malmal 0.1.0

Upload scanner for images, SVG, PDF and PostScript files
Documentation
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
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
//! PDF analysis. The cross-reference table is deliberately ignored: it is the
//! first thing a malicious file lies about. Instead every `N G obj` in the
//! file is parsed where it stands, object streams are unpacked, and every
//! dictionary found either way is inspected for active content.

pub(crate) mod filters;
pub(crate) mod parser;

use std::borrow::Cow;
use std::collections::HashMap;

use memchr::memmem;

use crate::limits::{Budget, LimitExceeded};
use crate::report::{Finding, Severity};
use filters::{DecodeError, Params};
use parser::{Dict, Name, Parser, Value, is_regular, is_whitespace};

/// Callback that scans a file extracted from the PDF.
pub(crate) type NestedScan<'a> = dyn FnMut(&[u8], &mut Budget, &mut Vec<Finding>) + 'a;

/// Parsed values kept in memory for one document.
const MAX_NODES: usize = 2_000_000;
const MAX_FILTERS: usize = 8;
/// How far past an unparsable object's start to look for names.
const FALLBACK_WINDOW: usize = 64 << 10;

/// Names whose `#xx`-escaped spelling is reported as obfuscation.
const SENSITIVE_NAMES: &[&[u8]] = &[
    b"JS",
    b"JavaScript",
    b"OpenAction",
    b"AA",
    b"Launch",
    b"EmbeddedFile",
    b"EmbeddedFiles",
    b"EF",
    b"RichMedia",
    b"XFA",
    b"Encrypt",
    b"ObjStm",
    b"SubmitForm",
    b"ImportData",
    b"GoToR",
    b"GoToE",
    b"URI",
    b"S",
    b"Type",
    b"Subtype",
    b"Filter",
    b"Length",
    b"Names",
    b"AcroForm",
    b"FlateDecode",
    b"ASCIIHexDecode",
    b"ASCII85Decode",
    b"LZWDecode",
    b"RunLengthDecode",
];

static NULL: Value = Value::Null;

struct Object {
    number: u32,
    /// Where the object (or the object stream holding it) starts in the file.
    offset: usize,
    value: Value,
    /// First byte of stream data, for stream objects.
    stream_start: Option<usize>,
}

struct Document<'a> {
    data: &'a [u8],
    objects: Vec<Object>,
    /// Latest definition of each object number, as incremental updates intend.
    by_number: HashMap<u32, usize>,
}

impl<'a> Document<'a> {
    fn add(&mut self, object: Object) {
        self.by_number.insert(object.number, self.objects.len());
        self.objects.push(object);
    }

    /// Follow indirect references; missing values and targets resolve to null.
    fn resolve<'v>(&'v self, value: Option<&'v Value>) -> &'v Value {
        let mut value = value.unwrap_or(&NULL);
        for _ in 0..16 {
            match value {
                Value::Ref(number) => match self.by_number.get(number) {
                    Some(&index) => value = &self.objects[index].value,
                    None => return &NULL,
                },
                _ => return value,
            }
        }
        &NULL
    }

    fn name_is(&self, value: Option<&Value>, name: &[u8]) -> bool {
        self.resolve(value).as_name() == Some(name)
    }

    /// Raw bytes of a stream object: `/Length` when it checks out, otherwise
    /// everything up to the `endstream` keyword.
    fn stream_data(&self, object: &Object) -> Option<&'a [u8]> {
        let dict = object.value.as_dict()?;
        let rest = self.data.get(object.stream_start?..)?;
        if let Some(length) = self.resolve(dict.get(b"Length")).as_usize()
            && let Some(after) = rest.get(length..)
            && after.trim_ascii_start().starts_with(b"endstream")
        {
            return Some(&rest[..length]);
        }
        let body = &rest[..memmem::find(rest, b"endstream").unwrap_or(rest.len())];
        Some(
            body.strip_suffix(b"\r\n")
                .or_else(|| body.strip_suffix(b"\n"))
                .or_else(|| body.strip_suffix(b"\r"))
                .unwrap_or(body),
        )
    }

    /// Stream contents with the whole filter chain undone.
    fn decode(&self, object: &Object, budget: &mut Budget) -> Result<Vec<u8>, DecodeError> {
        let dict = object.value.as_dict().ok_or(DecodeError::Corrupt)?;
        let raw = self.stream_data(object).ok_or(DecodeError::Corrupt)?;
        let filters: Vec<&[u8]> = match self.resolve(dict.get(b"Filter")) {
            Value::Null => Vec::new(),
            Value::Name(name) => vec![name.text.as_slice()],
            Value::Array(items) => items
                .iter()
                .map(|item| {
                    self.resolve(Some(item))
                        .as_name()
                        .ok_or(DecodeError::Corrupt)
                })
                .collect::<Result<_, _>>()?,
            _ => return Err(DecodeError::Corrupt),
        };
        if filters.len() > MAX_FILTERS {
            return Err(DecodeError::Unsupported);
        }
        let parms = self.resolve(dict.get(b"DecodeParms"));

        let mut data = Cow::Borrowed(raw);
        for (index, filter) in filters.iter().enumerate() {
            let parm_dict = match parms {
                Value::Dict(dict) if index == 0 => Some(dict),
                Value::Array(items) => self.resolve(items.get(index)).as_dict(),
                _ => None,
            };
            let limit = usize::try_from(budget.remaining()).unwrap_or(usize::MAX);
            let decoded = filters::apply(filter, &data, &self.params(parm_dict), limit)?;
            budget
                .consume(decoded.len() as u64)
                .map_err(|_| DecodeError::Limit)?;
            data = Cow::Owned(decoded);
        }
        Ok(data.into_owned())
    }

    fn params(&self, dict: Option<&Dict>) -> Params {
        let defaults = Params::default();
        let Some(dict) = dict else {
            return defaults;
        };
        let get = |key: &[u8], default| self.resolve(dict.get(key)).as_usize().unwrap_or(default);
        Params {
            predictor: get(b"Predictor", defaults.predictor),
            columns: get(b"Columns", defaults.columns),
            colors: get(b"Colors", defaults.colors),
            bits_per_component: get(b"BitsPerComponent", defaults.bits_per_component),
            early_change: get(b"EarlyChange", 1) != 0,
        }
    }
}

struct Output<'a> {
    findings: &'a mut Vec<Finding>,
    /// Object numbers that file specifications name as embedded files.
    embedded: Vec<u32>,
    /// Object numbers that pages, fonts and patterns name as content streams.
    contents: Vec<u32>,
    encrypted: bool,
}

impl Output<'_> {
    /// Record a finding once per code and severity.
    fn report(&mut self, code: &'static str, severity: Severity, message: &str, offset: usize) {
        let seen = |f: &Finding| f.code == code && f.severity == severity;
        if !self.findings.iter().any(seen) {
            self.findings
                .push(Finding::new(code, severity, message).at(offset));
        }
    }

    fn malformed(&mut self, message: &str, offset: usize) {
        self.report("pdf.malformed", Severity::Medium, message, offset);
    }

    fn limit(&mut self, limit: LimitExceeded) {
        if !self.findings.iter().any(|f| f.code == limit.code()) {
            self.findings.push(limit.finding());
        }
    }

    fn encrypted(&mut self, offset: usize) {
        self.encrypted = true;
        self.report(
            "pdf.encrypted",
            Severity::Medium,
            "document is encrypted; strings, streams and object streams cannot be inspected",
            offset,
        );
    }
}

pub(crate) fn analyze(
    data: &[u8],
    budget: &mut Budget,
    findings: &mut Vec<Finding>,
    nested: &mut NestedScan<'_>,
) {
    let mut out = Output {
        findings,
        embedded: Vec::new(),
        contents: Vec::new(),
        encrypted: false,
    };
    let mut doc = Document {
        data,
        objects: Vec::new(),
        by_number: HashMap::new(),
    };
    let mut nodes_left = MAX_NODES;

    check_header(data, &mut out);
    discover_objects(&mut doc, &mut nodes_left, budget, &mut out);
    for at in memmem::find_iter(data, b"trailer") {
        let mut parser = Parser::new(data, at + b"trailer".len(), nodes_left);
        if let Some(trailer) = parser.parse_value() {
            walk(&doc, &trailer, at, &mut out);
        }
        nodes_left = parser.nodes_left;
    }
    for object in &doc.objects {
        walk(&doc, &object.value, object.offset, &mut out);
    }

    // In an encrypted file, object streams and embedded files are ciphertext.
    if !out.encrypted {
        let top_level = doc.objects.len();
        expand_object_streams(&mut doc, &mut nodes_left, budget, &mut out);
        for object in &doc.objects[top_level..] {
            walk(&doc, &object.value, object.offset, &mut out);
        }
        scan_embedded_files(&doc, budget, &mut out, nested);
        scan_content_streams(&doc, budget, &mut out);
    }

    if doc.objects.is_empty() {
        out.malformed("PDF contains no objects", 0);
    }
    check_end(data, &mut out);
}

fn check_header(data: &[u8], out: &mut Output<'_>) {
    let head = &data[..data.len().min(1024)];
    let Some(at) = memmem::find(head, b"%PDF-") else {
        return out.malformed("PDF header not found", 0);
    };
    let before = head[..at]
        .strip_prefix(b"\xEF\xBB\xBF")
        .unwrap_or(&head[..at]);
    if !before.iter().all(|&b| is_whitespace(b)) {
        out.malformed("data before the PDF header", 0);
    }
}

fn check_end(data: &[u8], out: &mut Output<'_>) {
    let mut markers = memmem::find_iter(data, b"%%EOF");
    let Some(first) = markers.next() else {
        return out.malformed("PDF has no %%EOF marker", data.len());
    };
    if markers.next().is_some() {
        out.report(
            "pdf.incremental_update",
            Severity::Info,
            "several %%EOF markers: the file was updated incrementally or is linearized",
            first,
        );
    }
}

/// Parse every `N G obj` in the file, wherever it is.
fn discover_objects(
    doc: &mut Document<'_>,
    nodes_left: &mut usize,
    budget: &Budget,
    out: &mut Output<'_>,
) {
    let data = doc.data;
    // Objects cannot legitimately overlap, so total parsing work is bounded
    // by a small multiple of the file size.
    let work_limit = data.len() as u64 * 8 + (1 << 20);
    let mut work = 0u64;
    for (count, at) in memmem::find_iter(data, b"obj").enumerate() {
        if count % 256 == 0
            && let Err(limit) = budget.check_time()
        {
            return out.limit(limit);
        }
        let body = at + b"obj".len();
        if data.get(body).is_some_and(|&b| is_regular(b)) {
            continue;
        }
        let Some((number, offset)) = object_header(data, at) else {
            continue;
        };

        let mut parser = Parser::new(data, body, *nodes_left);
        parser.skip_whitespace();
        let value = if parser.rest().starts_with(b"endobj") {
            Some(Value::Null)
        } else {
            parser.parse_value()
        };
        *nodes_left = parser.nodes_left;
        work += (parser.pos - at) as u64;
        if work > work_limit || *nodes_left == 0 {
            return out.limit(LimitExceeded::Complexity);
        }

        let Some(value) = value else {
            out.malformed("object could not be parsed", offset);
            fallback_names(&data[body..], offset, out);
            continue;
        };
        parser.skip_whitespace();
        let is_stream = value.as_dict().is_some() && parser.rest().starts_with(b"stream");
        let stream_start = is_stream.then(|| {
            let start = parser.pos + b"stream".len();
            match data.get(start..) {
                Some([b'\r', b'\n', ..]) => start + 2,
                Some([b'\r' | b'\n', ..]) => start + 1,
                _ => start,
            }
        });
        doc.add(Object {
            number,
            offset,
            value,
            stream_start,
        });
    }
}

/// Read `N G` backwards from the `obj` keyword at `obj_at`; returns the
/// object number and where the header starts.
fn object_header(data: &[u8], obj_at: usize) -> Option<(u32, usize)> {
    let field = |pos: &mut usize| -> Option<u64> {
        let after_space = *pos;
        while *pos > 0 && is_whitespace(data[*pos - 1]) {
            *pos -= 1;
        }
        if *pos == after_space {
            return None;
        }
        let end = *pos;
        while *pos > 0 && data[*pos - 1].is_ascii_digit() && end - *pos < 10 {
            *pos -= 1;
        }
        std::str::from_utf8(&data[*pos..end]).ok()?.parse().ok()
    };
    let mut pos = obj_at;
    let _generation = field(&mut pos)?;
    let number = field(&mut pos)?;
    Some((u32::try_from(number).ok()?, pos))
}

/// Unpack `/Type /ObjStm` streams and add the objects stored inside them.
fn expand_object_streams(
    doc: &mut Document<'_>,
    nodes_left: &mut usize,
    budget: &mut Budget,
    out: &mut Output<'_>,
) {
    let mut unpacked = Vec::new();
    for object in &doc.objects {
        let Some(dict) = object.value.as_dict() else {
            continue;
        };
        if object.stream_start.is_none() || !doc.name_is(dict.get(b"Type"), b"ObjStm") {
            continue;
        }
        let offset = object.offset;
        let decoded = match doc.decode(object, budget) {
            Ok(decoded) => decoded,
            Err(DecodeError::Limit) => {
                out.limit(LimitExceeded::Decompression);
                continue;
            }
            // Whatever is inside stays unseen, so the file cannot be cleared.
            Err(_) => {
                out.malformed("object stream cannot be decoded", offset);
                continue;
            }
        };
        let count = doc.resolve(dict.get(b"N")).as_usize();
        let first = doc.resolve(dict.get(b"First")).as_usize();
        let (Some(count), Some(first)) = (count, first) else {
            out.malformed("object stream has no index", offset);
            continue;
        };

        // The index is `count` pairs of object number and relative offset.
        let mut index = Parser::new(&decoded, 0, usize::MAX);
        let mut entries = Vec::new();
        for _ in 0..count.min(decoded.len()) {
            let (Some(number), Some(relative)) = (index.integer(), index.integer()) else {
                break;
            };
            entries.push((number, first.saturating_add(relative)));
        }
        for (number, start) in entries {
            let mut parser = Parser::new(&decoded, start, *nodes_left);
            let value = parser.parse_value();
            *nodes_left = parser.nodes_left;
            if *nodes_left == 0 {
                return out.limit(LimitExceeded::Complexity);
            }
            match (value, u32::try_from(number)) {
                (Some(value), Ok(number)) => unpacked.push(Object {
                    number,
                    offset,
                    value,
                    stream_start: None,
                }),
                _ => {
                    out.malformed("object inside an object stream could not be parsed", offset);
                    fallback_names(decoded.get(start..).unwrap_or(&[]), offset, out);
                }
            }
        }
    }
    for object in unpacked {
        doc.add(object);
    }
}

/// Inspect every dictionary reachable inside one object.
fn walk(doc: &Document<'_>, value: &Value, offset: usize, out: &mut Output<'_>) {
    match value {
        Value::Dict(dict) => {
            check_postscript(doc, dict, offset, out);
            for (key, item) in &dict.0 {
                check_name(key, offset, out);
                check_entry(doc, key, item, offset, out);
                walk(doc, item, offset, out);
            }
        }
        Value::Array(items) => {
            for item in items {
                walk(doc, item, offset, out);
            }
        }
        Value::Name(name) => check_name(name, offset, out),
        _ => {}
    }
}

fn check_name(name: &Name, offset: usize, out: &mut Output<'_>) {
    if name.obfuscated && SENSITIVE_NAMES.contains(&name.text.as_slice()) {
        out.report(
            "pdf.obfuscated_name",
            Severity::High,
            "keyword disguised with hex escapes",
            offset,
        );
    }
}

fn check_entry(doc: &Document<'_>, key: &Name, value: &Value, offset: usize, out: &mut Output<'_>) {
    match key.text.as_slice() {
        b"JS" | b"JavaScript" => javascript(offset, out),
        b"OpenAction" => match classify_action(doc, value) {
            Action::Harmless => {}
            Action::Other => out.report(
                "pdf.open_action",
                Severity::Medium,
                "performs an action when opened",
                offset,
            ),
            Action::Code => out.report(
                "pdf.open_action",
                Severity::High,
                "runs a script or program when opened",
                offset,
            ),
        },
        b"AA" => {
            let worst = match doc.resolve(Some(value)) {
                Value::Dict(actions) => actions
                    .0
                    .iter()
                    .map(|(_, action)| classify_action(doc, action))
                    .max()
                    .unwrap_or(Action::Harmless),
                _ => Action::Other,
            };
            match worst {
                Action::Harmless => {}
                Action::Other => out.report(
                    "pdf.additional_action",
                    Severity::Medium,
                    "performs an action on a document or page event",
                    offset,
                ),
                Action::Code => out.report(
                    "pdf.additional_action",
                    Severity::High,
                    "runs a script or program on a document or page event",
                    offset,
                ),
            }
        }
        b"EF" => {
            embedded_file(offset, out);
            if let Value::Dict(files) = doc.resolve(Some(value)) {
                out.embedded
                    .extend(files.0.iter().filter_map(|(_, file)| match file {
                        Value::Ref(number) => Some(*number),
                        _ => None,
                    }));
            }
        }
        b"EmbeddedFiles" => embedded_file(offset, out),
        // A page's content, or the glyph procedures of a Type 3 font. An
        // annotation's /Contents is a string and names nothing.
        b"Contents" => match (value, doc.resolve(Some(value))) {
            (_, Value::Array(items)) => out.contents.extend(items.iter().filter_map(as_ref)),
            (Value::Ref(number), _) => out.contents.push(*number),
            _ => {}
        },
        b"CharProcs" => {
            if let Value::Dict(procedures) = doc.resolve(Some(value)) {
                let streams = procedures.0.iter().filter_map(|(_, item)| as_ref(item));
                out.contents.extend(streams);
            }
        }
        b"XFA" => xfa(offset, out),
        b"Encrypt" => out.encrypted(offset),
        b"URI" => uri(offset, out),
        b"S" => match doc.resolve(Some(value)).as_name() {
            Some(b"JavaScript") => javascript(offset, out),
            Some(b"Launch") => launch(offset, out),
            Some(b"SubmitForm" | b"ImportData") => out.report(
                "pdf.submit_form",
                Severity::Low,
                "form action exchanges data with a remote server",
                offset,
            ),
            Some(b"GoToR" | b"GoToE") => out.report(
                "pdf.remote_goto",
                Severity::Low,
                "action opens another document",
                offset,
            ),
            Some(b"URI") => uri(offset, out),
            _ => {}
        },
        b"Type" if doc.name_is(Some(value), b"EmbeddedFile") => embedded_file(offset, out),
        b"Subtype" => {
            if let Some(b"RichMedia" | b"3D") = doc.resolve(Some(value)).as_name() {
                rich_media(offset, out);
            }
        }
        _ => {}
    }
}

fn as_ref(value: &Value) -> Option<u32> {
    match value {
        Value::Ref(number) => Some(*number),
        _ => None,
    }
}

/// PostScript XObjects are passed to the interpreter of a PostScript
/// printer or converter as they are. They come as `/Subtype /PS`, or as a
/// form XObject marked with `/Subtype2 /PS` or carrying a `/PS` stream.
fn check_postscript(doc: &Document<'_>, dict: &Dict, offset: usize, out: &mut Output<'_>) {
    let is = |key: &[u8], name: &[u8]| doc.name_is(dict.get(key), name);
    let form_with_alternative =
        dict.get(b"PS").is_some() && (is(b"Subtype", b"Form") || is(b"Type", b"XObject"));
    if is(b"Subtype", b"PS") || is(b"Subtype2", b"PS") || form_with_alternative {
        // The stream is scanned with the embedded files.
        out.embedded.extend(dict.get(b"PS").and_then(as_ref));
        postscript("embeds a PostScript XObject", offset, out);
    }
}

fn postscript(message: &str, offset: usize, out: &mut Output<'_>) {
    out.report("pdf.postscript", Severity::High, message, offset);
}

fn javascript(offset: usize, out: &mut Output<'_>) {
    out.report(
        "pdf.javascript",
        Severity::High,
        "document contains JavaScript",
        offset,
    );
}

fn launch(offset: usize, out: &mut Output<'_>) {
    out.report(
        "pdf.launch",
        Severity::High,
        "action launches an external program or file",
        offset,
    );
}

fn embedded_file(offset: usize, out: &mut Output<'_>) {
    out.report(
        "pdf.embedded_file",
        Severity::Medium,
        "document carries embedded files",
        offset,
    );
}

fn rich_media(offset: usize, out: &mut Output<'_>) {
    out.report(
        "pdf.rich_media",
        Severity::Medium,
        "document embeds Flash, video or 3D content",
        offset,
    );
}

fn xfa(offset: usize, out: &mut Output<'_>) {
    out.report(
        "pdf.xfa",
        Severity::Medium,
        "document contains an XFA form",
        offset,
    );
}

fn uri(offset: usize, out: &mut Output<'_>) {
    out.report(
        "pdf.uri",
        Severity::Info,
        "document links to external URIs",
        offset,
    );
}

#[derive(Clone, Copy, PartialEq, Eq, PartialOrd, Ord)]
enum Action {
    /// A destination or a jump within the document.
    Harmless,
    Other,
    /// JavaScript or a program launch.
    Code,
}

fn classify_action(doc: &Document<'_>, value: &Value) -> Action {
    match doc.resolve(Some(value)) {
        // Explicit and named destinations.
        Value::Array(_) | Value::Name(_) | Value::String => Action::Harmless,
        Value::Dict(action) => match doc.resolve(action.get(b"S")).as_name() {
            Some(b"JavaScript" | b"Launch") => Action::Code,
            Some(b"GoTo") => Action::Harmless,
            None if action.get(b"D").is_some() => Action::Harmless,
            _ => Action::Other,
        },
        // Unresolvable: it may live in a stream that could not be read.
        _ => Action::Other,
    }
}

/// For an object the parser rejected: readers are more forgiving than this
/// parser, so every name token in the object is still checked by itself.
fn fallback_names(region: &[u8], offset: usize, out: &mut Output<'_>) {
    let end = [b"endobj".as_slice(), b"stream"]
        .iter()
        .filter_map(|keyword| memmem::find(region, keyword))
        .min()
        .unwrap_or(region.len())
        .min(FALLBACK_WINDOW);
    let mut parser = Parser::new(&region[..end], 0, usize::MAX);
    while let Some(slash) = memchr::memchr(b'/', parser.rest()) {
        parser.pos += slash;
        let Some(name) = parser.name() else {
            break;
        };
        check_name(&name, offset, out);
        match name.text.as_slice() {
            b"JS" | b"JavaScript" => javascript(offset, out),
            b"Launch" => launch(offset, out),
            b"OpenAction" => out.report(
                "pdf.open_action",
                Severity::Medium,
                "performs an action when opened",
                offset,
            ),
            b"AA" => out.report(
                "pdf.additional_action",
                Severity::Medium,
                "performs an action on a document or page event",
                offset,
            ),
            b"EmbeddedFile" | b"EmbeddedFiles" | b"EF" => embedded_file(offset, out),
            b"RichMedia" => rich_media(offset, out),
            b"XFA" => xfa(offset, out),
            b"Encrypt" => out.encrypted(offset),
            b"Subtype2" => postscript("embeds a PostScript XObject", offset, out),
            _ => {}
        }
    }
}

/// Decode each embedded file and hand it back to the scanner.
fn scan_embedded_files(
    doc: &Document<'_>,
    budget: &mut Budget,
    out: &mut Output<'_>,
    nested: &mut NestedScan<'_>,
) {
    let typed = doc
        .objects
        .iter()
        .enumerate()
        .filter_map(|(index, object)| {
            let dict = object.value.as_dict()?;
            let is = |key: &[u8], name: &[u8]| doc.name_is(dict.get(key), name);
            (is(b"Type", b"EmbeddedFile") || is(b"Subtype", b"PS") || is(b"Subtype2", b"PS"))
                .then_some(index)
        });
    let referenced = out
        .embedded
        .iter()
        .filter_map(|number| doc.by_number.get(number).copied());
    let mut targets: Vec<usize> = typed.chain(referenced).collect();
    targets.sort_unstable();
    targets.dedup();

    for index in targets {
        let object = &doc.objects[index];
        if object.stream_start.is_none() {
            continue;
        }
        if let Err(limit) = budget.enter() {
            return out.limit(limit);
        }
        match doc.decode(object, budget) {
            Ok(content) => {
                let mut inner = Vec::new();
                nested(&content, budget, &mut inner);
                for finding in inner {
                    if finding.code.starts_with("limit.") {
                        out.findings.push(finding);
                    } else {
                        let message = format!("embedded file: {}", finding.message);
                        out.report(finding.code, finding.severity, &message, object.offset);
                    }
                }
            }
            Err(DecodeError::Limit) => out.limit(LimitExceeded::Decompression),
            Err(_) => out.malformed("embedded file cannot be decoded", object.offset),
        }
        budget.leave();
    }
}

/// Search page content, form XObjects, patterns and glyph procedures for the
/// `PS` operator, which hands its string operand to a PostScript interpreter.
fn scan_content_streams(doc: &Document<'_>, budget: &mut Budget, out: &mut Output<'_>) {
    let named = out
        .contents
        .iter()
        .filter_map(|number| doc.by_number.get(number).copied());
    let typed = doc
        .objects
        .iter()
        .enumerate()
        .filter_map(|(index, object)| {
            let dict = object.value.as_dict()?;
            (doc.name_is(dict.get(b"Subtype"), b"Form") || dict.get(b"PatternType").is_some())
                .then_some(index)
        });
    let mut targets: Vec<usize> = named.chain(typed).collect();
    targets.sort_unstable();
    targets.dedup();

    for index in targets {
        let object = &doc.objects[index];
        if object.stream_start.is_none() {
            continue;
        }
        match doc.decode(object, budget) {
            Ok(content) if uses_operator(&content, b"PS") => {
                return postscript(
                    "content stream uses the PostScript operator",
                    object.offset,
                    out,
                );
            }
            Err(DecodeError::Limit) => return out.limit(LimitExceeded::Decompression),
            // A stream no reader can decode draws nothing.
            _ => {}
        }
    }
}

/// Whether a content stream has `operator` as a token of its own, outside
/// strings, names and comments.
fn uses_operator(content: &[u8], operator: &[u8]) -> bool {
    let mut unclosed_string = false;
    let mut pos = 0;
    while let Some(&byte) = content.get(pos) {
        match byte {
            b'%' => pos += memchr::memchr2(b'\n', b'\r', &content[pos..]).unwrap_or(content.len()),
            // Once a string runs to the end, so does every later one: step
            // over the parenthesis so that it cannot hide what follows.
            b'(' if unclosed_string => pos += 1,
            b'(' => {
                let start = pos;
                let mut depth = 0usize;
                while let Some(&inner) = content.get(pos) {
                    pos += 1;
                    match inner {
                        b'\\' => pos += 1,
                        b'(' => depth += 1,
                        b')' => depth -= 1,
                        _ => {}
                    }
                    if depth == 0 {
                        break;
                    }
                }
                if depth > 0 {
                    unclosed_string = true;
                    pos = start + 1;
                }
            }
            b'<' if content.get(pos + 1) != Some(&b'<') => {
                pos += memchr::memchr(b'>', &content[pos..]).unwrap_or(content.len());
            }
            _ if is_regular(byte) => {
                let length = content[pos..]
                    .iter()
                    .take_while(|&&b| is_regular(b))
                    .count();
                let token = &content[pos..pos + length];
                if token == operator {
                    return true;
                }
                pos += length;
                // Inline image data is binary up to the `EI` that closes it.
                if token == b"ID" {
                    let end = memmem::find_iter(&content[pos..], b"EI").find(|&at| {
                        let before = content[pos + at - 1];
                        let after = content.get(pos + at + 2);
                        is_whitespace(before) && after.is_none_or(|&b| !is_regular(b))
                    });
                    pos += end.unwrap_or(0);
                }
            }
            b'/' => {
                pos += 1;
                pos += content[pos..]
                    .iter()
                    .take_while(|&&b| is_regular(b))
                    .count();
            }
            _ => pos += 1,
        }
    }
    false
}

#[cfg(test)]
mod tests {
    use miniz_oxide::deflate::compress_to_vec_zlib;

    use super::*;
    use crate::limits::Limits;

    fn pdf(objects: &[&[u8]]) -> Vec<u8> {
        let mut out = b"%PDF-1.7\n".to_vec();
        for (index, body) in objects.iter().enumerate() {
            out.extend_from_slice(format!("{} 0 obj\n", index + 1).as_bytes());
            out.extend_from_slice(body);
            out.extend_from_slice(b"\nendobj\n");
        }
        out.extend_from_slice(b"trailer\n<< /Root 1 0 R >>\nstartxref\n0\n%%EOF\n");
        out
    }

    fn stream(entries: &str, data: &[u8]) -> Vec<u8> {
        let mut out = format!("<< {entries} /Length {} >>\nstream\n", data.len()).into_bytes();
        out.extend_from_slice(data);
        out.extend_from_slice(b"\nendstream");
        out
    }

    fn run(data: &[u8]) -> Vec<(&'static str, Severity)> {
        let mut budget = Budget::new(&Limits::default());
        let mut findings = Vec::new();
        let mut nested = |inner: &[u8], _: &mut Budget, findings: &mut Vec<Finding>| {
            if inner.starts_with(b"\x7fELF") {
                findings.push(Finding::new("type.executable", Severity::Critical, "elf"));
            }
        };
        analyze(data, &mut budget, &mut findings, &mut nested);
        findings.iter().map(|f| (f.code, f.severity)).collect()
    }

    fn codes(data: &[u8]) -> Vec<&'static str> {
        run(data).into_iter().map(|(code, _)| code).collect()
    }

    const CATALOG: &[u8] = b"<< /Type /Catalog /Pages 2 0 R /OpenAction [3 0 R /Fit] >>";
    const PAGES: &[u8] = b"<< /Type /Pages /Kids [3 0 R] /Count 1 >>";
    const PAGE: &[u8] = b"<< /Type /Page /Parent 2 0 R /MediaBox [0 0 612 792] >>";

    #[test]
    fn ordinary_document_is_clean() {
        let content = stream("", b"BT /F1 12 Tf (Hello) Tj ET");
        assert_eq!(
            codes(&pdf(&[CATALOG, PAGES, PAGE, &content])),
            [] as [&str; 0]
        );
    }

    #[test]
    fn postscript_xobjects() {
        let program = b"%!PS\n(PAYLOAD) show";
        for entries in [
            "/Type /XObject /Subtype /PS",
            "/Type /XObject /Subtype /Form /Subtype2 /PS",
            "/Type /XObject /Subtype /Form /PS 9 0 R",
        ] {
            let xobject = stream(entries, program);
            assert_eq!(
                codes(&pdf(&[CATALOG, PAGES, PAGE, &xobject])),
                ["pdf.postscript"],
                "{entries}"
            );
        }
        // /PS is only an XObject entry; elsewhere it is a resource name.
        let resources = b"<< /Font << /PS 9 0 R >> >>";
        assert_eq!(
            codes(&pdf(&[CATALOG, PAGES, PAGE, resources])),
            [] as [&str; 0]
        );
        // What the program does is reported too.
        let xobject = stream("/Subtype /PS", b"\x7fELF");
        assert_eq!(
            codes(&pdf(&[CATALOG, PAGES, PAGE, &xobject])),
            ["pdf.postscript", "type.executable"]
        );
    }

    #[test]
    fn postscript_operator_in_content() {
        let page = b"<< /Type /Page /Parent 2 0 R /Contents 4 0 R >>";
        let content = stream("", b"BT (text) Tj ET\n(1 2 moveto) PS\n");
        assert_eq!(
            codes(&pdf(&[CATALOG, PAGES, page, &content])),
            ["pdf.postscript"]
        );
        let packed = stream(
            "/Filter /FlateDecode",
            &compress_to_vec_zlib(b"q (x) PS Q", 6),
        );
        assert_eq!(
            codes(&pdf(&[CATALOG, PAGES, page, &packed])),
            ["pdf.postscript"]
        );
        let form = stream("/Type /XObject /Subtype /Form", b"(x)PS");
        assert_eq!(
            codes(&pdf(&[CATALOG, PAGES, PAGE, &form])),
            ["pdf.postscript"]
        );
        // The same letters as a string, a name, a comment or part of a word.
        let harmless = stream("", b"(PS) Tj <5053> Tj /PS gs % PS\nPSx EPS BT ET");
        assert_eq!(
            codes(&pdf(&[CATALOG, PAGES, page, &harmless])),
            [] as [&str; 0]
        );
        // Neither inline image data nor a stray parenthesis hides it.
        for hidden in [
            &b"BI /W 1 /H 1 ID (\x00 EI (x) PS"[..],
            b"BI /W 1 /H 1 ID \x00) EI (x) PS",
            b"( (x) PS",
        ] {
            let content = stream("", hidden);
            assert_eq!(
                codes(&pdf(&[CATALOG, PAGES, page, &content])),
                ["pdf.postscript"],
                "{hidden:?}"
            );
        }
        // A stream nothing names as content is not searched.
        let other = stream("/Length1 5", b"(x) PS");
        assert_eq!(
            codes(&pdf(&[CATALOG, PAGES, PAGE, &other])),
            [] as [&str; 0]
        );
    }

    #[test]
    fn javascript_action() {
        let action = b"<< /S /JavaScript /JS (app.alert(1)) >>";
        assert_eq!(
            codes(&pdf(&[CATALOG, PAGES, PAGE, action])),
            ["pdf.javascript"]
        );

        let names = b"<< /Names << /JavaScript 9 0 R >> >>";
        assert_eq!(
            codes(&pdf(&[CATALOG, PAGES, PAGE, names])),
            ["pdf.javascript"]
        );
    }

    #[test]
    fn open_action_severity_follows_its_target() {
        let scripted = b"<< /Type /Catalog /Pages 2 0 R /OpenAction 4 0 R >>";
        let action = b"<< /S /JavaScript /JS 5 0 R >>";
        assert_eq!(
            run(&pdf(&[scripted, PAGES, PAGE, action])),
            [
                ("pdf.open_action", Severity::High),
                ("pdf.javascript", Severity::High),
            ]
        );

        let link = b"<< /Type /Catalog /OpenAction << /S /URI /URI (http://example.com) >> >>";
        assert_eq!(
            run(&pdf(&[link])),
            [
                ("pdf.open_action", Severity::Medium),
                ("pdf.uri", Severity::Info),
            ]
        );

        let dangling = b"<< /Type /Catalog /OpenAction 99 0 R >>";
        assert_eq!(
            run(&pdf(&[dangling])),
            [("pdf.open_action", Severity::Medium)]
        );
    }

    #[test]
    fn additional_actions_and_launch() {
        let page = b"<< /Type /Page /AA << /O << /S /Launch /F (cmd.exe) >> >> >>";
        assert_eq!(
            run(&pdf(&[CATALOG, PAGES, page])),
            [
                ("pdf.additional_action", Severity::High),
                ("pdf.launch", Severity::High),
            ]
        );
        let harmless = b"<< /Type /Page /AA << /O << /S /GoTo /D [3 0 R /Fit] >> >> >>";
        assert_eq!(codes(&pdf(&[CATALOG, PAGES, harmless])), [] as [&str; 0]);
    }

    #[test]
    fn hex_escaped_keywords() {
        let action = b"<< /S /J#61vaScript /J#53 (app.alert(1)) >>";
        assert_eq!(
            codes(&pdf(&[CATALOG, PAGES, PAGE, action])),
            ["pdf.javascript", "pdf.obfuscated_name"]
        );
        let font = b"<< /Type /Font /BaseFont /ABCDEF+Arial#20Bold >>";
        assert_eq!(codes(&pdf(&[CATALOG, PAGES, PAGE, font])), [] as [&str; 0]);
    }

    #[test]
    fn javascript_inside_object_streams() {
        let inner = b"10 0 11 16 << /Foo /Bar >> << /S /JavaScript /JS (x) >>";
        let packed = stream(
            "/Type /ObjStm /N 2 /First 11 /Filter /FlateDecode",
            &compress_to_vec_zlib(inner, 6),
        );
        assert_eq!(
            codes(&pdf(&[CATALOG, PAGES, PAGE, &packed])),
            ["pdf.javascript"]
        );

        // A filter chain, with the stream length stored in another object.
        let hex: String = compress_to_vec_zlib(inner, 6)
            .iter()
            .map(|byte| format!("{byte:02x}"))
            .collect();
        let mut chained =
            b"<< /Type /ObjStm /N 2 /First 11 /Filter [/ASCIIHexDecode /FlateDecode] /Length 5 0 R >>\nstream\n"
                .to_vec();
        chained.extend_from_slice(hex.as_bytes());
        chained.extend_from_slice(b">\nendstream");
        let length = format!("{}", hex.len() + 1);
        assert_eq!(
            codes(&pdf(&[CATALOG, PAGES, PAGE, &chained, length.as_bytes()])),
            ["pdf.javascript"]
        );
    }

    #[test]
    fn unreadable_object_streams_are_never_clean() {
        let garbage = stream(
            "/Type /ObjStm /N 1 /First 4 /Filter /FlateDecode",
            b"\xFF\xFF\xFF",
        );
        assert_eq!(codes(&pdf(&[CATALOG, &garbage])), ["pdf.malformed"]);
        let exotic = stream("/Type /ObjStm /N 1 /First 4 /Filter /JBIG2Decode", b"xx");
        assert_eq!(codes(&pdf(&[CATALOG, &exotic])), ["pdf.malformed"]);
    }

    #[test]
    fn embedded_files_are_scanned() {
        let filespec = b"<< /Type /Filespec /F (tool.bin) /EF << /F 3 0 R >> >>";
        let payload = stream(
            "/Filter /FlateDecode",
            &compress_to_vec_zlib(b"\x7fELF\x02\x01", 6),
        );
        assert_eq!(
            run(&pdf(&[CATALOG, filespec, &payload])),
            [
                ("pdf.embedded_file", Severity::Medium),
                ("type.executable", Severity::Critical),
            ]
        );
        let text = stream("/Type /EmbeddedFile", b"plain notes");
        assert_eq!(codes(&pdf(&[CATALOG, &text])), ["pdf.embedded_file"]);
    }

    #[test]
    fn encryption_and_other_features() {
        let mut encrypted = pdf(&[CATALOG]);
        encrypted.extend_from_slice(b"trailer\n<< /Root 1 0 R /Encrypt 7 0 R >>\n%%EOF\n");
        assert_eq!(
            codes(&encrypted),
            ["pdf.encrypted", "pdf.incremental_update"]
        );

        let form = b"<< /AcroForm << /XFA 8 0 R >> >>";
        let media = b"<< /Type /Annot /Subtype /RichMedia >>";
        let submit = b"<< /S /SubmitForm /F (http://example.com) >>";
        let remote = b"<< /S /GoToR /F (other.pdf) >>";
        assert_eq!(
            codes(&pdf(&[form, media, submit, remote])),
            [
                "pdf.xfa",
                "pdf.rich_media",
                "pdf.submit_form",
                "pdf.remote_goto"
            ]
        );
    }

    #[test]
    fn unparsable_objects_are_still_searched() {
        let broken = b"<< /S /JavaScript /JS (app.alert(1)) /Oops >";
        assert_eq!(
            codes(&pdf(&[CATALOG, broken])),
            ["pdf.malformed", "pdf.javascript"]
        );
    }

    #[test]
    fn structural_problems() {
        let mut prefixed = b"GIF89a junk ".to_vec();
        prefixed.extend(pdf(&[CATALOG]));
        assert_eq!(codes(&prefixed), ["pdf.malformed"]);

        assert_eq!(codes(b"%PDF-1.7\n%%EOF\n"), ["pdf.malformed"]);

        let complete = pdf(&[CATALOG]);
        let cut = &complete[..complete.len() - 7];
        assert_eq!(codes(cut), ["pdf.malformed"]);
    }

    #[test]
    fn objects_hidden_inside_stream_data_are_found() {
        let hidden = stream("", b"1 0 obj\n<< /S /JavaScript /JS (x) >>\nendobj");
        assert_eq!(codes(&pdf(&[CATALOG, &hidden])), ["pdf.javascript"]);
    }

    #[test]
    fn decompression_is_budgeted() {
        let limits = Limits {
            max_decompressed: 1000,
            ..Limits::default()
        };
        let bomb = stream(
            "/Type /ObjStm /N 1 /First 4 /Filter /FlateDecode",
            &compress_to_vec_zlib(&vec![b' '; 1_000_000], 6),
        );
        let mut budget = Budget::new(&limits);
        let mut findings = Vec::new();
        analyze(
            &pdf(&[CATALOG, &bomb]),
            &mut budget,
            &mut findings,
            &mut |_, _, _| {},
        );
        assert_eq!(findings[0].code, "limit.decompression");
    }
}