1 // script.cc -- handle linker scripts for gold.
2 
3 // Copyright (C) 2006-2020 Free Software Foundation, Inc.
4 // Written by Ian Lance Taylor <iant@google.com>.
5 
6 // This file is part of gold.
7 
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.
12 
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.
17 
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.
22 
23 #include "gold.h"
24 
25 #include <cstdio>
26 #include <cstdlib>
27 #include <cstring>
28 #include <fnmatch.h>
29 #include <string>
30 #include <vector>
31 #include "filenames.h"
32 
33 #include "elfcpp.h"
34 #include "demangle.h"
35 #include "dirsearch.h"
36 #include "options.h"
37 #include "fileread.h"
38 #include "workqueue.h"
39 #include "readsyms.h"
40 #include "parameters.h"
41 #include "layout.h"
42 #include "symtab.h"
43 #include "target-select.h"
44 #include "script.h"
45 #include "script-c.h"
46 #include "incremental.h"
47 
48 namespace gold
49 {
50 
51 // A token read from a script file.  We don't implement keywords here;
52 // all keywords are simply represented as a string.
53 
54 class Token
55 {
56  public:
57   // Token classification.
58   enum Classification
59   {
60     // Token is invalid.
61     TOKEN_INVALID,
62     // Token indicates end of input.
63     TOKEN_EOF,
64     // Token is a string of characters.
65     TOKEN_STRING,
66     // Token is a quoted string of characters.
67     TOKEN_QUOTED_STRING,
68     // Token is an operator.
69     TOKEN_OPERATOR,
70     // Token is a number (an integer).
71     TOKEN_INTEGER
72   };
73 
74   // We need an empty constructor so that we can put this STL objects.
75   Token()
76     : classification_(TOKEN_INVALID), value_(NULL), value_length_(0),
77       opcode_(0), lineno_(0), charpos_(0)
78   { }
79 
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)
84   {
85     gold_assert(classification == TOKEN_INVALID
86 		|| classification == TOKEN_EOF);
87   }
88 
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)
94   {
95     gold_assert(classification != TOKEN_INVALID
96 		&& classification != TOKEN_EOF);
97   }
98 
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)
103   { }
104 
105   // Return whether the token is invalid.
106   bool
107   is_invalid() const
108   { return this->classification_ == TOKEN_INVALID; }
109 
110   // Return whether this is an EOF token.
111   bool
112   is_eof() const
113   { return this->classification_ == TOKEN_EOF; }
114 
115   // Return the token classification.
116   Classification
117   classification() const
118   { return this->classification_; }
119 
120   // Return the line number at which the token starts.
121   int
122   lineno() const
123   { return this->lineno_; }
124 
125   // Return the character position at this the token starts.
126   int
127   charpos() const
128   { return this->charpos_; }
129 
130   // Get the value of a token.
131 
132   const char*
133   string_value(size_t* length) const
134   {
135     gold_assert(this->classification_ == TOKEN_STRING
136 		|| this->classification_ == TOKEN_QUOTED_STRING);
137     *length = this->value_length_;
138     return this->value_;
139   }
140 
141   int
142   operator_value() const
143   {
144     gold_assert(this->classification_ == TOKEN_OPERATOR);
145     return this->opcode_;
146   }
147 
148   uint64_t
149   integer_value() const;
150 
151  private:
152   // The token classification.
153   Classification classification_;
154   // The token value, for TOKEN_STRING or TOKEN_QUOTED_STRING or
155   // TOKEN_INTEGER.
156   const char* value_;
157   // The length of the token value.
158   size_t value_length_;
159   // The token value, for TOKEN_OPERATOR.
160   int opcode_;
161   // The line number where this token started (one based).
162   int lineno_;
163   // The character position within the line where this token started
164   // (one based).
165   int charpos_;
166 };
167 
168 // Return the value of a TOKEN_INTEGER.
169 
170 uint64_t
171 Token::integer_value() const
172 {
173   gold_assert(this->classification_ == TOKEN_INTEGER);
174 
175   size_t len = this->value_length_;
176 
177   uint64_t multiplier = 1;
178   char last = this->value_[len - 1];
179   if (last == 'm' || last == 'M')
180     {
181       multiplier = 1024 * 1024;
182       --len;
183     }
184   else if (last == 'k' || last == 'K')
185     {
186       multiplier = 1024;
187       --len;
188     }
189 
190   char *end;
191   uint64_t ret = strtoull(this->value_, &end, 0);
192   gold_assert(static_cast<size_t>(end - this->value_) == len);
193 
194   return ret * multiplier;
195 }
196 
197 // This class handles lexing a file into a sequence of tokens.
198 
199 class Lex
200 {
201  public:
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.
205   enum Mode
206   {
207     // Reading an ordinary linker script.
208     LINKER_SCRIPT,
209     // Reading an expression in a linker script.
210     EXPRESSION,
211     // Reading a version script.
212     VERSION_SCRIPT,
213     // Reading a --dynamic-list file.
214     DYNAMIC_LIST
215   };
216 
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)
222   { }
223 
224   // Read a file into a string.
225   static void
226   read_file(Input_file*, std::string*);
227 
228   // Return the next token.
229   const Token*
230   next_token();
231 
232   // Return the current lexing mode.
233   Lex::Mode
234   mode() const
235   { return this->mode_; }
236 
237   // Set the lexing mode.
238   void
239   set_mode(Mode mode)
240   { this->mode_ = mode; }
241 
242  private:
243   Lex(const Lex&);
244   Lex& operator=(const Lex&);
245 
246   // Make a general token with no value at the current location.
247   Token
248   make_token(Token::Classification c, const char* start) const
249   { return Token(c, this->lineno_, start - this->linestart_ + 1); }
250 
251   // Make a general token with a value at the current location.
252   Token
253   make_token(Token::Classification c, const char* v, size_t len,
254 	     const char* start)
255     const
256   { return Token(c, v, len, this->lineno_, start - this->linestart_ + 1); }
257 
258   // Make an operator token at the current location.
259   Token
260   make_token(int opcode, const char* start) const
261   { return Token(opcode, this->lineno_, start - this->linestart_ + 1); }
262 
263   // Make an invalid token at the current location.
264   Token
265   make_invalid_token(const char* start)
266   { return this->make_token(Token::TOKEN_INVALID, start); }
267 
268   // Make an EOF token at the current location.
269   Token
270   make_eof_token(const char* start)
271   { return this->make_token(Token::TOKEN_EOF, start); }
272 
273   // Return whether C can be the first character in a name.  C2 is the
274   // next character, since we sometimes need that.
275   inline bool
276   can_start_name(char c, char c2);
277 
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.
281   inline const char*
282   can_continue_name(const char* c);
283 
284   // Return whether C, C2, C3 can start a hex number.
285   inline bool
286   can_start_hex(char c, char c2, char c3);
287 
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.
291   inline const char*
292   can_continue_hex(const char* c);
293 
294   // Return whether C can start a non-hex number.
295   static inline bool
296   can_start_number(char c);
297 
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.
301   inline const char*
302   can_continue_number(const char* c)
303   { return Lex::can_start_number(*c) ? c + 1 : NULL; }
304 
305   // If C1 C2 C3 form a valid three character operator, return the
306   // opcode.  Otherwise return 0.
307   static inline int
308   three_char_operator(char c1, char c2, char c3);
309 
310   // If C1 C2 form a valid two character operator, return the opcode.
311   // Otherwise return 0.
312   static inline int
313   two_char_operator(char c1, char c2);
314 
315   // If C1 is a valid one character operator, return the opcode.
316   // Otherwise return 0.
317   static inline int
318   one_char_operator(char c1);
319 
320   // Read the next token.
321   Token
322   get_token(const char**);
323 
324   // Skip a C style /* */ comment.  Return false if the comment did
325   // not end.
326   bool
327   skip_c_comment(const char**);
328 
329   // Skip a line # comment.  Return false if there was no newline.
330   bool
331   skip_line_comment(const char**);
332 
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.
336   inline 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);
340 
341   // Build a token from a quoted string.
342   Token
343   gather_quoted_string(const char** pp);
344 
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.
352   Mode mode_;
353   // The code to use for the first token.  This is set to 0 after it
354   // is used.
355   int first_token_;
356   // The current token.
357   Token token_;
358   // The current line number.
359   int lineno_;
360   // The start of the current line in the string.
361   const char* linestart_;
362 };
363 
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.
367 
368 void
369 Lex::read_file(Input_file* input_file, std::string* contents)
370 {
371   off_t filesize = input_file->file().filesize();
372   contents->clear();
373   contents->reserve(filesize);
374 
375   off_t off = 0;
376   unsigned char buf[BUFSIZ];
377   while (off < filesize)
378     {
379       off_t get = BUFSIZ;
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);
384       off += get;
385     }
386 }
387 
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
397 // compatible.
398 
399 inline bool
400 Lex::can_start_name(char c, char c2)
401 {
402   switch (c)
403     {
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':
408     case 'Y': case 'Z':
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':
413     case 'y': case 'z':
414     case '_': case '.': case '$':
415       return true;
416 
417     case '/': case '\\':
418       return this->mode_ == LINKER_SCRIPT;
419 
420     case '~':
421       return this->mode_ == LINKER_SCRIPT && can_continue_name(&c2);
422 
423     case '*': case '[':
424       return (this->mode_ == VERSION_SCRIPT
425               || this->mode_ == DYNAMIC_LIST
426 	      || (this->mode_ == LINKER_SCRIPT
427 		  && can_continue_name(&c2)));
428 
429     default:
430       return false;
431     }
432 }
433 
434 // Return whether C can continue a name which has already started.
435 // Subsequent characters in a name are the same as the leading
436 // characters, plus digits and "=+-:[],?*".  So in general the linker
437 // script language requires spaces around operators, unless we know
438 // that we are parsing an expression.
439 
440 inline const char*
441 Lex::can_continue_name(const char* c)
442 {
443   switch (*c)
444     {
445     case 'A': case 'B': case 'C': case 'D': case 'E': case 'F':
446     case 'G': case 'H': case 'I': case 'J': case 'K': case 'L':
447     case 'M': case 'N': case 'O': case 'Q': case 'P': case 'R':
448     case 'S': case 'T': case 'U': case 'V': case 'W': case 'X':
449     case 'Y': case 'Z':
450     case 'a': case 'b': case 'c': case 'd': case 'e': case 'f':
451     case 'g': case 'h': case 'i': case 'j': case 'k': case 'l':
452     case 'm': case 'n': case 'o': case 'q': case 'p': case 'r':
453     case 's': case 't': case 'u': case 'v': case 'w': case 'x':
454     case 'y': case 'z':
455     case '_': case '.': case '$':
456     case '0': case '1': case '2': case '3': case '4':
457     case '5': case '6': case '7': case '8': case '9':
458       return c + 1;
459 
460     // TODO(csilvers): why not allow ~ in names for version-scripts?
461     case '/': case '\\': case '~':
462     case '=': case '+':
463     case ',':
464       if (this->mode_ == LINKER_SCRIPT)
465         return c + 1;
466       return NULL;
467 
468     case '[': case ']': case '*': case '?': case '-':
469       if (this->mode_ == LINKER_SCRIPT || this->mode_ == VERSION_SCRIPT
470           || this->mode_ == DYNAMIC_LIST)
471         return c + 1;
472       return NULL;
473 
474     // TODO(csilvers): why allow this?  ^ is meaningless in version scripts.
475     case '^':
476       if (this->mode_ == VERSION_SCRIPT || this->mode_ == DYNAMIC_LIST)
477         return c + 1;
478       return NULL;
479 
480     case ':':
481       if (this->mode_ == LINKER_SCRIPT)
482         return c + 1;
483       else if ((this->mode_ == VERSION_SCRIPT || this->mode_ == DYNAMIC_LIST)
484                && (c[1] == ':'))
485         {
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.
488           return c + 2;
489         }
490       return NULL;
491 
492     default:
493       return NULL;
494     }
495 }
496 
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
500 // accept them.
501 
502 // Return whether C1 C2 C3 can start a hex number.
503 
504 inline bool
505 Lex::can_start_hex(char c1, char c2, char c3)
506 {
507   if (c1 == '0' && (c2 == 'x' || c2 == 'X'))
508     return this->can_continue_hex(&c3);
509   return false;
510 }
511 
512 // Return whether C can appear in a hex number.
513 
514 inline const char*
515 Lex::can_continue_hex(const char* c)
516 {
517   switch (*c)
518     {
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':
523       return c + 1;
524 
525     default:
526       return NULL;
527     }
528 }
529 
530 // Return whether C can start a non-hex number.
531 
532 inline bool
533 Lex::can_start_number(char c)
534 {
535   switch (c)
536     {
537     case '0': case '1': case '2': case '3': case '4':
538     case '5': case '6': case '7': case '8': case '9':
539       return true;
540 
541     default:
542       return false;
543     }
544 }
545 
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.
549 
550 inline int
551 Lex::three_char_operator(char c1, char c2, char c3)
552 {
553   switch (c1)
554     {
555     case '<':
556       if (c2 == '<' && c3 == '=')
557 	return LSHIFTEQ;
558       break;
559     case '>':
560       if (c2 == '>' && c3 == '=')
561 	return RSHIFTEQ;
562       break;
563     default:
564       break;
565     }
566   return 0;
567 }
568 
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.
572 
573 inline int
574 Lex::two_char_operator(char c1, char c2)
575 {
576   switch (c1)
577     {
578     case '=':
579       if (c2 == '=')
580 	return EQ;
581       break;
582     case '!':
583       if (c2 == '=')
584 	return NE;
585       break;
586     case '+':
587       if (c2 == '=')
588 	return PLUSEQ;
589       break;
590     case '-':
591       if (c2 == '=')
592 	return MINUSEQ;
593       break;
594     case '*':
595       if (c2 == '=')
596 	return MULTEQ;
597       break;
598     case '/':
599       if (c2 == '=')
600 	return DIVEQ;
601       break;
602     case '|':
603       if (c2 == '=')
604 	return OREQ;
605       if (c2 == '|')
606 	return OROR;
607       break;
608     case '&':
609       if (c2 == '=')
610 	return ANDEQ;
611       if (c2 == '&')
612 	return ANDAND;
613       break;
614     case '>':
615       if (c2 == '=')
616 	return GE;
617       if (c2 == '>')
618 	return RSHIFT;
619       break;
620     case '<':
621       if (c2 == '=')
622 	return LE;
623       if (c2 == '<')
624 	return LSHIFT;
625       break;
626     default:
627       break;
628     }
629   return 0;
630 }
631 
632 // If C1 is a valid operator, return the opcode.  Otherwise return 0.
633 
634 inline int
635 Lex::one_char_operator(char c1)
636 {
637   switch (c1)
638     {
639     case '+':
640     case '-':
641     case '*':
642     case '/':
643     case '%':
644     case '!':
645     case '&':
646     case '|':
647     case '^':
648     case '~':
649     case '<':
650     case '>':
651     case '=':
652     case '?':
653     case ',':
654     case '(':
655     case ')':
656     case '{':
657     case '}':
658     case '[':
659     case ']':
660     case ':':
661     case ';':
662       return c1;
663     default:
664       return 0;
665     }
666 }
667 
668 // Skip a C style comment.  *PP points to just after the "/*".  Return
669 // false if the comment did not end.
670 
671 bool
672 Lex::skip_c_comment(const char** pp)
673 {
674   const char* p = *pp;
675   while (p[0] != '*' || p[1] != '/')
676     {
677       if (*p == '\0')
678 	{
679 	  *pp = p;
680 	  return false;
681 	}
682 
683       if (*p == '\n')
684 	{
685 	  ++this->lineno_;
686 	  this->linestart_ = p + 1;
687 	}
688       ++p;
689     }
690 
691   *pp = p + 2;
692   return true;
693 }
694 
695 // Skip a line # comment.  Return false if there was no newline.
696 
697 bool
698 Lex::skip_line_comment(const char** pp)
699 {
700   const char* p = *pp;
701   size_t skip = strcspn(p, "\n");
702   if (p[skip] == '\0')
703     {
704       *pp = p + skip;
705       return false;
706     }
707 
708   p += skip + 1;
709   ++this->lineno_;
710   this->linestart_ = p;
711   *pp = p;
712 
713   return true;
714 }
715 
716 // Build a token CLASSIFICATION from all characters that match
717 // CAN_CONTINUE_FN.  Update *PP.
718 
719 inline Token
720 Lex::gather_token(Token::Classification classification,
721 		  const char* (Lex::*can_continue_fn)(const char*),
722 		  const char* start,
723 		  const char* match,
724 		  const char** pp)
725 {
726   const char* new_match = NULL;
727   while ((new_match = (this->*can_continue_fn)(match)) != NULL)
728     match = new_match;
729 
730   // A special case: integers may be followed by a single M or K,
731   // case-insensitive.
732   if (classification == Token::TOKEN_INTEGER
733       && (*match == 'm' || *match == 'M' || *match == 'k' || *match == 'K'))
734     ++match;
735 
736   *pp = match;
737   return this->make_token(classification, start, match - start, start);
738 }
739 
740 // Build a token from a quoted string.
741 
742 Token
743 Lex::gather_quoted_string(const char** pp)
744 {
745   const char* start = *pp;
746   const char* p = start;
747   ++p;
748   size_t skip = strcspn(p, "\"\n");
749   if (p[skip] != '"')
750     return this->make_invalid_token(start);
751   *pp = p + skip + 1;
752   return this->make_token(Token::TOKEN_QUOTED_STRING, p, skip, start);
753 }
754 
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
758 // the input.
759 
760 Token
761 Lex::get_token(const char** pp)
762 {
763   const char* p = *pp;
764 
765   while (true)
766     {
767       // Skip whitespace quickly.
768       while (*p == ' ' || *p == '\t' || *p == '\r')
769 	++p;
770 
771       if (*p == '\n')
772 	{
773 	  ++p;
774 	  ++this->lineno_;
775 	  this->linestart_ = p;
776 	  continue;
777 	}
778 
779       char c0 = *p;
780 
781       if (c0 == '\0')
782 	{
783 	  *pp = p;
784 	  return this->make_eof_token(p);
785 	}
786 
787       char c1 = p[1];
788 
789       // Skip C style comments.
790       if (c0 == '/' && c1 == '*')
791 	{
792 	  int lineno = this->lineno_;
793 	  int charpos = p - this->linestart_ + 1;
794 
795 	  *pp = p + 2;
796 	  if (!this->skip_c_comment(pp))
797 	    return Token(Token::TOKEN_INVALID, lineno, charpos);
798 	  p = *pp;
799 
800 	  continue;
801 	}
802 
803       // Skip line comments.
804       if (c0 == '#')
805 	{
806 	  *pp = p + 1;
807 	  if (!this->skip_line_comment(pp))
808 	    return this->make_eof_token(p);
809 	  p = *pp;
810 	  continue;
811 	}
812 
813       // Check for a name.
814       if (this->can_start_name(c0, c1))
815 	return this->gather_token(Token::TOKEN_STRING,
816 				  &Lex::can_continue_name,
817 				  p, p + 1, pp);
818 
819       // We accept any arbitrary name in double quotes, as long as it
820       // does not cross a line boundary.
821       if (*p == '"')
822 	{
823 	  *pp = p;
824 	  return this->gather_quoted_string(pp);
825 	}
826 
827       // Be careful not to lookahead past the end of the buffer.
828       char c2 = (c1 == '\0' ? '\0' : p[2]);
829 
830       // Check for a number.
831 
832       if (this->can_start_hex(c0, c1, c2))
833 	return this->gather_token(Token::TOKEN_INTEGER,
834 				  &Lex::can_continue_hex,
835 				  p, p + 3, pp);
836 
837       if (Lex::can_start_number(c0))
838 	return this->gather_token(Token::TOKEN_INTEGER,
839 				  &Lex::can_continue_number,
840 				  p, p + 1, pp);
841 
842       // Check for operators.
843 
844       int opcode = Lex::three_char_operator(c0, c1, c2);
845       if (opcode != 0)
846 	{
847 	  *pp = p + 3;
848 	  return this->make_token(opcode, p);
849 	}
850 
851       opcode = Lex::two_char_operator(c0, c1);
852       if (opcode != 0)
853 	{
854 	  *pp = p + 2;
855 	  return this->make_token(opcode, p);
856 	}
857 
858       opcode = Lex::one_char_operator(c0);
859       if (opcode != 0)
860 	{
861 	  *pp = p + 1;
862 	  return this->make_token(opcode, p);
863 	}
864 
865       return this->make_token(Token::TOKEN_INVALID, p);
866     }
867 }
868 
869 // Return the next token.
870 
871 const Token*
872 Lex::next_token()
873 {
874   // The first token is special.
875   if (this->first_token_ != 0)
876     {
877       this->token_ = Token(this->first_token_, 0, 0);
878       this->first_token_ = 0;
879       return &this->token_;
880     }
881 
882   this->token_ = this->get_token(&this->current_);
883 
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_);
890 
891   return &this->token_;
892 }
893 
894 // class Symbol_assignment.
895 
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
900 // been finalized.
901 
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?
905 
906 void
907 Symbol_assignment::add_to_table(Symbol_table* symtab)
908 {
909   elfcpp::STV vis = this->hidden_ ? elfcpp::STV_HIDDEN : elfcpp::STV_DEFAULT;
910   this->sym_ = symtab->define_as_constant(this->name_.c_str(),
911 					  NULL, // version
912 					  (this->is_defsym_
913 					   ? Symbol_table::DEFSYM
914 					   : Symbol_table::SCRIPT),
915 					  0, // value
916 					  0, // size
917 					  elfcpp::STT_NOTYPE,
918 					  elfcpp::STB_GLOBAL,
919 					  vis,
920 					  0, // nonvis
921 					  this->provide_,
922                                           true); // force_override
923 }
924 
925 // Finalize a symbol value.
926 
927 void
928 Symbol_assignment::finalize(Symbol_table* symtab, const Layout* layout)
929 {
930   this->finalize_maybe_dot(symtab, layout, false, 0, NULL);
931 }
932 
933 // Finalize a symbol value which can refer to the dot symbol.
934 
935 void
936 Symbol_assignment::finalize_with_dot(Symbol_table* symtab,
937 				     const Layout* layout,
938 				     uint64_t dot_value,
939 				     Output_section* dot_section)
940 {
941   this->finalize_maybe_dot(symtab, layout, true, dot_value, dot_section);
942 }
943 
944 // Finalize a symbol value, internal version.
945 
946 void
947 Symbol_assignment::finalize_maybe_dot(Symbol_table* symtab,
948 				      const Layout* layout,
949 				      bool is_dot_available,
950 				      uint64_t dot_value,
951 				      Output_section* dot_section)
952 {
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)
956     {
957       gold_assert(this->provide_);
958       return;
959     }
960 
961   if (parameters->target().get_size() == 32)
962     {
963 #if defined(HAVE_TARGET_32_LITTLE) || defined(HAVE_TARGET_32_BIG)
964       this->sized_finalize<32>(symtab, layout, is_dot_available, dot_value,
965 			       dot_section);
966 #else
967       gold_unreachable();
968 #endif
969     }
970   else if (parameters->target().get_size() == 64)
971     {
972 #if defined(HAVE_TARGET_64_LITTLE) || defined(HAVE_TARGET_64_BIG)
973       this->sized_finalize<64>(symtab, layout, is_dot_available, dot_value,
974 			       dot_section);
975 #else
976       gold_unreachable();
977 #endif
978     }
979   else
980     gold_unreachable();
981 }
982 
983 template<int size>
984 void
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)
988 {
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,
994 						  is_dot_available,
995 						  dot_value, dot_section,
996 						  &section, 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);
1005 }
1006 
1007 // Set the symbol value if the expression yields an absolute value or
1008 // a value relative to DOT_SECTION.
1009 
1010 void
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)
1014 {
1015   if (this->sym_ == NULL)
1016     return;
1017 
1018   Output_section* val_section;
1019   bool is_valid;
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))
1025     return;
1026 
1027   if (parameters->target().get_size() == 32)
1028     {
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);
1032 #else
1033       gold_unreachable();
1034 #endif
1035     }
1036   else if (parameters->target().get_size() == 64)
1037     {
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);
1041 #else
1042       gold_unreachable();
1043 #endif
1044     }
1045   else
1046     gold_unreachable();
1047   if (val_section != NULL)
1048     this->sym_->set_output_section(val_section);
1049 }
1050 
1051 // Print for debugging.
1052 
1053 void
1054 Symbol_assignment::print(FILE* f) const
1055 {
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_)
1061     gold_unreachable();
1062 
1063   fprintf(f, "%s = ", this->name_.c_str());
1064   this->val_->print(f);
1065 
1066   if (this->provide_ || this->hidden_)
1067     fprintf(f, ")");
1068 
1069   fprintf(f, "\n");
1070 }
1071 
1072 // Class Script_assertion.
1073 
1074 // Check the assertion.
1075 
1076 void
1077 Script_assertion::check(const Symbol_table* symtab, const Layout* layout)
1078 {
1079   if (!this->check_->eval(symtab, layout, true))
1080     gold_error("%s", this->message_.c_str());
1081 }
1082 
1083 // Print for debugging.
1084 
1085 void
1086 Script_assertion::print(FILE* f) const
1087 {
1088   fprintf(f, "ASSERT(");
1089   this->check_->print(f);
1090   fprintf(f, ", \"%s\")\n", this->message_.c_str());
1091 }
1092 
1093 // Class Script_options.
1094 
1095 Script_options::Script_options()
1096   : entry_(), symbol_assignments_(), symbol_definitions_(),
1097     symbol_references_(), version_script_info_(), script_sections_()
1098 {
1099 }
1100 
1101 // Returns true if NAME is on the list of symbol assignments waiting
1102 // to be processed.
1103 
1104 bool
1105 Script_options::is_pending_assignment(const char* name)
1106 {
1107   for (Symbol_assignments::iterator p = this->symbol_assignments_.begin();
1108        p != this->symbol_assignments_.end();
1109        ++p)
1110     if ((*p)->name() == name)
1111       return true;
1112   return false;
1113 }
1114 
1115 // Populates the set with symbols defined in defsym LHS.
1116 
1117 void Script_options::find_defsym_defs(Unordered_set<std::string>& defsym_set)
1118 {
1119   for (Symbol_assignments::const_iterator p = this->symbol_assignments_.begin();
1120        p != this->symbol_assignments_.end();
1121        ++p)
1122     {
1123       defsym_set.insert((*p)->name());
1124     }
1125 }
1126 
1127 void
1128 Script_options::set_defsym_uses_in_real_elf(Symbol_table* symtab) const
1129 {
1130   for (Symbol_assignments::const_iterator p = this->symbol_assignments_.begin();
1131        p != this->symbol_assignments_.end();
1132        ++p)
1133     {
1134       (*p)->value()->set_expr_sym_in_real_elf(symtab);
1135     }
1136 }
1137 
1138 // Add a symbol to be defined.
1139 
1140 void
1141 Script_options::add_symbol_assignment(const char* name, size_t length,
1142 				      bool is_defsym, Expression* value,
1143 				      bool provide, bool hidden)
1144 {
1145   if (length != 1 || name[0] != '.')
1146     {
1147       if (this->script_sections_.in_sections_clause())
1148 	{
1149 	  gold_assert(!is_defsym);
1150 	  this->script_sections_.add_symbol_assignment(name, length, value,
1151 						       provide, hidden);
1152 	}
1153       else
1154 	{
1155 	  Symbol_assignment* p = new Symbol_assignment(name, length, is_defsym,
1156 						       value, provide, hidden);
1157 	  this->symbol_assignments_.push_back(p);
1158 	}
1159 
1160       if (!provide)
1161 	{
1162 	  std::string n(name, length);
1163 	  this->symbol_definitions_.insert(n);
1164 	  this->symbol_references_.erase(n);
1165 	}
1166     }
1167   else
1168     {
1169       if (provide || hidden)
1170 	gold_error(_("invalid use of PROVIDE for dot symbol"));
1171 
1172       // The GNU linker permits assignments to dot outside of SECTIONS
1173       // clauses and treats them as occurring inside, so we don't
1174       // check in_sections_clause here.
1175       this->script_sections_.add_dot_assignment(value);
1176     }
1177 }
1178 
1179 // Add a reference to a symbol.
1180 
1181 void
1182 Script_options::add_symbol_reference(const char* name, size_t length)
1183 {
1184   if (length != 1 || name[0] != '.')
1185     {
1186       std::string n(name, length);
1187       if (this->symbol_definitions_.find(n) == this->symbol_definitions_.end())
1188 	this->symbol_references_.insert(n);
1189     }
1190 }
1191 
1192 // Add an assertion.
1193 
1194 void
1195 Script_options::add_assertion(Expression* check, const char* message,
1196 			      size_t messagelen)
1197 {
1198   if (this->script_sections_.in_sections_clause())
1199     this->script_sections_.add_assertion(check, message, messagelen);
1200   else
1201     {
1202       Script_assertion* p = new Script_assertion(check, message, messagelen);
1203       this->assertions_.push_back(p);
1204     }
1205 }
1206 
1207 // Create sections required by any linker scripts.
1208 
1209 void
1210 Script_options::create_script_sections(Layout* layout)
1211 {
1212   if (this->saw_sections_clause())
1213     this->script_sections_.create_sections(layout);
1214 }
1215 
1216 // Add any symbols we are defining to the symbol table.
1217 
1218 void
1219 Script_options::add_symbols_to_table(Symbol_table* symtab)
1220 {
1221   for (Symbol_assignments::iterator p = this->symbol_assignments_.begin();
1222        p != this->symbol_assignments_.end();
1223        ++p)
1224     (*p)->add_to_table(symtab);
1225   this->script_sections_.add_symbols_to_table(symtab);
1226 }
1227 
1228 // Finalize symbol values.  Also check assertions.
1229 
1230 void
1231 Script_options::finalize_symbols(Symbol_table* symtab, const Layout* layout)
1232 {
1233   // We finalize the symbols defined in SECTIONS first, because they
1234   // are the ones which may have changed.  This way if symbol outside
1235   // SECTIONS are defined in terms of symbols inside SECTIONS, they
1236   // will get the right value.
1237   this->script_sections_.finalize_symbols(symtab, layout);
1238 
1239   for (Symbol_assignments::iterator p = this->symbol_assignments_.begin();
1240        p != this->symbol_assignments_.end();
1241        ++p)
1242     (*p)->finalize(symtab, layout);
1243 
1244   for (Assertions::iterator p = this->assertions_.begin();
1245        p != this->assertions_.end();
1246        ++p)
1247     (*p)->check(symtab, layout);
1248 }
1249 
1250 // Set section addresses.  We set all the symbols which have absolute
1251 // values.  Then we let the SECTIONS clause do its thing.  This
1252 // returns the segment which holds the file header and segment
1253 // headers, if any.
1254 
1255 Output_segment*
1256 Script_options::set_section_addresses(Symbol_table* symtab, Layout* layout)
1257 {
1258   for (Symbol_assignments::iterator p = this->symbol_assignments_.begin();
1259        p != this->symbol_assignments_.end();
1260        ++p)
1261     (*p)->set_if_absolute(symtab, layout, false, 0, NULL);
1262 
1263   return this->script_sections_.set_section_addresses(symtab, layout);
1264 }
1265 
1266 // This class holds data passed through the parser to the lexer and to
1267 // the parser support functions.  This avoids global variables.  We
1268 // can't use global variables because we need not be called by a
1269 // singleton thread.
1270 
1271 class Parser_closure
1272 {
1273  public:
1274   Parser_closure(const char* filename,
1275 		 const Position_dependent_options& posdep_options,
1276 		 bool parsing_defsym, bool in_group, bool is_in_sysroot,
1277                  Command_line* command_line,
1278 		 Script_options* script_options,
1279 		 Lex* lex,
1280 		 bool skip_on_incompatible_target,
1281 		 Script_info* script_info)
1282     : filename_(filename), posdep_options_(posdep_options),
1283       parsing_defsym_(parsing_defsym), in_group_(in_group),
1284       is_in_sysroot_(is_in_sysroot),
1285       skip_on_incompatible_target_(skip_on_incompatible_target),
1286       found_incompatible_target_(false),
1287       command_line_(command_line), script_options_(script_options),
1288       version_script_info_(script_options->version_script_info()),
1289       lex_(lex), lineno_(0), charpos_(0), lex_mode_stack_(), inputs_(NULL),
1290       script_info_(script_info)
1291   {
1292     // We start out processing C symbols in the default lex mode.
1293     this->language_stack_.push_back(Version_script_info::LANGUAGE_C);
1294     this->lex_mode_stack_.push_back(lex->mode());
1295   }
1296 
1297   // Return the file name.
1298   const char*
1299   filename() const
1300   { return this->filename_; }
1301 
1302   // Return the position dependent options.  The caller may modify
1303   // this.
1304   Position_dependent_options&
1305   position_dependent_options()
1306   { return this->posdep_options_; }
1307 
1308   // Whether we are parsing a --defsym.
1309   bool
1310   parsing_defsym() const
1311   { return this->parsing_defsym_; }
1312 
1313   // Return whether this script is being run in a group.
1314   bool
1315   in_group() const
1316   { return this->in_group_; }
1317 
1318   // Return whether this script was found using a directory in the
1319   // sysroot.
1320   bool
1321   is_in_sysroot() const
1322   { return this->is_in_sysroot_; }
1323 
1324   // Whether to skip to the next file with the same name if we find an
1325   // incompatible target in an OUTPUT_FORMAT statement.
1326   bool
1327   skip_on_incompatible_target() const
1328   { return this->skip_on_incompatible_target_; }
1329 
1330   // Stop skipping to the next file on an incompatible target.  This
1331   // is called when we make some unrevocable change to the data
1332   // structures.
1333   void
1334   clear_skip_on_incompatible_target()
1335   { this->skip_on_incompatible_target_ = false; }
1336 
1337   // Whether we found an incompatible target in an OUTPUT_FORMAT
1338   // statement.
1339   bool
1340   found_incompatible_target() const
1341   { return this->found_incompatible_target_; }
1342 
1343   // Note that we found an incompatible target.
1344   void
1345   set_found_incompatible_target()
1346   { this->found_incompatible_target_ = true; }
1347 
1348   // Returns the Command_line structure passed in at constructor time.
1349   // This value may be NULL.  The caller may modify this, which modifies
1350   // the passed-in Command_line object (not a copy).
1351   Command_line*
1352   command_line()
1353   { return this->command_line_; }
1354 
1355   // Return the options which may be set by a script.
1356   Script_options*
1357   script_options()
1358   { return this->script_options_; }
1359 
1360   // Return the object in which version script information should be stored.
1361   Version_script_info*
1362   version_script()
1363   { return this->version_script_info_; }
1364 
1365   // Return the next token, and advance.
1366   const Token*
1367   next_token()
1368   {
1369     const Token* token = this->lex_->next_token();
1370     this->lineno_ = token->lineno();
1371     this->charpos_ = token->charpos();
1372     return token;
1373   }
1374 
1375   // Set a new lexer mode, pushing the current one.
1376   void
1377   push_lex_mode(Lex::Mode mode)
1378   {
1379     this->lex_mode_stack_.push_back(this->lex_->mode());
1380     this->lex_->set_mode(mode);
1381   }
1382 
1383   // Pop the lexer mode.
1384   void
1385   pop_lex_mode()
1386   {
1387     gold_assert(!this->lex_mode_stack_.empty());
1388     this->lex_->set_mode(this->lex_mode_stack_.back());
1389     this->lex_mode_stack_.pop_back();
1390   }
1391 
1392   // Return the current lexer mode.
1393   Lex::Mode
1394   lex_mode() const
1395   { return this->lex_mode_stack_.back(); }
1396 
1397   // Return the line number of the last token.
1398   int
1399   lineno() const
1400   { return this->lineno_; }
1401 
1402   // Return the character position in the line of the last token.
1403   int
1404   charpos() const
1405   { return this->charpos_; }
1406 
1407   // Return the list of input files, creating it if necessary.  This
1408   // is a space leak--we never free the INPUTS_ pointer.
1409   Input_arguments*
1410   inputs()
1411   {
1412     if (this->inputs_ == NULL)
1413       this->inputs_ = new Input_arguments();
1414     return this->inputs_;
1415   }
1416 
1417   // Return whether we saw any input files.
1418   bool
1419   saw_inputs() const
1420   { return this->inputs_ != NULL && !this->inputs_->empty(); }
1421 
1422   // Return the current language being processed in a version script
1423   // (eg, "C++").  The empty string represents unmangled C names.
1424   Version_script_info::Language
1425   get_current_language() const
1426   { return this->language_stack_.back(); }
1427 
1428   // Push a language onto the stack when entering an extern block.
1429   void
1430   push_language(Version_script_info::Language lang)
1431   { this->language_stack_.push_back(lang); }
1432 
1433   // Pop a language off of the stack when exiting an extern block.
1434   void
1435   pop_language()
1436   {
1437     gold_assert(!this->language_stack_.empty());
1438     this->language_stack_.pop_back();
1439   }
1440 
1441   // Return a pointer to the incremental info.
1442   Script_info*
1443   script_info()
1444   { return this->script_info_; }
1445 
1446  private:
1447   // The name of the file we are reading.
1448   const char* filename_;
1449   // The position dependent options.
1450   Position_dependent_options posdep_options_;
1451   // True if we are parsing a --defsym.
1452   bool parsing_defsym_;
1453   // Whether we are currently in a --start-group/--end-group.
1454   bool in_group_;
1455   // Whether the script was found in a sysrooted directory.
1456   bool is_in_sysroot_;
1457   // If this is true, then if we find an OUTPUT_FORMAT with an
1458   // incompatible target, then we tell the parser to abort so that we
1459   // can search for the next file with the same name.
1460   bool skip_on_incompatible_target_;
1461   // True if we found an OUTPUT_FORMAT with an incompatible target.
1462   bool found_incompatible_target_;
1463   // May be NULL if the user chooses not to pass one in.
1464   Command_line* command_line_;
1465   // Options which may be set from any linker script.
1466   Script_options* script_options_;
1467   // Information parsed from a version script.
1468   Version_script_info* version_script_info_;
1469   // The lexer.
1470   Lex* lex_;
1471   // The line number of the last token returned by next_token.
1472   int lineno_;
1473   // The column number of the last token returned by next_token.
1474   int charpos_;
1475   // A stack of lexer modes.
1476   std::vector<Lex::Mode> lex_mode_stack_;
1477   // A stack of which extern/language block we're inside. Can be C++,
1478   // java, or empty for C.
1479   std::vector<Version_script_info::Language> language_stack_;
1480   // New input files found to add to the link.
1481   Input_arguments* inputs_;
1482   // Pointer to incremental linking info.
1483   Script_info* script_info_;
1484 };
1485 
1486 // FILE was found as an argument on the command line.  Try to read it
1487 // as a script.  Return true if the file was handled.
1488 
1489 bool
1490 read_input_script(Workqueue* workqueue, Symbol_table* symtab, Layout* layout,
1491 		  Dirsearch* dirsearch, int dirindex,
1492 		  Input_objects* input_objects, Mapfile* mapfile,
1493 		  Input_group* input_group,
1494 		  const Input_argument* input_argument,
1495 		  Input_file* input_file, Task_token* next_blocker,
1496 		  bool* used_next_blocker)
1497 {
1498   *used_next_blocker = false;
1499 
1500   std::string input_string;
1501   Lex::read_file(input_file, &input_string);
1502 
1503   Lex lex(input_string.c_str(), input_string.length(), PARSING_LINKER_SCRIPT);
1504 
1505   Script_info* script_info = NULL;
1506   if (layout->incremental_inputs() != NULL)
1507     {
1508       const std::string& filename = input_file->filename();
1509       Timespec mtime = input_file->file().get_mtime();
1510       unsigned int arg_serial = input_argument->file().arg_serial();
1511       script_info = new Script_info(filename);
1512       layout->incremental_inputs()->report_script(script_info, arg_serial,
1513 						  mtime);
1514     }
1515 
1516   Parser_closure closure(input_file->filename().c_str(),
1517 			 input_argument->file().options(),
1518 			 false,
1519 			 input_group != NULL,
1520 			 input_file->is_in_sysroot(),
1521                          NULL,
1522 			 layout->script_options(),
1523 			 &lex,
1524 			 input_file->will_search_for(),
1525 			 script_info);
1526 
1527   bool old_saw_sections_clause =
1528     layout->script_options()->saw_sections_clause();
1529 
1530   if (yyparse(&closure) != 0)
1531     {
1532       if (closure.found_incompatible_target())
1533 	{
1534 	  Read_symbols::incompatible_warning(input_argument, input_file);
1535 	  Read_symbols::requeue(workqueue, input_objects, symtab, layout,
1536 				dirsearch, dirindex, mapfile, input_argument,
1537 				input_group, next_blocker);
1538 	  return true;
1539 	}
1540       return false;
1541     }
1542 
1543   if (!old_saw_sections_clause
1544       && layout->script_options()->saw_sections_clause()
1545       && layout->have_added_input_section())
1546     gold_error(_("%s: SECTIONS seen after other input files; try -T/--script"),
1547 	       input_file->filename().c_str());
1548 
1549   if (!closure.saw_inputs())
1550     return true;
1551 
1552   Task_token* this_blocker = NULL;
1553   for (Input_arguments::const_iterator p = closure.inputs()->begin();
1554        p != closure.inputs()->end();
1555        ++p)
1556     {
1557       Task_token* nb;
1558       if (p + 1 == closure.inputs()->end())
1559 	nb = next_blocker;
1560       else
1561 	{
1562 	  nb = new Task_token(true);
1563 	  nb->add_blocker();
1564 	}
1565       workqueue->queue_soon(new Read_symbols(input_objects, symtab,
1566 					     layout, dirsearch, 0, mapfile, &*p,
1567 					     input_group, NULL, this_blocker, nb));
1568       this_blocker = nb;
1569     }
1570 
1571   *used_next_blocker = true;
1572 
1573   return true;
1574 }
1575 
1576 // Helper function for read_version_script(), read_commandline_script() and
1577 // script_include_directive().  Processes the given file in the mode indicated
1578 // by first_token and lex_mode.
1579 
1580 static bool
1581 read_script_file(const char* filename, Command_line* cmdline,
1582                  Script_options* script_options,
1583                  int first_token, Lex::Mode lex_mode)
1584 {
1585   Dirsearch dirsearch;
1586   std::string name = filename;
1587 
1588   // If filename is a relative filename, search for it manually using "." +
1589   // cmdline->options()->library_path() -- not dirsearch.
1590   if (!IS_ABSOLUTE_PATH(filename))
1591     {
1592       const General_options::Dir_list& search_path =
1593           cmdline->options().library_path();
1594       name = Dirsearch::find_file_in_dir_list(name, search_path, ".");
1595     }
1596 
1597   // The file locking code wants to record a Task, but we haven't
1598   // started the workqueue yet.  This is only for debugging purposes,
1599   // so we invent a fake value.
1600   const Task* task = reinterpret_cast<const Task*>(-1);
1601 
1602   // We don't want this file to be opened in binary mode.
1603   Position_dependent_options posdep = cmdline->position_dependent_options();
1604   if (posdep.format_enum() == General_options::OBJECT_FORMAT_BINARY)
1605     posdep.set_format_enum(General_options::OBJECT_FORMAT_ELF);
1606   Input_file_argument input_argument(name.c_str(),
1607 				     Input_file_argument::INPUT_FILE_TYPE_FILE,
1608 				     "", false, posdep);
1609   Input_file input_file(&input_argument);
1610   int dummy = 0;
1611   if (!input_file.open(dirsearch, task, &dummy))
1612     return false;
1613 
1614   std::string input_string;
1615   Lex::read_file(&input_file, &input_string);
1616 
1617   Lex lex(input_string.c_str(), input_string.length(), first_token);
1618   lex.set_mode(lex_mode);
1619 
1620   Parser_closure closure(filename,
1621 			 cmdline->position_dependent_options(),
1622 			 first_token == Lex::DYNAMIC_LIST,
1623 			 false,
1624 			 input_file.is_in_sysroot(),
1625                          cmdline,
1626 			 script_options,
1627 			 &lex,
1628 			 false,
1629 			 NULL);
1630   if (yyparse(&closure) != 0)
1631     {
1632       input_file.file().unlock(task);
1633       return false;
1634     }
1635 
1636   input_file.file().unlock(task);
1637 
1638   gold_assert(!closure.saw_inputs());
1639 
1640   return true;
1641 }
1642 
1643 // FILENAME was found as an argument to --script (-T).
1644 // Read it as a script, and execute its contents immediately.
1645 
1646 bool
1647 read_commandline_script(const char* filename, Command_line* cmdline)
1648 {
1649   return read_script_file(filename, cmdline, &cmdline->script_options(),
1650                           PARSING_LINKER_SCRIPT, Lex::LINKER_SCRIPT);
1651 }
1652 
1653 // FILENAME was found as an argument to --version-script.  Read it as
1654 // a version script, and store its contents in
1655 // cmdline->script_options()->version_script_info().
1656 
1657 bool
1658 read_version_script(const char* filename, Command_line* cmdline)
1659 {
1660   return read_script_file(filename, cmdline, &cmdline->script_options(),
1661                           PARSING_VERSION_SCRIPT, Lex::VERSION_SCRIPT);
1662 }
1663 
1664 // FILENAME was found as an argument to --dynamic-list.  Read it as a
1665 // list of symbols, and store its contents in DYNAMIC_LIST.
1666 
1667 bool
1668 read_dynamic_list(const char* filename, Command_line* cmdline,
1669                   Script_options* dynamic_list)
1670 {
1671   return read_script_file(filename, cmdline, dynamic_list,
1672                           PARSING_DYNAMIC_LIST, Lex::DYNAMIC_LIST);
1673 }
1674 
1675 // Implement the --defsym option on the command line.  Return true if
1676 // all is well.
1677 
1678 bool
1679 Script_options::define_symbol(const char* definition)
1680 {
1681   Lex lex(definition, strlen(definition), PARSING_DEFSYM);
1682   lex.set_mode(Lex::EXPRESSION);
1683 
1684   // Dummy value.
1685   Position_dependent_options posdep_options;
1686 
1687   Parser_closure closure("command line", posdep_options, true,
1688 			 false, false, NULL, this, &lex, false, NULL);
1689 
1690   if (yyparse(&closure) != 0)
1691     return false;
1692 
1693   gold_assert(!closure.saw_inputs());
1694 
1695   return true;
1696 }
1697 
1698 // Print the script to F for debugging.
1699 
1700 void
1701 Script_options::print(FILE* f) const
1702 {
1703   fprintf(f, "%s: Dumping linker script\n", program_name);
1704 
1705   if (!this->entry_.empty())
1706     fprintf(f, "ENTRY(%s)\n", this->entry_.c_str());
1707 
1708   for (Symbol_assignments::const_iterator p =
1709 	 this->symbol_assignments_.begin();
1710        p != this->symbol_assignments_.end();
1711        ++p)
1712     (*p)->print(f);
1713 
1714   for (Assertions::const_iterator p = this->assertions_.begin();
1715        p != this->assertions_.end();
1716        ++p)
1717     (*p)->print(f);
1718 
1719   this->script_sections_.print(f);
1720 
1721   this->version_script_info_.print(f);
1722 }
1723 
1724 // Manage mapping from keywords to the codes expected by the bison
1725 // parser.  We construct one global object for each lex mode with
1726 // keywords.
1727 
1728 class Keyword_to_parsecode
1729 {
1730  public:
1731   // The structure which maps keywords to parsecodes.
1732   struct Keyword_parsecode
1733   {
1734     // Keyword.
1735     const char* keyword;
1736     // Corresponding parsecode.
1737     int parsecode;
1738   };
1739 
1740   Keyword_to_parsecode(const Keyword_parsecode* keywords,
1741                        int keyword_count)
1742       : keyword_parsecodes_(keywords), keyword_count_(keyword_count)
1743   { }
1744 
1745   // Return the parsecode corresponding KEYWORD, or 0 if it is not a
1746   // keyword.
1747   int
1748   keyword_to_parsecode(const char* keyword, size_t len) const;
1749 
1750  private:
1751   const Keyword_parsecode* keyword_parsecodes_;
1752   const int keyword_count_;
1753 };
1754 
1755 // Mapping from keyword string to keyword parsecode.  This array must
1756 // be kept in sorted order.  Parsecodes are looked up using bsearch.
1757 // This array must correspond to the list of parsecodes in yyscript.y.
1758 
1759 static const Keyword_to_parsecode::Keyword_parsecode
1760 script_keyword_parsecodes[] =
1761 {
1762   { "ABSOLUTE", ABSOLUTE },
1763   { "ADDR", ADDR },
1764   { "ALIGN", ALIGN_K },
1765   { "ALIGNOF", ALIGNOF },
1766   { "ASSERT", ASSERT_K },
1767   { "AS_NEEDED", AS_NEEDED },
1768   { "AT", AT },
1769   { "BIND", BIND },
1770   { "BLOCK", BLOCK },
1771   { "BYTE", BYTE },
1772   { "CONSTANT", CONSTANT },
1773   { "CONSTRUCTORS", CONSTRUCTORS },
1774   { "COPY", COPY },
1775   { "CREATE_OBJECT_SYMBOLS", CREATE_OBJECT_SYMBOLS },
1776   { "DATA_SEGMENT_ALIGN", DATA_SEGMENT_ALIGN },
1777   { "DATA_SEGMENT_END", DATA_SEGMENT_END },
1778   { "DATA_SEGMENT_RELRO_END", DATA_SEGMENT_RELRO_END },
1779   { "DEFINED", DEFINED },
1780   { "DSECT", DSECT },
1781   { "ENTRY", ENTRY },
1782   { "EXCLUDE_FILE", EXCLUDE_FILE },
1783   { "EXTERN", EXTERN },
1784   { "FILL", FILL },
1785   { "FLOAT", FLOAT },
1786   { "FORCE_COMMON_ALLOCATION", FORCE_COMMON_ALLOCATION },
1787   { "GROUP", GROUP },
1788   { "HIDDEN", HIDDEN },
1789   { "HLL", HLL },
1790   { "INCLUDE", INCLUDE },
1791   { "INFO", INFO },
1792   { "INHIBIT_COMMON_ALLOCATION", INHIBIT_COMMON_ALLOCATION },
1793   { "INPUT", INPUT },
1794   { "KEEP", KEEP },
1795   { "LENGTH", LENGTH },
1796   { "LOADADDR", LOADADDR },
1797   { "LONG", LONG },
1798   { "MAP", MAP },
1799   { "MAX", MAX_K },
1800   { "MEMORY", MEMORY },
1801   { "MIN", MIN_K },
1802   { "NEXT", NEXT },
1803   { "NOCROSSREFS", NOCROSSREFS },
1804   { "NOFLOAT", NOFLOAT },
1805   { "NOLOAD", NOLOAD },
1806   { "ONLY_IF_RO", ONLY_IF_RO },
1807   { "ONLY_IF_RW", ONLY_IF_RW },
1808   { "OPTION", OPTION },
1809   { "ORIGIN", ORIGIN },
1810   { "OUTPUT", OUTPUT },
1811   { "OUTPUT_ARCH", OUTPUT_ARCH },
1812   { "OUTPUT_FORMAT", OUTPUT_FORMAT },
1813   { "OVERLAY", OVERLAY },
1814   { "PHDRS", PHDRS },
1815   { "PROVIDE", PROVIDE },
1816   { "PROVIDE_HIDDEN", PROVIDE_HIDDEN },
1817   { "QUAD", QUAD },
1818   { "SEARCH_DIR", SEARCH_DIR },
1819   { "SECTIONS", SECTIONS },
1820   { "SEGMENT_START", SEGMENT_START },
1821   { "SHORT", SHORT },
1822   { "SIZEOF", SIZEOF },
1823   { "SIZEOF_HEADERS", SIZEOF_HEADERS },
1824   { "SORT", SORT_BY_NAME },
1825   { "SORT_BY_ALIGNMENT", SORT_BY_ALIGNMENT },
1826   { "SORT_BY_INIT_PRIORITY", SORT_BY_INIT_PRIORITY },
1827   { "SORT_BY_NAME", SORT_BY_NAME },
1828   { "SPECIAL", SPECIAL },
1829   { "SQUAD", SQUAD },
1830   { "STARTUP", STARTUP },
1831   { "SUBALIGN", SUBALIGN },
1832   { "SYSLIB", SYSLIB },
1833   { "TARGET", TARGET_K },
1834   { "TRUNCATE", TRUNCATE },
1835   { "VERSION", VERSIONK },
1836   { "global", GLOBAL },
1837   { "l", LENGTH },
1838   { "len", LENGTH },
1839   { "local", LOCAL },
1840   { "o", ORIGIN },
1841   { "org", ORIGIN },
1842   { "sizeof_headers", SIZEOF_HEADERS },
1843 };
1844 
1845 static const Keyword_to_parsecode
1846 script_keywords(&script_keyword_parsecodes[0],
1847                 (sizeof(script_keyword_parsecodes)
1848                  / sizeof(script_keyword_parsecodes[0])));
1849 
1850 static const Keyword_to_parsecode::Keyword_parsecode
1851 version_script_keyword_parsecodes[] =
1852 {
1853   { "extern", EXTERN },
1854   { "global", GLOBAL },
1855   { "local", LOCAL },
1856 };
1857 
1858 static const Keyword_to_parsecode
1859 version_script_keywords(&version_script_keyword_parsecodes[0],
1860                         (sizeof(version_script_keyword_parsecodes)
1861                          / sizeof(version_script_keyword_parsecodes[0])));
1862 
1863 static const Keyword_to_parsecode::Keyword_parsecode
1864 dynamic_list_keyword_parsecodes[] =
1865 {
1866   { "extern", EXTERN },
1867 };
1868 
1869 static const Keyword_to_parsecode
1870 dynamic_list_keywords(&dynamic_list_keyword_parsecodes[0],
1871                       (sizeof(dynamic_list_keyword_parsecodes)
1872                        / sizeof(dynamic_list_keyword_parsecodes[0])));
1873 
1874 
1875 
1876 // Comparison function passed to bsearch.
1877 
1878 extern "C"
1879 {
1880 
1881 struct Ktt_key
1882 {
1883   const char* str;
1884   size_t len;
1885 };
1886 
1887 static int
1888 ktt_compare(const void* keyv, const void* kttv)
1889 {
1890   const Ktt_key* key = static_cast<const Ktt_key*>(keyv);
1891   const Keyword_to_parsecode::Keyword_parsecode* ktt =
1892     static_cast<const Keyword_to_parsecode::Keyword_parsecode*>(kttv);
1893   int i = strncmp(key->str, ktt->keyword, key->len);
1894   if (i != 0)
1895     return i;
1896   if (ktt->keyword[key->len] != '\0')
1897     return -1;
1898   return 0;
1899 }
1900 
1901 } // End extern "C".
1902 
1903 int
1904 Keyword_to_parsecode::keyword_to_parsecode(const char* keyword,
1905                                            size_t len) const
1906 {
1907   Ktt_key key;
1908   key.str = keyword;
1909   key.len = len;
1910   void* kttv = bsearch(&key,
1911                        this->keyword_parsecodes_,
1912                        this->keyword_count_,
1913                        sizeof(this->keyword_parsecodes_[0]),
1914                        ktt_compare);
1915   if (kttv == NULL)
1916     return 0;
1917   Keyword_parsecode* ktt = static_cast<Keyword_parsecode*>(kttv);
1918   return ktt->parsecode;
1919 }
1920 
1921 // The following structs are used within the VersionInfo class as well
1922 // as in the bison helper functions.  They store the information
1923 // parsed from the version script.
1924 
1925 // A single version expression.
1926 // For example, pattern="std::map*" and language="C++".
1927 struct Version_expression
1928 {
1929   Version_expression(const std::string& a_pattern,
1930 		     Version_script_info::Language a_language,
1931                      bool a_exact_match)
1932     : pattern(a_pattern), language(a_language), exact_match(a_exact_match),
1933       was_matched_by_symbol(false)
1934   { }
1935 
1936   std::string pattern;
1937   Version_script_info::Language language;
1938   // If false, we use glob() to match pattern.  If true, we use strcmp().
1939   bool exact_match;
1940   // True if --no-undefined-version is in effect and we found this
1941   // version in get_symbol_version.  We use mutable because this
1942   // struct is generally not modifiable after it has been created.
1943   mutable bool was_matched_by_symbol;
1944 };
1945 
1946 // A list of expressions.
1947 struct Version_expression_list
1948 {
1949   std::vector<struct Version_expression> expressions;
1950 };
1951 
1952 // A list of which versions upon which another version depends.
1953 // Strings should be from the Stringpool.
1954 struct Version_dependency_list
1955 {
1956   std::vector<std::string> dependencies;
1957 };
1958 
1959 // The total definition of a version.  It includes the tag for the
1960 // version, its global and local expressions, and any dependencies.
1961 struct Version_tree
1962 {
1963   Version_tree()
1964       : tag(), global(NULL), local(NULL), dependencies(NULL)
1965   { }
1966 
1967   std::string tag;
1968   const struct Version_expression_list* global;
1969   const struct Version_expression_list* local;
1970   const struct Version_dependency_list* dependencies;
1971 };
1972 
1973 // Helper class that calls cplus_demangle when needed and takes care of freeing
1974 // the result.
1975 
1976 class Lazy_demangler
1977 {
1978  public:
1979   Lazy_demangler(const char* symbol, int options)
1980     : symbol_(symbol), options_(options), demangled_(NULL), did_demangle_(false)
1981   { }
1982 
1983   ~Lazy_demangler()
1984   { free(this->demangled_); }
1985 
1986   // Return the demangled name. The actual demangling happens on the first call,
1987   // and the result is later cached.
1988   inline char*
1989   get();
1990 
1991  private:
1992   // The symbol to demangle.
1993   const char* symbol_;
1994   // Option flags to pass to cplus_demagle.
1995   const int options_;
1996   // The cached demangled value, or NULL if demangling didn't happen yet or
1997   // failed.
1998   char* demangled_;
1999   // Whether we already called cplus_demangle
2000   bool did_demangle_;
2001 };
2002 
2003 // Return the demangled name. The actual demangling happens on the first call,
2004 // and the result is later cached. Returns NULL if the symbol cannot be
2005 // demangled.
2006 
2007 inline char*
2008 Lazy_demangler::get()
2009 {
2010   if (!this->did_demangle_)
2011     {
2012       this->demangled_ = cplus_demangle(this->symbol_, this->options_);
2013       this->did_demangle_ = true;
2014     }
2015   return this->demangled_;
2016 }
2017 
2018 // Class Version_script_info.
2019 
2020 Version_script_info::Version_script_info()
2021   : dependency_lists_(), expression_lists_(), version_trees_(), globs_(),
2022     default_version_(NULL), default_is_global_(false), is_finalized_(false)
2023 {
2024   for (int i = 0; i < LANGUAGE_COUNT; ++i)
2025     this->exact_[i] = NULL;
2026 }
2027 
2028 Version_script_info::~Version_script_info()
2029 {
2030 }
2031 
2032 // Forget all the known version script information.
2033 
2034 void
2035 Version_script_info::clear()
2036 {
2037   for (size_t k = 0; k < this->dependency_lists_.size(); ++k)
2038     delete this->dependency_lists_[k];
2039   this->dependency_lists_.clear();
2040   for (size_t k = 0; k < this->version_trees_.size(); ++k)
2041     delete this->version_trees_[k];
2042   this->version_trees_.clear();
2043   for (size_t k = 0; k < this->expression_lists_.size(); ++k)
2044     delete this->expression_lists_[k];
2045   this->expression_lists_.clear();
2046 }
2047 
2048 // Finalize the version script information.
2049 
2050 void
2051 Version_script_info::finalize()
2052 {
2053   if (!this->is_finalized_)
2054     {
2055       this->build_lookup_tables();
2056       this->is_finalized_ = true;
2057     }
2058 }
2059 
2060 // Return all the versions.
2061 
2062 std::vector<std::string>
2063 Version_script_info::get_versions() const
2064 {
2065   std::vector<std::string> ret;
2066   for (size_t j = 0; j < this->version_trees_.size(); ++j)
2067     if (!this->version_trees_[j]->tag.empty())
2068       ret.push_back(this->version_trees_[j]->tag);
2069   return ret;
2070 }
2071 
2072 // Return the dependencies of VERSION.
2073 
2074 std::vector<std::string>
2075 Version_script_info::get_dependencies(const char* version) const
2076 {
2077   std::vector<std::string> ret;
2078   for (size_t j = 0; j < this->version_trees_.size(); ++j)
2079     if (this->version_trees_[j]->tag == version)
2080       {
2081         const struct Version_dependency_list* deps =
2082           this->version_trees_[j]->dependencies;
2083         if (deps != NULL)
2084           for (size_t k = 0; k < deps->dependencies.size(); ++k)
2085             ret.push_back(deps->dependencies[k]);
2086         return ret;
2087       }
2088   return ret;
2089 }
2090 
2091 // A version script essentially maps a symbol name to a version tag
2092 // and an indication of whether symbol is global or local within that
2093 // version tag.  Each symbol maps to at most one version tag.
2094 // Unfortunately, in practice, version scripts are ambiguous, and list
2095 // symbols multiple times.  Thus, we have to document the matching
2096 // process.
2097 
2098 // This is a description of what the GNU linker does as of 2010-01-11.
2099 // It walks through the version tags in the order in which they appear
2100 // in the version script.  For each tag, it first walks through the
2101 // global patterns for that tag, then the local patterns.  When
2102 // looking at a single pattern, it first applies any language specific
2103 // demangling as specified for the pattern, and then matches the
2104 // resulting symbol name to the pattern.  If it finds an exact match
2105 // for a literal pattern (a pattern enclosed in quotes or with no
2106 // wildcard characters), then that is the match that it uses.  If
2107 // finds a match with a wildcard pattern, then it saves it and
2108 // continues searching.  Wildcard patterns that are exactly "*" are
2109 // saved separately.
2110 
2111 // If no exact match with a literal pattern is ever found, then if a
2112 // wildcard match with a global pattern was found it is used,
2113 // otherwise if a wildcard match with a local pattern was found it is
2114 // used.
2115 
2116 // This is the result:
2117 //   * If there is an exact match, then we use the first tag in the
2118 //     version script where it matches.
2119 //     + If the exact match in that tag is global, it is used.
2120 //     + Otherwise the exact match in that tag is local, and is used.
2121 //   * Otherwise, if there is any match with a global wildcard pattern:
2122 //     + If there is any match with a wildcard pattern which is not
2123 //       "*", then we use the tag in which the *last* such pattern
2124 //       appears.
2125 //     + Otherwise, we matched "*".  If there is no match with a local
2126 //       wildcard pattern which is not "*", then we use the *last*
2127 //       match with a global "*".  Otherwise, continue.
2128 //   * Otherwise, if there is any match with a local wildcard pattern:
2129 //     + If there is any match with a wildcard pattern which is not
2130 //       "*", then we use the tag in which the *last* such pattern
2131 //       appears.
2132 //     + Otherwise, we matched "*", and we use the tag in which the
2133 //       *last* such match occurred.
2134 
2135 // There is an additional wrinkle.  When the GNU linker finds a symbol
2136 // with a version defined in an object file due to a .symver
2137 // directive, it looks up that symbol name in that version tag.  If it
2138 // finds it, it matches the symbol name against the patterns for that
2139 // version.  If there is no match with a global pattern, but there is
2140 // a match with a local pattern, then the GNU linker marks the symbol
2141 // as local.
2142 
2143 // We want gold to be generally compatible, but we also want gold to
2144 // be fast.  These are the rules that gold implements:
2145 //   * If there is an exact match for the mangled name, we use it.
2146 //     + If there is more than one exact match, we give a warning, and
2147 //       we use the first tag in the script which matches.
2148 //     + If a symbol has an exact match as both global and local for
2149 //       the same version tag, we give an error.
2150 //   * Otherwise, we look for an extern C++ or an extern Java exact
2151 //     match.  If we find an exact match, we use it.
2152 //     + If there is more than one exact match, we give a warning, and
2153 //       we use the first tag in the script which matches.
2154 //     + If a symbol has an exact match as both global and local for
2155 //       the same version tag, we give an error.
2156 //   * Otherwise, we look through the wildcard patterns, ignoring "*"
2157 //     patterns.  We look through the version tags in reverse order.
2158 //     For each version tag, we look through the global patterns and
2159 //     then the local patterns.  We use the first match we find (i.e.,
2160 //     the last matching version tag in the file).
2161 //   * Otherwise, we use the "*" pattern if there is one.  We give an
2162 //     error if there are multiple "*" patterns.
2163 
2164 // At least for now, gold does not look up the version tag for a
2165 // symbol version found in an object file to see if it should be
2166 // forced local.  There are other ways to force a symbol to be local,
2167 // and I don't understand why this one is useful.
2168 
2169 // Build a set of fast lookup tables for a version script.
2170 
2171 void
2172 Version_script_info::build_lookup_tables()
2173 {
2174   size_t size = this->version_trees_.size();
2175   for (size_t j = 0; j < size; ++j)
2176     {
2177       const Version_tree* v = this->version_trees_[j];
2178       this->build_expression_list_lookup(v->local, v, false);
2179       this->build_expression_list_lookup(v->global, v, true);
2180     }
2181 }
2182 
2183 // If a pattern has backlashes but no unquoted wildcard characters,
2184 // then we apply backslash unquoting and look for an exact match.
2185 // Otherwise we treat it as a wildcard pattern.  This function returns
2186 // true for a wildcard pattern.  Otherwise, it does backslash
2187 // unquoting on *PATTERN and returns false.  If this returns true,
2188 // *PATTERN may have been partially unquoted.
2189 
2190 bool
2191 Version_script_info::unquote(std::string* pattern) const
2192 {
2193   bool saw_backslash = false;
2194   size_t len = pattern->length();
2195   size_t j = 0;
2196   for (size_t i = 0; i < len; ++i)
2197     {
2198       if (saw_backslash)
2199 	saw_backslash = false;
2200       else
2201 	{
2202 	  switch ((*pattern)[i])
2203 	    {
2204 	    case '?': case '[': case '*':
2205 	      return true;
2206 	    case '\\':
2207 	      saw_backslash = true;
2208 	      continue;
2209 	    default:
2210 	      break;
2211 	    }
2212 	}
2213 
2214       if (i != j)
2215 	(*pattern)[j] = (*pattern)[i];
2216       ++j;
2217     }
2218   return false;
2219 }
2220 
2221 // Add an exact match for MATCH to *PE.  The result of the match is
2222 // V/IS_GLOBAL.
2223 
2224 void
2225 Version_script_info::add_exact_match(const std::string& match,
2226 				     const Version_tree* v, bool is_global,
2227 				     const Version_expression* ve,
2228 				     Exact* pe)
2229 {
2230   std::pair<Exact::iterator, bool> ins =
2231     pe->insert(std::make_pair(match, Version_tree_match(v, is_global, ve)));
2232   if (ins.second)
2233     {
2234       // This is the first time we have seen this match.
2235       return;
2236     }
2237 
2238   Version_tree_match& vtm(ins.first->second);
2239   if (vtm.real->tag != v->tag)
2240     {
2241       // This is an ambiguous match.  We still return the
2242       // first version that we found in the script, but we
2243       // record the new version to issue a warning if we
2244       // wind up looking up this symbol.
2245       if (vtm.ambiguous == NULL)
2246 	vtm.ambiguous = v;
2247     }
2248   else if (is_global != vtm.is_global)
2249     {
2250       // We have a match for both the global and local entries for a
2251       // version tag.  That's got to be wrong.
2252       gold_error(_("'%s' appears as both a global and a local symbol "
2253 		   "for version '%s' in script"),
2254 		 match.c_str(), v->tag.c_str());
2255     }
2256 }
2257 
2258 // Build fast lookup information for EXPLIST and store it in LOOKUP.
2259 // All matches go to V, and IS_GLOBAL is true if they are global
2260 // matches.
2261 
2262 void
2263 Version_script_info::build_expression_list_lookup(
2264     const Version_expression_list* explist,
2265     const Version_tree* v,
2266     bool is_global)
2267 {
2268   if (explist == NULL)
2269     return;
2270   size_t size = explist->expressions.size();
2271   for (size_t i = 0; i < size; ++i)
2272     {
2273       const Version_expression& exp(explist->expressions[i]);
2274 
2275       if (exp.pattern.length() == 1 && exp.pattern[0] == '*')
2276 	{
2277 	  if (this->default_version_ != NULL
2278 	      && this->default_version_->tag != v->tag)
2279 	    gold_warning(_("wildcard match appears in both version '%s' "
2280 			   "and '%s' in script"),
2281 			 this->default_version_->tag.c_str(), v->tag.c_str());
2282 	  else if (this->default_version_ != NULL
2283 		   && this->default_is_global_ != is_global)
2284 	    gold_error(_("wildcard match appears as both global and local "
2285 			 "in version '%s' in script"),
2286 		       v->tag.c_str());
2287 	  this->default_version_ = v;
2288 	  this->default_is_global_ = is_global;
2289 	  continue;
2290 	}
2291 
2292       std::string pattern = exp.pattern;
2293       if (!exp.exact_match)
2294 	{
2295 	  if (this->unquote(&pattern))
2296 	    {
2297 	      this->globs_.push_back(Glob(&exp, v, is_global));
2298 	      continue;
2299 	    }
2300 	}
2301 
2302       if (this->exact_[exp.language] == NULL)
2303 	this->exact_[exp.language] = new Exact();
2304       this->add_exact_match(pattern, v, is_global, &exp,
2305 			    this->exact_[exp.language]);
2306     }
2307 }
2308 
2309 // Return the name to match given a name, a language code, and two
2310 // lazy demanglers.
2311 
2312 const char*
2313 Version_script_info::get_name_to_match(const char* name,
2314 				       int language,
2315 				       Lazy_demangler* cpp_demangler,
2316 				       Lazy_demangler* java_demangler) const
2317 {
2318   switch (language)
2319     {
2320     case LANGUAGE_C:
2321       return name;
2322     case LANGUAGE_CXX:
2323       return cpp_demangler->get();
2324     case LANGUAGE_JAVA:
2325       return java_demangler->get();
2326     default:
2327       gold_unreachable();
2328     }
2329 }
2330 
2331 // Look up SYMBOL_NAME in the list of versions.  Return true if the
2332 // symbol is found, false if not.  If the symbol is found, then if
2333 // PVERSION is not NULL, set *PVERSION to the version tag, and if
2334 // P_IS_GLOBAL is not NULL, set *P_IS_GLOBAL according to whether the
2335 // symbol is global or not.
2336 
2337 bool
2338 Version_script_info::get_symbol_version(const char* symbol_name,
2339 					std::string* pversion,
2340 					bool* p_is_global) const
2341 {
2342   Lazy_demangler cpp_demangled_name(symbol_name, DMGL_ANSI | DMGL_PARAMS);
2343   Lazy_demangler java_demangled_name(symbol_name,
2344 				     DMGL_ANSI | DMGL_PARAMS | DMGL_JAVA);
2345 
2346   gold_assert(this->is_finalized_);
2347   for (int i = 0; i < LANGUAGE_COUNT; ++i)
2348     {
2349       Exact* exact = this->exact_[i];
2350       if (exact == NULL)
2351 	continue;
2352 
2353       const char* name_to_match = this->get_name_to_match(symbol_name, i,
2354 							  &cpp_demangled_name,
2355 							  &java_demangled_name);
2356       if (name_to_match == NULL)
2357 	{
2358 	  // If the name can not be demangled, the GNU linker goes
2359 	  // ahead and tries to match it anyhow.  That does not
2360 	  // make sense to me and I have not implemented it.
2361 	  continue;
2362 	}
2363 
2364       Exact::const_iterator pe = exact->find(name_to_match);
2365       if (pe != exact->end())
2366 	{
2367 	  const Version_tree_match& vtm(pe->second);
2368 	  if (vtm.ambiguous != NULL)
2369 	    gold_warning(_("using '%s' as version for '%s' which is also "
2370 			   "named in version '%s' in script"),
2371 			 vtm.real->tag.c_str(), name_to_match,
2372 			 vtm.ambiguous->tag.c_str());
2373 
2374 	  if (pversion != NULL)
2375 	    *pversion = vtm.real->tag;
2376 	  if (p_is_global != NULL)
2377 	    *p_is_global = vtm.is_global;
2378 
2379 	  // If we are using --no-undefined-version, and this is a
2380 	  // global symbol, we have to record that we have found this
2381 	  // symbol, so that we don't warn about it.  We have to do
2382 	  // this now, because otherwise we have no way to get from a
2383 	  // non-C language back to the demangled name that we
2384 	  // matched.
2385 	  if (p_is_global != NULL && vtm.is_global)
2386 	    vtm.expression->was_matched_by_symbol = true;
2387 
2388 	  return true;
2389 	}
2390     }
2391 
2392   // Look through the glob patterns in reverse order.
2393 
2394   for (Globs::const_reverse_iterator p = this->globs_.rbegin();
2395        p != this->globs_.rend();
2396        ++p)
2397     {
2398       int language = p->expression->language;
2399       const char* name_to_match = this->get_name_to_match(symbol_name,
2400 							  language,
2401 							  &cpp_demangled_name,
2402 							  &java_demangled_name);
2403       if (name_to_match == NULL)
2404 	continue;
2405 
2406       if (fnmatch(p->expression->pattern.c_str(), name_to_match,
2407 		  FNM_NOESCAPE) == 0)
2408 	{
2409 	  if (pversion != NULL)
2410 	    *pversion = p->version->tag;
2411 	  if (p_is_global != NULL)
2412 	    *p_is_global = p->is_global;
2413 	  return true;
2414 	}
2415     }
2416 
2417   // Finally, there may be a wildcard.
2418   if (this->default_version_ != NULL)
2419     {
2420       if (pversion != NULL)
2421 	*pversion = this->default_version_->tag;
2422       if (p_is_global != NULL)
2423 	*p_is_global = this->default_is_global_;
2424       return true;
2425     }
2426 
2427   return false;
2428 }
2429 
2430 // Give an error if any exact symbol names (not wildcards) appear in a
2431 // version script, but there is no such symbol.
2432 
2433 void
2434 Version_script_info::check_unmatched_names(const Symbol_table* symtab) const
2435 {
2436   for (size_t i = 0; i < this->version_trees_.size(); ++i)
2437     {
2438       const Version_tree* vt = this->version_trees_[i];
2439       if (vt->global == NULL)
2440 	continue;
2441       for (size_t j = 0; j < vt->global->expressions.size(); ++j)
2442 	{
2443 	  const Version_expression& expression(vt->global->expressions[j]);
2444 
2445 	  // Ignore cases where we used the version because we saw a
2446 	  // symbol that we looked up.  Note that
2447 	  // WAS_MATCHED_BY_SYMBOL will be true even if the symbol was
2448 	  // not a definition.  That's OK as in that case we most
2449 	  // likely gave an undefined symbol error anyhow.
2450 	  if (expression.was_matched_by_symbol)
2451 	    continue;
2452 
2453 	  // Just ignore names which are in languages other than C.
2454 	  // We have no way to look them up in the symbol table.
2455 	  if (expression.language != LANGUAGE_C)
2456 	    continue;
2457 
2458 	  // Remove backslash quoting, and ignore wildcard patterns.
2459 	  std::string pattern = expression.pattern;
2460 	  if (!expression.exact_match)
2461 	    {
2462 	      if (this->unquote(&pattern))
2463 		continue;
2464 	    }
2465 
2466 	  if (symtab->lookup(pattern.c_str(), vt->tag.c_str()) == NULL)
2467 	    gold_error(_("version script assignment of %s to symbol %s "
2468 			 "failed: symbol not defined"),
2469 		       vt->tag.c_str(), pattern.c_str());
2470 	}
2471     }
2472 }
2473 
2474 struct Version_dependency_list*
2475 Version_script_info::allocate_dependency_list()
2476 {
2477   dependency_lists_.push_back(new Version_dependency_list);
2478   return dependency_lists_.back();
2479 }
2480 
2481 struct Version_expression_list*
2482 Version_script_info::allocate_expression_list()
2483 {
2484   expression_lists_.push_back(new Version_expression_list);
2485   return expression_lists_.back();
2486 }
2487 
2488 struct Version_tree*
2489 Version_script_info::allocate_version_tree()
2490 {
2491   version_trees_.push_back(new Version_tree);
2492   return version_trees_.back();
2493 }
2494 
2495 // Print for debugging.
2496 
2497 void
2498 Version_script_info::print(FILE* f) const
2499 {
2500   if (this->empty())
2501     return;
2502 
2503   fprintf(f, "VERSION {");
2504 
2505   for (size_t i = 0; i < this->version_trees_.size(); ++i)
2506     {
2507       const Version_tree* vt = this->version_trees_[i];
2508 
2509       if (vt->tag.empty())
2510 	fprintf(f, "  {\n");
2511       else
2512 	fprintf(f, "  %s {\n", vt->tag.c_str());
2513 
2514       if (vt->global != NULL)
2515 	{
2516 	  fprintf(f, "    global :\n");
2517 	  this->print_expression_list(f, vt->global);
2518 	}
2519 
2520       if (vt->local != NULL)
2521 	{
2522 	  fprintf(f, "    local :\n");
2523 	  this->print_expression_list(f, vt->local);
2524 	}
2525 
2526       fprintf(f, "  }");
2527       if (vt->dependencies != NULL)
2528 	{
2529 	  const Version_dependency_list* deps = vt->dependencies;
2530 	  for (size_t j = 0; j < deps->dependencies.size(); ++j)
2531 	    {
2532 	      if (j < deps->dependencies.size() - 1)
2533 		fprintf(f, "\n");
2534 	      fprintf(f, "    %s", deps->dependencies[j].c_str());
2535 	    }
2536 	}
2537       fprintf(f, ";\n");
2538     }
2539 
2540   fprintf(f, "}\n");
2541 }
2542 
2543 void
2544 Version_script_info::print_expression_list(
2545     FILE* f,
2546     const Version_expression_list* vel) const
2547 {
2548   Version_script_info::Language current_language = LANGUAGE_C;
2549   for (size_t i = 0; i < vel->expressions.size(); ++i)
2550     {
2551       const Version_expression& ve(vel->expressions[i]);
2552 
2553       if (ve.language != current_language)
2554 	{
2555 	  if (current_language != LANGUAGE_C)
2556 	    fprintf(f, "      }\n");
2557 	  switch (ve.language)
2558 	    {
2559 	    case LANGUAGE_C:
2560 	      break;
2561 	    case LANGUAGE_CXX:
2562 	      fprintf(f, "      extern \"C++\" {\n");
2563 	      break;
2564 	    case LANGUAGE_JAVA:
2565 	      fprintf(f, "      extern \"Java\" {\n");
2566 	      break;
2567 	    default:
2568 	      gold_unreachable();
2569 	    }
2570 	  current_language = ve.language;
2571 	}
2572 
2573       fprintf(f, "      ");
2574       if (current_language != LANGUAGE_C)
2575 	fprintf(f, "  ");
2576 
2577       if (ve.exact_match)
2578 	fprintf(f, "\"");
2579       fprintf(f, "%s", ve.pattern.c_str());
2580       if (ve.exact_match)
2581 	fprintf(f, "\"");
2582 
2583       fprintf(f, "\n");
2584     }
2585 
2586   if (current_language != LANGUAGE_C)
2587     fprintf(f, "      }\n");
2588 }
2589 
2590 } // End namespace gold.
2591 
2592 // The remaining functions are extern "C", so it's clearer to not put
2593 // them in namespace gold.
2594 
2595 using namespace gold;
2596 
2597 // This function is called by the bison parser to return the next
2598 // token.
2599 
2600 extern "C" int
2601 yylex(YYSTYPE* lvalp, void* closurev)
2602 {
2603   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2604   const Token* token = closure->next_token();
2605   switch (token->classification())
2606     {
2607     default:
2608       gold_unreachable();
2609 
2610     case Token::TOKEN_INVALID:
2611       yyerror(closurev, "invalid character");
2612       return 0;
2613 
2614     case Token::TOKEN_EOF:
2615       return 0;
2616 
2617     case Token::TOKEN_STRING:
2618       {
2619 	// This is either a keyword or a STRING.
2620 	size_t len;
2621 	const char* str = token->string_value(&len);
2622 	int parsecode = 0;
2623         switch (closure->lex_mode())
2624           {
2625           case Lex::LINKER_SCRIPT:
2626             parsecode = script_keywords.keyword_to_parsecode(str, len);
2627             break;
2628           case Lex::VERSION_SCRIPT:
2629             parsecode = version_script_keywords.keyword_to_parsecode(str, len);
2630             break;
2631           case Lex::DYNAMIC_LIST:
2632             parsecode = dynamic_list_keywords.keyword_to_parsecode(str, len);
2633             break;
2634           default:
2635             break;
2636           }
2637 	if (parsecode != 0)
2638 	  return parsecode;
2639 	lvalp->string.value = str;
2640 	lvalp->string.length = len;
2641 	return STRING;
2642       }
2643 
2644     case Token::TOKEN_QUOTED_STRING:
2645       lvalp->string.value = token->string_value(&lvalp->string.length);
2646       return QUOTED_STRING;
2647 
2648     case Token::TOKEN_OPERATOR:
2649       return token->operator_value();
2650 
2651     case Token::TOKEN_INTEGER:
2652       lvalp->integer = token->integer_value();
2653       return INTEGER;
2654     }
2655 }
2656 
2657 // This function is called by the bison parser to report an error.
2658 
2659 extern "C" void
2660 yyerror(void* closurev, const char* message)
2661 {
2662   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2663   gold_error(_("%s:%d:%d: %s"), closure->filename(), closure->lineno(),
2664 	     closure->charpos(), message);
2665 }
2666 
2667 // Called by the bison parser to add an external symbol to the link.
2668 
2669 extern "C" void
2670 script_add_extern(void* closurev, const char* name, size_t length)
2671 {
2672   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2673   closure->script_options()->add_symbol_reference(name, length);
2674 }
2675 
2676 // Called by the bison parser to add a file to the link.
2677 
2678 extern "C" void
2679 script_add_file(void* closurev, const char* name, size_t length)
2680 {
2681   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2682 
2683   // If this is an absolute path, and we found the script in the
2684   // sysroot, then we want to prepend the sysroot to the file name.
2685   // For example, this is how we handle a cross link to the x86_64
2686   // libc.so, which refers to /lib/libc.so.6.
2687   std::string name_string(name, length);
2688   const char* extra_search_path = ".";
2689   std::string script_directory;
2690   if (IS_ABSOLUTE_PATH(name_string.c_str()))
2691     {
2692       if (closure->is_in_sysroot())
2693 	{
2694 	  const std::string& sysroot(parameters->options().sysroot());
2695 	  gold_assert(!sysroot.empty());
2696 	  name_string = sysroot + name_string;
2697 	}
2698     }
2699   else
2700     {
2701       // In addition to checking the normal library search path, we
2702       // also want to check in the script-directory.
2703       const char* slash = strrchr(closure->filename(), '/');
2704       if (slash != NULL)
2705 	{
2706 	  script_directory.assign(closure->filename(),
2707 				  slash - closure->filename() + 1);
2708 	  extra_search_path = script_directory.c_str();
2709 	}
2710     }
2711 
2712   Input_file_argument file(name_string.c_str(),
2713 			   Input_file_argument::INPUT_FILE_TYPE_FILE,
2714 			   extra_search_path, false,
2715 			   closure->position_dependent_options());
2716   Input_argument& arg = closure->inputs()->add_file(file);
2717   arg.set_script_info(closure->script_info());
2718 }
2719 
2720 // Called by the bison parser to add a library to the link.
2721 
2722 extern "C" void
2723 script_add_library(void* closurev, const char* name, size_t length)
2724 {
2725   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2726   std::string name_string(name, length);
2727 
2728   if (name_string[0] != 'l')
2729     gold_error(_("library name must be prefixed with -l"));
2730 
2731   Input_file_argument file(name_string.c_str() + 1,
2732 			   Input_file_argument::INPUT_FILE_TYPE_LIBRARY,
2733 			   "", false,
2734 			   closure->position_dependent_options());
2735   Input_argument& arg = closure->inputs()->add_file(file);
2736   arg.set_script_info(closure->script_info());
2737 }
2738 
2739 // Called by the bison parser to start a group.  If we are already in
2740 // a group, that means that this script was invoked within a
2741 // --start-group --end-group sequence on the command line, or that
2742 // this script was found in a GROUP of another script.  In that case,
2743 // we simply continue the existing group, rather than starting a new
2744 // one.  It is possible to construct a case in which this will do
2745 // something other than what would happen if we did a recursive group,
2746 // but it's hard to imagine why the different behaviour would be
2747 // useful for a real program.  Avoiding recursive groups is simpler
2748 // and more efficient.
2749 
2750 extern "C" void
2751 script_start_group(void* closurev)
2752 {
2753   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2754   if (!closure->in_group())
2755     closure->inputs()->start_group();
2756 }
2757 
2758 // Called by the bison parser at the end of a group.
2759 
2760 extern "C" void
2761 script_end_group(void* closurev)
2762 {
2763   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2764   if (!closure->in_group())
2765     closure->inputs()->end_group();
2766 }
2767 
2768 // Called by the bison parser to start an AS_NEEDED list.
2769 
2770 extern "C" void
2771 script_start_as_needed(void* closurev)
2772 {
2773   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2774   closure->position_dependent_options().set_as_needed(true);
2775 }
2776 
2777 // Called by the bison parser at the end of an AS_NEEDED list.
2778 
2779 extern "C" void
2780 script_end_as_needed(void* closurev)
2781 {
2782   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2783   closure->position_dependent_options().set_as_needed(false);
2784 }
2785 
2786 // Called by the bison parser to set the entry symbol.
2787 
2788 extern "C" void
2789 script_set_entry(void* closurev, const char* entry, size_t length)
2790 {
2791   // We'll parse this exactly the same as --entry=ENTRY on the commandline
2792   // TODO(csilvers): FIXME -- call set_entry directly.
2793   std::string arg("--entry=");
2794   arg.append(entry, length);
2795   script_parse_option(closurev, arg.c_str(), arg.size());
2796 }
2797 
2798 // Called by the bison parser to set whether to define common symbols.
2799 
2800 extern "C" void
2801 script_set_common_allocation(void* closurev, int set)
2802 {
2803   const char* arg = set != 0 ? "--define-common" : "--no-define-common";
2804   script_parse_option(closurev, arg, strlen(arg));
2805 }
2806 
2807 // Called by the bison parser to refer to a symbol.
2808 
2809 extern "C" Expression*
2810 script_symbol(void* closurev, const char* name, size_t length)
2811 {
2812   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2813   if (length != 1 || name[0] != '.')
2814     closure->script_options()->add_symbol_reference(name, length);
2815   return script_exp_string(name, length);
2816 }
2817 
2818 // Called by the bison parser to define a symbol.
2819 
2820 extern "C" void
2821 script_set_symbol(void* closurev, const char* name, size_t length,
2822 		  Expression* value, int providei, int hiddeni)
2823 {
2824   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2825   const bool provide = providei != 0;
2826   const bool hidden = hiddeni != 0;
2827   closure->script_options()->add_symbol_assignment(name, length,
2828 						   closure->parsing_defsym(),
2829 						   value, provide, hidden);
2830   closure->clear_skip_on_incompatible_target();
2831 }
2832 
2833 // Called by the bison parser to add an assertion.
2834 
2835 extern "C" void
2836 script_add_assertion(void* closurev, Expression* check, const char* message,
2837 		     size_t messagelen)
2838 {
2839   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2840   closure->script_options()->add_assertion(check, message, messagelen);
2841   closure->clear_skip_on_incompatible_target();
2842 }
2843 
2844 // Called by the bison parser to parse an OPTION.
2845 
2846 extern "C" void
2847 script_parse_option(void* closurev, const char* option, size_t length)
2848 {
2849   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2850   // We treat the option as a single command-line option, even if
2851   // it has internal whitespace.
2852   if (closure->command_line() == NULL)
2853     {
2854       // There are some options that we could handle here--e.g.,
2855       // -lLIBRARY.  Should we bother?
2856       gold_warning(_("%s:%d:%d: ignoring command OPTION; OPTION is only valid"
2857 		     " for scripts specified via -T/--script"),
2858 		   closure->filename(), closure->lineno(), closure->charpos());
2859     }
2860   else
2861     {
2862       bool past_a_double_dash_option = false;
2863       const char* mutable_option = strndup(option, length);
2864       gold_assert(mutable_option != NULL);
2865       closure->command_line()->process_one_option(1, &mutable_option, 0,
2866                                                   &past_a_double_dash_option);
2867       // The General_options class will quite possibly store a pointer
2868       // into mutable_option, so we can't free it.  In cases the class
2869       // does not store such a pointer, this is a memory leak.  Alas. :(
2870     }
2871   closure->clear_skip_on_incompatible_target();
2872 }
2873 
2874 // Called by the bison parser to handle OUTPUT_FORMAT.  OUTPUT_FORMAT
2875 // takes either one or three arguments.  In the three argument case,
2876 // the format depends on the endianness option, which we don't
2877 // currently support (FIXME).  If we see an OUTPUT_FORMAT for the
2878 // wrong format, then we want to search for a new file.  Returning 0
2879 // here will cause the parser to immediately abort.
2880 
2881 extern "C" int
2882 script_check_output_format(void* closurev,
2883 			   const char* default_name, size_t default_length,
2884 			   const char*, size_t, const char*, size_t)
2885 {
2886   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2887   std::string name(default_name, default_length);
2888   Target* target = select_target_by_bfd_name(name.c_str());
2889   if (target == NULL || !parameters->is_compatible_target(target))
2890     {
2891       if (closure->skip_on_incompatible_target())
2892 	{
2893 	  closure->set_found_incompatible_target();
2894 	  return 0;
2895 	}
2896       // FIXME: Should we warn about the unknown target?
2897     }
2898   return 1;
2899 }
2900 
2901 // Called by the bison parser to handle TARGET.
2902 
2903 extern "C" void
2904 script_set_target(void* closurev, const char* target, size_t len)
2905 {
2906   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2907   std::string s(target, len);
2908   General_options::Object_format format_enum;
2909   format_enum = General_options::string_to_object_format(s.c_str());
2910   closure->position_dependent_options().set_format_enum(format_enum);
2911 }
2912 
2913 // Called by the bison parser to handle SEARCH_DIR.  This is handled
2914 // exactly like a -L option.
2915 
2916 extern "C" void
2917 script_add_search_dir(void* closurev, const char* option, size_t length)
2918 {
2919   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2920   if (closure->command_line() == NULL)
2921     gold_warning(_("%s:%d:%d: ignoring SEARCH_DIR; SEARCH_DIR is only valid"
2922 		   " for scripts specified via -T/--script"),
2923 		 closure->filename(), closure->lineno(), closure->charpos());
2924   else if (!closure->command_line()->options().nostdlib())
2925     {
2926       std::string s = "-L" + std::string(option, length);
2927       script_parse_option(closurev, s.c_str(), s.size());
2928     }
2929 }
2930 
2931 /* Called by the bison parser to push the lexer into expression
2932    mode.  */
2933 
2934 extern "C" void
2935 script_push_lex_into_expression_mode(void* closurev)
2936 {
2937   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2938   closure->push_lex_mode(Lex::EXPRESSION);
2939 }
2940 
2941 /* Called by the bison parser to push the lexer into version
2942    mode.  */
2943 
2944 extern "C" void
2945 script_push_lex_into_version_mode(void* closurev)
2946 {
2947   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2948   if (closure->version_script()->is_finalized())
2949     gold_error(_("%s:%d:%d: invalid use of VERSION in input file"),
2950 	       closure->filename(), closure->lineno(), closure->charpos());
2951   closure->push_lex_mode(Lex::VERSION_SCRIPT);
2952 }
2953 
2954 /* Called by the bison parser to pop the lexer mode.  */
2955 
2956 extern "C" void
2957 script_pop_lex_mode(void* closurev)
2958 {
2959   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2960   closure->pop_lex_mode();
2961 }
2962 
2963 // Register an entire version node. For example:
2964 //
2965 // GLIBC_2.1 {
2966 //   global: foo;
2967 // } GLIBC_2.0;
2968 //
2969 // - tag is "GLIBC_2.1"
2970 // - tree contains the information "global: foo"
2971 // - deps contains "GLIBC_2.0"
2972 
2973 extern "C" void
2974 script_register_vers_node(void*,
2975 			  const char* tag,
2976 			  int taglen,
2977 			  struct Version_tree* tree,
2978 			  struct Version_dependency_list* deps)
2979 {
2980   gold_assert(tree != NULL);
2981   tree->dependencies = deps;
2982   if (tag != NULL)
2983     tree->tag = std::string(tag, taglen);
2984 }
2985 
2986 // Add a dependencies to the list of existing dependencies, if any,
2987 // and return the expanded list.
2988 
2989 extern "C" struct Version_dependency_list*
2990 script_add_vers_depend(void* closurev,
2991 		       struct Version_dependency_list* all_deps,
2992 		       const char* depend_to_add, int deplen)
2993 {
2994   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
2995   if (all_deps == NULL)
2996     all_deps = closure->version_script()->allocate_dependency_list();
2997   all_deps->dependencies.push_back(std::string(depend_to_add, deplen));
2998   return all_deps;
2999 }
3000 
3001 // Add a pattern expression to an existing list of expressions, if any.
3002 
3003 extern "C" struct Version_expression_list*
3004 script_new_vers_pattern(void* closurev,
3005 			struct Version_expression_list* expressions,
3006 			const char* pattern, int patlen, int exact_match)
3007 {
3008   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3009   if (expressions == NULL)
3010     expressions = closure->version_script()->allocate_expression_list();
3011   expressions->expressions.push_back(
3012       Version_expression(std::string(pattern, patlen),
3013                          closure->get_current_language(),
3014                          static_cast<bool>(exact_match)));
3015   return expressions;
3016 }
3017 
3018 // Attaches b to the end of a, and clears b.  So a = a + b and b = {}.
3019 
3020 extern "C" struct Version_expression_list*
3021 script_merge_expressions(struct Version_expression_list* a,
3022                          struct Version_expression_list* b)
3023 {
3024   a->expressions.insert(a->expressions.end(),
3025                         b->expressions.begin(), b->expressions.end());
3026   // We could delete b and remove it from expressions_lists_, but
3027   // that's a lot of work.  This works just as well.
3028   b->expressions.clear();
3029   return a;
3030 }
3031 
3032 // Combine the global and local expressions into a a Version_tree.
3033 
3034 extern "C" struct Version_tree*
3035 script_new_vers_node(void* closurev,
3036 		     struct Version_expression_list* global,
3037 		     struct Version_expression_list* local)
3038 {
3039   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3040   Version_tree* tree = closure->version_script()->allocate_version_tree();
3041   tree->global = global;
3042   tree->local = local;
3043   return tree;
3044 }
3045 
3046 // Handle a transition in language, such as at the
3047 // start or end of 'extern "C++"'
3048 
3049 extern "C" void
3050 version_script_push_lang(void* closurev, const char* lang, int langlen)
3051 {
3052   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3053   std::string language(lang, langlen);
3054   Version_script_info::Language code;
3055   if (language.empty() || language == "C")
3056     code = Version_script_info::LANGUAGE_C;
3057   else if (language == "C++")
3058     code = Version_script_info::LANGUAGE_CXX;
3059   else if (language == "Java")
3060     code = Version_script_info::LANGUAGE_JAVA;
3061   else
3062     {
3063       char* buf = new char[langlen + 100];
3064       snprintf(buf, langlen + 100,
3065 	       _("unrecognized version script language '%s'"),
3066 	       language.c_str());
3067       yyerror(closurev, buf);
3068       delete[] buf;
3069       code = Version_script_info::LANGUAGE_C;
3070     }
3071   closure->push_language(code);
3072 }
3073 
3074 extern "C" void
3075 version_script_pop_lang(void* closurev)
3076 {
3077   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3078   closure->pop_language();
3079 }
3080 
3081 // Called by the bison parser to start a SECTIONS clause.
3082 
3083 extern "C" void
3084 script_start_sections(void* closurev)
3085 {
3086   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3087   closure->script_options()->script_sections()->start_sections();
3088   closure->clear_skip_on_incompatible_target();
3089 }
3090 
3091 // Called by the bison parser to finish a SECTIONS clause.
3092 
3093 extern "C" void
3094 script_finish_sections(void* closurev)
3095 {
3096   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3097   closure->script_options()->script_sections()->finish_sections();
3098 }
3099 
3100 // Start processing entries for an output section.
3101 
3102 extern "C" void
3103 script_start_output_section(void* closurev, const char* name, size_t namelen,
3104 			    const struct Parser_output_section_header* header)
3105 {
3106   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3107   closure->script_options()->script_sections()->start_output_section(name,
3108 								     namelen,
3109 								     header);
3110 }
3111 
3112 // Finish processing entries for an output section.
3113 
3114 extern "C" void
3115 script_finish_output_section(void* closurev,
3116 			     const struct Parser_output_section_trailer* trail)
3117 {
3118   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3119   closure->script_options()->script_sections()->finish_output_section(trail);
3120 }
3121 
3122 // Add a data item (e.g., "WORD (0)") to the current output section.
3123 
3124 extern "C" void
3125 script_add_data(void* closurev, int data_token, Expression* val)
3126 {
3127   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3128   int size;
3129   bool is_signed = true;
3130   switch (data_token)
3131     {
3132     case QUAD:
3133       size = 8;
3134       is_signed = false;
3135       break;
3136     case SQUAD:
3137       size = 8;
3138       break;
3139     case LONG:
3140       size = 4;
3141       break;
3142     case SHORT:
3143       size = 2;
3144       break;
3145     case BYTE:
3146       size = 1;
3147       break;
3148     default:
3149       gold_unreachable();
3150     }
3151   closure->script_options()->script_sections()->add_data(size, is_signed, val);
3152 }
3153 
3154 // Add a clause setting the fill value to the current output section.
3155 
3156 extern "C" void
3157 script_add_fill(void* closurev, Expression* val)
3158 {
3159   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3160   closure->script_options()->script_sections()->add_fill(val);
3161 }
3162 
3163 // Add a new input section specification to the current output
3164 // section.
3165 
3166 extern "C" void
3167 script_add_input_section(void* closurev,
3168 			 const struct Input_section_spec* spec,
3169 			 int keepi)
3170 {
3171   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3172   bool keep = keepi != 0;
3173   closure->script_options()->script_sections()->add_input_section(spec, keep);
3174 }
3175 
3176 // When we see DATA_SEGMENT_ALIGN we record that following output
3177 // sections may be relro.
3178 
3179 extern "C" void
3180 script_data_segment_align(void* closurev)
3181 {
3182   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3183   if (!closure->script_options()->saw_sections_clause())
3184     gold_error(_("%s:%d:%d: DATA_SEGMENT_ALIGN not in SECTIONS clause"),
3185 	       closure->filename(), closure->lineno(), closure->charpos());
3186   else
3187     closure->script_options()->script_sections()->data_segment_align();
3188 }
3189 
3190 // When we see DATA_SEGMENT_RELRO_END we know that all output sections
3191 // since DATA_SEGMENT_ALIGN should be relro.
3192 
3193 extern "C" void
3194 script_data_segment_relro_end(void* closurev)
3195 {
3196   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3197   if (!closure->script_options()->saw_sections_clause())
3198     gold_error(_("%s:%d:%d: DATA_SEGMENT_ALIGN not in SECTIONS clause"),
3199 	       closure->filename(), closure->lineno(), closure->charpos());
3200   else
3201     closure->script_options()->script_sections()->data_segment_relro_end();
3202 }
3203 
3204 // Create a new list of string/sort pairs.
3205 
3206 extern "C" String_sort_list_ptr
3207 script_new_string_sort_list(const struct Wildcard_section* string_sort)
3208 {
3209   return new String_sort_list(1, *string_sort);
3210 }
3211 
3212 // Add an entry to a list of string/sort pairs.  The way the parser
3213 // works permits us to simply modify the first parameter, rather than
3214 // copy the vector.
3215 
3216 extern "C" String_sort_list_ptr
3217 script_string_sort_list_add(String_sort_list_ptr pv,
3218 			    const struct Wildcard_section* string_sort)
3219 {
3220   if (pv == NULL)
3221     return script_new_string_sort_list(string_sort);
3222   else
3223     {
3224       pv->push_back(*string_sort);
3225       return pv;
3226     }
3227 }
3228 
3229 // Create a new list of strings.
3230 
3231 extern "C" String_list_ptr
3232 script_new_string_list(const char* str, size_t len)
3233 {
3234   return new String_list(1, std::string(str, len));
3235 }
3236 
3237 // Add an element to a list of strings.  The way the parser works
3238 // permits us to simply modify the first parameter, rather than copy
3239 // the vector.
3240 
3241 extern "C" String_list_ptr
3242 script_string_list_push_back(String_list_ptr pv, const char* str, size_t len)
3243 {
3244   if (pv == NULL)
3245     return script_new_string_list(str, len);
3246   else
3247     {
3248       pv->push_back(std::string(str, len));
3249       return pv;
3250     }
3251 }
3252 
3253 // Concatenate two string lists.  Either or both may be NULL.  The way
3254 // the parser works permits us to modify the parameters, rather than
3255 // copy the vector.
3256 
3257 extern "C" String_list_ptr
3258 script_string_list_append(String_list_ptr pv1, String_list_ptr pv2)
3259 {
3260   if (pv1 == NULL)
3261     return pv2;
3262   if (pv2 == NULL)
3263     return pv1;
3264   pv1->insert(pv1->end(), pv2->begin(), pv2->end());
3265   return pv1;
3266 }
3267 
3268 // Add a new program header.
3269 
3270 extern "C" void
3271 script_add_phdr(void* closurev, const char* name, size_t namelen,
3272 		unsigned int type, const Phdr_info* info)
3273 {
3274   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3275   bool includes_filehdr = info->includes_filehdr != 0;
3276   bool includes_phdrs = info->includes_phdrs != 0;
3277   bool is_flags_valid = info->is_flags_valid != 0;
3278   Script_sections* ss = closure->script_options()->script_sections();
3279   ss->add_phdr(name, namelen, type, includes_filehdr, includes_phdrs,
3280 	       is_flags_valid, info->flags, info->load_address);
3281   closure->clear_skip_on_incompatible_target();
3282 }
3283 
3284 // Convert a program header string to a type.
3285 
3286 #define PHDR_TYPE(NAME) { #NAME, sizeof(#NAME) - 1, elfcpp::NAME }
3287 
3288 static struct
3289 {
3290   const char* name;
3291   size_t namelen;
3292   unsigned int val;
3293 } phdr_type_names[] =
3294 {
3295   PHDR_TYPE(PT_NULL),
3296   PHDR_TYPE(PT_LOAD),
3297   PHDR_TYPE(PT_DYNAMIC),
3298   PHDR_TYPE(PT_INTERP),
3299   PHDR_TYPE(PT_NOTE),
3300   PHDR_TYPE(PT_SHLIB),
3301   PHDR_TYPE(PT_PHDR),
3302   PHDR_TYPE(PT_TLS),
3303   PHDR_TYPE(PT_GNU_EH_FRAME),
3304   PHDR_TYPE(PT_GNU_STACK),
3305   PHDR_TYPE(PT_GNU_RELRO)
3306 };
3307 
3308 extern "C" unsigned int
3309 script_phdr_string_to_type(void* closurev, const char* name, size_t namelen)
3310 {
3311   for (unsigned int i = 0;
3312        i < sizeof(phdr_type_names) / sizeof(phdr_type_names[0]);
3313        ++i)
3314     if (namelen == phdr_type_names[i].namelen
3315 	&& strncmp(name, phdr_type_names[i].name, namelen) == 0)
3316       return phdr_type_names[i].val;
3317   yyerror(closurev, _("unknown PHDR type (try integer)"));
3318   return elfcpp::PT_NULL;
3319 }
3320 
3321 extern "C" void
3322 script_saw_segment_start_expression(void* closurev)
3323 {
3324   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3325   Script_sections* ss = closure->script_options()->script_sections();
3326   ss->set_saw_segment_start_expression(true);
3327 }
3328 
3329 extern "C" void
3330 script_set_section_region(void* closurev, const char* name, size_t namelen,
3331 			  int set_vma)
3332 {
3333   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3334   if (!closure->script_options()->saw_sections_clause())
3335     {
3336       gold_error(_("%s:%d:%d: MEMORY region '%.*s' referred to outside of "
3337 		   "SECTIONS clause"),
3338 		 closure->filename(), closure->lineno(), closure->charpos(),
3339 		 static_cast<int>(namelen), name);
3340       return;
3341     }
3342 
3343   Script_sections* ss = closure->script_options()->script_sections();
3344   Memory_region* mr = ss->find_memory_region(name, namelen);
3345   if (mr == NULL)
3346     {
3347       gold_error(_("%s:%d:%d: MEMORY region '%.*s' not declared"),
3348 		 closure->filename(), closure->lineno(), closure->charpos(),
3349 		 static_cast<int>(namelen), name);
3350       return;
3351     }
3352 
3353   ss->set_memory_region(mr, set_vma);
3354 }
3355 
3356 extern "C" void
3357 script_add_memory(void* closurev, const char* name, size_t namelen,
3358 		  unsigned int attrs, Expression* origin, Expression* length)
3359 {
3360   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3361   Script_sections* ss = closure->script_options()->script_sections();
3362   ss->add_memory_region(name, namelen, attrs, origin, length);
3363 }
3364 
3365 extern "C" unsigned int
3366 script_parse_memory_attr(void* closurev, const char* attrs, size_t attrlen,
3367 			 int invert)
3368 {
3369   int attributes = 0;
3370 
3371   while (attrlen--)
3372     switch (*attrs++)
3373       {
3374       case 'R':
3375       case 'r':
3376 	attributes |= MEM_READABLE; break;
3377       case 'W':
3378       case 'w':
3379 	attributes |= MEM_READABLE | MEM_WRITEABLE; break;
3380       case 'X':
3381       case 'x':
3382 	attributes |= MEM_EXECUTABLE; break;
3383       case 'A':
3384       case 'a':
3385 	attributes |= MEM_ALLOCATABLE; break;
3386       case 'I':
3387       case 'i':
3388       case 'L':
3389       case 'l':
3390 	attributes |= MEM_INITIALIZED; break;
3391       default:
3392 	yyerror(closurev, _("unknown MEMORY attribute"));
3393       }
3394 
3395   if (invert)
3396     attributes = (~ attributes) & MEM_ATTR_MASK;
3397 
3398   return attributes;
3399 }
3400 
3401 extern "C" void
3402 script_include_directive(int first_token, void* closurev,
3403 			 const char* filename, size_t length)
3404 {
3405   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3406   std::string name(filename, length);
3407   Command_line* cmdline = closure->command_line();
3408   read_script_file(name.c_str(), cmdline, &cmdline->script_options(),
3409                    first_token, Lex::LINKER_SCRIPT);
3410 }
3411 
3412 // Functions for memory regions.
3413 
3414 extern "C" Expression*
3415 script_exp_function_origin(void* closurev, const char* name, size_t namelen)
3416 {
3417   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3418   Script_sections* ss = closure->script_options()->script_sections();
3419   Expression* origin = ss->find_memory_region_origin(name, namelen);
3420 
3421   if (origin == NULL)
3422     {
3423       gold_error(_("undefined memory region '%s' referenced "
3424 		   "in ORIGIN expression"),
3425 		 name);
3426       // Create a dummy expression to prevent crashes later on.
3427       origin = script_exp_integer(0);
3428     }
3429 
3430   return origin;
3431 }
3432 
3433 extern "C" Expression*
3434 script_exp_function_length(void* closurev, const char* name, size_t namelen)
3435 {
3436   Parser_closure* closure = static_cast<Parser_closure*>(closurev);
3437   Script_sections* ss = closure->script_options()->script_sections();
3438   Expression* length = ss->find_memory_region_length(name, namelen);
3439 
3440   if (length == NULL)
3441     {
3442       gold_error(_("undefined memory region '%s' referenced "
3443 		   "in LENGTH expression"),
3444 		 name);
3445       // Create a dummy expression to prevent crashes later on.
3446       length = script_exp_integer(0);
3447     }
3448 
3449   return length;
3450 }
3451