aboutsummaryrefslogtreecommitdiffstats
path: root/lualib/lua_content/pdf.lua
blob: d3d4b9d8522dabdad2050ddcec70413007015e76 (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
--[[
Copyright (c) 2019, Vsevolod Stakhov <vsevolod@highsecure.ru>

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.
]]--

--[[[
-- @module lua_content/pdf
-- This module contains some heuristics for PDF files
--]]

local rspamd_trie = require "rspamd_trie"
local rspamd_util = require "rspamd_util"
local rspamd_text = require "rspamd_text"
local bit = require "bit"
local N = "lua_content"
local lua_util = require "lua_util"
local rspamd_regexp = require "rspamd_regexp"
local lpeg = require "lpeg"
local pdf_patterns = {
  trailer = {
    patterns = {
      [[\ntrailer\r?\n]]
    }
  },
  javascript = {
    patterns = {
      [[\/JS(?:[\s/><])]],
      [[\/JavaScript(?:[\s/><])]],
    }
  },
  openaction = {
    patterns = {
      [[\/OpenAction(?:[\s/><])]],
      [[\/AA(?:[\s/><])]],
    }
  },
  suspicious = {
    patterns = {
      [[netsh\s]],
      [[echo\s]],
      [[\/[A-Za-z]*#\d\d(?:[#A-Za-z<>/\s])]], -- Hex encode obfuscation
    }
  },
  start_object = {
    patterns = {
      [=[[\r\n\0]\s*\d+\s+\d+\s+obj[\r\n]]=]
    }
  },
  end_object = {
    patterns = {
      [=[endobj[\r\n]]=]
    }
  },
  start_stream = {
    patterns = {
      [=[>\s*stream[\r\n]]=],
    }
  },
  end_stream = {
    patterns = {
      [=[endstream[\r\n]]=]
    }
  }
}

local pdf_text_patterns = {
  start = {
    patterns = {
      [[\sBT\s]]
    }
  },
  stop = {
    patterns = {
      [[\sET\b]]
    }
  }
}

-- index[n] ->
--  t[1] - pattern,
--  t[2] - key in patterns table,
--  t[3] - value in patterns table
--  t[4] - local pattern index
local pdf_indexes = {}
local pdf_text_indexes = {}

local pdf_trie
local pdf_text_trie

local exports = {}

local config = {
  max_extraction_size = 512 * 1024,
  max_processing_size = 32 * 1024,
  enabled = true,
}

-- Used to process patterns found in PDF
-- positions for functional processors should be a iter/table from trie matcher in form
---- [{n1, pat_idx1}, ... {nn, pat_idxn}] where
---- pat_idxn is pattern index and n1 ... nn are match positions
local processors = {}
-- PDF objects outer grammar in LPEG style (performing table captures)
local pdf_outer_grammar
local pdf_text_grammar

-- Used to match objects
local object_re = rspamd_regexp.create_cached([=[/(\d+)\s+(\d+)\s+obj\s*/]=])

local function config_module()
  local opts = rspamd_config:get_all_opt('lua_content')

  if opts and opts.pdf then
    config = lua_util.override_defaults(config, opts.pdf)
  end
end

local function compile_tries()
  local default_compile_flags = bit.bor(rspamd_trie.flags.re,
      rspamd_trie.flags.dot_all,
      rspamd_trie.flags.no_start)
  local function compile_pats(patterns, indexes, compile_flags)
    local strs = {}
    for what,data in pairs(patterns) do
      for i,pat in ipairs(data.patterns) do
        strs[#strs + 1] = pat
        indexes[#indexes + 1] = {what, data, pat, i}
      end
    end

    return rspamd_trie.create(strs, compile_flags or default_compile_flags)
  end

  if not pdf_trie then
    pdf_trie = compile_pats(pdf_patterns, pdf_indexes)
  end
  if not pdf_text_trie then
    pdf_text_trie = compile_pats(pdf_text_patterns, pdf_text_indexes)
  end
end

-- Returns a table with generic grammar elements for PDF
local function generic_grammar_elts()
  local P = lpeg.P
  local R = lpeg.R
  local S = lpeg.S
  local V = lpeg.V
  local C = lpeg.C
  local D = R'09' -- Digits

  local grammar_elts = {}

  -- Helper functions
  local function pdf_hexstring_unescape(s)
    local function ue(cc)
      return string.char(tonumber(cc, 16))
    end
    if #s % 2 == 0 then
      -- Sane hex string
      return s:gsub('..', ue)
    end

    -- WTF hex string
    -- Append '0' to it and unescape...
    return s:sub(1, #s - 1):gsub('..' , ue) .. (s:sub(#s) .. '0'):gsub('..' , ue)
  end

  local function pdf_string_unescape(s)
    -- TODO: add unescaping logic
    return s
  end

  local function pdf_id_unescape(s)
    return (s:gsub('#%d%d', function (cc)
      return string.char(tonumber(cc:sub(2), 16))
    end))
  end

  local delim = S'()<>[]{}/%'
  grammar_elts.ws = S'\0 \r\n\t\f'
  local hex = R'af' + R'AF' + D
  -- Comments.
  local eol = P'\r\n' + '\n'
  local line = (1 - S'\r\n\f')^0 * eol^-1
  grammar_elts.comment = P'%' * line

  -- Numbers.
  local sign = S'+-'^-1
  local decimal = D^1
  local float = D^1 * P'.' * D^0 + P'.' * D^1
  grammar_elts.number = C(sign * (float + decimal)) / tonumber

  -- String
  grammar_elts.str = P{ "(" * C(((1 - S"()") + V(1))^0) / pdf_string_unescape * ")" }
  grammar_elts.hexstr = P{"<" * C(hex^0) / pdf_hexstring_unescape * ">"}

  -- Identifier
  grammar_elts.id = P{'/' * C((1-(delim + grammar_elts.ws))^1) / pdf_id_unescape}

  -- Booleans (who care about them?)
  grammar_elts.boolean = C(P("true") + P("false"))

  -- Stupid references
  grammar_elts.ref = lpeg.Ct{lpeg.Cc("%REF%") * C(D^1) * " " * C(D^1) * " " * "R"}

  return grammar_elts
end


-- Generates a grammar to parse outer elements (external objects in PDF notation)
local function gen_outer_grammar()
  local V = lpeg.V
  local gen = generic_grammar_elts()

  return lpeg.P{
    "EXPR";
    EXPR = gen.ws^0 * V("ELT")^0 * gen.ws^0,
    ELT = V("ARRAY") + V("DICT") + V("ATOM"),
    ATOM = gen.ws^0 * (gen.comment + gen.boolean + gen.ref +
        gen.number + V("STRING") + gen.id) * gen.ws^0,
    DICT = "<<" * gen.ws^0  * lpeg.Cf(lpeg.Ct("") * V("KV_PAIR")^0, rawset) * gen.ws^0 * ">>",
    KV_PAIR = lpeg.Cg(gen.id * gen.ws^0 * V("ELT") * gen.ws^0),
    ARRAY = "[" * gen.ws^0 * lpeg.Ct(V("ELT")^0) * gen.ws^0 * "]",
    STRING = lpeg.P{gen.str + gen.hexstr},
  }
end

-- Graphic state in PDF
local function gen_graphics_unary()
  local P = lpeg.P
  local S = lpeg.S

  return P("q") + P("Q") + P("h")
      + S("WSsFfBb") * P("*")^0 + P("n")

end
local function gen_graphics_binary()
  local P = lpeg.P
  local S = lpeg.S

  return S("gGwJjMi") +
      P("M") + P("ri") + P("gs") +
      P("CS") + P("cs") + P("sh")
end
local function gen_graphics_ternary()
  local P = lpeg.P
  local S = lpeg.S

  return P("d") + P("m") + S("lm")
end
local function gen_graphics_nary()
  local P = lpeg.P
  local S = lpeg.S

  return P("SC") + P("sc") + P("SCN") + P("scn") + P("k") + P("K") + P("re") + S("cvy") +
      P("RG") + P("rg")
end

-- Generates a grammar to parse text blocks (between BT and ET)
local function gen_text_grammar()
  local V = lpeg.V
  local P = lpeg.P
  local C = lpeg.C
  local gen = generic_grammar_elts()

  local empty = ""
  local unary_ops = C("T*") / "\n" +
      C(gen_graphics_unary()) / empty
  local binary_ops = P("Tc") + P("Tw") + P("Tz") + P("TL") + P("Tr") + P("Ts") +
      gen_graphics_binary()
  local ternary_ops = P("TD") + P("Td") + gen_graphics_ternary()
  local nary_op = P("Tm") + gen_graphics_nary()
  local text_binary_op = P("Tj") + P("TJ") + P("'")
  local text_quote_op = P('"')
  local font_op = P("Tf")

  return lpeg.P{
    "EXPR";
    EXPR = gen.ws^0 * lpeg.Ct(V("COMMAND")^0),
    COMMAND = (V("UNARY") + V("BINARY") + V("TERNARY") + V("NARY") + V("TEXT") +
        V("FONT") + gen.comment) * gen.ws^0,
    UNARY = unary_ops,
    BINARY = V("ARG") / empty * gen.ws^1 * binary_ops,
    TERNARY = V("ARG") / empty * gen.ws^1 * V("ARG") / empty * gen.ws^1 * ternary_ops,
    NARY = (gen.number / 0 * gen.ws^1)^1 * (gen.id / empty * gen.ws^0)^-1 * nary_op,
    ARG = V("ARRAY") + V("DICT") + V("ATOM"),
    ATOM = (gen.comment + gen.boolean + gen.ref +
        gen.number + V("STRING") + gen.id),
    DICT = "<<" * gen.ws^0  * lpeg.Cf(lpeg.Ct("") * V("KV_PAIR")^0, rawset) * gen.ws^0 * ">>",
    KV_PAIR = lpeg.Cg(gen.id * gen.ws^0 * V("ARG") * gen.ws^0),
    ARRAY = "[" * gen.ws^0 * lpeg.Ct(V("ARG")^0) * gen.ws^0 * "]",
    STRING = lpeg.P{gen.str + gen.hexstr},
    TEXT = (V("TEXT_ARG") * gen.ws^1 * text_binary_op) +
        (V("ARG") / 0 * gen.ws^1 * V("ARG") / 0 * gen.ws^1 * V("TEXT_ARG") * gen.ws^1 * text_quote_op),
    FONT = (V("FONT_ARG") * gen.ws^1 * (gen.number / 0) * gen.ws^1 * font_op),
    FONT_ARG = lpeg.Ct(lpeg.Cc("%font%") * gen.id),
    TEXT_ARG = lpeg.Ct(V("STRING")) + V("TEXT_ARRAY"),
    TEXT_ARRAY = "[" *
        lpeg.Ct(((gen.ws^0 * (gen.ws^0 * (gen.number / 0)^0 * gen.ws^0 * (gen.str + gen.hexstr)))^1)) * gen.ws^0 * "]",
  }
end

-- Call immediately on require
compile_tries()
config_module()
pdf_outer_grammar = gen_outer_grammar()
pdf_text_grammar = gen_text_grammar()

local function extract_text_data(specific)
  return nil -- NYI
end

-- Generates index for major/minor pair
local function obj_ref(major, minor)
  return major * 10.0 + 1.0 / (minor + 1.0)
end

-- Return indirect object reference (if needed)
local function maybe_dereference_object(elt, pdf, task)
  if type(elt) == 'table' and elt[1] == '%REF%' then
    local ref = obj_ref(elt[2], elt[3])

    if pdf.ref[ref] then
      -- No recursion!
      return pdf.ref[ref]
    else
      lua_util.debugm(N, task, 'cannot dereference %s:%s -> %s',
          elt[2], elt[3], obj_ref(elt[2], elt[3]))
      return nil
    end
  end

  return elt
end

-- Enforced dereference
local function dereference_object(elt, pdf)
  if type(elt) == 'table' and elt[1] == '%REF%' then
    local ref = obj_ref(elt[2], elt[3])

    if pdf.ref[ref] then
      -- Not a dict but the object!
      return pdf.ref[ref]
    end
  end

  return nil
end

local function process_dict(task, pdf, obj, dict)
  if not obj.type and type(dict) == 'table' then
    if dict.Type and type(dict.Type) == 'string' then
      -- Common stuff
      obj.type = dict.Type
    else
      -- Fucking pdf, we need to guess a type (or ignore that crap)...
      lua_util.debugm(N, task, 'no explicit type for %s:%s',
          obj.major, obj.minor)
      if dict.Parent then
        -- Guess by parent
        local parent = dereference_object(dict.Parent, pdf)

        if parent and parent.type then
          if parent.type == 'Catalog' then
            obj.type = 'Pages'
          elseif parent.type == 'Pages' then
            obj.type = 'Page'
          end

          if obj.type then
            lua_util.debugm(N, task, 'guessed type for %s:%s (%s) from parent %s:%s (%s)',
                obj.major, obj.minor, obj.type, parent.major, parent.minor, parent.type)

          end
        end
      end
    end

    lua_util.debugm(N, task, 'process stream dictionary for object %s:%s -> %s',
        obj.major, obj.minor, obj.type)
    local contents = dict.Contents
    if contents and type(contents) == 'table' then
      if contents[1] == '%REF%' then
        -- Single reference
        contents = {contents}
      end
      obj.contents = {}

      for _,c in ipairs(contents) do
        local cobj = maybe_dereference_object(c, pdf, task)
        if cobj then
          obj.contents[#obj.contents + 1] = cobj
          cobj.parent = obj
          cobj.type = 'content'
        end
      end

      lua_util.debugm(N, task, 'found content objects for %s:%s -> %s',
          obj.major, obj.minor, #obj.contents)
    end

    local resources = dict.Resources
    if resources and type(resources) == 'table' then
      obj.resources = maybe_dereference_object(resources, pdf, task)
    else
      -- Fucking pdf: we need to inherit from parent
      resources = {}
      if dict.Parent then
        local parent = maybe_dereference_object(dict.Parent, pdf, task)

        if parent and type(parent) == 'table' and parent.dict then
          if parent.resources then
            lua_util.debugm(N, task, 'propagated resources from %s:%s to %s:%s',
                parent.major, parent.minor, obj.major, obj.minor)
            resources = parent.resources
          end
        end
      end

      obj.resources = resources
    end

    local fonts = obj.resources.Font

    if fonts and type(fonts) == 'table' then
      obj.fonts = {}
      for k,v in pairs(fonts) do
        obj.fonts[k] = maybe_dereference_object(v, pdf, task)

        if obj.fonts[k] then
          local font = obj.fonts[k]
          lua_util.debugm(N, task, 'found font for object %s:%s -> %s',
              obj.major, obj.minor, font)
        end
      end
    end

    lua_util.debugm(N, task, 'found resources for object %s:%s: %s',
        obj.major, obj.minor, obj.resources)

    if dict.Type == 'FontDescriptor' then

      lua_util.debugm(N, task, "obj %s:%s is a font descriptor",
         obj.major, obj.minor)

      local stream_ref
      if dict.FontFile then
        stream_ref = dereference_object(dict.FontFile, pdf)
      end
      if dict.FontFile2 then
        stream_ref = dereference_object(dict.FontFile2, pdf)
      end
      if dict.FontFile3 then
        stream_ref = dereference_object(dict.FontFile3, pdf)
      end

      if stream_ref then
        if not stream_ref.dict then
          stream_ref.dict = {}
        end
        stream_ref.dict.type = 'font_data'
        stream_ref.dict.ignore = true

        lua_util.debugm(N, task, "obj %s:%s is a font data stream",
            stream_ref.major, stream_ref.minor)
      end
    end
  end
end

-- PDF 1.5 ObjStmt
local function extract_pdf_compound_objects(task, pdf)
  for _,obj in ipairs(pdf.objects or {}) do
    if obj.stream and obj.dict and type(obj.dict) == 'table' then
      local t = obj.dict.Type
      if t and t == 'ObjStm' then
        -- We are in troubles sir...
        local nobjs = tonumber(maybe_dereference_object(obj.dict.N, pdf, task))
        local first = tonumber(maybe_dereference_object(obj.dict.First, pdf, task))

        if nobjs and first then
          local extend = maybe_dereference_object(obj.dict.Extends, pdf, task)
          lua_util.debugm(N, task, 'extract ObjStm with %s objects (%s first) %s extend',
              nobjs, first, extend)
        else
          lua_util.debugm(N, task, 'ObjStm object %s:%s has bad dict: %s',
              obj.major, obj.minor, obj.dict)
        end
      end
    end
  end
end

-- This function arranges starts and ends of all objects and process them into initial
-- set of objects
local function extract_outer_objects(task, input, pdf)
  local start_pos, end_pos = 1, 1
  local obj_count = 0
  while start_pos <= #pdf.start_objects and end_pos <= #pdf.end_objects do
    local first = pdf.start_objects[start_pos]
    local last = pdf.end_objects[end_pos]

    -- 7 is length of `endobj\n`
    if first + 7 < last then
      local len = last - first - 7

      -- Also get the starting span and try to match it versus obj re to get numbers
      local obj_line_potential = first - 32
      if obj_line_potential < 1 then obj_line_potential = 1 end

      if end_pos > 1 and pdf.end_objects[end_pos - 1] >= obj_line_potential then
        obj_line_potential = pdf.end_objects[end_pos - 1] + 1
      end

      local obj_line_span = input:span(obj_line_potential, first - obj_line_potential + 1)
      local matches = object_re:search(obj_line_span, true, true)

      if matches and matches[1] then
        local nobj = {
          start = first,
          len = len,
          data = input:span(first, len),
          major = tonumber(matches[1][2]),
          minor = tonumber(matches[1][3]),
        }
        pdf.objects[obj_count + 1] = nobj
        if nobj.major and nobj.minor then
          -- Add reference
          local ref = obj_ref(nobj.major, nobj.minor)
          nobj.ref = ref -- Our internal reference
          pdf.ref[ref] = nobj
        end
      end

      obj_count = obj_count + 1
      start_pos = start_pos + 1
      end_pos = end_pos + 1
    elseif first > last then
      end_pos = end_pos + 1
    else
      start_pos = start_pos + 1
      end_pos = end_pos + 1
    end
  end
end

local function parse_object_grammar(obj, task, pdf)
  -- Parse grammar
  local obj_dict_span
  if obj.stream then
    obj_dict_span = obj.data:span(1, obj.stream.start - obj.start)
  else
    obj_dict_span = obj.data
  end

  if obj_dict_span:len() < config.max_processing_size then
    local ret,obj_or_err = pcall(pdf_outer_grammar.match, pdf_outer_grammar, obj_dict_span)

    if ret then
      if obj.stream then
        obj.dict = obj_or_err
        lua_util.debugm(N, task, 'stream object %s:%s is parsed to: %s',
            obj.major, obj.minor, obj_or_err)
      else
        -- Direct object
        pdf.ref[obj_ref(obj.major, obj.minor)] = obj_or_err
        if type(obj_or_err) == 'table' then
          obj.dict = obj_or_err
        end
        obj.uncompressed = obj_or_err
        lua_util.debugm(N, task, 'direct object %s:%s is parsed to: %s',
            obj.major, obj.minor, obj_or_err)
      end
    else
      lua_util.debugm(N, task, 'object %s:%s cannot be parsed: %s',
          obj.major, obj.minor, obj_or_err)
    end
  else
    lua_util.debugm(N, task, 'object %s:%s cannot be parsed: too large %s',
        obj.major, obj.minor, obj_dict_span:len())
  end
end

-- This function attaches streams to objects and processes outer pdf grammar
local function attach_pdf_streams(task, input, pdf)
  if pdf.start_streams and pdf.end_streams then
    local start_pos, end_pos = 1, 1

    for _,obj in ipairs(pdf.objects) do
      while start_pos <= #pdf.start_streams and end_pos <= #pdf.end_streams do
        local first = pdf.start_streams[start_pos]
        local last = pdf.end_streams[end_pos]
        last = last - 10 -- Exclude endstream\n pattern
        lua_util.debugm(N, task, "start: %s, end: %s; obj: %s-%s",
            first, last, obj.start, obj.start + obj.len)
        if first > obj.start and last < obj.start + obj.len and last > first then
          -- In case if we have fake endstream :(
          while pdf.end_streams[end_pos + 1] and pdf.end_streams[end_pos + 1] < obj.start + obj.len do
            end_pos = end_pos + 1
            last = pdf.end_streams[end_pos]
          end
          -- Strip the first \n
          while first < last do
            local chr = input:at(first)
            if chr ~= 13 and chr ~= 10 then break end
            first = first + 1
          end
          local len = last - first
          obj.stream = {
            start = first,
            len = len,
            data = input:span(first, len)
          }
          start_pos = start_pos + 1
          end_pos = end_pos + 1
          break
        elseif first < obj.start then
          start_pos = start_pos + 1
        elseif last > obj.start + obj.len then
          -- Not this object
          break
        else
          start_pos = start_pos + 1
          end_pos = end_pos + 1
        end
      end
      if obj.stream then
        lua_util.debugm(N, task, 'found object %s:%s %s start %s len, %s stream start, %s stream length',
            obj.major, obj.minor, obj.start, obj.len, obj.stream.start, obj.stream.len)
      else
        lua_util.debugm(N, task, 'found object %s:%s %s start %s len, no stream',
            obj.major, obj.minor, obj.start, obj.len)
      end
    end
  end
end

-- Processes PDF objects: extracts streams, object numbers, process outer grammar,
-- augment object types
local function postprocess_pdf_objects(task, input, pdf)
  pdf.objects = {} -- objects table
  pdf.ref = {} -- references table
  extract_outer_objects(task, input, pdf)

  -- Now we have objects and we need to attach streams that are in bounds
  attach_pdf_streams(task, input, pdf)
  -- Parse grammar for outer objects
  for _,obj in ipairs(pdf.objects) do
    if obj.ref then
      parse_object_grammar(obj, task, pdf)
    end
  end
  extract_pdf_compound_objects(task, pdf)

  -- Now we might probably have all objects being processed
  for _,obj in ipairs(pdf.objects) do
    if obj.dict then
      -- Types processing
      process_dict(task, pdf, obj, obj.dict)
    end
  end
end

local function apply_pdf_filter(input, filt)
  if filt == 'FlateDecode' then
    return rspamd_util.inflate(input, config.max_extraction_size)
  end

  return nil
end

local function maybe_apply_filter(dict, data)
  local uncompressed = data

  if dict.Filter then
    local filt = dict.Filter
    if type(filt) == 'string' then
      filt = {filt}
    end

    for _,f in ipairs(filt) do
      uncompressed = apply_pdf_filter(uncompressed, f)

      if not uncompressed then break end
    end
  end

  return uncompressed
end

local function maybe_extract_object_stream(obj, pdf, task)
  local dict = obj.dict
  if dict.Filter and dict.Length then
    local len = math.min(obj.stream.len,
        tonumber(maybe_dereference_object(dict.Length, pdf, task)) or 0)
    local real_stream = obj.stream.data:span(1, len)

    local uncompressed = maybe_apply_filter(dict, real_stream)

    if uncompressed then
      obj.uncompressed = uncompressed
      lua_util.debugm(N, task, 'extracted object %s:%s: (%s -> %s)',
          obj.major, obj.minor, len, uncompressed:len())
    else
      lua_util.debugm(N, task, 'cannot extract object %s:%s; len = %s; filter = %s',
          obj.major, obj.minor, len, dict.Filter)
    end
  end
end

local function offsets_to_blocks(starts, ends, out)
  local start_pos, end_pos = 1, 1

  while start_pos <= #starts and end_pos <= #ends do
    local first = starts[start_pos]
    local last = ends[end_pos]

    if first < last then
      local len = last - first
      out[#out + 1] = {
        start = first,
        len = len,
      }
      start_pos = start_pos + 1
      end_pos = end_pos + 1
    elseif first > last then
      end_pos = end_pos + 1
    else
      -- Not ordered properly!
      break
    end
  end
end

local function search_text(task, pdf)
  for _,obj in ipairs(pdf.objects) do
    if obj.type == 'Page' and obj.contents then
      local text = {}
      for _,tobj in ipairs(obj.contents) do
        maybe_extract_object_stream(tobj, pdf, task)
        local matches = pdf_text_trie:match(tobj.uncompressed or '')
        if matches then
          local text_blocks = {}
          local starts = {}
          local ends = {}

          for npat,matched_positions in pairs(matches) do
            if npat == 1 then
              for _,pos in ipairs(matched_positions) do
                starts[#starts + 1] = pos
              end
            else
              for _,pos in ipairs(matched_positions) do
                ends[#ends + 1] = pos
              end
            end
          end

          offsets_to_blocks(starts, ends, text_blocks)
          for _,bl in ipairs(text_blocks) do
            if bl.len > 2 then
              -- To remove \s+ET\b pattern (it can leave trailing space or not but it doesn't matter)
              bl.len = bl.len - 2
            end

            bl.data = tobj.uncompressed:span(bl.start, bl.len)
            --lua_util.debugm(N, task, 'extracted text from object %s:%s: %s',
            --    tobj.major, tobj.minor, bl.data)

            if bl.len < config.max_processing_size then
              local ret,obj_or_err = pcall(pdf_text_grammar.match, pdf_text_grammar,
                  bl.data)

              if ret then
                text[#text + 1] = obj_or_err
                lua_util.debugm(N, task, 'attached %s from content object %s:%s to %s:%s',
                    obj_or_err, tobj.major, tobj.minor, obj.major, obj.minor)
              else
                lua_util.debugm(N, task, 'object %s:%s cannot be parsed: %s',
                    obj.major, obj.minor, obj_or_err)
              end

            end
          end
        end
      end

      -- Join all text data together
      if #text > 0 then
        obj.text = rspamd_text.fromtable(text)
        lua_util.debugm(N, task, 'object %s:%s is parsed to: %s',
            obj.major, obj.minor, obj.text)
      end
    end
  end
end

local function process_pdf(input, _, task)

  if not config.enabled then
    -- Skip processing
    return {}
  end

  local matches = pdf_trie:match(input)

  if matches then
    local pdf_output = {
      tag = 'pdf',
      extract_text = extract_text_data,
    }
    local grouped_processors = {}
    for npat,matched_positions in pairs(matches) do
      local index = pdf_indexes[npat]

      local proc_key,loc_npat = index[1], index[4]

      if not grouped_processors[proc_key] then
        grouped_processors[proc_key] = {
          processor_func = processors[proc_key],
          offsets = {},
        }
      end
      local proc = grouped_processors[proc_key]
      -- Fill offsets
      for _,pos in ipairs(matched_positions) do
        proc.offsets[#proc.offsets + 1] = {pos, loc_npat}
      end
    end

    for name,processor in pairs(grouped_processors) do
      -- Sort by offset
      lua_util.debugm(N, task, "pdf: process group %s with %s matches",
          name, #processor.offsets)
      table.sort(processor.offsets, function(e1, e2) return e1[1] < e2[1] end)
      processor.processor_func(input, task, processor.offsets, pdf_output)
    end

    pdf_output.flags = {}

    if pdf_output.start_objects and pdf_output.end_objects then
      -- Postprocess objects
      postprocess_pdf_objects(task, input, pdf_output)
      search_text(task, pdf_output)
    else
      pdf_output.flags.no_objects = true
    end

    return pdf_output
  end
end

-- Processes the PDF trailer
processors.trailer = function(input, task, positions, output)
  local last_pos = positions[#positions]

  local last_span = input:span(last_pos[1])
  for line in last_span:lines(true) do
    if line:find('/Encrypt ') then
      lua_util.debugm(N, task, "pdf: found encrypted line in trailer: %s",
          line)
      output.encrypted = true
    end
  end
end

processors.javascript = function(_, task, _, output)
  lua_util.debugm(N, task, "pdf: found javascript tag")
  output.javascript = true
end

processors.openaction = function(_, task, _, output)
  lua_util.debugm(N, task, "pdf: found openaction tag")
  output.openaction = true
end

processors.suspicious = function(_, task, _, output)
  lua_util.debugm(N, task, "pdf: found a suspicious pattern")
  output.suspicious = true
end

local function generic_table_inserter(positions, output, output_key)
  if not output[output_key] then
    output[output_key] = {}
  end
  local shift = #output[output_key]
  for i,pos in ipairs(positions) do
    output[output_key][i + shift] = pos[1]
  end
end

processors.start_object = function(_, task, positions, output)
  generic_table_inserter(positions, output, 'start_objects')
end

processors.end_object = function(_, task, positions, output)
  generic_table_inserter(positions, output, 'end_objects')
end

processors.start_stream = function(_, task, positions, output)
  generic_table_inserter(positions, output, 'start_streams')
end

processors.end_stream = function(_, task, positions, output)
  generic_table_inserter(positions, output, 'end_streams')
end

exports.process = process_pdf

return exports