1//===-- RuntimeDyldImpl.h - Run-time dynamic linker for MC-JIT --*- 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// Interface for the implementations of runtime dynamic linker facilities.
11//
12//===----------------------------------------------------------------------===//
13
14#ifndef LLVM_RUNTIME_DYLD_IMPL_H
15#define LLVM_RUNTIME_DYLD_IMPL_H
16
17#include "llvm/ADT/DenseMap.h"
18#include "llvm/ADT/SmallVector.h"
19#include "llvm/ADT/StringMap.h"
20#include "llvm/ADT/Triple.h"
21#include "llvm/ExecutionEngine/ObjectImage.h"
22#include "llvm/ExecutionEngine/RuntimeDyld.h"
23#include "llvm/ExecutionEngine/RuntimeDyldChecker.h"
24#include "llvm/Object/ObjectFile.h"
25#include "llvm/Support/Debug.h"
26#include "llvm/Support/ErrorHandling.h"
27#include "llvm/Support/Format.h"
28#include "llvm/Support/Host.h"
29#include "llvm/Support/Mutex.h"
30#include "llvm/Support/SwapByteOrder.h"
31#include "llvm/Support/raw_ostream.h"
32#include <map>
33#include <system_error>
34
35using namespace llvm;
36using namespace llvm::object;
37
38namespace llvm {
39
40class ObjectBuffer;
41class Twine;
42
43/// SectionEntry - represents a section emitted into memory by the dynamic
44/// linker.
45class SectionEntry {
46public:
47  /// Name - section name.
48  StringRef Name;
49
50  /// Address - address in the linker's memory where the section resides.
51  uint8_t *Address;
52
53  /// Size - section size. Doesn't include the stubs.
54  size_t Size;
55
56  /// LoadAddress - the address of the section in the target process's memory.
57  /// Used for situations in which JIT-ed code is being executed in the address
58  /// space of a separate process.  If the code executes in the same address
59  /// space where it was JIT-ed, this just equals Address.
60  uint64_t LoadAddress;
61
62  /// StubOffset - used for architectures with stub functions for far
63  /// relocations (like ARM).
64  uintptr_t StubOffset;
65
66  /// ObjAddress - address of the section in the in-memory object file.  Used
67  /// for calculating relocations in some object formats (like MachO).
68  uintptr_t ObjAddress;
69
70  SectionEntry(StringRef name, uint8_t *address, size_t size,
71               uintptr_t objAddress)
72      : Name(name), Address(address), Size(size),
73        LoadAddress((uintptr_t)address), StubOffset(size),
74        ObjAddress(objAddress) {}
75};
76
77/// RelocationEntry - used to represent relocations internally in the dynamic
78/// linker.
79class RelocationEntry {
80public:
81  /// SectionID - the section this relocation points to.
82  unsigned SectionID;
83
84  /// Offset - offset into the section.
85  uint64_t Offset;
86
87  /// RelType - relocation type.
88  uint32_t RelType;
89
90  /// Addend - the relocation addend encoded in the instruction itself.  Also
91  /// used to make a relocation section relative instead of symbol relative.
92  int64_t Addend;
93
94  struct SectionPair {
95      uint32_t SectionA;
96      uint32_t SectionB;
97  };
98
99  /// SymOffset - Section offset of the relocation entry's symbol (used for GOT
100  /// lookup).
101  union {
102    uint64_t SymOffset;
103    SectionPair Sections;
104  };
105
106  /// True if this is a PCRel relocation (MachO specific).
107  bool IsPCRel;
108
109  /// The size of this relocation (MachO specific).
110  unsigned Size;
111
112  RelocationEntry(unsigned id, uint64_t offset, uint32_t type, int64_t addend)
113      : SectionID(id), Offset(offset), RelType(type), Addend(addend),
114        SymOffset(0), IsPCRel(false), Size(0) {}
115
116  RelocationEntry(unsigned id, uint64_t offset, uint32_t type, int64_t addend,
117                  uint64_t symoffset)
118      : SectionID(id), Offset(offset), RelType(type), Addend(addend),
119        SymOffset(symoffset), IsPCRel(false), Size(0) {}
120
121  RelocationEntry(unsigned id, uint64_t offset, uint32_t type, int64_t addend,
122                  bool IsPCRel, unsigned Size)
123      : SectionID(id), Offset(offset), RelType(type), Addend(addend),
124        SymOffset(0), IsPCRel(IsPCRel), Size(Size) {}
125
126  RelocationEntry(unsigned id, uint64_t offset, uint32_t type, int64_t addend,
127                  unsigned SectionA, uint64_t SectionAOffset, unsigned SectionB,
128                  uint64_t SectionBOffset, bool IsPCRel, unsigned Size)
129      : SectionID(id), Offset(offset), RelType(type),
130        Addend(SectionAOffset - SectionBOffset + addend), IsPCRel(IsPCRel),
131        Size(Size) {
132    Sections.SectionA = SectionA;
133    Sections.SectionB = SectionB;
134  }
135};
136
137class RelocationValueRef {
138public:
139  unsigned SectionID;
140  uint64_t Offset;
141  int64_t Addend;
142  const char *SymbolName;
143  RelocationValueRef() : SectionID(0), Offset(0), Addend(0),
144                         SymbolName(nullptr) {}
145
146  inline bool operator==(const RelocationValueRef &Other) const {
147    return SectionID == Other.SectionID && Offset == Other.Offset &&
148           Addend == Other.Addend && SymbolName == Other.SymbolName;
149  }
150  inline bool operator<(const RelocationValueRef &Other) const {
151    if (SectionID != Other.SectionID)
152      return SectionID < Other.SectionID;
153    if (Offset != Other.Offset)
154      return Offset < Other.Offset;
155    if (Addend != Other.Addend)
156      return Addend < Other.Addend;
157    return SymbolName < Other.SymbolName;
158  }
159};
160
161class RuntimeDyldImpl {
162  friend class RuntimeDyldChecker;
163private:
164
165  uint64_t getAnySymbolRemoteAddress(StringRef Symbol) {
166    if (uint64_t InternalSymbolAddr = getSymbolLoadAddress(Symbol))
167      return InternalSymbolAddr;
168    return MemMgr->getSymbolAddress(Symbol);
169  }
170
171protected:
172  // The MemoryManager to load objects into.
173  RTDyldMemoryManager *MemMgr;
174
175  // A list of all sections emitted by the dynamic linker.  These sections are
176  // referenced in the code by means of their index in this list - SectionID.
177  typedef SmallVector<SectionEntry, 64> SectionList;
178  SectionList Sections;
179
180  typedef unsigned SID; // Type for SectionIDs
181#define RTDYLD_INVALID_SECTION_ID ((SID)(-1))
182
183  // Keep a map of sections from object file to the SectionID which
184  // references it.
185  typedef std::map<SectionRef, unsigned> ObjSectionToIDMap;
186
187  // A global symbol table for symbols from all loaded modules.  Maps the
188  // symbol name to a (SectionID, offset in section) pair.
189  typedef std::pair<unsigned, uintptr_t> SymbolLoc;
190  typedef StringMap<SymbolLoc> SymbolTableMap;
191  SymbolTableMap GlobalSymbolTable;
192
193  // Pair representing the size and alignment requirement for a common symbol.
194  typedef std::pair<unsigned, unsigned> CommonSymbolInfo;
195  // Keep a map of common symbols to their info pairs
196  typedef std::map<SymbolRef, CommonSymbolInfo> CommonSymbolMap;
197
198  // For each symbol, keep a list of relocations based on it. Anytime
199  // its address is reassigned (the JIT re-compiled the function, e.g.),
200  // the relocations get re-resolved.
201  // The symbol (or section) the relocation is sourced from is the Key
202  // in the relocation list where it's stored.
203  typedef SmallVector<RelocationEntry, 64> RelocationList;
204  // Relocations to sections already loaded. Indexed by SectionID which is the
205  // source of the address. The target where the address will be written is
206  // SectionID/Offset in the relocation itself.
207  DenseMap<unsigned, RelocationList> Relocations;
208
209  // Relocations to external symbols that are not yet resolved.  Symbols are
210  // external when they aren't found in the global symbol table of all loaded
211  // modules.  This map is indexed by symbol name.
212  StringMap<RelocationList> ExternalSymbolRelocations;
213
214  typedef std::map<RelocationValueRef, uintptr_t> StubMap;
215
216  Triple::ArchType Arch;
217  bool IsTargetLittleEndian;
218
219  // True if all sections should be passed to the memory manager, false if only
220  // sections containing relocations should be. Defaults to 'false'.
221  bool ProcessAllSections;
222
223  // This mutex prevents simultaneously loading objects from two different
224  // threads.  This keeps us from having to protect individual data structures
225  // and guarantees that section allocation requests to the memory manager
226  // won't be interleaved between modules.  It is also used in mapSectionAddress
227  // and resolveRelocations to protect write access to internal data structures.
228  //
229  // loadObject may be called on the same thread during the handling of of
230  // processRelocations, and that's OK.  The handling of the relocation lists
231  // is written in such a way as to work correctly if new elements are added to
232  // the end of the list while the list is being processed.
233  sys::Mutex lock;
234
235  virtual unsigned getMaxStubSize() = 0;
236  virtual unsigned getStubAlignment() = 0;
237
238  bool HasError;
239  std::string ErrorStr;
240
241  // Set the error state and record an error string.
242  bool Error(const Twine &Msg) {
243    ErrorStr = Msg.str();
244    HasError = true;
245    return true;
246  }
247
248  uint64_t getSectionLoadAddress(unsigned SectionID) {
249    return Sections[SectionID].LoadAddress;
250  }
251
252  uint8_t *getSectionAddress(unsigned SectionID) {
253    return (uint8_t *)Sections[SectionID].Address;
254  }
255
256  void writeInt16BE(uint8_t *Addr, uint16_t Value) {
257    if (IsTargetLittleEndian)
258      sys::swapByteOrder(Value);
259    *Addr       = (Value >> 8) & 0xFF;
260    *(Addr + 1) = Value & 0xFF;
261  }
262
263  void writeInt32BE(uint8_t *Addr, uint32_t Value) {
264    if (IsTargetLittleEndian)
265      sys::swapByteOrder(Value);
266    *Addr       = (Value >> 24) & 0xFF;
267    *(Addr + 1) = (Value >> 16) & 0xFF;
268    *(Addr + 2) = (Value >> 8) & 0xFF;
269    *(Addr + 3) = Value & 0xFF;
270  }
271
272  void writeInt64BE(uint8_t *Addr, uint64_t Value) {
273    if (IsTargetLittleEndian)
274      sys::swapByteOrder(Value);
275    *Addr       = (Value >> 56) & 0xFF;
276    *(Addr + 1) = (Value >> 48) & 0xFF;
277    *(Addr + 2) = (Value >> 40) & 0xFF;
278    *(Addr + 3) = (Value >> 32) & 0xFF;
279    *(Addr + 4) = (Value >> 24) & 0xFF;
280    *(Addr + 5) = (Value >> 16) & 0xFF;
281    *(Addr + 6) = (Value >> 8) & 0xFF;
282    *(Addr + 7) = Value & 0xFF;
283  }
284
285  /// \brief Given the common symbols discovered in the object file, emit a
286  /// new section for them and update the symbol mappings in the object and
287  /// symbol table.
288  void emitCommonSymbols(ObjectImage &Obj, const CommonSymbolMap &CommonSymbols,
289                         uint64_t TotalSize, SymbolTableMap &SymbolTable);
290
291  /// \brief Emits section data from the object file to the MemoryManager.
292  /// \param IsCode if it's true then allocateCodeSection() will be
293  ///        used for emits, else allocateDataSection() will be used.
294  /// \return SectionID.
295  unsigned emitSection(ObjectImage &Obj, const SectionRef &Section,
296                       bool IsCode);
297
298  /// \brief Find Section in LocalSections. If the secton is not found - emit
299  ///        it and store in LocalSections.
300  /// \param IsCode if it's true then allocateCodeSection() will be
301  ///        used for emmits, else allocateDataSection() will be used.
302  /// \return SectionID.
303  unsigned findOrEmitSection(ObjectImage &Obj, const SectionRef &Section,
304                             bool IsCode, ObjSectionToIDMap &LocalSections);
305
306  // \brief Add a relocation entry that uses the given section.
307  void addRelocationForSection(const RelocationEntry &RE, unsigned SectionID);
308
309  // \brief Add a relocation entry that uses the given symbol.  This symbol may
310  // be found in the global symbol table, or it may be external.
311  void addRelocationForSymbol(const RelocationEntry &RE, StringRef SymbolName);
312
313  /// \brief Emits long jump instruction to Addr.
314  /// \return Pointer to the memory area for emitting target address.
315  uint8_t *createStubFunction(uint8_t *Addr);
316
317  /// \brief Resolves relocations from Relocs list with address from Value.
318  void resolveRelocationList(const RelocationList &Relocs, uint64_t Value);
319
320  /// \brief A object file specific relocation resolver
321  /// \param RE The relocation to be resolved
322  /// \param Value Target symbol address to apply the relocation action
323  virtual void resolveRelocation(const RelocationEntry &RE, uint64_t Value) = 0;
324
325  /// \brief Parses one or more object file relocations (some object files use
326  ///        relocation pairs) and stores it to Relocations or SymbolRelocations
327  ///        (this depends on the object file type).
328  /// \return Iterator to the next relocation that needs to be parsed.
329  virtual relocation_iterator
330  processRelocationRef(unsigned SectionID, relocation_iterator RelI,
331                       ObjectImage &Obj, ObjSectionToIDMap &ObjSectionToID,
332                       const SymbolTableMap &Symbols, StubMap &Stubs) = 0;
333
334  /// \brief Resolve relocations to external symbols.
335  void resolveExternalSymbols();
336
337  /// \brief Update GOT entries for external symbols.
338  // The base class does nothing.  ELF overrides this.
339  virtual void updateGOTEntries(StringRef Name, uint64_t Addr) {}
340
341  // \brief Compute an upper bound of the memory that is required to load all
342  // sections
343  void computeTotalAllocSize(ObjectImage &Obj, uint64_t &CodeSize,
344                             uint64_t &DataSizeRO, uint64_t &DataSizeRW);
345
346  // \brief Compute the stub buffer size required for a section
347  unsigned computeSectionStubBufSize(ObjectImage &Obj,
348                                     const SectionRef &Section);
349
350public:
351  RuntimeDyldImpl(RTDyldMemoryManager *mm)
352      : MemMgr(mm), ProcessAllSections(false), HasError(false) {
353  }
354
355  virtual ~RuntimeDyldImpl();
356
357  void setProcessAllSections(bool ProcessAllSections) {
358    this->ProcessAllSections = ProcessAllSections;
359  }
360
361  ObjectImage *loadObject(ObjectImage *InputObject);
362
363  uint8_t* getSymbolAddress(StringRef Name) {
364    // FIXME: Just look up as a function for now. Overly simple of course.
365    // Work in progress.
366    SymbolTableMap::const_iterator pos = GlobalSymbolTable.find(Name);
367    if (pos == GlobalSymbolTable.end())
368      return nullptr;
369    SymbolLoc Loc = pos->second;
370    return getSectionAddress(Loc.first) + Loc.second;
371  }
372
373  uint64_t getSymbolLoadAddress(StringRef Name) {
374    // FIXME: Just look up as a function for now. Overly simple of course.
375    // Work in progress.
376    SymbolTableMap::const_iterator pos = GlobalSymbolTable.find(Name);
377    if (pos == GlobalSymbolTable.end())
378      return 0;
379    SymbolLoc Loc = pos->second;
380    return getSectionLoadAddress(Loc.first) + Loc.second;
381  }
382
383  void resolveRelocations();
384
385  void reassignSectionAddress(unsigned SectionID, uint64_t Addr);
386
387  void mapSectionAddress(const void *LocalAddress, uint64_t TargetAddress);
388
389  // Is the linker in an error state?
390  bool hasError() { return HasError; }
391
392  // Mark the error condition as handled and continue.
393  void clearError() { HasError = false; }
394
395  // Get the error message.
396  StringRef getErrorString() { return ErrorStr; }
397
398  virtual bool isCompatibleFormat(const ObjectBuffer *Buffer) const = 0;
399  virtual bool isCompatibleFile(const ObjectFile *Obj) const = 0;
400
401  virtual void registerEHFrames();
402
403  virtual void deregisterEHFrames();
404
405  virtual void finalizeLoad(ObjectImage &ObjImg, ObjSectionToIDMap &SectionMap) {}
406};
407
408} // end namespace llvm
409
410#endif
411