-
Notifications
You must be signed in to change notification settings - Fork 190
/
final.pl
455 lines (390 loc) · 21.5 KB
/
final.pl
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
% Copyright 2017-2020 Carnegie Mellon University.
% ============================================================================================
% Final reporting API.
% ============================================================================================
:- use_module(library(lists), [member/2, min_list/2, max_list/2]).
% Long ago, Ed wrote: A solution is <list of classes, a mapping of classes to methods, a
% mapping of classes to members, a mapping of each class to its immediate parents and offset, a
% list of vtables, mapping of vtables to class x offset x size, map of virtual function calls
% to classes and methods, special methods of class>. Now it's more correct to say something
% like: "A solution is a list of finalXXX() predicates contained in this section".
% These rules are for formatting the results into "final" format. These rules are not used for
% reasoning, and do not involve guessing or backtracking in any way. As a consequence it's not
% obvious that they need to be tabled (although there's probably no harm in that).
% ejs: It sure would be nice if there was a consistent prolog library...
% ============================================================================================
% We've found that under some circumstances (see https://github.com/cmu-sei/pharos/issues/101)
% vftables can be reused on different classes. We believe this is only when no methods are
% overridden. Here is an example.
%% Concluding factVFTableWrite(0x70d682, 0x70d650, 0, 0xa2c85c).
%% Concluding factVFTableWrite(0x70d147, 0x70d0a0, 0x16c, 0x9ae574).
%% Concluding factVFTableWrite(0x70d133, 0x70d0a0, 0, 0xa2c85c).
% Ed thought there were two possibilities:
%% 1. Both classes inherit from a common base class but don't override any functions. Class
%% 0x70d0a0 inherits from another class as well.
%% 2. Class 0x70d0a0 inherits from 0x70d650 and
%% another class, and does not override any methods from 0x70d650. Normally we'd see a vftable
%% overwrite, but since the vftables are the same we might not.
%% This predicate adds an extra check to make sure that there is exactly one class that claims
%% ownership of the VFTable. Right now we only envision it being used in final.pl, hence the
%% name.
reasonPrimaryVFTableForClassFinal(V, C) :-
nonvar(C),
% Hey we found one!
findVFTable(V, 0, C),
!,
% No one else better also claim it
forall(findVFTable(V, 0, OC), C = OC).
% ============================================================================================
% Class Identification...
% ============================================================================================
% Once we've finished assigning methods to constructors, we can assign class identifiers. This
% version of the rule is a bit nasty because we can't rely on multiple implementations of the
% rule and tabling to give us just one answer... I think.
:- table classIdentifier/2 as opaque.
classIdentifier(Method, ID) :-
var(Method),
nonvar(ID),
throw_with_backtrace(error(instantiation_error, classIdentifier/2)).
classIdentifier(Method, ID) :-
find(Method, Class),
((
% Must be the VFTable at offset zero to be the "master" table for the class.
reasonPrimaryVFTableForClassFinal(VFTable, Class),
logtraceln('Setting classID of ~Q to primary vftable ~Q', [Class, VFTable]),
true
)
->
ID is VFTable
;
% This will use any VFTable on the class as an identifier but only if there is a single
% VFTable.
(
once((findVFTable(VFTable, Class),
forall(findVFTable(OtherVFTable, Class),
VFTable = OtherVFTable),
logtraceln('Setting classID of ~Q to only vftable ~Q', [Class, VFTable])))
) ->
ID is VFTable
;
(
find(RealDestructor, Class),
factRealDestructor(RealDestructor),
logtraceln('Setting classID of ~Q to real destructuor ~Q', [Class, RealDestructor]),
true
)
->
ID is RealDestructor
;
(
findallMethods(Class, MethodSet),
%logwarnln('trying to pick class ID from method set1 ...'),
%logwarnln('picking class ID from method set: ~Q', MethodSet),
setof(C, (member(C, MethodSet), factConstructor(C)), ConstructorSet),
%logwarnln('constructor set was: ~Q', ConstructorSet),
min_list(ConstructorSet, MinimumConstructor),
logtraceln('Setting classID of ~Q to minimum constructor ~Q', [Class, MinimumConstructor]),
true
)
-> ID is MinimumConstructor
;
(
findallMethods(Method, MethodSet),
%logwarnln('trying to pick class ID from method set2 ...'),
%logwarnln('picking class ID from method set: ~Q', MethodSet),
min_list(MethodSet, MinimumMethod),
logtraceln('Setting classID of ~Q to minimum method ~Q', [Class, MinimumMethod]),
true
)
-> ID is MinimumMethod).
% --------------------------------------------------------------------------------------------
% A helper for identifying "worthless" classes to reduce noise in the output.
:- table worthlessClass/1 as opaque.
% A class is useful if it appears in a relationship.
usefulClass(Class) :-
factEmbeddedObject(_, Class, _);
factEmbeddedObject(Class, _, _);
factDerivedClass(_, Class, _);
factDerivedClass(Class, _, _).
% A class is useful if it has a VFTable.
usefulClass(Class) :-
findVFTable(_VFTable, Class).
% A class is useful if it is not size zero.
usefulClass(Class) :-
reasonMinimumPossibleClassSize(Class, Size), Size > 0.
% A class containing a constructor is useful.
usefulClass(Class) :-
findMethod(Method, Class),
factConstructor(Method).
% A class containing a real destructor is useful.
usefulClass(Class) :-
findMethod(Method, Class),
factRealDestructor(Method).
% A class with more than one method is useful.
usefulClass(Class) :-
findallMethods(Class, List),
length(List, Len),
Len > 1.
% We want to reject classes with no useful information. In practice this means having a single
% method, and no other "useful" data like a real destructor, a vftable, etc. Sadly the final
% class rule is pretty complicated, this rule is largely a copy of that rule. Maybe in the
% future we can find some witty way to structure this that results in better code reuse.
worthlessClass(Class) :-
var(Class),
throw_with_backtrace(error(instantiation_error, worthlessClass/1)).
% A class is worthless if (1) it's not useful, or (2) consists only of purecall.
worthlessClass(Class) :-
not(usefulClass(Class)),
!,
logtraceln('Rejecting worthless finalClass ~Q', [Class]).
worthlessClass(Class) :-
purecall(Class),
is_singleton(Class),
!,
logtraceln('Rejecting worthless finalClass ~Q because it contains purecall and is a singleton', [Class]).
finalFileInfo(FileMD5, Filename) :-
fileInfo(FileMD5, Filename).
% --------------------------------------------------------------------------------------------
% This final result defines the existance of a class. More details in results.txt.
%:- finalClass/6 as incremental.
finalClass(ClassID, VFTableOrNull, CSize, LSize, RealDestructorOrNull, MethodList) :-
class(Class),
not(worthlessClass(Class)),
classIdentifier(Class, ClassID),
% If there's a certain VFTableWrite, use the VFTable value from it. On the other hand if
% there is a single VFTable in the class, use that. Otherwise return zero (null).
((findVFTable(VFTable, 0, Class);
(findVFTable(VFTable, Class),
forall(findVFTable(OtherVFTable, Class), VFTable = OtherVFTable)))
->
VFTableOrNull=VFTable
;
VFTableOrNull=0),
% Get the certain and likely class sizes.
reasonMinimumPossibleClassSize(Class, CSize),
LSize is CSize,
% Optionally find the the real destructor as well.
((find(RealDestructor, Class),
factRealDestructor(RealDestructor))
-> RealDestructorOrNull=RealDestructor; RealDestructorOrNull=0),
findallMethods(Class, UnsortedMethodList),
sort(UnsortedMethodList, MethodList).
% --------------------------------------------------------------------------------------------
% This final result defines the properties of a VFTable. More details in results.txt.
%:- finalVFTable/5 as incremental.
finalVFTable(VFTable, CertainSize, LikelySize, RTTIAddressOrNull, RTTINameOrNull) :-
factVFTable(VFTable),
(reasonRTTIInformation(VFTable, RTTIAddress, RTTIName) ->
RTTIAddressOrNull=RTTIAddress, RTTINameOrNull=RTTIName;
RTTIAddressOrNull=0, RTTINameOrNull=''),
findall(CertainOffset, factVFTableEntry(VFTable, CertainOffset, _Method), CertainOffsets),
max_list(CertainOffsets, CertainMax),
CertainSize is CertainMax + 4,
% It's a little unclear what the likely size means in a proper guessing framework.
LikelySize is CertainSize.
finalVFTableEntry(VFTable, Offset, Method) :-
factVFTableEntry(VFTable, Offset, Method).
% --------------------------------------------------------------------------------------------
%:- finalVBTable/4 as incremental.
finalVBTable(VBTable, Class, Size, Offset) :-
factVBTable(VBTable),
factVBTableWrite(_Insn, Method, Offset, VBTable),
findall(CertainOffset, factVBTableEntry(VBTable, CertainOffset, _Value), CertainOffsets),
max_list(CertainOffsets, Size),
classIdentifier(Method, Class).
finalVBTableEntry(VBTable, Offset, Value) :-
factVBTableEntry(VBTable, Offset, Value).
% --------------------------------------------------------------------------------------------
%:- finalEmbeddedObject/4 as incremental.
% In the Outer class identifier at the specified offset is an object instance of the type
% specified by EmbeddedClass indeitifier. The embedded object is not believed to be a base
% class (via an inheritance relationship), for that relationship see finalInheritance().
finalEmbeddedObject(OuterClass, Offset, EmbeddedClass, likely) :-
factEmbeddedObject(OuterClass1, EmbeddedClass1, Offset),
classIdentifier(OuterClass1, OuterClass),
classIdentifier(EmbeddedClass1, EmbeddedClass),
iso_dif(OuterClass, EmbeddedClass).
% --------------------------------------------------------------------------------------------
%:- finalInheritance/5 as incremental.
% In the Derived class at the specified offset is an object instance of the type specified by
% Base. If the base class is not the first base class (non-zero offset) and the base class has
% a virtual function table, the VFTable field will contain the derived class instance of that
% virtual function table.
finalInheritance(DerivedClassID, BaseClassID, ObjectOffset, VFTableOrNull, false) :-
factDerivedClass(DerivedClass, BaseClass, ObjectOffset),
% The following line stops us from reporting inheritances from A to C if there is also a
% relation from A to B and B to C. We have this because virtual inheritance is known to
% introduce "false" A to C relationships depending on the order that guessing occurs in.
once((not((factDerivedClass(DerivedClass, OtherClass, _Off1), factDerivedClass(OtherClass, BaseClass, _Off2))),
classIdentifier(DerivedClass, DerivedClassID),
classIdentifier(BaseClass, BaseClassID),
% Try to identify the relevant VFTable
((find(VFTable, DerivedClass),
factVFTableWrite(_Insn, DerivedConstructor, ObjectOffset, VFTable),
find(DerivedConstructor, DerivedClass))
->
VFTableOrNull = VFTable
;
VFTableOrNull is 0))).
% Cory's a little uncertain about this rule because it's unclear if we'll ever assert a certain
% inheritance relationship without the virtual table fact that would trigger the rule above.
% Specifically, there's currently no likelyDerivedConstructor() or likelyDerivedClass(), and
% it's unclear that there ever will be. Perhaps we'll end up guessing based on knowing that
% both are constructors, and on calls the other, and there's no contradictory member access.
% Or something equally complicated. For now, we'll just use the confidence field to indicate
% that there's not a virtual function table in this result, but I don't think this rule can
% ever trigger currently either.
%finalInheritance(DerivedClass, BaseClass, ObjectOffset, likely, 0) :-
% certainDerivedClass(DerivedClass, BaseClass, ObjectOffset),
% % Unify with a specific constructor to unify with the VFTable value.
% classIdentifier(DerivedConstructor, DerivedClass),
% not_exists(factVFTableWrite(_Insn, DerivedConstructor, ObjectOffset, _VFTable)).
% --------------------------------------------------------------------------------------------
%%:- certainMemberAccessEvidence/4 as incremental.
% Find all methods that access offsets in the class.
certainMemberAccessEvidence(Class, Offset, Size, Insn) :-
certainMemberOnClass(Class, Offset, Size),
find(Method, Class),
factMethod(Method),
validMethodMemberAccess(Insn, Method, Offset, Size).
% --------------------------------------------------------------------------------------------
%:- finalMemberAccess/4 as incremental.
% The member at Offset in Class was accessed using the given Size by the list of evidence
% instructions provided. The list of evidence instructions only contains instructions from the
% methods assigned to the class. Other accesses of base class members will appear in the
% accesses for their respective classes. There is no certainty field, which must be derived
% from the certainty of the method that the instruction is located in. This certainty may vary
% for individual evidence instructions within the list. Note that that presentation does not
% present knowledge about the class and subclass relationships particularly clearly. For those
% opbservations refer to finalMember() instead.
finalMemberAccess(ClassID, Offset, Size, EvidenceList) :-
certainMemberOnClass(Class, Offset, Size),
not(worthlessClass(Class)),
classIdentifier(Class, ClassID),
setof(I, certainMemberAccessEvidence(Class, Offset, Size, I), UnsortedEvidenceList),
sort(UnsortedEvidenceList, EvidenceList).
% --------------------------------------------------------------------------------------------
%:- finalMember/4 as incremental.
% The finalMember() result documents the existence of the definition of a member on a specific
% class. The intention (possibly with a currently incorrect implementation) is that we will
% only report the members defined on the class from a C++ source code perspective. The Class
% is specified with a class identifier (not a constructor address or other specific type of
% address). Offset is the positive offset into the specified class (and not it's base or
% embedded classes). Sizes is a list of all the different sizes through which the member has
% been accessed anywhere in the program. The final field indicates our confidence in the
% existence of member, and may be 'certain' or 'likely'. Embedded object and inherited bases
% are not listed again as finalMembers. Instead they are list in the finalEmbeddedObject and
% finalInheritance results.
finalMember(ClassID, Offset, Sizes, certain) :-
certainMemberOnExactClass(Class, Offset, EarlySize),
not(worthlessClass(Class)),
classIdentifier(Class, ClassID),
setof(Size, certainMemberOnExactClass(Class, Offset, Size), UnsortedSizes),
%subtract(UnsortedSizes, [0], FilteredSizes),
%sort(FilteredSizes, Sizes).
sort(UnsortedSizes, Sizes),
% As a hack to prevent us from outputting a Class, Offset multiple times, only proceed if
% EarlySize == Sizes[0]
UnsortedSizes = [EarlySize|_].
% ============================================================================================
% Final Method Properties
% ============================================================================================
% --------------------------------------------------------------------------------------------
%:- finalMethodProperty/3 as incremental.
% In many cases, no qualifiers of any kind are required for specific methods. However, there
% are multiple properties of a method that can all be determined at different confidence levels
% independently.
% This result marks that a method is certain to be a constructor. The certain field is
% currently scheduled for elimination.
finalMethodProperty(Method, constructor, certain) :-
factConstructor(Method),
find(Method, Class),
not(worthlessClass(Class)).
% --------------------------------------------------------------------------------------------
% This result marks that a method is certain to be a deleting destructor. The certain field is
% currently scheduled for elimination.. The single method that is a real destructor for any
% given class is determined from the finalClass result.
finalMethodProperty(Method, deletingDestructor, certain) :-
factDeletingDestructor(Method),
find(Method, Class),
not(worthlessClass(Class)).
% --------------------------------------------------------------------------------------------
% At one point I had though that this was duplicative, and therefore should not be included in
% the output. After having been annoyed by it's absence repeatedly, we're going to at least
% try it with it on for a while.
finalMethodProperty(Method, realDestructor, certain) :-
factRealDestructor(Method),
find(Method, Class),
not(worthlessClass(Class)).
% --------------------------------------------------------------------------------------------
% This result marks whether a method is believed to be virtual, and what our confidence in that
% assertion is. It's likely that this is exactly equivalent to whether the method appears in
% any virtual function table, and what the confidence in that offset of that virtual function
% table is. This fact may be equally easy to compute from the finalVFTable results after
% importation back into C++. In that case, this result may be eliminated.
% The obvious rule is that if we're certain that the method is virtual, we should report so.
finalMethodProperty(Method, virtual, certain) :-
% By unifying Method first, we make sure that we don't produce duplicate results. It may
% not be the most efficient thing to do here.
factMethod(Method),
once((symbolProperty(Method, virtual);
(factMethodInVFTable(_VFTable, _Offset, Method), not(symbolProperty(Method, virtual))))),
find(Method, Class),
not(worthlessClass(Class)),
not(purecall(Method)).
% Unimplemented rule: There might be a reasoning rule about transferring knowledge from one
% virtual function to another. Specifically that the virtual function table of a Derived class
% must be as large (or larger than it's base class?)
% --------------------------------------------------------------------------------------------
%:- finalResolvedVirtualCall/3 as incremental.
% The call at Insn can be resolved to the Target through the virtual function table at VFTable.
% Most of the information describing a virtual function call is already avilable in C++ once
% you know the address of the call instruction. This result communicates just the required
% information for simplicity and clarity.
finalResolvedVirtualCall(Insn, VFTable, Target) :-
% Have we already confirmed that the Insn is legitimately a virtual function call?
factVirtualFunctionCall(Insn, _Method, _ObjectOffset, VFTable, VFTableOffset),
% Now actually resolve the call using the VftableOffset.
factVFTableEntry(VFTable, VFTableOffset, Entry),
dethunk(Entry, Target).
% --------------------------------------------------------------------------------------------
%:- table finalThunk/2 as incremental.
finalThunk(From, To) :-
% First there needs to be a thunk.
thunk(From, To),
% But it also needs to appear in a finalVFTableEntry.
once(finalVFTableEntry(_Address, _Offset, From)),
% And the target needs to be associated with finalClass method.
once((finalClass(_ClassID, _VFTable, _MinSize, _MaxSize, _RealDestructor, MethodList),
member(To, MethodList))),
% Is this approach more efficient? Less accurate because of worthless classes?
% find(To, _ClassID),
true.
% --------------------------------------------------------------------------------------------
%:- table finalDemangledName/4 as incremental.
finalDemangledName(Address, MangledName, ClassName, '') :-
rTTITypeDescriptor(TDA, _Unclear, MangledName, ClassName),
rTTITDA2VFTable(TDA, Address),
finalClass(_ClassID, Address, _MinSize, _MaxSize, _RealDestructor, _MethodList).
finalDemangledName(Address, MangledName, ClassName, MethodName) :-
finalClass(_ClassID, _VFTable, _MinSize, _MaxSize, _RealDestructor, MethodList),
member(Address, MethodList),
symbolClass(Address, MangledName, ClassName, MethodName).
% --------------------------------------------------------------------------------------------
generateResults :-
writeln('% Prolog results autogenerated by OOAnalyzer.'),
(setof((M3, N2), finalFileInfo(M3, N2), _Set0); true),
(setof((V1, C1, L1, A1, N1), finalVFTable(V1, C1, L1, A1, N1), _Set1); true),
(setof((C2, V2, S2, L2, R2, M2), finalClass(C2, V2, S2, L2, R2, M2), _Set2); true),
(setof((I3, V3, T3), finalResolvedVirtualCall(I3, V3, T3), _Set3); true),
(setof((C4, O4, E4, X4), finalEmbeddedObject(C4, O4, E4, X4), _Set4); true),
(setof((D5, B5, O5, C5, V5), finalInheritance(D5, B5, O5, C5, V5), _Set5); true),
(setof((C6, O6, S6, L6), finalMember(C6, O6, S6, L6), _Set6); true),
(setof((C7, O7, S7, E7), finalMemberAccess(C7, O7, S7, E7), _Set7); true),
(setof((M8, P8, C8), finalMethodProperty(M8, P8, C8), _Set8); true),
writeln('% Object detection reporting complete.'),
reportStage('Complete'),
ws_end.
%% Local Variables:
%% mode: prolog
%% End: