 // The name of the file we are reading.
1334 const char* filename_;
1335 // The position dependent options.
1336 Position_dependent_options posdep_options_;
1337 // True if we are parsing a --defsym.
1338 bool parsing_defsym_;
1339 // Whether we are currently in a --start-group/--end-group.
1341 // Whether the script was found in a sysrooted directory.
1342 bool is_in_sysroot_;
1343 // If this is true, then if we find an OUTPUT_FORMAT with an
1344 // incompatible target, then we tell the parser to abort so that we
1345 // can search for the next file with the same name.
1346 bool skip_on_incompatible_target_;
1347 // True if we found an OUTPUT_FORMAT with an incompatible target.
1348 bool found_incompatible_target_;
1349 // May be NULL if the user chooses not to pass one in.
1350 Command_line* command_line_;
1351 // Options which may be set from any linker script.
1352 Script_options* script_options_;
1353 // Information parsed from a version script.
1354 Version_script_info* version_script_info_;
1357 // The line number of the last token returned by next_token.
1359 // The column number of the last token returned by next_token.
1361 // A stack of lexer modes.
1362 std::vector<Lex::Mode> lex_mode_stack_;
1363 // A stack of which extern/language block we're inside. Can be C++,
1364 // java, or empty for C.
1365 std::vector<std::string> language_stack_;
1366 // New input files found to add to the link.
1367 Input_arguments* inputs_;
1370 // FILE was found as an argument on the command line. Try to read it
1371 // as a script. Return true if the file was handled.
1374 read_input_script(Workqueue* workqueue, Symbol_table* symtab, Layout* layout,
1375 Dirsearch* dirsearch, int dirindex,
1376 Input_objects* input_objects, Mapfile* mapfile,
1377 Input_group* input_group,
1378 const Input_argument* input_argument,
1379 Input_file* input_file, Task_token* next_blocker,
1380 bool* used_next_blocker)
1382 *used_next_blocker = false;
1384 std::string input_string;
1385 Lex::read_file(input_file, &input_string);
1387 Lex lex(input_string.c_str(), input_string.length(), PARSING_LINKER_SCRIPT);
1389 Parser_closure closure(input_file->filename().c_str(),
1390 input_argument->file().options(),
1392 input_group != NULL,
1393 input_file->is_in_sysroot(),
1395 layout->script_options(),
1397 input_file->will_search_for());
1399 if (yyparse(&closure) != 0)
1401 if (closure.found_incompatible_target())
1403 Read_symbols::incompatible_warning(input_argument, input_file);
1404 Read_symbols::requeue(workqueue, input_objects, symtab, layout,
1405 dirsearch, dirindex, mapfile, input_argument,
1406 input_group, next_blocker);
1412 if (!closure.saw_inputs())
1415 Task_token* this_blocker = NULL;
1416 for (Input_arguments::const_iterator p = closure.inputs()->begin();
1417 p != closure.inputs()->end();
1421 if (p + 1 == closure.inputs()->end())
1425 nb = new Task_token(true);
1428 workqueue->queue_soon(new Read_symbols(input_objects, symtab,
1429 layout, dirsearch, 0, mapfile, &*p,
1430 input_group, this_blocker, nb));
1434 if (layout->incremental_inputs())
1436 // Like new Read_symbols(...) above, we rely on close.inputs()
1437 // getting leaked by closure.
1438 Script_info* info = new Script_info(closure.inputs());
1439 layout->incremental_inputs()->report_script(
1441 input_file->file().get_mtime(),
1444 *used_next_blocker = true;
1449 // Helper function for read_version_script() and
1450 // read_commandline_script(). Processes the given file in the mode
1451 // indicated by first_token and lex_mode.
1454 read_script_file(const char* filename, Command_line* cmdline,
1455 Script_options* script_options,
1456 int first_token, Lex::Mode lex_mode)
1458 // TODO: if filename is a relative filename, search for it manually
1459 // using "." + cmdline->options()->search_path() -- not dirsearch.
1460 Dirsearch dirsearch;
1462 // The file locking code wants to record a Task, but we haven't
1463 // started the workqueue yet. This is only for debugging purposes,
1464 // so we invent a fake value.
1465 const Task* task = reinterpret_cast<const Task*>(-1);
1467 // We don't want this file to be opened in binary mode.
1468 Position_dependent_options posdep = cmdline->position_dependent_options();
1469 if (posdep.format_enum() == General_options::OBJECT_FORMAT_BINARY)
1470 posdep.set_format_enum(General_options::OBJECT_FORMAT_ELF);
1471 Input_file_argument input_argument(filename,
1472 Input_file_argument::INPUT_FILE_TYPE_FILE,
1474 Input_file input_file(&input_argument);
1476 if (!input_file.open(dirsearch, task, &dummy))
1479 std::string input_string;
1480 Lex::read_file(&input_file, &input_string);
1482 Lex lex(input_string.c_str(), input_string.length(), first_token);
1483 lex.set_mode(lex_mode);
1485 Parser_closure closure(filename,
1486 cmdline->position_dependent_options(),
1487 first_token == Lex::DYNAMIC_LIST,
1489 input_file.is_in_sysroot(),
1494 if (yyparse(&closure) != 0)
1496 input_file.file().unlock(task);
1500 input_file.file().unlock(task);
1502 gold_assert(!closure.saw_inputs());
1507 // FILENAME was found as an argument to --script (-T).
1508 // Read it as a script, and execute its contents immediately.
1511 read_commandline_script(const char* filename, Command_line* cmdline)
1513 return read_script_file(filename, cmdline, &cmdline->script_options(),
1514 PARSING_LINKER_SCRIPT, Lex::LINKER_SCRIPT);
1517 // FILENAME was found as an argument to --version-script. Read it as
1518 // a version script, and store its contents in
1519 // cmdline->script_options()->version_script_info().
1522 read_version_script(const char* filename, Command_line* cmdline)
1524 return read_script_file(filename, cmdline, &cmdline->script_options(),
1525 PARSING_VERSION_SCRIPT, Lex::VERSION_SCRIPT);
1528 // FILENAME was found as an argument to --dynamic-list. Read it as a
1529 // list of symbols, and store its contents in DYNAMIC_LIST.
1532 read_dynamic_list(const char* filename, Command_line* cmdline,
1533 Script_options* dynamic_list)
1535 return read_script_file(filename, cmdline, dynamic_list,
1536 PARSING_DYNAMIC_LIST, Lex::DYNAMIC_LIST);
1539 // Implement the --defsym option on the command line. Return true if
1543 Script_options::define_symbol(const char* definition)
1545 Lex lex(definition, strlen(definition), PARSING_DEFSYM);
1546 lex.set_mode(Lex::EXPRESSION);
1549 Position_dependent_options posdep_options;
1551 Parser_closure closure("command line", posdep_options, true,
1552 false, false, NULL, this, &lex, false);
1554 if (yyparse(&closure) != 0)
1557 gold_assert(!closure.saw_inputs());
1562 // Print the script to F for debugging.
1565 Script_options::print(FILE* f) const
1567 fprintf(f, "%s: Dumping linker script\n", program_name);
1569 if (!this->entry_.empty())
1570 fprintf(f, "ENTRY(%s)\n", this->entry_.c_str());
1572 for (Symbol_assignments::const_iterator p =
1573 this->symbol_assignments_.begin();
1574 p != this->symbol_assignments_.end();
1578 for (Assertions::const_iterator p = this->assertions_.begin();
1579 p != this->assertions_.end();
1583 this->script_sections_.print(f);
1585 this->version_script_info_.print(f);
1588 // Manage mapping from keywords to the codes expected by the bison
1589 // parser. We construct one global object for each lex mode with
1592 class Keyword_to_parsecode
1595 // The structure which maps keywords to parsecodes.
1596 struct Keyword_parsecode
1599 const char* keyword;
1600 // Corresponding parsecode.
1604 Keyword_to_parsecode(const Keyword_parsecode* keywords,
1606 : keyword_parsecodes_(keywords), keyword_count_(keyword_count)
1609 // Return the parsecode corresponding KEYWORD, or 0 if it is not a
1612 keyword_to_parsecode(const char* keyword, size_t len) const;
1615 const Keyword_parsecode* keyword_parsecodes_;
1616 const int keyword_count_;
1619 // Mapping from keyword string to keyword parsecode. This array must
1620 // be kept in sorted order. Parsecodes are looked up using bsearch.
1621 // This array must correspond to the list of parsecodes in yyscript.y.
1623 static const Keyword_to_parsecode::Keyword_parsecode
1624 script_keyword_parsecodes[] =
1626 { "ABSOLUTE", ABSOLUTE },
1628 { "ALIGN", ALIGN_K },
1629 { "ALIGNOF", ALIGNOF },
1630 { "ASSERT", ASSERT_K },
1631 { "AS_NEEDED", AS_NEEDED },
1636 { "CONSTANT", CONSTANT },
1637 { "CONSTRUCTORS", CONSTRUCTORS },
1638 { "CREATE_OBJECT_SYMBOLS", CREATE_OBJECT_SYMBOLS },
1639 { "DATA_SEGMENT_ALIGN", DATA_SEGMENT_ALIGN },
1640 { "DATA_SEGMENT_END", DATA_SEGMENT_END },
1641 { "DATA_SEGMENT_RELRO_END", DATA_SEGMENT_RELRO_END },
1642 { "DEFINED", DEFINED },
1644 { "EXCLUDE_FILE", EXCLUDE_FILE },
1645 { "EXTERN", EXTERN },
1648 { "FORCE_COMMON_ALLOCATION", FORCE_COMMON_ALLOCATION },
1651 { "INCLUDE", INCLUDE },
1652 { "INHIBIT_COMMON_ALLOCATION", INHIBIT_COMMON_ALLOCATION },
1655 { "LENGTH", LENGTH },
1656 { "LOADADDR", LOADADDR },
1660 { "MEMORY", MEMORY },
1663 { "NOCROSSREFS", NOCROSSREFS },
1664 { "NOFLOAT", NOFLOAT },
1665 { "ONLY_IF_RO", ONLY_IF_RO },
1666 { "ONLY_IF_RW", ONLY_IF_RW },
1667 { "OPTION", OPTION },
1668 { "ORIGIN", ORIGIN },
1669 { "OUTPUT", OUTPUT },
1670 { "OUTPUT_ARCH", OUTPUT_ARCH },
1671 { "OUTPUT_FORMAT", OUTPUT_FORMAT },
1672 { "OVERLAY", OVERLAY },
1674 { "PROVIDE", PROVIDE },
1675 { "PROVIDE_HIDDEN", PROVIDE_HIDDEN },
1677 { "SEARCH_DIR", SEARCH_DIR },
1678 { "SECTIONS", SECTIONS },
1679 { "SEGMENT_START", SEGMENT_START },
1681 { "SIZEOF", SIZEOF },
1682 { "SIZEOF_HEADERS", SIZEOF_HEADERS },
1683 { "SORT", SORT_BY_NAME },
1684 { "SORT_BY_ALIGNMENT", SORT_BY_ALIGNMENT },
1685 { "SORT_BY_NAME", SORT_BY_NAME },
1686 { "SPECIAL", SPECIAL },
1688 { "STARTUP", STARTUP },
1689 { "SUBALIGN", SUBALIGN },
1690 { "SYSLIB", SYSLIB },
1691 { "TARGET", TARGET_K },
1692 { "TRUNCATE", TRUNCATE },
1693 { "VERSION", VERSIONK },
1694 { "global", GLOBAL },
1700 { "sizeof_headers", SIZEOF_HEADERS },
1703 static const Keyword_to_parsecode
1704 script_keywords(&script_keyword_parsecodes[0],
1705 (sizeof(script_keyword_parsecodes)
1706 / sizeof(script_keyword_parsecodes[0])));
1708 static const Keyword_to_parsecode::Keyword_parsecode
1709 version_script_keyword_parsecodes[] =
1711 { "extern", EXTERN },
1712 { "global", GLOBAL },
1716 static const Keyword_to_parsecode
1717 version_script_keywords(&version_script_keyword_parsecodes[0],
1718 (sizeof(version_script_keyword_parsecodes)
1719 / sizeof(version_script_keyword_parsecodes[0])));
1721 static const Keyword_to_parsecode::Keyword_parsecode
1722 dynamic_list_keyword_parsecodes[] =
1724 { "extern", EXTERN },
1727 static const Keyword_to_parsecode
1728 dynamic_list_keywords(&dynamic_list_keyword_parsecodes[0],
1729 (sizeof(dynamic_list_keyword_parsecodes)
1730 / sizeof(dynamic_list_keyword_parsecodes[0])));
1734 // Comparison function passed to bsearch.
1746 ktt_compare(const void* keyv, const void* kttv)
1748 const Ktt_key* key = static_cast<const Ktt_key*>(keyv);
1749 const Keyword_to_parsecode::Keyword_parsecode* ktt =
1750 static_cast<const Keyword_to_parsecode::Keyword_parsecode*>(kttv);
1751 int i = strncmp(key->str, ktt->keyword, key->len);
1754 if (ktt->keyword[key->len] != '\0')
1759 } // End extern "C".
1762 Keyword_to_parsecode::keyword_to_parsecode(const char* keyword,
1768 void* kttv = bsearch(&key,
1769 this->keyword_parsecodes_,
1770 this->keyword_count_,
1771 sizeof(this->keyword_parsecodes_[0]),
1775 Keyword_parsecode* ktt = static_cast<Keyword_parsecode*>(kttv);
1776 return ktt->parsecode;
1779 // Helper class that calls cplus_demangle when needed and takes care of freeing
1782 class Lazy_demangler
1785 Lazy_demangler(const char* symbol, int options)
1786 : symbol_(symbol), options_(options), demangled_(NULL), did_demangle_(false)
1790 { free(this->demangled_); }
1792 // Return the demangled name. The actual demangling happens on the first call,
1793 // and the result is later cached.
1799 // The symbol to demangle.
1800 const char *symbol_;
1801 // Option flags to pass to cplus_demagle.
1803 // The cached demangled value, or NULL if demangling didn't happen yet or
1806 // Whether we already called cplus_demangle
1810 // Return the demangled name. The actual demangling happens on the first call,
1811 // and the result is later cached. Returns NULL if the symbol cannot be
1815 Lazy_demangler::get()
1817 if (!this->did_demangle_)
1819 this->demangled_ = cplus_demangle(this->symbol_, this->options_);
1820 this->did_demangle_ = true;
1822 return this->demangled_;
1825 // The following structs are used within the VersionInfo class as well
1826 // as in the bison helper functions. They store the information
1827 // parsed from the version script.
1829 // A single version expression.
1830 // For example, pattern="std::map*" and language="C++".
1831 // pattern and language should be from the stringpool
1832 struct Version_expression {
1833 Version_expression(const std::string& pattern,
1834 const std::string& language,
1836 : pattern(pattern), language(language), exact_match(exact_match) {}
1838 std::string pattern;
1839 std::string language;
1840 // If false, we use glob() to match pattern. If true, we use strcmp().
1845 // A list of expressions.
1846 struct Version_expression_list {
1847 std::vector<struct Version_expression> expressions;
1851 // A list of which versions upon which another version depends.
1852 // Strings should be from the Stringpool.
1853 struct Version_dependency_list {
1854 std::vector<std::string> dependencies;
1858 // The total definition of a version. It includes the tag for the
1859 // version, its global and local expressions, and any dependencies.
1860 struct Version_tree {
1862 : tag(), global(NULL), local(NULL), dependencies(NULL) {}
1865 const struct Version_expression_list* global;
1866 const struct Version_expression_list* local;
1867 const struct Version_dependency_list* dependencies;
1870 Version_script_info::~Version_script_info()
1876 Version_script_info::clear()
1878 for (size_t k = 0; k < dependency_lists_.size(); ++k)
1879 delete dependency_lists_[k];
1880 this->dependency_lists_.clear();
1881 for (size_t k = 0; k < version_trees_.size(); ++k)
1882 delete version_trees_[k];
1883 this->version_trees_.clear();
1884 for (size_t k = 0; k < expression_lists_.size(); ++k)
1885 delete expression_lists_[k];
1886 this->expression_lists_.clear();
1889 std::vector<std::string>
1890 Version_script_info::get_versions() const
1892 std::vector<std::string> ret;
1893 for (size_t j = 0; j < version_trees_.size(); ++j)
1894 if (!this->version_trees_[j]->tag.empty())
1895 ret.push_back(this->version_trees_[j]->tag);
1899 std::vector<std::string>
1900 Version_script_info::get_dependencies(const char* version) const
1902 std::vector<std::string> ret;
1903 for (size_t j = 0; j < version_trees_.size(); ++j)
1904 if (version_trees_[j]->tag == version)
1906 const struct Version_dependency_list* deps =
1907 version_trees_[j]->dependencies;
1909 for (size_t k = 0; k < deps->dependencies.size(); ++k)
1910 ret.push_back(deps->dependencies[k]);
1916 // Look up SYMBOL_NAME in the list of versions. If CHECK_GLOBAL is
1917 // true look at the globally visible symbols, otherwise look at the
1918 // symbols listed as "local:". Return true if the symbol is found,
1919 // false otherwise. If the symbol is found, then if PVERSION is not
1920 // NULL, set *PVERSION to the version.
1923 Version_script_info::get_symbol_version_helper(const char* symbol_name,
1925 std::string* pversion) const
1927 Lazy_demangler cpp_demangled_name(symbol_name, DMGL_ANSI | DMGL_PARAMS);
1928 Lazy_demangler java_demangled_name(symbol_name,
1929 DMGL_ANSI | DMGL_PARAMS | DMGL_JAVA);
1930 for (size_t j = 0; j < version_trees_.size(); ++j)
1932 // Is it a global symbol for this version?
1933 const Version_expression_list* explist =
1934 check_global ? version_trees_[j]->global : version_trees_[j]->local;
1935 if (explist != NULL)
1936 for (size_t k = 0; k < explist->expressions.size(); ++k)
1938 const char* name_to_match = symbol_name;
1939 const struct Version_expression& exp = explist->expressions[k];
1940 if (exp.language == "C++")
1942 name_to_match = cpp_demangled_name.get();
1943 // This isn't a C++ symbol.
1944 if (name_to_match == NULL)
1947 else if (exp.language == "Java")
1949 name_to_match = java_demangled_name.get();
1950 // This isn't a Java symbol.
1951 if (name_to_match == NULL)
1955 if (exp.exact_match)
1956 matched = strcmp(exp.pattern.c_str(), name_to_match) == 0;
1958 matched = fnmatch(exp.pattern.c_str(), name_to_match,
1962 if (pversion != NULL)
1963 *pversion = this->version_trees_[j]->tag;
1971 struct Version_dependency_list*
1972 Version_script_info::allocate_dependency_list()
1974 dependency_lists_.push_back(new Version_dependency_list);
1975 return dependency_lists_.back();
1978 struct Version_expression_list*
1979 Version_script_info::allocate_expression_list()
1981 expression_lists_.push_back(new Version_expression_list);
1982 return expression_lists_.back();
1985 struct Version_tree*
1986 Version_script_info::allocate_version_tree()
1988 version_trees_.push_back(new Version_tree);
1989 return version_trees_.back();
1992 // Print for debugging.
1995 Version_script_info::print(FILE* f) const
2000 fprintf(f, "VERSION {");
2002 for (size_t i = 0; i < this->version_trees_.size(); ++i)
2004 const Version_tree* vt = this->version_trees_[i];
2006 if (vt->tag.empty())
2009 fprintf(f, " %s {\n", vt->tag.c_str());
2011 if (vt->global != NULL)
2013 fprintf(f, " global :\n");
2014 this->print_expression_list(f, vt->global);
2017 if (vt->local != NULL)
2019 fprintf(f, " local :\n");
2020 this->print_expression_list(f, vt->local);
2024 if (vt->dependencies != NULL)
2026 const Version_dependency_list* deps = vt->dependencies;
2027 for (size_t j = 0; j < deps->dependencies.size(); ++j)
2029 if (j < deps->dependencies.size() - 1)
2031 fprintf(f, " %s", deps->dependencies[j].c_str());
2041 Version_script_info::print_expression_list(
2043 const Version_expression_list* vel) const
2045 std::string current_language;
2046 for (size_t i = 0; i < vel->expressions.size(); ++i)
2048 const Version_expression& ve(vel->expressions[i]);
2050 if (ve.language != current_language)
2052 if (!current_language.empty())
2054 fprintf(f, " extern \"%s\" {\n", ve.language.c_str());
2055 current_language = ve.language;
2059 if (!current_language.empty())
2064 fprintf(f, "%s", ve.pattern.c_str());
2071 if (!current_language.empty())
2075 } // End namespace gold.
2077 // The remaining functions are extern "C", so it's clearer to not put
2078 // them in namespace gold.
2080 using namespace gold;
2082 // This function is called by the bison parser to return the next
2086 yylex(YYSTYPE* lvalp, void* closurev)
2088 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2089 const Token* token = closure->next_token();
2090 switch (token->classification())
2095 case Token::TOKEN_INVALID:
2096 yyerror(closurev, "invalid character");
2099 case Token::TOKEN_EOF:
2102 case Token::TOKEN_STRING:
2104 // This is either a keyword or a STRING.
2106 const char* str = token->string_value(&len);
2108 switch (closure->lex_mode())
2110 case Lex::LINKER_SCRIPT:
2111 parsecode = script_keywords.keyword_to_parsecode(str, len);
2113 case Lex::VERSION_SCRIPT:
2114 parsecode = version_script_keywords.keyword_to_parsecode(str, len);
2116 case Lex::DYNAMIC_LIST:
2117 parsecode = dynamic_list_keywords.keyword_to_parsecode(str, len);
2124 lvalp->string.value = str;
2125 lvalp->string.length = len;
2129 case Token::TOKEN_QUOTED_STRING:
2130 lvalp->string.value = token->string_value(&lvalp->string.length);
2131 return QUOTED_STRING;
2133 case Token::TOKEN_OPERATOR:
2134 return token->operator_value();
2136 case Token::TOKEN_INTEGER:
2137 lvalp->integer = token->integer_value();
2142 // This function is called by the bison parser to report an error.
2145 yyerror(void* closurev, const char* message)
2147 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2148 gold_error(_("%s:%d:%d: %s"), closure->filename(), closure->lineno(),
2149 closure->charpos(), message);
2152 // Called by the bison parser to add an external symbol to the link.
2155 script_add_extern(void* closurev, const char* name, size_t length)
2157 // We treat exactly like -u NAME. FIXME: If it seems useful, we
2158 // could handle this after the command line has been read, by adding
2159 // entries to the symbol table directly.
2160 std::string arg("--undefined=");
2161 arg.append(name, length);
2162 script_parse_option(closurev, arg.c_str(), arg.size());
2165 // Called by the bison parser to add a file to the link.
2168 script_add_file(void* closurev, const char* name, size_t length)
2170 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2172 // If this is an absolute path, and we found the script in the
2173 // sysroot, then we want to prepend the sysroot to the file name.
2174 // For example, this is how we handle a cross link to the x86_64
2175 // libc.so, which refers to /lib/libc.so.6.
2176 std::string name_string(name, length);
2177 const char* extra_search_path = ".";
2178 std::string script_directory;
2179 if (IS_ABSOLUTE_PATH(name_string.c_str()))
2181 if (closure->is_in_sysroot())
2183 const std::string& sysroot(parameters->options().sysroot());
2184 gold_assert(!sysroot.empty());
2185 name_string = sysroot + name_string;
2190 // In addition to checking the normal library search path, we
2191 // also want to check in the script-directory.
2192 const char *slash = strrchr(closure->filename(), '/');
2195 script_directory.assign(closure->filename(),
2196 slash - closure->filename() + 1);
2197 extra_search_path = script_directory.c_str();
2201 Input_file_argument file(name_string.c_str(),
2202 Input_file_argument::INPUT_FILE_TYPE_FILE,
2203 extra_search_path, false,
2204 closure->position_dependent_options());
2205 closure->inputs()->add_file(file);
2208 // Called by the bison parser to start a group. If we are already in
2209 // a group, that means that this script was invoked within a
2210 // --start-group --end-group sequence on the command line, or that
2211 // this script was found in a GROUP of another script. In that case,
2212 // we simply continue the existing group, rather than starting a new
2213 // one. It is possible to construct a case in which this will do
2214 // something other than what would happen if we did a recursive group,
2215 // but it's hard to imagine why the different behaviour would be
2216 // useful for a real program. Avoiding recursive groups is simpler
2217 // and more efficient.
2220 script_start_group(void* closurev)
2222 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2223 if (!closure->in_group())
2224 closure->inputs()->start_group();
2227 // Called by the bison parser at the end of a group.
2230 script_end_group(void* closurev)
2232 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2233 if (!closure->in_group())
2234 closure->inputs()->end_group();
2237 // Called by the bison parser to start an AS_NEEDED list.
2240 script_start_as_needed(void* closurev)
2242 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2243 closure->position_dependent_options().set_as_needed(true);
2246 // Called by the bison parser at the end of an AS_NEEDED list.
2249 script_end_as_needed(void* closurev)
2251 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2252 closure->position_dependent_options().set_as_needed(false);
2255 // Called by the bison parser to set the entry symbol.
2258 script_set_entry(void* closurev, const char* entry, size_t length)
2260 // We'll parse this exactly the same as --entry=ENTRY on the commandline
2261 // TODO(csilvers): FIXME -- call set_entry directly.
2262 std::string arg("--entry=");
2263 arg.append(entry, length);
2264 script_parse_option(closurev, arg.c_str(), arg.size());
2267 // Called by the bison parser to set whether to define common symbols.
2270 script_set_common_allocation(void* closurev, int set)
2272 const char* arg = set != 0 ? "--define-common" : "--no-define-common";
2273 script_parse_option(closurev, arg, strlen(arg));
2276 // Called by the bison parser to define a symbol.
2279 script_set_symbol(void* closurev, const char* name, size_t length,
2280 Expression* value, int providei, int hiddeni)
2282 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2283 const bool provide = providei != 0;
2284 const bool hidden = hiddeni != 0;
2285 closure->script_options()->add_symbol_assignment(name, length,
2286 closure->parsing_defsym(),
2287 value, provide, hidden);
2288 closure->clear_skip_on_incompatible_target();
2291 // Called by the bison parser to add an assertion.
2294 script_add_assertion(void* closurev, Expression* check, const char* message,
2297 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2298 closure->script_options()->add_assertion(check, message, messagelen);
2299 closure->clear_skip_on_incompatible_target();
2302 // Called by the bison parser to parse an OPTION.
2305 script_parse_option(void* closurev, const char* option, size_t length)
2307 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2308 // We treat the option as a single command-line option, even if
2309 // it has internal whitespace.
2310 if (closure->command_line() == NULL)
2312 // There are some options that we could handle here--e.g.,
2313 // -lLIBRARY. Should we bother?
2314 gold_warning(_("%s:%d:%d: ignoring command OPTION; OPTION is only valid"
2315 " for scripts specified via -T/--script"),
2316 closure->filename(), closure->lineno(), closure->charpos());
2320 bool past_a_double_dash_option = false;
2321 const char* mutable_option = strndup(option, length);
2322 gold_assert(mutable_option != NULL);
2323 closure->command_line()->process_one_option(1, &mutable_option, 0,
2324 &past_a_double_dash_option);
2325 // The General_options class will quite possibly store a pointer
2326 // into mutable_option, so we can't free it. In cases the class
2327 // does not store such a pointer, this is a memory leak. Alas. :(
2329 closure->clear_skip_on_incompatible_target();
2332 // Called by the bison parser to handle OUTPUT_FORMAT. OUTPUT_FORMAT
2333 // takes either one or three arguments. In the three argument case,
2334 // the format depends on the endianness option, which we don't
2335 // currently support (FIXME). If we see an OUTPUT_FORMAT for the
2336 // wrong format, then we want to search for a new file. Returning 0
2337 // here will cause the parser to immediately abort.
2340 script_check_output_format(void* closurev,
2341 const char* default_name, size_t default_length,
2342 const char*, size_t, const char*, size_t)
2344 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2345 std::string name(default_name, default_length);
2346 Target* target = select_target_by_name(name.c_str());
2347 if (target == NULL || !parameters->is_compatible_target(target))
2349 if (closure->skip_on_incompatible_target())
2351 closure->set_found_incompatible_target();
2354 // FIXME: Should we warn about the unknown target?
2359 // Called by the bison parser to handle TARGET.
2362 script_set_target(void* closurev, const char* target, size_t len)
2364 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2365 std::string s(target, len);
2366 General_options::Object_format format_enum;
2367 format_enum = General_options::string_to_object_format(s.c_str());
2368 closure->position_dependent_options().set_format_enum(format_enum);
2371 // Called by the bison parser to handle SEARCH_DIR. This is handled
2372 // exactly like a -L option.
2375 script_add_search_dir(void* closurev, const char* option, size_t length)
2377 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2378 if (closure->command_line() == NULL)
2379 gold_warning(_("%s:%d:%d: ignoring SEARCH_DIR; SEARCH_DIR is only valid"
2380 " for scripts specified via -T/--script"),
2381 closure->filename(), closure->lineno(), closure->charpos());
2384 std::string s = "-L" + std::string(option, length);
2385 script_parse_option(closurev, s.c_str(), s.size());
2389 /* Called by the bison parser to push the lexer into expression
2393 script_push_lex_into_expression_mode(void* closurev)
2395 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2396 closure->push_lex_mode(Lex::EXPRESSION);
2399 /* Called by the bison parser to push the lexer into version
2403 script_push_lex_into_version_mode(void* closurev)
2405 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2406 closure->push_lex_mode(Lex::VERSION_SCRIPT);
2409 /* Called by the bison parser to pop the lexer mode. */
2412 script_pop_lex_mode(void* closurev)
2414 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2415 closure->pop_lex_mode();
2418 // Register an entire version node. For example:
2424 // - tag is "GLIBC_2.1"
2425 // - tree contains the information "global: foo"
2426 // - deps contains "GLIBC_2.0"
2429 script_register_vers_node(void*,
2432 struct Version_tree *tree,
2433 struct Version_dependency_list *deps)
2435 gold_assert(tree != NULL);
2436 tree->dependencies = deps;
2438 tree->tag = std::string(tag, taglen);
2441 // Add a dependencies to the list of existing dependencies, if any,
2442 // and return the expanded list.
2444 extern "C" struct Version_dependency_list *
2445 script_add_vers_depend(void* closurev,
2446 struct Version_dependency_list *all_deps,
2447 const char *depend_to_add, int deplen)
2449 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2450 if (all_deps == NULL)
2451 all_deps = closure->version_script()->allocate_dependency_list();
2452 all_deps->dependencies.push_back(std::string(depend_to_add, deplen));
2456 // Add a pattern expression to an existing list of expressions, if any.
2457 // TODO: In the old linker, the last argument used to be a bool, but I
2458 // don't know what it meant.
2460 extern "C" struct Version_expression_list *
2461 script_new_vers_pattern(void* closurev,
2462 struct Version_expression_list *expressions,
2463 const char *pattern, int patlen, int exact_match)
2465 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2466 if (expressions == NULL)
2467 expressions = closure->version_script()->allocate_expression_list();
2468 expressions->expressions.push_back(
2469 Version_expression(std::string(pattern, patlen),
2470 closure->get_current_language(),
2471 static_cast<bool>(exact_match)));
2475 // Attaches b to the end of a, and clears b. So a = a + b and b = {}.
2477 extern "C" struct Version_expression_list*
2478 script_merge_expressions(struct Version_expression_list *a,
2479 struct Version_expression_list *b)
2481 a->expressions.insert(a->expressions.end(),
2482 b->expressions.begin(), b->expressions.end());
2483 // We could delete b and remove it from expressions_lists_, but
2484 // that's a lot of work. This works just as well.
2485 b->expressions.clear();
2489 // Combine the global and local expressions into a a Version_tree.
2491 extern "C" struct Version_tree *
2492 script_new_vers_node(void* closurev,
2493 struct Version_expression_list *global,
2494 struct Version_expression_list *local)
2496 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2497 Version_tree* tree = closure->version_script()->allocate_version_tree();
2498 tree->global = global;
2499 tree->local = local;
2503 // Handle a transition in language, such as at the
2504 // start or end of 'extern "C++"'
2507 version_script_push_lang(void* closurev, const char* lang, int langlen)
2509 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2510 closure->push_language(std::string(lang, langlen));
2514 version_script_pop_lang(void* closurev)
2516 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2517 closure->pop_language();
2520 // Called by the bison parser to start a SECTIONS clause.
2523 script_start_sections(void* closurev)
2525 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2526 closure->script_options()->script_sections()->start_sections();
2527 closure->clear_skip_on_incompatible_target();
2530 // Called by the bison parser to finish a SECTIONS clause.
2533 script_finish_sections(void* closurev)
2535 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2536 closure->script_options()->script_sections()->finish_sections();
2539 // Start processing entries for an output section.
2542 script_start_output_section(void* closurev, const char* name, size_t namelen,
2543 const struct Parser_output_section_header* header)
2545 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2546 closure->script_options()->script_sections()->start_output_section(name,
2551 // Finish processing entries for an output section.
2554 script_finish_output_section(void* closurev,
2555 const struct Parser_output_section_trailer* trail)
2557 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2558 closure->script_options()->script_sections()->finish_output_section(trail);
2561 // Add a data item (e.g., "WORD (0)") to the current output section.
2564 script_add_data(void* closurev, int data_token, Expression* val)
2566 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2568 bool is_signed = true;
2590 closure->script_options()->script_sections()->add_data(size, is_signed, val);
2593 // Add a clause setting the fill value to the current output section.
2596 script_add_fill(void* closurev, Expression* val)
2598 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2599 closure->script_options()->script_sections()->add_fill(val);
2602 // Add a new input section specification to the current output
2606 script_add_input_section(void* closurev,
2607 const struct Input_section_spec* spec,
2610 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2611 bool keep = keepi != 0;
2612 closure->script_options()->script_sections()->add_input_section(spec, keep);
2615 // When we see DATA_SEGMENT_ALIGN we record that following output
2616 // sections may be relro.
2619 script_data_segment_align(void* closurev)
2621 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2622 if (!closure->script_options()->saw_sections_clause())
2623 gold_error(_("%s:%d:%d: DATA_SEGMENT_ALIGN not in SECTIONS clause"),
2624 closure->filename(), closure->lineno(), closure->charpos());
2626 closure->script_options()->script_sections()->data_segment_align();
2629 // When we see DATA_SEGMENT_RELRO_END we know that all output sections
2630 // since DATA_SEGMENT_ALIGN should be relro.
2633 script_data_segment_relro_end(void* closurev)
2635 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2636 if (!closure->script_options()->saw_sections_clause())
2637 gold_error(_("%s:%d:%d: DATA_SEGMENT_ALIGN not in SECTIONS clause"),
2638 closure->filename(), closure->lineno(), closure->charpos());
2640 closure->script_options()->script_sections()->data_segment_relro_end();
2643 // Create a new list of string/sort pairs.
2645 extern "C" String_sort_list_ptr
2646 script_new_string_sort_list(const struct Wildcard_section* string_sort)
2648 return new String_sort_list(1, *string_sort);
2651 // Add an entry to a list of string/sort pairs. The way the parser
2652 // works permits us to simply modify the first parameter, rather than
2655 extern "C" String_sort_list_ptr
2656 script_string_sort_list_add(String_sort_list_ptr pv,
2657 const struct Wildcard_section* string_sort)
2660 return script_new_string_sort_list(string_sort);
2663 pv->push_back(*string_sort);
2668 // Create a new list of strings.
2670 extern "C" String_list_ptr
2671 script_new_string_list(const char* str, size_t len)
2673 return new String_list(1, std::string(str, len));
2676 // Add an element to a list of strings. The way the parser works
2677 // permits us to simply modify the first parameter, rather than copy
2680 extern "C" String_list_ptr
2681 script_string_list_push_back(String_list_ptr pv, const char* str, size_t len)
2684 return script_new_string_list(str, len);
2687 pv->push_back(std::string(str, len));
2692 // Concatenate two string lists. Either or both may be NULL. The way
2693 // the parser works permits us to modify the parameters, rather than
2696 extern "C" String_list_ptr
2697 script_string_list_append(String_list_ptr pv1, String_list_ptr pv2)
2703 pv1->insert(pv1->end(), pv2->begin(), pv2->end());
2707 // Add a new program header.
2710 script_add_phdr(void* closurev, const char* name, size_t namelen,
2711 unsigned int type, const Phdr_info* info)
2713 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2714 bool includes_filehdr = info->includes_filehdr != 0;
2715 bool includes_phdrs = info->includes_phdrs != 0;
2716 bool is_flags_valid = info->is_flags_valid != 0;
2717 Script_sections* ss = closure->script_options()->script_sections();
2718 ss->add_phdr(name, namelen, type, includes_filehdr, includes_phdrs,
2719 is_flags_valid, info->flags, info->load_address);
2720 closure->clear_skip_on_incompatible_target();
2723 // Convert a program header string to a type.
2725 #define PHDR_TYPE(NAME) { #NAME, sizeof(#NAME) - 1, elfcpp::NAME }
2732 } phdr_type_names[] =
2736 PHDR_TYPE(PT_DYNAMIC),
2737 PHDR_TYPE(PT_INTERP),
2739 PHDR_TYPE(PT_SHLIB),
2742 PHDR_TYPE(PT_GNU_EH_FRAME),
2743 PHDR_TYPE(PT_GNU_STACK),
2744 PHDR_TYPE(PT_GNU_RELRO)
2747 extern "C" unsigned int
2748 script_phdr_string_to_type(void* closurev, const char* name, size_t namelen)
2750 for (unsigned int i = 0;
2751 i < sizeof(phdr_type_names) / sizeof(phdr_type_names[0]);
2753 if (namelen == phdr_type_names[i].namelen
2754 && strncmp(name, phdr_type_names[i].name, namelen) == 0)
2755 return phdr_type_names[i].val;
2756 yyerror(closurev, _("unknown PHDR type (try integer)"));
2757 return elfcpp::PT_NULL;
2761 script_saw_segment_start_expression(void* closurev)
2763 Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2764 Script_sections* ss = closure->script_options()->script_sections();
2765 ss->set_saw_segment_start_expression(true);