Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Introspection #58

Merged
merged 24 commits into from
May 7, 2021
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
24 commits
Select commit Hold shift + click to select a range
12b9584
Rework to put `ClientPerfStats` in `State` and pass that along. Still…
ctfhacker Apr 7, 2021
c017d72
Add perf_stats feature to libafl/Cargo.toml
ctfhacker Apr 7, 2021
23061ef
Update feedbacks to have with_perf
ctfhacker Apr 12, 2021
9469776
Merge branch 'real_time_benchmarks' of github.com:AFLplusplus/LibAFL …
ctfhacker Apr 12, 2021
542738c
Merge branch 'dev' into real_time_benchmarks
ctfhacker Apr 18, 2021
fedb5f8
Remove unneeeded print statement
ctfhacker Apr 18, 2021
7423cef
cargo fmt all the things
ctfhacker Apr 18, 2021
4ad9c10
use local llvmint vs cpu specific asm for reading cycle counter
ctfhacker Apr 19, 2021
b9a0c7c
Remove debug testing code
ctfhacker Apr 19, 2021
2a96fe3
Stats timeout to 3 seconds
ctfhacker Apr 19, 2021
55703cb
Inline smallish functions for ClientPerfStats
ctfhacker Apr 19, 2021
042bfce
Remove .libs/llvmint and have the correct conditional compilation of …
ctfhacker Apr 19, 2021
2b6b369
pub(crate) the NUM_FEEDBACK and NUM_STAGES consts
ctfhacker Apr 19, 2021
8219da3
Merge branch 'dev' into real_time_benchmarks
domenukk Apr 26, 2021
33df973
perf_stats->instrospection; merged dev; moved back to x86 stable
domenukk May 1, 2021
b9e75c0
Tcp Broker to Broker Communication (#66)
domenukk May 1, 2021
3fb4626
merged dev
domenukk May 1, 2021
0ac407a
adapted to new feedback interface;;
domenukk May 6, 2021
6cc1e96
clippy fixes
domenukk May 6, 2021
1fa893e
fallback to systemtime on non-x86
domenukk May 6, 2021
5d1eb27
make clippy more strict
domenukk May 7, 2021
f56f340
small fixes
andreafioraldi May 7, 2021
1e61748
bump 0.2.1
andreafioraldi May 7, 2021
92116b9
readme
andreafioraldi May 7, 2021
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 0 additions & 6 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -42,11 +42,6 @@ Rust directly, instructions can be found [here](https://www.rust-lang.org/tools/
git clone https://github.com/AFLplusplus/LibAFL
```

If you want to get the latest and greatest features,
```
git checkout dev
```

Build the library using

```
Expand All @@ -65,7 +60,6 @@ cargo doc
cd docs && mdbook serve
```


We collect all example fuzzers in [`./fuzzers`](./fuzzers/).
Be sure to read their documentation (and source), this is *the natural way to get started!*

Expand Down
4 changes: 0 additions & 4 deletions clippy.sh
Original file line number Diff line number Diff line change
Expand Up @@ -3,16 +3,12 @@
cargo clean -p libafl
RUST_BACKTRACE=full cargo clippy --all --all-features --tests -- \
-D clippy::pedantic \
-W clippy::cast_sign_loss \
-W clippy::similar-names \
-W clippy::cast_ptr_alignment \
-W clippy::cast_possible_wrap \
-W clippy::unused_self \
-W clippy::too_many_lines \
-W clippy::option_if_let_else \
-W clippy::must-use-candidate \
-W clippy::if-not-else \
-W clippy::doc-markdown \
-A clippy::type_repetition_in_bounds \
-A clippy::missing-errors-doc \
-A clippy::cast-possible-truncation \
Expand Down
3 changes: 2 additions & 1 deletion fuzzers/libfuzzer_libpng/Cargo.toml
Original file line number Diff line number Diff line change
Expand Up @@ -20,7 +20,8 @@ which = { version = "4.0.2" }
num_cpus = "1.0"

[dependencies]
libafl = { path = "../../libafl/" }
libafl = { path = "../../libafl/", features = ["default", "introspection"] }
# libafl = { path = "../../libafl/", features = ["default"] }
libafl_targets = { path = "../../libafl_targets/", features = ["pcguard_hitcounts", "libfuzzer"] }
# TODO Include it only when building cc
libafl_cc = { path = "../../libafl_cc/" }
Expand Down
63 changes: 63 additions & 0 deletions fuzzers/libfuzzer_libpng/Makefile
Original file line number Diff line number Diff line change
@@ -0,0 +1,63 @@
PWD=`pwd`
FUZZER_NAME="fuzzer_libpng"

all:
# Build the libpng libfuzzer library
cargo build --release

# Build the libpng harness
$(PWD)/target/release/libafl_cxx \
$(PWD)/harness.cc \
$(PWD)/libpng-1.6.37/.libs/libpng16.a \
-I$(PWD)/libpng-1.6.37/ \
-o $(FUZZER_NAME) \
-lm -lz

run: all
./$(FUZZER_NAME) &
sleep 0.2
./$(FUZZER_NAME) >/dev/null 2>/dev/null &

short_test: all
timeout 11s ./$(FUZZER_NAME) &
sleep 0.2
timeout 10s taskset -c 0 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
timeout 10s taskset -c 1 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
timeout 10s taskset -c 2 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
timeout 10s taskset -c 3 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &

test: all
timeout 60s ./$(FUZZER_NAME) &
sleep 0.2
timeout 59s taskset 0x00000001 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
timeout 59s taskset 0x00000002 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
timeout 59s taskset 0x00000004 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
timeout 59s taskset 0x00000008 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
# timeout 59s taskset 0x00000010 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
# timeout 59s taskset 0x00000020 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
# timeout 59s taskset 0x00000040 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
# timeout 59s taskset 0x00000080 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
# timeout 59s taskset 0x00000100 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
# timeout 59s taskset 0x00000200 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
# timeout 59s taskset 0x00000400 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
# timeout 59s taskset 0x00000800 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
# timeout 59s taskset 0x00001000 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
# timeout 59s taskset 0x00002000 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
# timeout 59s taskset 0x00004000 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
# timeout 59s taskset 0x00008000 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
# timeout 59s taskset 0x00010000 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
# timeout 59s taskset 0x00020000 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
# timeout 59s taskset 0x00040000 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
# timeout 59s taskset 0x00080000 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
# timeout 59s taskset 0x00100000 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
# timeout 59s taskset 0x00200000 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
# timeout 59s taskset 0x00400000 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
# timeout 59s taskset 0x00800000 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
# timeout 59s taskset 0x01000000 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
# timeout 59s taskset 0x02000000 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
# timeout 59s taskset 0x04000000 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
# timeout 59s taskset 0x08000000 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
# timeout 59s taskset 0x10000000 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
# timeout 59s taskset 0x20000000 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
# timeout 59s taskset 0x40000000 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
# timeout 59s taskset 0x80000000 ./$(FUZZER_NAME) >/dev/null 2>/dev/null &
4 changes: 3 additions & 1 deletion fuzzers/libfuzzer_libpng/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -6,6 +6,8 @@ In contrast to other fuzzer examples, this setup uses `fuzz_loop_for`, to occasi
While this costs performance, it can be useful for targets with memory leaks or other instabilities.
If your target is really instable, however, consider exchanging the `InProcessExecutor` for a `ForkserverExecutor` instead.

It also uses the `introspection` feature, printing fuzzer stats during execution.

To show off crash detection, we added a `ud2` instruction to the harness, edit harness.cc if you want a non-crashing example.
It has been tested on Linux.

Expand Down Expand Up @@ -51,7 +53,7 @@ This allows you to run multiple different builds of the same fuzzer alongside, f

## Run

The first time you run the binary, the broker will open a tcp port (currently on port `1337`), waiting for fuzzer clients to connect. This port is local and only used for the initial handshake. All further communication happens via shared map, to be independent of the kernel. Currently you must run the clients from the libfuzzer_libpng directory for them to be able to access the PNG corpus.
The first time you run the binary, the broker will open a tcp port (currently on port `1337`), waiting for fuzzer clients to connect. This port is local and only used for the initial handshake. All further communication happens via shared map, to be independent of the kernel. Currently, you must run the clients from the libfuzzer_libpng directory for them to be able to access the PNG corpus.

```
./fuzzer_libpng
Expand Down
2 changes: 1 addition & 1 deletion fuzzers/libfuzzer_libpng/src/lib.rs
Original file line number Diff line number Diff line change
Expand Up @@ -63,7 +63,7 @@ fn fuzz(corpus_dirs: Vec<PathBuf>, objective_dir: PathBuf, broker_port: u16) ->
}
},
};

// Create an observation channel using the coverage map
let edges = unsafe { &mut EDGES_MAP[0..MAX_EDGES_NUM] };
let edges_observer = HitcountsMapObserver::new(StdMapObserver::new("edges", edges));
Expand Down
11 changes: 6 additions & 5 deletions libafl/Cargo.toml
Original file line number Diff line number Diff line change
@@ -1,6 +1,6 @@
[package]
name = "libafl"
version = "0.2.0"
version = "0.2.1"
authors = ["Andrea Fioraldi <andreafioraldi@gmail.com>", "Dominik Maier <domenukk@gmail.com>"]
description = "Slot your own fuzzers together and extend their features using Rust"
documentation = "https://docs.rs/libafl"
Expand Down Expand Up @@ -40,10 +40,11 @@ default = ["std", "anymap_debug", "derive", "llmp_compression"]
std = [] # print, sharedmap, ... support
anymap_debug = ["serde_json"] # uses serde_json to Debug the anymap trait. Disable for smaller footprint.
derive = ["libafl_derive"] # provide derive(SerdeAny) macro.
llmp_small_maps = [] # reduces initial map size for llmp
llmp_debug = ["backtrace"] # Enables debug output for LLMP
llmp_compression = [] # llmp compression using GZip
llmp_bind_public = [] # If set, llmp will bind to 0.0.0.0, allowing cross-device communication. Binds to localhost by default.
llmp_compression = [] # llmp compression using GZip
llmp_debug = ["backtrace"] # Enables debug output for LLMP
llmp_small_maps = [] # reduces initial map size for llmp
introspection = [] # Include performance statistics of the fuzzing pipeline

[[example]]
name = "llmp_test"
Expand All @@ -60,7 +61,7 @@ erased-serde = "0.3.12"
postcard = { version = "0.5.1", features = ["alloc"] } # no_std compatible serde serialization fromat
static_assertions = "1.1.0"
ctor = "0.1.20"
libafl_derive = { version = "0.1.0", optional = true, path = "../libafl_derive" }
libafl_derive = { optional = true, path = "../libafl_derive", version = "0.2.1" }
serde_json = { version = "1.0", optional = true, default-features = false, features = ["alloc"] } # an easy way to debug print SerdeAnyMap
compression = { version = "0.1.5" }
num_enum = "0.5.1"
Expand Down
2 changes: 2 additions & 0 deletions libafl/src/bolts/llmp.rs
Original file line number Diff line number Diff line change
Expand Up @@ -55,6 +55,8 @@ For broker2broker communication, all messages are forwarded via network sockets.

Check out the `llmp_test` example in ./examples, or build it with `cargo run --example llmp_test`.

For broker2broker communication, all messages are forwarded via network sockets.

*/

use alloc::{string::String, vec::Vec};
Expand Down
31 changes: 31 additions & 0 deletions libafl/src/cpu.rs
Original file line number Diff line number Diff line change
@@ -0,0 +1,31 @@
//! Architecture agnostic processor features

#[cfg(not(any(target_arch = "x86_64", target_arch = "x86")))]
use crate::utils::current_nanos;

// TODO: Add more architectures, using C code, see
// https://github.com/google/benchmark/blob/master/src/cycleclock.h
// Or using llvm intrinsics (if they ever should become available in stable rust?)

/// Read a timestamp for measurements.
///
/// This function is a wrapper around different ways to get a timestamp, fast
/// In this way, an experiment only has to
/// change this implementation rather than every instead of [`cpu::read_time_counter`]
/// It is using [`rdtsc`] on `x86_64` and `x86`.
#[cfg(any(target_arch = "x86_64", target_arch = "x86"))]
#[must_use]
pub fn read_time_counter() -> u64 {
unsafe { core::arch::x86_64::_rdtsc() }
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Might be worth using https://docs.rs/llvmint/0.0.3/llvmint/fn.readcyclecounter.html, as that will automatically do the right thing for each arch.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sounds really useful for more obscure platforms!
We should only add the dependency for this feature flag though (and maybe even only for the platforms we can not reach otherwise?)

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Oh sure! Didn't know about the llvm intrinsic. If we're okay with the llvmint dependency then I'm fine with using it.

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The original thought was to add those specific instructions similar to Google's benchmark: https://github.com/google/benchmark/blob/master/src/cycleclock.h

I think that would mitigate the external dependency just for one instruction per arch.

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

But that would require asm... which is not stable.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

let's go with

  • libc, if available for this arch
  • llvmint dependency otherwise
    ?

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Or since it's a feature flag anyway, just llvmint is fine too, if it's too much work otherwise

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Just pushed a fix using llvmint::readcyclecounter. Looks like the feature needed requires nightly. I think you guys are wanting to stay on stable as far as I'm aware, which would mean neither this or the asm! fix would work. I guess we could also require nightly for the perf_stats feature as well.

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Edit from the last comment.. i had the wrong cfg_attr. I've ripped out only the needed extern from llvmint so we don't need to add that dependency directly to src/cpu.rs and it seems to work fine now.

}

/// Read a timestamp for measurements.
///
/// This function is a wrapper around different ways to get a timestamp, fast
/// In this way, an experiment only has to
/// change this implementation rather than every instead of [`cpu::read_time_counter`]
/// On unsupported architectures, it's falling back to normal system time, in millis.
#[cfg(not(any(target_arch = "x86_64", target_arch = "x86")))]
pub fn read_time_counter() -> u64 {
current_nanos()
}
32 changes: 30 additions & 2 deletions libafl/src/events/llmp.rs
Original file line number Diff line number Diff line change
Expand Up @@ -230,7 +230,7 @@ where
let client = stats.client_stats_mut_for(sender_id);
client.update_corpus_size(*corpus_size as u64);
client.update_executions(*executions as u64, *time);
stats.display(event.name().to_string() + " #" + &sender_id.to_string());
// stats.display(event.name().to_string() + " #" + &sender_id.to_string());
Copy link
Member

@andreafioraldi andreafioraldi May 20, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@domenukk porcodio, now we have to release again because this line was commented

Ok(BrokerEventResult::Forward)
}
Event::UpdateStats {
Expand All @@ -241,7 +241,35 @@ where
// TODO: The stats buffer should be added on client add.
let client = stats.client_stats_mut_for(sender_id);
client.update_executions(*executions as u64, *time);
stats.display(event.name().to_string() + " #" + &sender_id.to_string());
if sender_id == 1 {
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Why this?

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Shouldn't be needed, I agree

stats.display(event.name().to_string() + " #" + &sender_id.to_string());
}
Ok(BrokerEventResult::Handled)
}
#[cfg(feature = "introspection")]
Event::UpdatePerfStats {
time,
executions,
introspection_stats,
phantom: _,
} => {
// TODO: The stats buffer should be added on client add.

// Get the client for the sender ID
let client = stats.client_stats_mut_for(sender_id);

// Update the normal stats for this client
client.update_executions(*executions as u64, *time);

// Update the performance stats for this client
client.update_introspection_stats(**introspection_stats);

// Display the stats via `.display` only on core #1
if sender_id == 1 {
stats.display(event.name().to_string() + " #" + &sender_id.to_string());
}

// Correctly handled the event
Ok(BrokerEventResult::Handled)
}
Event::Objective { objective_size } => {
Expand Down
24 changes: 24 additions & 0 deletions libafl/src/events/mod.rs
Original file line number Diff line number Diff line change
Expand Up @@ -17,6 +17,9 @@ use crate::{
Error,
};

#[cfg(feature = "introspection")]
use crate::stats::ClientPerfStats;

/// The log event severity
#[derive(Serialize, Deserialize, Debug, Clone, Copy)]
pub enum LogSeverity {
Expand Down Expand Up @@ -98,6 +101,20 @@ where
/// [`PhantomData`]
phantom: PhantomData<I>,
},
/// New stats with performance stats.
#[cfg(feature = "introspection")]
UpdatePerfStats {
/// The time of generation of the event
time: Duration,

/// The executions of this client
executions: usize,

/// Current performance statistics
introspection_stats: Box<ClientPerfStats>,

phantom: PhantomData<I>,
},
/// A new objective was found
Objective {
/// Objective corpus size
Expand Down Expand Up @@ -138,6 +155,13 @@ where
executions: _,
phantom: _,
} => "Stats",
#[cfg(feature = "introspection")]
Event::UpdatePerfStats {
time: _,
executions: _,
introspection_stats: _,
phantom: _,
} => "PerfStats",
Event::Objective { objective_size: _ } => "Objective",
Event::Log {
severity_level: _,
Expand Down
13 changes: 13 additions & 0 deletions libafl/src/events/simple.rs
Original file line number Diff line number Diff line change
Expand Up @@ -106,6 +106,19 @@ where
stats.display(event.name().to_string());
Ok(BrokerEventResult::Handled)
}
#[cfg(feature = "introspection")]
Event::UpdatePerfStats {
time,
executions,
introspection_stats,
phantom: _,
} => {
// TODO: The stats buffer should be added on client add.
stats.client_stats_mut()[0].update_executions(*executions as u64, *time);
stats.client_stats_mut()[0].update_introspection_stats(**introspection_stats);
stats.display(event.name().to_string());
Ok(BrokerEventResult::Handled)
}
Event::Objective { objective_size } => {
stats
.client_stats_mut_for(0)
Expand Down
Loading