• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 //===-- ArchiveReader.cpp - Read LLVM archive files -------------*- C++ -*-===//
2 //
3 //                     The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // Builds up standard unix archive files (.a) containing LLVM bitcode.
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #include "ArchiveInternals.h"
15 #include "llvm/Bitcode/ReaderWriter.h"
16 #include "llvm/Support/MemoryBuffer.h"
17 #include "llvm/Module.h"
18 #include <cstdlib>
19 #include <memory>
20 using namespace llvm;
21 
22 /// Read a variable-bit-rate encoded unsigned integer
readInteger(const char * & At,const char * End)23 static inline unsigned readInteger(const char*&At, const char*End) {
24   unsigned Shift = 0;
25   unsigned Result = 0;
26 
27   do {
28     if (At == End)
29       return Result;
30     Result |= (unsigned)((*At++) & 0x7F) << Shift;
31     Shift += 7;
32   } while (At[-1] & 0x80);
33   return Result;
34 }
35 
36 // Completely parse the Archive's symbol table and populate symTab member var.
37 bool
parseSymbolTable(const void * data,unsigned size,std::string * error)38 Archive::parseSymbolTable(const void* data, unsigned size, std::string* error) {
39   const char* At = (const char*) data;
40   const char* End = At + size;
41   while (At < End) {
42     unsigned offset = readInteger(At, End);
43     if (At == End) {
44       if (error)
45         *error = "Ran out of data reading vbr_uint for symtab offset!";
46       return false;
47     }
48     unsigned length = readInteger(At, End);
49     if (At == End) {
50       if (error)
51         *error = "Ran out of data reading vbr_uint for symtab length!";
52       return false;
53     }
54     if (At + length > End) {
55       if (error)
56         *error = "Malformed symbol table: length not consistent with size";
57       return false;
58     }
59     // we don't care if it can't be inserted (duplicate entry)
60     symTab.insert(std::make_pair(std::string(At, length), offset));
61     At += length;
62   }
63   symTabSize = size;
64   return true;
65 }
66 
67 // This member parses an ArchiveMemberHeader that is presumed to be pointed to
68 // by At. The At pointer is updated to the byte just after the header, which
69 // can be variable in size.
70 ArchiveMember*
parseMemberHeader(const char * & At,const char * End,std::string * error)71 Archive::parseMemberHeader(const char*& At, const char* End, std::string* error)
72 {
73   if (At + sizeof(ArchiveMemberHeader) >= End) {
74     if (error)
75       *error = "Unexpected end of file";
76     return 0;
77   }
78 
79   // Cast archive member header
80   ArchiveMemberHeader* Hdr = (ArchiveMemberHeader*)At;
81   At += sizeof(ArchiveMemberHeader);
82 
83   // Extract the size and determine if the file is
84   // compressed or not (negative length).
85   int flags = 0;
86   int MemberSize = atoi(Hdr->size);
87   if (MemberSize < 0) {
88     flags |= ArchiveMember::CompressedFlag;
89     MemberSize = -MemberSize;
90   }
91 
92   // Check the size of the member for sanity
93   if (At + MemberSize > End) {
94     if (error)
95       *error = "invalid member length in archive file";
96     return 0;
97   }
98 
99   // Check the member signature
100   if (!Hdr->checkSignature()) {
101     if (error)
102       *error = "invalid file member signature";
103     return 0;
104   }
105 
106   // Convert and check the member name
107   // The empty name ( '/' and 15 blanks) is for a foreign (non-LLVM) symbol
108   // table. The special name "//" and 14 blanks is for a string table, used
109   // for long file names. This library doesn't generate either of those but
110   // it will accept them. If the name starts with #1/ and the remainder is
111   // digits, then those digits specify the length of the name that is
112   // stored immediately following the header. The special name
113   // __LLVM_SYM_TAB__ identifies the symbol table for LLVM bitcode.
114   // Anything else is a regular, short filename that is terminated with
115   // a '/' and blanks.
116 
117   std::string pathname;
118   switch (Hdr->name[0]) {
119     case '#':
120       if (Hdr->name[1] == '1' && Hdr->name[2] == '/') {
121         if (isdigit(Hdr->name[3])) {
122           unsigned len = atoi(&Hdr->name[3]);
123           const char *nulp = (const char *)memchr(At, '\0', len);
124           pathname.assign(At, nulp != 0 ? (uintptr_t)(nulp - At) : len);
125           At += len;
126           MemberSize -= len;
127           flags |= ArchiveMember::HasLongFilenameFlag;
128         } else {
129           if (error)
130             *error = "invalid long filename";
131           return 0;
132         }
133       } else if (Hdr->name[1] == '_' &&
134                  (0 == memcmp(Hdr->name, ARFILE_LLVM_SYMTAB_NAME, 16))) {
135         // The member is using a long file name (>15 chars) format.
136         // This format is standard for 4.4BSD and Mac OSX operating
137         // systems. LLVM uses it similarly. In this format, the
138         // remainder of the name field (after #1/) specifies the
139         // length of the file name which occupy the first bytes of
140         // the member's data. The pathname already has the #1/ stripped.
141         pathname.assign(ARFILE_LLVM_SYMTAB_NAME);
142         flags |= ArchiveMember::LLVMSymbolTableFlag;
143       }
144       break;
145     case '/':
146       if (Hdr->name[1]== '/') {
147         if (0 == memcmp(Hdr->name, ARFILE_STRTAB_NAME, 16)) {
148           pathname.assign(ARFILE_STRTAB_NAME);
149           flags |= ArchiveMember::StringTableFlag;
150         } else {
151           if (error)
152             *error = "invalid string table name";
153           return 0;
154         }
155       } else if (Hdr->name[1] == ' ') {
156         if (0 == memcmp(Hdr->name, ARFILE_SVR4_SYMTAB_NAME, 16)) {
157           pathname.assign(ARFILE_SVR4_SYMTAB_NAME);
158           flags |= ArchiveMember::SVR4SymbolTableFlag;
159         } else {
160           if (error)
161             *error = "invalid SVR4 symbol table name";
162           return 0;
163         }
164       } else if (isdigit(Hdr->name[1])) {
165         unsigned index = atoi(&Hdr->name[1]);
166         if (index < strtab.length()) {
167           const char* namep = strtab.c_str() + index;
168           const char* endp = strtab.c_str() + strtab.length();
169           const char* p = namep;
170           const char* last_p = p;
171           while (p < endp) {
172             if (*p == '\n' && *last_p == '/') {
173               pathname.assign(namep, last_p - namep);
174               flags |= ArchiveMember::HasLongFilenameFlag;
175               break;
176             }
177             last_p = p;
178             p++;
179           }
180           if (p >= endp) {
181             if (error)
182               *error = "missing name termiantor in string table";
183             return 0;
184           }
185         } else {
186           if (error)
187             *error = "name index beyond string table";
188           return 0;
189         }
190       }
191       break;
192     case '_':
193       if (Hdr->name[1] == '_' &&
194           (0 == memcmp(Hdr->name, ARFILE_BSD4_SYMTAB_NAME, 16))) {
195         pathname.assign(ARFILE_BSD4_SYMTAB_NAME);
196         flags |= ArchiveMember::BSD4SymbolTableFlag;
197         break;
198       }
199       /* FALL THROUGH */
200 
201     default:
202       char* slash = (char*) memchr(Hdr->name, '/', 16);
203       if (slash == 0)
204         slash = Hdr->name + 16;
205       pathname.assign(Hdr->name, slash - Hdr->name);
206       break;
207   }
208 
209   // Determine if this is a bitcode file
210   switch (sys::IdentifyFileType(At, 4)) {
211     case sys::Bitcode_FileType:
212       flags |= ArchiveMember::BitcodeFlag;
213       break;
214     default:
215       flags &= ~ArchiveMember::BitcodeFlag;
216       break;
217   }
218 
219   // Instantiate the ArchiveMember to be filled
220   ArchiveMember* member = new ArchiveMember(this);
221 
222   // Fill in fields of the ArchiveMember
223   member->parent = this;
224   member->path.set(pathname);
225   member->info.fileSize = MemberSize;
226   member->info.modTime.fromEpochTime(atoi(Hdr->date));
227   unsigned int mode;
228   sscanf(Hdr->mode, "%o", &mode);
229   member->info.mode = mode;
230   member->info.user = atoi(Hdr->uid);
231   member->info.group = atoi(Hdr->gid);
232   member->flags = flags;
233   member->data = At;
234 
235   return member;
236 }
237 
238 bool
checkSignature(std::string * error)239 Archive::checkSignature(std::string* error) {
240   // Check the magic string at file's header
241   if (mapfile->getBufferSize() < 8 || memcmp(base, ARFILE_MAGIC, 8)) {
242     if (error)
243       *error = "invalid signature for an archive file";
244     return false;
245   }
246   return true;
247 }
248 
249 // This function loads the entire archive and fully populates its ilist with
250 // the members of the archive file. This is typically used in preparation for
251 // editing the contents of the archive.
252 bool
loadArchive(std::string * error)253 Archive::loadArchive(std::string* error) {
254 
255   // Set up parsing
256   members.clear();
257   symTab.clear();
258   const char *At = base;
259   const char *End = mapfile->getBufferEnd();
260 
261   if (!checkSignature(error))
262     return false;
263 
264   At += 8;  // Skip the magic string.
265 
266   bool seenSymbolTable = false;
267   bool foundFirstFile = false;
268   while (At < End) {
269     // parse the member header
270     const char* Save = At;
271     ArchiveMember* mbr = parseMemberHeader(At, End, error);
272     if (!mbr)
273       return false;
274 
275     // check if this is the foreign symbol table
276     if (mbr->isSVR4SymbolTable() || mbr->isBSD4SymbolTable()) {
277       // We just save this but don't do anything special
278       // with it. It doesn't count as the "first file".
279       if (foreignST) {
280         // What? Multiple foreign symbol tables? Just chuck it
281         // and retain the last one found.
282         delete foreignST;
283       }
284       foreignST = mbr;
285       At += mbr->getSize();
286       if ((intptr_t(At) & 1) == 1)
287         At++;
288     } else if (mbr->isStringTable()) {
289       // Simply suck the entire string table into a string
290       // variable. This will be used to get the names of the
291       // members that use the "/ddd" format for their names
292       // (SVR4 style long names).
293       strtab.assign(At, mbr->getSize());
294       At += mbr->getSize();
295       if ((intptr_t(At) & 1) == 1)
296         At++;
297       delete mbr;
298     } else if (mbr->isLLVMSymbolTable()) {
299       // This is the LLVM symbol table for the archive. If we've seen it
300       // already, its an error. Otherwise, parse the symbol table and move on.
301       if (seenSymbolTable) {
302         if (error)
303           *error = "invalid archive: multiple symbol tables";
304         return false;
305       }
306       if (!parseSymbolTable(mbr->getData(), mbr->getSize(), error))
307         return false;
308       seenSymbolTable = true;
309       At += mbr->getSize();
310       if ((intptr_t(At) & 1) == 1)
311         At++;
312       delete mbr; // We don't need this member in the list of members.
313     } else {
314       // This is just a regular file. If its the first one, save its offset.
315       // Otherwise just push it on the list and move on to the next file.
316       if (!foundFirstFile) {
317         firstFileOffset = Save - base;
318         foundFirstFile = true;
319       }
320       members.push_back(mbr);
321       At += mbr->getSize();
322       if ((intptr_t(At) & 1) == 1)
323         At++;
324     }
325   }
326   return true;
327 }
328 
329 // Open and completely load the archive file.
330 Archive*
OpenAndLoad(const sys::Path & file,LLVMContext & C,std::string * ErrorMessage)331 Archive::OpenAndLoad(const sys::Path& file, LLVMContext& C,
332                      std::string* ErrorMessage) {
333   std::auto_ptr<Archive> result ( new Archive(file, C));
334   if (result->mapToMemory(ErrorMessage))
335     return 0;
336   if (!result->loadArchive(ErrorMessage))
337     return 0;
338   return result.release();
339 }
340 
341 // Get all the bitcode modules from the archive
342 bool
getAllModules(std::vector<Module * > & Modules,std::string * ErrMessage)343 Archive::getAllModules(std::vector<Module*>& Modules,
344                        std::string* ErrMessage) {
345 
346   for (iterator I=begin(), E=end(); I != E; ++I) {
347     if (I->isBitcode()) {
348       std::string FullMemberName = archPath.str() +
349         "(" + I->getPath().str() + ")";
350       MemoryBuffer *Buffer =
351         MemoryBuffer::getMemBufferCopy(StringRef(I->getData(), I->getSize()),
352                                        FullMemberName.c_str());
353 
354       Module *M = ParseBitcodeFile(Buffer, Context, ErrMessage);
355       delete Buffer;
356       if (!M)
357         return true;
358 
359       Modules.push_back(M);
360     }
361   }
362   return false;
363 }
364 
365 // Load just the symbol table from the archive file
366 bool
loadSymbolTable(std::string * ErrorMsg)367 Archive::loadSymbolTable(std::string* ErrorMsg) {
368 
369   // Set up parsing
370   members.clear();
371   symTab.clear();
372   const char *At = base;
373   const char *End = mapfile->getBufferEnd();
374 
375   // Make sure we're dealing with an archive
376   if (!checkSignature(ErrorMsg))
377     return false;
378 
379   At += 8; // Skip signature
380 
381   // Parse the first file member header
382   const char* FirstFile = At;
383   ArchiveMember* mbr = parseMemberHeader(At, End, ErrorMsg);
384   if (!mbr)
385     return false;
386 
387   if (mbr->isSVR4SymbolTable() || mbr->isBSD4SymbolTable()) {
388     // Skip the foreign symbol table, we don't do anything with it
389     At += mbr->getSize();
390     if ((intptr_t(At) & 1) == 1)
391       At++;
392     delete mbr;
393 
394     // Read the next one
395     FirstFile = At;
396     mbr = parseMemberHeader(At, End, ErrorMsg);
397     if (!mbr) {
398       delete mbr;
399       return false;
400     }
401   }
402 
403   if (mbr->isStringTable()) {
404     // Process the string table entry
405     strtab.assign((const char*)mbr->getData(), mbr->getSize());
406     At += mbr->getSize();
407     if ((intptr_t(At) & 1) == 1)
408       At++;
409     delete mbr;
410     // Get the next one
411     FirstFile = At;
412     mbr = parseMemberHeader(At, End, ErrorMsg);
413     if (!mbr) {
414       delete mbr;
415       return false;
416     }
417   }
418 
419   // See if its the symbol table
420   if (mbr->isLLVMSymbolTable()) {
421     if (!parseSymbolTable(mbr->getData(), mbr->getSize(), ErrorMsg)) {
422       delete mbr;
423       return false;
424     }
425 
426     At += mbr->getSize();
427     if ((intptr_t(At) & 1) == 1)
428       At++;
429     delete mbr;
430     // Can't be any more symtab headers so just advance
431     FirstFile = At;
432   } else {
433     // There's no symbol table in the file. We have to rebuild it from scratch
434     // because the intent of this method is to get the symbol table loaded so
435     // it can be searched efficiently.
436     // Add the member to the members list
437     members.push_back(mbr);
438   }
439 
440   firstFileOffset = FirstFile - base;
441   return true;
442 }
443 
444 // Open the archive and load just the symbol tables
OpenAndLoadSymbols(const sys::Path & file,LLVMContext & C,std::string * ErrorMessage)445 Archive* Archive::OpenAndLoadSymbols(const sys::Path& file,
446                                      LLVMContext& C,
447                                      std::string* ErrorMessage) {
448   std::auto_ptr<Archive> result ( new Archive(file, C) );
449   if (result->mapToMemory(ErrorMessage))
450     return 0;
451   if (!result->loadSymbolTable(ErrorMessage))
452     return 0;
453   return result.release();
454 }
455 
456 // Look up one symbol in the symbol table and return the module that defines
457 // that symbol.
458 Module*
findModuleDefiningSymbol(const std::string & symbol,std::string * ErrMsg)459 Archive::findModuleDefiningSymbol(const std::string& symbol,
460                                   std::string* ErrMsg) {
461   SymTabType::iterator SI = symTab.find(symbol);
462   if (SI == symTab.end())
463     return 0;
464 
465   // The symbol table was previously constructed assuming that the members were
466   // written without the symbol table header. Because VBR encoding is used, the
467   // values could not be adjusted to account for the offset of the symbol table
468   // because that could affect the size of the symbol table due to VBR encoding.
469   // We now have to account for this by adjusting the offset by the size of the
470   // symbol table and its header.
471   unsigned fileOffset =
472     SI->second +                // offset in symbol-table-less file
473     firstFileOffset;            // add offset to first "real" file in archive
474 
475   // See if the module is already loaded
476   ModuleMap::iterator MI = modules.find(fileOffset);
477   if (MI != modules.end())
478     return MI->second.first;
479 
480   // Module hasn't been loaded yet, we need to load it
481   const char* modptr = base + fileOffset;
482   ArchiveMember* mbr = parseMemberHeader(modptr, mapfile->getBufferEnd(),
483                                          ErrMsg);
484   if (!mbr)
485     return 0;
486 
487   // Now, load the bitcode module to get the Module.
488   std::string FullMemberName = archPath.str() + "(" +
489     mbr->getPath().str() + ")";
490   MemoryBuffer *Buffer =
491     MemoryBuffer::getMemBufferCopy(StringRef(mbr->getData(), mbr->getSize()),
492                                    FullMemberName.c_str());
493 
494   Module *m = getLazyBitcodeModule(Buffer, Context, ErrMsg);
495   if (!m)
496     return 0;
497 
498   modules.insert(std::make_pair(fileOffset, std::make_pair(m, mbr)));
499 
500   return m;
501 }
502 
503 // Look up multiple symbols in the symbol table and return a set of
504 // Modules that define those symbols.
505 bool
findModulesDefiningSymbols(std::set<std::string> & symbols,std::set<Module * > & result,std::string * error)506 Archive::findModulesDefiningSymbols(std::set<std::string>& symbols,
507                                     std::set<Module*>& result,
508                                     std::string* error) {
509   if (!mapfile || !base) {
510     if (error)
511       *error = "Empty archive invalid for finding modules defining symbols";
512     return false;
513   }
514 
515   if (symTab.empty()) {
516     // We don't have a symbol table, so we must build it now but lets also
517     // make sure that we populate the modules table as we do this to ensure
518     // that we don't load them twice when findModuleDefiningSymbol is called
519     // below.
520 
521     // Get a pointer to the first file
522     const char* At  = base + firstFileOffset;
523     const char* End = mapfile->getBufferEnd();
524 
525     while ( At < End) {
526       // Compute the offset to be put in the symbol table
527       unsigned offset = At - base - firstFileOffset;
528 
529       // Parse the file's header
530       ArchiveMember* mbr = parseMemberHeader(At, End, error);
531       if (!mbr)
532         return false;
533 
534       // If it contains symbols
535       if (mbr->isBitcode()) {
536         // Get the symbols
537         std::vector<std::string> symbols;
538         std::string FullMemberName = archPath.str() + "(" +
539           mbr->getPath().str() + ")";
540         Module* M =
541           GetBitcodeSymbols(At, mbr->getSize(), FullMemberName, Context,
542                             symbols, error);
543 
544         if (M) {
545           // Insert the module's symbols into the symbol table
546           for (std::vector<std::string>::iterator I = symbols.begin(),
547                E=symbols.end(); I != E; ++I ) {
548             symTab.insert(std::make_pair(*I, offset));
549           }
550           // Insert the Module and the ArchiveMember into the table of
551           // modules.
552           modules.insert(std::make_pair(offset, std::make_pair(M, mbr)));
553         } else {
554           if (error)
555             *error = "Can't parse bitcode member: " +
556               mbr->getPath().str() + ": " + *error;
557           delete mbr;
558           return false;
559         }
560       }
561 
562       // Go to the next file location
563       At += mbr->getSize();
564       if ((intptr_t(At) & 1) == 1)
565         At++;
566     }
567   }
568 
569   // At this point we have a valid symbol table (one way or another) so we
570   // just use it to quickly find the symbols requested.
571 
572   for (std::set<std::string>::iterator I=symbols.begin(),
573        E=symbols.end(); I != E;) {
574     // See if this symbol exists
575     Module* m = findModuleDefiningSymbol(*I,error);
576     if (m) {
577       // The symbol exists, insert the Module into our result, duplicates will
578       // be ignored.
579       result.insert(m);
580 
581       // Remove the symbol now that its been resolved, being careful to
582       // post-increment the iterator.
583       symbols.erase(I++);
584     } else {
585       ++I;
586     }
587   }
588   return true;
589 }
590 
isBitcodeArchive()591 bool Archive::isBitcodeArchive() {
592   // Make sure the symTab has been loaded. In most cases this should have been
593   // done when the archive was constructed, but still,  this is just in case.
594   if (symTab.empty())
595     if (!loadSymbolTable(0))
596       return false;
597 
598   // Now that we know it's been loaded, return true
599   // if it has a size
600   if (symTab.size()) return true;
601 
602   // We still can't be sure it isn't a bitcode archive
603   if (!loadArchive(0))
604     return false;
605 
606   std::vector<Module *> Modules;
607   std::string ErrorMessage;
608 
609   // Scan the archive, trying to load a bitcode member.  We only load one to
610   // see if this works.
611   for (iterator I = begin(), E = end(); I != E; ++I) {
612     if (!I->isBitcode())
613       continue;
614 
615     std::string FullMemberName =
616       archPath.str() + "(" + I->getPath().str() + ")";
617 
618     MemoryBuffer *Buffer =
619       MemoryBuffer::getMemBufferCopy(StringRef(I->getData(), I->getSize()),
620                                      FullMemberName.c_str());
621     Module *M = ParseBitcodeFile(Buffer, Context);
622     delete Buffer;
623     if (!M)
624       return false;  // Couldn't parse bitcode, not a bitcode archive.
625     delete M;
626     return true;
627   }
628 
629   return false;
630 }
631