aboutsummaryrefslogtreecommitdiffstats
path: root/applypatch/imgdiff.c
blob: 6b9ebee5c5e577c075c66c946dbf04f370e675a0 (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
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
/*
 * Copyright (C) 2009 The Android Open Source Project
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *      http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

/*
 * This program constructs binary patches for images -- such as boot.img
 * and recovery.img -- that consist primarily of large chunks of gzipped
 * data interspersed with uncompressed data.  Doing a naive bsdiff of
 * these files is not useful because small changes in the data lead to
 * large changes in the compressed bitstream; bsdiff patches of gzipped
 * data are typically as large as the data itself.
 *
 * To patch these usefully, we break the source and target images up into
 * chunks of two types: "normal" and "gzip".  Normal chunks are simply
 * patched using a plain bsdiff.  Gzip chunks are first expanded, then a
 * bsdiff is applied to the uncompressed data, then the patched data is
 * gzipped using the same encoder parameters.  Patched chunks are
 * concatenated together to create the output file; the output image
 * should be *exactly* the same series of bytes as the target image used
 * originally to generate the patch.
 *
 * To work well with this tool, the gzipped sections of the target
 * image must have been generated using the same deflate encoder that
 * is available in applypatch, namely, the one in the zlib library.
 * In practice this means that images should be compressed using the
 * "minigzip" tool included in the zlib distribution, not the GNU gzip
 * program.
 *
 * An "imgdiff" patch consists of a header describing the chunk structure
 * of the file and any encoding parameters needed for the gzipped
 * chunks, followed by N bsdiff patches, one per chunk.
 *
 * For a diff to be generated, the source and target images must have the
 * same "chunk" structure: that is, the same number of gzipped and normal
 * chunks in the same order.  Android boot and recovery images currently
 * consist of five chunks:  a small normal header, a gzipped kernel, a
 * small normal section, a gzipped ramdisk, and finally a small normal
 * footer.
 *
 * Caveats:  we locate gzipped sections within the source and target
 * images by searching for the byte sequence 1f8b0800:  1f8b is the gzip
 * magic number; 08 specifies the "deflate" encoding [the only encoding
 * supported by the gzip standard]; and 00 is the flags byte.  We do not
 * currently support any extra header fields (which would be indicated by
 * a nonzero flags byte).  We also don't handle the case when that byte
 * sequence appears spuriously in the file.  (Note that it would have to
 * occur spuriously within a normal chunk to be a problem.)
 *
 *
 * The imgdiff patch header looks like this:
 *
 *    "IMGDIFF1"                  (8)   [magic number and version]
 *    chunk count                 (4)
 *    for each chunk:
 *        chunk type              (4)   [CHUNK_{NORMAL, GZIP, DEFLATE, RAW}]
 *        if chunk type == CHUNK_NORMAL:
 *           source start         (8)
 *           source len           (8)
 *           bsdiff patch offset  (8)   [from start of patch file]
 *        if chunk type == CHUNK_GZIP:      (version 1 only)
 *           source start         (8)
 *           source len           (8)
 *           bsdiff patch offset  (8)   [from start of patch file]
 *           source expanded len  (8)   [size of uncompressed source]
 *           target expected len  (8)   [size of uncompressed target]
 *           gzip level           (4)
 *                method          (4)
 *                windowBits      (4)
 *                memLevel        (4)
 *                strategy        (4)
 *           gzip header len      (4)
 *           gzip header          (gzip header len)
 *           gzip footer          (8)
 *        if chunk type == CHUNK_DEFLATE:   (version 2 only)
 *           source start         (8)
 *           source len           (8)
 *           bsdiff patch offset  (8)   [from start of patch file]
 *           source expanded len  (8)   [size of uncompressed source]
 *           target expected len  (8)   [size of uncompressed target]
 *           gzip level           (4)
 *                method          (4)
 *                windowBits      (4)
 *                memLevel        (4)
 *                strategy        (4)
 *        if chunk type == RAW:             (version 2 only)
 *           target len           (4)
 *           data                 (target len)
 *
 * All integers are little-endian.  "source start" and "source len"
 * specify the section of the input image that comprises this chunk,
 * including the gzip header and footer for gzip chunks.  "source
 * expanded len" is the size of the uncompressed source data.  "target
 * expected len" is the size of the uncompressed data after applying
 * the bsdiff patch.  The next five parameters specify the zlib
 * parameters to be used when compressing the patched data, and the
 * next three specify the header and footer to be wrapped around the
 * compressed data to create the output chunk (so that header contents
 * like the timestamp are recreated exactly).
 *
 * After the header there are 'chunk count' bsdiff patches; the offset
 * of each from the beginning of the file is specified in the header.
 */

#include <errno.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/stat.h>
#include <unistd.h>
#include <sys/types.h>

#include "zlib.h"
#include "imgdiff.h"
#include "utils.h"

typedef struct {
  int type;             // CHUNK_NORMAL, CHUNK_DEFLATE
  size_t start;         // offset of chunk in original image file

  size_t len;
  unsigned char* data;  // data to be patched (uncompressed, for deflate chunks)

  size_t source_start;
  size_t source_len;

  off_t* I;             // used by bsdiff

  // --- for CHUNK_DEFLATE chunks only: ---

  // original (compressed) deflate data
  size_t deflate_len;
  unsigned char* deflate_data;

  char* filename;       // used for zip entries

  // deflate encoder parameters
  int level, method, windowBits, memLevel, strategy;

  size_t source_uncompressed_len;
} ImageChunk;

typedef struct {
  int data_offset;
  int deflate_len;
  int uncomp_len;
  char* filename;
} ZipFileEntry;

static int fileentry_compare(const void* a, const void* b) {
  int ao = ((ZipFileEntry*)a)->data_offset;
  int bo = ((ZipFileEntry*)b)->data_offset;
  if (ao < bo) {
    return -1;
  } else if (ao > bo) {
    return 1;
  } else {
    return 0;
  }
}

// from bsdiff.c
int bsdiff(u_char* old, off_t oldsize, off_t** IP, u_char* new, off_t newsize,
           const char* patch_filename);

unsigned char* ReadZip(const char* filename,
                       int* num_chunks, ImageChunk** chunks,
                       int include_pseudo_chunk) {
  struct stat st;
  if (stat(filename, &st) != 0) {
    printf("failed to stat \"%s\": %s\n", filename, strerror(errno));
    return NULL;
  }

  unsigned char* img = malloc(st.st_size);
  FILE* f = fopen(filename, "rb");
  if (fread(img, 1, st.st_size, f) != st.st_size) {
    printf("failed to read \"%s\" %s\n", filename, strerror(errno));
    fclose(f);
    return NULL;
  }
  fclose(f);

  // look for the end-of-central-directory record.

  int i;
  for (i = st.st_size-20; i >= 0 && i > st.st_size - 65600; --i) {
    if (img[i] == 0x50 && img[i+1] == 0x4b &&
        img[i+2] == 0x05 && img[i+3] == 0x06) {
      break;
    }
  }
  // double-check: this archive consists of a single "disk"
  if (!(img[i+4] == 0 && img[i+5] == 0 && img[i+6] == 0 && img[i+7] == 0)) {
    printf("can't process multi-disk archive\n");
    return NULL;
  }

  int cdcount = Read2(img+i+8);
  int cdoffset = Read4(img+i+16);

  ZipFileEntry* temp_entries = malloc(cdcount * sizeof(ZipFileEntry));
  int entrycount = 0;

  unsigned char* cd = img+cdoffset;
  for (i = 0; i < cdcount; ++i) {
    if (!(cd[0] == 0x50 && cd[1] == 0x4b && cd[2] == 0x01 && cd[3] == 0x02)) {
      printf("bad central directory entry %d\n", i);
      return NULL;
    }

    int clen = Read4(cd+20);   // compressed len
    int ulen = Read4(cd+24);   // uncompressed len
    int nlen = Read2(cd+28);   // filename len
    int xlen = Read2(cd+30);   // extra field len
    int mlen = Read2(cd+32);   // file comment len
    int hoffset = Read4(cd+42);   // local header offset

    char* filename = malloc(nlen+1);
    memcpy(filename, cd+46, nlen);
    filename[nlen] = '\0';

    int method = Read2(cd+10);

    cd += 46 + nlen + xlen + mlen;

    if (method != 8) {  // 8 == deflate
      free(filename);
      continue;
    }

    unsigned char* lh = img + hoffset;

    if (!(lh[0] == 0x50 && lh[1] == 0x4b && lh[2] == 0x03 && lh[3] == 0x04)) {
      printf("bad local file header entry %d\n", i);
      return NULL;
    }

    if (Read2(lh+26) != nlen || memcmp(lh+30, filename, nlen) != 0) {
      printf("central dir filename doesn't match local header\n");
      return NULL;
    }

    xlen = Read2(lh+28);   // extra field len; might be different from CD entry?

    temp_entries[entrycount].data_offset = hoffset+30+nlen+xlen;
    temp_entries[entrycount].deflate_len = clen;
    temp_entries[entrycount].uncomp_len = ulen;
    temp_entries[entrycount].filename = filename;
    ++entrycount;
  }

  qsort(temp_entries, entrycount, sizeof(ZipFileEntry), fileentry_compare);

#if 0
  printf("found %d deflated entries\n", entrycount);
  for (i = 0; i < entrycount; ++i) {
    printf("off %10d  len %10d unlen %10d   %p %s\n",
           temp_entries[i].data_offset,
           temp_entries[i].deflate_len,
           temp_entries[i].uncomp_len,
           temp_entries[i].filename,
           temp_entries[i].filename);
  }
#endif

  *num_chunks = 0;
  *chunks = malloc((entrycount*2+2) * sizeof(ImageChunk));
  ImageChunk* curr = *chunks;

  if (include_pseudo_chunk) {
    curr->type = CHUNK_NORMAL;
    curr->start = 0;
    curr->len = st.st_size;
    curr->data = img;
    curr->filename = NULL;
    curr->I = NULL;
    ++curr;
    ++*num_chunks;
  }

  int pos = 0;
  int nextentry = 0;

  while (pos < st.st_size) {
    if (nextentry < entrycount && pos == temp_entries[nextentry].data_offset) {
      curr->type = CHUNK_DEFLATE;
      curr->start = pos;
      curr->deflate_len = temp_entries[nextentry].deflate_len;
      curr->deflate_data = img + pos;
      curr->filename = temp_entries[nextentry].filename;
      curr->I = NULL;

      curr->len = temp_entries[nextentry].uncomp_len;
      curr->data = malloc(curr->len);

      z_stream strm;
      strm.zalloc = Z_NULL;
      strm.zfree = Z_NULL;
      strm.opaque = Z_NULL;
      strm.avail_in = curr->deflate_len;
      strm.next_in = curr->deflate_data;

      // -15 means we are decoding a 'raw' deflate stream; zlib will
      // not expect zlib headers.
      int ret = inflateInit2(&strm, -15);

      strm.avail_out = curr->len;
      strm.next_out = curr->data;
      ret = inflate(&strm, Z_NO_FLUSH);
      if (ret != Z_STREAM_END) {
        printf("failed to inflate \"%s\"; %d\n", curr->filename, ret);
        return NULL;
      }

      inflateEnd(&strm);

      pos += curr->deflate_len;
      ++nextentry;
      ++*num_chunks;
      ++curr;
      continue;
    }

    // use a normal chunk to take all the data up to the start of the
    // next deflate section.

    curr->type = CHUNK_NORMAL;
    curr->start = pos;
    if (nextentry < entrycount) {
      curr->len = temp_entries[nextentry].data_offset - pos;
    } else {
      curr->len = st.st_size - pos;
    }
    curr->data = img + pos;
    curr->filename = NULL;
    curr->I = NULL;
    pos += curr->len;

    ++*num_chunks;
    ++curr;
  }

  free(temp_entries);
  return img;
}

/*
 * Read the given file and break it up into chunks, putting the number
 * of chunks and their info in *num_chunks and **chunks,
 * respectively.  Returns a malloc'd block of memory containing the
 * contents of the file; various pointers in the output chunk array
 * will point into this block of memory.  The caller should free the
 * return value when done with all the chunks.  Returns NULL on
 * failure.
 */
unsigned char* ReadImage(const char* filename,
                         int* num_chunks, ImageChunk** chunks) {
  struct stat st;
  if (stat(filename, &st) != 0) {
    printf("failed to stat \"%s\": %s\n", filename, strerror(errno));
    return NULL;
  }

  unsigned char* img = malloc(st.st_size + 4);
  FILE* f = fopen(filename, "rb");
  if (fread(img, 1, st.st_size, f) != st.st_size) {
    printf("failed to read \"%s\" %s\n", filename, strerror(errno));
    fclose(f);
    return NULL;
  }
  fclose(f);

  // append 4 zero bytes to the data so we can always search for the
  // four-byte string 1f8b0800 starting at any point in the actual
  // file data, without special-casing the end of the data.
  memset(img+st.st_size, 0, 4);

  size_t pos = 0;

  *num_chunks = 0;
  *chunks = NULL;

  while (pos < st.st_size) {
    unsigned char* p = img+pos;

    if (st.st_size - pos >= 4 &&
        p[0] == 0x1f && p[1] == 0x8b &&
        p[2] == 0x08 &&    // deflate compression
        p[3] == 0x00) {    // no header flags
      // 'pos' is the offset of the start of a gzip chunk.

      *num_chunks += 3;
      *chunks = realloc(*chunks, *num_chunks * sizeof(ImageChunk));
      ImageChunk* curr = *chunks + (*num_chunks-3);

      // create a normal chunk for the header.
      curr->start = pos;
      curr->type = CHUNK_NORMAL;
      curr->len = GZIP_HEADER_LEN;
      curr->data = p;
      curr->I = NULL;

      pos += curr->len;
      p += curr->len;
      ++curr;

      curr->type = CHUNK_DEFLATE;
      curr->filename = NULL;
      curr->I = NULL;

      // We must decompress this chunk in order to discover where it
      // ends, and so we can put the uncompressed data and its length
      // into curr->data and curr->len.

      size_t allocated = 32768;
      curr->len = 0;
      curr->data = malloc(allocated);
      curr->start = pos;
      curr->deflate_data = p;

      z_stream strm;
      strm.zalloc = Z_NULL;
      strm.zfree = Z_NULL;
      strm.opaque = Z_NULL;
      strm.avail_in = st.st_size - pos;
      strm.next_in = p;

      // -15 means we are decoding a 'raw' deflate stream; zlib will
      // not expect zlib headers.
      int ret = inflateInit2(&strm, -15);

      do {
        strm.avail_out = allocated - curr->len;
        strm.next_out = curr->data + curr->len;
        ret = inflate(&strm, Z_NO_FLUSH);
        curr->len = allocated - strm.avail_out;
        if (strm.avail_out == 0) {
          allocated *= 2;
          curr->data = realloc(curr->data, allocated);
        }
      } while (ret != Z_STREAM_END);

      curr->deflate_len = st.st_size - strm.avail_in - pos;
      inflateEnd(&strm);
      pos += curr->deflate_len;
      p += curr->deflate_len;
      ++curr;

      // create a normal chunk for the footer

      curr->type = CHUNK_NORMAL;
      curr->start = pos;
      curr->len = GZIP_FOOTER_LEN;
      curr->data = img+pos;
      curr->I = NULL;

      pos += curr->len;
      p += curr->len;
      ++curr;

      // The footer (that we just skipped over) contains the size of
      // the uncompressed data.  Double-check to make sure that it
      // matches the size of the data we got when we actually did
      // the decompression.
      size_t footer_size = Read4(p-4);
      if (footer_size != curr[-2].len) {
        printf("Error: footer size %d != decompressed size %d\n",
                footer_size, curr[-2].len);
        free(img);
        return NULL;
      }
    } else {
      // Reallocate the list for every chunk; we expect the number of
      // chunks to be small (5 for typical boot and recovery images).
      ++*num_chunks;
      *chunks = realloc(*chunks, *num_chunks * sizeof(ImageChunk));
      ImageChunk* curr = *chunks + (*num_chunks-1);
      curr->start = pos;
      curr->I = NULL;

      // 'pos' is not the offset of the start of a gzip chunk, so scan
      // forward until we find a gzip header.
      curr->type = CHUNK_NORMAL;
      curr->data = p;

      for (curr->len = 0; curr->len < (st.st_size - pos); ++curr->len) {
        if (p[curr->len] == 0x1f &&
            p[curr->len+1] == 0x8b &&
            p[curr->len+2] == 0x08 &&
            p[curr->len+3] == 0x00) {
          break;
        }
      }
      pos += curr->len;
    }
  }

  return img;
}

#define BUFFER_SIZE 32768

/*
 * Takes the uncompressed data stored in the chunk, compresses it
 * using the zlib parameters stored in the chunk, and checks that it
 * matches exactly the compressed data we started with (also stored in
 * the chunk).  Return 0 on success.
 */
int TryReconstruction(ImageChunk* chunk, unsigned char* out) {
  size_t p = 0;

#if 0
  printf("trying %d %d %d %d %d\n",
          chunk->level, chunk->method, chunk->windowBits,
          chunk->memLevel, chunk->strategy);
#endif

  z_stream strm;
  strm.zalloc = Z_NULL;
  strm.zfree = Z_NULL;
  strm.opaque = Z_NULL;
  strm.avail_in = chunk->len;
  strm.next_in = chunk->data;
  int ret;
  ret = deflateInit2(&strm, chunk->level, chunk->method, chunk->windowBits,
                     chunk->memLevel, chunk->strategy);
  do {
    strm.avail_out = BUFFER_SIZE;
    strm.next_out = out;
    ret = deflate(&strm, Z_FINISH);
    size_t have = BUFFER_SIZE - strm.avail_out;

    if (memcmp(out, chunk->deflate_data+p, have) != 0) {
      // mismatch; data isn't the same.
      deflateEnd(&strm);
      return -1;
    }
    p += have;
  } while (ret != Z_STREAM_END);
  deflateEnd(&strm);
  if (p != chunk->deflate_len) {
    // mismatch; ran out of data before we should have.
    return -1;
  }
  return 0;
}

/*
 * Verify that we can reproduce exactly the same compressed data that
 * we started with.  Sets the level, method, windowBits, memLevel, and
 * strategy fields in the chunk to the encoding parameters needed to
 * produce the right output.  Returns 0 on success.
 */
int ReconstructDeflateChunk(ImageChunk* chunk) {
  if (chunk->type != CHUNK_DEFLATE) {
    printf("attempt to reconstruct non-deflate chunk\n");
    return -1;
  }

  size_t p = 0;
  unsigned char* out = malloc(BUFFER_SIZE);

  // We only check two combinations of encoder parameters:  level 6
  // (the default) and level 9 (the maximum).
  for (chunk->level = 6; chunk->level <= 9; chunk->level += 3) {
    chunk->windowBits = -15;  // 32kb window; negative to indicate a raw stream.
    chunk->memLevel = 8;      // the default value.
    chunk->method = Z_DEFLATED;
    chunk->strategy = Z_DEFAULT_STRATEGY;

    if (TryReconstruction(chunk, out) == 0) {
      free(out);
      return 0;
    }
  }

  free(out);
  return -1;
}

/*
 * Given source and target chunks, compute a bsdiff patch between them
 * by running bsdiff in a subprocess.  Return the patch data, placing
 * its length in *size.  Return NULL on failure.  We expect the bsdiff
 * program to be in the path.
 */
unsigned char* MakePatch(ImageChunk* src, ImageChunk* tgt, size_t* size) {
  if (tgt->type == CHUNK_NORMAL) {
    if (tgt->len <= 160) {
      tgt->type = CHUNK_RAW;
      *size = tgt->len;
      return tgt->data;
    }
  }

  char ptemp[] = "/tmp/imgdiff-patch-XXXXXX";
  mkstemp(ptemp);

  int r = bsdiff(src->data, src->len, &(src->I), tgt->data, tgt->len, ptemp);
  if (r != 0) {
    printf("bsdiff() failed: %d\n", r);
    return NULL;
  }

  struct stat st;
  if (stat(ptemp, &st) != 0) {
    printf("failed to stat patch file %s: %s\n",
            ptemp, strerror(errno));
    return NULL;
  }

  unsigned char* data = malloc(st.st_size);

  if (tgt->type == CHUNK_NORMAL && tgt->len <= st.st_size) {
    unlink(ptemp);

    tgt->type = CHUNK_RAW;
    *size = tgt->len;
    return tgt->data;
  }

  *size = st.st_size;

  FILE* f = fopen(ptemp, "rb");
  if (f == NULL) {
    printf("failed to open patch %s: %s\n", ptemp, strerror(errno));
    return NULL;
  }
  if (fread(data, 1, st.st_size, f) != st.st_size) {
    printf("failed to read patch %s: %s\n", ptemp, strerror(errno));
    return NULL;
  }
  fclose(f);

  unlink(ptemp);

  tgt->source_start = src->start;
  switch (tgt->type) {
    case CHUNK_NORMAL:
      tgt->source_len = src->len;
      break;
    case CHUNK_DEFLATE:
      tgt->source_len = src->deflate_len;
      tgt->source_uncompressed_len = src->len;
      break;
  }

  return data;
}

/*
 * Cause a gzip chunk to be treated as a normal chunk (ie, as a blob
 * of uninterpreted data).  The resulting patch will likely be about
 * as big as the target file, but it lets us handle the case of images
 * where some gzip chunks are reconstructible but others aren't (by
 * treating the ones that aren't as normal chunks).
 */
void ChangeDeflateChunkToNormal(ImageChunk* ch) {
  if (ch->type != CHUNK_DEFLATE) return;
  ch->type = CHUNK_NORMAL;
  free(ch->data);
  ch->data = ch->deflate_data;
  ch->len = ch->deflate_len;
}

/*
 * Return true if the data in the chunk is identical (including the
 * compressed representation, for gzip chunks).
 */
int AreChunksEqual(ImageChunk* a, ImageChunk* b) {
    if (a->type != b->type) return 0;

    switch (a->type) {
        case CHUNK_NORMAL:
            return a->len == b->len && memcmp(a->data, b->data, a->len) == 0;

        case CHUNK_DEFLATE:
            return a->deflate_len == b->deflate_len &&
                memcmp(a->deflate_data, b->deflate_data, a->deflate_len) == 0;

        default:
            printf("unknown chunk type %d\n", a->type);
            return 0;
    }
}

/*
 * Look for runs of adjacent normal chunks and compress them down into
 * a single chunk.  (Such runs can be produced when deflate chunks are
 * changed to normal chunks.)
 */
void MergeAdjacentNormalChunks(ImageChunk* chunks, int* num_chunks) {
  int out = 0;
  int in_start = 0, in_end;
  while (in_start < *num_chunks) {
    if (chunks[in_start].type != CHUNK_NORMAL) {
      in_end = in_start+1;
    } else {
      // in_start is a normal chunk.  Look for a run of normal chunks
      // that constitute a solid block of data (ie, each chunk begins
      // where the previous one ended).
      for (in_end = in_start+1;
           in_end < *num_chunks && chunks[in_end].type == CHUNK_NORMAL &&
             (chunks[in_end].start ==
              chunks[in_end-1].start + chunks[in_end-1].len &&
              chunks[in_end].data ==
              chunks[in_end-1].data + chunks[in_end-1].len);
           ++in_end);
    }

    if (in_end == in_start+1) {
#if 0
      printf("chunk %d is now %d\n", in_start, out);
#endif
      if (out != in_start) {
        memcpy(chunks+out, chunks+in_start, sizeof(ImageChunk));
      }
    } else {
#if 0
      printf("collapse normal chunks %d-%d into %d\n", in_start, in_end-1, out);
#endif

      // Merge chunks [in_start, in_end-1] into one chunk.  Since the
      // data member of each chunk is just a pointer into an in-memory
      // copy of the file, this can be done without recopying (the
      // output chunk has the first chunk's start location and data
      // pointer, and length equal to the sum of the input chunk
      // lengths).
      chunks[out].type = CHUNK_NORMAL;
      chunks[out].start = chunks[in_start].start;
      chunks[out].data = chunks[in_start].data;
      chunks[out].len = chunks[in_end-1].len +
        (chunks[in_end-1].start - chunks[in_start].start);
    }

    ++out;
    in_start = in_end;
  }
  *num_chunks = out;
}

ImageChunk* FindChunkByName(const char* name,
                            ImageChunk* chunks, int num_chunks) {
  int i;
  for (i = 0; i < num_chunks; ++i) {
    if (chunks[i].type == CHUNK_DEFLATE && chunks[i].filename &&
        strcmp(name, chunks[i].filename) == 0) {
      return chunks+i;
    }
  }
  return NULL;
}

void DumpChunks(ImageChunk* chunks, int num_chunks) {
    int i;
    for (i = 0; i < num_chunks; ++i) {
        printf("chunk %d: type %d start %d len %d\n",
               i, chunks[i].type, chunks[i].start, chunks[i].len);
    }
}

int main(int argc, char** argv) {
  if (argc != 4 && argc != 5) {
    usage:
    printf("usage: %s [-z] <src-img> <tgt-img> <patch-file>\n",
            argv[0]);
    return 2;
  }

  int zip_mode = 0;

  if (strcmp(argv[1], "-z") == 0) {
    zip_mode = 1;
    --argc;
    ++argv;
  }


  int num_src_chunks;
  ImageChunk* src_chunks;
  int num_tgt_chunks;
  ImageChunk* tgt_chunks;
  int i;

  if (zip_mode) {
    if (ReadZip(argv[1], &num_src_chunks, &src_chunks, 1) == NULL) {
      printf("failed to break apart source zip file\n");
      return 1;
    }
    if (ReadZip(argv[2], &num_tgt_chunks, &tgt_chunks, 0) == NULL) {
      printf("failed to break apart target zip file\n");
      return 1;
    }
  } else {
    if (ReadImage(argv[1], &num_src_chunks, &src_chunks) == NULL) {
      printf("failed to break apart source image\n");
      return 1;
    }
    if (ReadImage(argv[2], &num_tgt_chunks, &tgt_chunks) == NULL) {
      printf("failed to break apart target image\n");
      return 1;
    }

    // Verify that the source and target images have the same chunk
    // structure (ie, the same sequence of deflate and normal chunks).

    if (!zip_mode) {
        // Merge the gzip header and footer in with any adjacent
        // normal chunks.
        MergeAdjacentNormalChunks(tgt_chunks, &num_tgt_chunks);
        MergeAdjacentNormalChunks(src_chunks, &num_src_chunks);
    }

    if (num_src_chunks != num_tgt_chunks) {
      printf("source and target don't have same number of chunks!\n");
      printf("source chunks:\n");
      DumpChunks(src_chunks, num_src_chunks);
      printf("target chunks:\n");
      DumpChunks(tgt_chunks, num_tgt_chunks);
      return 1;
    }
    for (i = 0; i < num_src_chunks; ++i) {
      if (src_chunks[i].type != tgt_chunks[i].type) {
        printf("source and target don't have same chunk "
                "structure! (chunk %d)\n", i);
        printf("source chunks:\n");
        DumpChunks(src_chunks, num_src_chunks);
        printf("target chunks:\n");
        DumpChunks(tgt_chunks, num_tgt_chunks);
        return 1;
      }
    }
  }

  for (i = 0; i < num_tgt_chunks; ++i) {
    if (tgt_chunks[i].type == CHUNK_DEFLATE) {
      // Confirm that given the uncompressed chunk data in the target, we
      // can recompress it and get exactly the same bits as are in the
      // input target image.  If this fails, treat the chunk as a normal
      // non-deflated chunk.
      if (ReconstructDeflateChunk(tgt_chunks+i) < 0) {
        printf("failed to reconstruct target deflate chunk %d [%s]; "
               "treating as normal\n", i, tgt_chunks[i].filename);
        ChangeDeflateChunkToNormal(tgt_chunks+i);
        if (zip_mode) {
          ImageChunk* src = FindChunkByName(tgt_chunks[i].filename, src_chunks, num_src_chunks);
          if (src) {
            ChangeDeflateChunkToNormal(src);
          }
        } else {
          ChangeDeflateChunkToNormal(src_chunks+i);
        }
        continue;
      }

      // If two deflate chunks are identical (eg, the kernel has not
      // changed between two builds), treat them as normal chunks.
      // This makes applypatch much faster -- it can apply a trivial
      // patch to the compressed data, rather than uncompressing and
      // recompressing to apply the trivial patch to the uncompressed
      // data.
      ImageChunk* src;
      if (zip_mode) {
        src = FindChunkByName(tgt_chunks[i].filename, src_chunks, num_src_chunks);
      } else {
        src = src_chunks+i;
      }

      if (src == NULL || AreChunksEqual(tgt_chunks+i, src)) {
        ChangeDeflateChunkToNormal(tgt_chunks+i);
        if (src) {
          ChangeDeflateChunkToNormal(src);
        }
      }
    }
  }

  // Merging neighboring normal chunks.
  if (zip_mode) {
    // For zips, we only need to do this to the target:  deflated
    // chunks are matched via filename, and normal chunks are patched
    // using the entire source file as the source.
    MergeAdjacentNormalChunks(tgt_chunks, &num_tgt_chunks);
  } else {
    // For images, we need to maintain the parallel structure of the
    // chunk lists, so do the merging in both the source and target
    // lists.
    MergeAdjacentNormalChunks(tgt_chunks, &num_tgt_chunks);
    MergeAdjacentNormalChunks(src_chunks, &num_src_chunks);
    if (num_src_chunks != num_tgt_chunks) {
      // This shouldn't happen.
      printf("merging normal chunks went awry\n");
      return 1;
    }
  }

  // Compute bsdiff patches for each chunk's data (the uncompressed
  // data, in the case of deflate chunks).

  printf("Construct patches for %d chunks...\n", num_tgt_chunks);
  unsigned char** patch_data = malloc(num_tgt_chunks * sizeof(unsigned char*));
  size_t* patch_size = malloc(num_tgt_chunks * sizeof(size_t));
  for (i = 0; i < num_tgt_chunks; ++i) {
    if (zip_mode) {
      ImageChunk* src;
      if (tgt_chunks[i].type == CHUNK_DEFLATE &&
          (src = FindChunkByName(tgt_chunks[i].filename, src_chunks,
                                 num_src_chunks))) {
        patch_data[i] = MakePatch(src, tgt_chunks+i, patch_size+i);
      } else {
        patch_data[i] = MakePatch(src_chunks, tgt_chunks+i, patch_size+i);
      }
    } else {
      patch_data[i] = MakePatch(src_chunks+i, tgt_chunks+i, patch_size+i);
    }
    printf("patch %3d is %d bytes (of %d)\n",
           i, patch_size[i], tgt_chunks[i].source_len);
  }

  // Figure out how big the imgdiff file header is going to be, so
  // that we can correctly compute the offset of each bsdiff patch
  // within the file.

  size_t total_header_size = 12;
  for (i = 0; i < num_tgt_chunks; ++i) {
    total_header_size += 4;
    switch (tgt_chunks[i].type) {
      case CHUNK_NORMAL:
        total_header_size += 8*3;
        break;
      case CHUNK_DEFLATE:
        total_header_size += 8*5 + 4*5;
        break;
      case CHUNK_RAW:
        total_header_size += 4 + patch_size[i];
        break;
    }
  }

  size_t offset = total_header_size;

  FILE* f = fopen(argv[3], "wb");

  // Write out the headers.

  fwrite("IMGDIFF2", 1, 8, f);
  Write4(num_tgt_chunks, f);
  for (i = 0; i < num_tgt_chunks; ++i) {
    Write4(tgt_chunks[i].type, f);

    switch (tgt_chunks[i].type) {
      case CHUNK_NORMAL:
        printf("chunk %3d: normal   (%10d, %10d)  %10d\n", i,
               tgt_chunks[i].start, tgt_chunks[i].len, patch_size[i]);
        Write8(tgt_chunks[i].source_start, f);
        Write8(tgt_chunks[i].source_len, f);
        Write8(offset, f);
        offset += patch_size[i];
        break;

      case CHUNK_DEFLATE:
        printf("chunk %3d: deflate  (%10d, %10d)  %10d  %s\n", i,
               tgt_chunks[i].start, tgt_chunks[i].deflate_len, patch_size[i],
               tgt_chunks[i].filename);
        Write8(tgt_chunks[i].source_start, f);
        Write8(tgt_chunks[i].source_len, f);
        Write8(offset, f);
        Write8(tgt_chunks[i].source_uncompressed_len, f);
        Write8(tgt_chunks[i].len, f);
        Write4(tgt_chunks[i].level, f);
        Write4(tgt_chunks[i].method, f);
        Write4(tgt_chunks[i].windowBits, f);
        Write4(tgt_chunks[i].memLevel, f);
        Write4(tgt_chunks[i].strategy, f);
        offset += patch_size[i];
        break;

      case CHUNK_RAW:
        printf("chunk %3d: raw      (%10d, %10d)\n", i,
               tgt_chunks[i].start, tgt_chunks[i].len);
        Write4(patch_size[i], f);
        fwrite(patch_data[i], 1, patch_size[i], f);
        break;
    }
  }

  // Append each chunk's bsdiff patch, in order.

  for (i = 0; i < num_tgt_chunks; ++i) {
    if (tgt_chunks[i].type != CHUNK_RAW) {
      fwrite(patch_data[i], 1, patch_size[i], f);
    }
  }

  fclose(f);

  return 0;
}