1 // script.cc -- handle linker scripts for gold.
3 // Copyright 2006, 2007, 2008 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')
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(),
884 true); // force_override
887 // Finalize a symbol value.
890 Symbol_assignment::finalize(Symbol_table
* symtab
, const Layout
* layout
)
892 this->finalize_maybe_dot(symtab
, layout
, false, 0, NULL
);
895 // Finalize a symbol value which can refer to the dot symbol.
898 Symbol_assignment::finalize_with_dot(Symbol_table
* symtab
,
899 const Layout
* layout
,
901 Output_section
* dot_section
)
903 this->finalize_maybe_dot(symtab
, layout
, true, dot_value
, dot_section
);
906 // Finalize a symbol value, internal version.
909 Symbol_assignment::finalize_maybe_dot(Symbol_table
* symtab
,
910 const Layout
* layout
,
911 bool is_dot_available
,
913 Output_section
* dot_section
)
915 // If we were only supposed to provide this symbol, the sym_ field
916 // will be NULL if the symbol was not referenced.
917 if (this->sym_
== NULL
)
919 gold_assert(this->provide_
);
923 if (parameters
->target().get_size() == 32)
925 #if defined(HAVE_TARGET_32_LITTLE) || defined(HAVE_TARGET_32_BIG)
926 this->sized_finalize
<32>(symtab
, layout
, is_dot_available
, dot_value
,
932 else if (parameters
->target().get_size() == 64)
934 #if defined(HAVE_TARGET_64_LITTLE) || defined(HAVE_TARGET_64_BIG)
935 this->sized_finalize
<64>(symtab
, layout
, is_dot_available
, dot_value
,
947 Symbol_assignment::sized_finalize(Symbol_table
* symtab
, const Layout
* layout
,
948 bool is_dot_available
, uint64_t dot_value
,
949 Output_section
* dot_section
)
951 Output_section
* section
;
952 uint64_t final_val
= this->val_
->eval_maybe_dot(symtab
, layout
, true,
954 dot_value
, dot_section
,
956 Sized_symbol
<size
>* ssym
= symtab
->get_sized_symbol
<size
>(this->sym_
);
957 ssym
->set_value(final_val
);
959 ssym
->set_output_section(section
);
962 // Set the symbol value if the expression yields an absolute value.
965 Symbol_assignment::set_if_absolute(Symbol_table
* symtab
, const Layout
* layout
,
966 bool is_dot_available
, uint64_t dot_value
)
968 if (this->sym_
== NULL
)
971 Output_section
* val_section
;
972 uint64_t val
= this->val_
->eval_maybe_dot(symtab
, layout
, false,
973 is_dot_available
, dot_value
,
975 if (val_section
!= NULL
)
978 if (parameters
->target().get_size() == 32)
980 #if defined(HAVE_TARGET_32_LITTLE) || defined(HAVE_TARGET_32_BIG)
981 Sized_symbol
<32>* ssym
= symtab
->get_sized_symbol
<32>(this->sym_
);
982 ssym
->set_value(val
);
987 else if (parameters
->target().get_size() == 64)
989 #if defined(HAVE_TARGET_64_LITTLE) || defined(HAVE_TARGET_64_BIG)
990 Sized_symbol
<64>* ssym
= symtab
->get_sized_symbol
<64>(this->sym_
);
991 ssym
->set_value(val
);
1000 // Print for debugging.
1003 Symbol_assignment::print(FILE* f
) const
1005 if (this->provide_
&& this->hidden_
)
1006 fprintf(f
, "PROVIDE_HIDDEN(");
1007 else if (this->provide_
)
1008 fprintf(f
, "PROVIDE(");
1009 else if (this->hidden_
)
1012 fprintf(f
, "%s = ", this->name_
.c_str());
1013 this->val_
->print(f
);
1015 if (this->provide_
|| this->hidden_
)
1021 // Class Script_assertion.
1023 // Check the assertion.
1026 Script_assertion::check(const Symbol_table
* symtab
, const Layout
* layout
)
1028 if (!this->check_
->eval(symtab
, layout
, true))
1029 gold_error("%s", this->message_
.c_str());
1032 // Print for debugging.
1035 Script_assertion::print(FILE* f
) const
1037 fprintf(f
, "ASSERT(");
1038 this->check_
->print(f
);
1039 fprintf(f
, ", \"%s\")\n", this->message_
.c_str());
1042 // Class Script_options.
1044 Script_options::Script_options()
1045 : entry_(), symbol_assignments_(), version_script_info_(),
1050 // Add a symbol to be defined.
1053 Script_options::add_symbol_assignment(const char* name
, size_t length
,
1054 Expression
* value
, bool provide
,
1057 if (length
!= 1 || name
[0] != '.')
1059 if (this->script_sections_
.in_sections_clause())
1060 this->script_sections_
.add_symbol_assignment(name
, length
, value
,
1064 Symbol_assignment
* p
= new Symbol_assignment(name
, length
, value
,
1066 this->symbol_assignments_
.push_back(p
);
1071 if (provide
|| hidden
)
1072 gold_error(_("invalid use of PROVIDE for dot symbol"));
1073 if (!this->script_sections_
.in_sections_clause())
1074 gold_error(_("invalid assignment to dot outside of SECTIONS"));
1076 this->script_sections_
.add_dot_assignment(value
);
1080 // Add an assertion.
1083 Script_options::add_assertion(Expression
* check
, const char* message
,
1086 if (this->script_sections_
.in_sections_clause())
1087 this->script_sections_
.add_assertion(check
, message
, messagelen
);
1090 Script_assertion
* p
= new Script_assertion(check
, message
, messagelen
);
1091 this->assertions_
.push_back(p
);
1095 // Create sections required by any linker scripts.
1098 Script_options::create_script_sections(Layout
* layout
)
1100 if (this->saw_sections_clause())
1101 this->script_sections_
.create_sections(layout
);
1104 // Add any symbols we are defining to the symbol table.
1107 Script_options::add_symbols_to_table(Symbol_table
* symtab
)
1109 for (Symbol_assignments::iterator p
= this->symbol_assignments_
.begin();
1110 p
!= this->symbol_assignments_
.end();
1112 (*p
)->add_to_table(symtab
);
1113 this->script_sections_
.add_symbols_to_table(symtab
);
1116 // Finalize symbol values. Also check assertions.
1119 Script_options::finalize_symbols(Symbol_table
* symtab
, const Layout
* layout
)
1121 // We finalize the symbols defined in SECTIONS first, because they
1122 // are the ones which may have changed. This way if symbol outside
1123 // SECTIONS are defined in terms of symbols inside SECTIONS, they
1124 // will get the right value.
1125 this->script_sections_
.finalize_symbols(symtab
, layout
);
1127 for (Symbol_assignments::iterator p
= this->symbol_assignments_
.begin();
1128 p
!= this->symbol_assignments_
.end();
1130 (*p
)->finalize(symtab
, layout
);
1132 for (Assertions::iterator p
= this->assertions_
.begin();
1133 p
!= this->assertions_
.end();
1135 (*p
)->check(symtab
, layout
);
1138 // Set section addresses. We set all the symbols which have absolute
1139 // values. Then we let the SECTIONS clause do its thing. This
1140 // returns the segment which holds the file header and segment
1144 Script_options::set_section_addresses(Symbol_table
* symtab
, Layout
* layout
)
1146 for (Symbol_assignments::iterator p
= this->symbol_assignments_
.begin();
1147 p
!= this->symbol_assignments_
.end();
1149 (*p
)->set_if_absolute(symtab
, layout
, false, 0);
1151 return this->script_sections_
.set_section_addresses(symtab
, layout
);
1154 // This class holds data passed through the parser to the lexer and to
1155 // the parser support functions. This avoids global variables. We
1156 // can't use global variables because we need not be called by a
1157 // singleton thread.
1159 class Parser_closure
1162 Parser_closure(const char* filename
,
1163 const Position_dependent_options
& posdep_options
,
1164 bool in_group
, bool is_in_sysroot
,
1165 Command_line
* command_line
,
1166 Script_options
* script_options
,
1168 bool skip_on_incompatible_target
)
1169 : filename_(filename
), posdep_options_(posdep_options
),
1170 in_group_(in_group
), is_in_sysroot_(is_in_sysroot
),
1171 skip_on_incompatible_target_(skip_on_incompatible_target
),
1172 found_incompatible_target_(false),
1173 command_line_(command_line
), script_options_(script_options
),
1174 version_script_info_(script_options
->version_script_info()),
1175 lex_(lex
), lineno_(0), charpos_(0), lex_mode_stack_(), inputs_(NULL
)
1177 // We start out processing C symbols in the default lex mode.
1178 language_stack_
.push_back("");
1179 lex_mode_stack_
.push_back(lex
->mode());
1182 // Return the file name.
1185 { return this->filename_
; }
1187 // Return the position dependent options. The caller may modify
1189 Position_dependent_options
&
1190 position_dependent_options()
1191 { return this->posdep_options_
; }
1193 // Return whether this script is being run in a group.
1196 { return this->in_group_
; }
1198 // Return whether this script was found using a directory in the
1201 is_in_sysroot() const
1202 { return this->is_in_sysroot_
; }
1204 // Whether to skip to the next file with the same name if we find an
1205 // incompatible target in an OUTPUT_FORMAT statement.
1207 skip_on_incompatible_target() const
1208 { return this->skip_on_incompatible_target_
; }
1210 // Stop skipping to the next flie on an incompatible target. This
1211 // is called when we make some unrevocable change to the data
1214 clear_skip_on_incompatible_target()
1215 { this->skip_on_incompatible_target_
= false; }
1217 // Whether we found an incompatible target in an OUTPUT_FORMAT
1220 found_incompatible_target() const
1221 { return this->found_incompatible_target_
; }
1223 // Note that we found an incompatible target.
1225 set_found_incompatible_target()
1226 { this->found_incompatible_target_
= true; }
1228 // Returns the Command_line structure passed in at constructor time.
1229 // This value may be NULL. The caller may modify this, which modifies
1230 // the passed-in Command_line object (not a copy).
1233 { return this->command_line_
; }
1235 // Return the options which may be set by a script.
1238 { return this->script_options_
; }
1240 // Return the object in which version script information should be stored.
1241 Version_script_info
*
1243 { return this->version_script_info_
; }
1245 // Return the next token, and advance.
1249 const Token
* token
= this->lex_
->next_token();
1250 this->lineno_
= token
->lineno();
1251 this->charpos_
= token
->charpos();
1255 // Set a new lexer mode, pushing the current one.
1257 push_lex_mode(Lex::Mode mode
)
1259 this->lex_mode_stack_
.push_back(this->lex_
->mode());
1260 this->lex_
->set_mode(mode
);
1263 // Pop the lexer mode.
1267 gold_assert(!this->lex_mode_stack_
.empty());
1268 this->lex_
->set_mode(this->lex_mode_stack_
.back());
1269 this->lex_mode_stack_
.pop_back();
1272 // Return the current lexer mode.
1275 { return this->lex_mode_stack_
.back(); }
1277 // Return the line number of the last token.
1280 { return this->lineno_
; }
1282 // Return the character position in the line of the last token.
1285 { return this->charpos_
; }
1287 // Return the list of input files, creating it if necessary. This
1288 // is a space leak--we never free the INPUTS_ pointer.
1292 if (this->inputs_
== NULL
)
1293 this->inputs_
= new Input_arguments();
1294 return this->inputs_
;
1297 // Return whether we saw any input files.
1300 { return this->inputs_
!= NULL
&& !this->inputs_
->empty(); }
1302 // Return the current language being processed in a version script
1303 // (eg, "C++"). The empty string represents unmangled C names.
1305 get_current_language() const
1306 { return this->language_stack_
.back(); }
1308 // Push a language onto the stack when entering an extern block.
1309 void push_language(const std::string
& lang
)
1310 { this->language_stack_
.push_back(lang
); }
1312 // Pop a language off of the stack when exiting an extern block.
1315 gold_assert(!this->language_stack_
.empty());
1316 this->language_stack_
.pop_back();
1320 // The name of the file we are reading.
1321 const char* filename_
;
1322 // The position dependent options.
1323 Position_dependent_options posdep_options_
;
1324 // Whether we are currently in a --start-group/--end-group.
1326 // Whether the script was found in a sysrooted directory.
1327 bool is_in_sysroot_
;
1328 // If this is true, then if we find an OUTPUT_FORMAT with an
1329 // incompatible target, then we tell the parser to abort so that we
1330 // can search for the next file with the same name.
1331 bool skip_on_incompatible_target_
;
1332 // True if we found an OUTPUT_FORMAT with an incompatible target.
1333 bool found_incompatible_target_
;
1334 // May be NULL if the user chooses not to pass one in.
1335 Command_line
* command_line_
;
1336 // Options which may be set from any linker script.
1337 Script_options
* script_options_
;
1338 // Information parsed from a version script.
1339 Version_script_info
* version_script_info_
;
1342 // The line number of the last token returned by next_token.
1344 // The column number of the last token returned by next_token.
1346 // A stack of lexer modes.
1347 std::vector
<Lex::Mode
> lex_mode_stack_
;
1348 // A stack of which extern/language block we're inside. Can be C++,
1349 // java, or empty for C.
1350 std::vector
<std::string
> language_stack_
;
1351 // New input files found to add to the link.
1352 Input_arguments
* inputs_
;
1355 // FILE was found as an argument on the command line. Try to read it
1356 // as a script. Return true if the file was handled.
1359 read_input_script(Workqueue
* workqueue
, Symbol_table
* symtab
, Layout
* layout
,
1360 Dirsearch
* dirsearch
, int dirindex
,
1361 Input_objects
* input_objects
, Mapfile
* mapfile
,
1362 Input_group
* input_group
,
1363 const Input_argument
* input_argument
,
1364 Input_file
* input_file
, Task_token
* next_blocker
,
1365 bool* used_next_blocker
)
1367 *used_next_blocker
= false;
1369 std::string input_string
;
1370 Lex::read_file(input_file
, &input_string
);
1372 Lex
lex(input_string
.c_str(), input_string
.length(), PARSING_LINKER_SCRIPT
);
1374 Parser_closure
closure(input_file
->filename().c_str(),
1375 input_argument
->file().options(),
1376 input_group
!= NULL
,
1377 input_file
->is_in_sysroot(),
1379 layout
->script_options(),
1381 input_file
->will_search_for());
1383 if (yyparse(&closure
) != 0)
1385 if (closure
.found_incompatible_target())
1387 Read_symbols::incompatible_warning(input_argument
, input_file
);
1388 Read_symbols::requeue(workqueue
, input_objects
, symtab
, layout
,
1389 dirsearch
, dirindex
, mapfile
, input_argument
,
1390 input_group
, next_blocker
);
1396 if (!closure
.saw_inputs())
1399 Task_token
* this_blocker
= NULL
;
1400 for (Input_arguments::const_iterator p
= closure
.inputs()->begin();
1401 p
!= closure
.inputs()->end();
1405 if (p
+ 1 == closure
.inputs()->end())
1409 nb
= new Task_token(true);
1412 workqueue
->queue_soon(new Read_symbols(input_objects
, symtab
,
1413 layout
, dirsearch
, 0, mapfile
, &*p
,
1414 input_group
, this_blocker
, nb
));
1418 if (layout
->incremental_inputs())
1420 // Like new Read_symbols(...) above, we rely on close.inputs()
1421 // getting leaked by closure.
1422 Script_info
* info
= new Script_info(closure
.inputs());
1423 layout
->incremental_inputs()->report_script(input_argument
, info
);
1425 *used_next_blocker
= true;
1430 // Helper function for read_version_script() and
1431 // read_commandline_script(). Processes the given file in the mode
1432 // indicated by first_token and lex_mode.
1435 read_script_file(const char* filename
, Command_line
* cmdline
,
1436 Script_options
* script_options
,
1437 int first_token
, Lex::Mode lex_mode
)
1439 // TODO: if filename is a relative filename, search for it manually
1440 // using "." + cmdline->options()->search_path() -- not dirsearch.
1441 Dirsearch dirsearch
;
1443 // The file locking code wants to record a Task, but we haven't
1444 // started the workqueue yet. This is only for debugging purposes,
1445 // so we invent a fake value.
1446 const Task
* task
= reinterpret_cast<const Task
*>(-1);
1448 // We don't want this file to be opened in binary mode.
1449 Position_dependent_options posdep
= cmdline
->position_dependent_options();
1450 if (posdep
.format_enum() == General_options::OBJECT_FORMAT_BINARY
)
1451 posdep
.set_format_enum(General_options::OBJECT_FORMAT_ELF
);
1452 Input_file_argument
input_argument(filename
, false, "", false, posdep
);
1453 Input_file
input_file(&input_argument
);
1455 if (!input_file
.open(dirsearch
, task
, &dummy
))
1458 std::string input_string
;
1459 Lex::read_file(&input_file
, &input_string
);
1461 Lex
lex(input_string
.c_str(), input_string
.length(), first_token
);
1462 lex
.set_mode(lex_mode
);
1464 Parser_closure
closure(filename
,
1465 cmdline
->position_dependent_options(),
1467 input_file
.is_in_sysroot(),
1472 if (yyparse(&closure
) != 0)
1474 input_file
.file().unlock(task
);
1478 input_file
.file().unlock(task
);
1480 gold_assert(!closure
.saw_inputs());
1485 // FILENAME was found as an argument to --script (-T).
1486 // Read it as a script, and execute its contents immediately.
1489 read_commandline_script(const char* filename
, Command_line
* cmdline
)
1491 return read_script_file(filename
, cmdline
, &cmdline
->script_options(),
1492 PARSING_LINKER_SCRIPT
, Lex::LINKER_SCRIPT
);
1495 // FILENAME was found as an argument to --version-script. Read it as
1496 // a version script, and store its contents in
1497 // cmdline->script_options()->version_script_info().
1500 read_version_script(const char* filename
, Command_line
* cmdline
)
1502 return read_script_file(filename
, cmdline
, &cmdline
->script_options(),
1503 PARSING_VERSION_SCRIPT
, Lex::VERSION_SCRIPT
);
1506 // FILENAME was found as an argument to --dynamic-list. Read it as a
1507 // list of symbols, and store its contents in DYNAMIC_LIST.
1510 read_dynamic_list(const char* filename
, Command_line
* cmdline
,
1511 Script_options
* dynamic_list
)
1513 return read_script_file(filename
, cmdline
, dynamic_list
,
1514 PARSING_DYNAMIC_LIST
, Lex::DYNAMIC_LIST
);
1517 // Implement the --defsym option on the command line. Return true if
1521 Script_options::define_symbol(const char* definition
)
1523 Lex
lex(definition
, strlen(definition
), PARSING_DEFSYM
);
1524 lex
.set_mode(Lex::EXPRESSION
);
1527 Position_dependent_options posdep_options
;
1529 Parser_closure
closure("command line", posdep_options
, false, false, NULL
,
1532 if (yyparse(&closure
) != 0)
1535 gold_assert(!closure
.saw_inputs());
1540 // Print the script to F for debugging.
1543 Script_options::print(FILE* f
) const
1545 fprintf(f
, "%s: Dumping linker script\n", program_name
);
1547 if (!this->entry_
.empty())
1548 fprintf(f
, "ENTRY(%s)\n", this->entry_
.c_str());
1550 for (Symbol_assignments::const_iterator p
=
1551 this->symbol_assignments_
.begin();
1552 p
!= this->symbol_assignments_
.end();
1556 for (Assertions::const_iterator p
= this->assertions_
.begin();
1557 p
!= this->assertions_
.end();
1561 this->script_sections_
.print(f
);
1563 this->version_script_info_
.print(f
);
1566 // Manage mapping from keywords to the codes expected by the bison
1567 // parser. We construct one global object for each lex mode with
1570 class Keyword_to_parsecode
1573 // The structure which maps keywords to parsecodes.
1574 struct Keyword_parsecode
1577 const char* keyword
;
1578 // Corresponding parsecode.
1582 Keyword_to_parsecode(const Keyword_parsecode
* keywords
,
1584 : keyword_parsecodes_(keywords
), keyword_count_(keyword_count
)
1587 // Return the parsecode corresponding KEYWORD, or 0 if it is not a
1590 keyword_to_parsecode(const char* keyword
, size_t len
) const;
1593 const Keyword_parsecode
* keyword_parsecodes_
;
1594 const int keyword_count_
;
1597 // Mapping from keyword string to keyword parsecode. This array must
1598 // be kept in sorted order. Parsecodes are looked up using bsearch.
1599 // This array must correspond to the list of parsecodes in yyscript.y.
1601 static const Keyword_to_parsecode::Keyword_parsecode
1602 script_keyword_parsecodes
[] =
1604 { "ABSOLUTE", ABSOLUTE
},
1606 { "ALIGN", ALIGN_K
},
1607 { "ALIGNOF", ALIGNOF
},
1608 { "ASSERT", ASSERT_K
},
1609 { "AS_NEEDED", AS_NEEDED
},
1614 { "CONSTANT", CONSTANT
},
1615 { "CONSTRUCTORS", CONSTRUCTORS
},
1616 { "CREATE_OBJECT_SYMBOLS", CREATE_OBJECT_SYMBOLS
},
1617 { "DATA_SEGMENT_ALIGN", DATA_SEGMENT_ALIGN
},
1618 { "DATA_SEGMENT_END", DATA_SEGMENT_END
},
1619 { "DATA_SEGMENT_RELRO_END", DATA_SEGMENT_RELRO_END
},
1620 { "DEFINED", DEFINED
},
1622 { "EXCLUDE_FILE", EXCLUDE_FILE
},
1623 { "EXTERN", EXTERN
},
1626 { "FORCE_COMMON_ALLOCATION", FORCE_COMMON_ALLOCATION
},
1629 { "INCLUDE", INCLUDE
},
1630 { "INHIBIT_COMMON_ALLOCATION", INHIBIT_COMMON_ALLOCATION
},
1633 { "LENGTH", LENGTH
},
1634 { "LOADADDR", LOADADDR
},
1638 { "MEMORY", MEMORY
},
1641 { "NOCROSSREFS", NOCROSSREFS
},
1642 { "NOFLOAT", NOFLOAT
},
1643 { "ONLY_IF_RO", ONLY_IF_RO
},
1644 { "ONLY_IF_RW", ONLY_IF_RW
},
1645 { "OPTION", OPTION
},
1646 { "ORIGIN", ORIGIN
},
1647 { "OUTPUT", OUTPUT
},
1648 { "OUTPUT_ARCH", OUTPUT_ARCH
},
1649 { "OUTPUT_FORMAT", OUTPUT_FORMAT
},
1650 { "OVERLAY", OVERLAY
},
1652 { "PROVIDE", PROVIDE
},
1653 { "PROVIDE_HIDDEN", PROVIDE_HIDDEN
},
1655 { "SEARCH_DIR", SEARCH_DIR
},
1656 { "SECTIONS", SECTIONS
},
1657 { "SEGMENT_START", SEGMENT_START
},
1659 { "SIZEOF", SIZEOF
},
1660 { "SIZEOF_HEADERS", SIZEOF_HEADERS
},
1661 { "SORT", SORT_BY_NAME
},
1662 { "SORT_BY_ALIGNMENT", SORT_BY_ALIGNMENT
},
1663 { "SORT_BY_NAME", SORT_BY_NAME
},
1664 { "SPECIAL", SPECIAL
},
1666 { "STARTUP", STARTUP
},
1667 { "SUBALIGN", SUBALIGN
},
1668 { "SYSLIB", SYSLIB
},
1669 { "TARGET", TARGET_K
},
1670 { "TRUNCATE", TRUNCATE
},
1671 { "VERSION", VERSIONK
},
1672 { "global", GLOBAL
},
1678 { "sizeof_headers", SIZEOF_HEADERS
},
1681 static const Keyword_to_parsecode
1682 script_keywords(&script_keyword_parsecodes
[0],
1683 (sizeof(script_keyword_parsecodes
)
1684 / sizeof(script_keyword_parsecodes
[0])));
1686 static const Keyword_to_parsecode::Keyword_parsecode
1687 version_script_keyword_parsecodes
[] =
1689 { "extern", EXTERN
},
1690 { "global", GLOBAL
},
1694 static const Keyword_to_parsecode
1695 version_script_keywords(&version_script_keyword_parsecodes
[0],
1696 (sizeof(version_script_keyword_parsecodes
)
1697 / sizeof(version_script_keyword_parsecodes
[0])));
1699 static const Keyword_to_parsecode::Keyword_parsecode
1700 dynamic_list_keyword_parsecodes
[] =
1702 { "extern", EXTERN
},
1705 static const Keyword_to_parsecode
1706 dynamic_list_keywords(&dynamic_list_keyword_parsecodes
[0],
1707 (sizeof(dynamic_list_keyword_parsecodes
)
1708 / sizeof(dynamic_list_keyword_parsecodes
[0])));
1712 // Comparison function passed to bsearch.
1724 ktt_compare(const void* keyv
, const void* kttv
)
1726 const Ktt_key
* key
= static_cast<const Ktt_key
*>(keyv
);
1727 const Keyword_to_parsecode::Keyword_parsecode
* ktt
=
1728 static_cast<const Keyword_to_parsecode::Keyword_parsecode
*>(kttv
);
1729 int i
= strncmp(key
->str
, ktt
->keyword
, key
->len
);
1732 if (ktt
->keyword
[key
->len
] != '\0')
1737 } // End extern "C".
1740 Keyword_to_parsecode::keyword_to_parsecode(const char* keyword
,
1746 void* kttv
= bsearch(&key
,
1747 this->keyword_parsecodes_
,
1748 this->keyword_count_
,
1749 sizeof(this->keyword_parsecodes_
[0]),
1753 Keyword_parsecode
* ktt
= static_cast<Keyword_parsecode
*>(kttv
);
1754 return ktt
->parsecode
;
1757 // Helper class that calls cplus_demangle when needed and takes care of freeing
1760 class Lazy_demangler
1763 Lazy_demangler(const char* symbol
, int options
)
1764 : symbol_(symbol
), options_(options
), demangled_(NULL
), did_demangle_(false)
1768 { free(this->demangled_
); }
1770 // Return the demangled name. The actual demangling happens on the first call,
1771 // and the result is later cached.
1777 // The symbol to demangle.
1778 const char *symbol_
;
1779 // Option flags to pass to cplus_demagle.
1781 // The cached demangled value, or NULL if demangling didn't happen yet or
1784 // Whether we already called cplus_demangle
1788 // Return the demangled name. The actual demangling happens on the first call,
1789 // and the result is later cached. Returns NULL if the symbol cannot be
1793 Lazy_demangler::get()
1795 if (!this->did_demangle_
)
1797 this->demangled_
= cplus_demangle(this->symbol_
, this->options_
);
1798 this->did_demangle_
= true;
1800 return this->demangled_
;
1803 // The following structs are used within the VersionInfo class as well
1804 // as in the bison helper functions. They store the information
1805 // parsed from the version script.
1807 // A single version expression.
1808 // For example, pattern="std::map*" and language="C++".
1809 // pattern and language should be from the stringpool
1810 struct Version_expression
{
1811 Version_expression(const std::string
& pattern
,
1812 const std::string
& language
,
1814 : pattern(pattern
), language(language
), exact_match(exact_match
) {}
1816 std::string pattern
;
1817 std::string language
;
1818 // If false, we use glob() to match pattern. If true, we use strcmp().
1823 // A list of expressions.
1824 struct Version_expression_list
{
1825 std::vector
<struct Version_expression
> expressions
;
1829 // A list of which versions upon which another version depends.
1830 // Strings should be from the Stringpool.
1831 struct Version_dependency_list
{
1832 std::vector
<std::string
> dependencies
;
1836 // The total definition of a version. It includes the tag for the
1837 // version, its global and local expressions, and any dependencies.
1838 struct Version_tree
{
1840 : tag(), global(NULL
), local(NULL
), dependencies(NULL
) {}
1843 const struct Version_expression_list
* global
;
1844 const struct Version_expression_list
* local
;
1845 const struct Version_dependency_list
* dependencies
;
1848 Version_script_info::~Version_script_info()
1854 Version_script_info::clear()
1856 for (size_t k
= 0; k
< dependency_lists_
.size(); ++k
)
1857 delete dependency_lists_
[k
];
1858 this->dependency_lists_
.clear();
1859 for (size_t k
= 0; k
< version_trees_
.size(); ++k
)
1860 delete version_trees_
[k
];
1861 this->version_trees_
.clear();
1862 for (size_t k
= 0; k
< expression_lists_
.size(); ++k
)
1863 delete expression_lists_
[k
];
1864 this->expression_lists_
.clear();
1867 std::vector
<std::string
>
1868 Version_script_info::get_versions() const
1870 std::vector
<std::string
> ret
;
1871 for (size_t j
= 0; j
< version_trees_
.size(); ++j
)
1872 if (!this->version_trees_
[j
]->tag
.empty())
1873 ret
.push_back(this->version_trees_
[j
]->tag
);
1877 std::vector
<std::string
>
1878 Version_script_info::get_dependencies(const char* version
) const
1880 std::vector
<std::string
> ret
;
1881 for (size_t j
= 0; j
< version_trees_
.size(); ++j
)
1882 if (version_trees_
[j
]->tag
== version
)
1884 const struct Version_dependency_list
* deps
=
1885 version_trees_
[j
]->dependencies
;
1887 for (size_t k
= 0; k
< deps
->dependencies
.size(); ++k
)
1888 ret
.push_back(deps
->dependencies
[k
]);
1894 // Look up SYMBOL_NAME in the list of versions. If CHECK_GLOBAL is
1895 // true look at the globally visible symbols, otherwise look at the
1896 // symbols listed as "local:". Return true if the symbol is found,
1897 // false otherwise. If the symbol is found, then if PVERSION is not
1898 // NULL, set *PVERSION to the version.
1901 Version_script_info::get_symbol_version_helper(const char* symbol_name
,
1903 std::string
* pversion
) const
1905 Lazy_demangler
cpp_demangled_name(symbol_name
, DMGL_ANSI
| DMGL_PARAMS
);
1906 Lazy_demangler
java_demangled_name(symbol_name
,
1907 DMGL_ANSI
| DMGL_PARAMS
| DMGL_JAVA
);
1908 for (size_t j
= 0; j
< version_trees_
.size(); ++j
)
1910 // Is it a global symbol for this version?
1911 const Version_expression_list
* explist
=
1912 check_global
? version_trees_
[j
]->global
: version_trees_
[j
]->local
;
1913 if (explist
!= NULL
)
1914 for (size_t k
= 0; k
< explist
->expressions
.size(); ++k
)
1916 const char* name_to_match
= symbol_name
;
1917 const struct Version_expression
& exp
= explist
->expressions
[k
];
1918 if (exp
.language
== "C++")
1920 name_to_match
= cpp_demangled_name
.get();
1921 // This isn't a C++ symbol.
1922 if (name_to_match
== NULL
)
1925 else if (exp
.language
== "Java")
1927 name_to_match
= java_demangled_name
.get();
1928 // This isn't a Java symbol.
1929 if (name_to_match
== NULL
)
1933 if (exp
.exact_match
)
1934 matched
= strcmp(exp
.pattern
.c_str(), name_to_match
) == 0;
1936 matched
= fnmatch(exp
.pattern
.c_str(), name_to_match
,
1940 if (pversion
!= NULL
)
1941 *pversion
= this->version_trees_
[j
]->tag
;
1949 struct Version_dependency_list
*
1950 Version_script_info::allocate_dependency_list()
1952 dependency_lists_
.push_back(new Version_dependency_list
);
1953 return dependency_lists_
.back();
1956 struct Version_expression_list
*
1957 Version_script_info::allocate_expression_list()
1959 expression_lists_
.push_back(new Version_expression_list
);
1960 return expression_lists_
.back();
1963 struct Version_tree
*
1964 Version_script_info::allocate_version_tree()
1966 version_trees_
.push_back(new Version_tree
);
1967 return version_trees_
.back();
1970 // Print for debugging.
1973 Version_script_info::print(FILE* f
) const
1978 fprintf(f
, "VERSION {");
1980 for (size_t i
= 0; i
< this->version_trees_
.size(); ++i
)
1982 const Version_tree
* vt
= this->version_trees_
[i
];
1984 if (vt
->tag
.empty())
1987 fprintf(f
, " %s {\n", vt
->tag
.c_str());
1989 if (vt
->global
!= NULL
)
1991 fprintf(f
, " global :\n");
1992 this->print_expression_list(f
, vt
->global
);
1995 if (vt
->local
!= NULL
)
1997 fprintf(f
, " local :\n");
1998 this->print_expression_list(f
, vt
->local
);
2002 if (vt
->dependencies
!= NULL
)
2004 const Version_dependency_list
* deps
= vt
->dependencies
;
2005 for (size_t j
= 0; j
< deps
->dependencies
.size(); ++j
)
2007 if (j
< deps
->dependencies
.size() - 1)
2009 fprintf(f
, " %s", deps
->dependencies
[j
].c_str());
2019 Version_script_info::print_expression_list(
2021 const Version_expression_list
* vel
) const
2023 std::string current_language
;
2024 for (size_t i
= 0; i
< vel
->expressions
.size(); ++i
)
2026 const Version_expression
& ve(vel
->expressions
[i
]);
2028 if (ve
.language
!= current_language
)
2030 if (!current_language
.empty())
2032 fprintf(f
, " extern \"%s\" {\n", ve
.language
.c_str());
2033 current_language
= ve
.language
;
2037 if (!current_language
.empty())
2042 fprintf(f
, "%s", ve
.pattern
.c_str());
2049 if (!current_language
.empty())
2053 } // End namespace gold.
2055 // The remaining functions are extern "C", so it's clearer to not put
2056 // them in namespace gold.
2058 using namespace gold
;
2060 // This function is called by the bison parser to return the next
2064 yylex(YYSTYPE
* lvalp
, void* closurev
)
2066 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2067 const Token
* token
= closure
->next_token();
2068 switch (token
->classification())
2073 case Token::TOKEN_INVALID
:
2074 yyerror(closurev
, "invalid character");
2077 case Token::TOKEN_EOF
:
2080 case Token::TOKEN_STRING
:
2082 // This is either a keyword or a STRING.
2084 const char* str
= token
->string_value(&len
);
2086 switch (closure
->lex_mode())
2088 case Lex::LINKER_SCRIPT
:
2089 parsecode
= script_keywords
.keyword_to_parsecode(str
, len
);
2091 case Lex::VERSION_SCRIPT
:
2092 parsecode
= version_script_keywords
.keyword_to_parsecode(str
, len
);
2094 case Lex::DYNAMIC_LIST
:
2095 parsecode
= dynamic_list_keywords
.keyword_to_parsecode(str
, len
);
2102 lvalp
->string
.value
= str
;
2103 lvalp
->string
.length
= len
;
2107 case Token::TOKEN_QUOTED_STRING
:
2108 lvalp
->string
.value
= token
->string_value(&lvalp
->string
.length
);
2109 return QUOTED_STRING
;
2111 case Token::TOKEN_OPERATOR
:
2112 return token
->operator_value();
2114 case Token::TOKEN_INTEGER
:
2115 lvalp
->integer
= token
->integer_value();
2120 // This function is called by the bison parser to report an error.
2123 yyerror(void* closurev
, const char* message
)
2125 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2126 gold_error(_("%s:%d:%d: %s"), closure
->filename(), closure
->lineno(),
2127 closure
->charpos(), message
);
2130 // Called by the bison parser to add an external symbol to the link.
2133 script_add_extern(void* closurev
, const char* name
, size_t length
)
2135 // We treat exactly like -u NAME. FIXME: If it seems useful, we
2136 // could handle this after the command line has been read, by adding
2137 // entries to the symbol table directly.
2138 std::string
arg("--undefined=");
2139 arg
.append(name
, length
);
2140 script_parse_option(closurev
, arg
.c_str(), arg
.size());
2143 // Called by the bison parser to add a file to the link.
2146 script_add_file(void* closurev
, const char* name
, size_t length
)
2148 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2150 // If this is an absolute path, and we found the script in the
2151 // sysroot, then we want to prepend the sysroot to the file name.
2152 // For example, this is how we handle a cross link to the x86_64
2153 // libc.so, which refers to /lib/libc.so.6.
2154 std::string
name_string(name
, length
);
2155 const char* extra_search_path
= ".";
2156 std::string script_directory
;
2157 if (IS_ABSOLUTE_PATH(name_string
.c_str()))
2159 if (closure
->is_in_sysroot())
2161 const std::string
& sysroot(parameters
->options().sysroot());
2162 gold_assert(!sysroot
.empty());
2163 name_string
= sysroot
+ name_string
;
2168 // In addition to checking the normal library search path, we
2169 // also want to check in the script-directory.
2170 const char *slash
= strrchr(closure
->filename(), '/');
2173 script_directory
.assign(closure
->filename(),
2174 slash
- closure
->filename() + 1);
2175 extra_search_path
= script_directory
.c_str();
2179 Input_file_argument
file(name_string
.c_str(), false, extra_search_path
,
2180 false, closure
->position_dependent_options());
2181 closure
->inputs()->add_file(file
);
2184 // Called by the bison parser to start a group. If we are already in
2185 // a group, that means that this script was invoked within a
2186 // --start-group --end-group sequence on the command line, or that
2187 // this script was found in a GROUP of another script. In that case,
2188 // we simply continue the existing group, rather than starting a new
2189 // one. It is possible to construct a case in which this will do
2190 // something other than what would happen if we did a recursive group,
2191 // but it's hard to imagine why the different behaviour would be
2192 // useful for a real program. Avoiding recursive groups is simpler
2193 // and more efficient.
2196 script_start_group(void* closurev
)
2198 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2199 if (!closure
->in_group())
2200 closure
->inputs()->start_group();
2203 // Called by the bison parser at the end of a group.
2206 script_end_group(void* closurev
)
2208 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2209 if (!closure
->in_group())
2210 closure
->inputs()->end_group();
2213 // Called by the bison parser to start an AS_NEEDED list.
2216 script_start_as_needed(void* closurev
)
2218 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2219 closure
->position_dependent_options().set_as_needed(true);
2222 // Called by the bison parser at the end of an AS_NEEDED list.
2225 script_end_as_needed(void* closurev
)
2227 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2228 closure
->position_dependent_options().set_as_needed(false);
2231 // Called by the bison parser to set the entry symbol.
2234 script_set_entry(void* closurev
, const char* entry
, size_t length
)
2236 // We'll parse this exactly the same as --entry=ENTRY on the commandline
2237 // TODO(csilvers): FIXME -- call set_entry directly.
2238 std::string
arg("--entry=");
2239 arg
.append(entry
, length
);
2240 script_parse_option(closurev
, arg
.c_str(), arg
.size());
2243 // Called by the bison parser to set whether to define common symbols.
2246 script_set_common_allocation(void* closurev
, int set
)
2248 const char* arg
= set
!= 0 ? "--define-common" : "--no-define-common";
2249 script_parse_option(closurev
, arg
, strlen(arg
));
2252 // Called by the bison parser to define a symbol.
2255 script_set_symbol(void* closurev
, const char* name
, size_t length
,
2256 Expression
* value
, int providei
, int hiddeni
)
2258 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2259 const bool provide
= providei
!= 0;
2260 const bool hidden
= hiddeni
!= 0;
2261 closure
->script_options()->add_symbol_assignment(name
, length
, value
,
2263 closure
->clear_skip_on_incompatible_target();
2266 // Called by the bison parser to add an assertion.
2269 script_add_assertion(void* closurev
, Expression
* check
, const char* message
,
2272 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2273 closure
->script_options()->add_assertion(check
, message
, messagelen
);
2274 closure
->clear_skip_on_incompatible_target();
2277 // Called by the bison parser to parse an OPTION.
2280 script_parse_option(void* closurev
, const char* option
, size_t length
)
2282 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2283 // We treat the option as a single command-line option, even if
2284 // it has internal whitespace.
2285 if (closure
->command_line() == NULL
)
2287 // There are some options that we could handle here--e.g.,
2288 // -lLIBRARY. Should we bother?
2289 gold_warning(_("%s:%d:%d: ignoring command OPTION; OPTION is only valid"
2290 " for scripts specified via -T/--script"),
2291 closure
->filename(), closure
->lineno(), closure
->charpos());
2295 bool past_a_double_dash_option
= false;
2296 const char* mutable_option
= strndup(option
, length
);
2297 gold_assert(mutable_option
!= NULL
);
2298 closure
->command_line()->process_one_option(1, &mutable_option
, 0,
2299 &past_a_double_dash_option
);
2300 // The General_options class will quite possibly store a pointer
2301 // into mutable_option, so we can't free it. In cases the class
2302 // does not store such a pointer, this is a memory leak. Alas. :(
2304 closure
->clear_skip_on_incompatible_target();
2307 // Called by the bison parser to handle OUTPUT_FORMAT. OUTPUT_FORMAT
2308 // takes either one or three arguments. In the three argument case,
2309 // the format depends on the endianness option, which we don't
2310 // currently support (FIXME). If we see an OUTPUT_FORMAT for the
2311 // wrong format, then we want to search for a new file. Returning 0
2312 // here will cause the parser to immediately abort.
2315 script_check_output_format(void* closurev
,
2316 const char* default_name
, size_t default_length
,
2317 const char*, size_t, const char*, size_t)
2319 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2320 std::string
name(default_name
, default_length
);
2321 Target
* target
= select_target_by_name(name
.c_str());
2322 if (target
== NULL
|| !parameters
->is_compatible_target(target
))
2324 if (closure
->skip_on_incompatible_target())
2326 closure
->set_found_incompatible_target();
2329 // FIXME: Should we warn about the unknown target?
2334 // Called by the bison parser to handle SEARCH_DIR. This is handled
2335 // exactly like a -L option.
2338 script_add_search_dir(void* closurev
, const char* option
, size_t length
)
2340 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2341 if (closure
->command_line() == NULL
)
2342 gold_warning(_("%s:%d:%d: ignoring SEARCH_DIR; SEARCH_DIR is only valid"
2343 " for scripts specified via -T/--script"),
2344 closure
->filename(), closure
->lineno(), closure
->charpos());
2347 std::string s
= "-L" + std::string(option
, length
);
2348 script_parse_option(closurev
, s
.c_str(), s
.size());
2352 /* Called by the bison parser to push the lexer into expression
2356 script_push_lex_into_expression_mode(void* closurev
)
2358 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2359 closure
->push_lex_mode(Lex::EXPRESSION
);
2362 /* Called by the bison parser to push the lexer into version
2366 script_push_lex_into_version_mode(void* closurev
)
2368 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2369 closure
->push_lex_mode(Lex::VERSION_SCRIPT
);
2372 /* Called by the bison parser to pop the lexer mode. */
2375 script_pop_lex_mode(void* closurev
)
2377 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2378 closure
->pop_lex_mode();
2381 // Register an entire version node. For example:
2387 // - tag is "GLIBC_2.1"
2388 // - tree contains the information "global: foo"
2389 // - deps contains "GLIBC_2.0"
2392 script_register_vers_node(void*,
2395 struct Version_tree
*tree
,
2396 struct Version_dependency_list
*deps
)
2398 gold_assert(tree
!= NULL
);
2399 tree
->dependencies
= deps
;
2401 tree
->tag
= std::string(tag
, taglen
);
2404 // Add a dependencies to the list of existing dependencies, if any,
2405 // and return the expanded list.
2407 extern "C" struct Version_dependency_list
*
2408 script_add_vers_depend(void* closurev
,
2409 struct Version_dependency_list
*all_deps
,
2410 const char *depend_to_add
, int deplen
)
2412 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2413 if (all_deps
== NULL
)
2414 all_deps
= closure
->version_script()->allocate_dependency_list();
2415 all_deps
->dependencies
.push_back(std::string(depend_to_add
, deplen
));
2419 // Add a pattern expression to an existing list of expressions, if any.
2420 // TODO: In the old linker, the last argument used to be a bool, but I
2421 // don't know what it meant.
2423 extern "C" struct Version_expression_list
*
2424 script_new_vers_pattern(void* closurev
,
2425 struct Version_expression_list
*expressions
,
2426 const char *pattern
, int patlen
, int exact_match
)
2428 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2429 if (expressions
== NULL
)
2430 expressions
= closure
->version_script()->allocate_expression_list();
2431 expressions
->expressions
.push_back(
2432 Version_expression(std::string(pattern
, patlen
),
2433 closure
->get_current_language(),
2434 static_cast<bool>(exact_match
)));
2438 // Attaches b to the end of a, and clears b. So a = a + b and b = {}.
2440 extern "C" struct Version_expression_list
*
2441 script_merge_expressions(struct Version_expression_list
*a
,
2442 struct Version_expression_list
*b
)
2444 a
->expressions
.insert(a
->expressions
.end(),
2445 b
->expressions
.begin(), b
->expressions
.end());
2446 // We could delete b and remove it from expressions_lists_, but
2447 // that's a lot of work. This works just as well.
2448 b
->expressions
.clear();
2452 // Combine the global and local expressions into a a Version_tree.
2454 extern "C" struct Version_tree
*
2455 script_new_vers_node(void* closurev
,
2456 struct Version_expression_list
*global
,
2457 struct Version_expression_list
*local
)
2459 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2460 Version_tree
* tree
= closure
->version_script()->allocate_version_tree();
2461 tree
->global
= global
;
2462 tree
->local
= local
;
2466 // Handle a transition in language, such as at the
2467 // start or end of 'extern "C++"'
2470 version_script_push_lang(void* closurev
, const char* lang
, int langlen
)
2472 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2473 closure
->push_language(std::string(lang
, langlen
));
2477 version_script_pop_lang(void* closurev
)
2479 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2480 closure
->pop_language();
2483 // Called by the bison parser to start a SECTIONS clause.
2486 script_start_sections(void* closurev
)
2488 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2489 closure
->script_options()->script_sections()->start_sections();
2490 closure
->clear_skip_on_incompatible_target();
2493 // Called by the bison parser to finish a SECTIONS clause.
2496 script_finish_sections(void* closurev
)
2498 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2499 closure
->script_options()->script_sections()->finish_sections();
2502 // Start processing entries for an output section.
2505 script_start_output_section(void* closurev
, const char* name
, size_t namelen
,
2506 const struct Parser_output_section_header
* header
)
2508 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2509 closure
->script_options()->script_sections()->start_output_section(name
,
2514 // Finish processing entries for an output section.
2517 script_finish_output_section(void* closurev
,
2518 const struct Parser_output_section_trailer
* trail
)
2520 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2521 closure
->script_options()->script_sections()->finish_output_section(trail
);
2524 // Add a data item (e.g., "WORD (0)") to the current output section.
2527 script_add_data(void* closurev
, int data_token
, Expression
* val
)
2529 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2531 bool is_signed
= true;
2553 closure
->script_options()->script_sections()->add_data(size
, is_signed
, val
);
2556 // Add a clause setting the fill value to the current output section.
2559 script_add_fill(void* closurev
, Expression
* val
)
2561 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2562 closure
->script_options()->script_sections()->add_fill(val
);
2565 // Add a new input section specification to the current output
2569 script_add_input_section(void* closurev
,
2570 const struct Input_section_spec
* spec
,
2573 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2574 bool keep
= keepi
!= 0;
2575 closure
->script_options()->script_sections()->add_input_section(spec
, keep
);
2578 // When we see DATA_SEGMENT_ALIGN we record that following output
2579 // sections may be relro.
2582 script_data_segment_align(void* closurev
)
2584 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2585 if (!closure
->script_options()->saw_sections_clause())
2586 gold_error(_("%s:%d:%d: DATA_SEGMENT_ALIGN not in SECTIONS clause"),
2587 closure
->filename(), closure
->lineno(), closure
->charpos());
2589 closure
->script_options()->script_sections()->data_segment_align();
2592 // When we see DATA_SEGMENT_RELRO_END we know that all output sections
2593 // since DATA_SEGMENT_ALIGN should be relro.
2596 script_data_segment_relro_end(void* closurev
)
2598 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2599 if (!closure
->script_options()->saw_sections_clause())
2600 gold_error(_("%s:%d:%d: DATA_SEGMENT_ALIGN not in SECTIONS clause"),
2601 closure
->filename(), closure
->lineno(), closure
->charpos());
2603 closure
->script_options()->script_sections()->data_segment_relro_end();
2606 // Create a new list of string/sort pairs.
2608 extern "C" String_sort_list_ptr
2609 script_new_string_sort_list(const struct Wildcard_section
* string_sort
)
2611 return new String_sort_list(1, *string_sort
);
2614 // Add an entry to a list of string/sort pairs. The way the parser
2615 // works permits us to simply modify the first parameter, rather than
2618 extern "C" String_sort_list_ptr
2619 script_string_sort_list_add(String_sort_list_ptr pv
,
2620 const struct Wildcard_section
* string_sort
)
2623 return script_new_string_sort_list(string_sort
);
2626 pv
->push_back(*string_sort
);
2631 // Create a new list of strings.
2633 extern "C" String_list_ptr
2634 script_new_string_list(const char* str
, size_t len
)
2636 return new String_list(1, std::string(str
, len
));
2639 // Add an element to a list of strings. The way the parser works
2640 // permits us to simply modify the first parameter, rather than copy
2643 extern "C" String_list_ptr
2644 script_string_list_push_back(String_list_ptr pv
, const char* str
, size_t len
)
2647 return script_new_string_list(str
, len
);
2650 pv
->push_back(std::string(str
, len
));
2655 // Concatenate two string lists. Either or both may be NULL. The way
2656 // the parser works permits us to modify the parameters, rather than
2659 extern "C" String_list_ptr
2660 script_string_list_append(String_list_ptr pv1
, String_list_ptr pv2
)
2666 pv1
->insert(pv1
->end(), pv2
->begin(), pv2
->end());
2670 // Add a new program header.
2673 script_add_phdr(void* closurev
, const char* name
, size_t namelen
,
2674 unsigned int type
, const Phdr_info
* info
)
2676 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2677 bool includes_filehdr
= info
->includes_filehdr
!= 0;
2678 bool includes_phdrs
= info
->includes_phdrs
!= 0;
2679 bool is_flags_valid
= info
->is_flags_valid
!= 0;
2680 Script_sections
* ss
= closure
->script_options()->script_sections();
2681 ss
->add_phdr(name
, namelen
, type
, includes_filehdr
, includes_phdrs
,
2682 is_flags_valid
, info
->flags
, info
->load_address
);
2683 closure
->clear_skip_on_incompatible_target();
2686 // Convert a program header string to a type.
2688 #define PHDR_TYPE(NAME) { #NAME, sizeof(#NAME) - 1, elfcpp::NAME }
2695 } phdr_type_names
[] =
2699 PHDR_TYPE(PT_DYNAMIC
),
2700 PHDR_TYPE(PT_INTERP
),
2702 PHDR_TYPE(PT_SHLIB
),
2705 PHDR_TYPE(PT_GNU_EH_FRAME
),
2706 PHDR_TYPE(PT_GNU_STACK
),
2707 PHDR_TYPE(PT_GNU_RELRO
)
2710 extern "C" unsigned int
2711 script_phdr_string_to_type(void* closurev
, const char* name
, size_t namelen
)
2713 for (unsigned int i
= 0;
2714 i
< sizeof(phdr_type_names
) / sizeof(phdr_type_names
[0]);
2716 if (namelen
== phdr_type_names
[i
].namelen
2717 && strncmp(name
, phdr_type_names
[i
].name
, namelen
) == 0)
2718 return phdr_type_names
[i
].val
;
2719 yyerror(closurev
, _("unknown PHDR type (try integer)"));
2720 return elfcpp::PT_NULL
;