1 // script.cc -- handle linker scripts for gold.
3 // Copyright 2006, 2007, 2008, 2009, 2010 Free Software Foundation, Inc.
4 // Written by Ian Lance Taylor <iant@google.com>.
6 // This file is part of gold.
8 // This program is free software; you can redistribute it and/or modify
9 // it under the terms of the GNU General Public License as published by
10 // the Free Software Foundation; either version 3 of the License, or
11 // (at your option) any later version.
13 // This program is distributed in the hope that it will be useful,
14 // but WITHOUT ANY WARRANTY; without even the implied warranty of
15 // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 // GNU General Public License for more details.
18 // You should have received a copy of the GNU General Public License
19 // along with this program; if not, write to the Free Software
20 // Foundation, Inc., 51 Franklin Street - Fifth Floor, Boston,
21 // MA 02110-1301, USA.
31 #include "filenames.h"
35 #include "dirsearch.h"
38 #include "workqueue.h"
40 #include "parameters.h"
43 #include "target-select.h"
46 #include "incremental.h"
51 // A token read from a script file. We don't implement keywords here;
52 // all keywords are simply represented as a string.
57 // Token classification.
62 // Token indicates end of input.
64 // Token is a string of characters.
66 // Token is a quoted string of characters.
68 // Token is an operator.
70 // Token is a number (an integer).
74 // We need an empty constructor so that we can put this STL objects.
76 : classification_(TOKEN_INVALID
), value_(NULL
), value_length_(0),
77 opcode_(0), lineno_(0), charpos_(0)
80 // A general token with no value.
81 Token(Classification classification
, int lineno
, int charpos
)
82 : classification_(classification
), value_(NULL
), value_length_(0),
83 opcode_(0), lineno_(lineno
), charpos_(charpos
)
85 gold_assert(classification
== TOKEN_INVALID
86 || classification
== TOKEN_EOF
);
89 // A general token with a value.
90 Token(Classification classification
, const char* value
, size_t length
,
91 int lineno
, int charpos
)
92 : classification_(classification
), value_(value
), value_length_(length
),
93 opcode_(0), lineno_(lineno
), charpos_(charpos
)
95 gold_assert(classification
!= TOKEN_INVALID
96 && classification
!= TOKEN_EOF
);
99 // A token representing an operator.
100 Token(int opcode
, int lineno
, int charpos
)
101 : classification_(TOKEN_OPERATOR
), value_(NULL
), value_length_(0),
102 opcode_(opcode
), lineno_(lineno
), charpos_(charpos
)
105 // Return whether the token is invalid.
108 { return this->classification_
== TOKEN_INVALID
; }
110 // Return whether this is an EOF token.
113 { return this->classification_
== TOKEN_EOF
; }
115 // Return the token classification.
117 classification() const
118 { return this->classification_
; }
120 // Return the line number at which the token starts.
123 { return this->lineno_
; }
125 // Return the character position at this the token starts.
128 { return this->charpos_
; }
130 // Get the value of a token.
133 string_value(size_t* length
) const
135 gold_assert(this->classification_
== TOKEN_STRING
136 || this->classification_
== TOKEN_QUOTED_STRING
);
137 *length
= this->value_length_
;
142 operator_value() const
144 gold_assert(this->classification_
== TOKEN_OPERATOR
);
145 return this->opcode_
;
149 integer_value() const
151 gold_assert(this->classification_
== TOKEN_INTEGER
);
153 std::string
s(this->value_
, this->value_length_
);
154 return strtoull(s
.c_str(), NULL
, 0);
158 // The token classification.
159 Classification classification_
;
160 // The token value, for TOKEN_STRING or TOKEN_QUOTED_STRING or
163 // The length of the token value.
164 size_t value_length_
;
165 // The token value, for TOKEN_OPERATOR.
167 // The line number where this token started (one based).
169 // The character position within the line where this token started
174 // This class handles lexing a file into a sequence of tokens.
179 // We unfortunately have to support different lexing modes, because
180 // when reading different parts of a linker script we need to parse
181 // things differently.
184 // Reading an ordinary linker script.
186 // Reading an expression in a linker script.
188 // Reading a version script.
190 // Reading a --dynamic-list file.
194 Lex(const char* input_string
, size_t input_length
, int parsing_token
)
195 : input_string_(input_string
), input_length_(input_length
),
196 current_(input_string
), mode_(LINKER_SCRIPT
),
197 first_token_(parsing_token
), token_(),
198 lineno_(1), linestart_(input_string
)
201 // Read a file into a string.
203 read_file(Input_file
*, std::string
*);
205 // Return the next token.
209 // Return the current lexing mode.
212 { return this->mode_
; }
214 // Set the lexing mode.
217 { this->mode_
= mode
; }
221 Lex
& operator=(const Lex
&);
223 // Make a general token with no value at the current location.
225 make_token(Token::Classification c
, const char* start
) const
226 { return Token(c
, this->lineno_
, start
- this->linestart_
+ 1); }
228 // Make a general token with a value at the current location.
230 make_token(Token::Classification c
, const char* v
, size_t len
,
233 { return Token(c
, v
, len
, this->lineno_
, start
- this->linestart_
+ 1); }
235 // Make an operator token at the current location.
237 make_token(int opcode
, const char* start
) const
238 { return Token(opcode
, this->lineno_
, start
- this->linestart_
+ 1); }
240 // Make an invalid token at the current location.
242 make_invalid_token(const char* start
)
243 { return this->make_token(Token::TOKEN_INVALID
, start
); }
245 // Make an EOF token at the current location.
247 make_eof_token(const char* start
)
248 { return this->make_token(Token::TOKEN_EOF
, start
); }
250 // Return whether C can be the first character in a name. C2 is the
251 // next character, since we sometimes need that.
253 can_start_name(char c
, char c2
);
255 // If C can appear in a name which has already started, return a
256 // pointer to a character later in the token or just past
257 // it. Otherwise, return NULL.
259 can_continue_name(const char* c
);
261 // Return whether C, C2, C3 can start a hex number.
263 can_start_hex(char c
, char c2
, char c3
);
265 // If C can appear in a hex number which has already started, return
266 // a pointer to a character later in the token or just past
267 // it. Otherwise, return NULL.
269 can_continue_hex(const char* c
);
271 // Return whether C can start a non-hex number.
273 can_start_number(char c
);
275 // If C can appear in a decimal number which has already started,
276 // return a pointer to a character later in the token or just past
277 // it. Otherwise, return NULL.
279 can_continue_number(const char* c
)
280 { return Lex::can_start_number(*c
) ? c
+ 1 : NULL
; }
282 // If C1 C2 C3 form a valid three character operator, return the
283 // opcode. Otherwise return 0.
285 three_char_operator(char c1
, char c2
, char c3
);
287 // If C1 C2 form a valid two character operator, return the opcode.
288 // Otherwise return 0.
290 two_char_operator(char c1
, char c2
);
292 // If C1 is a valid one character operator, return the opcode.
293 // Otherwise return 0.
295 one_char_operator(char c1
);
297 // Read the next token.
299 get_token(const char**);
301 // Skip a C style /* */ comment. Return false if the comment did
304 skip_c_comment(const char**);
306 // Skip a line # comment. Return false if there was no newline.
308 skip_line_comment(const char**);
310 // Build a token CLASSIFICATION from all characters that match
311 // CAN_CONTINUE_FN. The token starts at START. Start matching from
312 // MATCH. Set *PP to the character following the token.
314 gather_token(Token::Classification
,
315 const char* (Lex::*can_continue_fn
)(const char*),
316 const char* start
, const char* match
, const char** pp
);
318 // Build a token from a quoted string.
320 gather_quoted_string(const char** pp
);
322 // The string we are tokenizing.
323 const char* input_string_
;
324 // The length of the string.
325 size_t input_length_
;
326 // The current offset into the string.
327 const char* current_
;
328 // The current lexing mode.
330 // The code to use for the first token. This is set to 0 after it
333 // The current token.
335 // The current line number.
337 // The start of the current line in the string.
338 const char* linestart_
;
341 // Read the whole file into memory. We don't expect linker scripts to
342 // be large, so we just use a std::string as a buffer. We ignore the
343 // data we've already read, so that we read aligned buffers.
346 Lex::read_file(Input_file
* input_file
, std::string
* contents
)
348 off_t filesize
= input_file
->file().filesize();
350 contents
->reserve(filesize
);
353 unsigned char buf
[BUFSIZ
];
354 while (off
< filesize
)
357 if (get
> filesize
- off
)
358 get
= filesize
- off
;
359 input_file
->file().read(off
, get
, buf
);
360 contents
->append(reinterpret_cast<char*>(&buf
[0]), get
);
365 // Return whether C can be the start of a name, if the next character
366 // is C2. A name can being with a letter, underscore, period, or
367 // dollar sign. Because a name can be a file name, we also permit
368 // forward slash, backslash, and tilde. Tilde is the tricky case
369 // here; GNU ld also uses it as a bitwise not operator. It is only
370 // recognized as the operator if it is not immediately followed by
371 // some character which can appear in a symbol. That is, when we
372 // don't know that we are looking at an expression, "~0" is a file
373 // name, and "~ 0" is an expression using bitwise not. We are
377 Lex::can_start_name(char c
, char c2
)
381 case 'A': case 'B': case 'C': case 'D': case 'E': case 'F':
382 case 'G': case 'H': case 'I': case 'J': case 'K': case 'L':
383 case 'M': case 'N': case 'O': case 'Q': case 'P': case 'R':
384 case 'S': case 'T': case 'U': case 'V': case 'W': case 'X':
386 case 'a': case 'b': case 'c': case 'd': case 'e': case 'f':
387 case 'g': case 'h': case 'i': case 'j': case 'k': case 'l':
388 case 'm': case 'n': case 'o': case 'q': case 'p': case 'r':
389 case 's': case 't': case 'u': case 'v': case 'w': case 'x':
391 case '_': case '.': case '$':
395 return this->mode_
== LINKER_SCRIPT
;
398 return this->mode_
== LINKER_SCRIPT
&& can_continue_name(&c2
);
401 return (this->mode_
== VERSION_SCRIPT
402 || this->mode_
== DYNAMIC_LIST
403 || (this->mode_
== LINKER_SCRIPT
404 && can_continue_name(&c2
)));
411 // Return whether C can continue a name which has already started.
412 // Subsequent characters in a name are the same as the leading
413 // characters, plus digits and "=+-:[],?*". So in general the linker
414 // script language requires spaces around operators, unless we know
415 // that we are parsing an expression.
418 Lex::can_continue_name(const char* c
)
422 case 'A': case 'B': case 'C': case 'D': case 'E': case 'F':
423 case 'G': case 'H': case 'I': case 'J': case 'K': case 'L':
424 case 'M': case 'N': case 'O': case 'Q': case 'P': case 'R':
425 case 'S': case 'T': case 'U': case 'V': case 'W': case 'X':
427 case 'a': case 'b': case 'c': case 'd': case 'e': case 'f':
428 case 'g': case 'h': case 'i': case 'j': case 'k': case 'l':
429 case 'm': case 'n': case 'o': case 'q': case 'p': case 'r':
430 case 's': case 't': case 'u': case 'v': case 'w': case 'x':
432 case '_': case '.': case '$':
433 case '0': case '1': case '2': case '3': case '4':
434 case '5': case '6': case '7': case '8': case '9':
437 // TODO(csilvers): why not allow ~ in names for version-scripts?
438 case '/': case '\\': case '~':
441 if (this->mode_
== LINKER_SCRIPT
)
445 case '[': case ']': case '*': case '?': case '-':
446 if (this->mode_
== LINKER_SCRIPT
|| this->mode_
== VERSION_SCRIPT
447 || this->mode_
== DYNAMIC_LIST
)
451 // TODO(csilvers): why allow this? ^ is meaningless in version scripts.
453 if (this->mode_
== VERSION_SCRIPT
|| this->mode_
== DYNAMIC_LIST
)
458 if (this->mode_
== LINKER_SCRIPT
)
460 else if ((this->mode_
== VERSION_SCRIPT
|| this->mode_
== DYNAMIC_LIST
)
463 // A name can have '::' in it, as that's a c++ namespace
464 // separator. But a single colon is not part of a name.
474 // For a number we accept 0x followed by hex digits, or any sequence
475 // of digits. The old linker accepts leading '$' for hex, and
476 // trailing HXBOD. Those are for MRI compatibility and we don't
477 // accept them. The old linker also accepts trailing MK for mega or
478 // kilo. FIXME: Those are mentioned in the documentation, and we
479 // should accept them.
481 // Return whether C1 C2 C3 can start a hex number.
484 Lex::can_start_hex(char c1
, char c2
, char c3
)
486 if (c1
== '0' && (c2
== 'x' || c2
== 'X'))
487 return this->can_continue_hex(&c3
);
491 // Return whether C can appear in a hex number.
494 Lex::can_continue_hex(const char* c
)
498 case '0': case '1': case '2': case '3': case '4':
499 case '5': case '6': case '7': case '8': case '9':
500 case 'A': case 'B': case 'C': case 'D': case 'E': case 'F':
501 case 'a': case 'b': case 'c': case 'd': case 'e': case 'f':
509 // Return whether C can start a non-hex number.
512 Lex::can_start_number(char c
)
516 case '0': case '1': case '2': case '3': case '4':
517 case '5': case '6': case '7': case '8': case '9':
525 // If C1 C2 C3 form a valid three character operator, return the
526 // opcode (defined in the yyscript.h file generated from yyscript.y).
527 // Otherwise return 0.
530 Lex::three_char_operator(char c1
, char c2
, char c3
)
535 if (c2
== '<' && c3
== '=')
539 if (c2
== '>' && c3
== '=')
548 // If C1 C2 form a valid two character operator, return the opcode
549 // (defined in the yyscript.h file generated from yyscript.y).
550 // Otherwise return 0.
553 Lex::two_char_operator(char c1
, char c2
)
611 // If C1 is a valid operator, return the opcode. Otherwise return 0.
614 Lex::one_char_operator(char c1
)
647 // Skip a C style comment. *PP points to just after the "/*". Return
648 // false if the comment did not end.
651 Lex::skip_c_comment(const char** pp
)
654 while (p
[0] != '*' || p
[1] != '/')
665 this->linestart_
= p
+ 1;
674 // Skip a line # comment. Return false if there was no newline.
677 Lex::skip_line_comment(const char** pp
)
680 size_t skip
= strcspn(p
, "\n");
689 this->linestart_
= p
;
695 // Build a token CLASSIFICATION from all characters that match
696 // CAN_CONTINUE_FN. Update *PP.
699 Lex::gather_token(Token::Classification classification
,
700 const char* (Lex::*can_continue_fn
)(const char*),
705 const char* new_match
= NULL
;
706 while ((new_match
= (this->*can_continue_fn
)(match
)))
709 return this->make_token(classification
, start
, match
- start
, start
);
712 // Build a token from a quoted string.
715 Lex::gather_quoted_string(const char** pp
)
717 const char* start
= *pp
;
718 const char* p
= start
;
720 size_t skip
= strcspn(p
, "\"\n");
722 return this->make_invalid_token(start
);
724 return this->make_token(Token::TOKEN_QUOTED_STRING
, p
, skip
, start
);
727 // Return the next token at *PP. Update *PP. General guideline: we
728 // require linker scripts to be simple ASCII. No unicode linker
729 // scripts. In particular we can assume that any '\0' is the end of
733 Lex::get_token(const char** pp
)
742 return this->make_eof_token(p
);
745 // Skip whitespace quickly.
746 while (*p
== ' ' || *p
== '\t' || *p
== '\r')
753 this->linestart_
= p
;
757 // Skip C style comments.
758 if (p
[0] == '/' && p
[1] == '*')
760 int lineno
= this->lineno_
;
761 int charpos
= p
- this->linestart_
+ 1;
764 if (!this->skip_c_comment(pp
))
765 return Token(Token::TOKEN_INVALID
, lineno
, charpos
);
771 // Skip line comments.
775 if (!this->skip_line_comment(pp
))
776 return this->make_eof_token(p
);
782 if (this->can_start_name(p
[0], p
[1]))
783 return this->gather_token(Token::TOKEN_STRING
,
784 &Lex::can_continue_name
,
787 // We accept any arbitrary name in double quotes, as long as it
788 // does not cross a line boundary.
792 return this->gather_quoted_string(pp
);
795 // Check for a number.
797 if (this->can_start_hex(p
[0], p
[1], p
[2]))
798 return this->gather_token(Token::TOKEN_INTEGER
,
799 &Lex::can_continue_hex
,
802 if (Lex::can_start_number(p
[0]))
803 return this->gather_token(Token::TOKEN_INTEGER
,
804 &Lex::can_continue_number
,
807 // Check for operators.
809 int opcode
= Lex::three_char_operator(p
[0], p
[1], p
[2]);
813 return this->make_token(opcode
, p
);
816 opcode
= Lex::two_char_operator(p
[0], p
[1]);
820 return this->make_token(opcode
, p
);
823 opcode
= Lex::one_char_operator(p
[0]);
827 return this->make_token(opcode
, p
);
830 return this->make_token(Token::TOKEN_INVALID
, p
);
834 // Return the next token.
839 // The first token is special.
840 if (this->first_token_
!= 0)
842 this->token_
= Token(this->first_token_
, 0, 0);
843 this->first_token_
= 0;
844 return &this->token_
;
847 this->token_
= this->get_token(&this->current_
);
849 // Don't let an early null byte fool us into thinking that we've
850 // reached the end of the file.
851 if (this->token_
.is_eof()
852 && (static_cast<size_t>(this->current_
- this->input_string_
)
853 < this->input_length_
))
854 this->token_
= this->make_invalid_token(this->current_
);
856 return &this->token_
;
859 // class Symbol_assignment.
861 // Add the symbol to the symbol table. This makes sure the symbol is
862 // there and defined. The actual value is stored later. We can't
863 // determine the actual value at this point, because we can't
864 // necessarily evaluate the expression until all ordinary symbols have
867 // The GNU linker lets symbol assignments in the linker script
868 // silently override defined symbols in object files. We are
869 // compatible. FIXME: Should we issue a warning?
872 Symbol_assignment::add_to_table(Symbol_table
* symtab
)
874 elfcpp::STV vis
= this->hidden_
? elfcpp::STV_HIDDEN
: elfcpp::STV_DEFAULT
;
875 this->sym_
= symtab
->define_as_constant(this->name_
.c_str(),
878 ? Symbol_table::DEFSYM
879 : Symbol_table::SCRIPT
),
887 true); // force_override
890 // Finalize a symbol value.
893 Symbol_assignment::finalize(Symbol_table
* symtab
, const Layout
* layout
)
895 this->finalize_maybe_dot(symtab
, layout
, false, 0, NULL
);
898 // Finalize a symbol value which can refer to the dot symbol.
901 Symbol_assignment::finalize_with_dot(Symbol_table
* symtab
,
902 const Layout
* layout
,
904 Output_section
* dot_section
)
906 this->finalize_maybe_dot(symtab
, layout
, true, dot_value
, dot_section
);
909 // Finalize a symbol value, internal version.
912 Symbol_assignment::finalize_maybe_dot(Symbol_table
* symtab
,
913 const Layout
* layout
,
914 bool is_dot_available
,
916 Output_section
* dot_section
)
918 // If we were only supposed to provide this symbol, the sym_ field
919 // will be NULL if the symbol was not referenced.
920 if (this->sym_
== NULL
)
922 gold_assert(this->provide_
);
926 if (parameters
->target().get_size() == 32)
928 #if defined(HAVE_TARGET_32_LITTLE) || defined(HAVE_TARGET_32_BIG)
929 this->sized_finalize
<32>(symtab
, layout
, is_dot_available
, dot_value
,
935 else if (parameters
->target().get_size() == 64)
937 #if defined(HAVE_TARGET_64_LITTLE) || defined(HAVE_TARGET_64_BIG)
938 this->sized_finalize
<64>(symtab
, layout
, is_dot_available
, dot_value
,
950 Symbol_assignment::sized_finalize(Symbol_table
* symtab
, const Layout
* layout
,
951 bool is_dot_available
, uint64_t dot_value
,
952 Output_section
* dot_section
)
954 Output_section
* section
;
955 uint64_t final_val
= this->val_
->eval_maybe_dot(symtab
, layout
, true,
957 dot_value
, dot_section
,
959 Sized_symbol
<size
>* ssym
= symtab
->get_sized_symbol
<size
>(this->sym_
);
960 ssym
->set_value(final_val
);
962 ssym
->set_output_section(section
);
965 // Set the symbol value if the expression yields an absolute value.
968 Symbol_assignment::set_if_absolute(Symbol_table
* symtab
, const Layout
* layout
,
969 bool is_dot_available
, uint64_t dot_value
)
971 if (this->sym_
== NULL
)
974 Output_section
* val_section
;
975 uint64_t val
= this->val_
->eval_maybe_dot(symtab
, layout
, false,
976 is_dot_available
, dot_value
,
977 NULL
, &val_section
, NULL
);
978 if (val_section
!= NULL
)
981 if (parameters
->target().get_size() == 32)
983 #if defined(HAVE_TARGET_32_LITTLE) || defined(HAVE_TARGET_32_BIG)
984 Sized_symbol
<32>* ssym
= symtab
->get_sized_symbol
<32>(this->sym_
);
985 ssym
->set_value(val
);
990 else if (parameters
->target().get_size() == 64)
992 #if defined(HAVE_TARGET_64_LITTLE) || defined(HAVE_TARGET_64_BIG)
993 Sized_symbol
<64>* ssym
= symtab
->get_sized_symbol
<64>(this->sym_
);
994 ssym
->set_value(val
);
1003 // Print for debugging.
1006 Symbol_assignment::print(FILE* f
) const
1008 if (this->provide_
&& this->hidden_
)
1009 fprintf(f
, "PROVIDE_HIDDEN(");
1010 else if (this->provide_
)
1011 fprintf(f
, "PROVIDE(");
1012 else if (this->hidden_
)
1015 fprintf(f
, "%s = ", this->name_
.c_str());
1016 this->val_
->print(f
);
1018 if (this->provide_
|| this->hidden_
)
1024 // Class Script_assertion.
1026 // Check the assertion.
1029 Script_assertion::check(const Symbol_table
* symtab
, const Layout
* layout
)
1031 if (!this->check_
->eval(symtab
, layout
, true))
1032 gold_error("%s", this->message_
.c_str());
1035 // Print for debugging.
1038 Script_assertion::print(FILE* f
) const
1040 fprintf(f
, "ASSERT(");
1041 this->check_
->print(f
);
1042 fprintf(f
, ", \"%s\")\n", this->message_
.c_str());
1045 // Class Script_options.
1047 Script_options::Script_options()
1048 : entry_(), symbol_assignments_(), symbol_definitions_(),
1049 symbol_references_(), version_script_info_(), script_sections_()
1053 // Returns true if NAME is on the list of symbol assignments waiting
1057 Script_options::is_pending_assignment(const char* name
)
1059 for (Symbol_assignments::iterator p
= this->symbol_assignments_
.begin();
1060 p
!= this->symbol_assignments_
.end();
1062 if ((*p
)->name() == name
)
1067 // Add a symbol to be defined.
1070 Script_options::add_symbol_assignment(const char* name
, size_t length
,
1071 bool is_defsym
, Expression
* value
,
1072 bool provide
, bool hidden
)
1074 if (length
!= 1 || name
[0] != '.')
1076 if (this->script_sections_
.in_sections_clause())
1078 gold_assert(!is_defsym
);
1079 this->script_sections_
.add_symbol_assignment(name
, length
, value
,
1084 Symbol_assignment
* p
= new Symbol_assignment(name
, length
, is_defsym
,
1085 value
, provide
, hidden
);
1086 this->symbol_assignments_
.push_back(p
);
1091 std::string
n(name
, length
);
1092 this->symbol_definitions_
.insert(n
);
1093 this->symbol_references_
.erase(n
);
1098 if (provide
|| hidden
)
1099 gold_error(_("invalid use of PROVIDE for dot symbol"));
1101 // The GNU linker permits assignments to dot outside of SECTIONS
1102 // clauses and treats them as occurring inside, so we don't
1103 // check in_sections_clause here.
1104 this->script_sections_
.add_dot_assignment(value
);
1108 // Add a reference to a symbol.
1111 Script_options::add_symbol_reference(const char* name
, size_t length
)
1113 if (length
!= 1 || name
[0] != '.')
1115 std::string
n(name
, length
);
1116 if (this->symbol_definitions_
.find(n
) == this->symbol_definitions_
.end())
1117 this->symbol_references_
.insert(n
);
1121 // Add an assertion.
1124 Script_options::add_assertion(Expression
* check
, const char* message
,
1127 if (this->script_sections_
.in_sections_clause())
1128 this->script_sections_
.add_assertion(check
, message
, messagelen
);
1131 Script_assertion
* p
= new Script_assertion(check
, message
, messagelen
);
1132 this->assertions_
.push_back(p
);
1136 // Create sections required by any linker scripts.
1139 Script_options::create_script_sections(Layout
* layout
)
1141 if (this->saw_sections_clause())
1142 this->script_sections_
.create_sections(layout
);
1145 // Add any symbols we are defining to the symbol table.
1148 Script_options::add_symbols_to_table(Symbol_table
* symtab
)
1150 for (Symbol_assignments::iterator p
= this->symbol_assignments_
.begin();
1151 p
!= this->symbol_assignments_
.end();
1153 (*p
)->add_to_table(symtab
);
1154 this->script_sections_
.add_symbols_to_table(symtab
);
1157 // Finalize symbol values. Also check assertions.
1160 Script_options::finalize_symbols(Symbol_table
* symtab
, const Layout
* layout
)
1162 // We finalize the symbols defined in SECTIONS first, because they
1163 // are the ones which may have changed. This way if symbol outside
1164 // SECTIONS are defined in terms of symbols inside SECTIONS, they
1165 // will get the right value.
1166 this->script_sections_
.finalize_symbols(symtab
, layout
);
1168 for (Symbol_assignments::iterator p
= this->symbol_assignments_
.begin();
1169 p
!= this->symbol_assignments_
.end();
1171 (*p
)->finalize(symtab
, layout
);
1173 for (Assertions::iterator p
= this->assertions_
.begin();
1174 p
!= this->assertions_
.end();
1176 (*p
)->check(symtab
, layout
);
1179 // Set section addresses. We set all the symbols which have absolute
1180 // values. Then we let the SECTIONS clause do its thing. This
1181 // returns the segment which holds the file header and segment
1185 Script_options::set_section_addresses(Symbol_table
* symtab
, Layout
* layout
)
1187 for (Symbol_assignments::iterator p
= this->symbol_assignments_
.begin();
1188 p
!= this->symbol_assignments_
.end();
1190 (*p
)->set_if_absolute(symtab
, layout
, false, 0);
1192 return this->script_sections_
.set_section_addresses(symtab
, layout
);
1195 // This class holds data passed through the parser to the lexer and to
1196 // the parser support functions. This avoids global variables. We
1197 // can't use global variables because we need not be called by a
1198 // singleton thread.
1200 class Parser_closure
1203 Parser_closure(const char* filename
,
1204 const Position_dependent_options
& posdep_options
,
1205 bool parsing_defsym
, bool in_group
, bool is_in_sysroot
,
1206 Command_line
* command_line
,
1207 Script_options
* script_options
,
1209 bool skip_on_incompatible_target
,
1210 Script_info
* script_info
)
1211 : filename_(filename
), posdep_options_(posdep_options
),
1212 parsing_defsym_(parsing_defsym
), in_group_(in_group
),
1213 is_in_sysroot_(is_in_sysroot
),
1214 skip_on_incompatible_target_(skip_on_incompatible_target
),
1215 found_incompatible_target_(false),
1216 command_line_(command_line
), script_options_(script_options
),
1217 version_script_info_(script_options
->version_script_info()),
1218 lex_(lex
), lineno_(0), charpos_(0), lex_mode_stack_(), inputs_(NULL
),
1219 script_info_(script_info
)
1221 // We start out processing C symbols in the default lex mode.
1222 this->language_stack_
.push_back(Version_script_info::LANGUAGE_C
);
1223 this->lex_mode_stack_
.push_back(lex
->mode());
1226 // Return the file name.
1229 { return this->filename_
; }
1231 // Return the position dependent options. The caller may modify
1233 Position_dependent_options
&
1234 position_dependent_options()
1235 { return this->posdep_options_
; }
1237 // Whether we are parsing a --defsym.
1239 parsing_defsym() const
1240 { return this->parsing_defsym_
; }
1242 // Return whether this script is being run in a group.
1245 { return this->in_group_
; }
1247 // Return whether this script was found using a directory in the
1250 is_in_sysroot() const
1251 { return this->is_in_sysroot_
; }
1253 // Whether to skip to the next file with the same name if we find an
1254 // incompatible target in an OUTPUT_FORMAT statement.
1256 skip_on_incompatible_target() const
1257 { return this->skip_on_incompatible_target_
; }
1259 // Stop skipping to the next file on an incompatible target. This
1260 // is called when we make some unrevocable change to the data
1263 clear_skip_on_incompatible_target()
1264 { this->skip_on_incompatible_target_
= false; }
1266 // Whether we found an incompatible target in an OUTPUT_FORMAT
1269 found_incompatible_target() const
1270 { return this->found_incompatible_target_
; }
1272 // Note that we found an incompatible target.
1274 set_found_incompatible_target()
1275 { this->found_incompatible_target_
= true; }
1277 // Returns the Command_line structure passed in at constructor time.
1278 // This value may be NULL. The caller may modify this, which modifies
1279 // the passed-in Command_line object (not a copy).
1282 { return this->command_line_
; }
1284 // Return the options which may be set by a script.
1287 { return this->script_options_
; }
1289 // Return the object in which version script information should be stored.
1290 Version_script_info
*
1292 { return this->version_script_info_
; }
1294 // Return the next token, and advance.
1298 const Token
* token
= this->lex_
->next_token();
1299 this->lineno_
= token
->lineno();
1300 this->charpos_
= token
->charpos();
1304 // Set a new lexer mode, pushing the current one.
1306 push_lex_mode(Lex::Mode mode
)
1308 this->lex_mode_stack_
.push_back(this->lex_
->mode());
1309 this->lex_
->set_mode(mode
);
1312 // Pop the lexer mode.
1316 gold_assert(!this->lex_mode_stack_
.empty());
1317 this->lex_
->set_mode(this->lex_mode_stack_
.back());
1318 this->lex_mode_stack_
.pop_back();
1321 // Return the current lexer mode.
1324 { return this->lex_mode_stack_
.back(); }
1326 // Return the line number of the last token.
1329 { return this->lineno_
; }
1331 // Return the character position in the line of the last token.
1334 { return this->charpos_
; }
1336 // Return the list of input files, creating it if necessary. This
1337 // is a space leak--we never free the INPUTS_ pointer.
1341 if (this->inputs_
== NULL
)
1342 this->inputs_
= new Input_arguments();
1343 return this->inputs_
;
1346 // Return whether we saw any input files.
1349 { return this->inputs_
!= NULL
&& !this->inputs_
->empty(); }
1351 // Return the current language being processed in a version script
1352 // (eg, "C++"). The empty string represents unmangled C names.
1353 Version_script_info::Language
1354 get_current_language() const
1355 { return this->language_stack_
.back(); }
1357 // Push a language onto the stack when entering an extern block.
1359 push_language(Version_script_info::Language lang
)
1360 { this->language_stack_
.push_back(lang
); }
1362 // Pop a language off of the stack when exiting an extern block.
1366 gold_assert(!this->language_stack_
.empty());
1367 this->language_stack_
.pop_back();
1370 // Return a pointer to the incremental info.
1373 { return this->script_info_
; }
1376 // The name of the file we are reading.
1377 const char* filename_
;
1378 // The position dependent options.
1379 Position_dependent_options posdep_options_
;
1380 // True if we are parsing a --defsym.
1381 bool parsing_defsym_
;
1382 // Whether we are currently in a --start-group/--end-group.
1384 // Whether the script was found in a sysrooted directory.
1385 bool is_in_sysroot_
;
1386 // If this is true, then if we find an OUTPUT_FORMAT with an
1387 // incompatible target, then we tell the parser to abort so that we
1388 // can search for the next file with the same name.
1389 bool skip_on_incompatible_target_
;
1390 // True if we found an OUTPUT_FORMAT with an incompatible target.
1391 bool found_incompatible_target_
;
1392 // May be NULL if the user chooses not to pass one in.
1393 Command_line
* command_line_
;
1394 // Options which may be set from any linker script.
1395 Script_options
* script_options_
;
1396 // Information parsed from a version script.
1397 Version_script_info
* version_script_info_
;
1400 // The line number of the last token returned by next_token.
1402 // The column number of the last token returned by next_token.
1404 // A stack of lexer modes.
1405 std::vector
<Lex::Mode
> lex_mode_stack_
;
1406 // A stack of which extern/language block we're inside. Can be C++,
1407 // java, or empty for C.
1408 std::vector
<Version_script_info::Language
> language_stack_
;
1409 // New input files found to add to the link.
1410 Input_arguments
* inputs_
;
1411 // Pointer to incremental linking info.
1412 Script_info
* script_info_
;
1415 // FILE was found as an argument on the command line. Try to read it
1416 // as a script. Return true if the file was handled.
1419 read_input_script(Workqueue
* workqueue
, Symbol_table
* symtab
, Layout
* layout
,
1420 Dirsearch
* dirsearch
, int dirindex
,
1421 Input_objects
* input_objects
, Mapfile
* mapfile
,
1422 Input_group
* input_group
,
1423 const Input_argument
* input_argument
,
1424 Input_file
* input_file
, Task_token
* next_blocker
,
1425 bool* used_next_blocker
)
1427 *used_next_blocker
= false;
1429 std::string input_string
;
1430 Lex::read_file(input_file
, &input_string
);
1432 Lex
lex(input_string
.c_str(), input_string
.length(), PARSING_LINKER_SCRIPT
);
1434 Script_info
* script_info
= NULL
;
1435 if (layout
->incremental_inputs() != NULL
)
1437 const std::string
& filename
= input_file
->filename();
1438 Timespec mtime
= input_file
->file().get_mtime();
1439 unsigned int arg_serial
= input_argument
->file().arg_serial();
1440 script_info
= new Script_info(filename
);
1441 layout
->incremental_inputs()->report_script(script_info
, arg_serial
,
1445 Parser_closure
closure(input_file
->filename().c_str(),
1446 input_argument
->file().options(),
1448 input_group
!= NULL
,
1449 input_file
->is_in_sysroot(),
1451 layout
->script_options(),
1453 input_file
->will_search_for(),
1456 bool old_saw_sections_clause
=
1457 layout
->script_options()->saw_sections_clause();
1459 if (yyparse(&closure
) != 0)
1461 if (closure
.found_incompatible_target())
1463 Read_symbols::incompatible_warning(input_argument
, input_file
);
1464 Read_symbols::requeue(workqueue
, input_objects
, symtab
, layout
,
1465 dirsearch
, dirindex
, mapfile
, input_argument
,
1466 input_group
, next_blocker
);
1472 if (!old_saw_sections_clause
1473 && layout
->script_options()->saw_sections_clause()
1474 && layout
->have_added_input_section())
1475 gold_error(_("%s: SECTIONS seen after other input files; try -T/--script"),
1476 input_file
->filename().c_str());
1478 if (!closure
.saw_inputs())
1481 Task_token
* this_blocker
= NULL
;
1482 for (Input_arguments::const_iterator p
= closure
.inputs()->begin();
1483 p
!= closure
.inputs()->end();
1487 if (p
+ 1 == closure
.inputs()->end())
1491 nb
= new Task_token(true);
1494 workqueue
->queue_soon(new Read_symbols(input_objects
, symtab
,
1495 layout
, dirsearch
, 0, mapfile
, &*p
,
1496 input_group
, NULL
, this_blocker
, nb
));
1500 *used_next_blocker
= true;
1505 // Helper function for read_version_script() and
1506 // read_commandline_script(). Processes the given file in the mode
1507 // indicated by first_token and lex_mode.
1510 read_script_file(const char* filename
, Command_line
* cmdline
,
1511 Script_options
* script_options
,
1512 int first_token
, Lex::Mode lex_mode
)
1514 // TODO: if filename is a relative filename, search for it manually
1515 // using "." + cmdline->options()->search_path() -- not dirsearch.
1516 Dirsearch dirsearch
;
1518 // The file locking code wants to record a Task, but we haven't
1519 // started the workqueue yet. This is only for debugging purposes,
1520 // so we invent a fake value.
1521 const Task
* task
= reinterpret_cast<const Task
*>(-1);
1523 // We don't want this file to be opened in binary mode.
1524 Position_dependent_options posdep
= cmdline
->position_dependent_options();
1525 if (posdep
.format_enum() == General_options::OBJECT_FORMAT_BINARY
)
1526 posdep
.set_format_enum(General_options::OBJECT_FORMAT_ELF
);
1527 Input_file_argument
input_argument(filename
,
1528 Input_file_argument::INPUT_FILE_TYPE_FILE
,
1530 Input_file
input_file(&input_argument
);
1532 if (!input_file
.open(dirsearch
, task
, &dummy
))
1535 std::string input_string
;
1536 Lex::read_file(&input_file
, &input_string
);
1538 Lex
lex(input_string
.c_str(), input_string
.length(), first_token
);
1539 lex
.set_mode(lex_mode
);
1541 Parser_closure
closure(filename
,
1542 cmdline
->position_dependent_options(),
1543 first_token
== Lex::DYNAMIC_LIST
,
1545 input_file
.is_in_sysroot(),
1551 if (yyparse(&closure
) != 0)
1553 input_file
.file().unlock(task
);
1557 input_file
.file().unlock(task
);
1559 gold_assert(!closure
.saw_inputs());
1564 // FILENAME was found as an argument to --script (-T).
1565 // Read it as a script, and execute its contents immediately.
1568 read_commandline_script(const char* filename
, Command_line
* cmdline
)
1570 return read_script_file(filename
, cmdline
, &cmdline
->script_options(),
1571 PARSING_LINKER_SCRIPT
, Lex::LINKER_SCRIPT
);
1574 // FILENAME was found as an argument to --version-script. Read it as
1575 // a version script, and store its contents in
1576 // cmdline->script_options()->version_script_info().
1579 read_version_script(const char* filename
, Command_line
* cmdline
)
1581 return read_script_file(filename
, cmdline
, &cmdline
->script_options(),
1582 PARSING_VERSION_SCRIPT
, Lex::VERSION_SCRIPT
);
1585 // FILENAME was found as an argument to --dynamic-list. Read it as a
1586 // list of symbols, and store its contents in DYNAMIC_LIST.
1589 read_dynamic_list(const char* filename
, Command_line
* cmdline
,
1590 Script_options
* dynamic_list
)
1592 return read_script_file(filename
, cmdline
, dynamic_list
,
1593 PARSING_DYNAMIC_LIST
, Lex::DYNAMIC_LIST
);
1596 // Implement the --defsym option on the command line. Return true if
1600 Script_options::define_symbol(const char* definition
)
1602 Lex
lex(definition
, strlen(definition
), PARSING_DEFSYM
);
1603 lex
.set_mode(Lex::EXPRESSION
);
1606 Position_dependent_options posdep_options
;
1608 Parser_closure
closure("command line", posdep_options
, true,
1609 false, false, NULL
, this, &lex
, false, NULL
);
1611 if (yyparse(&closure
) != 0)
1614 gold_assert(!closure
.saw_inputs());
1619 // Print the script to F for debugging.
1622 Script_options::print(FILE* f
) const
1624 fprintf(f
, "%s: Dumping linker script\n", program_name
);
1626 if (!this->entry_
.empty())
1627 fprintf(f
, "ENTRY(%s)\n", this->entry_
.c_str());
1629 for (Symbol_assignments::const_iterator p
=
1630 this->symbol_assignments_
.begin();
1631 p
!= this->symbol_assignments_
.end();
1635 for (Assertions::const_iterator p
= this->assertions_
.begin();
1636 p
!= this->assertions_
.end();
1640 this->script_sections_
.print(f
);
1642 this->version_script_info_
.print(f
);
1645 // Manage mapping from keywords to the codes expected by the bison
1646 // parser. We construct one global object for each lex mode with
1649 class Keyword_to_parsecode
1652 // The structure which maps keywords to parsecodes.
1653 struct Keyword_parsecode
1656 const char* keyword
;
1657 // Corresponding parsecode.
1661 Keyword_to_parsecode(const Keyword_parsecode
* keywords
,
1663 : keyword_parsecodes_(keywords
), keyword_count_(keyword_count
)
1666 // Return the parsecode corresponding KEYWORD, or 0 if it is not a
1669 keyword_to_parsecode(const char* keyword
, size_t len
) const;
1672 const Keyword_parsecode
* keyword_parsecodes_
;
1673 const int keyword_count_
;
1676 // Mapping from keyword string to keyword parsecode. This array must
1677 // be kept in sorted order. Parsecodes are looked up using bsearch.
1678 // This array must correspond to the list of parsecodes in yyscript.y.
1680 static const Keyword_to_parsecode::Keyword_parsecode
1681 script_keyword_parsecodes
[] =
1683 { "ABSOLUTE", ABSOLUTE
},
1685 { "ALIGN", ALIGN_K
},
1686 { "ALIGNOF", ALIGNOF
},
1687 { "ASSERT", ASSERT_K
},
1688 { "AS_NEEDED", AS_NEEDED
},
1693 { "CONSTANT", CONSTANT
},
1694 { "CONSTRUCTORS", CONSTRUCTORS
},
1696 { "CREATE_OBJECT_SYMBOLS", CREATE_OBJECT_SYMBOLS
},
1697 { "DATA_SEGMENT_ALIGN", DATA_SEGMENT_ALIGN
},
1698 { "DATA_SEGMENT_END", DATA_SEGMENT_END
},
1699 { "DATA_SEGMENT_RELRO_END", DATA_SEGMENT_RELRO_END
},
1700 { "DEFINED", DEFINED
},
1703 { "EXCLUDE_FILE", EXCLUDE_FILE
},
1704 { "EXTERN", EXTERN
},
1707 { "FORCE_COMMON_ALLOCATION", FORCE_COMMON_ALLOCATION
},
1710 { "INCLUDE", INCLUDE
},
1712 { "INHIBIT_COMMON_ALLOCATION", INHIBIT_COMMON_ALLOCATION
},
1715 { "LENGTH", LENGTH
},
1716 { "LOADADDR", LOADADDR
},
1720 { "MEMORY", MEMORY
},
1723 { "NOCROSSREFS", NOCROSSREFS
},
1724 { "NOFLOAT", NOFLOAT
},
1725 { "NOLOAD", NOLOAD
},
1726 { "ONLY_IF_RO", ONLY_IF_RO
},
1727 { "ONLY_IF_RW", ONLY_IF_RW
},
1728 { "OPTION", OPTION
},
1729 { "ORIGIN", ORIGIN
},
1730 { "OUTPUT", OUTPUT
},
1731 { "OUTPUT_ARCH", OUTPUT_ARCH
},
1732 { "OUTPUT_FORMAT", OUTPUT_FORMAT
},
1733 { "OVERLAY", OVERLAY
},
1735 { "PROVIDE", PROVIDE
},
1736 { "PROVIDE_HIDDEN", PROVIDE_HIDDEN
},
1738 { "SEARCH_DIR", SEARCH_DIR
},
1739 { "SECTIONS", SECTIONS
},
1740 { "SEGMENT_START", SEGMENT_START
},
1742 { "SIZEOF", SIZEOF
},
1743 { "SIZEOF_HEADERS", SIZEOF_HEADERS
},
1744 { "SORT", SORT_BY_NAME
},
1745 { "SORT_BY_ALIGNMENT", SORT_BY_ALIGNMENT
},
1746 { "SORT_BY_NAME", SORT_BY_NAME
},
1747 { "SPECIAL", SPECIAL
},
1749 { "STARTUP", STARTUP
},
1750 { "SUBALIGN", SUBALIGN
},
1751 { "SYSLIB", SYSLIB
},
1752 { "TARGET", TARGET_K
},
1753 { "TRUNCATE", TRUNCATE
},
1754 { "VERSION", VERSIONK
},
1755 { "global", GLOBAL
},
1761 { "sizeof_headers", SIZEOF_HEADERS
},
1764 static const Keyword_to_parsecode
1765 script_keywords(&script_keyword_parsecodes
[0],
1766 (sizeof(script_keyword_parsecodes
)
1767 / sizeof(script_keyword_parsecodes
[0])));
1769 static const Keyword_to_parsecode::Keyword_parsecode
1770 version_script_keyword_parsecodes
[] =
1772 { "extern", EXTERN
},
1773 { "global", GLOBAL
},
1777 static const Keyword_to_parsecode
1778 version_script_keywords(&version_script_keyword_parsecodes
[0],
1779 (sizeof(version_script_keyword_parsecodes
)
1780 / sizeof(version_script_keyword_parsecodes
[0])));
1782 static const Keyword_to_parsecode::Keyword_parsecode
1783 dynamic_list_keyword_parsecodes
[] =
1785 { "extern", EXTERN
},
1788 static const Keyword_to_parsecode
1789 dynamic_list_keywords(&dynamic_list_keyword_parsecodes
[0],
1790 (sizeof(dynamic_list_keyword_parsecodes
)
1791 / sizeof(dynamic_list_keyword_parsecodes
[0])));
1795 // Comparison function passed to bsearch.
1807 ktt_compare(const void* keyv
, const void* kttv
)
1809 const Ktt_key
* key
= static_cast<const Ktt_key
*>(keyv
);
1810 const Keyword_to_parsecode::Keyword_parsecode
* ktt
=
1811 static_cast<const Keyword_to_parsecode::Keyword_parsecode
*>(kttv
);
1812 int i
= strncmp(key
->str
, ktt
->keyword
, key
->len
);
1815 if (ktt
->keyword
[key
->len
] != '\0')
1820 } // End extern "C".
1823 Keyword_to_parsecode::keyword_to_parsecode(const char* keyword
,
1829 void* kttv
= bsearch(&key
,
1830 this->keyword_parsecodes_
,
1831 this->keyword_count_
,
1832 sizeof(this->keyword_parsecodes_
[0]),
1836 Keyword_parsecode
* ktt
= static_cast<Keyword_parsecode
*>(kttv
);
1837 return ktt
->parsecode
;
1840 // The following structs are used within the VersionInfo class as well
1841 // as in the bison helper functions. They store the information
1842 // parsed from the version script.
1844 // A single version expression.
1845 // For example, pattern="std::map*" and language="C++".
1846 struct Version_expression
1848 Version_expression(const std::string
& a_pattern
,
1849 Version_script_info::Language a_language
,
1851 : pattern(a_pattern
), language(a_language
), exact_match(a_exact_match
),
1852 was_matched_by_symbol(false)
1855 std::string pattern
;
1856 Version_script_info::Language language
;
1857 // If false, we use glob() to match pattern. If true, we use strcmp().
1859 // True if --no-undefined-version is in effect and we found this
1860 // version in get_symbol_version. We use mutable because this
1861 // struct is generally not modifiable after it has been created.
1862 mutable bool was_matched_by_symbol
;
1865 // A list of expressions.
1866 struct Version_expression_list
1868 std::vector
<struct Version_expression
> expressions
;
1871 // A list of which versions upon which another version depends.
1872 // Strings should be from the Stringpool.
1873 struct Version_dependency_list
1875 std::vector
<std::string
> dependencies
;
1878 // The total definition of a version. It includes the tag for the
1879 // version, its global and local expressions, and any dependencies.
1883 : tag(), global(NULL
), local(NULL
), dependencies(NULL
)
1887 const struct Version_expression_list
* global
;
1888 const struct Version_expression_list
* local
;
1889 const struct Version_dependency_list
* dependencies
;
1892 // Helper class that calls cplus_demangle when needed and takes care of freeing
1895 class Lazy_demangler
1898 Lazy_demangler(const char* symbol
, int options
)
1899 : symbol_(symbol
), options_(options
), demangled_(NULL
), did_demangle_(false)
1903 { free(this->demangled_
); }
1905 // Return the demangled name. The actual demangling happens on the first call,
1906 // and the result is later cached.
1911 // The symbol to demangle.
1912 const char* symbol_
;
1913 // Option flags to pass to cplus_demagle.
1915 // The cached demangled value, or NULL if demangling didn't happen yet or
1918 // Whether we already called cplus_demangle
1922 // Return the demangled name. The actual demangling happens on the first call,
1923 // and the result is later cached. Returns NULL if the symbol cannot be
1927 Lazy_demangler::get()
1929 if (!this->did_demangle_
)
1931 this->demangled_
= cplus_demangle(this->symbol_
, this->options_
);
1932 this->did_demangle_
= true;
1934 return this->demangled_
;
1937 // Class Version_script_info.
1939 Version_script_info::Version_script_info()
1940 : dependency_lists_(), expression_lists_(), version_trees_(), globs_(),
1941 default_version_(NULL
), default_is_global_(false), is_finalized_(false)
1943 for (int i
= 0; i
< LANGUAGE_COUNT
; ++i
)
1944 this->exact_
[i
] = NULL
;
1947 Version_script_info::~Version_script_info()
1951 // Forget all the known version script information.
1954 Version_script_info::clear()
1956 for (size_t k
= 0; k
< this->dependency_lists_
.size(); ++k
)
1957 delete this->dependency_lists_
[k
];
1958 this->dependency_lists_
.clear();
1959 for (size_t k
= 0; k
< this->version_trees_
.size(); ++k
)
1960 delete this->version_trees_
[k
];
1961 this->version_trees_
.clear();
1962 for (size_t k
= 0; k
< this->expression_lists_
.size(); ++k
)
1963 delete this->expression_lists_
[k
];
1964 this->expression_lists_
.clear();
1967 // Finalize the version script information.
1970 Version_script_info::finalize()
1972 if (!this->is_finalized_
)
1974 this->build_lookup_tables();
1975 this->is_finalized_
= true;
1979 // Return all the versions.
1981 std::vector
<std::string
>
1982 Version_script_info::get_versions() const
1984 std::vector
<std::string
> ret
;
1985 for (size_t j
= 0; j
< this->version_trees_
.size(); ++j
)
1986 if (!this->version_trees_
[j
]->tag
.empty())
1987 ret
.push_back(this->version_trees_
[j
]->tag
);
1991 // Return the dependencies of VERSION.
1993 std::vector
<std::string
>
1994 Version_script_info::get_dependencies(const char* version
) const
1996 std::vector
<std::string
> ret
;
1997 for (size_t j
= 0; j
< this->version_trees_
.size(); ++j
)
1998 if (this->version_trees_
[j
]->tag
== version
)
2000 const struct Version_dependency_list
* deps
=
2001 this->version_trees_
[j
]->dependencies
;
2003 for (size_t k
= 0; k
< deps
->dependencies
.size(); ++k
)
2004 ret
.push_back(deps
->dependencies
[k
]);
2010 // A version script essentially maps a symbol name to a version tag
2011 // and an indication of whether symbol is global or local within that
2012 // version tag. Each symbol maps to at most one version tag.
2013 // Unfortunately, in practice, version scripts are ambiguous, and list
2014 // symbols multiple times. Thus, we have to document the matching
2017 // This is a description of what the GNU linker does as of 2010-01-11.
2018 // It walks through the version tags in the order in which they appear
2019 // in the version script. For each tag, it first walks through the
2020 // global patterns for that tag, then the local patterns. When
2021 // looking at a single pattern, it first applies any language specific
2022 // demangling as specified for the pattern, and then matches the
2023 // resulting symbol name to the pattern. If it finds an exact match
2024 // for a literal pattern (a pattern enclosed in quotes or with no
2025 // wildcard characters), then that is the match that it uses. If
2026 // finds a match with a wildcard pattern, then it saves it and
2027 // continues searching. Wildcard patterns that are exactly "*" are
2028 // saved separately.
2030 // If no exact match with a literal pattern is ever found, then if a
2031 // wildcard match with a global pattern was found it is used,
2032 // otherwise if a wildcard match with a local pattern was found it is
2035 // This is the result:
2036 // * If there is an exact match, then we use the first tag in the
2037 // version script where it matches.
2038 // + If the exact match in that tag is global, it is used.
2039 // + Otherwise the exact match in that tag is local, and is used.
2040 // * Otherwise, if there is any match with a global wildcard pattern:
2041 // + If there is any match with a wildcard pattern which is not
2042 // "*", then we use the tag in which the *last* such pattern
2044 // + Otherwise, we matched "*". If there is no match with a local
2045 // wildcard pattern which is not "*", then we use the *last*
2046 // match with a global "*". Otherwise, continue.
2047 // * Otherwise, if there is any match with a local wildcard pattern:
2048 // + If there is any match with a wildcard pattern which is not
2049 // "*", then we use the tag in which the *last* such pattern
2051 // + Otherwise, we matched "*", and we use the tag in which the
2052 // *last* such match occurred.
2054 // There is an additional wrinkle. When the GNU linker finds a symbol
2055 // with a version defined in an object file due to a .symver
2056 // directive, it looks up that symbol name in that version tag. If it
2057 // finds it, it matches the symbol name against the patterns for that
2058 // version. If there is no match with a global pattern, but there is
2059 // a match with a local pattern, then the GNU linker marks the symbol
2062 // We want gold to be generally compatible, but we also want gold to
2063 // be fast. These are the rules that gold implements:
2064 // * If there is an exact match for the mangled name, we use it.
2065 // + If there is more than one exact match, we give a warning, and
2066 // we use the first tag in the script which matches.
2067 // + If a symbol has an exact match as both global and local for
2068 // the same version tag, we give an error.
2069 // * Otherwise, we look for an extern C++ or an extern Java exact
2070 // match. If we find an exact match, we use it.
2071 // + If there is more than one exact match, we give a warning, and
2072 // we use the first tag in the script which matches.
2073 // + If a symbol has an exact match as both global and local for
2074 // the same version tag, we give an error.
2075 // * Otherwise, we look through the wildcard patterns, ignoring "*"
2076 // patterns. We look through the version tags in reverse order.
2077 // For each version tag, we look through the global patterns and
2078 // then the local patterns. We use the first match we find (i.e.,
2079 // the last matching version tag in the file).
2080 // * Otherwise, we use the "*" pattern if there is one. We give an
2081 // error if there are multiple "*" patterns.
2083 // At least for now, gold does not look up the version tag for a
2084 // symbol version found in an object file to see if it should be
2085 // forced local. There are other ways to force a symbol to be local,
2086 // and I don't understand why this one is useful.
2088 // Build a set of fast lookup tables for a version script.
2091 Version_script_info::build_lookup_tables()
2093 size_t size
= this->version_trees_
.size();
2094 for (size_t j
= 0; j
< size
; ++j
)
2096 const Version_tree
* v
= this->version_trees_
[j
];
2097 this->build_expression_list_lookup(v
->local
, v
, false);
2098 this->build_expression_list_lookup(v
->global
, v
, true);
2102 // If a pattern has backlashes but no unquoted wildcard characters,
2103 // then we apply backslash unquoting and look for an exact match.
2104 // Otherwise we treat it as a wildcard pattern. This function returns
2105 // true for a wildcard pattern. Otherwise, it does backslash
2106 // unquoting on *PATTERN and returns false. If this returns true,
2107 // *PATTERN may have been partially unquoted.
2110 Version_script_info::unquote(std::string
* pattern
) const
2112 bool saw_backslash
= false;
2113 size_t len
= pattern
->length();
2115 for (size_t i
= 0; i
< len
; ++i
)
2118 saw_backslash
= false;
2121 switch ((*pattern
)[i
])
2123 case '?': case '[': case '*':
2126 saw_backslash
= true;
2134 (*pattern
)[j
] = (*pattern
)[i
];
2140 // Add an exact match for MATCH to *PE. The result of the match is
2144 Version_script_info::add_exact_match(const std::string
& match
,
2145 const Version_tree
* v
, bool is_global
,
2146 const Version_expression
* ve
,
2149 std::pair
<Exact::iterator
, bool> ins
=
2150 pe
->insert(std::make_pair(match
, Version_tree_match(v
, is_global
, ve
)));
2153 // This is the first time we have seen this match.
2157 Version_tree_match
& vtm(ins
.first
->second
);
2158 if (vtm
.real
->tag
!= v
->tag
)
2160 // This is an ambiguous match. We still return the
2161 // first version that we found in the script, but we
2162 // record the new version to issue a warning if we
2163 // wind up looking up this symbol.
2164 if (vtm
.ambiguous
== NULL
)
2167 else if (is_global
!= vtm
.is_global
)
2169 // We have a match for both the global and local entries for a
2170 // version tag. That's got to be wrong.
2171 gold_error(_("'%s' appears as both a global and a local symbol "
2172 "for version '%s' in script"),
2173 match
.c_str(), v
->tag
.c_str());
2177 // Build fast lookup information for EXPLIST and store it in LOOKUP.
2178 // All matches go to V, and IS_GLOBAL is true if they are global
2182 Version_script_info::build_expression_list_lookup(
2183 const Version_expression_list
* explist
,
2184 const Version_tree
* v
,
2187 if (explist
== NULL
)
2189 size_t size
= explist
->expressions
.size();
2190 for (size_t i
= 0; i
< size
; ++i
)
2192 const Version_expression
& exp(explist
->expressions
[i
]);
2194 if (exp
.pattern
.length() == 1 && exp
.pattern
[0] == '*')
2196 if (this->default_version_
!= NULL
2197 && this->default_version_
->tag
!= v
->tag
)
2198 gold_warning(_("wildcard match appears in both version '%s' "
2199 "and '%s' in script"),
2200 this->default_version_
->tag
.c_str(), v
->tag
.c_str());
2201 else if (this->default_version_
!= NULL
2202 && this->default_is_global_
!= is_global
)
2203 gold_error(_("wildcard match appears as both global and local "
2204 "in version '%s' in script"),
2206 this->default_version_
= v
;
2207 this->default_is_global_
= is_global
;
2211 std::string pattern
= exp
.pattern
;
2212 if (!exp
.exact_match
)
2214 if (this->unquote(&pattern
))
2216 this->globs_
.push_back(Glob(&exp
, v
, is_global
));
2221 if (this->exact_
[exp
.language
] == NULL
)
2222 this->exact_
[exp
.language
] = new Exact();
2223 this->add_exact_match(pattern
, v
, is_global
, &exp
,
2224 this->exact_
[exp
.language
]);
2228 // Return the name to match given a name, a language code, and two
2232 Version_script_info::get_name_to_match(const char* name
,
2234 Lazy_demangler
* cpp_demangler
,
2235 Lazy_demangler
* java_demangler
) const
2242 return cpp_demangler
->get();
2244 return java_demangler
->get();
2250 // Look up SYMBOL_NAME in the list of versions. Return true if the
2251 // symbol is found, false if not. If the symbol is found, then if
2252 // PVERSION is not NULL, set *PVERSION to the version tag, and if
2253 // P_IS_GLOBAL is not NULL, set *P_IS_GLOBAL according to whether the
2254 // symbol is global or not.
2257 Version_script_info::get_symbol_version(const char* symbol_name
,
2258 std::string
* pversion
,
2259 bool* p_is_global
) const
2261 Lazy_demangler
cpp_demangled_name(symbol_name
, DMGL_ANSI
| DMGL_PARAMS
);
2262 Lazy_demangler
java_demangled_name(symbol_name
,
2263 DMGL_ANSI
| DMGL_PARAMS
| DMGL_JAVA
);
2265 gold_assert(this->is_finalized_
);
2266 for (int i
= 0; i
< LANGUAGE_COUNT
; ++i
)
2268 Exact
* exact
= this->exact_
[i
];
2272 const char* name_to_match
= this->get_name_to_match(symbol_name
, i
,
2273 &cpp_demangled_name
,
2274 &java_demangled_name
);
2275 if (name_to_match
== NULL
)
2277 // If the name can not be demangled, the GNU linker goes
2278 // ahead and tries to match it anyhow. That does not
2279 // make sense to me and I have not implemented it.
2283 Exact::const_iterator pe
= exact
->find(name_to_match
);
2284 if (pe
!= exact
->end())
2286 const Version_tree_match
& vtm(pe
->second
);
2287 if (vtm
.ambiguous
!= NULL
)
2288 gold_warning(_("using '%s' as version for '%s' which is also "
2289 "named in version '%s' in script"),
2290 vtm
.real
->tag
.c_str(), name_to_match
,
2291 vtm
.ambiguous
->tag
.c_str());
2293 if (pversion
!= NULL
)
2294 *pversion
= vtm
.real
->tag
;
2295 if (p_is_global
!= NULL
)
2296 *p_is_global
= vtm
.is_global
;
2298 // If we are using --no-undefined-version, and this is a
2299 // global symbol, we have to record that we have found this
2300 // symbol, so that we don't warn about it. We have to do
2301 // this now, because otherwise we have no way to get from a
2302 // non-C language back to the demangled name that we
2304 if (p_is_global
!= NULL
&& vtm
.is_global
)
2305 vtm
.expression
->was_matched_by_symbol
= true;
2311 // Look through the glob patterns in reverse order.
2313 for (Globs::const_reverse_iterator p
= this->globs_
.rbegin();
2314 p
!= this->globs_
.rend();
2317 int language
= p
->expression
->language
;
2318 const char* name_to_match
= this->get_name_to_match(symbol_name
,
2320 &cpp_demangled_name
,
2321 &java_demangled_name
);
2322 if (name_to_match
== NULL
)
2325 if (fnmatch(p
->expression
->pattern
.c_str(), name_to_match
,
2328 if (pversion
!= NULL
)
2329 *pversion
= p
->version
->tag
;
2330 if (p_is_global
!= NULL
)
2331 *p_is_global
= p
->is_global
;
2336 // Finally, there may be a wildcard.
2337 if (this->default_version_
!= NULL
)
2339 if (pversion
!= NULL
)
2340 *pversion
= this->default_version_
->tag
;
2341 if (p_is_global
!= NULL
)
2342 *p_is_global
= this->default_is_global_
;
2349 // Give an error if any exact symbol names (not wildcards) appear in a
2350 // version script, but there is no such symbol.
2353 Version_script_info::check_unmatched_names(const Symbol_table
* symtab
) const
2355 for (size_t i
= 0; i
< this->version_trees_
.size(); ++i
)
2357 const Version_tree
* vt
= this->version_trees_
[i
];
2358 if (vt
->global
== NULL
)
2360 for (size_t j
= 0; j
< vt
->global
->expressions
.size(); ++j
)
2362 const Version_expression
& expression(vt
->global
->expressions
[j
]);
2364 // Ignore cases where we used the version because we saw a
2365 // symbol that we looked up. Note that
2366 // WAS_MATCHED_BY_SYMBOL will be true even if the symbol was
2367 // not a definition. That's OK as in that case we most
2368 // likely gave an undefined symbol error anyhow.
2369 if (expression
.was_matched_by_symbol
)
2372 // Just ignore names which are in languages other than C.
2373 // We have no way to look them up in the symbol table.
2374 if (expression
.language
!= LANGUAGE_C
)
2377 // Remove backslash quoting, and ignore wildcard patterns.
2378 std::string pattern
= expression
.pattern
;
2379 if (!expression
.exact_match
)
2381 if (this->unquote(&pattern
))
2385 if (symtab
->lookup(pattern
.c_str(), vt
->tag
.c_str()) == NULL
)
2386 gold_error(_("version script assignment of %s to symbol %s "
2387 "failed: symbol not defined"),
2388 vt
->tag
.c_str(), pattern
.c_str());
2393 struct Version_dependency_list
*
2394 Version_script_info::allocate_dependency_list()
2396 dependency_lists_
.push_back(new Version_dependency_list
);
2397 return dependency_lists_
.back();
2400 struct Version_expression_list
*
2401 Version_script_info::allocate_expression_list()
2403 expression_lists_
.push_back(new Version_expression_list
);
2404 return expression_lists_
.back();
2407 struct Version_tree
*
2408 Version_script_info::allocate_version_tree()
2410 version_trees_
.push_back(new Version_tree
);
2411 return version_trees_
.back();
2414 // Print for debugging.
2417 Version_script_info::print(FILE* f
) const
2422 fprintf(f
, "VERSION {");
2424 for (size_t i
= 0; i
< this->version_trees_
.size(); ++i
)
2426 const Version_tree
* vt
= this->version_trees_
[i
];
2428 if (vt
->tag
.empty())
2431 fprintf(f
, " %s {\n", vt
->tag
.c_str());
2433 if (vt
->global
!= NULL
)
2435 fprintf(f
, " global :\n");
2436 this->print_expression_list(f
, vt
->global
);
2439 if (vt
->local
!= NULL
)
2441 fprintf(f
, " local :\n");
2442 this->print_expression_list(f
, vt
->local
);
2446 if (vt
->dependencies
!= NULL
)
2448 const Version_dependency_list
* deps
= vt
->dependencies
;
2449 for (size_t j
= 0; j
< deps
->dependencies
.size(); ++j
)
2451 if (j
< deps
->dependencies
.size() - 1)
2453 fprintf(f
, " %s", deps
->dependencies
[j
].c_str());
2463 Version_script_info::print_expression_list(
2465 const Version_expression_list
* vel
) const
2467 Version_script_info::Language current_language
= LANGUAGE_C
;
2468 for (size_t i
= 0; i
< vel
->expressions
.size(); ++i
)
2470 const Version_expression
& ve(vel
->expressions
[i
]);
2472 if (ve
.language
!= current_language
)
2474 if (current_language
!= LANGUAGE_C
)
2476 switch (ve
.language
)
2481 fprintf(f
, " extern \"C++\" {\n");
2484 fprintf(f
, " extern \"Java\" {\n");
2489 current_language
= ve
.language
;
2493 if (current_language
!= LANGUAGE_C
)
2498 fprintf(f
, "%s", ve
.pattern
.c_str());
2505 if (current_language
!= LANGUAGE_C
)
2509 } // End namespace gold.
2511 // The remaining functions are extern "C", so it's clearer to not put
2512 // them in namespace gold.
2514 using namespace gold
;
2516 // This function is called by the bison parser to return the next
2520 yylex(YYSTYPE
* lvalp
, void* closurev
)
2522 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2523 const Token
* token
= closure
->next_token();
2524 switch (token
->classification())
2529 case Token::TOKEN_INVALID
:
2530 yyerror(closurev
, "invalid character");
2533 case Token::TOKEN_EOF
:
2536 case Token::TOKEN_STRING
:
2538 // This is either a keyword or a STRING.
2540 const char* str
= token
->string_value(&len
);
2542 switch (closure
->lex_mode())
2544 case Lex::LINKER_SCRIPT
:
2545 parsecode
= script_keywords
.keyword_to_parsecode(str
, len
);
2547 case Lex::VERSION_SCRIPT
:
2548 parsecode
= version_script_keywords
.keyword_to_parsecode(str
, len
);
2550 case Lex::DYNAMIC_LIST
:
2551 parsecode
= dynamic_list_keywords
.keyword_to_parsecode(str
, len
);
2558 lvalp
->string
.value
= str
;
2559 lvalp
->string
.length
= len
;
2563 case Token::TOKEN_QUOTED_STRING
:
2564 lvalp
->string
.value
= token
->string_value(&lvalp
->string
.length
);
2565 return QUOTED_STRING
;
2567 case Token::TOKEN_OPERATOR
:
2568 return token
->operator_value();
2570 case Token::TOKEN_INTEGER
:
2571 lvalp
->integer
= token
->integer_value();
2576 // This function is called by the bison parser to report an error.
2579 yyerror(void* closurev
, const char* message
)
2581 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2582 gold_error(_("%s:%d:%d: %s"), closure
->filename(), closure
->lineno(),
2583 closure
->charpos(), message
);
2586 // Called by the bison parser to add an external symbol to the link.
2589 script_add_extern(void* closurev
, const char* name
, size_t length
)
2591 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2592 closure
->script_options()->add_symbol_reference(name
, length
);
2595 // Called by the bison parser to add a file to the link.
2598 script_add_file(void* closurev
, const char* name
, size_t length
)
2600 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2602 // If this is an absolute path, and we found the script in the
2603 // sysroot, then we want to prepend the sysroot to the file name.
2604 // For example, this is how we handle a cross link to the x86_64
2605 // libc.so, which refers to /lib/libc.so.6.
2606 std::string
name_string(name
, length
);
2607 const char* extra_search_path
= ".";
2608 std::string script_directory
;
2609 if (IS_ABSOLUTE_PATH(name_string
.c_str()))
2611 if (closure
->is_in_sysroot())
2613 const std::string
& sysroot(parameters
->options().sysroot());
2614 gold_assert(!sysroot
.empty());
2615 name_string
= sysroot
+ name_string
;
2620 // In addition to checking the normal library search path, we
2621 // also want to check in the script-directory.
2622 const char* slash
= strrchr(closure
->filename(), '/');
2625 script_directory
.assign(closure
->filename(),
2626 slash
- closure
->filename() + 1);
2627 extra_search_path
= script_directory
.c_str();
2631 Input_file_argument
file(name_string
.c_str(),
2632 Input_file_argument::INPUT_FILE_TYPE_FILE
,
2633 extra_search_path
, false,
2634 closure
->position_dependent_options());
2635 Input_argument
& arg
= closure
->inputs()->add_file(file
);
2636 arg
.set_script_info(closure
->script_info());
2639 // Called by the bison parser to add a library to the link.
2642 script_add_library(void* closurev
, const char* name
, size_t length
)
2644 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2645 std::string
name_string(name
, length
);
2647 if (name_string
[0] != 'l')
2648 gold_error(_("library name must be prefixed with -l"));
2650 Input_file_argument
file(name_string
.c_str() + 1,
2651 Input_file_argument::INPUT_FILE_TYPE_LIBRARY
,
2653 closure
->position_dependent_options());
2654 Input_argument
& arg
= closure
->inputs()->add_file(file
);
2655 arg
.set_script_info(closure
->script_info());
2658 // Called by the bison parser to start a group. If we are already in
2659 // a group, that means that this script was invoked within a
2660 // --start-group --end-group sequence on the command line, or that
2661 // this script was found in a GROUP of another script. In that case,
2662 // we simply continue the existing group, rather than starting a new
2663 // one. It is possible to construct a case in which this will do
2664 // something other than what would happen if we did a recursive group,
2665 // but it's hard to imagine why the different behaviour would be
2666 // useful for a real program. Avoiding recursive groups is simpler
2667 // and more efficient.
2670 script_start_group(void* closurev
)
2672 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2673 if (!closure
->in_group())
2674 closure
->inputs()->start_group();
2677 // Called by the bison parser at the end of a group.
2680 script_end_group(void* closurev
)
2682 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2683 if (!closure
->in_group())
2684 closure
->inputs()->end_group();
2687 // Called by the bison parser to start an AS_NEEDED list.
2690 script_start_as_needed(void* closurev
)
2692 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2693 closure
->position_dependent_options().set_as_needed(true);
2696 // Called by the bison parser at the end of an AS_NEEDED list.
2699 script_end_as_needed(void* closurev
)
2701 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2702 closure
->position_dependent_options().set_as_needed(false);
2705 // Called by the bison parser to set the entry symbol.
2708 script_set_entry(void* closurev
, const char* entry
, size_t length
)
2710 // We'll parse this exactly the same as --entry=ENTRY on the commandline
2711 // TODO(csilvers): FIXME -- call set_entry directly.
2712 std::string
arg("--entry=");
2713 arg
.append(entry
, length
);
2714 script_parse_option(closurev
, arg
.c_str(), arg
.size());
2717 // Called by the bison parser to set whether to define common symbols.
2720 script_set_common_allocation(void* closurev
, int set
)
2722 const char* arg
= set
!= 0 ? "--define-common" : "--no-define-common";
2723 script_parse_option(closurev
, arg
, strlen(arg
));
2726 // Called by the bison parser to refer to a symbol.
2728 extern "C" Expression
*
2729 script_symbol(void* closurev
, const char* name
, size_t length
)
2731 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2732 if (length
!= 1 || name
[0] != '.')
2733 closure
->script_options()->add_symbol_reference(name
, length
);
2734 return script_exp_string(name
, length
);
2737 // Called by the bison parser to define a symbol.
2740 script_set_symbol(void* closurev
, const char* name
, size_t length
,
2741 Expression
* value
, int providei
, int hiddeni
)
2743 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2744 const bool provide
= providei
!= 0;
2745 const bool hidden
= hiddeni
!= 0;
2746 closure
->script_options()->add_symbol_assignment(name
, length
,
2747 closure
->parsing_defsym(),
2748 value
, provide
, hidden
);
2749 closure
->clear_skip_on_incompatible_target();
2752 // Called by the bison parser to add an assertion.
2755 script_add_assertion(void* closurev
, Expression
* check
, const char* message
,
2758 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2759 closure
->script_options()->add_assertion(check
, message
, messagelen
);
2760 closure
->clear_skip_on_incompatible_target();
2763 // Called by the bison parser to parse an OPTION.
2766 script_parse_option(void* closurev
, const char* option
, size_t length
)
2768 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2769 // We treat the option as a single command-line option, even if
2770 // it has internal whitespace.
2771 if (closure
->command_line() == NULL
)
2773 // There are some options that we could handle here--e.g.,
2774 // -lLIBRARY. Should we bother?
2775 gold_warning(_("%s:%d:%d: ignoring command OPTION; OPTION is only valid"
2776 " for scripts specified via -T/--script"),
2777 closure
->filename(), closure
->lineno(), closure
->charpos());
2781 bool past_a_double_dash_option
= false;
2782 const char* mutable_option
= strndup(option
, length
);
2783 gold_assert(mutable_option
!= NULL
);
2784 closure
->command_line()->process_one_option(1, &mutable_option
, 0,
2785 &past_a_double_dash_option
);
2786 // The General_options class will quite possibly store a pointer
2787 // into mutable_option, so we can't free it. In cases the class
2788 // does not store such a pointer, this is a memory leak. Alas. :(
2790 closure
->clear_skip_on_incompatible_target();
2793 // Called by the bison parser to handle OUTPUT_FORMAT. OUTPUT_FORMAT
2794 // takes either one or three arguments. In the three argument case,
2795 // the format depends on the endianness option, which we don't
2796 // currently support (FIXME). If we see an OUTPUT_FORMAT for the
2797 // wrong format, then we want to search for a new file. Returning 0
2798 // here will cause the parser to immediately abort.
2801 script_check_output_format(void* closurev
,
2802 const char* default_name
, size_t default_length
,
2803 const char*, size_t, const char*, size_t)
2805 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2806 std::string
name(default_name
, default_length
);
2807 Target
* target
= select_target_by_name(name
.c_str());
2808 if (target
== NULL
|| !parameters
->is_compatible_target(target
))
2810 if (closure
->skip_on_incompatible_target())
2812 closure
->set_found_incompatible_target();
2815 // FIXME: Should we warn about the unknown target?
2820 // Called by the bison parser to handle TARGET.
2823 script_set_target(void* closurev
, const char* target
, size_t len
)
2825 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2826 std::string
s(target
, len
);
2827 General_options::Object_format format_enum
;
2828 format_enum
= General_options::string_to_object_format(s
.c_str());
2829 closure
->position_dependent_options().set_format_enum(format_enum
);
2832 // Called by the bison parser to handle SEARCH_DIR. This is handled
2833 // exactly like a -L option.
2836 script_add_search_dir(void* closurev
, const char* option
, size_t length
)
2838 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2839 if (closure
->command_line() == NULL
)
2840 gold_warning(_("%s:%d:%d: ignoring SEARCH_DIR; SEARCH_DIR is only valid"
2841 " for scripts specified via -T/--script"),
2842 closure
->filename(), closure
->lineno(), closure
->charpos());
2843 else if (!closure
->command_line()->options().nostdlib())
2845 std::string s
= "-L" + std::string(option
, length
);
2846 script_parse_option(closurev
, s
.c_str(), s
.size());
2850 /* Called by the bison parser to push the lexer into expression
2854 script_push_lex_into_expression_mode(void* closurev
)
2856 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2857 closure
->push_lex_mode(Lex::EXPRESSION
);
2860 /* Called by the bison parser to push the lexer into version
2864 script_push_lex_into_version_mode(void* closurev
)
2866 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2867 if (closure
->version_script()->is_finalized())
2868 gold_error(_("%s:%d:%d: invalid use of VERSION in input file"),
2869 closure
->filename(), closure
->lineno(), closure
->charpos());
2870 closure
->push_lex_mode(Lex::VERSION_SCRIPT
);
2873 /* Called by the bison parser to pop the lexer mode. */
2876 script_pop_lex_mode(void* closurev
)
2878 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2879 closure
->pop_lex_mode();
2882 // Register an entire version node. For example:
2888 // - tag is "GLIBC_2.1"
2889 // - tree contains the information "global: foo"
2890 // - deps contains "GLIBC_2.0"
2893 script_register_vers_node(void*,
2896 struct Version_tree
* tree
,
2897 struct Version_dependency_list
* deps
)
2899 gold_assert(tree
!= NULL
);
2900 tree
->dependencies
= deps
;
2902 tree
->tag
= std::string(tag
, taglen
);
2905 // Add a dependencies to the list of existing dependencies, if any,
2906 // and return the expanded list.
2908 extern "C" struct Version_dependency_list
*
2909 script_add_vers_depend(void* closurev
,
2910 struct Version_dependency_list
* all_deps
,
2911 const char* depend_to_add
, int deplen
)
2913 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2914 if (all_deps
== NULL
)
2915 all_deps
= closure
->version_script()->allocate_dependency_list();
2916 all_deps
->dependencies
.push_back(std::string(depend_to_add
, deplen
));
2920 // Add a pattern expression to an existing list of expressions, if any.
2922 extern "C" struct Version_expression_list
*
2923 script_new_vers_pattern(void* closurev
,
2924 struct Version_expression_list
* expressions
,
2925 const char* pattern
, int patlen
, int exact_match
)
2927 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2928 if (expressions
== NULL
)
2929 expressions
= closure
->version_script()->allocate_expression_list();
2930 expressions
->expressions
.push_back(
2931 Version_expression(std::string(pattern
, patlen
),
2932 closure
->get_current_language(),
2933 static_cast<bool>(exact_match
)));
2937 // Attaches b to the end of a, and clears b. So a = a + b and b = {}.
2939 extern "C" struct Version_expression_list
*
2940 script_merge_expressions(struct Version_expression_list
* a
,
2941 struct Version_expression_list
* b
)
2943 a
->expressions
.insert(a
->expressions
.end(),
2944 b
->expressions
.begin(), b
->expressions
.end());
2945 // We could delete b and remove it from expressions_lists_, but
2946 // that's a lot of work. This works just as well.
2947 b
->expressions
.clear();
2951 // Combine the global and local expressions into a a Version_tree.
2953 extern "C" struct Version_tree
*
2954 script_new_vers_node(void* closurev
,
2955 struct Version_expression_list
* global
,
2956 struct Version_expression_list
* local
)
2958 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2959 Version_tree
* tree
= closure
->version_script()->allocate_version_tree();
2960 tree
->global
= global
;
2961 tree
->local
= local
;
2965 // Handle a transition in language, such as at the
2966 // start or end of 'extern "C++"'
2969 version_script_push_lang(void* closurev
, const char* lang
, int langlen
)
2971 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2972 std::string
language(lang
, langlen
);
2973 Version_script_info::Language code
;
2974 if (language
.empty() || language
== "C")
2975 code
= Version_script_info::LANGUAGE_C
;
2976 else if (language
== "C++")
2977 code
= Version_script_info::LANGUAGE_CXX
;
2978 else if (language
== "Java")
2979 code
= Version_script_info::LANGUAGE_JAVA
;
2982 char* buf
= new char[langlen
+ 100];
2983 snprintf(buf
, langlen
+ 100,
2984 _("unrecognized version script language '%s'"),
2986 yyerror(closurev
, buf
);
2988 code
= Version_script_info::LANGUAGE_C
;
2990 closure
->push_language(code
);
2994 version_script_pop_lang(void* closurev
)
2996 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2997 closure
->pop_language();
3000 // Called by the bison parser to start a SECTIONS clause.
3003 script_start_sections(void* closurev
)
3005 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
3006 closure
->script_options()->script_sections()->start_sections();
3007 closure
->clear_skip_on_incompatible_target();
3010 // Called by the bison parser to finish a SECTIONS clause.
3013 script_finish_sections(void* closurev
)
3015 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
3016 closure
->script_options()->script_sections()->finish_sections();
3019 // Start processing entries for an output section.
3022 script_start_output_section(void* closurev
, const char* name
, size_t namelen
,
3023 const struct Parser_output_section_header
* header
)
3025 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
3026 closure
->script_options()->script_sections()->start_output_section(name
,
3031 // Finish processing entries for an output section.
3034 script_finish_output_section(void* closurev
,
3035 const struct Parser_output_section_trailer
* trail
)
3037 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
3038 closure
->script_options()->script_sections()->finish_output_section(trail
);
3041 // Add a data item (e.g., "WORD (0)") to the current output section.
3044 script_add_data(void* closurev
, int data_token
, Expression
* val
)
3046 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
3048 bool is_signed
= true;
3070 closure
->script_options()->script_sections()->add_data(size
, is_signed
, val
);
3073 // Add a clause setting the fill value to the current output section.
3076 script_add_fill(void* closurev
, Expression
* val
)
3078 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
3079 closure
->script_options()->script_sections()->add_fill(val
);
3082 // Add a new input section specification to the current output
3086 script_add_input_section(void* closurev
,
3087 const struct Input_section_spec
* spec
,
3090 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
3091 bool keep
= keepi
!= 0;
3092 closure
->script_options()->script_sections()->add_input_section(spec
, keep
);
3095 // When we see DATA_SEGMENT_ALIGN we record that following output
3096 // sections may be relro.
3099 script_data_segment_align(void* closurev
)
3101 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
3102 if (!closure
->script_options()->saw_sections_clause())
3103 gold_error(_("%s:%d:%d: DATA_SEGMENT_ALIGN not in SECTIONS clause"),
3104 closure
->filename(), closure
->lineno(), closure
->charpos());
3106 closure
->script_options()->script_sections()->data_segment_align();
3109 // When we see DATA_SEGMENT_RELRO_END we know that all output sections
3110 // since DATA_SEGMENT_ALIGN should be relro.
3113 script_data_segment_relro_end(void* closurev
)
3115 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
3116 if (!closure
->script_options()->saw_sections_clause())
3117 gold_error(_("%s:%d:%d: DATA_SEGMENT_ALIGN not in SECTIONS clause"),
3118 closure
->filename(), closure
->lineno(), closure
->charpos());
3120 closure
->script_options()->script_sections()->data_segment_relro_end();
3123 // Create a new list of string/sort pairs.
3125 extern "C" String_sort_list_ptr
3126 script_new_string_sort_list(const struct Wildcard_section
* string_sort
)
3128 return new String_sort_list(1, *string_sort
);
3131 // Add an entry to a list of string/sort pairs. The way the parser
3132 // works permits us to simply modify the first parameter, rather than
3135 extern "C" String_sort_list_ptr
3136 script_string_sort_list_add(String_sort_list_ptr pv
,
3137 const struct Wildcard_section
* string_sort
)
3140 return script_new_string_sort_list(string_sort
);
3143 pv
->push_back(*string_sort
);
3148 // Create a new list of strings.
3150 extern "C" String_list_ptr
3151 script_new_string_list(const char* str
, size_t len
)
3153 return new String_list(1, std::string(str
, len
));
3156 // Add an element to a list of strings. The way the parser works
3157 // permits us to simply modify the first parameter, rather than copy
3160 extern "C" String_list_ptr
3161 script_string_list_push_back(String_list_ptr pv
, const char* str
, size_t len
)
3164 return script_new_string_list(str
, len
);
3167 pv
->push_back(std::string(str
, len
));
3172 // Concatenate two string lists. Either or both may be NULL. The way
3173 // the parser works permits us to modify the parameters, rather than
3176 extern "C" String_list_ptr
3177 script_string_list_append(String_list_ptr pv1
, String_list_ptr pv2
)
3183 pv1
->insert(pv1
->end(), pv2
->begin(), pv2
->end());
3187 // Add a new program header.
3190 script_add_phdr(void* closurev
, const char* name
, size_t namelen
,
3191 unsigned int type
, const Phdr_info
* info
)
3193 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
3194 bool includes_filehdr
= info
->includes_filehdr
!= 0;
3195 bool includes_phdrs
= info
->includes_phdrs
!= 0;
3196 bool is_flags_valid
= info
->is_flags_valid
!= 0;
3197 Script_sections
* ss
= closure
->script_options()->script_sections();
3198 ss
->add_phdr(name
, namelen
, type
, includes_filehdr
, includes_phdrs
,
3199 is_flags_valid
, info
->flags
, info
->load_address
);
3200 closure
->clear_skip_on_incompatible_target();
3203 // Convert a program header string to a type.
3205 #define PHDR_TYPE(NAME) { #NAME, sizeof(#NAME) - 1, elfcpp::NAME }
3212 } phdr_type_names
[] =
3216 PHDR_TYPE(PT_DYNAMIC
),
3217 PHDR_TYPE(PT_INTERP
),
3219 PHDR_TYPE(PT_SHLIB
),
3222 PHDR_TYPE(PT_GNU_EH_FRAME
),
3223 PHDR_TYPE(PT_GNU_STACK
),
3224 PHDR_TYPE(PT_GNU_RELRO
)
3227 extern "C" unsigned int
3228 script_phdr_string_to_type(void* closurev
, const char* name
, size_t namelen
)
3230 for (unsigned int i
= 0;
3231 i
< sizeof(phdr_type_names
) / sizeof(phdr_type_names
[0]);
3233 if (namelen
== phdr_type_names
[i
].namelen
3234 && strncmp(name
, phdr_type_names
[i
].name
, namelen
) == 0)
3235 return phdr_type_names
[i
].val
;
3236 yyerror(closurev
, _("unknown PHDR type (try integer)"));
3237 return elfcpp::PT_NULL
;
3241 script_saw_segment_start_expression(void* closurev
)
3243 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
3244 Script_sections
* ss
= closure
->script_options()->script_sections();
3245 ss
->set_saw_segment_start_expression(true);
3249 script_set_section_region(void* closurev
, const char* name
, size_t namelen
,
3252 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
3253 if (!closure
->script_options()->saw_sections_clause())
3255 gold_error(_("%s:%d:%d: MEMORY region '%.*s' referred to outside of "
3257 closure
->filename(), closure
->lineno(), closure
->charpos(),
3258 static_cast<int>(namelen
), name
);
3262 Script_sections
* ss
= closure
->script_options()->script_sections();
3263 Memory_region
* mr
= ss
->find_memory_region(name
, namelen
);
3266 gold_error(_("%s:%d:%d: MEMORY region '%.*s' not declared"),
3267 closure
->filename(), closure
->lineno(), closure
->charpos(),
3268 static_cast<int>(namelen
), name
);
3272 ss
->set_memory_region(mr
, set_vma
);
3276 script_add_memory(void* closurev
, const char* name
, size_t namelen
,
3277 unsigned int attrs
, Expression
* origin
, Expression
* length
)
3279 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
3280 Script_sections
* ss
= closure
->script_options()->script_sections();
3281 ss
->add_memory_region(name
, namelen
, attrs
, origin
, length
);
3284 extern "C" unsigned int
3285 script_parse_memory_attr(void* closurev
, const char* attrs
, size_t attrlen
,
3295 attributes
|= MEM_READABLE
; break;
3298 attributes
|= MEM_READABLE
| MEM_WRITEABLE
; break;
3301 attributes
|= MEM_EXECUTABLE
; break;
3304 attributes
|= MEM_ALLOCATABLE
; break;
3309 attributes
|= MEM_INITIALIZED
; break;
3311 yyerror(closurev
, _("unknown MEMORY attribute"));
3315 attributes
= (~ attributes
) & MEM_ATTR_MASK
;
3321 script_include_directive(void* closurev
, const char*, size_t)
3323 // FIXME: Implement ?
3324 yyerror (closurev
, _("GOLD does not currently support INCLUDE directives"));
3327 // Functions for memory regions.
3329 extern "C" Expression
*
3330 script_exp_function_origin(void* closurev
, const char* name
, size_t namelen
)
3332 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
3333 Script_sections
* ss
= closure
->script_options()->script_sections();
3334 Expression
* origin
= ss
->find_memory_region_origin(name
, namelen
);
3338 gold_error(_("undefined memory region '%s' referenced "
3339 "in ORIGIN expression"),
3341 // Create a dummy expression to prevent crashes later on.
3342 origin
= script_exp_integer(0);
3348 extern "C" Expression
*
3349 script_exp_function_length(void* closurev
, const char* name
, size_t namelen
)
3351 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
3352 Script_sections
* ss
= closure
->script_options()->script_sections();
3353 Expression
* length
= ss
->find_memory_region_length(name
, namelen
);
3357 gold_error(_("undefined memory region '%s' referenced "
3358 "in LENGTH expression"),
3360 // Create a dummy expression to prevent crashes later on.
3361 length
= script_exp_integer(0);