1 //===- BitstreamReader.cpp - BitstreamReader implementation ---------------===//
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 #include "llvm/Bitcode/BitstreamReader.h"
11 
12 using namespace llvm;
13 
14 //===----------------------------------------------------------------------===//
15 //  BitstreamCursor implementation
16 //===----------------------------------------------------------------------===//
17 
freeState()18 void BitstreamCursor::freeState() {
19   // Free all the Abbrevs.
20   CurAbbrevs.clear();
21 
22   // Free all the Abbrevs in the block scope.
23   BlockScope.clear();
24 }
25 
26 /// EnterSubBlock - Having read the ENTER_SUBBLOCK abbrevid, enter
27 /// the block, and return true if the block has an error.
EnterSubBlock(unsigned BlockID,unsigned * NumWordsP)28 bool BitstreamCursor::EnterSubBlock(unsigned BlockID, unsigned *NumWordsP) {
29   // Save the current block's state on BlockScope.
30   BlockScope.push_back(Block(CurCodeSize));
31   BlockScope.back().PrevAbbrevs.swap(CurAbbrevs);
32 
33   // Add the abbrevs specific to this block to the CurAbbrevs list.
34   if (const BitstreamReader::BlockInfo *Info =
35       BitStream->getBlockInfo(BlockID)) {
36     CurAbbrevs.insert(CurAbbrevs.end(), Info->Abbrevs.begin(),
37                       Info->Abbrevs.end());
38   }
39 
40   // Get the codesize of this block.
41   CurCodeSize = ReadVBR(bitc::CodeLenWidth);
42   // We can't read more than MaxChunkSize at a time
43   if (CurCodeSize > MaxChunkSize)
44     return true;
45 
46   SkipToFourByteBoundary();
47   unsigned NumWords = Read(bitc::BlockSizeWidth);
48   if (NumWordsP) *NumWordsP = NumWords;
49 
50   // Validate that this block is sane.
51   return CurCodeSize == 0 || AtEndOfStream();
52 }
53 
readAbbreviatedField(BitstreamCursor & Cursor,const BitCodeAbbrevOp & Op)54 static uint64_t readAbbreviatedField(BitstreamCursor &Cursor,
55                                      const BitCodeAbbrevOp &Op) {
56   assert(!Op.isLiteral() && "Not to be used with literals!");
57 
58   // Decode the value as we are commanded.
59   switch (Op.getEncoding()) {
60   case BitCodeAbbrevOp::Array:
61   case BitCodeAbbrevOp::Blob:
62     llvm_unreachable("Should not reach here");
63   case BitCodeAbbrevOp::Fixed:
64     assert((unsigned)Op.getEncodingData() <= Cursor.MaxChunkSize);
65     return Cursor.Read((unsigned)Op.getEncodingData());
66   case BitCodeAbbrevOp::VBR:
67     assert((unsigned)Op.getEncodingData() <= Cursor.MaxChunkSize);
68     return Cursor.ReadVBR64((unsigned)Op.getEncodingData());
69   case BitCodeAbbrevOp::Char6:
70     return BitCodeAbbrevOp::DecodeChar6(Cursor.Read(6));
71   }
72   llvm_unreachable("invalid abbreviation encoding");
73 }
74 
skipAbbreviatedField(BitstreamCursor & Cursor,const BitCodeAbbrevOp & Op)75 static void skipAbbreviatedField(BitstreamCursor &Cursor,
76                                  const BitCodeAbbrevOp &Op) {
77   assert(!Op.isLiteral() && "Not to be used with literals!");
78 
79   // Decode the value as we are commanded.
80   switch (Op.getEncoding()) {
81   case BitCodeAbbrevOp::Array:
82   case BitCodeAbbrevOp::Blob:
83     llvm_unreachable("Should not reach here");
84   case BitCodeAbbrevOp::Fixed:
85     assert((unsigned)Op.getEncodingData() <= Cursor.MaxChunkSize);
86     Cursor.Read((unsigned)Op.getEncodingData());
87     break;
88   case BitCodeAbbrevOp::VBR:
89     assert((unsigned)Op.getEncodingData() <= Cursor.MaxChunkSize);
90     Cursor.ReadVBR64((unsigned)Op.getEncodingData());
91     break;
92   case BitCodeAbbrevOp::Char6:
93     Cursor.Read(6);
94     break;
95   }
96 }
97 
98 
99 
100 /// skipRecord - Read the current record and discard it.
skipRecord(unsigned AbbrevID)101 void BitstreamCursor::skipRecord(unsigned AbbrevID) {
102   // Skip unabbreviated records by reading past their entries.
103   if (AbbrevID == bitc::UNABBREV_RECORD) {
104     unsigned Code = ReadVBR(6);
105     (void)Code;
106     unsigned NumElts = ReadVBR(6);
107     for (unsigned i = 0; i != NumElts; ++i)
108       (void)ReadVBR64(6);
109     return;
110   }
111 
112   const BitCodeAbbrev *Abbv = getAbbrev(AbbrevID);
113 
114   for (unsigned i = 0, e = Abbv->getNumOperandInfos(); i != e; ++i) {
115     const BitCodeAbbrevOp &Op = Abbv->getOperandInfo(i);
116     if (Op.isLiteral())
117       continue;
118 
119     if (Op.getEncoding() != BitCodeAbbrevOp::Array &&
120         Op.getEncoding() != BitCodeAbbrevOp::Blob) {
121       skipAbbreviatedField(*this, Op);
122       continue;
123     }
124 
125     if (Op.getEncoding() == BitCodeAbbrevOp::Array) {
126       // Array case.  Read the number of elements as a vbr6.
127       unsigned NumElts = ReadVBR(6);
128 
129       // Get the element encoding.
130       assert(i+2 == e && "array op not second to last?");
131       const BitCodeAbbrevOp &EltEnc = Abbv->getOperandInfo(++i);
132 
133       // Read all the elements.
134       for (; NumElts; --NumElts)
135         skipAbbreviatedField(*this, EltEnc);
136       continue;
137     }
138 
139     assert(Op.getEncoding() == BitCodeAbbrevOp::Blob);
140     // Blob case.  Read the number of bytes as a vbr6.
141     unsigned NumElts = ReadVBR(6);
142     SkipToFourByteBoundary();  // 32-bit alignment
143 
144     // Figure out where the end of this blob will be including tail padding.
145     size_t NewEnd = GetCurrentBitNo()+((NumElts+3)&~3)*8;
146 
147     // If this would read off the end of the bitcode file, just set the
148     // record to empty and return.
149     if (!canSkipToPos(NewEnd/8)) {
150       NextChar = BitStream->getBitcodeBytes().getExtent();
151       break;
152     }
153 
154     // Skip over the blob.
155     JumpToBit(NewEnd);
156   }
157 }
158 
readRecord(unsigned AbbrevID,SmallVectorImpl<uint64_t> & Vals,StringRef * Blob)159 unsigned BitstreamCursor::readRecord(unsigned AbbrevID,
160                                      SmallVectorImpl<uint64_t> &Vals,
161                                      StringRef *Blob) {
162   if (AbbrevID == bitc::UNABBREV_RECORD) {
163     unsigned Code = ReadVBR(6);
164     unsigned NumElts = ReadVBR(6);
165     for (unsigned i = 0; i != NumElts; ++i)
166       Vals.push_back(ReadVBR64(6));
167     return Code;
168   }
169 
170   const BitCodeAbbrev *Abbv = getAbbrev(AbbrevID);
171 
172   // Read the record code first.
173   assert(Abbv->getNumOperandInfos() != 0 && "no record code in abbreviation?");
174   const BitCodeAbbrevOp &CodeOp = Abbv->getOperandInfo(0);
175   unsigned Code;
176   if (CodeOp.isLiteral())
177     Code = CodeOp.getLiteralValue();
178   else {
179     if (CodeOp.getEncoding() == BitCodeAbbrevOp::Array ||
180         CodeOp.getEncoding() == BitCodeAbbrevOp::Blob)
181       report_fatal_error("Abbreviation starts with an Array or a Blob");
182     Code = readAbbreviatedField(*this, CodeOp);
183   }
184 
185   for (unsigned i = 1, e = Abbv->getNumOperandInfos(); i != e; ++i) {
186     const BitCodeAbbrevOp &Op = Abbv->getOperandInfo(i);
187     if (Op.isLiteral()) {
188       Vals.push_back(Op.getLiteralValue());
189       continue;
190     }
191 
192     if (Op.getEncoding() != BitCodeAbbrevOp::Array &&
193         Op.getEncoding() != BitCodeAbbrevOp::Blob) {
194       Vals.push_back(readAbbreviatedField(*this, Op));
195       continue;
196     }
197 
198     if (Op.getEncoding() == BitCodeAbbrevOp::Array) {
199       // Array case.  Read the number of elements as a vbr6.
200       unsigned NumElts = ReadVBR(6);
201 
202       // Get the element encoding.
203       if (i + 2 != e)
204         report_fatal_error("Array op not second to last");
205       const BitCodeAbbrevOp &EltEnc = Abbv->getOperandInfo(++i);
206       if (!EltEnc.isEncoding())
207         report_fatal_error(
208             "Array element type has to be an encoding of a type");
209       if (EltEnc.getEncoding() == BitCodeAbbrevOp::Array ||
210           EltEnc.getEncoding() == BitCodeAbbrevOp::Blob)
211         report_fatal_error("Array element type can't be an Array or a Blob");
212 
213       // Read all the elements.
214       for (; NumElts; --NumElts)
215         Vals.push_back(readAbbreviatedField(*this, EltEnc));
216       continue;
217     }
218 
219     assert(Op.getEncoding() == BitCodeAbbrevOp::Blob);
220     // Blob case.  Read the number of bytes as a vbr6.
221     unsigned NumElts = ReadVBR(6);
222     SkipToFourByteBoundary();  // 32-bit alignment
223 
224     // Figure out where the end of this blob will be including tail padding.
225     size_t CurBitPos = GetCurrentBitNo();
226     size_t NewEnd = CurBitPos+((NumElts+3)&~3)*8;
227 
228     // If this would read off the end of the bitcode file, just set the
229     // record to empty and return.
230     if (!canSkipToPos(NewEnd/8)) {
231       Vals.append(NumElts, 0);
232       NextChar = BitStream->getBitcodeBytes().getExtent();
233       break;
234     }
235 
236     // Otherwise, inform the streamer that we need these bytes in memory.
237     const char *Ptr = (const char*)
238       BitStream->getBitcodeBytes().getPointer(CurBitPos/8, NumElts);
239 
240     // If we can return a reference to the data, do so to avoid copying it.
241     if (Blob) {
242       *Blob = StringRef(Ptr, NumElts);
243     } else {
244       // Otherwise, unpack into Vals with zero extension.
245       for (; NumElts; --NumElts)
246         Vals.push_back((unsigned char)*Ptr++);
247     }
248     // Skip over tail padding.
249     JumpToBit(NewEnd);
250   }
251 
252   return Code;
253 }
254 
255 
ReadAbbrevRecord()256 void BitstreamCursor::ReadAbbrevRecord() {
257   BitCodeAbbrev *Abbv = new BitCodeAbbrev();
258   unsigned NumOpInfo = ReadVBR(5);
259   for (unsigned i = 0; i != NumOpInfo; ++i) {
260     bool IsLiteral = Read(1);
261     if (IsLiteral) {
262       Abbv->Add(BitCodeAbbrevOp(ReadVBR64(8)));
263       continue;
264     }
265 
266     BitCodeAbbrevOp::Encoding E = (BitCodeAbbrevOp::Encoding)Read(3);
267     if (BitCodeAbbrevOp::hasEncodingData(E)) {
268       uint64_t Data = ReadVBR64(5);
269 
270       // As a special case, handle fixed(0) (i.e., a fixed field with zero bits)
271       // and vbr(0) as a literal zero.  This is decoded the same way, and avoids
272       // a slow path in Read() to have to handle reading zero bits.
273       if ((E == BitCodeAbbrevOp::Fixed || E == BitCodeAbbrevOp::VBR) &&
274           Data == 0) {
275         Abbv->Add(BitCodeAbbrevOp(0));
276         continue;
277       }
278 
279       if ((E == BitCodeAbbrevOp::Fixed || E == BitCodeAbbrevOp::VBR) &&
280           Data > MaxChunkSize)
281         report_fatal_error(
282             "Fixed or VBR abbrev record with size > MaxChunkData");
283 
284       Abbv->Add(BitCodeAbbrevOp(E, Data));
285     } else
286       Abbv->Add(BitCodeAbbrevOp(E));
287   }
288 
289   if (Abbv->getNumOperandInfos() == 0)
290     report_fatal_error("Abbrev record with no operands");
291   CurAbbrevs.push_back(Abbv);
292 }
293 
ReadBlockInfoBlock()294 bool BitstreamCursor::ReadBlockInfoBlock() {
295   // If this is the second stream to get to the block info block, skip it.
296   if (BitStream->hasBlockInfoRecords())
297     return SkipBlock();
298 
299   if (EnterSubBlock(bitc::BLOCKINFO_BLOCK_ID)) return true;
300 
301   SmallVector<uint64_t, 64> Record;
302   BitstreamReader::BlockInfo *CurBlockInfo = nullptr;
303 
304   // Read all the records for this module.
305   while (1) {
306     BitstreamEntry Entry = advanceSkippingSubblocks(AF_DontAutoprocessAbbrevs);
307 
308     switch (Entry.Kind) {
309     case llvm::BitstreamEntry::SubBlock: // Handled for us already.
310     case llvm::BitstreamEntry::Error:
311       return true;
312     case llvm::BitstreamEntry::EndBlock:
313       return false;
314     case llvm::BitstreamEntry::Record:
315       // The interesting case.
316       break;
317     }
318 
319     // Read abbrev records, associate them with CurBID.
320     if (Entry.ID == bitc::DEFINE_ABBREV) {
321       if (!CurBlockInfo) return true;
322       ReadAbbrevRecord();
323 
324       // ReadAbbrevRecord installs the abbrev in CurAbbrevs.  Move it to the
325       // appropriate BlockInfo.
326       CurBlockInfo->Abbrevs.push_back(std::move(CurAbbrevs.back()));
327       CurAbbrevs.pop_back();
328       continue;
329     }
330 
331     // Read a record.
332     Record.clear();
333     switch (readRecord(Entry.ID, Record)) {
334       default: break;  // Default behavior, ignore unknown content.
335       case bitc::BLOCKINFO_CODE_SETBID:
336         if (Record.size() < 1) return true;
337         CurBlockInfo = &BitStream->getOrCreateBlockInfo((unsigned)Record[0]);
338         break;
339       case bitc::BLOCKINFO_CODE_BLOCKNAME: {
340         if (!CurBlockInfo) return true;
341         if (BitStream->isIgnoringBlockInfoNames()) break;  // Ignore name.
342         std::string Name;
343         for (unsigned i = 0, e = Record.size(); i != e; ++i)
344           Name += (char)Record[i];
345         CurBlockInfo->Name = Name;
346         break;
347       }
348       case bitc::BLOCKINFO_CODE_SETRECORDNAME: {
349         if (!CurBlockInfo) return true;
350         if (BitStream->isIgnoringBlockInfoNames()) break;  // Ignore name.
351         std::string Name;
352         for (unsigned i = 1, e = Record.size(); i != e; ++i)
353           Name += (char)Record[i];
354         CurBlockInfo->RecordNames.push_back(std::make_pair((unsigned)Record[0],
355                                                            Name));
356         break;
357       }
358     }
359   }
360 }
361 
362