1 // script.cc -- handle linker scripts for gold.
3 // Copyright (C) 2006-2017 Free Software Foundation, Inc.
4 // Written by Ian Lance Taylor <iant@google.com>.
6 // This file is part of gold.
8 // This program is free software; you can redistribute it and/or modify
9 // it under the terms of the GNU General Public License as published by
10 // the Free Software Foundation; either version 3 of the License, or
11 // (at your option) any later version.
13 // This program is distributed in the hope that it will be useful,
14 // but WITHOUT ANY WARRANTY; without even the implied warranty of
15 // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 // GNU General Public License for more details.
18 // You should have received a copy of the GNU General Public License
19 // along with this program; if not, write to the Free Software
20 // Foundation, Inc., 51 Franklin Street - Fifth Floor, Boston,
21 // MA 02110-1301, USA.
31 #include "filenames.h"
35 #include "dirsearch.h"
38 #include "workqueue.h"
40 #include "parameters.h"
43 #include "target-select.h"
46 #include "incremental.h"
51 // A token read from a script file. We don't implement keywords here;
52 // all keywords are simply represented as a string.
57 // Token classification.
62 // Token indicates end of input.
64 // Token is a string of characters.
66 // Token is a quoted string of characters.
68 // Token is an operator.
70 // Token is a number (an integer).
74 // We need an empty constructor so that we can put this STL objects.
76 : classification_(TOKEN_INVALID
), value_(NULL
), value_length_(0),
77 opcode_(0), lineno_(0), charpos_(0)
80 // A general token with no value.
81 Token(Classification classification
, int lineno
, int charpos
)
82 : classification_(classification
), value_(NULL
), value_length_(0),
83 opcode_(0), lineno_(lineno
), charpos_(charpos
)
85 gold_assert(classification
== TOKEN_INVALID
86 || classification
== TOKEN_EOF
);
89 // A general token with a value.
90 Token(Classification classification
, const char* value
, size_t length
,
91 int lineno
, int charpos
)
92 : classification_(classification
), value_(value
), value_length_(length
),
93 opcode_(0), lineno_(lineno
), charpos_(charpos
)
95 gold_assert(classification
!= TOKEN_INVALID
96 && classification
!= TOKEN_EOF
);
99 // A token representing an operator.
100 Token(int opcode
, int lineno
, int charpos
)
101 : classification_(TOKEN_OPERATOR
), value_(NULL
), value_length_(0),
102 opcode_(opcode
), lineno_(lineno
), charpos_(charpos
)
105 // Return whether the token is invalid.
108 { return this->classification_
== TOKEN_INVALID
; }
110 // Return whether this is an EOF token.
113 { return this->classification_
== TOKEN_EOF
; }
115 // Return the token classification.
117 classification() const
118 { return this->classification_
; }
120 // Return the line number at which the token starts.
123 { return this->lineno_
; }
125 // Return the character position at this the token starts.
128 { return this->charpos_
; }
130 // Get the value of a token.
133 string_value(size_t* length
) const
135 gold_assert(this->classification_
== TOKEN_STRING
136 || this->classification_
== TOKEN_QUOTED_STRING
);
137 *length
= this->value_length_
;
142 operator_value() const
144 gold_assert(this->classification_
== TOKEN_OPERATOR
);
145 return this->opcode_
;
149 integer_value() const;
152 // The token classification.
153 Classification classification_
;
154 // The token value, for TOKEN_STRING or TOKEN_QUOTED_STRING or
157 // The length of the token value.
158 size_t value_length_
;
159 // The token value, for TOKEN_OPERATOR.
161 // The line number where this token started (one based).
163 // The character position within the line where this token started
168 // Return the value of a TOKEN_INTEGER.
171 Token::integer_value() const
173 gold_assert(this->classification_
== TOKEN_INTEGER
);
175 size_t len
= this->value_length_
;
177 uint64_t multiplier
= 1;
178 char last
= this->value_
[len
- 1];
179 if (last
== 'm' || last
== 'M')
181 multiplier
= 1024 * 1024;
184 else if (last
== 'k' || last
== 'K')
191 uint64_t ret
= strtoull(this->value_
, &end
, 0);
192 gold_assert(static_cast<size_t>(end
- this->value_
) == len
);
194 return ret
* multiplier
;
197 // This class handles lexing a file into a sequence of tokens.
202 // We unfortunately have to support different lexing modes, because
203 // when reading different parts of a linker script we need to parse
204 // things differently.
207 // Reading an ordinary linker script.
209 // Reading an expression in a linker script.
211 // Reading a version script.
213 // Reading a --dynamic-list file.
217 Lex(const char* input_string
, size_t input_length
, int parsing_token
)
218 : input_string_(input_string
), input_length_(input_length
),
219 current_(input_string
), mode_(LINKER_SCRIPT
),
220 first_token_(parsing_token
), token_(),
221 lineno_(1), linestart_(input_string
)
224 // Read a file into a string.
226 read_file(Input_file
*, std::string
*);
228 // Return the next token.
232 // Return the current lexing mode.
235 { return this->mode_
; }
237 // Set the lexing mode.
240 { this->mode_
= mode
; }
244 Lex
& operator=(const Lex
&);
246 // Make a general token with no value at the current location.
248 make_token(Token::Classification c
, const char* start
) const
249 { return Token(c
, this->lineno_
, start
- this->linestart_
+ 1); }
251 // Make a general token with a value at the current location.
253 make_token(Token::Classification c
, const char* v
, size_t len
,
256 { return Token(c
, v
, len
, this->lineno_
, start
- this->linestart_
+ 1); }
258 // Make an operator token at the current location.
260 make_token(int opcode
, const char* start
) const
261 { return Token(opcode
, this->lineno_
, start
- this->linestart_
+ 1); }
263 // Make an invalid token at the current location.
265 make_invalid_token(const char* start
)
266 { return this->make_token(Token::TOKEN_INVALID
, start
); }
268 // Make an EOF token at the current location.
270 make_eof_token(const char* start
)
271 { return this->make_token(Token::TOKEN_EOF
, start
); }
273 // Return whether C can be the first character in a name. C2 is the
274 // next character, since we sometimes need that.
276 can_start_name(char c
, char c2
);
278 // If C can appear in a name which has already started, return a
279 // pointer to a character later in the token or just past
280 // it. Otherwise, return NULL.
282 can_continue_name(const char* c
);
284 // Return whether C, C2, C3 can start a hex number.
286 can_start_hex(char c
, char c2
, char c3
);
288 // If C can appear in a hex number which has already started, return
289 // a pointer to a character later in the token or just past
290 // it. Otherwise, return NULL.
292 can_continue_hex(const char* c
);
294 // Return whether C can start a non-hex number.
296 can_start_number(char c
);
298 // If C can appear in a decimal number which has already started,
299 // return a pointer to a character later in the token or just past
300 // it. Otherwise, return NULL.
302 can_continue_number(const char* c
)
303 { return Lex::can_start_number(*c
) ? c
+ 1 : NULL
; }
305 // If C1 C2 C3 form a valid three character operator, return the
306 // opcode. Otherwise return 0.
308 three_char_operator(char c1
, char c2
, char c3
);
310 // If C1 C2 form a valid two character operator, return the opcode.
311 // Otherwise return 0.
313 two_char_operator(char c1
, char c2
);
315 // If C1 is a valid one character operator, return the opcode.
316 // Otherwise return 0.
318 one_char_operator(char c1
);
320 // Read the next token.
322 get_token(const char**);
324 // Skip a C style /* */ comment. Return false if the comment did
327 skip_c_comment(const char**);
329 // Skip a line # comment. Return false if there was no newline.
331 skip_line_comment(const char**);
333 // Build a token CLASSIFICATION from all characters that match
334 // CAN_CONTINUE_FN. The token starts at START. Start matching from
335 // MATCH. Set *PP to the character following the token.
337 gather_token(Token::Classification
,
338 const char* (Lex::*can_continue_fn
)(const char*),
339 const char* start
, const char* match
, const char** pp
);
341 // Build a token from a quoted string.
343 gather_quoted_string(const char** pp
);
345 // The string we are tokenizing.
346 const char* input_string_
;
347 // The length of the string.
348 size_t input_length_
;
349 // The current offset into the string.
350 const char* current_
;
351 // The current lexing mode.
353 // The code to use for the first token. This is set to 0 after it
356 // The current token.
358 // The current line number.
360 // The start of the current line in the string.
361 const char* linestart_
;
364 // Read the whole file into memory. We don't expect linker scripts to
365 // be large, so we just use a std::string as a buffer. We ignore the
366 // data we've already read, so that we read aligned buffers.
369 Lex::read_file(Input_file
* input_file
, std::string
* contents
)
371 off_t filesize
= input_file
->file().filesize();
373 contents
->reserve(filesize
);
376 unsigned char buf
[BUFSIZ
];
377 while (off
< filesize
)
380 if (get
> filesize
- off
)
381 get
= filesize
- off
;
382 input_file
->file().read(off
, get
, buf
);
383 contents
->append(reinterpret_cast<char*>(&buf
[0]), get
);
388 // Return whether C can be the start of a name, if the next character
389 // is C2. A name can being with a letter, underscore, period, or
390 // dollar sign. Because a name can be a file name, we also permit
391 // forward slash, backslash, and tilde. Tilde is the tricky case
392 // here; GNU ld also uses it as a bitwise not operator. It is only
393 // recognized as the operator if it is not immediately followed by
394 // some character which can appear in a symbol. That is, when we
395 // don't know that we are looking at an expression, "~0" is a file
396 // name, and "~ 0" is an expression using bitwise not. We are
400 Lex::can_start_name(char c
, char c2
)
404 case 'A': case 'B': case 'C': case 'D': case 'E': case 'F':
405 case 'G': case 'H': case 'I': case 'J': case 'K': case 'L':
406 case 'M': case 'N': case 'O': case 'Q': case 'P': case 'R':
407 case 'S': case 'T': case 'U': case 'V': case 'W': case 'X':
409 case 'a': case 'b': case 'c': case 'd': case 'e': case 'f':
410 case 'g': case 'h': case 'i': case 'j': case 'k': case 'l':
411 case 'm': case 'n': case 'o': case 'q': case 'p': case 'r':
412 case 's': case 't': case 'u': case 'v': case 'w': case 'x':
414 case '_': case '.': case '$':
418 return this->mode_
== LINKER_SCRIPT
;
421 return this->mode_
== LINKER_SCRIPT
&& can_continue_name(&c2
);
424 return (this->mode_
== VERSION_SCRIPT
425 || this->mode_
== DYNAMIC_LIST
426 || (this->mode_
== LINKER_SCRIPT
427 && can_continue_name(&c2
)));
434 // Return whether C can continue a name which has already started.
435 // Subsequent characters in a name are the same as the leading
436 // characters, plus digits and "=+-:[],?*". So in general the linker
437 // script language requires spaces around operators, unless we know
438 // that we are parsing an expression.
441 Lex::can_continue_name(const char* c
)
445 case 'A': case 'B': case 'C': case 'D': case 'E': case 'F':
446 case 'G': case 'H': case 'I': case 'J': case 'K': case 'L':
447 case 'M': case 'N': case 'O': case 'Q': case 'P': case 'R':
448 case 'S': case 'T': case 'U': case 'V': case 'W': case 'X':
450 case 'a': case 'b': case 'c': case 'd': case 'e': case 'f':
451 case 'g': case 'h': case 'i': case 'j': case 'k': case 'l':
452 case 'm': case 'n': case 'o': case 'q': case 'p': case 'r':
453 case 's': case 't': case 'u': case 'v': case 'w': case 'x':
455 case '_': case '.': case '$':
456 case '0': case '1': case '2': case '3': case '4':
457 case '5': case '6': case '7': case '8': case '9':
460 // TODO(csilvers): why not allow ~ in names for version-scripts?
461 case '/': case '\\': case '~':
464 if (this->mode_
== LINKER_SCRIPT
)
468 case '[': case ']': case '*': case '?': case '-':
469 if (this->mode_
== LINKER_SCRIPT
|| this->mode_
== VERSION_SCRIPT
470 || this->mode_
== DYNAMIC_LIST
)
474 // TODO(csilvers): why allow this? ^ is meaningless in version scripts.
476 if (this->mode_
== VERSION_SCRIPT
|| this->mode_
== DYNAMIC_LIST
)
481 if (this->mode_
== LINKER_SCRIPT
)
483 else if ((this->mode_
== VERSION_SCRIPT
|| this->mode_
== DYNAMIC_LIST
)
486 // A name can have '::' in it, as that's a c++ namespace
487 // separator. But a single colon is not part of a name.
497 // For a number we accept 0x followed by hex digits, or any sequence
498 // of digits. The old linker accepts leading '$' for hex, and
499 // trailing HXBOD. Those are for MRI compatibility and we don't
502 // Return whether C1 C2 C3 can start a hex number.
505 Lex::can_start_hex(char c1
, char c2
, char c3
)
507 if (c1
== '0' && (c2
== 'x' || c2
== 'X'))
508 return this->can_continue_hex(&c3
);
512 // Return whether C can appear in a hex number.
515 Lex::can_continue_hex(const char* c
)
519 case '0': case '1': case '2': case '3': case '4':
520 case '5': case '6': case '7': case '8': case '9':
521 case 'A': case 'B': case 'C': case 'D': case 'E': case 'F':
522 case 'a': case 'b': case 'c': case 'd': case 'e': case 'f':
530 // Return whether C can start a non-hex number.
533 Lex::can_start_number(char c
)
537 case '0': case '1': case '2': case '3': case '4':
538 case '5': case '6': case '7': case '8': case '9':
546 // If C1 C2 C3 form a valid three character operator, return the
547 // opcode (defined in the yyscript.h file generated from yyscript.y).
548 // Otherwise return 0.
551 Lex::three_char_operator(char c1
, char c2
, char c3
)
556 if (c2
== '<' && c3
== '=')
560 if (c2
== '>' && c3
== '=')
569 // If C1 C2 form a valid two character operator, return the opcode
570 // (defined in the yyscript.h file generated from yyscript.y).
571 // Otherwise return 0.
574 Lex::two_char_operator(char c1
, char c2
)
632 // If C1 is a valid operator, return the opcode. Otherwise return 0.
635 Lex::one_char_operator(char c1
)
668 // Skip a C style comment. *PP points to just after the "/*". Return
669 // false if the comment did not end.
672 Lex::skip_c_comment(const char** pp
)
675 while (p
[0] != '*' || p
[1] != '/')
686 this->linestart_
= p
+ 1;
695 // Skip a line # comment. Return false if there was no newline.
698 Lex::skip_line_comment(const char** pp
)
701 size_t skip
= strcspn(p
, "\n");
710 this->linestart_
= p
;
716 // Build a token CLASSIFICATION from all characters that match
717 // CAN_CONTINUE_FN. Update *PP.
720 Lex::gather_token(Token::Classification classification
,
721 const char* (Lex::*can_continue_fn
)(const char*),
726 const char* new_match
= NULL
;
727 while ((new_match
= (this->*can_continue_fn
)(match
)) != NULL
)
730 // A special case: integers may be followed by a single M or K,
732 if (classification
== Token::TOKEN_INTEGER
733 && (*match
== 'm' || *match
== 'M' || *match
== 'k' || *match
== 'K'))
737 return this->make_token(classification
, start
, match
- start
, start
);
740 // Build a token from a quoted string.
743 Lex::gather_quoted_string(const char** pp
)
745 const char* start
= *pp
;
746 const char* p
= start
;
748 size_t skip
= strcspn(p
, "\"\n");
750 return this->make_invalid_token(start
);
752 return this->make_token(Token::TOKEN_QUOTED_STRING
, p
, skip
, start
);
755 // Return the next token at *PP. Update *PP. General guideline: we
756 // require linker scripts to be simple ASCII. No unicode linker
757 // scripts. In particular we can assume that any '\0' is the end of
761 Lex::get_token(const char** pp
)
767 // Skip whitespace quickly.
768 while (*p
== ' ' || *p
== '\t' || *p
== '\r')
775 this->linestart_
= p
;
784 return this->make_eof_token(p
);
789 // Skip C style comments.
790 if (c0
== '/' && c1
== '*')
792 int lineno
= this->lineno_
;
793 int charpos
= p
- this->linestart_
+ 1;
796 if (!this->skip_c_comment(pp
))
797 return Token(Token::TOKEN_INVALID
, lineno
, charpos
);
803 // Skip line comments.
807 if (!this->skip_line_comment(pp
))
808 return this->make_eof_token(p
);
814 if (this->can_start_name(c0
, c1
))
815 return this->gather_token(Token::TOKEN_STRING
,
816 &Lex::can_continue_name
,
819 // We accept any arbitrary name in double quotes, as long as it
820 // does not cross a line boundary.
824 return this->gather_quoted_string(pp
);
827 // Be careful not to lookahead past the end of the buffer.
828 char c2
= (c1
== '\0' ? '\0' : p
[2]);
830 // Check for a number.
832 if (this->can_start_hex(c0
, c1
, c2
))
833 return this->gather_token(Token::TOKEN_INTEGER
,
834 &Lex::can_continue_hex
,
837 if (Lex::can_start_number(c0
))
838 return this->gather_token(Token::TOKEN_INTEGER
,
839 &Lex::can_continue_number
,
842 // Check for operators.
844 int opcode
= Lex::three_char_operator(c0
, c1
, c2
);
848 return this->make_token(opcode
, p
);
851 opcode
= Lex::two_char_operator(c0
, c1
);
855 return this->make_token(opcode
, p
);
858 opcode
= Lex::one_char_operator(c0
);
862 return this->make_token(opcode
, p
);
865 return this->make_token(Token::TOKEN_INVALID
, p
);
869 // Return the next token.
874 // The first token is special.
875 if (this->first_token_
!= 0)
877 this->token_
= Token(this->first_token_
, 0, 0);
878 this->first_token_
= 0;
879 return &this->token_
;
882 this->token_
= this->get_token(&this->current_
);
884 // Don't let an early null byte fool us into thinking that we've
885 // reached the end of the file.
886 if (this->token_
.is_eof()
887 && (static_cast<size_t>(this->current_
- this->input_string_
)
888 < this->input_length_
))
889 this->token_
= this->make_invalid_token(this->current_
);
891 return &this->token_
;
894 // class Symbol_assignment.
896 // Add the symbol to the symbol table. This makes sure the symbol is
897 // there and defined. The actual value is stored later. We can't
898 // determine the actual value at this point, because we can't
899 // necessarily evaluate the expression until all ordinary symbols have
902 // The GNU linker lets symbol assignments in the linker script
903 // silently override defined symbols in object files. We are
904 // compatible. FIXME: Should we issue a warning?
907 Symbol_assignment::add_to_table(Symbol_table
* symtab
)
909 elfcpp::STV vis
= this->hidden_
? elfcpp::STV_HIDDEN
: elfcpp::STV_DEFAULT
;
910 this->sym_
= symtab
->define_as_constant(this->name_
.c_str(),
913 ? Symbol_table::DEFSYM
914 : Symbol_table::SCRIPT
),
922 true); // force_override
925 // Finalize a symbol value.
928 Symbol_assignment::finalize(Symbol_table
* symtab
, const Layout
* layout
)
930 this->finalize_maybe_dot(symtab
, layout
, false, 0, NULL
);
933 // Finalize a symbol value which can refer to the dot symbol.
936 Symbol_assignment::finalize_with_dot(Symbol_table
* symtab
,
937 const Layout
* layout
,
939 Output_section
* dot_section
)
941 this->finalize_maybe_dot(symtab
, layout
, true, dot_value
, dot_section
);
944 // Finalize a symbol value, internal version.
947 Symbol_assignment::finalize_maybe_dot(Symbol_table
* symtab
,
948 const Layout
* layout
,
949 bool is_dot_available
,
951 Output_section
* dot_section
)
953 // If we were only supposed to provide this symbol, the sym_ field
954 // will be NULL if the symbol was not referenced.
955 if (this->sym_
== NULL
)
957 gold_assert(this->provide_
);
961 if (parameters
->target().get_size() == 32)
963 #if defined(HAVE_TARGET_32_LITTLE) || defined(HAVE_TARGET_32_BIG)
964 this->sized_finalize
<32>(symtab
, layout
, is_dot_available
, dot_value
,
970 else if (parameters
->target().get_size() == 64)
972 #if defined(HAVE_TARGET_64_LITTLE) || defined(HAVE_TARGET_64_BIG)
973 this->sized_finalize
<64>(symtab
, layout
, is_dot_available
, dot_value
,
985 Symbol_assignment::sized_finalize(Symbol_table
* symtab
, const Layout
* layout
,
986 bool is_dot_available
, uint64_t dot_value
,
987 Output_section
* dot_section
)
989 Output_section
* section
;
990 elfcpp::STT type
= elfcpp::STT_NOTYPE
;
991 elfcpp::STV vis
= elfcpp::STV_DEFAULT
;
992 unsigned char nonvis
= 0;
993 uint64_t final_val
= this->val_
->eval_maybe_dot(symtab
, layout
, true,
995 dot_value
, dot_section
,
996 §ion
, NULL
, &type
,
997 &vis
, &nonvis
, false, NULL
);
998 Sized_symbol
<size
>* ssym
= symtab
->get_sized_symbol
<size
>(this->sym_
);
999 ssym
->set_value(final_val
);
1000 ssym
->set_type(type
);
1001 ssym
->set_visibility(vis
);
1002 ssym
->set_nonvis(nonvis
);
1003 if (section
!= NULL
)
1004 ssym
->set_output_section(section
);
1007 // Set the symbol value if the expression yields an absolute value or
1008 // a value relative to DOT_SECTION.
1011 Symbol_assignment::set_if_absolute(Symbol_table
* symtab
, const Layout
* layout
,
1012 bool is_dot_available
, uint64_t dot_value
,
1013 Output_section
* dot_section
)
1015 if (this->sym_
== NULL
)
1018 Output_section
* val_section
;
1020 uint64_t val
= this->val_
->eval_maybe_dot(symtab
, layout
, false,
1021 is_dot_available
, dot_value
,
1022 dot_section
, &val_section
, NULL
,
1023 NULL
, NULL
, NULL
, false, &is_valid
);
1024 if (!is_valid
|| (val_section
!= NULL
&& val_section
!= dot_section
))
1027 if (parameters
->target().get_size() == 32)
1029 #if defined(HAVE_TARGET_32_LITTLE) || defined(HAVE_TARGET_32_BIG)
1030 Sized_symbol
<32>* ssym
= symtab
->get_sized_symbol
<32>(this->sym_
);
1031 ssym
->set_value(val
);
1036 else if (parameters
->target().get_size() == 64)
1038 #if defined(HAVE_TARGET_64_LITTLE) || defined(HAVE_TARGET_64_BIG)
1039 Sized_symbol
<64>* ssym
= symtab
->get_sized_symbol
<64>(this->sym_
);
1040 ssym
->set_value(val
);
1047 if (val_section
!= NULL
)
1048 this->sym_
->set_output_section(val_section
);
1051 // Print for debugging.
1054 Symbol_assignment::print(FILE* f
) const
1056 if (this->provide_
&& this->hidden_
)
1057 fprintf(f
, "PROVIDE_HIDDEN(");
1058 else if (this->provide_
)
1059 fprintf(f
, "PROVIDE(");
1060 else if (this->hidden_
)
1063 fprintf(f
, "%s = ", this->name_
.c_str());
1064 this->val_
->print(f
);
1066 if (this->provide_
|| this->hidden_
)
1072 // Class Script_assertion.
1074 // Check the assertion.
1077 Script_assertion::check(const Symbol_table
* symtab
, const Layout
* layout
)
1079 if (!this->check_
->eval(symtab
, layout
, true))
1080 gold_error("%s", this->message_
.c_str());
1083 // Print for debugging.
1086 Script_assertion::print(FILE* f
) const
1088 fprintf(f
, "ASSERT(");
1089 this->check_
->print(f
);
1090 fprintf(f
, ", \"%s\")\n", this->message_
.c_str());
1093 // Class Script_options.
1095 Script_options::Script_options()
1096 : entry_(), symbol_assignments_(), symbol_definitions_(),
1097 symbol_references_(), version_script_info_(), script_sections_()
1101 // Returns true if NAME is on the list of symbol assignments waiting
1105 Script_options::is_pending_assignment(const char* name
)
1107 for (Symbol_assignments::iterator p
= this->symbol_assignments_
.begin();
1108 p
!= this->symbol_assignments_
.end();
1110 if ((*p
)->name() == name
)
1115 // Add a symbol to be defined.
1118 Script_options::add_symbol_assignment(const char* name
, size_t length
,
1119 bool is_defsym
, Expression
* value
,
1120 bool provide
, bool hidden
)
1122 if (length
!= 1 || name
[0] != '.')
1124 if (this->script_sections_
.in_sections_clause())
1126 gold_assert(!is_defsym
);
1127 this->script_sections_
.add_symbol_assignment(name
, length
, value
,
1132 Symbol_assignment
* p
= new Symbol_assignment(name
, length
, is_defsym
,
1133 value
, provide
, hidden
);
1134 this->symbol_assignments_
.push_back(p
);
1139 std::string
n(name
, length
);
1140 this->symbol_definitions_
.insert(n
);
1141 this->symbol_references_
.erase(n
);
1146 if (provide
|| hidden
)
1147 gold_error(_("invalid use of PROVIDE for dot symbol"));
1149 // The GNU linker permits assignments to dot outside of SECTIONS
1150 // clauses and treats them as occurring inside, so we don't
1151 // check in_sections_clause here.
1152 this->script_sections_
.add_dot_assignment(value
);
1156 // Add a reference to a symbol.
1159 Script_options::add_symbol_reference(const char* name
, size_t length
)
1161 if (length
!= 1 || name
[0] != '.')
1163 std::string
n(name
, length
);
1164 if (this->symbol_definitions_
.find(n
) == this->symbol_definitions_
.end())
1165 this->symbol_references_
.insert(n
);
1169 // Add an assertion.
1172 Script_options::add_assertion(Expression
* check
, const char* message
,
1175 if (this->script_sections_
.in_sections_clause())
1176 this->script_sections_
.add_assertion(check
, message
, messagelen
);
1179 Script_assertion
* p
= new Script_assertion(check
, message
, messagelen
);
1180 this->assertions_
.push_back(p
);
1184 // Create sections required by any linker scripts.
1187 Script_options::create_script_sections(Layout
* layout
)
1189 if (this->saw_sections_clause())
1190 this->script_sections_
.create_sections(layout
);
1193 // Add any symbols we are defining to the symbol table.
1196 Script_options::add_symbols_to_table(Symbol_table
* symtab
)
1198 for (Symbol_assignments::iterator p
= this->symbol_assignments_
.begin();
1199 p
!= this->symbol_assignments_
.end();
1201 (*p
)->add_to_table(symtab
);
1202 this->script_sections_
.add_symbols_to_table(symtab
);
1205 // Finalize symbol values. Also check assertions.
1208 Script_options::finalize_symbols(Symbol_table
* symtab
, const Layout
* layout
)
1210 // We finalize the symbols defined in SECTIONS first, because they
1211 // are the ones which may have changed. This way if symbol outside
1212 // SECTIONS are defined in terms of symbols inside SECTIONS, they
1213 // will get the right value.
1214 this->script_sections_
.finalize_symbols(symtab
, layout
);
1216 for (Symbol_assignments::iterator p
= this->symbol_assignments_
.begin();
1217 p
!= this->symbol_assignments_
.end();
1219 (*p
)->finalize(symtab
, layout
);
1221 for (Assertions::iterator p
= this->assertions_
.begin();
1222 p
!= this->assertions_
.end();
1224 (*p
)->check(symtab
, layout
);
1227 // Set section addresses. We set all the symbols which have absolute
1228 // values. Then we let the SECTIONS clause do its thing. This
1229 // returns the segment which holds the file header and segment
1233 Script_options::set_section_addresses(Symbol_table
* symtab
, Layout
* layout
)
1235 for (Symbol_assignments::iterator p
= this->symbol_assignments_
.begin();
1236 p
!= this->symbol_assignments_
.end();
1238 (*p
)->set_if_absolute(symtab
, layout
, false, 0, NULL
);
1240 return this->script_sections_
.set_section_addresses(symtab
, layout
);
1243 // This class holds data passed through the parser to the lexer and to
1244 // the parser support functions. This avoids global variables. We
1245 // can't use global variables because we need not be called by a
1246 // singleton thread.
1248 class Parser_closure
1251 Parser_closure(const char* filename
,
1252 const Position_dependent_options
& posdep_options
,
1253 bool parsing_defsym
, bool in_group
, bool is_in_sysroot
,
1254 Command_line
* command_line
,
1255 Script_options
* script_options
,
1257 bool skip_on_incompatible_target
,
1258 Script_info
* script_info
)
1259 : filename_(filename
), posdep_options_(posdep_options
),
1260 parsing_defsym_(parsing_defsym
), in_group_(in_group
),
1261 is_in_sysroot_(is_in_sysroot
),
1262 skip_on_incompatible_target_(skip_on_incompatible_target
),
1263 found_incompatible_target_(false),
1264 command_line_(command_line
), script_options_(script_options
),
1265 version_script_info_(script_options
->version_script_info()),
1266 lex_(lex
), lineno_(0), charpos_(0), lex_mode_stack_(), inputs_(NULL
),
1267 script_info_(script_info
)
1269 // We start out processing C symbols in the default lex mode.
1270 this->language_stack_
.push_back(Version_script_info::LANGUAGE_C
);
1271 this->lex_mode_stack_
.push_back(lex
->mode());
1274 // Return the file name.
1277 { return this->filename_
; }
1279 // Return the position dependent options. The caller may modify
1281 Position_dependent_options
&
1282 position_dependent_options()
1283 { return this->posdep_options_
; }
1285 // Whether we are parsing a --defsym.
1287 parsing_defsym() const
1288 { return this->parsing_defsym_
; }
1290 // Return whether this script is being run in a group.
1293 { return this->in_group_
; }
1295 // Return whether this script was found using a directory in the
1298 is_in_sysroot() const
1299 { return this->is_in_sysroot_
; }
1301 // Whether to skip to the next file with the same name if we find an
1302 // incompatible target in an OUTPUT_FORMAT statement.
1304 skip_on_incompatible_target() const
1305 { return this->skip_on_incompatible_target_
; }
1307 // Stop skipping to the next file on an incompatible target. This
1308 // is called when we make some unrevocable change to the data
1311 clear_skip_on_incompatible_target()
1312 { this->skip_on_incompatible_target_
= false; }
1314 // Whether we found an incompatible target in an OUTPUT_FORMAT
1317 found_incompatible_target() const
1318 { return this->found_incompatible_target_
; }
1320 // Note that we found an incompatible target.
1322 set_found_incompatible_target()
1323 { this->found_incompatible_target_
= true; }
1325 // Returns the Command_line structure passed in at constructor time.
1326 // This value may be NULL. The caller may modify this, which modifies
1327 // the passed-in Command_line object (not a copy).
1330 { return this->command_line_
; }
1332 // Return the options which may be set by a script.
1335 { return this->script_options_
; }
1337 // Return the object in which version script information should be stored.
1338 Version_script_info
*
1340 { return this->version_script_info_
; }
1342 // Return the next token, and advance.
1346 const Token
* token
= this->lex_
->next_token();
1347 this->lineno_
= token
->lineno();
1348 this->charpos_
= token
->charpos();
1352 // Set a new lexer mode, pushing the current one.
1354 push_lex_mode(Lex::Mode mode
)
1356 this->lex_mode_stack_
.push_back(this->lex_
->mode());
1357 this->lex_
->set_mode(mode
);
1360 // Pop the lexer mode.
1364 gold_assert(!this->lex_mode_stack_
.empty());
1365 this->lex_
->set_mode(this->lex_mode_stack_
.back());
1366 this->lex_mode_stack_
.pop_back();
1369 // Return the current lexer mode.
1372 { return this->lex_mode_stack_
.back(); }
1374 // Return the line number of the last token.
1377 { return this->lineno_
; }
1379 // Return the character position in the line of the last token.
1382 { return this->charpos_
; }
1384 // Return the list of input files, creating it if necessary. This
1385 // is a space leak--we never free the INPUTS_ pointer.
1389 if (this->inputs_
== NULL
)
1390 this->inputs_
= new Input_arguments();
1391 return this->inputs_
;
1394 // Return whether we saw any input files.
1397 { return this->inputs_
!= NULL
&& !this->inputs_
->empty(); }
1399 // Return the current language being processed in a version script
1400 // (eg, "C++"). The empty string represents unmangled C names.
1401 Version_script_info::Language
1402 get_current_language() const
1403 { return this->language_stack_
.back(); }
1405 // Push a language onto the stack when entering an extern block.
1407 push_language(Version_script_info::Language lang
)
1408 { this->language_stack_
.push_back(lang
); }
1410 // Pop a language off of the stack when exiting an extern block.
1414 gold_assert(!this->language_stack_
.empty());
1415 this->language_stack_
.pop_back();
1418 // Return a pointer to the incremental info.
1421 { return this->script_info_
; }
1424 // The name of the file we are reading.
1425 const char* filename_
;
1426 // The position dependent options.
1427 Position_dependent_options posdep_options_
;
1428 // True if we are parsing a --defsym.
1429 bool parsing_defsym_
;
1430 // Whether we are currently in a --start-group/--end-group.
1432 // Whether the script was found in a sysrooted directory.
1433 bool is_in_sysroot_
;
1434 // If this is true, then if we find an OUTPUT_FORMAT with an
1435 // incompatible target, then we tell the parser to abort so that we
1436 // can search for the next file with the same name.
1437 bool skip_on_incompatible_target_
;
1438 // True if we found an OUTPUT_FORMAT with an incompatible target.
1439 bool found_incompatible_target_
;
1440 // May be NULL if the user chooses not to pass one in.
1441 Command_line
* command_line_
;
1442 // Options which may be set from any linker script.
1443 Script_options
* script_options_
;
1444 // Information parsed from a version script.
1445 Version_script_info
* version_script_info_
;
1448 // The line number of the last token returned by next_token.
1450 // The column number of the last token returned by next_token.
1452 // A stack of lexer modes.
1453 std::vector
<Lex::Mode
> lex_mode_stack_
;
1454 // A stack of which extern/language block we're inside. Can be C++,
1455 // java, or empty for C.
1456 std::vector
<Version_script_info::Language
> language_stack_
;
1457 // New input files found to add to the link.
1458 Input_arguments
* inputs_
;
1459 // Pointer to incremental linking info.
1460 Script_info
* script_info_
;
1463 // FILE was found as an argument on the command line. Try to read it
1464 // as a script. Return true if the file was handled.
1467 read_input_script(Workqueue
* workqueue
, Symbol_table
* symtab
, Layout
* layout
,
1468 Dirsearch
* dirsearch
, int dirindex
,
1469 Input_objects
* input_objects
, Mapfile
* mapfile
,
1470 Input_group
* input_group
,
1471 const Input_argument
* input_argument
,
1472 Input_file
* input_file
, Task_token
* next_blocker
,
1473 bool* used_next_blocker
)
1475 *used_next_blocker
= false;
1477 std::string input_string
;
1478 Lex::read_file(input_file
, &input_string
);
1480 Lex
lex(input_string
.c_str(), input_string
.length(), PARSING_LINKER_SCRIPT
);
1482 Script_info
* script_info
= NULL
;
1483 if (layout
->incremental_inputs() != NULL
)
1485 const std::string
& filename
= input_file
->filename();
1486 Timespec mtime
= input_file
->file().get_mtime();
1487 unsigned int arg_serial
= input_argument
->file().arg_serial();
1488 script_info
= new Script_info(filename
);
1489 layout
->incremental_inputs()->report_script(script_info
, arg_serial
,
1493 Parser_closure
closure(input_file
->filename().c_str(),
1494 input_argument
->file().options(),
1496 input_group
!= NULL
,
1497 input_file
->is_in_sysroot(),
1499 layout
->script_options(),
1501 input_file
->will_search_for(),
1504 bool old_saw_sections_clause
=
1505 layout
->script_options()->saw_sections_clause();
1507 if (yyparse(&closure
) != 0)
1509 if (closure
.found_incompatible_target())
1511 Read_symbols::incompatible_warning(input_argument
, input_file
);
1512 Read_symbols::requeue(workqueue
, input_objects
, symtab
, layout
,
1513 dirsearch
, dirindex
, mapfile
, input_argument
,
1514 input_group
, next_blocker
);
1520 if (!old_saw_sections_clause
1521 && layout
->script_options()->saw_sections_clause()
1522 && layout
->have_added_input_section())
1523 gold_error(_("%s: SECTIONS seen after other input files; try -T/--script"),
1524 input_file
->filename().c_str());
1526 if (!closure
.saw_inputs())
1529 Task_token
* this_blocker
= NULL
;
1530 for (Input_arguments::const_iterator p
= closure
.inputs()->begin();
1531 p
!= closure
.inputs()->end();
1535 if (p
+ 1 == closure
.inputs()->end())
1539 nb
= new Task_token(true);
1542 workqueue
->queue_soon(new Read_symbols(input_objects
, symtab
,
1543 layout
, dirsearch
, 0, mapfile
, &*p
,
1544 input_group
, NULL
, this_blocker
, nb
));
1548 *used_next_blocker
= true;
1553 // Helper function for read_version_script(), read_commandline_script() and
1554 // script_include_directive(). Processes the given file in the mode indicated
1555 // by first_token and lex_mode.
1558 read_script_file(const char* filename
, Command_line
* cmdline
,
1559 Script_options
* script_options
,
1560 int first_token
, Lex::Mode lex_mode
)
1562 Dirsearch dirsearch
;
1563 std::string name
= filename
;
1565 // If filename is a relative filename, search for it manually using "." +
1566 // cmdline->options()->library_path() -- not dirsearch.
1567 if (!IS_ABSOLUTE_PATH(filename
))
1569 const General_options::Dir_list
& search_path
=
1570 cmdline
->options().library_path();
1571 name
= Dirsearch::find_file_in_dir_list(name
, search_path
, ".");
1574 // The file locking code wants to record a Task, but we haven't
1575 // started the workqueue yet. This is only for debugging purposes,
1576 // so we invent a fake value.
1577 const Task
* task
= reinterpret_cast<const Task
*>(-1);
1579 // We don't want this file to be opened in binary mode.
1580 Position_dependent_options posdep
= cmdline
->position_dependent_options();
1581 if (posdep
.format_enum() == General_options::OBJECT_FORMAT_BINARY
)
1582 posdep
.set_format_enum(General_options::OBJECT_FORMAT_ELF
);
1583 Input_file_argument
input_argument(name
.c_str(),
1584 Input_file_argument::INPUT_FILE_TYPE_FILE
,
1586 Input_file
input_file(&input_argument
);
1588 if (!input_file
.open(dirsearch
, task
, &dummy
))
1591 std::string input_string
;
1592 Lex::read_file(&input_file
, &input_string
);
1594 Lex
lex(input_string
.c_str(), input_string
.length(), first_token
);
1595 lex
.set_mode(lex_mode
);
1597 Parser_closure
closure(filename
,
1598 cmdline
->position_dependent_options(),
1599 first_token
== Lex::DYNAMIC_LIST
,
1601 input_file
.is_in_sysroot(),
1607 if (yyparse(&closure
) != 0)
1609 input_file
.file().unlock(task
);
1613 input_file
.file().unlock(task
);
1615 gold_assert(!closure
.saw_inputs());
1620 // FILENAME was found as an argument to --script (-T).
1621 // Read it as a script, and execute its contents immediately.
1624 read_commandline_script(const char* filename
, Command_line
* cmdline
)
1626 return read_script_file(filename
, cmdline
, &cmdline
->script_options(),
1627 PARSING_LINKER_SCRIPT
, Lex::LINKER_SCRIPT
);
1630 // FILENAME was found as an argument to --version-script. Read it as
1631 // a version script, and store its contents in
1632 // cmdline->script_options()->version_script_info().
1635 read_version_script(const char* filename
, Command_line
* cmdline
)
1637 return read_script_file(filename
, cmdline
, &cmdline
->script_options(),
1638 PARSING_VERSION_SCRIPT
, Lex::VERSION_SCRIPT
);
1641 // FILENAME was found as an argument to --dynamic-list. Read it as a
1642 // list of symbols, and store its contents in DYNAMIC_LIST.
1645 read_dynamic_list(const char* filename
, Command_line
* cmdline
,
1646 Script_options
* dynamic_list
)
1648 return read_script_file(filename
, cmdline
, dynamic_list
,
1649 PARSING_DYNAMIC_LIST
, Lex::DYNAMIC_LIST
);
1652 // Implement the --defsym option on the command line. Return true if
1656 Script_options::define_symbol(const char* definition
)
1658 Lex
lex(definition
, strlen(definition
), PARSING_DEFSYM
);
1659 lex
.set_mode(Lex::EXPRESSION
);
1662 Position_dependent_options posdep_options
;
1664 Parser_closure
closure("command line", posdep_options
, true,
1665 false, false, NULL
, this, &lex
, false, NULL
);
1667 if (yyparse(&closure
) != 0)
1670 gold_assert(!closure
.saw_inputs());
1675 // Print the script to F for debugging.
1678 Script_options::print(FILE* f
) const
1680 fprintf(f
, "%s: Dumping linker script\n", program_name
);
1682 if (!this->entry_
.empty())
1683 fprintf(f
, "ENTRY(%s)\n", this->entry_
.c_str());
1685 for (Symbol_assignments::const_iterator p
=
1686 this->symbol_assignments_
.begin();
1687 p
!= this->symbol_assignments_
.end();
1691 for (Assertions::const_iterator p
= this->assertions_
.begin();
1692 p
!= this->assertions_
.end();
1696 this->script_sections_
.print(f
);
1698 this->version_script_info_
.print(f
);
1701 // Manage mapping from keywords to the codes expected by the bison
1702 // parser. We construct one global object for each lex mode with
1705 class Keyword_to_parsecode
1708 // The structure which maps keywords to parsecodes.
1709 struct Keyword_parsecode
1712 const char* keyword
;
1713 // Corresponding parsecode.
1717 Keyword_to_parsecode(const Keyword_parsecode
* keywords
,
1719 : keyword_parsecodes_(keywords
), keyword_count_(keyword_count
)
1722 // Return the parsecode corresponding KEYWORD, or 0 if it is not a
1725 keyword_to_parsecode(const char* keyword
, size_t len
) const;
1728 const Keyword_parsecode
* keyword_parsecodes_
;
1729 const int keyword_count_
;
1732 // Mapping from keyword string to keyword parsecode. This array must
1733 // be kept in sorted order. Parsecodes are looked up using bsearch.
1734 // This array must correspond to the list of parsecodes in yyscript.y.
1736 static const Keyword_to_parsecode::Keyword_parsecode
1737 script_keyword_parsecodes
[] =
1739 { "ABSOLUTE", ABSOLUTE
},
1741 { "ALIGN", ALIGN_K
},
1742 { "ALIGNOF", ALIGNOF
},
1743 { "ASSERT", ASSERT_K
},
1744 { "AS_NEEDED", AS_NEEDED
},
1749 { "CONSTANT", CONSTANT
},
1750 { "CONSTRUCTORS", CONSTRUCTORS
},
1752 { "CREATE_OBJECT_SYMBOLS", CREATE_OBJECT_SYMBOLS
},
1753 { "DATA_SEGMENT_ALIGN", DATA_SEGMENT_ALIGN
},
1754 { "DATA_SEGMENT_END", DATA_SEGMENT_END
},
1755 { "DATA_SEGMENT_RELRO_END", DATA_SEGMENT_RELRO_END
},
1756 { "DEFINED", DEFINED
},
1759 { "EXCLUDE_FILE", EXCLUDE_FILE
},
1760 { "EXTERN", EXTERN
},
1763 { "FORCE_COMMON_ALLOCATION", FORCE_COMMON_ALLOCATION
},
1765 { "HIDDEN", HIDDEN
},
1767 { "INCLUDE", INCLUDE
},
1769 { "INHIBIT_COMMON_ALLOCATION", INHIBIT_COMMON_ALLOCATION
},
1772 { "LENGTH", LENGTH
},
1773 { "LOADADDR", LOADADDR
},
1777 { "MEMORY", MEMORY
},
1780 { "NOCROSSREFS", NOCROSSREFS
},
1781 { "NOFLOAT", NOFLOAT
},
1782 { "NOLOAD", NOLOAD
},
1783 { "ONLY_IF_RO", ONLY_IF_RO
},
1784 { "ONLY_IF_RW", ONLY_IF_RW
},
1785 { "OPTION", OPTION
},
1786 { "ORIGIN", ORIGIN
},
1787 { "OUTPUT", OUTPUT
},
1788 { "OUTPUT_ARCH", OUTPUT_ARCH
},
1789 { "OUTPUT_FORMAT", OUTPUT_FORMAT
},
1790 { "OVERLAY", OVERLAY
},
1792 { "PROVIDE", PROVIDE
},
1793 { "PROVIDE_HIDDEN", PROVIDE_HIDDEN
},
1795 { "SEARCH_DIR", SEARCH_DIR
},
1796 { "SECTIONS", SECTIONS
},
1797 { "SEGMENT_START", SEGMENT_START
},
1799 { "SIZEOF", SIZEOF
},
1800 { "SIZEOF_HEADERS", SIZEOF_HEADERS
},
1801 { "SORT", SORT_BY_NAME
},
1802 { "SORT_BY_ALIGNMENT", SORT_BY_ALIGNMENT
},
1803 { "SORT_BY_INIT_PRIORITY", SORT_BY_INIT_PRIORITY
},
1804 { "SORT_BY_NAME", SORT_BY_NAME
},
1805 { "SPECIAL", SPECIAL
},
1807 { "STARTUP", STARTUP
},
1808 { "SUBALIGN", SUBALIGN
},
1809 { "SYSLIB", SYSLIB
},
1810 { "TARGET", TARGET_K
},
1811 { "TRUNCATE", TRUNCATE
},
1812 { "VERSION", VERSIONK
},
1813 { "global", GLOBAL
},
1819 { "sizeof_headers", SIZEOF_HEADERS
},
1822 static const Keyword_to_parsecode
1823 script_keywords(&script_keyword_parsecodes
[0],
1824 (sizeof(script_keyword_parsecodes
)
1825 / sizeof(script_keyword_parsecodes
[0])));
1827 static const Keyword_to_parsecode::Keyword_parsecode
1828 version_script_keyword_parsecodes
[] =
1830 { "extern", EXTERN
},
1831 { "global", GLOBAL
},
1835 static const Keyword_to_parsecode
1836 version_script_keywords(&version_script_keyword_parsecodes
[0],
1837 (sizeof(version_script_keyword_parsecodes
)
1838 / sizeof(version_script_keyword_parsecodes
[0])));
1840 static const Keyword_to_parsecode::Keyword_parsecode
1841 dynamic_list_keyword_parsecodes
[] =
1843 { "extern", EXTERN
},
1846 static const Keyword_to_parsecode
1847 dynamic_list_keywords(&dynamic_list_keyword_parsecodes
[0],
1848 (sizeof(dynamic_list_keyword_parsecodes
)
1849 / sizeof(dynamic_list_keyword_parsecodes
[0])));
1853 // Comparison function passed to bsearch.
1865 ktt_compare(const void* keyv
, const void* kttv
)
1867 const Ktt_key
* key
= static_cast<const Ktt_key
*>(keyv
);
1868 const Keyword_to_parsecode::Keyword_parsecode
* ktt
=
1869 static_cast<const Keyword_to_parsecode::Keyword_parsecode
*>(kttv
);
1870 int i
= strncmp(key
->str
, ktt
->keyword
, key
->len
);
1873 if (ktt
->keyword
[key
->len
] != '\0')
1878 } // End extern "C".
1881 Keyword_to_parsecode::keyword_to_parsecode(const char* keyword
,
1887 void* kttv
= bsearch(&key
,
1888 this->keyword_parsecodes_
,
1889 this->keyword_count_
,
1890 sizeof(this->keyword_parsecodes_
[0]),
1894 Keyword_parsecode
* ktt
= static_cast<Keyword_parsecode
*>(kttv
);
1895 return ktt
->parsecode
;
1898 // The following structs are used within the VersionInfo class as well
1899 // as in the bison helper functions. They store the information
1900 // parsed from the version script.
1902 // A single version expression.
1903 // For example, pattern="std::map*" and language="C++".
1904 struct Version_expression
1906 Version_expression(const std::string
& a_pattern
,
1907 Version_script_info::Language a_language
,
1909 : pattern(a_pattern
), language(a_language
), exact_match(a_exact_match
),
1910 was_matched_by_symbol(false)
1913 std::string pattern
;
1914 Version_script_info::Language language
;
1915 // If false, we use glob() to match pattern. If true, we use strcmp().
1917 // True if --no-undefined-version is in effect and we found this
1918 // version in get_symbol_version. We use mutable because this
1919 // struct is generally not modifiable after it has been created.
1920 mutable bool was_matched_by_symbol
;
1923 // A list of expressions.
1924 struct Version_expression_list
1926 std::vector
<struct Version_expression
> expressions
;
1929 // A list of which versions upon which another version depends.
1930 // Strings should be from the Stringpool.
1931 struct Version_dependency_list
1933 std::vector
<std::string
> dependencies
;
1936 // The total definition of a version. It includes the tag for the
1937 // version, its global and local expressions, and any dependencies.
1941 : tag(), global(NULL
), local(NULL
), dependencies(NULL
)
1945 const struct Version_expression_list
* global
;
1946 const struct Version_expression_list
* local
;
1947 const struct Version_dependency_list
* dependencies
;
1950 // Helper class that calls cplus_demangle when needed and takes care of freeing
1953 class Lazy_demangler
1956 Lazy_demangler(const char* symbol
, int options
)
1957 : symbol_(symbol
), options_(options
), demangled_(NULL
), did_demangle_(false)
1961 { free(this->demangled_
); }
1963 // Return the demangled name. The actual demangling happens on the first call,
1964 // and the result is later cached.
1969 // The symbol to demangle.
1970 const char* symbol_
;
1971 // Option flags to pass to cplus_demagle.
1973 // The cached demangled value, or NULL if demangling didn't happen yet or
1976 // Whether we already called cplus_demangle
1980 // Return the demangled name. The actual demangling happens on the first call,
1981 // and the result is later cached. Returns NULL if the symbol cannot be
1985 Lazy_demangler::get()
1987 if (!this->did_demangle_
)
1989 this->demangled_
= cplus_demangle(this->symbol_
, this->options_
);
1990 this->did_demangle_
= true;
1992 return this->demangled_
;
1995 // Class Version_script_info.
1997 Version_script_info::Version_script_info()
1998 : dependency_lists_(), expression_lists_(), version_trees_(), globs_(),
1999 default_version_(NULL
), default_is_global_(false), is_finalized_(false)
2001 for (int i
= 0; i
< LANGUAGE_COUNT
; ++i
)
2002 this->exact_
[i
] = NULL
;
2005 Version_script_info::~Version_script_info()
2009 // Forget all the known version script information.
2012 Version_script_info::clear()
2014 for (size_t k
= 0; k
< this->dependency_lists_
.size(); ++k
)
2015 delete this->dependency_lists_
[k
];
2016 this->dependency_lists_
.clear();
2017 for (size_t k
= 0; k
< this->version_trees_
.size(); ++k
)
2018 delete this->version_trees_
[k
];
2019 this->version_trees_
.clear();
2020 for (size_t k
= 0; k
< this->expression_lists_
.size(); ++k
)
2021 delete this->expression_lists_
[k
];
2022 this->expression_lists_
.clear();
2025 // Finalize the version script information.
2028 Version_script_info::finalize()
2030 if (!this->is_finalized_
)
2032 this->build_lookup_tables();
2033 this->is_finalized_
= true;
2037 // Return all the versions.
2039 std::vector
<std::string
>
2040 Version_script_info::get_versions() const
2042 std::vector
<std::string
> ret
;
2043 for (size_t j
= 0; j
< this->version_trees_
.size(); ++j
)
2044 if (!this->version_trees_
[j
]->tag
.empty())
2045 ret
.push_back(this->version_trees_
[j
]->tag
);
2049 // Return the dependencies of VERSION.
2051 std::vector
<std::string
>
2052 Version_script_info::get_dependencies(const char* version
) const
2054 std::vector
<std::string
> ret
;
2055 for (size_t j
= 0; j
< this->version_trees_
.size(); ++j
)
2056 if (this->version_trees_
[j
]->tag
== version
)
2058 const struct Version_dependency_list
* deps
=
2059 this->version_trees_
[j
]->dependencies
;
2061 for (size_t k
= 0; k
< deps
->dependencies
.size(); ++k
)
2062 ret
.push_back(deps
->dependencies
[k
]);
2068 // A version script essentially maps a symbol name to a version tag
2069 // and an indication of whether symbol is global or local within that
2070 // version tag. Each symbol maps to at most one version tag.
2071 // Unfortunately, in practice, version scripts are ambiguous, and list
2072 // symbols multiple times. Thus, we have to document the matching
2075 // This is a description of what the GNU linker does as of 2010-01-11.
2076 // It walks through the version tags in the order in which they appear
2077 // in the version script. For each tag, it first walks through the
2078 // global patterns for that tag, then the local patterns. When
2079 // looking at a single pattern, it first applies any language specific
2080 // demangling as specified for the pattern, and then matches the
2081 // resulting symbol name to the pattern. If it finds an exact match
2082 // for a literal pattern (a pattern enclosed in quotes or with no
2083 // wildcard characters), then that is the match that it uses. If
2084 // finds a match with a wildcard pattern, then it saves it and
2085 // continues searching. Wildcard patterns that are exactly "*" are
2086 // saved separately.
2088 // If no exact match with a literal pattern is ever found, then if a
2089 // wildcard match with a global pattern was found it is used,
2090 // otherwise if a wildcard match with a local pattern was found it is
2093 // This is the result:
2094 // * If there is an exact match, then we use the first tag in the
2095 // version script where it matches.
2096 // + If the exact match in that tag is global, it is used.
2097 // + Otherwise the exact match in that tag is local, and is used.
2098 // * Otherwise, if there is any match with a global wildcard pattern:
2099 // + If there is any match with a wildcard pattern which is not
2100 // "*", then we use the tag in which the *last* such pattern
2102 // + Otherwise, we matched "*". If there is no match with a local
2103 // wildcard pattern which is not "*", then we use the *last*
2104 // match with a global "*". Otherwise, continue.
2105 // * Otherwise, if there is any match with a local wildcard pattern:
2106 // + If there is any match with a wildcard pattern which is not
2107 // "*", then we use the tag in which the *last* such pattern
2109 // + Otherwise, we matched "*", and we use the tag in which the
2110 // *last* such match occurred.
2112 // There is an additional wrinkle. When the GNU linker finds a symbol
2113 // with a version defined in an object file due to a .symver
2114 // directive, it looks up that symbol name in that version tag. If it
2115 // finds it, it matches the symbol name against the patterns for that
2116 // version. If there is no match with a global pattern, but there is
2117 // a match with a local pattern, then the GNU linker marks the symbol
2120 // We want gold to be generally compatible, but we also want gold to
2121 // be fast. These are the rules that gold implements:
2122 // * If there is an exact match for the mangled name, we use it.
2123 // + If there is more than one exact match, we give a warning, and
2124 // we use the first tag in the script which matches.
2125 // + If a symbol has an exact match as both global and local for
2126 // the same version tag, we give an error.
2127 // * Otherwise, we look for an extern C++ or an extern Java exact
2128 // match. If we find an exact match, we use it.
2129 // + If there is more than one exact match, we give a warning, and
2130 // we use the first tag in the script which matches.
2131 // + If a symbol has an exact match as both global and local for
2132 // the same version tag, we give an error.
2133 // * Otherwise, we look through the wildcard patterns, ignoring "*"
2134 // patterns. We look through the version tags in reverse order.
2135 // For each version tag, we look through the global patterns and
2136 // then the local patterns. We use the first match we find (i.e.,
2137 // the last matching version tag in the file).
2138 // * Otherwise, we use the "*" pattern if there is one. We give an
2139 // error if there are multiple "*" patterns.
2141 // At least for now, gold does not look up the version tag for a
2142 // symbol version found in an object file to see if it should be
2143 // forced local. There are other ways to force a symbol to be local,
2144 // and I don't understand why this one is useful.
2146 // Build a set of fast lookup tables for a version script.
2149 Version_script_info::build_lookup_tables()
2151 size_t size
= this->version_trees_
.size();
2152 for (size_t j
= 0; j
< size
; ++j
)
2154 const Version_tree
* v
= this->version_trees_
[j
];
2155 this->build_expression_list_lookup(v
->local
, v
, false);
2156 this->build_expression_list_lookup(v
->global
, v
, true);
2160 // If a pattern has backlashes but no unquoted wildcard characters,
2161 // then we apply backslash unquoting and look for an exact match.
2162 // Otherwise we treat it as a wildcard pattern. This function returns
2163 // true for a wildcard pattern. Otherwise, it does backslash
2164 // unquoting on *PATTERN and returns false. If this returns true,
2165 // *PATTERN may have been partially unquoted.
2168 Version_script_info::unquote(std::string
* pattern
) const
2170 bool saw_backslash
= false;
2171 size_t len
= pattern
->length();
2173 for (size_t i
= 0; i
< len
; ++i
)
2176 saw_backslash
= false;
2179 switch ((*pattern
)[i
])
2181 case '?': case '[': case '*':
2184 saw_backslash
= true;
2192 (*pattern
)[j
] = (*pattern
)[i
];
2198 // Add an exact match for MATCH to *PE. The result of the match is
2202 Version_script_info::add_exact_match(const std::string
& match
,
2203 const Version_tree
* v
, bool is_global
,
2204 const Version_expression
* ve
,
2207 std::pair
<Exact::iterator
, bool> ins
=
2208 pe
->insert(std::make_pair(match
, Version_tree_match(v
, is_global
, ve
)));
2211 // This is the first time we have seen this match.
2215 Version_tree_match
& vtm(ins
.first
->second
);
2216 if (vtm
.real
->tag
!= v
->tag
)
2218 // This is an ambiguous match. We still return the
2219 // first version that we found in the script, but we
2220 // record the new version to issue a warning if we
2221 // wind up looking up this symbol.
2222 if (vtm
.ambiguous
== NULL
)
2225 else if (is_global
!= vtm
.is_global
)
2227 // We have a match for both the global and local entries for a
2228 // version tag. That's got to be wrong.
2229 gold_error(_("'%s' appears as both a global and a local symbol "
2230 "for version '%s' in script"),
2231 match
.c_str(), v
->tag
.c_str());
2235 // Build fast lookup information for EXPLIST and store it in LOOKUP.
2236 // All matches go to V, and IS_GLOBAL is true if they are global
2240 Version_script_info::build_expression_list_lookup(
2241 const Version_expression_list
* explist
,
2242 const Version_tree
* v
,
2245 if (explist
== NULL
)
2247 size_t size
= explist
->expressions
.size();
2248 for (size_t i
= 0; i
< size
; ++i
)
2250 const Version_expression
& exp(explist
->expressions
[i
]);
2252 if (exp
.pattern
.length() == 1 && exp
.pattern
[0] == '*')
2254 if (this->default_version_
!= NULL
2255 && this->default_version_
->tag
!= v
->tag
)
2256 gold_warning(_("wildcard match appears in both version '%s' "
2257 "and '%s' in script"),
2258 this->default_version_
->tag
.c_str(), v
->tag
.c_str());
2259 else if (this->default_version_
!= NULL
2260 && this->default_is_global_
!= is_global
)
2261 gold_error(_("wildcard match appears as both global and local "
2262 "in version '%s' in script"),
2264 this->default_version_
= v
;
2265 this->default_is_global_
= is_global
;
2269 std::string pattern
= exp
.pattern
;
2270 if (!exp
.exact_match
)
2272 if (this->unquote(&pattern
))
2274 this->globs_
.push_back(Glob(&exp
, v
, is_global
));
2279 if (this->exact_
[exp
.language
] == NULL
)
2280 this->exact_
[exp
.language
] = new Exact();
2281 this->add_exact_match(pattern
, v
, is_global
, &exp
,
2282 this->exact_
[exp
.language
]);
2286 // Return the name to match given a name, a language code, and two
2290 Version_script_info::get_name_to_match(const char* name
,
2292 Lazy_demangler
* cpp_demangler
,
2293 Lazy_demangler
* java_demangler
) const
2300 return cpp_demangler
->get();
2302 return java_demangler
->get();
2308 // Look up SYMBOL_NAME in the list of versions. Return true if the
2309 // symbol is found, false if not. If the symbol is found, then if
2310 // PVERSION is not NULL, set *PVERSION to the version tag, and if
2311 // P_IS_GLOBAL is not NULL, set *P_IS_GLOBAL according to whether the
2312 // symbol is global or not.
2315 Version_script_info::get_symbol_version(const char* symbol_name
,
2316 std::string
* pversion
,
2317 bool* p_is_global
) const
2319 Lazy_demangler
cpp_demangled_name(symbol_name
, DMGL_ANSI
| DMGL_PARAMS
);
2320 Lazy_demangler
java_demangled_name(symbol_name
,
2321 DMGL_ANSI
| DMGL_PARAMS
| DMGL_JAVA
);
2323 gold_assert(this->is_finalized_
);
2324 for (int i
= 0; i
< LANGUAGE_COUNT
; ++i
)
2326 Exact
* exact
= this->exact_
[i
];
2330 const char* name_to_match
= this->get_name_to_match(symbol_name
, i
,
2331 &cpp_demangled_name
,
2332 &java_demangled_name
);
2333 if (name_to_match
== NULL
)
2335 // If the name can not be demangled, the GNU linker goes
2336 // ahead and tries to match it anyhow. That does not
2337 // make sense to me and I have not implemented it.
2341 Exact::const_iterator pe
= exact
->find(name_to_match
);
2342 if (pe
!= exact
->end())
2344 const Version_tree_match
& vtm(pe
->second
);
2345 if (vtm
.ambiguous
!= NULL
)
2346 gold_warning(_("using '%s' as version for '%s' which is also "
2347 "named in version '%s' in script"),
2348 vtm
.real
->tag
.c_str(), name_to_match
,
2349 vtm
.ambiguous
->tag
.c_str());
2351 if (pversion
!= NULL
)
2352 *pversion
= vtm
.real
->tag
;
2353 if (p_is_global
!= NULL
)
2354 *p_is_global
= vtm
.is_global
;
2356 // If we are using --no-undefined-version, and this is a
2357 // global symbol, we have to record that we have found this
2358 // symbol, so that we don't warn about it. We have to do
2359 // this now, because otherwise we have no way to get from a
2360 // non-C language back to the demangled name that we
2362 if (p_is_global
!= NULL
&& vtm
.is_global
)
2363 vtm
.expression
->was_matched_by_symbol
= true;
2369 // Look through the glob patterns in reverse order.
2371 for (Globs::const_reverse_iterator p
= this->globs_
.rbegin();
2372 p
!= this->globs_
.rend();
2375 int language
= p
->expression
->language
;
2376 const char* name_to_match
= this->get_name_to_match(symbol_name
,
2378 &cpp_demangled_name
,
2379 &java_demangled_name
);
2380 if (name_to_match
== NULL
)
2383 if (fnmatch(p
->expression
->pattern
.c_str(), name_to_match
,
2386 if (pversion
!= NULL
)
2387 *pversion
= p
->version
->tag
;
2388 if (p_is_global
!= NULL
)
2389 *p_is_global
= p
->is_global
;
2394 // Finally, there may be a wildcard.
2395 if (this->default_version_
!= NULL
)
2397 if (pversion
!= NULL
)
2398 *pversion
= this->default_version_
->tag
;
2399 if (p_is_global
!= NULL
)
2400 *p_is_global
= this->default_is_global_
;
2407 // Give an error if any exact symbol names (not wildcards) appear in a
2408 // version script, but there is no such symbol.
2411 Version_script_info::check_unmatched_names(const Symbol_table
* symtab
) const
2413 for (size_t i
= 0; i
< this->version_trees_
.size(); ++i
)
2415 const Version_tree
* vt
= this->version_trees_
[i
];
2416 if (vt
->global
== NULL
)
2418 for (size_t j
= 0; j
< vt
->global
->expressions
.size(); ++j
)
2420 const Version_expression
& expression(vt
->global
->expressions
[j
]);
2422 // Ignore cases where we used the version because we saw a
2423 // symbol that we looked up. Note that
2424 // WAS_MATCHED_BY_SYMBOL will be true even if the symbol was
2425 // not a definition. That's OK as in that case we most
2426 // likely gave an undefined symbol error anyhow.
2427 if (expression
.was_matched_by_symbol
)
2430 // Just ignore names which are in languages other than C.
2431 // We have no way to look them up in the symbol table.
2432 if (expression
.language
!= LANGUAGE_C
)
2435 // Remove backslash quoting, and ignore wildcard patterns.
2436 std::string pattern
= expression
.pattern
;
2437 if (!expression
.exact_match
)
2439 if (this->unquote(&pattern
))
2443 if (symtab
->lookup(pattern
.c_str(), vt
->tag
.c_str()) == NULL
)
2444 gold_error(_("version script assignment of %s to symbol %s "
2445 "failed: symbol not defined"),
2446 vt
->tag
.c_str(), pattern
.c_str());
2451 struct Version_dependency_list
*
2452 Version_script_info::allocate_dependency_list()
2454 dependency_lists_
.push_back(new Version_dependency_list
);
2455 return dependency_lists_
.back();
2458 struct Version_expression_list
*
2459 Version_script_info::allocate_expression_list()
2461 expression_lists_
.push_back(new Version_expression_list
);
2462 return expression_lists_
.back();
2465 struct Version_tree
*
2466 Version_script_info::allocate_version_tree()
2468 version_trees_
.push_back(new Version_tree
);
2469 return version_trees_
.back();
2472 // Print for debugging.
2475 Version_script_info::print(FILE* f
) const
2480 fprintf(f
, "VERSION {");
2482 for (size_t i
= 0; i
< this->version_trees_
.size(); ++i
)
2484 const Version_tree
* vt
= this->version_trees_
[i
];
2486 if (vt
->tag
.empty())
2489 fprintf(f
, " %s {\n", vt
->tag
.c_str());
2491 if (vt
->global
!= NULL
)
2493 fprintf(f
, " global :\n");
2494 this->print_expression_list(f
, vt
->global
);
2497 if (vt
->local
!= NULL
)
2499 fprintf(f
, " local :\n");
2500 this->print_expression_list(f
, vt
->local
);
2504 if (vt
->dependencies
!= NULL
)
2506 const Version_dependency_list
* deps
= vt
->dependencies
;
2507 for (size_t j
= 0; j
< deps
->dependencies
.size(); ++j
)
2509 if (j
< deps
->dependencies
.size() - 1)
2511 fprintf(f
, " %s", deps
->dependencies
[j
].c_str());
2521 Version_script_info::print_expression_list(
2523 const Version_expression_list
* vel
) const
2525 Version_script_info::Language current_language
= LANGUAGE_C
;
2526 for (size_t i
= 0; i
< vel
->expressions
.size(); ++i
)
2528 const Version_expression
& ve(vel
->expressions
[i
]);
2530 if (ve
.language
!= current_language
)
2532 if (current_language
!= LANGUAGE_C
)
2534 switch (ve
.language
)
2539 fprintf(f
, " extern \"C++\" {\n");
2542 fprintf(f
, " extern \"Java\" {\n");
2547 current_language
= ve
.language
;
2551 if (current_language
!= LANGUAGE_C
)
2556 fprintf(f
, "%s", ve
.pattern
.c_str());
2563 if (current_language
!= LANGUAGE_C
)
2567 } // End namespace gold.
2569 // The remaining functions are extern "C", so it's clearer to not put
2570 // them in namespace gold.
2572 using namespace gold
;
2574 // This function is called by the bison parser to return the next
2578 yylex(YYSTYPE
* lvalp
, void* closurev
)
2580 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2581 const Token
* token
= closure
->next_token();
2582 switch (token
->classification())
2587 case Token::TOKEN_INVALID
:
2588 yyerror(closurev
, "invalid character");
2591 case Token::TOKEN_EOF
:
2594 case Token::TOKEN_STRING
:
2596 // This is either a keyword or a STRING.
2598 const char* str
= token
->string_value(&len
);
2600 switch (closure
->lex_mode())
2602 case Lex::LINKER_SCRIPT
:
2603 parsecode
= script_keywords
.keyword_to_parsecode(str
, len
);
2605 case Lex::VERSION_SCRIPT
:
2606 parsecode
= version_script_keywords
.keyword_to_parsecode(str
, len
);
2608 case Lex::DYNAMIC_LIST
:
2609 parsecode
= dynamic_list_keywords
.keyword_to_parsecode(str
, len
);
2616 lvalp
->string
.value
= str
;
2617 lvalp
->string
.length
= len
;
2621 case Token::TOKEN_QUOTED_STRING
:
2622 lvalp
->string
.value
= token
->string_value(&lvalp
->string
.length
);
2623 return QUOTED_STRING
;
2625 case Token::TOKEN_OPERATOR
:
2626 return token
->operator_value();
2628 case Token::TOKEN_INTEGER
:
2629 lvalp
->integer
= token
->integer_value();
2634 // This function is called by the bison parser to report an error.
2637 yyerror(void* closurev
, const char* message
)
2639 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2640 gold_error(_("%s:%d:%d: %s"), closure
->filename(), closure
->lineno(),
2641 closure
->charpos(), message
);
2644 // Called by the bison parser to add an external symbol to the link.
2647 script_add_extern(void* closurev
, const char* name
, size_t length
)
2649 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2650 closure
->script_options()->add_symbol_reference(name
, length
);
2653 // Called by the bison parser to add a file to the link.
2656 script_add_file(void* closurev
, const char* name
, size_t length
)
2658 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2660 // If this is an absolute path, and we found the script in the
2661 // sysroot, then we want to prepend the sysroot to the file name.
2662 // For example, this is how we handle a cross link to the x86_64
2663 // libc.so, which refers to /lib/libc.so.6.
2664 std::string
name_string(name
, length
);
2665 const char* extra_search_path
= ".";
2666 std::string script_directory
;
2667 if (IS_ABSOLUTE_PATH(name_string
.c_str()))
2669 if (closure
->is_in_sysroot())
2671 const std::string
& sysroot(parameters
->options().sysroot());
2672 gold_assert(!sysroot
.empty());
2673 name_string
= sysroot
+ name_string
;
2678 // In addition to checking the normal library search path, we
2679 // also want to check in the script-directory.
2680 const char* slash
= strrchr(closure
->filename(), '/');
2683 script_directory
.assign(closure
->filename(),
2684 slash
- closure
->filename() + 1);
2685 extra_search_path
= script_directory
.c_str();
2689 Input_file_argument
file(name_string
.c_str(),
2690 Input_file_argument::INPUT_FILE_TYPE_FILE
,
2691 extra_search_path
, false,
2692 closure
->position_dependent_options());
2693 Input_argument
& arg
= closure
->inputs()->add_file(file
);
2694 arg
.set_script_info(closure
->script_info());
2697 // Called by the bison parser to add a library to the link.
2700 script_add_library(void* closurev
, const char* name
, size_t length
)
2702 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2703 std::string
name_string(name
, length
);
2705 if (name_string
[0] != 'l')
2706 gold_error(_("library name must be prefixed with -l"));
2708 Input_file_argument
file(name_string
.c_str() + 1,
2709 Input_file_argument::INPUT_FILE_TYPE_LIBRARY
,
2711 closure
->position_dependent_options());
2712 Input_argument
& arg
= closure
->inputs()->add_file(file
);
2713 arg
.set_script_info(closure
->script_info());
2716 // Called by the bison parser to start a group. If we are already in
2717 // a group, that means that this script was invoked within a
2718 // --start-group --end-group sequence on the command line, or that
2719 // this script was found in a GROUP of another script. In that case,
2720 // we simply continue the existing group, rather than starting a new
2721 // one. It is possible to construct a case in which this will do
2722 // something other than what would happen if we did a recursive group,
2723 // but it's hard to imagine why the different behaviour would be
2724 // useful for a real program. Avoiding recursive groups is simpler
2725 // and more efficient.
2728 script_start_group(void* closurev
)
2730 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2731 if (!closure
->in_group())
2732 closure
->inputs()->start_group();
2735 // Called by the bison parser at the end of a group.
2738 script_end_group(void* closurev
)
2740 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2741 if (!closure
->in_group())
2742 closure
->inputs()->end_group();
2745 // Called by the bison parser to start an AS_NEEDED list.
2748 script_start_as_needed(void* closurev
)
2750 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2751 closure
->position_dependent_options().set_as_needed(true);
2754 // Called by the bison parser at the end of an AS_NEEDED list.
2757 script_end_as_needed(void* closurev
)
2759 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2760 closure
->position_dependent_options().set_as_needed(false);
2763 // Called by the bison parser to set the entry symbol.
2766 script_set_entry(void* closurev
, const char* entry
, size_t length
)
2768 // We'll parse this exactly the same as --entry=ENTRY on the commandline
2769 // TODO(csilvers): FIXME -- call set_entry directly.
2770 std::string
arg("--entry=");
2771 arg
.append(entry
, length
);
2772 script_parse_option(closurev
, arg
.c_str(), arg
.size());
2775 // Called by the bison parser to set whether to define common symbols.
2778 script_set_common_allocation(void* closurev
, int set
)
2780 const char* arg
= set
!= 0 ? "--define-common" : "--no-define-common";
2781 script_parse_option(closurev
, arg
, strlen(arg
));
2784 // Called by the bison parser to refer to a symbol.
2786 extern "C" Expression
*
2787 script_symbol(void* closurev
, const char* name
, size_t length
)
2789 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2790 if (length
!= 1 || name
[0] != '.')
2791 closure
->script_options()->add_symbol_reference(name
, length
);
2792 return script_exp_string(name
, length
);
2795 // Called by the bison parser to define a symbol.
2798 script_set_symbol(void* closurev
, const char* name
, size_t length
,
2799 Expression
* value
, int providei
, int hiddeni
)
2801 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2802 const bool provide
= providei
!= 0;
2803 const bool hidden
= hiddeni
!= 0;
2804 closure
->script_options()->add_symbol_assignment(name
, length
,
2805 closure
->parsing_defsym(),
2806 value
, provide
, hidden
);
2807 closure
->clear_skip_on_incompatible_target();
2810 // Called by the bison parser to add an assertion.
2813 script_add_assertion(void* closurev
, Expression
* check
, const char* message
,
2816 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2817 closure
->script_options()->add_assertion(check
, message
, messagelen
);
2818 closure
->clear_skip_on_incompatible_target();
2821 // Called by the bison parser to parse an OPTION.
2824 script_parse_option(void* closurev
, const char* option
, size_t length
)
2826 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2827 // We treat the option as a single command-line option, even if
2828 // it has internal whitespace.
2829 if (closure
->command_line() == NULL
)
2831 // There are some options that we could handle here--e.g.,
2832 // -lLIBRARY. Should we bother?
2833 gold_warning(_("%s:%d:%d: ignoring command OPTION; OPTION is only valid"
2834 " for scripts specified via -T/--script"),
2835 closure
->filename(), closure
->lineno(), closure
->charpos());
2839 bool past_a_double_dash_option
= false;
2840 const char* mutable_option
= strndup(option
, length
);
2841 gold_assert(mutable_option
!= NULL
);
2842 closure
->command_line()->process_one_option(1, &mutable_option
, 0,
2843 &past_a_double_dash_option
);
2844 // The General_options class will quite possibly store a pointer
2845 // into mutable_option, so we can't free it. In cases the class
2846 // does not store such a pointer, this is a memory leak. Alas. :(
2848 closure
->clear_skip_on_incompatible_target();
2851 // Called by the bison parser to handle OUTPUT_FORMAT. OUTPUT_FORMAT
2852 // takes either one or three arguments. In the three argument case,
2853 // the format depends on the endianness option, which we don't
2854 // currently support (FIXME). If we see an OUTPUT_FORMAT for the
2855 // wrong format, then we want to search for a new file. Returning 0
2856 // here will cause the parser to immediately abort.
2859 script_check_output_format(void* closurev
,
2860 const char* default_name
, size_t default_length
,
2861 const char*, size_t, const char*, size_t)
2863 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2864 std::string
name(default_name
, default_length
);
2865 Target
* target
= select_target_by_bfd_name(name
.c_str());
2866 if (target
== NULL
|| !parameters
->is_compatible_target(target
))
2868 if (closure
->skip_on_incompatible_target())
2870 closure
->set_found_incompatible_target();
2873 // FIXME: Should we warn about the unknown target?
2878 // Called by the bison parser to handle TARGET.
2881 script_set_target(void* closurev
, const char* target
, size_t len
)
2883 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2884 std::string
s(target
, len
);
2885 General_options::Object_format format_enum
;
2886 format_enum
= General_options::string_to_object_format(s
.c_str());
2887 closure
->position_dependent_options().set_format_enum(format_enum
);
2890 // Called by the bison parser to handle SEARCH_DIR. This is handled
2891 // exactly like a -L option.
2894 script_add_search_dir(void* closurev
, const char* option
, size_t length
)
2896 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2897 if (closure
->command_line() == NULL
)
2898 gold_warning(_("%s:%d:%d: ignoring SEARCH_DIR; SEARCH_DIR is only valid"
2899 " for scripts specified via -T/--script"),
2900 closure
->filename(), closure
->lineno(), closure
->charpos());
2901 else if (!closure
->command_line()->options().nostdlib())
2903 std::string s
= "-L" + std::string(option
, length
);
2904 script_parse_option(closurev
, s
.c_str(), s
.size());
2908 /* Called by the bison parser to push the lexer into expression
2912 script_push_lex_into_expression_mode(void* closurev
)
2914 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2915 closure
->push_lex_mode(Lex::EXPRESSION
);
2918 /* Called by the bison parser to push the lexer into version
2922 script_push_lex_into_version_mode(void* closurev
)
2924 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2925 if (closure
->version_script()->is_finalized())
2926 gold_error(_("%s:%d:%d: invalid use of VERSION in input file"),
2927 closure
->filename(), closure
->lineno(), closure
->charpos());
2928 closure
->push_lex_mode(Lex::VERSION_SCRIPT
);
2931 /* Called by the bison parser to pop the lexer mode. */
2934 script_pop_lex_mode(void* closurev
)
2936 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2937 closure
->pop_lex_mode();
2940 // Register an entire version node. For example:
2946 // - tag is "GLIBC_2.1"
2947 // - tree contains the information "global: foo"
2948 // - deps contains "GLIBC_2.0"
2951 script_register_vers_node(void*,
2954 struct Version_tree
* tree
,
2955 struct Version_dependency_list
* deps
)
2957 gold_assert(tree
!= NULL
);
2958 tree
->dependencies
= deps
;
2960 tree
->tag
= std::string(tag
, taglen
);
2963 // Add a dependencies to the list of existing dependencies, if any,
2964 // and return the expanded list.
2966 extern "C" struct Version_dependency_list
*
2967 script_add_vers_depend(void* closurev
,
2968 struct Version_dependency_list
* all_deps
,
2969 const char* depend_to_add
, int deplen
)
2971 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2972 if (all_deps
== NULL
)
2973 all_deps
= closure
->version_script()->allocate_dependency_list();
2974 all_deps
->dependencies
.push_back(std::string(depend_to_add
, deplen
));
2978 // Add a pattern expression to an existing list of expressions, if any.
2980 extern "C" struct Version_expression_list
*
2981 script_new_vers_pattern(void* closurev
,
2982 struct Version_expression_list
* expressions
,
2983 const char* pattern
, int patlen
, int exact_match
)
2985 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
2986 if (expressions
== NULL
)
2987 expressions
= closure
->version_script()->allocate_expression_list();
2988 expressions
->expressions
.push_back(
2989 Version_expression(std::string(pattern
, patlen
),
2990 closure
->get_current_language(),
2991 static_cast<bool>(exact_match
)));
2995 // Attaches b to the end of a, and clears b. So a = a + b and b = {}.
2997 extern "C" struct Version_expression_list
*
2998 script_merge_expressions(struct Version_expression_list
* a
,
2999 struct Version_expression_list
* b
)
3001 a
->expressions
.insert(a
->expressions
.end(),
3002 b
->expressions
.begin(), b
->expressions
.end());
3003 // We could delete b and remove it from expressions_lists_, but
3004 // that's a lot of work. This works just as well.
3005 b
->expressions
.clear();
3009 // Combine the global and local expressions into a a Version_tree.
3011 extern "C" struct Version_tree
*
3012 script_new_vers_node(void* closurev
,
3013 struct Version_expression_list
* global
,
3014 struct Version_expression_list
* local
)
3016 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
3017 Version_tree
* tree
= closure
->version_script()->allocate_version_tree();
3018 tree
->global
= global
;
3019 tree
->local
= local
;
3023 // Handle a transition in language, such as at the
3024 // start or end of 'extern "C++"'
3027 version_script_push_lang(void* closurev
, const char* lang
, int langlen
)
3029 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
3030 std::string
language(lang
, langlen
);
3031 Version_script_info::Language code
;
3032 if (language
.empty() || language
== "C")
3033 code
= Version_script_info::LANGUAGE_C
;
3034 else if (language
== "C++")
3035 code
= Version_script_info::LANGUAGE_CXX
;
3036 else if (language
== "Java")
3037 code
= Version_script_info::LANGUAGE_JAVA
;
3040 char* buf
= new char[langlen
+ 100];
3041 snprintf(buf
, langlen
+ 100,
3042 _("unrecognized version script language '%s'"),
3044 yyerror(closurev
, buf
);
3046 code
= Version_script_info::LANGUAGE_C
;
3048 closure
->push_language(code
);
3052 version_script_pop_lang(void* closurev
)
3054 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
3055 closure
->pop_language();
3058 // Called by the bison parser to start a SECTIONS clause.
3061 script_start_sections(void* closurev
)
3063 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
3064 closure
->script_options()->script_sections()->start_sections();
3065 closure
->clear_skip_on_incompatible_target();
3068 // Called by the bison parser to finish a SECTIONS clause.
3071 script_finish_sections(void* closurev
)
3073 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
3074 closure
->script_options()->script_sections()->finish_sections();
3077 // Start processing entries for an output section.
3080 script_start_output_section(void* closurev
, const char* name
, size_t namelen
,
3081 const struct Parser_output_section_header
* header
)
3083 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
3084 closure
->script_options()->script_sections()->start_output_section(name
,
3089 // Finish processing entries for an output section.
3092 script_finish_output_section(void* closurev
,
3093 const struct Parser_output_section_trailer
* trail
)
3095 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
3096 closure
->script_options()->script_sections()->finish_output_section(trail
);
3099 // Add a data item (e.g., "WORD (0)") to the current output section.
3102 script_add_data(void* closurev
, int data_token
, Expression
* val
)
3104 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
3106 bool is_signed
= true;
3128 closure
->script_options()->script_sections()->add_data(size
, is_signed
, val
);
3131 // Add a clause setting the fill value to the current output section.
3134 script_add_fill(void* closurev
, Expression
* val
)
3136 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
3137 closure
->script_options()->script_sections()->add_fill(val
);
3140 // Add a new input section specification to the current output
3144 script_add_input_section(void* closurev
,
3145 const struct Input_section_spec
* spec
,
3148 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
3149 bool keep
= keepi
!= 0;
3150 closure
->script_options()->script_sections()->add_input_section(spec
, keep
);
3153 // When we see DATA_SEGMENT_ALIGN we record that following output
3154 // sections may be relro.
3157 script_data_segment_align(void* closurev
)
3159 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
3160 if (!closure
->script_options()->saw_sections_clause())
3161 gold_error(_("%s:%d:%d: DATA_SEGMENT_ALIGN not in SECTIONS clause"),
3162 closure
->filename(), closure
->lineno(), closure
->charpos());
3164 closure
->script_options()->script_sections()->data_segment_align();
3167 // When we see DATA_SEGMENT_RELRO_END we know that all output sections
3168 // since DATA_SEGMENT_ALIGN should be relro.
3171 script_data_segment_relro_end(void* closurev
)
3173 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
3174 if (!closure
->script_options()->saw_sections_clause())
3175 gold_error(_("%s:%d:%d: DATA_SEGMENT_ALIGN not in SECTIONS clause"),
3176 closure
->filename(), closure
->lineno(), closure
->charpos());
3178 closure
->script_options()->script_sections()->data_segment_relro_end();
3181 // Create a new list of string/sort pairs.
3183 extern "C" String_sort_list_ptr
3184 script_new_string_sort_list(const struct Wildcard_section
* string_sort
)
3186 return new String_sort_list(1, *string_sort
);
3189 // Add an entry to a list of string/sort pairs. The way the parser
3190 // works permits us to simply modify the first parameter, rather than
3193 extern "C" String_sort_list_ptr
3194 script_string_sort_list_add(String_sort_list_ptr pv
,
3195 const struct Wildcard_section
* string_sort
)
3198 return script_new_string_sort_list(string_sort
);
3201 pv
->push_back(*string_sort
);
3206 // Create a new list of strings.
3208 extern "C" String_list_ptr
3209 script_new_string_list(const char* str
, size_t len
)
3211 return new String_list(1, std::string(str
, len
));
3214 // Add an element to a list of strings. The way the parser works
3215 // permits us to simply modify the first parameter, rather than copy
3218 extern "C" String_list_ptr
3219 script_string_list_push_back(String_list_ptr pv
, const char* str
, size_t len
)
3222 return script_new_string_list(str
, len
);
3225 pv
->push_back(std::string(str
, len
));
3230 // Concatenate two string lists. Either or both may be NULL. The way
3231 // the parser works permits us to modify the parameters, rather than
3234 extern "C" String_list_ptr
3235 script_string_list_append(String_list_ptr pv1
, String_list_ptr pv2
)
3241 pv1
->insert(pv1
->end(), pv2
->begin(), pv2
->end());
3245 // Add a new program header.
3248 script_add_phdr(void* closurev
, const char* name
, size_t namelen
,
3249 unsigned int type
, const Phdr_info
* info
)
3251 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
3252 bool includes_filehdr
= info
->includes_filehdr
!= 0;
3253 bool includes_phdrs
= info
->includes_phdrs
!= 0;
3254 bool is_flags_valid
= info
->is_flags_valid
!= 0;
3255 Script_sections
* ss
= closure
->script_options()->script_sections();
3256 ss
->add_phdr(name
, namelen
, type
, includes_filehdr
, includes_phdrs
,
3257 is_flags_valid
, info
->flags
, info
->load_address
);
3258 closure
->clear_skip_on_incompatible_target();
3261 // Convert a program header string to a type.
3263 #define PHDR_TYPE(NAME) { #NAME, sizeof(#NAME) - 1, elfcpp::NAME }
3270 } phdr_type_names
[] =
3274 PHDR_TYPE(PT_DYNAMIC
),
3275 PHDR_TYPE(PT_INTERP
),
3277 PHDR_TYPE(PT_SHLIB
),
3280 PHDR_TYPE(PT_GNU_EH_FRAME
),
3281 PHDR_TYPE(PT_GNU_STACK
),
3282 PHDR_TYPE(PT_GNU_RELRO
)
3285 extern "C" unsigned int
3286 script_phdr_string_to_type(void* closurev
, const char* name
, size_t namelen
)
3288 for (unsigned int i
= 0;
3289 i
< sizeof(phdr_type_names
) / sizeof(phdr_type_names
[0]);
3291 if (namelen
== phdr_type_names
[i
].namelen
3292 && strncmp(name
, phdr_type_names
[i
].name
, namelen
) == 0)
3293 return phdr_type_names
[i
].val
;
3294 yyerror(closurev
, _("unknown PHDR type (try integer)"));
3295 return elfcpp::PT_NULL
;
3299 script_saw_segment_start_expression(void* closurev
)
3301 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
3302 Script_sections
* ss
= closure
->script_options()->script_sections();
3303 ss
->set_saw_segment_start_expression(true);
3307 script_set_section_region(void* closurev
, const char* name
, size_t namelen
,
3310 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
3311 if (!closure
->script_options()->saw_sections_clause())
3313 gold_error(_("%s:%d:%d: MEMORY region '%.*s' referred to outside of "
3315 closure
->filename(), closure
->lineno(), closure
->charpos(),
3316 static_cast<int>(namelen
), name
);
3320 Script_sections
* ss
= closure
->script_options()->script_sections();
3321 Memory_region
* mr
= ss
->find_memory_region(name
, namelen
);
3324 gold_error(_("%s:%d:%d: MEMORY region '%.*s' not declared"),
3325 closure
->filename(), closure
->lineno(), closure
->charpos(),
3326 static_cast<int>(namelen
), name
);
3330 ss
->set_memory_region(mr
, set_vma
);
3334 script_add_memory(void* closurev
, const char* name
, size_t namelen
,
3335 unsigned int attrs
, Expression
* origin
, Expression
* length
)
3337 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
3338 Script_sections
* ss
= closure
->script_options()->script_sections();
3339 ss
->add_memory_region(name
, namelen
, attrs
, origin
, length
);
3342 extern "C" unsigned int
3343 script_parse_memory_attr(void* closurev
, const char* attrs
, size_t attrlen
,
3353 attributes
|= MEM_READABLE
; break;
3356 attributes
|= MEM_READABLE
| MEM_WRITEABLE
; break;
3359 attributes
|= MEM_EXECUTABLE
; break;
3362 attributes
|= MEM_ALLOCATABLE
; break;
3367 attributes
|= MEM_INITIALIZED
; break;
3369 yyerror(closurev
, _("unknown MEMORY attribute"));
3373 attributes
= (~ attributes
) & MEM_ATTR_MASK
;
3379 script_include_directive(int first_token
, void* closurev
,
3380 const char* filename
, size_t length
)
3382 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
3383 std::string
name(filename
, length
);
3384 Command_line
* cmdline
= closure
->command_line();
3385 read_script_file(name
.c_str(), cmdline
, &cmdline
->script_options(),
3386 first_token
, Lex::LINKER_SCRIPT
);
3389 // Functions for memory regions.
3391 extern "C" Expression
*
3392 script_exp_function_origin(void* closurev
, const char* name
, size_t namelen
)
3394 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
3395 Script_sections
* ss
= closure
->script_options()->script_sections();
3396 Expression
* origin
= ss
->find_memory_region_origin(name
, namelen
);
3400 gold_error(_("undefined memory region '%s' referenced "
3401 "in ORIGIN expression"),
3403 // Create a dummy expression to prevent crashes later on.
3404 origin
= script_exp_integer(0);
3410 extern "C" Expression
*
3411 script_exp_function_length(void* closurev
, const char* name
, size_t namelen
)
3413 Parser_closure
* closure
= static_cast<Parser_closure
*>(closurev
);
3414 Script_sections
* ss
= closure
->script_options()->script_sections();
3415 Expression
* length
= ss
->find_memory_region_length(name
, namelen
);
3419 gold_error(_("undefined memory region '%s' referenced "
3420 "in LENGTH expression"),
3422 // Create a dummy expression to prevent crashes later on.
3423 length
= script_exp_integer(0);