summaryrefslogtreecommitdiffstats
path: root/courgette/encoded_program.cc
blob: 59800c50c99d9ec75d7cf2f70994bf9b0b1486c1 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
// Copyright (c) 2011 The Chromium Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.

#include "courgette/encoded_program.h"

#include <stddef.h>
#include <stdint.h>

#include <algorithm>
#include <map>
#include <string>
#include <vector>

#include "base/environment.h"
#include "base/logging.h"
#include "base/memory/scoped_ptr.h"
#include "base/numerics/safe_conversions.h"
#include "base/numerics/safe_math.h"
#include "base/strings/string_number_conversions.h"
#include "base/strings/string_util.h"
#include "courgette/courgette.h"
#include "courgette/disassembler_elf_32_arm.h"
#include "courgette/streams.h"
#include "courgette/types_elf.h"

namespace courgette {

namespace {

// Serializes a vector of integral values using Varint32 coding.
template<typename V>
CheckBool WriteVector(const V& items, SinkStream* buffer) {
  size_t count = items.size();
  bool ok = buffer->WriteSizeVarint32(count);
  for (size_t i = 0; ok && i < count;  ++i) {
    ok = buffer->WriteSizeVarint32(items[i]);
  }
  return ok;
}

template<typename V>
bool ReadVector(V* items, SourceStream* buffer) {
  uint32_t count;
  if (!buffer->ReadVarint32(&count))
    return false;

  items->clear();

  bool ok = items->reserve(count);
  for (size_t i = 0;  ok && i < count;  ++i) {
    uint32_t item;
    ok = buffer->ReadVarint32(&item);
    if (ok)
      ok = items->push_back(static_cast<typename V::value_type>(item));
  }

  return ok;
}

// Serializes a vector, using delta coding followed by Varint32Signed coding.
template<typename V>
CheckBool WriteSigned32Delta(const V& set, SinkStream* buffer) {
  size_t count = set.size();
  bool ok = buffer->WriteSizeVarint32(count);
  uint32_t prev = 0;
  for (size_t i = 0; ok && i < count; ++i) {
    uint32_t current = set[i];
    int32_t delta = current - prev;
    ok = buffer->WriteVarint32Signed(delta);
    prev = current;
  }
  return ok;
}

template <typename V>
static CheckBool ReadSigned32Delta(V* set, SourceStream* buffer) {
  uint32_t count;

  if (!buffer->ReadVarint32(&count))
    return false;

  set->clear();
  bool ok = set->reserve(count);
  uint32_t prev = 0;
  for (size_t i = 0; ok && i < count; ++i) {
    int32_t delta;
    ok = buffer->ReadVarint32Signed(&delta);
    if (ok) {
      uint32_t current = static_cast<uint32_t>(prev + delta);
      ok = set->push_back(current);
      prev = current;
    }
  }
  return ok;
}

// Write a vector as the byte representation of the contents.
//
// (This only really makes sense for a type T that has sizeof(T)==1, otherwise
// serialized representation is not endian-agnostic.  But it is useful to keep
// the possibility of a greater size for experiments comparing Varint32 encoding
// of a vector of larger integrals vs a plain form.)
//
template<typename V>
CheckBool WriteVectorU8(const V& items, SinkStream* buffer) {
  size_t count = items.size();
  bool ok = buffer->WriteSizeVarint32(count);
  if (count != 0 && ok) {
    size_t byte_count = count * sizeof(typename V::value_type);
    ok = buffer->Write(static_cast<const void*>(&items[0]), byte_count);
  }
  return ok;
}

template<typename V>
bool ReadVectorU8(V* items, SourceStream* buffer) {
  uint32_t count;
  if (!buffer->ReadVarint32(&count))
    return false;

  items->clear();
  bool ok = items->resize(count, 0);
  if (ok && count != 0) {
    size_t byte_count = count * sizeof(typename V::value_type);
    return buffer->Read(static_cast<void*>(&((*items)[0])), byte_count);
  }
  return ok;
}

}  // namespace

////////////////////////////////////////////////////////////////////////////////

// Constructor is here rather than in the header. Although the constructor
// appears to do nothing it is fact quite large because of the implicit calls to
// field constructors. Ditto for the destructor.
EncodedProgram::EncodedProgram() {}
EncodedProgram::~EncodedProgram() {}

CheckBool EncodedProgram::DefineLabels(const RVAToLabel& abs32_labels,
                                       const RVAToLabel& rel32_labels) {
  // which == 0 => abs32; which == 1 => rel32.
  for (int which = 0; which < 2; ++which) {
    const RVAToLabel& labels = which == 0 ? abs32_labels : rel32_labels;
    RvaVector& rvas = which == 0 ? abs32_rva_ : rel32_rva_;

    if (!rvas.resize(LabelManager::GetIndexBound(labels), kUnassignedRVA))
      return false;

    // For each Label, write its RVA to assigned index.
    for (const auto& rva_and_label : labels) {
      const Label& label = *rva_and_label.second;
      DCHECK_EQ(rva_and_label.first, label.rva_);
      DCHECK_NE(label.index_, Label::kNoIndex);
      DCHECK_EQ(rvas[label.index_], kUnassignedRVA)
          << "DefineLabels() double assigned " << label.index_;
      rvas[label.index_] = label.rva_;
    }

    // Replace all unassigned slots with the value at the previous index so they
    // delta-encode to zero. (There might be better values than zero. The way to
    // get that is have the higher level assembly program assign the unassigned
    // slots.)
    RVA previous = 0;
    for (RVA& rva : rvas) {
      if (rva == kUnassignedRVA)
        rva = previous;
      else
        previous = rva;
    }
  }
  return true;
}

CheckBool EncodedProgram::AddOrigin(RVA origin) {
  return ops_.push_back(ORIGIN) && origins_.push_back(origin);
}

CheckBool EncodedProgram::AddCopy(size_t count, const void* bytes) {
  const uint8_t* source = static_cast<const uint8_t*>(bytes);

  bool ok = true;

  // Fold adjacent COPY instructions into one.  This nearly halves the size of
  // an EncodedProgram with only COPY1 instructions since there are approx plain
  // 16 bytes per reloc.  This has a working-set benefit during decompression.
  // For compression of files with large differences this makes a small (4%)
  // improvement in size.  For files with small differences this degrades the
  // compressed size by 1.3%
  if (!ops_.empty()) {
    if (ops_.back() == COPY1) {
      ops_.back() = COPY;
      ok = copy_counts_.push_back(1);
    }
    if (ok && ops_.back() == COPY) {
      copy_counts_.back() += count;
      for (size_t i = 0; ok && i < count; ++i) {
        ok = copy_bytes_.push_back(source[i]);
      }
      return ok;
    }
  }

  if (ok) {
    if (count == 1) {
      ok = ops_.push_back(COPY1) && copy_bytes_.push_back(source[0]);
    } else {
      ok = ops_.push_back(COPY) && copy_counts_.push_back(count);
      for (size_t i = 0; ok && i < count; ++i) {
        ok = copy_bytes_.push_back(source[i]);
      }
    }
  }

  return ok;
}

CheckBool EncodedProgram::AddAbs32(int label_index) {
  return ops_.push_back(ABS32) && abs32_ix_.push_back(label_index);
}

CheckBool EncodedProgram::AddAbs64(int label_index) {
  return ops_.push_back(ABS64) && abs32_ix_.push_back(label_index);
}

CheckBool EncodedProgram::AddRel32(int label_index) {
  return ops_.push_back(REL32) && rel32_ix_.push_back(label_index);
}

CheckBool EncodedProgram::AddRel32ARM(uint16_t op, int label_index) {
  return ops_.push_back(static_cast<OP>(op)) &&
      rel32_ix_.push_back(label_index);
}

CheckBool EncodedProgram::AddPeMakeRelocs(ExecutableType kind) {
  if (kind == EXE_WIN_32_X86)
    return ops_.push_back(MAKE_PE_RELOCATION_TABLE);
  return ops_.push_back(MAKE_PE64_RELOCATION_TABLE);
}

CheckBool EncodedProgram::AddElfMakeRelocs() {
  return ops_.push_back(MAKE_ELF_RELOCATION_TABLE);
}

CheckBool EncodedProgram::AddElfARMMakeRelocs() {
  return ops_.push_back(MAKE_ELF_ARM_RELOCATION_TABLE);
}

void EncodedProgram::DebuggingSummary() {
  VLOG(1) << "EncodedProgram Summary"
          << "\n  image base  " << image_base_
          << "\n  abs32 rvas  " << abs32_rva_.size()
          << "\n  rel32 rvas  " << rel32_rva_.size()
          << "\n  ops         " << ops_.size()
          << "\n  origins     " << origins_.size()
          << "\n  copy_counts " << copy_counts_.size()
          << "\n  copy_bytes  " << copy_bytes_.size()
          << "\n  abs32_ix    " << abs32_ix_.size()
          << "\n  rel32_ix    " << rel32_ix_.size();
}

////////////////////////////////////////////////////////////////////////////////

// For algorithm refinement purposes it is useful to write subsets of the file
// format.  This gives us the ability to estimate the entropy of the
// differential compression of the individual streams, which can provide
// invaluable insights.  The default, of course, is to include all the streams.
//
enum FieldSelect {
  INCLUDE_ABS32_ADDRESSES = 0x0001,
  INCLUDE_REL32_ADDRESSES = 0x0002,
  INCLUDE_ABS32_INDEXES   = 0x0010,
  INCLUDE_REL32_INDEXES   = 0x0020,
  INCLUDE_OPS             = 0x0100,
  INCLUDE_BYTES           = 0x0200,
  INCLUDE_COPY_COUNTS     = 0x0400,
  INCLUDE_MISC            = 0x1000
};

static FieldSelect GetFieldSelect() {
  // TODO(sra): Use better configuration.
  scoped_ptr<base::Environment> env(base::Environment::Create());
  std::string s;
  env->GetVar("A_FIELDS", &s);
  uint64_t fields;
  if (!base::StringToUint64(s, &fields))
    return static_cast<FieldSelect>(~0);
  return static_cast<FieldSelect>(fields);
}

CheckBool EncodedProgram::WriteTo(SinkStreamSet* streams) {
  FieldSelect select = GetFieldSelect();

  // The order of fields must be consistent in WriteTo and ReadFrom, regardless
  // of the streams used.  The code can be configured with all kStreamXXX
  // constants the same.
  //
  // If we change the code to pipeline reading with assembly (to avoid temporary
  // storage vectors by consuming operands directly from the stream) then we
  // need to read the base address and the random access address tables first,
  // the rest can be interleaved.

  if (select & INCLUDE_MISC) {
    uint32_t high = static_cast<uint32_t>(image_base_ >> 32);
    uint32_t low = static_cast<uint32_t>(image_base_ & 0xffffffffU);

    if (!streams->stream(kStreamMisc)->WriteVarint32(high) ||
        !streams->stream(kStreamMisc)->WriteVarint32(low)) {
      return false;
    }
  }

  bool success = true;

  if (select & INCLUDE_ABS32_ADDRESSES) {
    success &= WriteSigned32Delta(abs32_rva_,
                                  streams->stream(kStreamAbs32Addresses));
  }

  if (select & INCLUDE_REL32_ADDRESSES) {
    success &= WriteSigned32Delta(rel32_rva_,
                                  streams->stream(kStreamRel32Addresses));
  }

  if (select & INCLUDE_MISC)
    success &= WriteVector(origins_, streams->stream(kStreamOriginAddresses));

  if (select & INCLUDE_OPS) {
    // 5 for length.
    success &= streams->stream(kStreamOps)->Reserve(ops_.size() + 5);
    success &= WriteVector(ops_, streams->stream(kStreamOps));
  }

  if (select & INCLUDE_COPY_COUNTS)
    success &= WriteVector(copy_counts_, streams->stream(kStreamCopyCounts));

  if (select & INCLUDE_BYTES)
    success &= WriteVectorU8(copy_bytes_, streams->stream(kStreamBytes));

  if (select & INCLUDE_ABS32_INDEXES)
    success &= WriteVector(abs32_ix_, streams->stream(kStreamAbs32Indexes));

  if (select & INCLUDE_REL32_INDEXES)
    success &= WriteVector(rel32_ix_, streams->stream(kStreamRel32Indexes));

  return success;
}

bool EncodedProgram::ReadFrom(SourceStreamSet* streams) {
  uint32_t high;
  uint32_t low;

  if (!streams->stream(kStreamMisc)->ReadVarint32(&high) ||
      !streams->stream(kStreamMisc)->ReadVarint32(&low)) {
    return false;
  }
  image_base_ = (static_cast<uint64_t>(high) << 32) | low;

  if (!ReadSigned32Delta(&abs32_rva_, streams->stream(kStreamAbs32Addresses)))
    return false;
  if (!ReadSigned32Delta(&rel32_rva_, streams->stream(kStreamRel32Addresses)))
    return false;
  if (!ReadVector(&origins_, streams->stream(kStreamOriginAddresses)))
    return false;
  if (!ReadVector(&ops_, streams->stream(kStreamOps)))
    return false;
  if (!ReadVector(&copy_counts_, streams->stream(kStreamCopyCounts)))
    return false;
  if (!ReadVectorU8(&copy_bytes_, streams->stream(kStreamBytes)))
    return false;
  if (!ReadVector(&abs32_ix_, streams->stream(kStreamAbs32Indexes)))
    return false;
  if (!ReadVector(&rel32_ix_, streams->stream(kStreamRel32Indexes)))
    return false;

  // Check that streams have been completely consumed.
  for (int i = 0;  i < kStreamLimit;  ++i) {
    if (streams->stream(i)->Remaining() > 0)
      return false;
  }

  return true;
}

// Safe, non-throwing version of std::vector::at().  Returns 'true' for success,
// 'false' for out-of-bounds index error.
template<typename V, typename T>
bool VectorAt(const V& v, size_t index, T* output) {
  if (index >= v.size())
    return false;
  *output = v[index];
  return true;
}

CheckBool EncodedProgram::EvaluateRel32ARM(OP op,
                                           size_t& ix_rel32_ix,
                                           RVA& current_rva,
                                           SinkStream* output) {
  switch (op & 0x0000F000) {
    case REL32ARM8: {
      uint32_t index;
      if (!VectorAt(rel32_ix_, ix_rel32_ix, &index))
        return false;
      ++ix_rel32_ix;
      RVA rva;
      if (!VectorAt(rel32_rva_, index, &rva))
        return false;
      uint32_t decompressed_op;
      if (!DisassemblerElf32ARM::Decompress(
              ARM_OFF8, static_cast<uint16_t>(op),
              static_cast<uint32_t>(rva - current_rva), &decompressed_op)) {
        return false;
      }
      uint16_t op16 = static_cast<uint16_t>(decompressed_op);
      if (!output->Write(&op16, 2))
        return false;
      current_rva += 2;
      break;
    }
    case REL32ARM11: {
      uint32_t index;
      if (!VectorAt(rel32_ix_, ix_rel32_ix, &index))
        return false;
      ++ix_rel32_ix;
      RVA rva;
      if (!VectorAt(rel32_rva_, index, &rva))
        return false;
      uint32_t decompressed_op;
      if (!DisassemblerElf32ARM::Decompress(ARM_OFF11, (uint16_t)op,
                                            (uint32_t)(rva - current_rva),
                                            &decompressed_op)) {
        return false;
      }
      uint16_t op16 = static_cast<uint16_t>(decompressed_op);
      if (!output->Write(&op16, 2))
        return false;
      current_rva += 2;
      break;
    }
    case REL32ARM24: {
      uint32_t index;
      if (!VectorAt(rel32_ix_, ix_rel32_ix, &index))
        return false;
      ++ix_rel32_ix;
      RVA rva;
      if (!VectorAt(rel32_rva_, index, &rva))
        return false;
      uint32_t decompressed_op;
      if (!DisassemblerElf32ARM::Decompress(ARM_OFF24, (uint16_t)op,
                                            (uint32_t)(rva - current_rva),
                                            &decompressed_op)) {
        return false;
      }
      if (!output->Write(&decompressed_op, 4))
        return false;
      current_rva += 4;
      break;
    }
    case REL32ARM25: {
      uint32_t index;
      if (!VectorAt(rel32_ix_, ix_rel32_ix, &index))
        return false;
      ++ix_rel32_ix;
      RVA rva;
      if (!VectorAt(rel32_rva_, index, &rva))
        return false;
      uint32_t decompressed_op;
      if (!DisassemblerElf32ARM::Decompress(ARM_OFF25, (uint16_t)op,
                                            (uint32_t)(rva - current_rva),
                                            &decompressed_op)) {
        return false;
      }
      uint32_t words = (decompressed_op << 16) | (decompressed_op >> 16);
      if (!output->Write(&words, 4))
        return false;
      current_rva += 4;
      break;
    }
    case REL32ARM21: {
      uint32_t index;
      if (!VectorAt(rel32_ix_, ix_rel32_ix, &index))
        return false;
      ++ix_rel32_ix;
      RVA rva;
      if (!VectorAt(rel32_rva_, index, &rva))
        return false;
      uint32_t decompressed_op;
      if (!DisassemblerElf32ARM::Decompress(ARM_OFF21, (uint16_t)op,
                                            (uint32_t)(rva - current_rva),
                                            &decompressed_op)) {
        return false;
      }
      uint32_t words = (decompressed_op << 16) | (decompressed_op >> 16);
      if (!output->Write(&words, 4))
        return false;
      current_rva += 4;
      break;
    }
    default:
      return false;
  }

  return true;
}

CheckBool EncodedProgram::AssembleTo(SinkStream* final_buffer) {
  // For the most part, the assembly process walks the various tables.
  // ix_mumble is the index into the mumble table.
  size_t ix_origins = 0;
  size_t ix_copy_counts = 0;
  size_t ix_copy_bytes = 0;
  size_t ix_abs32_ix = 0;
  size_t ix_rel32_ix = 0;

  RVA current_rva = 0;

  bool pending_pe_relocation_table = false;
  uint8_t pending_pe_relocation_table_type = 0x03;  // IMAGE_REL_BASED_HIGHLOW
  Elf32_Word pending_elf_relocation_table_type = 0;
  SinkStream bytes_following_relocation_table;

  SinkStream* output = final_buffer;

  for (size_t ix_ops = 0;  ix_ops < ops_.size();  ++ix_ops) {
    OP op = ops_[ix_ops];

    switch (op) {
      default:
        if (!EvaluateRel32ARM(op, ix_rel32_ix, current_rva, output))
          return false;
        break;

      case ORIGIN: {
        RVA section_rva;
        if (!VectorAt(origins_, ix_origins, &section_rva))
          return false;
        ++ix_origins;
        current_rva = section_rva;
        break;
      }

      case COPY: {
        size_t count;
        if (!VectorAt(copy_counts_, ix_copy_counts, &count))
          return false;
        ++ix_copy_counts;
        for (size_t i = 0;  i < count;  ++i) {
          uint8_t b;
          if (!VectorAt(copy_bytes_, ix_copy_bytes, &b))
            return false;
          ++ix_copy_bytes;
          if (!output->Write(&b, 1))
            return false;
        }
        current_rva += static_cast<RVA>(count);
        break;
      }

      case COPY1: {
        uint8_t b;
        if (!VectorAt(copy_bytes_, ix_copy_bytes, &b))
          return false;
        ++ix_copy_bytes;
        if (!output->Write(&b, 1))
          return false;
        current_rva += 1;
        break;
      }

      case REL32: {
        uint32_t index;
        if (!VectorAt(rel32_ix_, ix_rel32_ix, &index))
          return false;
        ++ix_rel32_ix;
        RVA rva;
        if (!VectorAt(rel32_rva_, index, &rva))
          return false;
        uint32_t offset = (rva - (current_rva + 4));
        if (!output->Write(&offset, 4))
          return false;
        current_rva += 4;
        break;
      }

      case ABS32:
      case ABS64: {
        uint32_t index;
        if (!VectorAt(abs32_ix_, ix_abs32_ix, &index))
          return false;
        ++ix_abs32_ix;
        RVA rva;
        if (!VectorAt(abs32_rva_, index, &rva))
          return false;
        if (op == ABS32) {
          base::CheckedNumeric<uint32_t> abs32 = image_base_;
          abs32 += rva;
          uint32_t safe_abs32 = abs32.ValueOrDie();
          if (!abs32_relocs_.push_back(current_rva) ||
              !output->Write(&safe_abs32, 4)) {
            return false;
          }
          current_rva += 4;
        } else {
          base::CheckedNumeric<uint64_t> abs64 = image_base_;
          abs64 += rva;
          uint64_t safe_abs64 = abs64.ValueOrDie();
          if (!abs32_relocs_.push_back(current_rva) ||
              !output->Write(&safe_abs64, 8)) {
            return false;
          }
          current_rva += 8;
        }
        break;
      }

      case MAKE_PE_RELOCATION_TABLE: {
        // We can see the base relocation anywhere, but we only have the
        // information to generate it at the very end.  So we divert the bytes
        // we are generating to a temporary stream.
        if (pending_pe_relocation_table)
          return false;  // Can't have two base relocation tables.

        pending_pe_relocation_table = true;
        output = &bytes_following_relocation_table;
        break;
        // There is a potential problem *if* the instruction stream contains
        // some REL32 relocations following the base relocation and in the same
        // section.  We don't know the size of the table, so 'current_rva' will
        // be wrong, causing REL32 offsets to be miscalculated.  This never
        // happens; the base relocation table is usually in a section of its
        // own, a data-only section, and following everything else in the
        // executable except some padding zero bytes.  We could fix this by
        // emitting an ORIGIN after the MAKE_BASE_RELOCATION_TABLE.
      }

      case MAKE_PE64_RELOCATION_TABLE: {
        if (pending_pe_relocation_table)
          return false;  // Can't have two base relocation tables.

        pending_pe_relocation_table = true;
        pending_pe_relocation_table_type = 0x0A;  // IMAGE_REL_BASED_DIR64
        output = &bytes_following_relocation_table;
        break;
      }

      case MAKE_ELF_ARM_RELOCATION_TABLE: {
        // We can see the base relocation anywhere, but we only have the
        // information to generate it at the very end.  So we divert the bytes
        // we are generating to a temporary stream.
        if (pending_elf_relocation_table_type)
          return false;  // Can't have two base relocation tables.

        pending_elf_relocation_table_type = R_ARM_RELATIVE;
        output = &bytes_following_relocation_table;
        break;
      }

      case MAKE_ELF_RELOCATION_TABLE: {
        // We can see the base relocation anywhere, but we only have the
        // information to generate it at the very end.  So we divert the bytes
        // we are generating to a temporary stream.
        if (pending_elf_relocation_table_type)
          return false;  // Can't have two base relocation tables.

        pending_elf_relocation_table_type = R_386_RELATIVE;
        output = &bytes_following_relocation_table;
        break;
      }
    }
  }

  if (pending_pe_relocation_table) {
    if (!GeneratePeRelocations(final_buffer,
                               pending_pe_relocation_table_type) ||
        !final_buffer->Append(&bytes_following_relocation_table))
      return false;
  }

  if (pending_elf_relocation_table_type) {
    if (!GenerateElfRelocations(pending_elf_relocation_table_type,
                                final_buffer) ||
        !final_buffer->Append(&bytes_following_relocation_table))
      return false;
  }

  // Final verification check: did we consume all lists?
  if (ix_copy_counts != copy_counts_.size())
    return false;
  if (ix_copy_bytes != copy_bytes_.size())
    return false;
  if (ix_abs32_ix != abs32_ix_.size())
    return false;
  if (ix_rel32_ix != rel32_ix_.size())
    return false;

  return true;
}

// RelocBlock has the layout of a block of relocations in the base relocation
// table file format.
//
struct RelocBlockPOD {
  uint32_t page_rva;
  uint32_t block_size;
  uint16_t relocs[4096];  // Allow up to one relocation per byte of a 4k page.
};

static_assert(offsetof(RelocBlockPOD, relocs) == 8, "reloc block header size");

class RelocBlock {
 public:
  RelocBlock() {
    pod.page_rva = 0xFFFFFFFF;
    pod.block_size = 8;
  }

  void Add(uint16_t item) {
    pod.relocs[(pod.block_size-8)/2] = item;
    pod.block_size += 2;
  }

  CheckBool Flush(SinkStream* buffer) WARN_UNUSED_RESULT {
    bool ok = true;
    if (pod.block_size != 8) {
      if (pod.block_size % 4 != 0) {  // Pad to make size multiple of 4 bytes.
        Add(0);
      }
      ok = buffer->Write(&pod, pod.block_size);
      pod.block_size = 8;
    }
    return ok;
  }
  RelocBlockPOD pod;
};

CheckBool EncodedProgram::GeneratePeRelocations(SinkStream* buffer,
                                                uint8_t type) {
  std::sort(abs32_relocs_.begin(), abs32_relocs_.end());
  DCHECK(abs32_relocs_.empty() || abs32_relocs_.back() != kUnassignedRVA);

  RelocBlock block;

  bool ok = true;
  for (size_t i = 0;  ok && i < abs32_relocs_.size();  ++i) {
    uint32_t rva = abs32_relocs_[i];
    uint32_t page_rva = rva & ~0xFFF;
    if (page_rva != block.pod.page_rva) {
      ok &= block.Flush(buffer);
      block.pod.page_rva = page_rva;
    }
    if (ok)
      block.Add(((static_cast<uint16_t>(type)) << 12) | (rva & 0xFFF));
  }
  ok &= block.Flush(buffer);
  return ok;
}

CheckBool EncodedProgram::GenerateElfRelocations(Elf32_Word r_info,
                                                 SinkStream* buffer) {
  std::sort(abs32_relocs_.begin(), abs32_relocs_.end());
  DCHECK(abs32_relocs_.empty() || abs32_relocs_.back() != kUnassignedRVA);

  Elf32_Rel relocation_block;

  relocation_block.r_info = r_info;

  bool ok = true;
  for (size_t i = 0;  ok && i < abs32_relocs_.size();  ++i) {
    relocation_block.r_offset = abs32_relocs_[i];
    ok = buffer->Write(&relocation_block, sizeof(Elf32_Rel));
  }

  return ok;
}
////////////////////////////////////////////////////////////////////////////////

Status WriteEncodedProgram(EncodedProgram* encoded, SinkStreamSet* sink) {
  if (!encoded->WriteTo(sink))
    return C_STREAM_ERROR;
  return C_OK;
}

Status ReadEncodedProgram(SourceStreamSet* streams, EncodedProgram** output) {
  EncodedProgram* encoded = new EncodedProgram();
  if (encoded->ReadFrom(streams)) {
    *output = encoded;
    return C_OK;
  }
  delete encoded;
  return C_DESERIALIZATION_FAILED;
}

Status Assemble(EncodedProgram* encoded, SinkStream* buffer) {
  bool assembled = encoded->AssembleTo(buffer);
  if (assembled)
    return C_OK;
  return C_ASSEMBLY_FAILED;
}

void DeleteEncodedProgram(EncodedProgram* encoded) {
  delete encoded;
}

}  // namespace courgette