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 #include "llvm/ADT/StringRef.h"
12 #include <cassert>
13 #include <string>
14 
15 using namespace llvm;
16 
17 //===----------------------------------------------------------------------===//
18 //  BitstreamCursor implementation
19 //===----------------------------------------------------------------------===//
20 
21 /// EnterSubBlock - Having read the ENTER_SUBBLOCK abbrevid, enter
22 /// the block, and return true if the block has an error.
23 bool BitstreamCursor::EnterSubBlock(unsigned BlockID, unsigned *NumWordsP) {
24   // Save the current block's state on BlockScope.
25   BlockScope.push_back(Block(CurCodeSize));
26   BlockScope.back().PrevAbbrevs.swap(CurAbbrevs);
27 
28   // Add the abbrevs specific to this block to the CurAbbrevs list.
29   if (BlockInfo) {
30     if (const BitstreamBlockInfo::BlockInfo *Info =
31             BlockInfo->getBlockInfo(BlockID)) {
32       CurAbbrevs.insert(CurAbbrevs.end(), Info->Abbrevs.begin(),
33                         Info->Abbrevs.end());
34     }
35   }
36 
37   // Get the codesize of this block.
38   CurCodeSize = ReadVBR(bitc::CodeLenWidth);
39   // We can't read more than MaxChunkSize at a time
40   if (CurCodeSize > MaxChunkSize)
41     return true;
42 
43   SkipToFourByteBoundary();
44   unsigned NumWords = Read(bitc::BlockSizeWidth);
45   if (NumWordsP) *NumWordsP = NumWords;
46 
47   // Validate that this block is sane.
48   return CurCodeSize == 0 || AtEndOfStream();
49 }
50 
51 static uint64_t readAbbreviatedField(BitstreamCursor &Cursor,
52                                      const BitCodeAbbrevOp &Op) {
53   assert(!Op.isLiteral() && "Not to be used with literals!");
54 
55   // Decode the value as we are commanded.
56   switch (Op.getEncoding()) {
57   case BitCodeAbbrevOp::Array:
58   case BitCodeAbbrevOp::Blob:
59     llvm_unreachable("Should not reach here");
60   case BitCodeAbbrevOp::Fixed:
61     assert((unsigned)Op.getEncodingData() <= Cursor.MaxChunkSize);
62     return Cursor.Read((unsigned)Op.getEncodingData());
63   case BitCodeAbbrevOp::VBR:
64     assert((unsigned)Op.getEncodingData() <= Cursor.MaxChunkSize);
65     return Cursor.ReadVBR64((unsigned)Op.getEncodingData());
66   case BitCodeAbbrevOp::Char6:
67     return BitCodeAbbrevOp::DecodeChar6(Cursor.Read(6));
68   }
69   llvm_unreachable("invalid abbreviation encoding");
70 }
71 
72 static void skipAbbreviatedField(BitstreamCursor &Cursor,
73                                  const BitCodeAbbrevOp &Op) {
74   assert(!Op.isLiteral() && "Not to be used with literals!");
75 
76   // Decode the value as we are commanded.
77   switch (Op.getEncoding()) {
78   case BitCodeAbbrevOp::Array:
79   case BitCodeAbbrevOp::Blob:
80     llvm_unreachable("Should not reach here");
81   case BitCodeAbbrevOp::Fixed:
82     assert((unsigned)Op.getEncodingData() <= Cursor.MaxChunkSize);
83     Cursor.Read((unsigned)Op.getEncodingData());
84     break;
85   case BitCodeAbbrevOp::VBR:
86     assert((unsigned)Op.getEncodingData() <= Cursor.MaxChunkSize);
87     Cursor.ReadVBR64((unsigned)Op.getEncodingData());
88     break;
89   case BitCodeAbbrevOp::Char6:
90     Cursor.Read(6);
91     break;
92   }
93 }
94 
95 /// skipRecord - Read the current record and discard it.
96 unsigned BitstreamCursor::skipRecord(unsigned AbbrevID) {
97   // Skip unabbreviated records by reading past their entries.
98   if (AbbrevID == bitc::UNABBREV_RECORD) {
99     unsigned Code = ReadVBR(6);
100     unsigned NumElts = ReadVBR(6);
101     for (unsigned i = 0; i != NumElts; ++i)
102       (void)ReadVBR64(6);
103     return Code;
104   }
105 
106   const BitCodeAbbrev *Abbv = getAbbrev(AbbrevID);
107   const BitCodeAbbrevOp &CodeOp = Abbv->getOperandInfo(0);
108   unsigned Code;
109   if (CodeOp.isLiteral())
110     Code = CodeOp.getLiteralValue();
111   else {
112     if (CodeOp.getEncoding() == BitCodeAbbrevOp::Array ||
113         CodeOp.getEncoding() == BitCodeAbbrevOp::Blob)
114       report_fatal_error("Abbreviation starts with an Array or a Blob");
115     Code = readAbbreviatedField(*this, CodeOp);
116   }
117 
118   for (unsigned i = 1, e = Abbv->getNumOperandInfos(); i < e; ++i) {
119     const BitCodeAbbrevOp &Op = Abbv->getOperandInfo(i);
120     if (Op.isLiteral())
121       continue;
122 
123     if (Op.getEncoding() != BitCodeAbbrevOp::Array &&
124         Op.getEncoding() != BitCodeAbbrevOp::Blob) {
125       skipAbbreviatedField(*this, Op);
126       continue;
127     }
128 
129     if (Op.getEncoding() == BitCodeAbbrevOp::Array) {
130       // Array case.  Read the number of elements as a vbr6.
131       unsigned NumElts = ReadVBR(6);
132 
133       // Get the element encoding.
134       assert(i+2 == e && "array op not second to last?");
135       const BitCodeAbbrevOp &EltEnc = Abbv->getOperandInfo(++i);
136 
137       // Read all the elements.
138       // Decode the value as we are commanded.
139       switch (EltEnc.getEncoding()) {
140       default:
141         report_fatal_error("Array element type can't be an Array or a Blob");
142       case BitCodeAbbrevOp::Fixed:
143         assert((unsigned)EltEnc.getEncodingData() <= MaxChunkSize);
144         JumpToBit(GetCurrentBitNo() + NumElts * EltEnc.getEncodingData());
145         break;
146       case BitCodeAbbrevOp::VBR:
147         assert((unsigned)EltEnc.getEncodingData() <= MaxChunkSize);
148         for (; NumElts; --NumElts)
149           ReadVBR64((unsigned)EltEnc.getEncodingData());
150         break;
151       case BitCodeAbbrevOp::Char6:
152         JumpToBit(GetCurrentBitNo() + NumElts * 6);
153         break;
154       }
155       continue;
156     }
157 
158     assert(Op.getEncoding() == BitCodeAbbrevOp::Blob);
159     // Blob case.  Read the number of bytes as a vbr6.
160     unsigned NumElts = ReadVBR(6);
161     SkipToFourByteBoundary();  // 32-bit alignment
162 
163     // Figure out where the end of this blob will be including tail padding.
164     size_t NewEnd = GetCurrentBitNo()+((NumElts+3)&~3)*8;
165 
166     // If this would read off the end of the bitcode file, just set the
167     // record to empty and return.
168     if (!canSkipToPos(NewEnd/8)) {
169       skipToEnd();
170       break;
171     }
172 
173     // Skip over the blob.
174     JumpToBit(NewEnd);
175   }
176   return Code;
177 }
178 
179 unsigned BitstreamCursor::readRecord(unsigned AbbrevID,
180                                      SmallVectorImpl<uint64_t> &Vals,
181                                      StringRef *Blob) {
182   if (AbbrevID == bitc::UNABBREV_RECORD) {
183     unsigned Code = ReadVBR(6);
184     unsigned NumElts = ReadVBR(6);
185     for (unsigned i = 0; i != NumElts; ++i)
186       Vals.push_back(ReadVBR64(6));
187     return Code;
188   }
189 
190   const BitCodeAbbrev *Abbv = getAbbrev(AbbrevID);
191 
192   // Read the record code first.
193   assert(Abbv->getNumOperandInfos() != 0 && "no record code in abbreviation?");
194   const BitCodeAbbrevOp &CodeOp = Abbv->getOperandInfo(0);
195   unsigned Code;
196   if (CodeOp.isLiteral())
197     Code = CodeOp.getLiteralValue();
198   else {
199     if (CodeOp.getEncoding() == BitCodeAbbrevOp::Array ||
200         CodeOp.getEncoding() == BitCodeAbbrevOp::Blob)
201       report_fatal_error("Abbreviation starts with an Array or a Blob");
202     Code = readAbbreviatedField(*this, CodeOp);
203   }
204 
205   for (unsigned i = 1, e = Abbv->getNumOperandInfos(); i != e; ++i) {
206     const BitCodeAbbrevOp &Op = Abbv->getOperandInfo(i);
207     if (Op.isLiteral()) {
208       Vals.push_back(Op.getLiteralValue());
209       continue;
210     }
211 
212     if (Op.getEncoding() != BitCodeAbbrevOp::Array &&
213         Op.getEncoding() != BitCodeAbbrevOp::Blob) {
214       Vals.push_back(readAbbreviatedField(*this, Op));
215       continue;
216     }
217 
218     if (Op.getEncoding() == BitCodeAbbrevOp::Array) {
219       // Array case.  Read the number of elements as a vbr6.
220       unsigned NumElts = ReadVBR(6);
221 
222       // Get the element encoding.
223       if (i + 2 != e)
224         report_fatal_error("Array op not second to last");
225       const BitCodeAbbrevOp &EltEnc = Abbv->getOperandInfo(++i);
226       if (!EltEnc.isEncoding())
227         report_fatal_error(
228             "Array element type has to be an encoding of a type");
229 
230       // Read all the elements.
231       switch (EltEnc.getEncoding()) {
232       default:
233         report_fatal_error("Array element type can't be an Array or a Blob");
234       case BitCodeAbbrevOp::Fixed:
235         for (; NumElts; --NumElts)
236           Vals.push_back(Read((unsigned)EltEnc.getEncodingData()));
237         break;
238       case BitCodeAbbrevOp::VBR:
239         for (; NumElts; --NumElts)
240           Vals.push_back(ReadVBR64((unsigned)EltEnc.getEncodingData()));
241         break;
242       case BitCodeAbbrevOp::Char6:
243         for (; NumElts; --NumElts)
244           Vals.push_back(BitCodeAbbrevOp::DecodeChar6(Read(6)));
245       }
246       continue;
247     }
248 
249     assert(Op.getEncoding() == BitCodeAbbrevOp::Blob);
250     // Blob case.  Read the number of bytes as a vbr6.
251     unsigned NumElts = ReadVBR(6);
252     SkipToFourByteBoundary();  // 32-bit alignment
253 
254     // Figure out where the end of this blob will be including tail padding.
255     size_t CurBitPos = GetCurrentBitNo();
256     size_t NewEnd = CurBitPos+((NumElts+3)&~3)*8;
257 
258     // If this would read off the end of the bitcode file, just set the
259     // record to empty and return.
260     if (!canSkipToPos(NewEnd/8)) {
261       Vals.append(NumElts, 0);
262       skipToEnd();
263       break;
264     }
265 
266     // Otherwise, inform the streamer that we need these bytes in memory.  Skip
267     // over tail padding first, in case jumping to NewEnd invalidates the Blob
268     // pointer.
269     JumpToBit(NewEnd);
270     const char *Ptr = (const char *)getPointerToBit(CurBitPos, NumElts);
271 
272     // If we can return a reference to the data, do so to avoid copying it.
273     if (Blob) {
274       *Blob = StringRef(Ptr, NumElts);
275     } else {
276       // Otherwise, unpack into Vals with zero extension.
277       for (; NumElts; --NumElts)
278         Vals.push_back((unsigned char)*Ptr++);
279     }
280   }
281 
282   return Code;
283 }
284 
285 void BitstreamCursor::ReadAbbrevRecord() {
286   auto Abbv = std::make_shared<BitCodeAbbrev>();
287   unsigned NumOpInfo = ReadVBR(5);
288   for (unsigned i = 0; i != NumOpInfo; ++i) {
289     bool IsLiteral = Read(1);
290     if (IsLiteral) {
291       Abbv->Add(BitCodeAbbrevOp(ReadVBR64(8)));
292       continue;
293     }
294 
295     BitCodeAbbrevOp::Encoding E = (BitCodeAbbrevOp::Encoding)Read(3);
296     if (BitCodeAbbrevOp::hasEncodingData(E)) {
297       uint64_t Data = ReadVBR64(5);
298 
299       // As a special case, handle fixed(0) (i.e., a fixed field with zero bits)
300       // and vbr(0) as a literal zero.  This is decoded the same way, and avoids
301       // a slow path in Read() to have to handle reading zero bits.
302       if ((E == BitCodeAbbrevOp::Fixed || E == BitCodeAbbrevOp::VBR) &&
303           Data == 0) {
304         Abbv->Add(BitCodeAbbrevOp(0));
305         continue;
306       }
307 
308       if ((E == BitCodeAbbrevOp::Fixed || E == BitCodeAbbrevOp::VBR) &&
309           Data > MaxChunkSize)
310         report_fatal_error(
311             "Fixed or VBR abbrev record with size > MaxChunkData");
312 
313       Abbv->Add(BitCodeAbbrevOp(E, Data));
314     } else
315       Abbv->Add(BitCodeAbbrevOp(E));
316   }
317 
318   if (Abbv->getNumOperandInfos() == 0)
319     report_fatal_error("Abbrev record with no operands");
320   CurAbbrevs.push_back(std::move(Abbv));
321 }
322 
323 Optional<BitstreamBlockInfo>
324 BitstreamCursor::ReadBlockInfoBlock(bool ReadBlockInfoNames) {
325   if (EnterSubBlock(bitc::BLOCKINFO_BLOCK_ID)) return None;
326 
327   BitstreamBlockInfo NewBlockInfo;
328 
329   SmallVector<uint64_t, 64> Record;
330   BitstreamBlockInfo::BlockInfo *CurBlockInfo = nullptr;
331 
332   // Read all the records for this module.
333   while (true) {
334     BitstreamEntry Entry = advanceSkippingSubblocks(AF_DontAutoprocessAbbrevs);
335 
336     switch (Entry.Kind) {
337     case llvm::BitstreamEntry::SubBlock: // Handled for us already.
338     case llvm::BitstreamEntry::Error:
339       return None;
340     case llvm::BitstreamEntry::EndBlock:
341       return std::move(NewBlockInfo);
342     case llvm::BitstreamEntry::Record:
343       // The interesting case.
344       break;
345     }
346 
347     // Read abbrev records, associate them with CurBID.
348     if (Entry.ID == bitc::DEFINE_ABBREV) {
349       if (!CurBlockInfo) return None;
350       ReadAbbrevRecord();
351 
352       // ReadAbbrevRecord installs the abbrev in CurAbbrevs.  Move it to the
353       // appropriate BlockInfo.
354       CurBlockInfo->Abbrevs.push_back(std::move(CurAbbrevs.back()));
355       CurAbbrevs.pop_back();
356       continue;
357     }
358 
359     // Read a record.
360     Record.clear();
361     switch (readRecord(Entry.ID, Record)) {
362       default: break;  // Default behavior, ignore unknown content.
363       case bitc::BLOCKINFO_CODE_SETBID:
364         if (Record.size() < 1) return None;
365         CurBlockInfo = &NewBlockInfo.getOrCreateBlockInfo((unsigned)Record[0]);
366         break;
367       case bitc::BLOCKINFO_CODE_BLOCKNAME: {
368         if (!CurBlockInfo) return None;
369         if (!ReadBlockInfoNames)
370           break; // Ignore name.
371         std::string Name;
372         for (unsigned i = 0, e = Record.size(); i != e; ++i)
373           Name += (char)Record[i];
374         CurBlockInfo->Name = Name;
375         break;
376       }
377       case bitc::BLOCKINFO_CODE_SETRECORDNAME: {
378         if (!CurBlockInfo) return None;
379         if (!ReadBlockInfoNames)
380           break; // Ignore name.
381         std::string Name;
382         for (unsigned i = 1, e = Record.size(); i != e; ++i)
383           Name += (char)Record[i];
384         CurBlockInfo->RecordNames.push_back(std::make_pair((unsigned)Record[0],
385                                                            Name));
386         break;
387       }
388     }
389   }
390 }
391