|
| 1 | +# Announcing memflow 0.2.0 |
| 2 | + |
| 3 | +Today, we are proud to release the first stable version of memflow 0.2! 3 years in the making, this |
| 4 | +is certainly a monumental release. In this post, we will go through the key changes to the fastest |
| 5 | +and most flexible physical memory introspection and forensics framework to date. |
| 6 | + |
| 7 | +## Key changes |
| 8 | + |
| 9 | +### 0. [memflowup](https://github.com/memflow/memflowup) |
| 10 | + |
| 11 | +Not a library change, but the ecosystem change! We now have a rust-written memflowup utility that |
| 12 | +makes it much easier to manage your memflow installation. Key features: |
| 13 | + |
| 14 | +- Download binary builds (optional). |
| 15 | +- Split between stable and dev versions. |
| 16 | +- Custom install scripts, for more complicated plugins |
| 17 | + - Used by [`memflow-kvm`](https://github.com/memflow/memflow-kvm) for DKMS install. |
| 18 | + - Entry point for these is `install.rhai` script at the root of the package's repo. |
| 19 | + |
| 20 | +You can get started with memflowup by running the following: |
| 21 | + |
| 22 | +``` |
| 23 | +> curl --proto '=https' --tlsv1.2 -sSf https://sh.memflow.io | sh |
| 24 | +``` |
| 25 | + |
| 26 | +### 1. OS layers and modularity |
| 27 | + |
| 28 | +With the advent of 0.2 series, we now abstracted most of `memflow-win32` functionality behind |
| 29 | +shared set of traits. These traits allow the user to interact with the operating system in unified |
| 30 | +manner. In addition, we now made OS a plugin, just as Connectors were in 0.1! And finally, we do |
| 31 | +indeed have multiple OS backends available, right now: |
| 32 | + |
| 33 | +- [`memflow-win32`](https://github.com/memflow/memflow-win32), for Windows analysis, given physical |
| 34 | + memory access. |
| 35 | +- [`memflow-native`](https://github.com/memflow/memflow-native), for syscall based interaction with |
| 36 | + the running operating system. |
| 37 | +- WIP: `memflow-linux` |
| 38 | + - Don't expect much anytime soon, because the challenge of cross-version, zero-knowledge linux |
| 39 | + support is a tricky one. |
| 40 | + |
| 41 | +With this, OS-independent code that works with `memflow-win32`, should also work on local OS. Here's |
| 42 | +an example of such code: |
| 43 | + |
| 44 | +```rust |
| 45 | +use memflow::prelude::v1::*; |
| 46 | + |
| 47 | +// We don't care what type of process we get, so long as it's a process |
| 48 | +fn module_address(process: &mut impl Process, module: &str) -> Result<Address> { |
| 49 | + let module = process.module_by_name(module)?; |
| 50 | + Ok(module.base) |
| 51 | +} |
| 52 | +``` |
| 53 | + |
| 54 | +In addition, modularization of operating systems allows for greater portability of connectors. For |
| 55 | +instance, we have now split `memflow-qemu-procfs` into |
| 56 | +[`memflow-qemu`](https://github.com/memflow/memflow-qemu), which (optionally) accepts an OS layer. |
| 57 | +This way, you can not only analyze QEMU VMs running on your computer, but you can also open them up |
| 58 | +in a nested way on a machine that is already being analyzed through DMA. As seen in this chart: |
| 59 | + |
| 60 | +```asciirend |
| 61 | +dynamic_w = true |
| 62 | +dynamic_h = false |
| 63 | +fov = 4.5 |
| 64 | +ortho = true |
| 65 | +disable_zoom = true |
| 66 | +# Scene: |
| 67 | +{ |
| 68 | + "camera_props": { |
| 69 | + "proj_mode": "Orthographic", |
| 70 | + "fov": 1.0, |
| 71 | + "near": 0.01, |
| 72 | + "far": 100.0 |
| 73 | + }, |
| 74 | + "camera_controller":{"fov_y":1.0,"focus_point":[0.0,0.0,0.0],"rot":[-0.19996414, -0.08282786, 0.37361234, 0.90197986],"dist":2.0,"in_motion":"None","scroll_sensitivity":0.02,"orbit_sensitivity":1.0,"last_down":false,"pressed":false}, |
| 75 | + "objects":[ |
| 76 | + { |
| 77 | + "transform":[ |
| 78 | + 1.0,0.0,0.0,0.0, |
| 79 | + 0.0,1.0,0.0,0.0, |
| 80 | + 0.0,0.0,1.0,0.0, |
| 81 | + 0.0,0.0,0.0,1.0 |
| 82 | + ], |
| 83 | + "material":0, |
| 84 | + "ty":{ |
| 85 | + "Primitive":{ |
| 86 | + "Line":{ |
| 87 | + "start":[-0.75,0.0,1.5,1.0], |
| 88 | + "end":[-0.25,0.0,0.5,1.0] |
| 89 | + } |
| 90 | + } |
| 91 | + } |
| 92 | + }, |
| 93 | + { |
| 94 | + "transform":[ |
| 95 | + 1.0,0.0,0.0,0.0, |
| 96 | + 0.0,1.0,0.0,0.0, |
| 97 | + 0.0,0.0,1.0,0.0, |
| 98 | + 0.0,0.0,0.0,1.0 |
| 99 | + ], |
| 100 | + "material":0, |
| 101 | + "ty":{ |
| 102 | + "Primitive":{ |
| 103 | + "Line":{ |
| 104 | + "start":[-0.25,0.0,0.5,1.0], |
| 105 | + "end":[0.25,0.0,-0.5,1.0] |
| 106 | + } |
| 107 | + } |
| 108 | + } |
| 109 | + }, |
| 110 | + { |
| 111 | + "transform":[ |
| 112 | + 1.0,0.0,0.0,0.0, |
| 113 | + 0.0,1.0,0.0,0.0, |
| 114 | + 0.0,0.0,1.0,0.0, |
| 115 | + 0.0,0.0,0.0,1.0 |
| 116 | + ], |
| 117 | + "material":0, |
| 118 | + "ty":{ |
| 119 | + "Primitive":{ |
| 120 | + "Line":{ |
| 121 | + "start":[0.25,0.0,-0.5,1.0], |
| 122 | + "end":[0.75,0.0,-1.5,1.0] |
| 123 | + } |
| 124 | + } |
| 125 | + } |
| 126 | + }, |
| 127 | + { |
| 128 | + "transform":[ |
| 129 | + 1.0,0.0,0.0,0.0, |
| 130 | + 0.0,1.0,0.0,0.0, |
| 131 | + 0.0,0.0,1.0,0.0, |
| 132 | + -0.75,0.0,1.5,1.0 |
| 133 | + ],"material":1,"ty":{"Cube":{"size":[1.0,1.0,0.5]}},"text":"memflow-kvm" |
| 134 | + }, |
| 135 | + { |
| 136 | + "transform":[ |
| 137 | + 1.0,0.0,0.0,0.0, |
| 138 | + 0.0,1.0,0.0,0.0, |
| 139 | + 0.0,0.0,1.0,0.0, |
| 140 | + -0.25,0.0,0.5,1.0 |
| 141 | + ],"material":1,"ty":{"Cube":{"size":[1.0,1.0,0.5]}},"text":"memflow-win32" |
| 142 | + }, |
| 143 | + { |
| 144 | + "transform":[ |
| 145 | + 1.0,0.0,0.0,0.0, |
| 146 | + 0.0,1.0,0.0,0.0, |
| 147 | + 0.0,0.0,1.0,0.0, |
| 148 | + 0.25,0.0,-0.5,1.0 |
| 149 | + ],"material":1,"ty":{"Cube":{"size":[1.0,1.0,0.5]}},"text":"memflow-qemu" |
| 150 | + }, |
| 151 | + { |
| 152 | + "transform":[ |
| 153 | + 1.0,0.0,0.0,0.0, |
| 154 | + 0.0,1.0,0.0,0.0, |
| 155 | + 0.0,0.0,1.0,0.0, |
| 156 | + 0.75,0.0,-1.5,1.0 |
| 157 | + ],"material":1,"ty":{"Cube":{"size":[1.0,1.0,0.5]}},"text":"memflow-win32" |
| 158 | + } |
| 159 | + ], |
| 160 | + "bg":{"color":[0.0,0.0,0.0]}, |
| 161 | + "dithering":{"count_frames":false,"frame_cnt":4181} |
| 162 | +} |
| 163 | +``` |
| 164 | + |
| 165 | +### 2. Stable ABI |
| 166 | + |
| 167 | +In 0.1, the Connectors were turned into plugins through use of Rust trait objects. This was an okay |
| 168 | +solution at the time, however, we knew that it was not a safe one - changes in Rust versions could |
| 169 | +change the layout of those trait objects, leading to crashes or other misbehavior, in case of |
| 170 | +mismatch of plugin's `rustc` version and the one of the user's code. While the layout has remained |
| 171 | +stable most of the time, the tides started to shift a few years ago, as more effort was put into |
| 172 | +trait objects on the compiler front. |
| 173 | + |
| 174 | +For 0.2, we knew we could not keep the status quo, so, we built `cglue`. The crate allows for |
| 175 | +simple and flexible ABI safe code generation, suited for the needs of `memflow`. Throughout the |
| 176 | +(very long) beta period, we received 0 crash reports stemming from ABI instability, while 0.1 had |
| 177 | +such cases. Therefore, we can conclude that it was a good investment that already made memflow more |
| 178 | +stable. |
| 179 | + |
| 180 | +In `0.2.0-betaX` series, you may have encountered "invalid ABI" errors, well, fear not, because in |
| 181 | +stable series, we commit to not breaking the ABI across entirety of `0.2` series, so this problem |
| 182 | +should be a thing of the past for most users. |
| 183 | + |
| 184 | +### 3. Memory constrained vtop |
| 185 | + |
| 186 | +memflow 0.2 introduces the most scalable virtual address translation backend, period. The backend |
| 187 | +is able to walk entire page tree in milliseconds, targeting any modern memory architecture (x86 and |
| 188 | +ARM support out-of-the box, sufficient building blocks for RISC-V). In addition, compared to 0.1, |
| 189 | +the new backend uses fixed-size buffers, meaning RAM usage will no longer blow up on large |
| 190 | +translation ranges. |
| 191 | + |
| 192 | +### 4. 64-bit and 128-bit address spaces, on all architectures |
| 193 | + |
| 194 | +We now support analyzing 64-bit operating systems on 32-bit machines. In addition, if there was a |
| 195 | +theoretical 128-bit architecture, we would support that as well. However, it's more of a PoC and we |
| 196 | +do not expect this to be needed in the foreseeable future. |
| 197 | + |
| 198 | +The support can be toggled through `64_bit_mem` (default) and `128_bit_mem` features. Do note that |
| 199 | +these feature toggles do change memflow's ABI and it should not be possible to mix the plugin |
| 200 | +features. |
| 201 | + |
| 202 | +### 5. Shared `MemoryView` |
| 203 | + |
| 204 | +In 0.1, we have had a split between physical and virtual memory. The reason for the split is |
| 205 | +caching - we wish minimize latency by caching read-only memory in high-latency scenarios. |
| 206 | +However, to tell the cache what mode the memory is in (readable/writeable/executable), you must add |
| 207 | +metadata with each request. Meanwhile, this metadata may only be filled in by the virtual address |
| 208 | +translation backend. |
| 209 | + |
| 210 | +If user submits an I/O operation - they can't possibly know whether the request is going to a |
| 211 | +read-only, or a writeable page, therefore they just submit `UNKNOWN` page flags. This is |
| 212 | +complicated, therefore, we have lowered the gap between virtual and physical memory access through |
| 213 | +use of `MemoryView` trait. This trait not only removes the need for the user to explicitly submit |
| 214 | +the page flags, but also brings all I/O helpers that existed in virtual memory contexts. To use |
| 215 | +`MemoryView` on physical memory, just use the `phys_view` function: |
| 216 | + |
| 217 | +```rust |
| 218 | +use memflow::prelude::v1::*; |
| 219 | + |
| 220 | +fn main() -> Result<()> { |
| 221 | + let inventory = Inventory::scan(); |
| 222 | + let mut conn = inventory.create_connector("dummy", None, None)?; |
| 223 | + |
| 224 | + // Create a physical memory view |
| 225 | + let mut view = conn.phys_view(); |
| 226 | + |
| 227 | + // Read from phys addr 0 |
| 228 | + let value: u64 = view.read(0.into())?; |
| 229 | + |
| 230 | + Ok(()) |
| 231 | +} |
| 232 | +``` |
| 233 | + |
| 234 | +### 6. C and C++ are now first-class citizen |
| 235 | + |
| 236 | +The FFI is now automatically generated using `cbindgen` and `cglue-bindgen`. It may initially seem |
| 237 | +like a downgrade, however, this way we can ensure entirety of memflow's plugin-focused API surface |
| 238 | +can be both accessed, and implemented by foreign languages, such as C and C++. |
| 239 | + |
| 240 | +The key to using the new FFI, is reading Rust documentation and examples, and then finding the |
| 241 | +function equivalents in the headers. There are a few quirks here and there, but after understanding |
| 242 | +them, using the FFI should not be hard. For inspiration, see the following: |
| 243 | + |
| 244 | +- [C examples](https://github.com/memflow/memflow/tree/0.2.0/memflow-ffi/examples/c) |
| 245 | +- [C++ examples](https://github.com/memflow/memflow/tree/0.2.0/memflow-ffi/examples/cpp) |
| 246 | +- [CMake template](https://github.com/memflow/memflow-cmake-example) |
| 247 | + |
| 248 | +## Side projects |
| 249 | + |
| 250 | +`memflow` is as useful as the projects utilizing it. To get started with the new version faster, |
| 251 | +you may want to have a look at some of them. Here's the list of first-party releases: |
| 252 | + |
| 253 | +- [`reflow`](https://github.com/memflow/reflow) - execute code on top of virtual memory. |
| 254 | +- [`scanflow`](https://github.com/memflow/scanflow) - basic CheatEngine features in a command line |
| 255 | + interface. |
| 256 | +- [`cloudflow`](https://github.com/memflow/cloudflow) (WIP) - flexible filesystem based way to |
| 257 | + interact with memflow. |
| 258 | +- [`memflow-py`](https://github.com/memflow/memflow-py) - python bindings for memflow (courtesy of |
| 259 | + emesare). |
| 260 | + |
| 261 | +## Reflection |
| 262 | + |
| 263 | +0.2 took way longer than we originally anticipated. This is mostly due to changing living |
| 264 | +conditions and the fact that both ko1N and I are only working on the project in hobbyist capacity. |
| 265 | +In addition, we pushed for perfection from documentation and implementation front - a feat |
| 266 | +infeasible at the current point. We do believe memflow is the framework that is going to bring the |
| 267 | +most empowerement to users, however, there are still ways to go. |
| 268 | + |
| 269 | +## Next up - Async Metamorphosis |
| 270 | + |
| 271 | +Next, we will work towards integrating [`mfio`](https://github.com/memflow/mfio) into memflow, |
| 272 | +which will enable higher scalability and simplicity. The key change is going to be transition from |
| 273 | +synchronous to asynchronous API. There are still a lot of open questions regarding this, such as |
| 274 | +FFI handling, how much the individual pieces of memflow's code will have to change, and how |
| 275 | +multithreading needs to be handled. However, we are confident those questions are not impossible |
| 276 | +to solve. Once the metamorphosis is done, we can consider the structure of memflow done. What comes |
| 277 | +afterwards, is rapid feature development. It will definitely be an exciting time to be alive. So |
| 278 | +let's just get there, shall we? |
| 279 | + |
| 280 | +\- h33p |
0 commit comments