-
Notifications
You must be signed in to change notification settings - Fork 7
/
state.rs
412 lines (368 loc) · 16.5 KB
/
state.rs
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
use std::{
thread,
time::Duration,
cmp::min,
sync::Weak,
};
extern crate rand;
use rand::Rng;
use banana::bananaq;
use banana::bananaq::FuzzyQ;
use exec::call::Call;
use exec::fd_info::Fd;
use super::id::StateTableId;
pub trait IFdState {
fn invalid(&self) -> bool;
/// once we want to share FD between processes / threads, we should be able to check if those are
/// valid ( alive )
///
/// - can be default = true, but better manualy craft per state
/// - you can enhance effectivity of fuzzing this way
fn is_online(&mut self) -> bool;
}
/// expose interface for fuzzing framework - to fuzz & log & debug!
pub trait IFuzzyObj : Send + Sync + IFdState {//quite pitty that i can not do Box<IFuzzyObj + IFdState>...
/// lets do one fuzzing loop over specific State
///
/// - need to invoke do_fuzz_one -> will select and invoke (sys)call
/// - need to invoke do_fuzz_update -> change state of current State ( allow to fuzz other layers)
/// - in between place your fuzzy knowledge to boost fuzzing ( optionaly )
/// # Example :
/// ```
/// impl IFuzzyObj for TestState {
/// fn fuzzy_loop(&mut self) -> bool {
/// while !self.state.do_fuzz_one() {
/// }
///
/// { let data: &test_struct = self.state.call_mut().args(0).read_unsafe(); }
///
/// println!(">>{}", self.state.call_view().serialize(self.info.fd.data()));
/// self.state.do_fuzz_update()
/// }
/// }
/// ```
fn fuzzy_loop(&mut self, state_idx: u16) -> Result<(), String>;
/// will do essentialy same as fuzzy_loop, however is invoked only until state is not initialized
fn fuzzy_init(&mut self) -> Result<(), String>;
/// we want forward information about current state + calls to callbacks
///
/// - only immutable
/// - mutable state *must* be only at queue worker who actually do fuzzing
/// - therefore only one mutable point, rest are imutable ones!
/// # Example :
/// ```
/// impl IFuzzyObj for TestState {
/// fn state(&self) -> &State {
/// &self.state
/// }
/// }
/// ```
fn state(&self) -> &State;
}
/// sharable informations with FuzzyQueue (rnd fd arg, de-dups, ..) && with modules; cheap to copy
#[derive(Clone)]
pub struct StateInfo {
pub bananaq: Weak<FuzzyQ>,
pub name: String,//maybe too expensive to share, and can be skipped ?
/// id is specific per fuzzing target ( vmwp, vmbus, packets, w32k, ntos, alpc, io, .. )
pub id: StateTableId,
/// runtime id of state! unique identifiable because of POC and building connections between state and for races
pub fd: Fd,
/// num of total fuzzing iterations performed
pub total: usize,
/// num of sucessfull ( syscall return OK value ) fuzzing iterations performed; debug reasons
pub sucess: usize,
/// level in ccache -> knowledge based fuzzing
pub level: usize,
/// signaling when dtors apply
pub finished: bool,
}
impl StateInfo {
pub fn uid(&self) -> u64 { u64::from(thread::current().id().as_u64()) }
pub fn bananaq(&self) -> Weak<FuzzyQ> { self.bananaq.clone() }
}
/// user mode state ( representation ) of target ( kernel object, remote object, io device, .. )
pub struct State {
/// sharable informations with FuzzyQueue (rnd fd arg, de-dups, ..) && with modules
info: StateInfo,
/// hardcoded limit per object how many iteration of fuzzing to perform on it - then close
limit: usize,
dead_ratio: f64,
/// one of the main features of state, we can here calibrate behaviour which (set of)function should be called when
slopes: Vec<[isize; 2]>,
/// we need to preserve state, which level and which call is fuzzed
///
/// - after do_fuzz_one it implies that self.call_view() will return currently fuzzed object
/// - on this object was just performed syscall!
/// - after do_fuzz_update, implies :
/// - self.level() points to current level ( updated by slopes )
/// - self.call_view() now is invalid!
ccache: (usize, usize),
/// 2 dimensional vector of target syscalls / apicalls / operations
///
/// - grouped by researcher knowledge - he can make it SOTA fuzzer for target or limit it way too much and fail
/// - no more calls in group than 0x200! - (just empirical numero)
/// - by def is this classic groups :
/// group1 : ctors => calls creating particular object in kernel
/// group2 : workers => calls working over this object
/// - can be grouped also with getters, checkers ( as those seems little impact, so waste of time to fuzz it with same prio as setters ? )
/// - then grouping is here to overcome complicated set-state scenarios, to help fuzzer in
/// blackbox settings!
/// - note that dtors are *not* in group!!
groups: Vec< Vec<Call> >,
/// at the end of fuzzing ( do_fuzz_update will say NO MORE ) is invoked dtor callback, should close resource in kernel / target
///
/// - close, CloseHandle, DeleteDC, DeleteObject, ZwFreeVirtualMemory, ...
/// - i experienced that state has always only one dtor - or can be dtored by only one call / packet
/// - if in future will get knowledge about breaking this rule, then vec![dtors] should be provided!
dtor: Call,
}
impl State {
pub fn name(&self) -> &str { &self.info.name }
pub fn id(&self) -> StateTableId { self.info.id.clone() }
pub fn info(&self) -> StateInfo { self.info.clone() }
pub fn fd(&self) -> &Fd { &self.info.fd }
pub fn level(&self) -> usize { self.ccache.0 }
pub fn call_view(&self) -> &Call { &self.groups[self.ccache.0][self.ccache.1] }
pub fn total(&self) -> usize { self.info.total }
pub fn sucess(&self) -> usize { self.info.sucess }
/// usually we dont know fd at creation of object, therefore after first OK syscalls should be initialized by IFuzzyObj holder
///
/// - call per fuzzy obj / state in fuzzy_init exposed by trait owned by holder of this State!
pub fn init(&mut self, fd: &Fd) { self.info.fd.init(&fd) }
/// mainly because dummy + negate trick, other cases should be reconsidered because PoC
/// generation legacy, you do something too fancy you are likely not able to repro in poc ...
pub fn call_mut(&mut self) -> &mut Call { &mut self.groups[self.ccache.0][self.ccache.1] }
/// select random one from current level
///
/// - maybe intercorporate some ranking of OK syscalls - aka prefer some syscalls at current level
/// - heatmaps, reinforcement learning, other algorithm
/// - currently is opt-ed to do it via modules, but it is perf overkill
/// - therefore if once decide for particular algorithm should be implemented here and
/// switched by config
/// - also some blacklisting of too often rejected call - very OK by modules
/// - and maybe tracking uniformity of selection globaly per State - forbid prefering one syscall by thread_rng ?
/// - in general this can be crucial part of fuzzer which i neglected to tinker with yet..
pub fn do_fuzz_one(&mut self, shared: &mut[u8]) -> Result<(), String> {
if self.info.total > self.limit {
return Err(format!("[state] out of fuzzing limit total : {:?} > limit : {:?}",
self.info.total, self.limit))
}
self.info.total += 1;
let fd = self.info.fd.clone();
let mut dead = false;
let bananaq = &self.info.bananaq;
let mut i = 0;
for _ in 0u32.. {
i += 1;
if !bananaq::is_active(&self.info.bananaq())? {
break
}
self.ccache.1 = rand::thread_rng().gen_range(0..self.groups[self.ccache.0].len());
if !self.call_view().dead(self.dead_ratio)
&& self.groups[self.ccache.0][self.ccache.1]
.do_call(&bananaq, &fd.data(), shared)
{ return Ok(()) }
thread::sleep(Duration::from_nanos(1));
if self.groups[self.ccache.0]
.iter()
.any(|ref call| !call.dead(self.dead_ratio)) { continue }
dead = true;
break
}
self.call_dtor(shared);
Err(format!("[state] end of fuzzing cycle for this object <{:?}::{:?}>, is dead : {:?}; fuzzing attempts for this round : {:?}, total fuzzed calls : {:?} -> {}\n\t==> w/ shared<{:?}>", self.info.name, fd.data(), dead, i, self.info.total, if self.ccache.0 < self.groups.len() && self.ccache.1 < self.groups[self.ccache.0].len() { self.call_view().name() } else { "no-call-yet" }, &shared))
}
/// need to be called after do_fuzz_one, to change level based on slopes!
///
/// - call once do_fuzz_one will return true
/// - this basically wraps do_fuzz_update_impl, that it checks for end of fuzzing
/// - and then it performs closing of this State by calling dtor syscall!
pub fn do_fuzz_update(&mut self, shared: &mut[u8]) -> Result<(), String> {
thread::yield_now();
//maybe better weight it against #calls in current group
//this way we trust a lot that every call is implemented correctly == good way
//it is quick death then, like object is no longer online
//however, if some syscall poorely implemented this will kill fuzzing for whole fuzzy object most
// likely ...
let e = match self.do_fuzz_update_impl() {
Ok(_) => return Ok(()),
Err(e) => e,
};
self.call_dtor(shared);
Err(format!("[state] failing fuzzning cycle : update failed for fuzzed object <{:?}::{:?}> total fuzzed calls : {:?}, with details <{e}>", self.info.name, self.info.fd.data(), self.info.total))
}
fn call_dtor(&mut self, shared: &mut[u8]) {
thread::yield_now();
if self.info.fd.is_invalid() {
return
}
self.info.level = usize::MAX;
self.info.finished = true;
if let Ok(_) = bananaq::update(&self.info) {
self.dtor.do_call(&self.info.bananaq, self.info.fd.data(), shared);
}
}
/// update slopes - state of current State, that we can proceed to fuzz next layer of syscalls
fn do_fuzz_update_impl(&mut self) -> Result<(), String> {
while usize::MAX == self.info.level { println!("DTOR CALLED UPDATE CALLED TOO!!") }
assert!(usize::MAX != self.info.level);
let mut call = &mut self.groups[self.ccache.0][self.ccache.1];
let ok = if call.ok() { 1 } else { 0 };
self.ccache.0 = (self.ccache.0 as isize + self.slopes[self.ccache.0][ok]) as usize;
self.ccache.1 = !0;//invalidate!! - now self.{c/m}call() is pretty much invalid!
let level = self.info.level;
self.info.sucess += ok;
self.info.level = self.ccache.0;
if 0 != level {//write locked callback
bananaq::call_aftermath(&mut self.info, &mut call)?;
} // for ctors we have notify_ctor at core/banana/looper.rs
if self.info.total > self.limit {
return Err(format!("fuzzed over limit : total:{:?} vs limit:{:?}",
self.info.total, self.limit))
}
if !call.dead(self.dead_ratio) {
return Ok(())
}
let dead_calls = self.groups[self.ccache.0]
.iter()
.filter(|&call| call.dead(self.dead_ratio))
.map(|call| format!("<{}>", call.name()))
.collect::<Vec<String>>();
if dead_calls.len() * 2 < self.groups[self.ccache.0].len() {
return Ok(())
}
Err(format!("more thatn 1/2 of group #{:?} call are dead; with current calls {:?}",
self.ccache.0, dead_calls.join("")))
}
/// TODO :
///
/// here is how we will build our state by default :
///
/// # Example
/// ```
/// struct TestState {
/// state: State,
/// }
///
/// impl TestState {
/// pub fn new() -> TestState {
/// TestState {
/// state : State::new(
/// "test-state",
/// StateTableId::Id(2),
/// 40,
/// vec![[1, 0], [-1, -1]],
/// vec![
/// vec![
/// Call::test_call()
/// ],
/// vec![
/// Call::test_call(),
/// Call::test_call()
/// ]],
/// Call::test_call())
/// }
/// }
/// }
/// ```
/// later we want to manualy play with fuzzing at IFuzzyObj::fuzzy_loop
///
/// - that will be our main point how to improve fuzzing by knowledge
/// - basically we can set pre-callback in modules, and post-callback there
/// - have unique state informations gathered
/// - create arbitrary ( but connected ) objects at certain points of fuzzing
/// - setting breakpoints, and optionally lead fuzzing
/// - leading of fuzzing maybe needed update bit core of this :
/// - ability to set self.ccache.1 and self.ccache.0 by will - for now i dont see a point
/// - ability to update self.call_view()->args()[..]->load_unsafe() data - also for now seems more over-engineering
/// - prefered dont touch arguments, and leave that for their generators
/// - by konwledge based approach do all work before+afer syscall is invoked, more
/// logic level leading
pub fn new(
bananaq: Weak<FuzzyQ>,
name: &'static str,
id: StateTableId,
fd_size: usize,
limit: usize,
slopes: Vec<[isize; 2]>,
groups: Vec< Vec<Call> >,
dtor: Call
) -> State
{
assert!(slopes.len() == groups.len());
assert!(groups.iter().all(|ref group| group.len() < 0x200));
if slopes.len() != groups.len() {
panic!("slopes vs groups len problem at : {} => {} vs {}", name, slopes.len(), groups.len());
}
if !groups.iter().all(|ref group| group.len() < 0x200) {
panic!("one of the group for {} is oversized!", name);
}
let fzzconfig = bananaq::config(&bananaq).unwrap();
State {
info : StateInfo {
bananaq : bananaq,
name : String::from(name),
total : 0,
sucess : 0,
fd : Fd::empty(fd_size),
id : id,
level : 0,
finished : false,
},
dead_ratio : fzzconfig.dead_call,
limit : min(fzzconfig.new_limit, limit),
slopes : slopes,
groups : groups,
dtor: dtor,
ccache : (0, !0),
}
}
/// apply as for new, but here we create already existing object :
///
/// - introduce race conditions
/// - some calls return more than one FD, so one of them we need create State by this way
/// - meaning : some calls create multiple states in target space / kernel / ..
/// - not necessary FD, sometimes can mean other than file descriptor only, but any runtime unique ID
/// - id of allocation, id of font in global table, crc32(string), runtime memory pointer, of some name uuid ? ...
pub fn duped(
bananaq: Weak<FuzzyQ>,
name: &'static str,
id: StateTableId,
fd: &Fd,
limit: usize,
slopes: Vec<[isize; 2]>,
groups: Vec< Vec<Call> >,
dtor: Call
) -> State
{
assert!(slopes.len() == groups.len());
assert!(groups.iter().all(|ref group| group.len() < 0x200));
if slopes.len() != groups.len() {
panic!("slopes vs groups len problem at : {} => {} vs {}", name, slopes.len(), groups.len());
}
if !groups.iter().all(|ref group| group.len() < 0x200) {
panic!("one of the group for {} is oversized!", name);
}
let fzzconfig = bananaq::config(&bananaq).unwrap();
let level = slopes[0][1] as usize;
State {
info : StateInfo {
bananaq : bananaq,
name : String::from(name),
total : 0,
sucess : 0,
fd : fd.clone(),
id : id,
level : level,
finished : false,
},
dead_ratio : fzzconfig.dead_call,
limit : min(fzzconfig.dup_limit, limit),
slopes : slopes,
groups : groups,
dtor: dtor,
ccache : (level, !0),
}
}
}