init show-object-layout using assemblage

Fix IDA 9.0 / idalib

.github/dependabot.yml

@@ -4,3 +4,6 @@ updates:
     directory: "/"
     schedule:
       interval: "weekly"
+    ignore:
+      - dependency-name: "*"
+        update-types: ["version-update:semver-patch"]

.github/mypy/mypy.ini

@@ -1,8 +1,5 @@
 [mypy]
 
-[mypy-tqdm.*]
-ignore_missing_imports = True
-
 [mypy-ruamel.*]
 ignore_missing_imports = True
 

.github/pyinstaller/pyinstaller.spec

@@ -2,7 +2,6 @@
 # Copyright (C) 2020 Mandiant, Inc. All Rights Reserved.
 import sys
 
-import wcwidth
 import capa.rules.cache
 
 from pathlib import Path
@@ -29,13 +28,6 @@ a = Analysis(
         ("../../rules", "rules"),
         ("../../sigs", "sigs"),
         ("../../cache", "cache"),
-        # capa.render.default uses tabulate that depends on wcwidth.
-        # it seems wcwidth uses a json file `version.json`
-        # and this doesn't get picked up by pyinstaller automatically.
-        # so we manually embed the wcwidth resources here.
-        #
-        # ref: https://stackoverflow.com/a/62278462/87207
-        (Path(wcwidth.__file__).parent, "wcwidth"),
     ],
     # when invoking pyinstaller from the project root,
     # this gets run from the project root.
@@ -48,11 +40,6 @@ a = Analysis(
         "tkinter",
         "_tkinter",
         "Tkinter",
-        # tqdm provides renderers for ipython,
-        # however, this drags in a lot of dependencies.
-        # since we don't spawn a notebook, we can safely remove these.
-        "IPython",
-        "ipywidgets",
         # these are pulled in by networkx
         # but we don't need to compute the strongly connected components.
         "numpy",
@@ -70,7 +57,10 @@ a = Analysis(
         "qt5",
         "pyqtwebengine",
         "pyasn1",
+        # don't pull in Binary Ninja/IDA bindings that should
+        # only be installed locally.
         "binaryninja",
+        "ida",
     ],
 )
 

.github/workflows/build.yml

@@ -30,8 +30,8 @@ jobs:
             python_version: 3.8
           - os: ubuntu-20.04
             artifact_name: capa
-            asset_name: linux-py311
+            asset_name: linux-py312
-            python_version: 3.11
+            python_version: 3.12
           - os: windows-2019
             artifact_name: capa.exe
             asset_name: windows
@@ -88,7 +88,7 @@ jobs:
             asset_name: linux
           - os: ubuntu-22.04
             artifact_name: capa
-            asset_name: linux-py311
+            asset_name: linux-py312
           - os: windows-2022
             artifact_name: capa.exe
             asset_name: windows
@@ -114,7 +114,7 @@ jobs:
         include:
           - asset_name: linux
             artifact_name: capa
-          - asset_name: linux-py311
+          - asset_name: linux-py312
             artifact_name: capa
           - asset_name: windows
             artifact_name: capa.exe

.github/workflows/web-deploy.yml

@@ -2,7 +2,7 @@ name: deploy web to GitHub Pages
 
 on:
   push:
-    branches: [ master, "wb/webui-actions-1" ]
+    branches: [ master ]
     paths:
       - 'web/**'
 
@@ -22,6 +22,7 @@ concurrency:
 
 jobs:
   build-landing-page:
+    name: Build landing page
     runs-on: ubuntu-latest
     steps:
       - name: Checkout
@@ -32,6 +33,7 @@ jobs:
           path: './web/public'
 
   build-explorer:
+    name: Build capa Explorer Web
     runs-on: ubuntu-latest
     steps:
       - name: Checkout
@@ -41,7 +43,7 @@ jobs:
           fetch-depth: 1
           show-progress: true
       - name: Set up Node
-        uses: actions/setup-node@v4
+        uses: actions/setup-node@0a44ba7841725637a19e28fa30b79a866c81b0a6 # v4.0.4
         with:
           node-version: 20
           cache: 'npm'
@@ -63,12 +65,51 @@ jobs:
           name: explorer
           path: './web/explorer/dist'
 
+  build-rules:
+    name: Build rules site
+    runs-on: ubuntu-latest
+    steps:
+      - name: Check out the repository
+        uses: actions/checkout@v4
+        with:
+          submodules: 'recursive'
+          # full depth so that capa-rules has a full history
+          # and we can construct a timeline of rule updates.
+          fetch-depth: 0
+      - name: Set up Python
+        uses: actions/setup-python@0a5c61591373683505ea898e09a3ea4f39ef2b9c # v5.0.0
+        with:
+          python-version: '3.12'
+      - uses: extractions/setup-just@v2
+      - name: Install pagefind
+        uses: supplypike/setup-bin@v4
+        with:
+          uri: "https://github.com/CloudCannon/pagefind/releases/download/v1.1.0/pagefind-v1.1.0-x86_64-unknown-linux-musl.tar.gz"
+          name: "pagefind"
+          version: "1.1.0"
+      - name: Install dependencies
+        working-directory: ./web/rules
+        run: pip install -r requirements.txt
+      - name: Build the website
+        working-directory: ./web/rules
+        run: just build
+      - name: Index the website
+        working-directory: ./web/rules
+        run: pagefind --site "public"
+      # upload the build website to artifacts
+      # so that we can download and inspect, if desired.
+      - uses: actions/upload-artifact@v4
+        with:
+          name: rules
+          path: './web/rules/public'
+
   deploy:
+    name: Deploy site to GitHub Pages
     environment:
       name: github-pages
       url: ${{ steps.deployment.outputs.page_url }}
     runs-on: ubuntu-latest
-    needs: [build-landing-page, build-explorer]
+    needs: [build-landing-page, build-explorer, build-rules]
     steps:
       - uses: actions/download-artifact@v4
         with:
@@ -78,6 +119,10 @@ jobs:
         with:
           name: explorer
           path: './public/explorer'
+      - uses: actions/download-artifact@v4
+        with:
+          name: rules
+          path: './public/rules'
       - name: Setup Pages
         uses: actions/configure-pages@v4
       - name: Upload artifact

.github/workflows/web-tests.yml

@@ -19,7 +19,7 @@ jobs:
         show-progress: true
 
     - name: Set up Node
-      uses: actions/setup-node@v3
+      uses: actions/setup-node@0a44ba7841725637a19e28fa30b79a866c81b0a6 # v4.0.4
       with:
         node-version: 20
         cache: 'npm'

.gitignore

@@ -126,3 +126,5 @@ Pipfile.lock
 .github/binja/binaryninja
 .github/binja/download_headless.py
 .github/binja/BinaryNinja-headless.zip
+justfile
+data/

.justfile

@@ -0,0 +1,25 @@
+@isort:
+    pre-commit run isort --show-diff-on-failure --all-files
+
+@black:
+    pre-commit run black --show-diff-on-failure --all-files
+
+@ruff:
+    pre-commit run ruff --all-files
+
+@flake8:
+    pre-commit run flake8 --hook-stage manual --all-files
+
+@mypy:
+    pre-commit run mypy --hook-stage manual --all-files
+
+@deptry:
+    pre-commit run deptry --hook-stage manual --all-files
+
+@lint:
+    -just isort
+    -just black
+    -just ruff
+    -just flake8
+    -just mypy
+    -just deptry

.pre-commit-config.yaml

@@ -38,6 +38,7 @@ repos:
         -   "capa/"
         -   "scripts/"
         -   "tests/"
+        -   "web/rules/scripts/"
         always_run: true
         pass_filenames: false
 
@@ -55,6 +56,7 @@ repos:
         -   "capa/"
         -   "scripts/"
         -   "tests/"
+        -   "web/rules/scripts/"
         always_run: true
         pass_filenames: false
 
@@ -72,6 +74,7 @@ repos:
         -   "capa/"
         -   "scripts/"
         -   "tests/"
+        -   "web/rules/scripts/"
         always_run: true
         pass_filenames: false
 
@@ -86,10 +89,11 @@ repos:
         -   "--config"
         -   ".github/flake8.ini"
         -   "--extend-exclude"
-        -   "capa/render/proto/capa_pb2.py"
+        -   "capa/render/proto/capa_pb2.py,capa/features/extractors/binexport2/binexport2_pb2.py"
         -   "capa/"
         -   "scripts/"
         -   "tests/"
+        -   "web/rules/scripts/"
         always_run: true
         pass_filenames: false
 
@@ -104,9 +108,11 @@ repos:
         -   "--check-untyped-defs"
         -   "--ignore-missing-imports"
         -   "--config-file=.github/mypy/mypy.ini"
+        -   "--enable-incomplete-feature=NewGenericSyntax"
         -   "capa/"
         -   "scripts/"
         -   "tests/"
+        -   "web/rules/scripts/"
         always_run: true
         pass_filenames: false
 

README.md

@@ -1,4 +1,16 @@
-![capa](https://github.com/mandiant/capa/blob/master/.github/logo.png)
+<br />
+<div align="center">
+<a href="https://mandiant.github.io/capa/" target="_blank">
+  <img src="https://github.com/mandiant/capa/blob/master/.github/logo.png">
+</a>
+<p align="center">
+  <a href="https://mandiant.github.io/capa/" target="_blank">Website</a>
+  |
+  <a href="https://github.com/mandiant/capa/releases/latest" target="_blank">Download</a>
+  |
+  <a href="https://mandiant.github.io/capa/explorer/" target="_blank">Web Interface</a>
+</p>
+<div align="center">
 
 [![PyPI - Python Version](https://img.shields.io/pypi/pyversions/flare-capa)](https://pypi.org/project/flare-capa)
 [![Last release](https://img.shields.io/github/v/release/mandiant/capa)](https://github.com/mandiant/capa/releases)
@@ -7,11 +19,16 @@
 [![Downloads](https://img.shields.io/github/downloads/mandiant/capa/total)](https://github.com/mandiant/capa/releases)
 [![License](https://img.shields.io/badge/license-Apache--2.0-green.svg)](LICENSE.txt)
 
+</div>
+</div>
+
+---
+
 capa detects capabilities in executable files.
 You run it against a PE, ELF, .NET module, shellcode file, or a sandbox report and it tells you what it thinks the program can do.
 For example, it might suggest that the file is a backdoor, is capable of installing services, or relies on HTTP to communicate.
 
-To interactively inspect capa results in your browser use the [capa web explorer](https://mandiant.github.io/capa/explorer/).
+To interactively inspect capa results in your browser use the [capa Explorer Web](https://mandiant.github.io/capa/explorer/).
 
 If you want to inspect or write capa rules, head on over to the [capa-rules repository](https://github.com/mandiant/capa-rules). Otherwise, keep reading.
 
@@ -72,12 +89,12 @@ Download stable releases of the standalone capa binaries [here](https://github.c
 
 To use capa as a library or integrate with another tool, see [doc/installation.md](https://github.com/mandiant/capa/blob/master/doc/installation.md) for further setup instructions.
 
-# web explorer
+# capa Explorer Web
-The [capa web explorer](https://mandiant.github.io/capa/explorer/) enables you to interactively explore capa results in your web browser. Besides the online version you can download a standalone HTML file for local offline usage.
+The [capa Explorer Web](https://mandiant.github.io/capa/explorer/) enables you to interactively explore capa results in your web browser. Besides the online version you can download a standalone HTML file for local offline usage.
 
-![capa web explorer screenshot](https://github.com/mandiant/capa/blob/master/doc/img/capa_web_explorer.png)
+![capa Explorer Web screenshot](https://github.com/mandiant/capa/blob/master/doc/img/capa_web_explorer.png)
 
-More details on the web UI is available in the [capa web explorer README](https://github.com/mandiant/capa/blob/master/web/explorer/README.md).
+More details on the web UI is available in the [capa Explorer Web README](https://github.com/mandiant/capa/blob/master/web/explorer/README.md).
 
 # example
 
@@ -133,13 +150,15 @@ function @ 0x4011C0
 ...
 ```
 
-## analyzing sandbox reports
+capa also supports dynamic capabilities detection for multiple sandboxes including:
-Additionally, capa also supports analyzing sandbox reports for dynamic capability extraction.
+* [CAPE](https://github.com/kevoreilly/CAPEv2) (supported report formats: `.json`, `.json_`, `.json.gz`)
-In order to use this, you first submit your sample to one of supported sandboxes for analysis, and then run capa against the generated report file.
+* [DRAKVUF](https://github.com/CERT-Polska/drakvuf-sandbox/) (supported report formats: `.log`, `.log.gz`)
+* [VMRay](https://www.vmray.com/) (supported report formats: analysis archive `.zip`)
 
-Currently, capa supports the [CAPE sandbox](https://github.com/kevoreilly/CAPEv2) and the [DRAKVUF sandbox](https://github.com/CERT-Polska/drakvuf-sandbox/). In order to use either, simply run capa against the generated file (JSON for CAPE or LOG for DRAKVUF sandbox) and it will automatically detect the sandbox and extract capabilities from it.
 
-Here's an example of running capa against a packed binary, and then running capa against the CAPE report of that binary:
+To use this feature, submit your file to a supported sandbox and then download and run capa against the generated report file. This feature enables capa to match capabilities against dynamic and static features that the sandbox captured during execution.
+
+Here's an example of running capa against a packed file, and then running capa against the CAPE report generated for the same packed file:
 
 ```yaml
 $ capa 05be49819139a3fdcdbddbdefd298398779521f3d68daa25275cc77508e42310.exe

capa/analysis/flirt.py

@@ -0,0 +1,38 @@
+# Copyright (C) 2024 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+
+from pydantic import BaseModel
+
+import capa.features.extractors.ida.idalib as idalib
+
+if not idalib.has_idalib():
+    raise RuntimeError("cannot find IDA idalib module.")
+
+if not idalib.load_idalib():
+    raise RuntimeError("failed to load IDA idalib module.")
+
+import idaapi
+import idautils
+
+
+class FunctionId(BaseModel):
+    va: int
+    is_library: bool
+    name: str
+
+
+def get_flirt_matches(lib_only=True):
+    for fva in idautils.Functions():
+        f = idaapi.get_func(fva)
+        is_lib = bool(f.flags & idaapi.FUNC_LIB)
+        fname = idaapi.get_func_name(fva)
+
+        if lib_only and not is_lib:
+            continue
+
+        yield FunctionId(va=fva, is_library=is_lib, name=fname)

capa/analysis/libraries.py

@@ -0,0 +1,240 @@
+# Copyright (C) 2024 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+import io
+import sys
+import logging
+import argparse
+import tempfile
+import contextlib
+from enum import Enum
+from typing import List, Optional
+from pathlib import Path
+
+import rich
+from pydantic import BaseModel
+from rich.text import Text
+from rich.console import Console
+
+import capa.main
+import capa.helpers
+import capa.analysis.flirt
+import capa.analysis.strings
+import capa.features.extractors.ida.idalib as idalib
+
+if not idalib.has_idalib():
+    raise RuntimeError("cannot find IDA idalib module.")
+
+if not idalib.load_idalib():
+    raise RuntimeError("failed to load IDA idalib module.")
+
+import idaapi
+import idapro
+import ida_auto
+import idautils
+
+logger = logging.getLogger(__name__)
+
+
+class Classification(str, Enum):
+    USER = "user"
+    LIBRARY = "library"
+    UNKNOWN = "unknown"
+
+
+class Method(str, Enum):
+    FLIRT = "flirt"
+    STRINGS = "strings"
+    THUNK = "thunk"
+    ENTRYPOINT = "entrypoint"
+
+
+class FunctionClassification(BaseModel):
+    va: int
+    classification: Classification
+    # name per the disassembler/analysis tool
+    # may be combined with the recovered/suspected name TODO below
+    name: str
+
+    # if is library, this must be provided
+    method: Optional[Method]
+
+    # TODO if is library, recovered/suspected name?
+
+    # if is library, these can optionally be provided.
+    library_name: Optional[str] = None
+    library_version: Optional[str] = None
+
+
+class FunctionIdResults(BaseModel):
+    function_classifications: List[FunctionClassification]
+
+
+@contextlib.contextmanager
+def ida_session(input_path: Path, use_temp_dir=True):
+    if use_temp_dir:
+        t = Path(tempfile.mkdtemp(prefix="ida-")) / input_path.name
+    else:
+        t = input_path
+
+    logger.debug("using %s", str(t))
+    # stderr=True is used here to redirect the spinner banner to stderr,
+    # so that users can redirect capa's output.
+    console = Console(stderr=True, quiet=False)
+
+    try:
+        if use_temp_dir:
+            t.write_bytes(input_path.read_bytes())
+
+        # idalib writes to stdout (ugh), so we have to capture that
+        # so as not to screw up structured output.
+        with capa.helpers.stdout_redirector(io.BytesIO()):
+            idapro.enable_console_messages(False)
+            with capa.main.timing("analyze program"):
+                with console.status("analyzing program...", spinner="dots"):
+                    if idapro.open_database(str(t.absolute()), run_auto_analysis=True):
+                        raise RuntimeError("failed to analyze input file")
+
+            logger.debug("idalib: waiting for analysis...")
+            ida_auto.auto_wait()
+            logger.debug("idalib: opened database.")
+
+        yield
+    finally:
+        idapro.close_database()
+        if use_temp_dir:
+            t.unlink()
+
+
+def is_thunk_function(fva):
+    f = idaapi.get_func(fva)
+    return bool(f.flags & idaapi.FUNC_THUNK)
+
+
+def main(argv=None):
+    if argv is None:
+        argv = sys.argv[1:]
+
+    parser = argparse.ArgumentParser(description="Identify library functions using various strategies.")
+    capa.main.install_common_args(parser, wanted={"input_file"})
+    parser.add_argument("--store-idb", action="store_true", default=False, help="store IDA database file")
+    parser.add_argument("--min-string-length", type=int, default=8, help="minimum string length")
+    parser.add_argument("-j", "--json", action="store_true", help="emit JSON instead of text")
+    args = parser.parse_args(args=argv)
+
+    try:
+        capa.main.handle_common_args(args)
+    except capa.main.ShouldExitError as e:
+        return e.status_code
+
+    dbs = capa.analysis.strings.get_default_databases()
+    capa.analysis.strings.prune_databases(dbs, n=args.min_string_length)
+
+    function_classifications: List[FunctionClassification] = []
+    with ida_session(args.input_file, use_temp_dir=not args.store_idb):
+        with capa.main.timing("FLIRT-based library identification"):
+            # TODO: add more signature (files)
+            # TOOD: apply more signatures
+            for flirt_match in capa.analysis.flirt.get_flirt_matches():
+                function_classifications.append(
+                    FunctionClassification(
+                        va=flirt_match.va,
+                        name=flirt_match.name,
+                        classification=Classification.LIBRARY,
+                        method=Method.FLIRT,
+                        # note: we cannot currently include which signature matched per function via the IDA API
+                    )
+                )
+
+        # thunks
+        for fva in idautils.Functions():
+            if is_thunk_function(fva):
+                function_classifications.append(
+                    FunctionClassification(
+                        va=fva,
+                        name=idaapi.get_func_name(fva),
+                        classification=Classification.LIBRARY,
+                        method=Method.THUNK,
+                    )
+                )
+
+        with capa.main.timing("string-based library identification"):
+            for string_match in capa.analysis.strings.get_string_matches(dbs):
+                function_classifications.append(
+                    FunctionClassification(
+                        va=string_match.va,
+                        name=idaapi.get_func_name(string_match.va),
+                        classification=Classification.LIBRARY,
+                        method=Method.STRINGS,
+                        library_name=string_match.metadata.library_name,
+                        library_version=string_match.metadata.library_version,
+                    )
+                )
+
+        for va in idautils.Functions():
+            name = idaapi.get_func_name(va)
+            if name not in {"WinMain", }:
+                continue
+
+            function_classifications.append(
+                FunctionClassification(
+                    va=va,
+                    name=name,
+                    classification=Classification.USER,
+                    method=Method.ENTRYPOINT,
+                )
+            )
+
+        doc = FunctionIdResults(function_classifications=[])
+        classifications_by_va = capa.analysis.strings.create_index(function_classifications, "va")
+        for va in idautils.Functions():
+            if classifications := classifications_by_va.get(va):
+                doc.function_classifications.extend(classifications)
+            else:
+                doc.function_classifications.append(
+                    FunctionClassification(
+                        va=va,
+                        name=idaapi.get_func_name(va),
+                        classification=Classification.UNKNOWN,
+                        method=None,
+                    )
+                )
+
+        if args.json:
+            print(doc.model_dump_json())  # noqa: T201 print found
+
+        else:
+            table = rich.table.Table()
+            table.add_column("FVA")
+            table.add_column("CLASSIFICATION")
+            table.add_column("METHOD")
+            table.add_column("FNAME")
+            table.add_column("EXTRA INFO")
+
+            classifications_by_va = capa.analysis.strings.create_index(doc.function_classifications, "va", sorted_=True)
+            for va, classifications in classifications_by_va.items():
+                name = ", ".join({c.name for c in classifications})
+                if "sub_" in name:
+                    name = Text(name, style="grey53")
+
+                classification = {c.classification for c in classifications}
+                method = {c.method for c in classifications if c.method}
+                extra = {f"{c.library_name}@{c.library_version}" for c in classifications if c.library_name}
+
+                table.add_row(
+                    hex(va),
+                    ", ".join(classification) if classification != {"unknown"} else Text("unknown", style="grey53"),
+                    ", ".join(method),
+                    name,
+                    ", ".join(extra),
+                )
+
+            rich.print(table)
+
+
+if __name__ == "__main__":
+    sys.exit(main())

capa/analysis/requirements.txt

@@ -0,0 +1,2 @@
+# temporary extra file to track dependencies of the analysis directory
+nltk==3.9.1

capa/analysis/strings/__init__.py

@@ -0,0 +1,269 @@
+# Copyright (C) 2024 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+
+"""
+further requirements:
+  - nltk
+"""
+import gzip
+import logging
+import collections
+from typing import Any, Dict, Mapping
+from pathlib import Path
+from dataclasses import dataclass
+
+import msgspec
+
+import capa.features.extractors.strings
+
+logger = logging.getLogger(__name__)
+
+
+class LibraryString(msgspec.Struct):
+    string: str
+    library_name: str
+    library_version: str
+    file_path: str | None = None
+    function_name: str | None = None
+    line_number: int | None = None
+
+
+@dataclass
+class LibraryStringDatabase:
+    metadata_by_string: Dict[str, LibraryString]
+
+    def __len__(self) -> int:
+        return len(self.metadata_by_string)
+
+    @classmethod
+    def from_file(cls, path: Path) -> "LibraryStringDatabase":
+        metadata_by_string: Dict[str, LibraryString] = {}
+        decoder = msgspec.json.Decoder(type=LibraryString)
+        for line in gzip.decompress(path.read_bytes()).split(b"\n"):
+            if not line:
+                continue
+            s = decoder.decode(line)
+            metadata_by_string[s.string] = s
+
+        return cls(metadata_by_string=metadata_by_string)
+
+
+DEFAULT_FILENAMES = (
+    "brotli.jsonl.gz",
+    "bzip2.jsonl.gz",
+    "cryptopp.jsonl.gz",
+    "curl.jsonl.gz",
+    "detours.jsonl.gz",
+    "jemalloc.jsonl.gz",
+    "jsoncpp.jsonl.gz",
+    "kcp.jsonl.gz",
+    "liblzma.jsonl.gz",
+    "libsodium.jsonl.gz",
+    "libpcap.jsonl.gz",
+    "mbedtls.jsonl.gz",
+    "openssl.jsonl.gz",
+    "sqlite3.jsonl.gz",
+    "tomcrypt.jsonl.gz",
+    "wolfssl.jsonl.gz",
+    "zlib.jsonl.gz",
+)
+
+DEFAULT_PATHS = tuple(Path(__file__).parent / "data" / "oss" / filename for filename in DEFAULT_FILENAMES) + (
+    Path(__file__).parent / "data" / "crt" / "msvc_v143.jsonl.gz",
+)
+
+
+def get_default_databases() -> list[LibraryStringDatabase]:
+    return [LibraryStringDatabase.from_file(path) for path in DEFAULT_PATHS]
+
+
+@dataclass
+class WindowsApiStringDatabase:
+    dll_names: set[str]
+    api_names: set[str]
+
+    def __len__(self) -> int:
+        return len(self.dll_names) + len(self.api_names)
+
+    @classmethod
+    def from_dir(cls, path: Path) -> "WindowsApiStringDatabase":
+        dll_names: set[str] = set()
+        api_names: set[str] = set()
+
+        for line in gzip.decompress((path / "dlls.txt.gz").read_bytes()).decode("utf-8").splitlines():
+            if not line:
+                continue
+            dll_names.add(line)
+
+        for line in gzip.decompress((path / "apis.txt.gz").read_bytes()).decode("utf-8").splitlines():
+            if not line:
+                continue
+            api_names.add(line)
+
+        return cls(dll_names=dll_names, api_names=api_names)
+
+    @classmethod
+    def from_defaults(cls) -> "WindowsApiStringDatabase":
+        return cls.from_dir(Path(__file__).parent / "data" / "winapi")
+
+
+def extract_strings(buf, n=4):
+    yield from capa.features.extractors.strings.extract_ascii_strings(buf, n=n)
+    yield from capa.features.extractors.strings.extract_unicode_strings(buf, n=n)
+
+
+def prune_databases(dbs: list[LibraryStringDatabase], n=8):
+    """remove less trustyworthy database entries.
+
+    such as:
+      - those found in multiple databases
+      - those that are English words
+      - those that are too short
+      - Windows API and DLL names
+    """
+
+    # TODO: consider applying these filters directly to the persisted databases, not at load time.
+
+    winapi = WindowsApiStringDatabase.from_defaults()
+
+    try:
+        from nltk.corpus import words as nltk_words
+
+        nltk_words.words()
+    except (ImportError, LookupError):
+        # one-time download of dataset.
+        # this probably doesn't work well for embedded use.
+        import nltk
+
+        nltk.download("words")
+        from nltk.corpus import words as nltk_words
+    words = set(nltk_words.words())
+
+    counter: collections.Counter[str] = collections.Counter()
+    to_remove = set()
+    for db in dbs:
+        for string in db.metadata_by_string.keys():
+            counter[string] += 1
+
+            if string in words:
+                to_remove.add(string)
+                continue
+
+            if len(string) < n:
+                to_remove.add(string)
+                continue
+
+            if string in winapi.api_names:
+                to_remove.add(string)
+                continue
+
+            if string in winapi.dll_names:
+                to_remove.add(string)
+                continue
+
+    for string, count in counter.most_common():
+        if count <= 1:
+            break
+
+        # remove strings that are seen in more than one database
+        to_remove.add(string)
+
+    for db in dbs:
+        for string in to_remove:
+            if string in db.metadata_by_string:
+                del db.metadata_by_string[string]
+
+
+def get_function_strings():
+    import idaapi
+    import idautils
+
+    import capa.features.extractors.ida.helpers as ida_helpers
+
+    strings_by_function = collections.defaultdict(set)
+    for ea in idautils.Functions():
+        f = idaapi.get_func(ea)
+
+        # ignore library functions and thunk functions as identified by IDA
+        if f.flags & idaapi.FUNC_THUNK:
+            continue
+        if f.flags & idaapi.FUNC_LIB:
+            continue
+
+        for bb in ida_helpers.get_function_blocks(f):
+            for insn in ida_helpers.get_instructions_in_range(bb.start_ea, bb.end_ea):
+                ref = capa.features.extractors.ida.helpers.find_data_reference_from_insn(insn)
+                if ref == insn.ea:
+                    continue
+
+                string = capa.features.extractors.ida.helpers.find_string_at(ref)
+                if not string:
+                    continue
+
+                strings_by_function[ea].add(string)
+
+    return strings_by_function
+
+
+@dataclass
+class LibraryStringClassification:
+    va: int
+    string: str
+    library_name: str
+    metadata: LibraryString
+
+
+def create_index(s: list, k: str, sorted_: bool = False) -> Mapping[Any, list]:
+    """create an index of the elements in `s` using the key `k`, optionally sorted by `k`"""
+    if sorted_:
+        s = sorted(s, key=lambda x: getattr(x, k))
+
+    s_by_k = collections.defaultdict(list)
+    for v in s:
+        p = getattr(v, k)
+        s_by_k[p].append(v)
+    return s_by_k
+
+
+def get_string_matches(dbs: list[LibraryStringDatabase]) -> list[LibraryStringClassification]:
+    matches: list[LibraryStringClassification] = []
+
+    for function, strings in sorted(get_function_strings().items()):
+        for string in strings:
+            for db in dbs:
+                if metadata := db.metadata_by_string.get(string):
+                    matches.append(
+                        LibraryStringClassification(
+                            va=function,
+                            string=string,
+                            library_name=metadata.library_name,
+                            metadata=metadata,
+                        )
+                    )
+
+    # if there are less than N strings per library, ignore that library
+    matches_by_library = create_index(matches, "library_name")
+    for library_name, library_matches in matches_by_library.items():
+        if len(library_matches) > 5:
+            continue
+
+        logger.info("pruning library %s: only %d matched string", library_name, len(library_matches))
+        matches = [m for m in matches if m.library_name != library_name]
+
+    # if there are conflicts within a single function, don't label it
+    matches_by_function = create_index(matches, "va")
+    for va, function_matches in matches_by_function.items():
+        library_names = {m.library_name for m in function_matches}
+        if len(library_names) == 1:
+            continue
+
+        logger.info("conflicting matches: 0x%x: %s", va, sorted(library_names))
+        # this is potentially slow (O(n**2)) but hopefully fast enough in practice.
+        matches = [m for m in matches if m.va != va]
+
+    return matches

capa/analysis/strings/__main__.py

@@ -0,0 +1,130 @@
+# Copyright (C) 2024 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+import sys
+import logging
+import collections
+from pathlib import Path
+
+import rich
+from rich.text import Text
+
+import capa.analysis.strings
+import capa.features.extractors.strings
+import capa.features.extractors.ida.helpers as ida_helpers
+
+logger = logging.getLogger(__name__)
+
+
+def open_ida(input_path: Path):
+    import tempfile
+
+    import idapro
+
+    t = Path(tempfile.mkdtemp(prefix="ida-")) / input_path.name
+    t.write_bytes(input_path.read_bytes())
+    # resource leak: we should delete this upon exit
+
+    idapro.enable_console_messages(False)
+    idapro.open_database(str(t.absolute()), run_auto_analysis=True)
+
+    import ida_auto
+
+    ida_auto.auto_wait()
+
+
+def main():
+    logging.basicConfig(level=logging.DEBUG)
+
+    # use n=8 to ignore common words
+    N = 8
+
+    input_path = Path(sys.argv[1])
+
+    dbs = capa.analysis.strings.get_default_databases()
+    capa.analysis.strings.prune_databases(dbs, n=N)
+
+    strings_by_library = collections.defaultdict(set)
+    for string in capa.analysis.strings.extract_strings(input_path.read_bytes(), n=N):
+        for db in dbs:
+            if metadata := db.metadata_by_string.get(string.s):
+                strings_by_library[metadata.library_name].add(string.s)
+
+    console = rich.get_console()
+    console.print("found libraries:", style="bold")
+    for library, strings in sorted(strings_by_library.items(), key=lambda p: len(p[1]), reverse=True):
+        console.print(f"  - [b]{library}[/] ({len(strings)} strings)")
+
+        for string in sorted(strings)[:10]:
+            console.print(f"    - {string}", markup=False, style="grey37")
+
+        if len(strings) > 10:
+            console.print("    ...", style="grey37")
+
+    if not strings_by_library:
+        console.print("  (none)", style="grey37")
+        # since we're not going to find any strings
+        # return early and don't do IDA analysis
+        return
+
+    open_ida(input_path)
+
+    import idaapi
+    import idautils
+    import ida_funcs
+
+    strings_by_function = collections.defaultdict(set)
+    for ea in idautils.Functions():
+        f = idaapi.get_func(ea)
+
+        # ignore library functions and thunk functions as identified by IDA
+        if f.flags & idaapi.FUNC_THUNK:
+            continue
+        if f.flags & idaapi.FUNC_LIB:
+            continue
+
+        for bb in ida_helpers.get_function_blocks(f):
+            for insn in ida_helpers.get_instructions_in_range(bb.start_ea, bb.end_ea):
+                ref = capa.features.extractors.ida.helpers.find_data_reference_from_insn(insn)
+                if ref == insn.ea:
+                    continue
+
+                string = capa.features.extractors.ida.helpers.find_string_at(ref)
+                if not string:
+                    continue
+
+                for db in dbs:
+                    if metadata := db.metadata_by_string.get(string):
+                        strings_by_function[ea].add(string)
+
+    # ensure there are at least XXX functions renamed, or ignore those entries
+
+    console.print("functions:", style="bold")
+    for function, strings in sorted(strings_by_function.items()):
+        if strings:
+            name = ida_funcs.get_func_name(function)
+
+            console.print(f"  [b]{name}[/]@{function:08x}:")
+
+            for string in strings:
+                for db in dbs:
+                    if metadata := db.metadata_by_string.get(string):
+                        location = Text(
+                            f"{metadata.library_name}@{metadata.library_version}::{metadata.function_name}",
+                            style="grey37",
+                        )
+                        console.print("    - ", location, ": ", string.rstrip())
+
+    console.print()
+
+    console.print(
+        f"found {len(strings_by_function)} library functions across {len(list(idautils.Functions()))} functions"
+    )
+
+
+if __name__ == "__main__":
+    main()

capa/analysis/strings/data/oss/.gitignore

@@ -0,0 +1,3 @@
+*.csv
+*.jsonl
+*.jsonl.gz

capa/analysis/strings/data/oss/readme.md

@@ -0,0 +1,99 @@
+# Strings from Open Source libraries
+
+This directory contains databases of strings extracted from open soure software. 
+capa uses these databases to ignore functions that are likely library code.
+
+There is one file for each database. Each database is a gzip-compressed, JSONL (one JSON document per line) file.
+The JSON document looks like this:
+
+    string: "1.0.8, 13-Jul-2019"
+    library_name: "bzip2"
+    library_version: "1.0.8#3"
+    file_path: "CMakeFiles/bz2.dir/bzlib.c.obj"
+    function_name: "BZ2_bzlibVersion"
+    line_number: null
+
+The following databases were extracted via the vkpkg & jh technique:
+
+  - brotli 1.0.9#5
+  - bzip2 1.0.8#3
+  - cryptopp 8.7.0
+  - curl 7.86.0#1
+  - detours 4.0.1#7
+  - jemalloc 5.3.0#1
+  - jsoncpp 1.9.5
+  - kcp 1.7
+  - liblzma 5.2.5#6
+  - libsodium 1.0.18#8
+  - libpcap 1.10.1#3
+  - mbedtls 2.28.1
+  - openssl 3.0.7#1
+  - sqlite3 3.40.0#1
+  - tomcrypt 1.18.2#2
+  - wolfssl 5.5.0
+  - zlib 1.2.13
+
+This code was originally developed in FLOSS and imported into capa.
+
+## The vkpkg & jh technique
+
+Major steps:
+
+  1. build static libraries via vcpkg
+  2. extract features via jh
+  3. convert to JSONL format with `jh_to_qs.py`
+  4. compress with gzip
+
+### Build static libraries via vcpkg
+
+[vcpkg](https://vcpkg.io/en/) is a free C/C++ package manager for acquiring and managing libraries.
+We use it to easily build common open source libraries, like zlib.
+Use the triplet `x64-windows-static` to build static archives (.lib files that are AR archives containing COFF object files):
+
+```console
+PS > C:\vcpkg\vcpkg.exe install --triplet x64-windows-static zlib
+```
+
+### Extract features via jh
+
+[jh](https://github.com/williballenthin/lancelot/blob/master/bin/src/bin/jh.rs)
+is a lancelot-based utility that parses AR archives containing COFF object files,
+reconstructs their control flow, finds functions, and extracts features. 
+jh extracts numbers, API calls, and strings; we are only interested in the string features.
+
+For each feature, jh emits a CSV line with the fields 
+  - target triplet
+  - compiler 
+  - library
+  - version
+  - build profile
+  - path
+  - function
+  - feature type
+  - feature value
+
+For example:
+
+```csv
+x64-windows-static,msvc143,bzip2,1.0.8#3,release,CMakeFiles/bz2.dir/bzlib.c.obj,BZ2_bzBuffToBuffCompress,number,0x00000100
+```
+
+For example, to invoke jh:
+
+```console
+$ ~/lancelot/target/release/jh x64-windows-static msvc143 zlib 1.2.13 release /mnt/c/vcpkg/installed/x64-windows-static/lib/zlib.lib > ~/flare-floss/floss/qs/db/data/oss/zlib.csv
+```
+
+### Convert to OSS database format
+
+We use the script `jh_to_qs.py` to convert these CSV lines into JSONL file prepared for FLOSS:
+
+```console
+$ python3 jh_to_qs.py zlib.csv > zlib.jsonl
+```
+
+These files are then gzip'd:
+
+```console
+$  gzip -c zlib.jsonl > zlib.jsonl.gz
+```

capa/capabilities/dynamic.py

@@ -6,20 +6,16 @@
 # Unless required by applicable law or agreed to in writing, software distributed under the License
 #  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and limitations under the License.
-import sys
 import logging
 import itertools
 import collections
-from typing import Any, Tuple
+from typing import Any, List, Tuple
-
-import tqdm
 
 import capa.perf
 import capa.features.freeze as frz
 import capa.render.result_document as rdoc
 from capa.rules import Scope, RuleSet
 from capa.engine import FeatureSet, MatchResults
-from capa.helpers import redirecting_print_to_tqdm
 from capa.capabilities.common import find_file_capabilities
 from capa.features.extractors.base_extractor import CallHandle, ThreadHandle, ProcessHandle, DynamicFeatureExtractor
 
@@ -139,38 +135,30 @@ def find_dynamic_capabilities(
     feature_counts = rdoc.DynamicFeatureCounts(file=0, processes=())
 
     assert isinstance(extractor, DynamicFeatureExtractor)
-    with redirecting_print_to_tqdm(disable_progress):
+    processes: List[ProcessHandle] = list(extractor.get_processes())
-        with tqdm.contrib.logging.logging_redirect_tqdm():
+    n_processes: int = len(processes)
-            pbar = tqdm.tqdm
-            if disable_progress:
-                # do not use tqdm to avoid unnecessary side effects when caller intends
-                # to disable progress completely
-                def pbar(s, *args, **kwargs):
-                    return s
 
-            elif not sys.stderr.isatty():
+    with capa.helpers.CapaProgressBar(
-                # don't display progress bar when stderr is redirected to a file
+        console=capa.helpers.log_console, transient=True, disable=disable_progress
-                def pbar(s, *args, **kwargs):
+    ) as pbar:
-                    return s
+        task = pbar.add_task("matching", total=n_processes, unit="processes")
+        for p in processes:
+            process_matches, thread_matches, call_matches, feature_count = find_process_capabilities(
+                ruleset, extractor, p
+            )
+            feature_counts.processes += (
+                rdoc.ProcessFeatureCount(address=frz.Address.from_capa(p.address), count=feature_count),
+            )
+            logger.debug("analyzed %s and extracted %d features", p.address, feature_count)
 
-            processes = list(extractor.get_processes())
+            for rule_name, res in process_matches.items():
+                all_process_matches[rule_name].extend(res)
+            for rule_name, res in thread_matches.items():
+                all_thread_matches[rule_name].extend(res)
+            for rule_name, res in call_matches.items():
+                all_call_matches[rule_name].extend(res)
 
-            pb = pbar(processes, desc="matching", unit=" processes", leave=False)
+            pbar.advance(task)
-            for p in pb:
-                process_matches, thread_matches, call_matches, feature_count = find_process_capabilities(
-                    ruleset, extractor, p
-                )
-                feature_counts.processes += (
-                    rdoc.ProcessFeatureCount(address=frz.Address.from_capa(p.address), count=feature_count),
-                )
-                logger.debug("analyzed %s and extracted %d features", p.address, feature_count)
-
-                for rule_name, res in process_matches.items():
-                    all_process_matches[rule_name].extend(res)
-                for rule_name, res in thread_matches.items():
-                    all_thread_matches[rule_name].extend(res)
-                for rule_name, res in call_matches.items():
-                    all_call_matches[rule_name].extend(res)
 
     # collection of features that captures the rule matches within process and thread scopes.
     # mapping from feature (matched rule) to set of addresses at which it matched.

capa/capabilities/static.py

@@ -6,21 +6,18 @@
 # Unless required by applicable law or agreed to in writing, software distributed under the License
 #  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and limitations under the License.
-import sys
 import time
 import logging
 import itertools
 import collections
-from typing import Any, Tuple
+from typing import Any, List, Tuple
-
-import tqdm.contrib.logging
 
 import capa.perf
+import capa.helpers
 import capa.features.freeze as frz
 import capa.render.result_document as rdoc
 from capa.rules import Scope, RuleSet
 from capa.engine import FeatureSet, MatchResults
-from capa.helpers import redirecting_print_to_tqdm
 from capa.capabilities.common import find_file_capabilities
 from capa.features.extractors.base_extractor import BBHandle, InsnHandle, FunctionHandle, StaticFeatureExtractor
 
@@ -143,75 +140,58 @@ def find_static_capabilities(
     library_functions: Tuple[rdoc.LibraryFunction, ...] = ()
 
     assert isinstance(extractor, StaticFeatureExtractor)
-    with redirecting_print_to_tqdm(disable_progress):
+    functions: List[FunctionHandle] = list(extractor.get_functions())
-        with tqdm.contrib.logging.logging_redirect_tqdm():
+    n_funcs: int = len(functions)
-            pbar = tqdm.tqdm
+    n_libs: int = 0
-            if capa.helpers.is_runtime_ghidra():
+    percentage: float = 0
-                # Ghidrathon interpreter cannot properly handle
-                # the TMonitor thread that is created via a monitor_interval
-                # > 0
-                pbar.monitor_interval = 0
-            if disable_progress:
-                # do not use tqdm to avoid unnecessary side effects when caller intends
-                # to disable progress completely
-                def pbar(s, *args, **kwargs):
-                    return s
 
-            elif not sys.stderr.isatty():
+    with capa.helpers.CapaProgressBar(
-                # don't display progress bar when stderr is redirected to a file
+        console=capa.helpers.log_console, transient=True, disable=disable_progress
-                def pbar(s, *args, **kwargs):
+    ) as pbar:
-                    return s
+        task = pbar.add_task(
-
+            "matching", total=n_funcs, unit="functions", postfix=f"skipped {n_libs} library functions, {percentage}%"
-            functions = list(extractor.get_functions())
+        )
-            n_funcs = len(functions)
+        for f in functions:
-
+            t0 = time.time()
-            pb = pbar(functions, desc="matching", unit=" functions", postfix="skipped 0 library functions", leave=False)
+            if extractor.is_library_function(f.address):
-            for f in pb:
+                function_name = extractor.get_function_name(f.address)
-                t0 = time.time()
+                logger.debug("skipping library function 0x%x (%s)", f.address, function_name)
-                if extractor.is_library_function(f.address):
+                library_functions += (
-                    function_name = extractor.get_function_name(f.address)
+                    rdoc.LibraryFunction(address=frz.Address.from_capa(f.address), name=function_name),
-                    logger.debug("skipping library function 0x%x (%s)", f.address, function_name)
-                    library_functions += (
-                        rdoc.LibraryFunction(address=frz.Address.from_capa(f.address), name=function_name),
-                    )
-                    n_libs = len(library_functions)
-                    percentage = round(100 * (n_libs / n_funcs))
-                    if isinstance(pb, tqdm.tqdm):
-                        pb.set_postfix_str(f"skipped {n_libs} library functions ({percentage}%)")
-                    continue
-
-                function_matches, bb_matches, insn_matches, feature_count = find_code_capabilities(
-                    ruleset, extractor, f
                 )
-                feature_counts.functions += (
+                n_libs = len(library_functions)
-                    rdoc.FunctionFeatureCount(address=frz.Address.from_capa(f.address), count=feature_count),
+                percentage = round(100 * (n_libs / n_funcs))
-                )
+                pbar.update(task, postfix=f"skipped {n_libs} library functions, {percentage}%")
-                t1 = time.time()
+                pbar.advance(task)
+                continue
 
-                match_count = 0
+            function_matches, bb_matches, insn_matches, feature_count = find_code_capabilities(ruleset, extractor, f)
-                for name, matches_ in itertools.chain(
+            feature_counts.functions += (
-                    function_matches.items(), bb_matches.items(), insn_matches.items()
+                rdoc.FunctionFeatureCount(address=frz.Address.from_capa(f.address), count=feature_count),
-                ):
+            )
-                    # in practice, most matches are derived rules,
+            t1 = time.time()
-                    # like "check OS version/5bf4c7f39fd4492cbed0f6dc7d596d49"
-                    # but when we log to the human, they really care about "real" rules.
-                    if not ruleset.rules[name].is_subscope_rule():
-                        match_count += len(matches_)
 
-                logger.debug(
+            match_count = 0
-                    "analyzed function 0x%x and extracted %d features, %d matches in %0.02fs",
+            for name, matches_ in itertools.chain(function_matches.items(), bb_matches.items(), insn_matches.items()):
-                    f.address,
+                if not ruleset.rules[name].is_subscope_rule():
-                    feature_count,
+                    match_count += len(matches_)
-                    match_count,
-                    t1 - t0,
-                )
 
-                for rule_name, res in function_matches.items():
+            logger.debug(
-                    all_function_matches[rule_name].extend(res)
+                "analyzed function 0x%x and extracted %d features, %d matches in %0.02fs",
-                for rule_name, res in bb_matches.items():
+                f.address,
-                    all_bb_matches[rule_name].extend(res)
+                feature_count,
-                for rule_name, res in insn_matches.items():
+                match_count,
-                    all_insn_matches[rule_name].extend(res)
+                t1 - t0,
+            )
+
+            for rule_name, res in function_matches.items():
+                all_function_matches[rule_name].extend(res)
+            for rule_name, res in bb_matches.items():
+                all_bb_matches[rule_name].extend(res)
+            for rule_name, res in insn_matches.items():
+                all_insn_matches[rule_name].extend(res)
+
+            pbar.advance(task)
 
     # collection of features that captures the rule matches within function, BB, and instruction scopes.
     # mapping from feature (matched rule) to set of addresses at which it matched.

capa/features/common.py

@@ -424,10 +424,11 @@ class Arch(Feature):
 OS_WINDOWS = "windows"
 OS_LINUX = "linux"
 OS_MACOS = "macos"
+OS_ANDROID = "android"
 # dotnet
 OS_ANY = "any"
 VALID_OS = {os.value for os in capa.features.extractors.elf.OS}
-VALID_OS.update({OS_WINDOWS, OS_LINUX, OS_MACOS, OS_ANY})
+VALID_OS.update({OS_WINDOWS, OS_LINUX, OS_MACOS, OS_ANY, OS_ANDROID})
 # internal only, not to be used in rules
 OS_AUTO = "auto"
 
@@ -462,6 +463,8 @@ FORMAT_SC32 = "sc32"
 FORMAT_SC64 = "sc64"
 FORMAT_CAPE = "cape"
 FORMAT_DRAKVUF = "drakvuf"
+FORMAT_VMRAY = "vmray"
+FORMAT_BINEXPORT2 = "binexport2"
 FORMAT_FREEZE = "freeze"
 FORMAT_RESULT = "result"
 STATIC_FORMATS = {
@@ -472,10 +475,12 @@ STATIC_FORMATS = {
     FORMAT_DOTNET,
     FORMAT_FREEZE,
     FORMAT_RESULT,
+    FORMAT_BINEXPORT2,
 }
 DYNAMIC_FORMATS = {
     FORMAT_CAPE,
     FORMAT_DRAKVUF,
+    FORMAT_VMRAY,
     FORMAT_FREEZE,
     FORMAT_RESULT,
 }

capa/features/extractors/binexport2/__init__.py

@@ -0,0 +1,416 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+"""
+Proto files generated via protobuf v24.4:
+
+    protoc --python_out=. --mypy_out=. binexport2.proto
+
+from BinExport2 at 6916731d5f6693c4a4f0a052501fd3bd92cfd08b
+https://github.com/google/binexport/blob/6916731/binexport2.proto
+"""
+import io
+import hashlib
+import logging
+import contextlib
+from typing import Set, Dict, List, Tuple, Iterator
+from pathlib import Path
+from collections import defaultdict
+from dataclasses import dataclass
+
+from pefile import PE
+from elftools.elf.elffile import ELFFile
+
+import capa.features.common
+import capa.features.extractors.common
+import capa.features.extractors.binexport2.helpers
+from capa.features.extractors.binexport2.binexport2_pb2 import BinExport2
+
+logger = logging.getLogger(__name__)
+
+
+def get_binexport2(sample: Path) -> BinExport2:
+    be2: BinExport2 = BinExport2()
+    be2.ParseFromString(sample.read_bytes())
+    return be2
+
+
+def compute_common_prefix_length(m: str, n: str) -> int:
+    # ensure #m < #n
+    if len(n) < len(m):
+        m, n = n, m
+
+    for i, c in enumerate(m):
+        if n[i] != c:
+            return i
+
+    return len(m)
+
+
+def get_sample_from_binexport2(input_file: Path, be2: BinExport2, search_paths: List[Path]) -> Path:
+    """attempt to find the sample file, given a BinExport2 file.
+
+    searches in the same directory as the BinExport2 file, and then in search_paths.
+    """
+
+    def filename_similarity_key(p: Path) -> Tuple[int, str]:
+        # note closure over input_file.
+        # sort first by length of common prefix, then by name (for stability)
+        return (compute_common_prefix_length(p.name, input_file.name), p.name)
+
+    wanted_sha256: str = be2.meta_information.executable_id.lower()
+
+    input_directory: Path = input_file.parent
+    siblings: List[Path] = [p for p in input_directory.iterdir() if p.is_file()]
+    siblings.sort(key=filename_similarity_key, reverse=True)
+    for sibling in siblings:
+        # e.g. with open IDA files in the same directory on Windows
+        with contextlib.suppress(PermissionError):
+            if hashlib.sha256(sibling.read_bytes()).hexdigest().lower() == wanted_sha256:
+                return sibling
+
+    for search_path in search_paths:
+        candidates: List[Path] = [p for p in search_path.iterdir() if p.is_file()]
+        candidates.sort(key=filename_similarity_key, reverse=True)
+        for candidate in candidates:
+            with contextlib.suppress(PermissionError):
+                if hashlib.sha256(candidate.read_bytes()).hexdigest().lower() == wanted_sha256:
+                    return candidate
+
+    raise ValueError("cannot find sample, you may specify the path using the CAPA_SAMPLES_DIR environment variable")
+
+
+class BinExport2Index:
+    def __init__(self, be2: BinExport2):
+        self.be2: BinExport2 = be2
+
+        self.callers_by_vertex_index: Dict[int, List[int]] = defaultdict(list)
+        self.callees_by_vertex_index: Dict[int, List[int]] = defaultdict(list)
+
+        # note: flow graph != call graph (vertex)
+        self.flow_graph_index_by_address: Dict[int, int] = {}
+        self.flow_graph_address_by_index: Dict[int, int] = {}
+
+        # edges that come from the given basic block
+        self.source_edges_by_basic_block_index: Dict[int, List[BinExport2.FlowGraph.Edge]] = defaultdict(list)
+        # edges that end up at the given basic block
+        self.target_edges_by_basic_block_index: Dict[int, List[BinExport2.FlowGraph.Edge]] = defaultdict(list)
+
+        self.vertex_index_by_address: Dict[int, int] = {}
+
+        self.data_reference_index_by_source_instruction_index: Dict[int, List[int]] = defaultdict(list)
+        self.data_reference_index_by_target_address: Dict[int, List[int]] = defaultdict(list)
+        self.string_reference_index_by_source_instruction_index: Dict[int, List[int]] = defaultdict(list)
+
+        self.insn_address_by_index: Dict[int, int] = {}
+        self.insn_index_by_address: Dict[int, int] = {}
+        self.insn_by_address: Dict[int, BinExport2.Instruction] = {}
+
+        # must index instructions first
+        self._index_insn_addresses()
+        self._index_vertex_edges()
+        self._index_flow_graph_nodes()
+        self._index_flow_graph_edges()
+        self._index_call_graph_vertices()
+        self._index_data_references()
+        self._index_string_references()
+
+    def get_insn_address(self, insn_index: int) -> int:
+        assert insn_index in self.insn_address_by_index, f"insn must be indexed, missing {insn_index}"
+        return self.insn_address_by_index[insn_index]
+
+    def get_basic_block_address(self, basic_block_index: int) -> int:
+        basic_block: BinExport2.BasicBlock = self.be2.basic_block[basic_block_index]
+        first_instruction_index: int = next(self.instruction_indices(basic_block))
+        return self.get_insn_address(first_instruction_index)
+
+    def _index_vertex_edges(self):
+        for edge in self.be2.call_graph.edge:
+            if not edge.source_vertex_index:
+                continue
+            if not edge.target_vertex_index:
+                continue
+
+            self.callers_by_vertex_index[edge.target_vertex_index].append(edge.source_vertex_index)
+            self.callees_by_vertex_index[edge.source_vertex_index].append(edge.target_vertex_index)
+
+    def _index_flow_graph_nodes(self):
+        for flow_graph_index, flow_graph in enumerate(self.be2.flow_graph):
+            function_address: int = self.get_basic_block_address(flow_graph.entry_basic_block_index)
+            self.flow_graph_index_by_address[function_address] = flow_graph_index
+            self.flow_graph_address_by_index[flow_graph_index] = function_address
+
+    def _index_flow_graph_edges(self):
+        for flow_graph in self.be2.flow_graph:
+            for edge in flow_graph.edge:
+                if not edge.HasField("source_basic_block_index") or not edge.HasField("target_basic_block_index"):
+                    continue
+
+                self.source_edges_by_basic_block_index[edge.source_basic_block_index].append(edge)
+                self.target_edges_by_basic_block_index[edge.target_basic_block_index].append(edge)
+
+    def _index_call_graph_vertices(self):
+        for vertex_index, vertex in enumerate(self.be2.call_graph.vertex):
+            if not vertex.HasField("address"):
+                continue
+
+            vertex_address: int = vertex.address
+            self.vertex_index_by_address[vertex_address] = vertex_index
+
+    def _index_data_references(self):
+        for data_reference_index, data_reference in enumerate(self.be2.data_reference):
+            self.data_reference_index_by_source_instruction_index[data_reference.instruction_index].append(
+                data_reference_index
+            )
+            self.data_reference_index_by_target_address[data_reference.address].append(data_reference_index)
+
+    def _index_string_references(self):
+        for string_reference_index, string_reference in enumerate(self.be2.string_reference):
+            self.string_reference_index_by_source_instruction_index[string_reference.instruction_index].append(
+                string_reference_index
+            )
+
+    def _index_insn_addresses(self):
+        # see https://github.com/google/binexport/blob/39f6445c232bb5caf5c4a2a996de91dfa20c48e8/binexport.cc#L45
+        if len(self.be2.instruction) == 0:
+            return
+
+        assert self.be2.instruction[0].HasField("address"), "first insn must have explicit address"
+
+        addr: int = 0
+        next_addr: int = 0
+        for idx, insn in enumerate(self.be2.instruction):
+            if insn.HasField("address"):
+                addr = insn.address
+                next_addr = addr + len(insn.raw_bytes)
+            else:
+                addr = next_addr
+                next_addr += len(insn.raw_bytes)
+            self.insn_address_by_index[idx] = addr
+            self.insn_index_by_address[addr] = idx
+            self.insn_by_address[addr] = insn
+
+    @staticmethod
+    def instruction_indices(basic_block: BinExport2.BasicBlock) -> Iterator[int]:
+        """
+        For a given basic block, enumerate the instruction indices.
+        """
+        for index_range in basic_block.instruction_index:
+            if not index_range.HasField("end_index"):
+                yield index_range.begin_index
+                continue
+            else:
+                yield from range(index_range.begin_index, index_range.end_index)
+
+    def basic_block_instructions(
+        self, basic_block: BinExport2.BasicBlock
+    ) -> Iterator[Tuple[int, BinExport2.Instruction, int]]:
+        """
+        For a given basic block, enumerate the instruction indices,
+        the instruction instances, and their addresses.
+        """
+        for instruction_index in self.instruction_indices(basic_block):
+            instruction: BinExport2.Instruction = self.be2.instruction[instruction_index]
+            instruction_address: int = self.get_insn_address(instruction_index)
+
+            yield instruction_index, instruction, instruction_address
+
+    def get_function_name_by_vertex(self, vertex_index: int) -> str:
+        vertex: BinExport2.CallGraph.Vertex = self.be2.call_graph.vertex[vertex_index]
+        name: str = f"sub_{vertex.address:x}"
+        if vertex.HasField("mangled_name"):
+            name = vertex.mangled_name
+
+        if vertex.HasField("demangled_name"):
+            name = vertex.demangled_name
+
+        if vertex.HasField("library_index"):
+            library: BinExport2.Library = self.be2.library[vertex.library_index]
+            if library.HasField("name"):
+                name = f"{library.name}!{name}"
+
+        return name
+
+    def get_function_name_by_address(self, address: int) -> str:
+        if address not in self.vertex_index_by_address:
+            return ""
+
+        vertex_index: int = self.vertex_index_by_address[address]
+        return self.get_function_name_by_vertex(vertex_index)
+
+    def get_instruction_by_address(self, address: int) -> BinExport2.Instruction:
+        assert address in self.insn_by_address, f"address must be indexed, missing {address:x}"
+        return self.insn_by_address[address]
+
+
+class BinExport2Analysis:
+    def __init__(self, be2: BinExport2, idx: BinExport2Index, buf: bytes):
+        self.be2: BinExport2 = be2
+        self.idx: BinExport2Index = idx
+        self.buf: bytes = buf
+        self.base_address: int = 0
+        self.thunks: Dict[int, int] = {}
+
+        self._find_base_address()
+        self._compute_thunks()
+
+    def _find_base_address(self):
+        sections_with_perms: Iterator[BinExport2.Section] = filter(
+            lambda s: s.flag_r or s.flag_w or s.flag_x, self.be2.section
+        )
+        # assume the lowest address is the base address.
+        # this works as long as BinExport doesn't record other
+        # libraries mapped into memory.
+        self.base_address = min(s.address for s in sections_with_perms)
+
+        logger.debug("found base address: %x", self.base_address)
+
+    def _compute_thunks(self):
+        for addr, idx in self.idx.vertex_index_by_address.items():
+            vertex: BinExport2.CallGraph.Vertex = self.be2.call_graph.vertex[idx]
+            if not capa.features.extractors.binexport2.helpers.is_vertex_type(
+                vertex, BinExport2.CallGraph.Vertex.Type.THUNK
+            ):
+                continue
+
+            curr_idx: int = idx
+            for _ in range(capa.features.common.THUNK_CHAIN_DEPTH_DELTA):
+                thunk_callees: List[int] = self.idx.callees_by_vertex_index[curr_idx]
+                # if this doesn't hold, then it doesn't seem like this is a thunk,
+                # because either, len is:
+                #    0 and the thunk doesn't point to anything, or
+                #   >1 and the thunk may end up at many functions.
+                assert len(thunk_callees) == 1, f"thunk @ {hex(addr)} failed"
+
+                thunked_idx: int = thunk_callees[0]
+                thunked_vertex: BinExport2.CallGraph.Vertex = self.be2.call_graph.vertex[thunked_idx]
+
+                if not capa.features.extractors.binexport2.helpers.is_vertex_type(
+                    thunked_vertex, BinExport2.CallGraph.Vertex.Type.THUNK
+                ):
+                    assert thunked_vertex.HasField("address")
+
+                    self.thunks[addr] = thunked_vertex.address
+                    break
+
+                curr_idx = thunked_idx
+
+
+@dataclass
+class MemoryRegion:
+    # location of the bytes, potentially relative to a base address
+    address: int
+    buf: bytes
+
+    @property
+    def end(self) -> int:
+        return self.address + len(self.buf)
+
+    def contains(self, address: int) -> bool:
+        # note: address must be relative to any base address
+        return self.address <= address < self.end
+
+
+class ReadMemoryError(ValueError): ...
+
+
+class AddressNotMappedError(ReadMemoryError): ...
+
+
+@dataclass
+class AddressSpace:
+    base_address: int
+    memory_regions: Tuple[MemoryRegion, ...]
+
+    def read_memory(self, address: int, length: int) -> bytes:
+        rva: int = address - self.base_address
+        for region in self.memory_regions:
+            if region.contains(rva):
+                offset: int = rva - region.address
+                return region.buf[offset : offset + length]
+
+        raise AddressNotMappedError(address)
+
+    @classmethod
+    def from_pe(cls, pe: PE, base_address: int):
+        regions: List[MemoryRegion] = []
+        for section in pe.sections:
+            address: int = section.VirtualAddress
+            size: int = section.Misc_VirtualSize
+            buf: bytes = section.get_data()
+
+            if len(buf) != size:
+                # pad the section with NULLs
+                # assume page alignment is already handled.
+                # might need more hardening here.
+                buf += b"\x00" * (size - len(buf))
+
+            regions.append(MemoryRegion(address, buf))
+
+        return cls(base_address, tuple(regions))
+
+    @classmethod
+    def from_elf(cls, elf: ELFFile, base_address: int):
+        regions: List[MemoryRegion] = []
+
+        # ELF segments are for runtime data,
+        # ELF sections are for link-time data.
+        for segment in elf.iter_segments():
+            # assume p_align is consistent with addresses here.
+            # otherwise, should harden this loader.
+            segment_rva: int = segment.header.p_vaddr
+            segment_size: int = segment.header.p_memsz
+            segment_data: bytes = segment.data()
+
+            if len(segment_data) < segment_size:
+                # pad the section with NULLs
+                # assume page alignment is already handled.
+                # might need more hardening here.
+                segment_data += b"\x00" * (segment_size - len(segment_data))
+
+            regions.append(MemoryRegion(segment_rva, segment_data))
+
+        return cls(base_address, tuple(regions))
+
+    @classmethod
+    def from_buf(cls, buf: bytes, base_address: int):
+        if buf.startswith(capa.features.extractors.common.MATCH_PE):
+            pe: PE = PE(data=buf)
+            return cls.from_pe(pe, base_address)
+        elif buf.startswith(capa.features.extractors.common.MATCH_ELF):
+            elf: ELFFile = ELFFile(io.BytesIO(buf))
+            return cls.from_elf(elf, base_address)
+        else:
+            raise NotImplementedError("file format address space")
+
+
+@dataclass
+class AnalysisContext:
+    sample_bytes: bytes
+    be2: BinExport2
+    idx: BinExport2Index
+    analysis: BinExport2Analysis
+    address_space: AddressSpace
+
+
+@dataclass
+class FunctionContext:
+    ctx: AnalysisContext
+    flow_graph_index: int
+    format: Set[str]
+    os: Set[str]
+    arch: Set[str]
+
+
+@dataclass
+class BasicBlockContext:
+    basic_block_index: int
+
+
+@dataclass
+class InstructionContext:
+    instruction_index: int

capa/features/extractors/binexport2/arch/arm/helpers.py

@@ -0,0 +1,15 @@
+# Copyright (C) 2024 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+
+from capa.features.extractors.binexport2.binexport2_pb2 import BinExport2
+
+
+def is_stack_register_expression(be2: BinExport2, expression: BinExport2.Expression) -> bool:
+    return bool(
+        expression and expression.type == BinExport2.Expression.REGISTER and expression.symbol.lower().endswith("sp")
+    )

capa/features/extractors/binexport2/arch/arm/insn.py

@@ -0,0 +1,155 @@
+# Copyright (C) 2024 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+import logging
+from typing import List, Tuple, Iterator, Optional
+
+import capa.features.extractors.binexport2.helpers
+from capa.features.insn import MAX_STRUCTURE_SIZE, Number, Offset, OperandNumber, OperandOffset
+from capa.features.common import Feature, Characteristic
+from capa.features.address import Address
+from capa.features.extractors.binexport2 import FunctionContext, InstructionContext
+from capa.features.extractors.base_extractor import BBHandle, InsnHandle, FunctionHandle
+from capa.features.extractors.binexport2.helpers import (
+    BinExport2InstructionPatternMatcher,
+    mask_immediate,
+    is_address_mapped,
+    get_instruction_mnemonic,
+    get_operand_register_expression,
+    get_operand_immediate_expression,
+)
+from capa.features.extractors.binexport2.binexport2_pb2 import BinExport2
+from capa.features.extractors.binexport2.arch.arm.helpers import is_stack_register_expression
+
+logger = logging.getLogger(__name__)
+
+
+def extract_insn_number_features(
+    fh: FunctionHandle, _bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
+    fhi: FunctionContext = fh.inner
+    ii: InstructionContext = ih.inner
+
+    be2: BinExport2 = fhi.ctx.be2
+
+    instruction_index: int = ii.instruction_index
+    instruction: BinExport2.Instruction = be2.instruction[instruction_index]
+
+    if len(instruction.operand_index) == 0:
+        # skip things like:
+        #   .text:0040116e leave
+        return
+
+    mnemonic: str = get_instruction_mnemonic(be2, instruction)
+
+    if mnemonic in ("add", "sub"):
+        assert len(instruction.operand_index) == 3
+
+        operand1_expression: Optional[BinExport2.Expression] = get_operand_register_expression(
+            be2, be2.operand[instruction.operand_index[1]]
+        )
+        if operand1_expression and is_stack_register_expression(be2, operand1_expression):
+            # skip things like:
+            # add x0,sp,#0x8
+            return
+
+    for i, operand_index in enumerate(instruction.operand_index):
+        operand: BinExport2.Operand = be2.operand[operand_index]
+
+        immediate_expression: Optional[BinExport2.Expression] = get_operand_immediate_expression(be2, operand)
+        if not immediate_expression:
+            continue
+
+        value: int = mask_immediate(fhi.arch, immediate_expression.immediate)
+        if is_address_mapped(be2, value):
+            continue
+
+        yield Number(value), ih.address
+        yield OperandNumber(i, value), ih.address
+
+        if mnemonic == "add" and i == 2:
+            if 0 < value < MAX_STRUCTURE_SIZE:
+                yield Offset(value), ih.address
+                yield OperandOffset(i, value), ih.address
+
+
+OFFSET_PATTERNS = BinExport2InstructionPatternMatcher.from_str(
+    """
+    ldr|ldrb|ldrh|ldrsb|ldrsh|ldrex|ldrd|str|strb|strh|strex|strd reg, [reg(not-stack),  #int]                                 ; capture #int
+    ldr|ldrb|ldrh|ldrsb|ldrsh|ldrex|ldrd|str|strb|strh|strex|strd reg, [reg(not-stack),  #int]!                                ; capture #int
+    ldr|ldrb|ldrh|ldrsb|ldrsh|ldrex|ldrd|str|strb|strh|strex|strd reg, [reg(not-stack)],        #int                           ; capture #int
+    ldp|ldpd|stp|stpd                                             reg, reg,                     [reg(not-stack), #int]         ; capture #int
+    ldp|ldpd|stp|stpd                                             reg, reg,                     [reg(not-stack), #int]!        ; capture #int
+    ldp|ldpd|stp|stpd                                             reg, reg,                     [reg(not-stack)],       #int   ; capture #int
+    """
+)
+
+
+def extract_insn_offset_features(
+    fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
+    fhi: FunctionContext = fh.inner
+    ii: InstructionContext = ih.inner
+
+    be2: BinExport2 = fhi.ctx.be2
+
+    match = OFFSET_PATTERNS.match_with_be2(be2, ii.instruction_index)
+    if not match:
+        return
+
+    value = match.expression.immediate
+
+    value = mask_immediate(fhi.arch, value)
+    if not is_address_mapped(be2, value):
+        value = capa.features.extractors.binexport2.helpers.twos_complement(fhi.arch, value)
+        yield Offset(value), ih.address
+        yield OperandOffset(match.operand_index, value), ih.address
+
+
+NZXOR_PATTERNS = BinExport2InstructionPatternMatcher.from_str(
+    """
+    eor reg, reg, reg
+    eor reg, reg, #int
+    """
+)
+
+
+def extract_insn_nzxor_characteristic_features(
+    fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
+    fhi: FunctionContext = fh.inner
+    ii: InstructionContext = ih.inner
+    be2: BinExport2 = fhi.ctx.be2
+
+    if NZXOR_PATTERNS.match_with_be2(be2, ii.instruction_index) is None:
+        return
+
+    instruction: BinExport2.Instruction = be2.instruction[ii.instruction_index]
+    # guaranteed to be simple int/reg operands
+    # so we don't have to realize the tree/list.
+    operands: List[BinExport2.Operand] = [be2.operand[operand_index] for operand_index in instruction.operand_index]
+
+    if operands[1] != operands[2]:
+        yield Characteristic("nzxor"), ih.address
+
+
+INDIRECT_CALL_PATTERNS = BinExport2InstructionPatternMatcher.from_str(
+    """
+    blx|bx|blr reg
+    """
+)
+
+
+def extract_function_indirect_call_characteristic_features(
+    fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
+    fhi: FunctionContext = fh.inner
+    ii: InstructionContext = ih.inner
+    be2: BinExport2 = fhi.ctx.be2
+
+    if INDIRECT_CALL_PATTERNS.match_with_be2(be2, ii.instruction_index) is not None:
+        yield Characteristic("indirect call"), ih.address

capa/features/extractors/binexport2/arch/intel/helpers.py

@@ -0,0 +1,135 @@
+# Copyright (C) 2024 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+from typing import List, Optional
+from dataclasses import dataclass
+
+from capa.features.extractors.binexport2.helpers import get_operand_expressions
+from capa.features.extractors.binexport2.binexport2_pb2 import BinExport2
+
+# security cookie checks may perform non-zeroing XORs, these are expected within a certain
+# byte range within the first and returning basic blocks, this helps to reduce FP features
+SECURITY_COOKIE_BYTES_DELTA: int = 0x40
+
+
+@dataclass
+class OperandPhraseInfo:
+    scale: Optional[BinExport2.Expression] = None
+    index: Optional[BinExport2.Expression] = None
+    base: Optional[BinExport2.Expression] = None
+    displacement: Optional[BinExport2.Expression] = None
+
+
+def get_operand_phrase_info(be2: BinExport2, operand: BinExport2.Operand) -> Optional[OperandPhraseInfo]:
+    # assume the following (see https://blog.yossarian.net/2020/06/13/How-x86_64-addresses-memory):
+    #
+    # Scale: A 2-bit constant factor
+    # Index: Any general purpose register
+    # Base: Any general purpose register
+    # Displacement: An integral offset
+
+    expressions: List[BinExport2.Expression] = get_operand_expressions(be2, operand)
+
+    # skip expression up to and including BinExport2.Expression.DEREFERENCE, assume caller
+    # has checked for BinExport2.Expression.DEREFERENCE
+    for i, expression in enumerate(expressions):
+        if expression.type == BinExport2.Expression.DEREFERENCE:
+            expressions = expressions[i + 1 :]
+            break
+
+    expression0: BinExport2.Expression
+    expression1: BinExport2.Expression
+    expression2: BinExport2.Expression
+    expression3: BinExport2.Expression
+    expression4: BinExport2.Expression
+
+    if len(expressions) == 1:
+        expression0 = expressions[0]
+
+        assert (
+            expression0.type == BinExport2.Expression.IMMEDIATE_INT
+            or expression0.type == BinExport2.Expression.REGISTER
+        )
+
+        if expression0.type == BinExport2.Expression.IMMEDIATE_INT:
+            # Displacement
+            return OperandPhraseInfo(displacement=expression0)
+        elif expression0.type == BinExport2.Expression.REGISTER:
+            # Base
+            return OperandPhraseInfo(base=expression0)
+
+    elif len(expressions) == 3:
+        expression0 = expressions[0]
+        expression1 = expressions[1]
+        expression2 = expressions[2]
+
+        assert expression0.type == BinExport2.Expression.REGISTER
+        assert expression1.type == BinExport2.Expression.OPERATOR
+        assert (
+            expression2.type == BinExport2.Expression.IMMEDIATE_INT
+            or expression2.type == BinExport2.Expression.REGISTER
+        )
+
+        if expression2.type == BinExport2.Expression.REGISTER:
+            # Base + Index
+            return OperandPhraseInfo(base=expression0, index=expression2)
+        elif expression2.type == BinExport2.Expression.IMMEDIATE_INT:
+            # Base + Displacement
+            return OperandPhraseInfo(base=expression0, displacement=expression2)
+
+    elif len(expressions) == 5:
+        expression0 = expressions[0]
+        expression1 = expressions[1]
+        expression2 = expressions[2]
+        expression3 = expressions[3]
+        expression4 = expressions[4]
+
+        assert expression0.type == BinExport2.Expression.REGISTER
+        assert expression1.type == BinExport2.Expression.OPERATOR
+        assert (
+            expression2.type == BinExport2.Expression.REGISTER
+            or expression2.type == BinExport2.Expression.IMMEDIATE_INT
+        )
+        assert expression3.type == BinExport2.Expression.OPERATOR
+        assert expression4.type == BinExport2.Expression.IMMEDIATE_INT
+
+        if expression1.symbol == "+" and expression3.symbol == "+":
+            # Base + Index + Displacement
+            return OperandPhraseInfo(base=expression0, index=expression2, displacement=expression4)
+        elif expression1.symbol == "+" and expression3.symbol == "*":
+            # Base + (Index * Scale)
+            return OperandPhraseInfo(base=expression0, index=expression2, scale=expression3)
+        elif expression1.symbol == "*" and expression3.symbol == "+":
+            # (Index * Scale) + Displacement
+            return OperandPhraseInfo(index=expression0, scale=expression2, displacement=expression3)
+        else:
+            raise NotImplementedError(expression1.symbol, expression3.symbol)
+
+    elif len(expressions) == 7:
+        expression0 = expressions[0]
+        expression1 = expressions[1]
+        expression2 = expressions[2]
+        expression3 = expressions[3]
+        expression4 = expressions[4]
+        expression5 = expressions[5]
+        expression6 = expressions[6]
+
+        assert expression0.type == BinExport2.Expression.REGISTER
+        assert expression1.type == BinExport2.Expression.OPERATOR
+        assert expression2.type == BinExport2.Expression.REGISTER
+        assert expression3.type == BinExport2.Expression.OPERATOR
+        assert expression4.type == BinExport2.Expression.IMMEDIATE_INT
+        assert expression5.type == BinExport2.Expression.OPERATOR
+        assert expression6.type == BinExport2.Expression.IMMEDIATE_INT
+
+        # Base + (Index * Scale) + Displacement
+        return OperandPhraseInfo(base=expression0, index=expression2, scale=expression4, displacement=expression6)
+
+    else:
+        raise NotImplementedError(len(expressions))
+
+    return None

capa/features/extractors/binexport2/arch/intel/insn.py

@@ -0,0 +1,248 @@
+# Copyright (C) 2024 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+import logging
+from typing import List, Tuple, Iterator
+
+import capa.features.extractors.strings
+import capa.features.extractors.binexport2.helpers
+from capa.features.insn import MAX_STRUCTURE_SIZE, Number, Offset, OperandNumber, OperandOffset
+from capa.features.common import Feature, Characteristic
+from capa.features.address import Address
+from capa.features.extractors.binexport2 import BinExport2Index, FunctionContext, BasicBlockContext, InstructionContext
+from capa.features.extractors.base_extractor import BBHandle, InsnHandle, FunctionHandle
+from capa.features.extractors.binexport2.helpers import (
+    BinExport2InstructionPatternMatcher,
+    mask_immediate,
+    is_address_mapped,
+    get_instruction_mnemonic,
+)
+from capa.features.extractors.binexport2.binexport2_pb2 import BinExport2
+from capa.features.extractors.binexport2.arch.intel.helpers import SECURITY_COOKIE_BYTES_DELTA
+
+logger = logging.getLogger(__name__)
+
+
+IGNORE_NUMBER_PATTERNS = BinExport2InstructionPatternMatcher.from_str(
+    """
+    ret  #int
+    retn #int
+    add  reg(stack), #int
+    sub  reg(stack), #int
+    """
+)
+
+NUMBER_PATTERNS = BinExport2InstructionPatternMatcher.from_str(
+    """
+    push #int0            ; capture #int0
+
+    # its a little tedious to enumerate all the address forms
+    # but at least we are explicit
+    cmp|and|or|test|mov|add|adc|sub|shl|shr|sal|sar  reg,                      #int0  ; capture #int0
+    cmp|and|or|test|mov|add|adc|sub|shl|shr|sal|sar [reg],                     #int0  ; capture #int0
+    cmp|and|or|test|mov|add|adc|sub|shl|shr|sal|sar [#int],                    #int0  ; capture #int0
+    cmp|and|or|test|mov|add|adc|sub|shl|shr|sal|sar [reg + #int],              #int0  ; capture #int0
+    cmp|and|or|test|mov|add|adc|sub|shl|shr|sal|sar [reg + reg + #int],        #int0  ; capture #int0
+    cmp|and|or|test|mov|add|adc|sub|shl|shr|sal|sar [reg + reg * #int],        #int0  ; capture #int0
+    cmp|and|or|test|mov|add|adc|sub|shl|shr|sal|sar [reg + reg * #int + #int], #int0  ; capture #int0
+
+    imul reg, reg, #int  ; capture #int
+    # note that int is first
+    cmp|test #int0, reg   ; capture #int0
+
+    # imagine reg is zero'd out, then this is like `mov reg, #int`
+    # which is not uncommon.
+    lea reg, [reg + #int]  ; capture #int
+    """
+)
+
+
+def extract_insn_number_features(
+    fh: FunctionHandle, _bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
+    fhi: FunctionContext = fh.inner
+    ii: InstructionContext = ih.inner
+
+    be2: BinExport2 = fhi.ctx.be2
+
+    if IGNORE_NUMBER_PATTERNS.match_with_be2(be2, ii.instruction_index):
+        return
+
+    match = NUMBER_PATTERNS.match_with_be2(be2, ii.instruction_index)
+    if not match:
+        return
+
+    value: int = mask_immediate(fhi.arch, match.expression.immediate)
+    if is_address_mapped(be2, value):
+        return
+
+    yield Number(value), ih.address
+    yield OperandNumber(match.operand_index, value), ih.address
+
+    instruction_index: int = ii.instruction_index
+    instruction: BinExport2.Instruction = be2.instruction[instruction_index]
+
+    mnemonic: str = get_instruction_mnemonic(be2, instruction)
+    if mnemonic.startswith("add"):
+        if 0 < value < MAX_STRUCTURE_SIZE:
+            yield Offset(value), ih.address
+            yield OperandOffset(match.operand_index, value), ih.address
+
+
+OFFSET_PATTERNS = BinExport2InstructionPatternMatcher.from_str(
+    """
+    mov|movzx|movsb|cmp [reg            +  reg * #int + #int0], #int  ; capture #int0
+    mov|movzx|movsb|cmp [reg            * #int + #int0],        #int  ; capture #int0
+    mov|movzx|movsb|cmp [reg            +  reg + #int0],        #int  ; capture #int0
+    mov|movzx|movsb|cmp [reg(not-stack) + #int0],               #int  ; capture #int0
+    mov|movzx|movsb|cmp [reg            +  reg * #int + #int0], reg   ; capture #int0
+    mov|movzx|movsb|cmp [reg            * #int + #int0],        reg   ; capture #int0
+    mov|movzx|movsb|cmp [reg            +  reg + #int0],        reg   ; capture #int0
+    mov|movzx|movsb|cmp [reg(not-stack) + #int0],               reg   ; capture #int0
+    mov|movzx|movsb|cmp|lea reg, [reg            +  reg * #int + #int0]  ; capture #int0
+    mov|movzx|movsb|cmp|lea reg, [reg            * #int + #int0]         ; capture #int0
+    mov|movzx|movsb|cmp|lea reg, [reg            +  reg + #int0]         ; capture #int0
+    mov|movzx|movsb|cmp|lea reg, [reg(not-stack) + #int0]                ; capture #int0
+    """
+)
+
+# these are patterns that access offset 0 from some pointer
+# (pointer is not the stack pointer).
+OFFSET_ZERO_PATTERNS = BinExport2InstructionPatternMatcher.from_str(
+    """
+    mov|movzx|movsb [reg(not-stack)], reg
+    mov|movzx|movsb [reg(not-stack)], #int
+    lea             reg,              [reg(not-stack)]
+    """
+)
+
+
+def extract_insn_offset_features(
+    fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
+    fhi: FunctionContext = fh.inner
+    ii: InstructionContext = ih.inner
+
+    be2: BinExport2 = fhi.ctx.be2
+
+    match = OFFSET_PATTERNS.match_with_be2(be2, ii.instruction_index)
+    if not match:
+        match = OFFSET_ZERO_PATTERNS.match_with_be2(be2, ii.instruction_index)
+        if not match:
+            return
+
+        yield Offset(0), ih.address
+        yield OperandOffset(match.operand_index, 0), ih.address
+
+    value = mask_immediate(fhi.arch, match.expression.immediate)
+    if is_address_mapped(be2, value):
+        return
+
+    value = capa.features.extractors.binexport2.helpers.twos_complement(fhi.arch, value, 32)
+    yield Offset(value), ih.address
+    yield OperandOffset(match.operand_index, value), ih.address
+
+
+def is_security_cookie(
+    fhi: FunctionContext,
+    bbi: BasicBlockContext,
+    instruction_address: int,
+    instruction: BinExport2.Instruction,
+) -> bool:
+    """
+    check if an instruction is related to security cookie checks.
+    """
+    be2: BinExport2 = fhi.ctx.be2
+    idx: BinExport2Index = fhi.ctx.idx
+
+    # security cookie check should use SP or BP
+    op1: BinExport2.Operand = be2.operand[instruction.operand_index[1]]
+    op1_exprs: List[BinExport2.Expression] = [be2.expression[expr_i] for expr_i in op1.expression_index]
+    if all(expr.symbol.lower() not in ("bp", "esp", "ebp", "rbp", "rsp") for expr in op1_exprs):
+        return False
+
+    # check_nzxor_security_cookie_delta
+    # if insn falls at the start of first entry block of the parent function.
+    flow_graph: BinExport2.FlowGraph = be2.flow_graph[fhi.flow_graph_index]
+    basic_block_index: int = bbi.basic_block_index
+    bb: BinExport2.BasicBlock = be2.basic_block[basic_block_index]
+    if flow_graph.entry_basic_block_index == basic_block_index:
+        first_addr: int = min((idx.insn_address_by_index[ir.begin_index] for ir in bb.instruction_index))
+        if instruction_address < first_addr + SECURITY_COOKIE_BYTES_DELTA:
+            return True
+    # or insn falls at the end before return in a terminal basic block.
+    if basic_block_index not in (e.source_basic_block_index for e in flow_graph.edge):
+        last_addr: int = max((idx.insn_address_by_index[ir.end_index - 1] for ir in bb.instruction_index))
+        if instruction_address > last_addr - SECURITY_COOKIE_BYTES_DELTA:
+            return True
+    return False
+
+
+NZXOR_PATTERNS = BinExport2InstructionPatternMatcher.from_str(
+    """
+    xor|xorpd|xorps|pxor reg, reg
+    xor|xorpd|xorps|pxor reg, #int
+    """
+)
+
+
+def extract_insn_nzxor_characteristic_features(
+    fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
+    """
+    parse non-zeroing XOR instruction from the given instruction.
+    ignore expected non-zeroing XORs, e.g. security cookies.
+    """
+    fhi: FunctionContext = fh.inner
+    ii: InstructionContext = ih.inner
+
+    be2: BinExport2 = fhi.ctx.be2
+    idx: BinExport2Index = fhi.ctx.idx
+
+    if NZXOR_PATTERNS.match_with_be2(be2, ii.instruction_index) is None:
+        return
+
+    instruction: BinExport2.Instruction = be2.instruction[ii.instruction_index]
+    # guaranteed to be simple int/reg operands
+    # so we don't have to realize the tree/list.
+    operands: List[BinExport2.Operand] = [be2.operand[operand_index] for operand_index in instruction.operand_index]
+
+    if operands[0] == operands[1]:
+        return
+
+    instruction_address: int = idx.insn_address_by_index[ii.instruction_index]
+    if is_security_cookie(fhi, bbh.inner, instruction_address, instruction):
+        return
+
+    yield Characteristic("nzxor"), ih.address
+
+
+INDIRECT_CALL_PATTERNS = BinExport2InstructionPatternMatcher.from_str(
+    """
+    call|jmp reg0
+    call|jmp [reg + reg * #int + #int]
+    call|jmp [reg + reg * #int]
+    call|jmp [reg * #int + #int]
+    call|jmp [reg + reg + #int]
+    call|jmp [reg + #int]
+    call|jmp [reg]
+    """
+)
+
+
+def extract_function_indirect_call_characteristic_features(
+    fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
+    fhi: FunctionContext = fh.inner
+    ii: InstructionContext = ih.inner
+    be2: BinExport2 = fhi.ctx.be2
+
+    match = INDIRECT_CALL_PATTERNS.match_with_be2(be2, ii.instruction_index)
+    if match is None:
+        return
+
+    yield Characteristic("indirect call"), ih.address

capa/features/extractors/binexport2/basicblock.py

@@ -0,0 +1,40 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+
+from typing import List, Tuple, Iterator
+
+from capa.features.common import Feature, Characteristic
+from capa.features.address import Address, AbsoluteVirtualAddress
+from capa.features.basicblock import BasicBlock
+from capa.features.extractors.binexport2 import FunctionContext, BasicBlockContext
+from capa.features.extractors.base_extractor import BBHandle, FunctionHandle
+from capa.features.extractors.binexport2.binexport2_pb2 import BinExport2
+
+
+def extract_bb_tight_loop(fh: FunctionHandle, bbh: BBHandle) -> Iterator[Tuple[Feature, Address]]:
+    fhi: FunctionContext = fh.inner
+    bbi: BasicBlockContext = bbh.inner
+
+    idx = fhi.ctx.idx
+
+    basic_block_index: int = bbi.basic_block_index
+    target_edges: List[BinExport2.FlowGraph.Edge] = idx.target_edges_by_basic_block_index[basic_block_index]
+    if basic_block_index in (e.source_basic_block_index for e in target_edges):
+        basic_block_address: int = idx.get_basic_block_address(basic_block_index)
+        yield Characteristic("tight loop"), AbsoluteVirtualAddress(basic_block_address)
+
+
+def extract_features(fh: FunctionHandle, bbh: BBHandle) -> Iterator[Tuple[Feature, Address]]:
+    """extract basic block features"""
+    for bb_handler in BASIC_BLOCK_HANDLERS:
+        for feature, addr in bb_handler(fh, bbh):
+            yield feature, addr
+    yield BasicBlock(), bbh.address
+
+
+BASIC_BLOCK_HANDLERS = (extract_bb_tight_loop,)

capa/features/extractors/binexport2/binexport2_pb2.pyi

@@ -0,0 +1,784 @@
+"""
+@generated by mypy-protobuf.  Do not edit manually!
+isort:skip_file
+The representation is generic to accommodate various source architectures.
+In particular 32 and 64 bit versions of x86, ARM, PowerPC and MIPS have been
+tested.
+
+Multiple levels of deduping have been applied to make the format more compact
+and avoid redundant data duplication. Some of this due to hard-earned
+experience trying to cope with intentionally obfuscated malicious binaries.
+Note in particular that the same instruction may occur in multiple basic
+blocks and the same basic block in multiple functions (instruction and basic
+block sharing). Implemented naively, malware can use this to cause
+combinatorial explosion in memory usage, DOSing the analyst. This format
+should store every unique expression, mnemonic, operand, instruction and
+basic block only once instead of duplicating the information for every
+instance of it.
+
+This format does _not_ try to be 100% backwards compatible with the old
+version. In particular, we do not store IDA's comment types, making lossless
+porting of IDA comments impossible. We do however, store comments and
+expression substitutions, so porting the actual data is possible, just not
+the exact IDA type.
+
+While it would be more natural to use addresses when defining call graph and
+flow graph edges and other such references, it is more efficient to employ
+one more level of indirection and use indices into the basic block or
+function arrays instead. This is because addresses will usually use most of
+the available 64 bit space while indices will be much smaller and compress
+much better (less randomly distributed).
+
+We omit all fields that are set to their default value anyways. Note that
+this has two side effects:
+  - changing the defaults in this proto file will, in effect, change what's
+    read from disk
+  - the generated code has_* methods are somewhat less useful
+WARNING: We omit the defaults manually in the code writing the data. Do not
+         change the defaults here without changing the code!
+
+TODO(cblichmann): Link flow graphs to call graph nodes. The connection is
+                  there via the address, but tricky to extract.
+"""
+import builtins
+import collections.abc
+import google.protobuf.descriptor
+import google.protobuf.internal.containers
+import google.protobuf.internal.enum_type_wrapper
+import google.protobuf.message
+import sys
+import typing
+
+if sys.version_info >= (3, 10):
+    import typing as typing_extensions
+else:
+    import typing_extensions
+
+DESCRIPTOR: google.protobuf.descriptor.FileDescriptor
+
+@typing_extensions.final
+class BinExport2(google.protobuf.message.Message):
+    DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
+    @typing_extensions.final
+    class Meta(google.protobuf.message.Message):
+        DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
+        EXECUTABLE_NAME_FIELD_NUMBER: builtins.int
+        EXECUTABLE_ID_FIELD_NUMBER: builtins.int
+        ARCHITECTURE_NAME_FIELD_NUMBER: builtins.int
+        TIMESTAMP_FIELD_NUMBER: builtins.int
+        executable_name: builtins.str
+        """Input binary filename including file extension but excluding file path.
+        example: "insider_gcc.exe"
+        """
+        executable_id: builtins.str
+        """Application defined executable id. Often the SHA256 hash of the input
+        binary.
+        """
+        architecture_name: builtins.str
+        """Input architecture name, e.g. x86-32."""
+        timestamp: builtins.int
+        """When did this file get created? Unix time. This may be used for some
+        primitive versioning in case the file format ever changes.
+        """
+        def __init__(
+            self,
+            *,
+            executable_name: builtins.str | None = ...,
+            executable_id: builtins.str | None = ...,
+            architecture_name: builtins.str | None = ...,
+            timestamp: builtins.int | None = ...,
+        ) -> None: ...
+        def HasField(self, field_name: typing_extensions.Literal["architecture_name", b"architecture_name", "executable_id", b"executable_id", "executable_name", b"executable_name", "timestamp", b"timestamp"]) -> builtins.bool: ...
+        def ClearField(self, field_name: typing_extensions.Literal["architecture_name", b"architecture_name", "executable_id", b"executable_id", "executable_name", b"executable_name", "timestamp", b"timestamp"]) -> None: ...
+
+    @typing_extensions.final
+    class CallGraph(google.protobuf.message.Message):
+        DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
+        @typing_extensions.final
+        class Vertex(google.protobuf.message.Message):
+            DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
+            class _Type:
+                ValueType = typing.NewType("ValueType", builtins.int)
+                V: typing_extensions.TypeAlias = ValueType
+
+            class _TypeEnumTypeWrapper(google.protobuf.internal.enum_type_wrapper._EnumTypeWrapper[BinExport2.CallGraph.Vertex._Type.ValueType], builtins.type):
+                DESCRIPTOR: google.protobuf.descriptor.EnumDescriptor
+                NORMAL: BinExport2.CallGraph.Vertex._Type.ValueType  # 0
+                """Regular function with full disassembly."""
+                LIBRARY: BinExport2.CallGraph.Vertex._Type.ValueType  # 1
+                """This function is a well known library function."""
+                IMPORTED: BinExport2.CallGraph.Vertex._Type.ValueType  # 2
+                """Imported from a dynamic link library (e.g. dll)."""
+                THUNK: BinExport2.CallGraph.Vertex._Type.ValueType  # 3
+                """A thunk function, forwarding its work via an unconditional jump."""
+                INVALID: BinExport2.CallGraph.Vertex._Type.ValueType  # 4
+                """An invalid function (a function that contained invalid code or was
+                considered invalid by some heuristics).
+                """
+
+            class Type(_Type, metaclass=_TypeEnumTypeWrapper): ...
+            NORMAL: BinExport2.CallGraph.Vertex.Type.ValueType  # 0
+            """Regular function with full disassembly."""
+            LIBRARY: BinExport2.CallGraph.Vertex.Type.ValueType  # 1
+            """This function is a well known library function."""
+            IMPORTED: BinExport2.CallGraph.Vertex.Type.ValueType  # 2
+            """Imported from a dynamic link library (e.g. dll)."""
+            THUNK: BinExport2.CallGraph.Vertex.Type.ValueType  # 3
+            """A thunk function, forwarding its work via an unconditional jump."""
+            INVALID: BinExport2.CallGraph.Vertex.Type.ValueType  # 4
+            """An invalid function (a function that contained invalid code or was
+            considered invalid by some heuristics).
+            """
+
+            ADDRESS_FIELD_NUMBER: builtins.int
+            TYPE_FIELD_NUMBER: builtins.int
+            MANGLED_NAME_FIELD_NUMBER: builtins.int
+            DEMANGLED_NAME_FIELD_NUMBER: builtins.int
+            LIBRARY_INDEX_FIELD_NUMBER: builtins.int
+            MODULE_INDEX_FIELD_NUMBER: builtins.int
+            address: builtins.int
+            """The function's entry point address. Messages need to be sorted, see
+            comment below on `vertex`.
+            """
+            type: global___BinExport2.CallGraph.Vertex.Type.ValueType
+            mangled_name: builtins.str
+            """If the function has a user defined, real name it will be given here.
+            main() is a proper name, sub_BAADF00D is not (auto generated dummy
+            name).
+            """
+            demangled_name: builtins.str
+            """Demangled name if the function is a mangled C++ function and we could
+            demangle it.
+            """
+            library_index: builtins.int
+            """If this is a library function, what is its index in library arrays."""
+            module_index: builtins.int
+            """If module name, such as class name for DEX files, is present - index in
+            module table.
+            """
+            def __init__(
+                self,
+                *,
+                address: builtins.int | None = ...,
+                type: global___BinExport2.CallGraph.Vertex.Type.ValueType | None = ...,
+                mangled_name: builtins.str | None = ...,
+                demangled_name: builtins.str | None = ...,
+                library_index: builtins.int | None = ...,
+                module_index: builtins.int | None = ...,
+            ) -> None: ...
+            def HasField(self, field_name: typing_extensions.Literal["address", b"address", "demangled_name", b"demangled_name", "library_index", b"library_index", "mangled_name", b"mangled_name", "module_index", b"module_index", "type", b"type"]) -> builtins.bool: ...
+            def ClearField(self, field_name: typing_extensions.Literal["address", b"address", "demangled_name", b"demangled_name", "library_index", b"library_index", "mangled_name", b"mangled_name", "module_index", b"module_index", "type", b"type"]) -> None: ...
+
+        @typing_extensions.final
+        class Edge(google.protobuf.message.Message):
+            DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
+            SOURCE_VERTEX_INDEX_FIELD_NUMBER: builtins.int
+            TARGET_VERTEX_INDEX_FIELD_NUMBER: builtins.int
+            source_vertex_index: builtins.int
+            """source and target index into the vertex repeated field."""
+            target_vertex_index: builtins.int
+            def __init__(
+                self,
+                *,
+                source_vertex_index: builtins.int | None = ...,
+                target_vertex_index: builtins.int | None = ...,
+            ) -> None: ...
+            def HasField(self, field_name: typing_extensions.Literal["source_vertex_index", b"source_vertex_index", "target_vertex_index", b"target_vertex_index"]) -> builtins.bool: ...
+            def ClearField(self, field_name: typing_extensions.Literal["source_vertex_index", b"source_vertex_index", "target_vertex_index", b"target_vertex_index"]) -> None: ...
+
+        VERTEX_FIELD_NUMBER: builtins.int
+        EDGE_FIELD_NUMBER: builtins.int
+        @property
+        def vertex(self) -> google.protobuf.internal.containers.RepeatedCompositeFieldContainer[global___BinExport2.CallGraph.Vertex]:
+            """vertices == functions in the call graph.
+            Important: Most downstream tooling (notably BinDiff), need these to be
+                       sorted by `Vertex::address` (ascending). For C++, the
+                       `BinExport2Writer` class enforces this invariant.
+            """
+        @property
+        def edge(self) -> google.protobuf.internal.containers.RepeatedCompositeFieldContainer[global___BinExport2.CallGraph.Edge]:
+            """edges == calls in the call graph."""
+        def __init__(
+            self,
+            *,
+            vertex: collections.abc.Iterable[global___BinExport2.CallGraph.Vertex] | None = ...,
+            edge: collections.abc.Iterable[global___BinExport2.CallGraph.Edge] | None = ...,
+        ) -> None: ...
+        def ClearField(self, field_name: typing_extensions.Literal["edge", b"edge", "vertex", b"vertex"]) -> None: ...
+
+    @typing_extensions.final
+    class Expression(google.protobuf.message.Message):
+        """An operand consists of 1 or more expressions, linked together as a tree."""
+
+        DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
+        class _Type:
+            ValueType = typing.NewType("ValueType", builtins.int)
+            V: typing_extensions.TypeAlias = ValueType
+
+        class _TypeEnumTypeWrapper(google.protobuf.internal.enum_type_wrapper._EnumTypeWrapper[BinExport2.Expression._Type.ValueType], builtins.type):
+            DESCRIPTOR: google.protobuf.descriptor.EnumDescriptor
+            SYMBOL: BinExport2.Expression._Type.ValueType  # 1
+            IMMEDIATE_INT: BinExport2.Expression._Type.ValueType  # 2
+            IMMEDIATE_FLOAT: BinExport2.Expression._Type.ValueType  # 3
+            OPERATOR: BinExport2.Expression._Type.ValueType  # 4
+            REGISTER: BinExport2.Expression._Type.ValueType  # 5
+            SIZE_PREFIX: BinExport2.Expression._Type.ValueType  # 6
+            DEREFERENCE: BinExport2.Expression._Type.ValueType  # 7
+
+        class Type(_Type, metaclass=_TypeEnumTypeWrapper): ...
+        SYMBOL: BinExport2.Expression.Type.ValueType  # 1
+        IMMEDIATE_INT: BinExport2.Expression.Type.ValueType  # 2
+        IMMEDIATE_FLOAT: BinExport2.Expression.Type.ValueType  # 3
+        OPERATOR: BinExport2.Expression.Type.ValueType  # 4
+        REGISTER: BinExport2.Expression.Type.ValueType  # 5
+        SIZE_PREFIX: BinExport2.Expression.Type.ValueType  # 6
+        DEREFERENCE: BinExport2.Expression.Type.ValueType  # 7
+
+        TYPE_FIELD_NUMBER: builtins.int
+        SYMBOL_FIELD_NUMBER: builtins.int
+        IMMEDIATE_FIELD_NUMBER: builtins.int
+        PARENT_INDEX_FIELD_NUMBER: builtins.int
+        IS_RELOCATION_FIELD_NUMBER: builtins.int
+        type: global___BinExport2.Expression.Type.ValueType
+        """IMMEDIATE_INT is by far the most common type and thus we can save some
+        space by omitting it as the default.
+        """
+        symbol: builtins.str
+        """Symbol for this expression. Interpretation depends on type. Examples
+        include: "eax", "[", "+"
+        """
+        immediate: builtins.int
+        """If the expression can be interpreted as an integer value (IMMEDIATE_INT)
+        the value is given here.
+        """
+        parent_index: builtins.int
+        """The parent expression. Example expression tree for the second operand of:
+        mov eax, b4 [ebx + 12]
+        "b4" --- "[" --- "+" --- "ebx"
+                              \\  "12"
+        """
+        is_relocation: builtins.bool
+        """true if the expression has entry in relocation table"""
+        def __init__(
+            self,
+            *,
+            type: global___BinExport2.Expression.Type.ValueType | None = ...,
+            symbol: builtins.str | None = ...,
+            immediate: builtins.int | None = ...,
+            parent_index: builtins.int | None = ...,
+            is_relocation: builtins.bool | None = ...,
+        ) -> None: ...
+        def HasField(self, field_name: typing_extensions.Literal["immediate", b"immediate", "is_relocation", b"is_relocation", "parent_index", b"parent_index", "symbol", b"symbol", "type", b"type"]) -> builtins.bool: ...
+        def ClearField(self, field_name: typing_extensions.Literal["immediate", b"immediate", "is_relocation", b"is_relocation", "parent_index", b"parent_index", "symbol", b"symbol", "type", b"type"]) -> None: ...
+
+    @typing_extensions.final
+    class Operand(google.protobuf.message.Message):
+        """An instruction may have 0 or more operands."""
+
+        DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
+        EXPRESSION_INDEX_FIELD_NUMBER: builtins.int
+        @property
+        def expression_index(self) -> google.protobuf.internal.containers.RepeatedScalarFieldContainer[builtins.int]:
+            """Contains all expressions constituting this operand. All expressions
+            should be linked into a single tree, i.e. there should only be one
+            expression in this list with parent_index == NULL and all others should
+            descend from that. Rendering order for expressions on the same tree level
+            (siblings) is implicitly given by the order they are referenced in this
+            repeated field.
+            Implicit: expression sequence
+            """
+        def __init__(
+            self,
+            *,
+            expression_index: collections.abc.Iterable[builtins.int] | None = ...,
+        ) -> None: ...
+        def ClearField(self, field_name: typing_extensions.Literal["expression_index", b"expression_index"]) -> None: ...
+
+    @typing_extensions.final
+    class Mnemonic(google.protobuf.message.Message):
+        """An instruction has exactly 1 mnemonic."""
+
+        DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
+        NAME_FIELD_NUMBER: builtins.int
+        name: builtins.str
+        """Literal representation of the mnemonic, e.g.: "mov"."""
+        def __init__(
+            self,
+            *,
+            name: builtins.str | None = ...,
+        ) -> None: ...
+        def HasField(self, field_name: typing_extensions.Literal["name", b"name"]) -> builtins.bool: ...
+        def ClearField(self, field_name: typing_extensions.Literal["name", b"name"]) -> None: ...
+
+    @typing_extensions.final
+    class Instruction(google.protobuf.message.Message):
+        DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
+        ADDRESS_FIELD_NUMBER: builtins.int
+        CALL_TARGET_FIELD_NUMBER: builtins.int
+        MNEMONIC_INDEX_FIELD_NUMBER: builtins.int
+        OPERAND_INDEX_FIELD_NUMBER: builtins.int
+        RAW_BYTES_FIELD_NUMBER: builtins.int
+        COMMENT_INDEX_FIELD_NUMBER: builtins.int
+        address: builtins.int
+        """This will only be filled for instructions that do not just flow from the
+        immediately preceding instruction. Regular instructions will have to
+        calculate their own address by adding raw_bytes.size() to the previous
+        instruction's address.
+        """
+        @property
+        def call_target(self) -> google.protobuf.internal.containers.RepeatedScalarFieldContainer[builtins.int]:
+            """If this is a call instruction and call targets could be determined
+            they'll be given here. Note that we may or may not have a flow graph for
+            the target and thus cannot use an index into the flow graph table here.
+            We could potentially use call graph nodes, but linking instructions to
+            the call graph directly does not seem a good choice.
+            """
+        mnemonic_index: builtins.int
+        """Index into the mnemonic array of strings. Used for de-duping the data.
+        The default value is used for the most common mnemonic in the executable.
+        """
+        @property
+        def operand_index(self) -> google.protobuf.internal.containers.RepeatedScalarFieldContainer[builtins.int]:
+            """Indices into the operand tree. On X86 this can be 0, 1 or 2 elements
+            long, 3 elements with VEX/EVEX.
+            Implicit: operand sequence
+            """
+        raw_bytes: builtins.bytes
+        """The unmodified input bytes corresponding to this instruction."""
+        @property
+        def comment_index(self) -> google.protobuf.internal.containers.RepeatedScalarFieldContainer[builtins.int]:
+            """Implicit: comment sequence"""
+        def __init__(
+            self,
+            *,
+            address: builtins.int | None = ...,
+            call_target: collections.abc.Iterable[builtins.int] | None = ...,
+            mnemonic_index: builtins.int | None = ...,
+            operand_index: collections.abc.Iterable[builtins.int] | None = ...,
+            raw_bytes: builtins.bytes | None = ...,
+            comment_index: collections.abc.Iterable[builtins.int] | None = ...,
+        ) -> None: ...
+        def HasField(self, field_name: typing_extensions.Literal["address", b"address", "mnemonic_index", b"mnemonic_index", "raw_bytes", b"raw_bytes"]) -> builtins.bool: ...
+        def ClearField(self, field_name: typing_extensions.Literal["address", b"address", "call_target", b"call_target", "comment_index", b"comment_index", "mnemonic_index", b"mnemonic_index", "operand_index", b"operand_index", "raw_bytes", b"raw_bytes"]) -> None: ...
+
+    @typing_extensions.final
+    class BasicBlock(google.protobuf.message.Message):
+        DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
+        @typing_extensions.final
+        class IndexRange(google.protobuf.message.Message):
+            """This is a space optimization. The instructions for an individual basic
+            block will usually be in a continuous index range. Thus it is more
+            efficient to store the range instead of individual indices. However, this
+            does not hold true for all basic blocks, so we need to be able to store
+            multiple index ranges per block.
+            """
+
+            DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
+            BEGIN_INDEX_FIELD_NUMBER: builtins.int
+            END_INDEX_FIELD_NUMBER: builtins.int
+            begin_index: builtins.int
+            """These work like begin and end iterators, i.e. the sequence is
+            [begin_index, end_index). If the sequence only contains a single
+            element end_index will be omitted.
+            """
+            end_index: builtins.int
+            def __init__(
+                self,
+                *,
+                begin_index: builtins.int | None = ...,
+                end_index: builtins.int | None = ...,
+            ) -> None: ...
+            def HasField(self, field_name: typing_extensions.Literal["begin_index", b"begin_index", "end_index", b"end_index"]) -> builtins.bool: ...
+            def ClearField(self, field_name: typing_extensions.Literal["begin_index", b"begin_index", "end_index", b"end_index"]) -> None: ...
+
+        INSTRUCTION_INDEX_FIELD_NUMBER: builtins.int
+        @property
+        def instruction_index(self) -> google.protobuf.internal.containers.RepeatedCompositeFieldContainer[global___BinExport2.BasicBlock.IndexRange]:
+            """Implicit: instruction sequence"""
+        def __init__(
+            self,
+            *,
+            instruction_index: collections.abc.Iterable[global___BinExport2.BasicBlock.IndexRange] | None = ...,
+        ) -> None: ...
+        def ClearField(self, field_name: typing_extensions.Literal["instruction_index", b"instruction_index"]) -> None: ...
+
+    @typing_extensions.final
+    class FlowGraph(google.protobuf.message.Message):
+        DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
+        @typing_extensions.final
+        class Edge(google.protobuf.message.Message):
+            DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
+            class _Type:
+                ValueType = typing.NewType("ValueType", builtins.int)
+                V: typing_extensions.TypeAlias = ValueType
+
+            class _TypeEnumTypeWrapper(google.protobuf.internal.enum_type_wrapper._EnumTypeWrapper[BinExport2.FlowGraph.Edge._Type.ValueType], builtins.type):
+                DESCRIPTOR: google.protobuf.descriptor.EnumDescriptor
+                CONDITION_TRUE: BinExport2.FlowGraph.Edge._Type.ValueType  # 1
+                CONDITION_FALSE: BinExport2.FlowGraph.Edge._Type.ValueType  # 2
+                UNCONDITIONAL: BinExport2.FlowGraph.Edge._Type.ValueType  # 3
+                SWITCH: BinExport2.FlowGraph.Edge._Type.ValueType  # 4
+
+            class Type(_Type, metaclass=_TypeEnumTypeWrapper): ...
+            CONDITION_TRUE: BinExport2.FlowGraph.Edge.Type.ValueType  # 1
+            CONDITION_FALSE: BinExport2.FlowGraph.Edge.Type.ValueType  # 2
+            UNCONDITIONAL: BinExport2.FlowGraph.Edge.Type.ValueType  # 3
+            SWITCH: BinExport2.FlowGraph.Edge.Type.ValueType  # 4
+
+            SOURCE_BASIC_BLOCK_INDEX_FIELD_NUMBER: builtins.int
+            TARGET_BASIC_BLOCK_INDEX_FIELD_NUMBER: builtins.int
+            TYPE_FIELD_NUMBER: builtins.int
+            IS_BACK_EDGE_FIELD_NUMBER: builtins.int
+            source_basic_block_index: builtins.int
+            """Source instruction will always be the last instruction of the source
+            basic block, target instruction the first instruction of the target
+            basic block.
+            """
+            target_basic_block_index: builtins.int
+            type: global___BinExport2.FlowGraph.Edge.Type.ValueType
+            is_back_edge: builtins.bool
+            """Indicates whether this is a loop edge as determined by Lengauer-Tarjan."""
+            def __init__(
+                self,
+                *,
+                source_basic_block_index: builtins.int | None = ...,
+                target_basic_block_index: builtins.int | None = ...,
+                type: global___BinExport2.FlowGraph.Edge.Type.ValueType | None = ...,
+                is_back_edge: builtins.bool | None = ...,
+            ) -> None: ...
+            def HasField(self, field_name: typing_extensions.Literal["is_back_edge", b"is_back_edge", "source_basic_block_index", b"source_basic_block_index", "target_basic_block_index", b"target_basic_block_index", "type", b"type"]) -> builtins.bool: ...
+            def ClearField(self, field_name: typing_extensions.Literal["is_back_edge", b"is_back_edge", "source_basic_block_index", b"source_basic_block_index", "target_basic_block_index", b"target_basic_block_index", "type", b"type"]) -> None: ...
+
+        BASIC_BLOCK_INDEX_FIELD_NUMBER: builtins.int
+        ENTRY_BASIC_BLOCK_INDEX_FIELD_NUMBER: builtins.int
+        EDGE_FIELD_NUMBER: builtins.int
+        @property
+        def basic_block_index(self) -> google.protobuf.internal.containers.RepeatedScalarFieldContainer[builtins.int]:
+            """Basic blocks are sorted by address."""
+        entry_basic_block_index: builtins.int
+        """The flow graph's entry point address is the first instruction of the
+        entry_basic_block.
+        """
+        @property
+        def edge(self) -> google.protobuf.internal.containers.RepeatedCompositeFieldContainer[global___BinExport2.FlowGraph.Edge]: ...
+        def __init__(
+            self,
+            *,
+            basic_block_index: collections.abc.Iterable[builtins.int] | None = ...,
+            entry_basic_block_index: builtins.int | None = ...,
+            edge: collections.abc.Iterable[global___BinExport2.FlowGraph.Edge] | None = ...,
+        ) -> None: ...
+        def HasField(self, field_name: typing_extensions.Literal["entry_basic_block_index", b"entry_basic_block_index"]) -> builtins.bool: ...
+        def ClearField(self, field_name: typing_extensions.Literal["basic_block_index", b"basic_block_index", "edge", b"edge", "entry_basic_block_index", b"entry_basic_block_index"]) -> None: ...
+
+    @typing_extensions.final
+    class Reference(google.protobuf.message.Message):
+        """Generic reference class used for address comments (deprecated), string
+        references and expression substitutions. It allows referencing from an
+        instruction, operand, expression subtree tuple to a de-duped string in the
+        string table.
+        """
+
+        DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
+        INSTRUCTION_INDEX_FIELD_NUMBER: builtins.int
+        INSTRUCTION_OPERAND_INDEX_FIELD_NUMBER: builtins.int
+        OPERAND_EXPRESSION_INDEX_FIELD_NUMBER: builtins.int
+        STRING_TABLE_INDEX_FIELD_NUMBER: builtins.int
+        instruction_index: builtins.int
+        """Index into the global instruction table."""
+        instruction_operand_index: builtins.int
+        """Index into the operand array local to an instruction."""
+        operand_expression_index: builtins.int
+        """Index into the expression array local to an operand."""
+        string_table_index: builtins.int
+        """Index into the global string table."""
+        def __init__(
+            self,
+            *,
+            instruction_index: builtins.int | None = ...,
+            instruction_operand_index: builtins.int | None = ...,
+            operand_expression_index: builtins.int | None = ...,
+            string_table_index: builtins.int | None = ...,
+        ) -> None: ...
+        def HasField(self, field_name: typing_extensions.Literal["instruction_index", b"instruction_index", "instruction_operand_index", b"instruction_operand_index", "operand_expression_index", b"operand_expression_index", "string_table_index", b"string_table_index"]) -> builtins.bool: ...
+        def ClearField(self, field_name: typing_extensions.Literal["instruction_index", b"instruction_index", "instruction_operand_index", b"instruction_operand_index", "operand_expression_index", b"operand_expression_index", "string_table_index", b"string_table_index"]) -> None: ...
+
+    @typing_extensions.final
+    class DataReference(google.protobuf.message.Message):
+        DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
+        INSTRUCTION_INDEX_FIELD_NUMBER: builtins.int
+        ADDRESS_FIELD_NUMBER: builtins.int
+        instruction_index: builtins.int
+        """Index into the global instruction table."""
+        address: builtins.int
+        """Address being referred."""
+        def __init__(
+            self,
+            *,
+            instruction_index: builtins.int | None = ...,
+            address: builtins.int | None = ...,
+        ) -> None: ...
+        def HasField(self, field_name: typing_extensions.Literal["address", b"address", "instruction_index", b"instruction_index"]) -> builtins.bool: ...
+        def ClearField(self, field_name: typing_extensions.Literal["address", b"address", "instruction_index", b"instruction_index"]) -> None: ...
+
+    @typing_extensions.final
+    class Comment(google.protobuf.message.Message):
+        DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
+        class _Type:
+            ValueType = typing.NewType("ValueType", builtins.int)
+            V: typing_extensions.TypeAlias = ValueType
+
+        class _TypeEnumTypeWrapper(google.protobuf.internal.enum_type_wrapper._EnumTypeWrapper[BinExport2.Comment._Type.ValueType], builtins.type):
+            DESCRIPTOR: google.protobuf.descriptor.EnumDescriptor
+            DEFAULT: BinExport2.Comment._Type.ValueType  # 0
+            """A regular instruction comment. Typically displayed next to the
+            instruction disassembly.
+            """
+            ANTERIOR: BinExport2.Comment._Type.ValueType  # 1
+            """A comment line that is typically displayed before (above) the
+            instruction it refers to.
+            """
+            POSTERIOR: BinExport2.Comment._Type.ValueType  # 2
+            """Like ANTERIOR, but a typically displayed after (below)."""
+            FUNCTION: BinExport2.Comment._Type.ValueType  # 3
+            """Similar to an ANTERIOR comment, but applies to the beginning of an
+            identified function. Programs displaying the proto may choose to render
+            these differently (e.g. above an inferred function signature).
+            """
+            ENUM: BinExport2.Comment._Type.ValueType  # 4
+            """Named constants, bitfields and similar."""
+            LOCATION: BinExport2.Comment._Type.ValueType  # 5
+            """Named locations, usually the target of a jump."""
+            GLOBAL_REFERENCE: BinExport2.Comment._Type.ValueType  # 6
+            """Data cross references."""
+            LOCAL_REFERENCE: BinExport2.Comment._Type.ValueType  # 7
+            """Local/stack variables."""
+
+        class Type(_Type, metaclass=_TypeEnumTypeWrapper): ...
+        DEFAULT: BinExport2.Comment.Type.ValueType  # 0
+        """A regular instruction comment. Typically displayed next to the
+        instruction disassembly.
+        """
+        ANTERIOR: BinExport2.Comment.Type.ValueType  # 1
+        """A comment line that is typically displayed before (above) the
+        instruction it refers to.
+        """
+        POSTERIOR: BinExport2.Comment.Type.ValueType  # 2
+        """Like ANTERIOR, but a typically displayed after (below)."""
+        FUNCTION: BinExport2.Comment.Type.ValueType  # 3
+        """Similar to an ANTERIOR comment, but applies to the beginning of an
+        identified function. Programs displaying the proto may choose to render
+        these differently (e.g. above an inferred function signature).
+        """
+        ENUM: BinExport2.Comment.Type.ValueType  # 4
+        """Named constants, bitfields and similar."""
+        LOCATION: BinExport2.Comment.Type.ValueType  # 5
+        """Named locations, usually the target of a jump."""
+        GLOBAL_REFERENCE: BinExport2.Comment.Type.ValueType  # 6
+        """Data cross references."""
+        LOCAL_REFERENCE: BinExport2.Comment.Type.ValueType  # 7
+        """Local/stack variables."""
+
+        INSTRUCTION_INDEX_FIELD_NUMBER: builtins.int
+        INSTRUCTION_OPERAND_INDEX_FIELD_NUMBER: builtins.int
+        OPERAND_EXPRESSION_INDEX_FIELD_NUMBER: builtins.int
+        STRING_TABLE_INDEX_FIELD_NUMBER: builtins.int
+        REPEATABLE_FIELD_NUMBER: builtins.int
+        TYPE_FIELD_NUMBER: builtins.int
+        instruction_index: builtins.int
+        """Index into the global instruction table. This is here to enable
+        comment processing without having to iterate over all instructions.
+        There is an N:M mapping of instructions to comments.
+        """
+        instruction_operand_index: builtins.int
+        """Index into the operand array local to an instruction."""
+        operand_expression_index: builtins.int
+        """Index into the expression array local to an operand, like in Reference.
+        This is not currently used, but allows to implement expression
+        substitutions.
+        """
+        string_table_index: builtins.int
+        """Index into the global string table."""
+        repeatable: builtins.bool
+        """Comment is propagated to all locations that reference the original
+        location.
+        """
+        type: global___BinExport2.Comment.Type.ValueType
+        def __init__(
+            self,
+            *,
+            instruction_index: builtins.int | None = ...,
+            instruction_operand_index: builtins.int | None = ...,
+            operand_expression_index: builtins.int | None = ...,
+            string_table_index: builtins.int | None = ...,
+            repeatable: builtins.bool | None = ...,
+            type: global___BinExport2.Comment.Type.ValueType | None = ...,
+        ) -> None: ...
+        def HasField(self, field_name: typing_extensions.Literal["instruction_index", b"instruction_index", "instruction_operand_index", b"instruction_operand_index", "operand_expression_index", b"operand_expression_index", "repeatable", b"repeatable", "string_table_index", b"string_table_index", "type", b"type"]) -> builtins.bool: ...
+        def ClearField(self, field_name: typing_extensions.Literal["instruction_index", b"instruction_index", "instruction_operand_index", b"instruction_operand_index", "operand_expression_index", b"operand_expression_index", "repeatable", b"repeatable", "string_table_index", b"string_table_index", "type", b"type"]) -> None: ...
+
+    @typing_extensions.final
+    class Section(google.protobuf.message.Message):
+        DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
+        ADDRESS_FIELD_NUMBER: builtins.int
+        SIZE_FIELD_NUMBER: builtins.int
+        FLAG_R_FIELD_NUMBER: builtins.int
+        FLAG_W_FIELD_NUMBER: builtins.int
+        FLAG_X_FIELD_NUMBER: builtins.int
+        address: builtins.int
+        """Section start address."""
+        size: builtins.int
+        """Section size."""
+        flag_r: builtins.bool
+        """Read flag of the section, True when section is readable."""
+        flag_w: builtins.bool
+        """Write flag of the section, True when section is writable."""
+        flag_x: builtins.bool
+        """Execute flag of the section, True when section is executable."""
+        def __init__(
+            self,
+            *,
+            address: builtins.int | None = ...,
+            size: builtins.int | None = ...,
+            flag_r: builtins.bool | None = ...,
+            flag_w: builtins.bool | None = ...,
+            flag_x: builtins.bool | None = ...,
+        ) -> None: ...
+        def HasField(self, field_name: typing_extensions.Literal["address", b"address", "flag_r", b"flag_r", "flag_w", b"flag_w", "flag_x", b"flag_x", "size", b"size"]) -> builtins.bool: ...
+        def ClearField(self, field_name: typing_extensions.Literal["address", b"address", "flag_r", b"flag_r", "flag_w", b"flag_w", "flag_x", b"flag_x", "size", b"size"]) -> None: ...
+
+    @typing_extensions.final
+    class Library(google.protobuf.message.Message):
+        DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
+        IS_STATIC_FIELD_NUMBER: builtins.int
+        LOAD_ADDRESS_FIELD_NUMBER: builtins.int
+        NAME_FIELD_NUMBER: builtins.int
+        is_static: builtins.bool
+        """If this library is statically linked."""
+        load_address: builtins.int
+        """Address where this library was loaded, 0 if unknown."""
+        name: builtins.str
+        """Name of the library (format is platform-dependent)."""
+        def __init__(
+            self,
+            *,
+            is_static: builtins.bool | None = ...,
+            load_address: builtins.int | None = ...,
+            name: builtins.str | None = ...,
+        ) -> None: ...
+        def HasField(self, field_name: typing_extensions.Literal["is_static", b"is_static", "load_address", b"load_address", "name", b"name"]) -> builtins.bool: ...
+        def ClearField(self, field_name: typing_extensions.Literal["is_static", b"is_static", "load_address", b"load_address", "name", b"name"]) -> None: ...
+
+    @typing_extensions.final
+    class Module(google.protobuf.message.Message):
+        DESCRIPTOR: google.protobuf.descriptor.Descriptor
+
+        NAME_FIELD_NUMBER: builtins.int
+        name: builtins.str
+        """Name, such as Java class name. Platform-dependent."""
+        def __init__(
+            self,
+            *,
+            name: builtins.str | None = ...,
+        ) -> None: ...
+        def HasField(self, field_name: typing_extensions.Literal["name", b"name"]) -> builtins.bool: ...
+        def ClearField(self, field_name: typing_extensions.Literal["name", b"name"]) -> None: ...
+
+    META_INFORMATION_FIELD_NUMBER: builtins.int
+    EXPRESSION_FIELD_NUMBER: builtins.int
+    OPERAND_FIELD_NUMBER: builtins.int
+    MNEMONIC_FIELD_NUMBER: builtins.int
+    INSTRUCTION_FIELD_NUMBER: builtins.int
+    BASIC_BLOCK_FIELD_NUMBER: builtins.int
+    FLOW_GRAPH_FIELD_NUMBER: builtins.int
+    CALL_GRAPH_FIELD_NUMBER: builtins.int
+    STRING_TABLE_FIELD_NUMBER: builtins.int
+    ADDRESS_COMMENT_FIELD_NUMBER: builtins.int
+    COMMENT_FIELD_NUMBER: builtins.int
+    STRING_REFERENCE_FIELD_NUMBER: builtins.int
+    EXPRESSION_SUBSTITUTION_FIELD_NUMBER: builtins.int
+    SECTION_FIELD_NUMBER: builtins.int
+    LIBRARY_FIELD_NUMBER: builtins.int
+    DATA_REFERENCE_FIELD_NUMBER: builtins.int
+    MODULE_FIELD_NUMBER: builtins.int
+    @property
+    def meta_information(self) -> global___BinExport2.Meta: ...
+    @property
+    def expression(self) -> google.protobuf.internal.containers.RepeatedCompositeFieldContainer[global___BinExport2.Expression]: ...
+    @property
+    def operand(self) -> google.protobuf.internal.containers.RepeatedCompositeFieldContainer[global___BinExport2.Operand]: ...
+    @property
+    def mnemonic(self) -> google.protobuf.internal.containers.RepeatedCompositeFieldContainer[global___BinExport2.Mnemonic]: ...
+    @property
+    def instruction(self) -> google.protobuf.internal.containers.RepeatedCompositeFieldContainer[global___BinExport2.Instruction]: ...
+    @property
+    def basic_block(self) -> google.protobuf.internal.containers.RepeatedCompositeFieldContainer[global___BinExport2.BasicBlock]: ...
+    @property
+    def flow_graph(self) -> google.protobuf.internal.containers.RepeatedCompositeFieldContainer[global___BinExport2.FlowGraph]: ...
+    @property
+    def call_graph(self) -> global___BinExport2.CallGraph: ...
+    @property
+    def string_table(self) -> google.protobuf.internal.containers.RepeatedScalarFieldContainer[builtins.str]: ...
+    @property
+    def address_comment(self) -> google.protobuf.internal.containers.RepeatedCompositeFieldContainer[global___BinExport2.Reference]:
+        """No longer written. This is here so that BinDiff can work with older
+        BinExport files.
+        """
+    @property
+    def comment(self) -> google.protobuf.internal.containers.RepeatedCompositeFieldContainer[global___BinExport2.Comment]:
+        """Rich comment index used for BinDiff's comment porting."""
+    @property
+    def string_reference(self) -> google.protobuf.internal.containers.RepeatedCompositeFieldContainer[global___BinExport2.Reference]: ...
+    @property
+    def expression_substitution(self) -> google.protobuf.internal.containers.RepeatedCompositeFieldContainer[global___BinExport2.Reference]: ...
+    @property
+    def section(self) -> google.protobuf.internal.containers.RepeatedCompositeFieldContainer[global___BinExport2.Section]: ...
+    @property
+    def library(self) -> google.protobuf.internal.containers.RepeatedCompositeFieldContainer[global___BinExport2.Library]: ...
+    @property
+    def data_reference(self) -> google.protobuf.internal.containers.RepeatedCompositeFieldContainer[global___BinExport2.DataReference]: ...
+    @property
+    def module(self) -> google.protobuf.internal.containers.RepeatedCompositeFieldContainer[global___BinExport2.Module]: ...
+    def __init__(
+        self,
+        *,
+        meta_information: global___BinExport2.Meta | None = ...,
+        expression: collections.abc.Iterable[global___BinExport2.Expression] | None = ...,
+        operand: collections.abc.Iterable[global___BinExport2.Operand] | None = ...,
+        mnemonic: collections.abc.Iterable[global___BinExport2.Mnemonic] | None = ...,
+        instruction: collections.abc.Iterable[global___BinExport2.Instruction] | None = ...,
+        basic_block: collections.abc.Iterable[global___BinExport2.BasicBlock] | None = ...,
+        flow_graph: collections.abc.Iterable[global___BinExport2.FlowGraph] | None = ...,
+        call_graph: global___BinExport2.CallGraph | None = ...,
+        string_table: collections.abc.Iterable[builtins.str] | None = ...,
+        address_comment: collections.abc.Iterable[global___BinExport2.Reference] | None = ...,
+        comment: collections.abc.Iterable[global___BinExport2.Comment] | None = ...,
+        string_reference: collections.abc.Iterable[global___BinExport2.Reference] | None = ...,
+        expression_substitution: collections.abc.Iterable[global___BinExport2.Reference] | None = ...,
+        section: collections.abc.Iterable[global___BinExport2.Section] | None = ...,
+        library: collections.abc.Iterable[global___BinExport2.Library] | None = ...,
+        data_reference: collections.abc.Iterable[global___BinExport2.DataReference] | None = ...,
+        module: collections.abc.Iterable[global___BinExport2.Module] | None = ...,
+    ) -> None: ...
+    def HasField(self, field_name: typing_extensions.Literal["call_graph", b"call_graph", "meta_information", b"meta_information"]) -> builtins.bool: ...
+    def ClearField(self, field_name: typing_extensions.Literal["address_comment", b"address_comment", "basic_block", b"basic_block", "call_graph", b"call_graph", "comment", b"comment", "data_reference", b"data_reference", "expression", b"expression", "expression_substitution", b"expression_substitution", "flow_graph", b"flow_graph", "instruction", b"instruction", "library", b"library", "meta_information", b"meta_information", "mnemonic", b"mnemonic", "module", b"module", "operand", b"operand", "section", b"section", "string_reference", b"string_reference", "string_table", b"string_table"]) -> None: ...
+
+global___BinExport2 = BinExport2

capa/features/extractors/binexport2/extractor.py

@@ -0,0 +1,130 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+import logging
+from typing import Set, List, Tuple, Iterator
+
+import capa.features.extractors.elf
+import capa.features.extractors.common
+import capa.features.extractors.binexport2.file
+import capa.features.extractors.binexport2.insn
+import capa.features.extractors.binexport2.helpers
+import capa.features.extractors.binexport2.function
+import capa.features.extractors.binexport2.basicblock
+from capa.features.common import OS, Arch, Format, Feature
+from capa.features.address import Address, AbsoluteVirtualAddress
+from capa.features.extractors.binexport2 import (
+    AddressSpace,
+    AnalysisContext,
+    BinExport2Index,
+    FunctionContext,
+    BasicBlockContext,
+    BinExport2Analysis,
+    InstructionContext,
+)
+from capa.features.extractors.base_extractor import (
+    BBHandle,
+    InsnHandle,
+    SampleHashes,
+    FunctionHandle,
+    StaticFeatureExtractor,
+)
+from capa.features.extractors.binexport2.binexport2_pb2 import BinExport2
+
+logger = logging.getLogger(__name__)
+
+
+class BinExport2FeatureExtractor(StaticFeatureExtractor):
+    def __init__(self, be2: BinExport2, buf: bytes):
+        super().__init__(hashes=SampleHashes.from_bytes(buf))
+        self.be2: BinExport2 = be2
+        self.buf: bytes = buf
+        self.idx: BinExport2Index = BinExport2Index(self.be2)
+        self.analysis: BinExport2Analysis = BinExport2Analysis(self.be2, self.idx, self.buf)
+        address_space: AddressSpace = AddressSpace.from_buf(buf, self.analysis.base_address)
+        self.ctx: AnalysisContext = AnalysisContext(self.buf, self.be2, self.idx, self.analysis, address_space)
+
+        self.global_features: List[Tuple[Feature, Address]] = []
+        self.global_features.extend(list(capa.features.extractors.common.extract_format(self.buf)))
+        self.global_features.extend(list(capa.features.extractors.common.extract_os(self.buf)))
+        self.global_features.extend(list(capa.features.extractors.common.extract_arch(self.buf)))
+
+        self.format: Set[str] = set()
+        self.os: Set[str] = set()
+        self.arch: Set[str] = set()
+
+        for feature, _ in self.global_features:
+            assert isinstance(feature.value, str)
+
+            if isinstance(feature, Format):
+                self.format.add(feature.value)
+            elif isinstance(feature, OS):
+                self.os.add(feature.value)
+            elif isinstance(feature, Arch):
+                self.arch.add(feature.value)
+            else:
+                raise ValueError("unexpected global feature: %s", feature)
+
+    def get_base_address(self) -> AbsoluteVirtualAddress:
+        return AbsoluteVirtualAddress(self.analysis.base_address)
+
+    def extract_global_features(self) -> Iterator[Tuple[Feature, Address]]:
+        yield from self.global_features
+
+    def extract_file_features(self) -> Iterator[Tuple[Feature, Address]]:
+        yield from capa.features.extractors.binexport2.file.extract_features(self.be2, self.buf)
+
+    def get_functions(self) -> Iterator[FunctionHandle]:
+        for flow_graph_index, flow_graph in enumerate(self.be2.flow_graph):
+            entry_basic_block_index: int = flow_graph.entry_basic_block_index
+            flow_graph_address: int = self.idx.get_basic_block_address(entry_basic_block_index)
+
+            vertex_idx: int = self.idx.vertex_index_by_address[flow_graph_address]
+            be2_vertex: BinExport2.CallGraph.Vertex = self.be2.call_graph.vertex[vertex_idx]
+
+            # skip thunks
+            if capa.features.extractors.binexport2.helpers.is_vertex_type(
+                be2_vertex, BinExport2.CallGraph.Vertex.Type.THUNK
+            ):
+                continue
+
+            yield FunctionHandle(
+                AbsoluteVirtualAddress(flow_graph_address),
+                inner=FunctionContext(self.ctx, flow_graph_index, self.format, self.os, self.arch),
+            )
+
+    def extract_function_features(self, fh: FunctionHandle) -> Iterator[Tuple[Feature, Address]]:
+        yield from capa.features.extractors.binexport2.function.extract_features(fh)
+
+    def get_basic_blocks(self, fh: FunctionHandle) -> Iterator[BBHandle]:
+        fhi: FunctionContext = fh.inner
+        flow_graph_index: int = fhi.flow_graph_index
+        flow_graph: BinExport2.FlowGraph = self.be2.flow_graph[flow_graph_index]
+
+        for basic_block_index in flow_graph.basic_block_index:
+            basic_block_address: int = self.idx.get_basic_block_address(basic_block_index)
+            yield BBHandle(
+                address=AbsoluteVirtualAddress(basic_block_address),
+                inner=BasicBlockContext(basic_block_index),
+            )
+
+    def extract_basic_block_features(self, fh: FunctionHandle, bbh: BBHandle) -> Iterator[Tuple[Feature, Address]]:
+        yield from capa.features.extractors.binexport2.basicblock.extract_features(fh, bbh)
+
+    def get_instructions(self, fh: FunctionHandle, bbh: BBHandle) -> Iterator[InsnHandle]:
+        bbi: BasicBlockContext = bbh.inner
+        basic_block: BinExport2.BasicBlock = self.be2.basic_block[bbi.basic_block_index]
+        for instruction_index, _, instruction_address in self.idx.basic_block_instructions(basic_block):
+            yield InsnHandle(
+                address=AbsoluteVirtualAddress(instruction_address),
+                inner=InstructionContext(instruction_index),
+            )
+
+    def extract_insn_features(
+        self, fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle
+    ) -> Iterator[Tuple[Feature, Address]]:
+        yield from capa.features.extractors.binexport2.insn.extract_features(fh, bbh, ih)

capa/features/extractors/binexport2/file.py

@@ -0,0 +1,80 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+import io
+import logging
+from typing import Tuple, Iterator
+
+import pefile
+from elftools.elf.elffile import ELFFile
+
+import capa.features.common
+import capa.features.extractors.common
+import capa.features.extractors.pefile
+import capa.features.extractors.elffile
+from capa.features.common import Feature
+from capa.features.address import Address
+from capa.features.extractors.binexport2.binexport2_pb2 import BinExport2
+
+logger = logging.getLogger(__name__)
+
+
+def extract_file_export_names(_be2: BinExport2, buf: bytes) -> Iterator[Tuple[Feature, Address]]:
+    if buf.startswith(capa.features.extractors.common.MATCH_PE):
+        pe: pefile.PE = pefile.PE(data=buf)
+        yield from capa.features.extractors.pefile.extract_file_export_names(pe)
+    elif buf.startswith(capa.features.extractors.common.MATCH_ELF):
+        elf: ELFFile = ELFFile(io.BytesIO(buf))
+        yield from capa.features.extractors.elffile.extract_file_export_names(elf)
+    else:
+        logger.warning("unsupported format")
+
+
+def extract_file_import_names(_be2: BinExport2, buf: bytes) -> Iterator[Tuple[Feature, Address]]:
+    if buf.startswith(capa.features.extractors.common.MATCH_PE):
+        pe: pefile.PE = pefile.PE(data=buf)
+        yield from capa.features.extractors.pefile.extract_file_import_names(pe)
+    elif buf.startswith(capa.features.extractors.common.MATCH_ELF):
+        elf: ELFFile = ELFFile(io.BytesIO(buf))
+        yield from capa.features.extractors.elffile.extract_file_import_names(elf)
+    else:
+        logger.warning("unsupported format")
+
+
+def extract_file_section_names(_be2: BinExport2, buf: bytes) -> Iterator[Tuple[Feature, Address]]:
+    if buf.startswith(capa.features.extractors.common.MATCH_PE):
+        pe: pefile.PE = pefile.PE(data=buf)
+        yield from capa.features.extractors.pefile.extract_file_section_names(pe)
+    elif buf.startswith(capa.features.extractors.common.MATCH_ELF):
+        elf: ELFFile = ELFFile(io.BytesIO(buf))
+        yield from capa.features.extractors.elffile.extract_file_section_names(elf)
+    else:
+        logger.warning("unsupported format")
+
+
+def extract_file_strings(_be2: BinExport2, buf: bytes) -> Iterator[Tuple[Feature, Address]]:
+    yield from capa.features.extractors.common.extract_file_strings(buf)
+
+
+def extract_file_format(_be2: BinExport2, buf: bytes) -> Iterator[Tuple[Feature, Address]]:
+    yield from capa.features.extractors.common.extract_format(buf)
+
+
+def extract_features(be2: BinExport2, buf: bytes) -> Iterator[Tuple[Feature, Address]]:
+    """extract file features"""
+    for file_handler in FILE_HANDLERS:
+        for feature, addr in file_handler(be2, buf):
+            yield feature, addr
+
+
+FILE_HANDLERS = (
+    extract_file_export_names,
+    extract_file_import_names,
+    extract_file_strings,
+    extract_file_section_names,
+    extract_file_format,
+)

capa/features/extractors/binexport2/function.py

@@ -0,0 +1,72 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+from typing import List, Tuple, Iterator
+
+from capa.features.file import FunctionName
+from capa.features.common import Feature, Characteristic
+from capa.features.address import Address, AbsoluteVirtualAddress
+from capa.features.extractors import loops
+from capa.features.extractors.binexport2 import BinExport2Index, FunctionContext
+from capa.features.extractors.base_extractor import FunctionHandle
+from capa.features.extractors.binexport2.binexport2_pb2 import BinExport2
+
+
+def extract_function_calls_to(fh: FunctionHandle) -> Iterator[Tuple[Feature, Address]]:
+    fhi: FunctionContext = fh.inner
+
+    be2: BinExport2 = fhi.ctx.be2
+    idx: BinExport2Index = fhi.ctx.idx
+
+    flow_graph_index: int = fhi.flow_graph_index
+    flow_graph_address: int = idx.flow_graph_address_by_index[flow_graph_index]
+    vertex_index: int = idx.vertex_index_by_address[flow_graph_address]
+
+    for caller_index in idx.callers_by_vertex_index[vertex_index]:
+        caller: BinExport2.CallGraph.Vertex = be2.call_graph.vertex[caller_index]
+        caller_address: int = caller.address
+        yield Characteristic("calls to"), AbsoluteVirtualAddress(caller_address)
+
+
+def extract_function_loop(fh: FunctionHandle) -> Iterator[Tuple[Feature, Address]]:
+    fhi: FunctionContext = fh.inner
+
+    be2: BinExport2 = fhi.ctx.be2
+
+    flow_graph_index: int = fhi.flow_graph_index
+    flow_graph: BinExport2.FlowGraph = be2.flow_graph[flow_graph_index]
+
+    edges: List[Tuple[int, int]] = []
+    for edge in flow_graph.edge:
+        edges.append((edge.source_basic_block_index, edge.target_basic_block_index))
+
+    if loops.has_loop(edges):
+        yield Characteristic("loop"), fh.address
+
+
+def extract_function_name(fh: FunctionHandle) -> Iterator[Tuple[Feature, Address]]:
+    fhi: FunctionContext = fh.inner
+
+    be2: BinExport2 = fhi.ctx.be2
+    idx: BinExport2Index = fhi.ctx.idx
+    flow_graph_index: int = fhi.flow_graph_index
+
+    flow_graph_address: int = idx.flow_graph_address_by_index[flow_graph_index]
+    vertex_index: int = idx.vertex_index_by_address[flow_graph_address]
+    vertex: BinExport2.CallGraph.Vertex = be2.call_graph.vertex[vertex_index]
+
+    if vertex.HasField("mangled_name"):
+        yield FunctionName(vertex.mangled_name), fh.address
+
+
+def extract_features(fh: FunctionHandle) -> Iterator[Tuple[Feature, Address]]:
+    for func_handler in FUNCTION_HANDLERS:
+        for feature, addr in func_handler(fh):
+            yield feature, addr
+
+
+FUNCTION_HANDLERS = (extract_function_calls_to, extract_function_loop, extract_function_name)

capa/features/extractors/binexport2/helpers.py

@@ -0,0 +1,650 @@
+# Copyright (C) 2024 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+import re
+from typing import Set, Dict, List, Tuple, Union, Iterator, Optional
+from collections import defaultdict
+from dataclasses import dataclass
+
+import capa.features.extractors.helpers
+import capa.features.extractors.binexport2.helpers
+from capa.features.common import ARCH_I386, ARCH_AMD64, ARCH_AARCH64
+from capa.features.extractors.binexport2.binexport2_pb2 import BinExport2
+
+HAS_ARCH32 = {ARCH_I386}
+HAS_ARCH64 = {ARCH_AARCH64, ARCH_AMD64}
+
+HAS_ARCH_INTEL = {ARCH_I386, ARCH_AMD64}
+HAS_ARCH_ARM = {ARCH_AARCH64}
+
+
+def mask_immediate(arch: Set[str], immediate: int) -> int:
+    if arch & HAS_ARCH64:
+        immediate &= 0xFFFFFFFFFFFFFFFF
+    elif arch & HAS_ARCH32:
+        immediate &= 0xFFFFFFFF
+    return immediate
+
+
+def twos_complement(arch: Set[str], immediate: int, default: Optional[int] = None) -> int:
+    if default is not None:
+        return capa.features.extractors.helpers.twos_complement(immediate, default)
+    elif arch & HAS_ARCH64:
+        return capa.features.extractors.helpers.twos_complement(immediate, 64)
+    elif arch & HAS_ARCH32:
+        return capa.features.extractors.helpers.twos_complement(immediate, 32)
+    return immediate
+
+
+def is_address_mapped(be2: BinExport2, address: int) -> bool:
+    """return True if the given address is mapped"""
+    sections_with_perms: Iterator[BinExport2.Section] = filter(lambda s: s.flag_r or s.flag_w or s.flag_x, be2.section)
+    return any(section.address <= address < section.address + section.size for section in sections_with_perms)
+
+
+def is_vertex_type(vertex: BinExport2.CallGraph.Vertex, type_: BinExport2.CallGraph.Vertex.Type.ValueType) -> bool:
+    return vertex.HasField("type") and vertex.type == type_
+
+
+# internal to `build_expression_tree`
+# this is unstable: it is subject to change, so don't rely on it!
+def _prune_expression_tree_empty_shifts(
+    be2: BinExport2,
+    operand: BinExport2.Operand,
+    expression_tree: List[List[int]],
+    tree_index: int,
+):
+    expression_index = operand.expression_index[tree_index]
+    expression = be2.expression[expression_index]
+    children_tree_indexes: List[int] = expression_tree[tree_index]
+
+    if expression.type == BinExport2.Expression.OPERATOR:
+        if len(children_tree_indexes) == 0 and expression.symbol in ("lsl", "lsr"):
+            # Ghidra may emit superfluous lsl nodes with no children.
+            # https://github.com/mandiant/capa/pull/2340/files#r1750003919
+            # Which is maybe: https://github.com/NationalSecurityAgency/ghidra/issues/6821#issuecomment-2295394697
+            #
+            # Which seems to be as if the shift wasn't there (shift of #0)
+            # so we want to remove references to this node from any parent nodes.
+            for tree_node in expression_tree:
+                if tree_index in tree_node:
+                    tree_node.remove(tree_index)
+
+            return
+
+    for child_tree_index in children_tree_indexes:
+        _prune_expression_tree_empty_shifts(be2, operand, expression_tree, child_tree_index)
+
+
+# internal to `build_expression_tree`
+# this is unstable: it is subject to change, so don't rely on it!
+def _prune_expression_tree_empty_commas(
+    be2: BinExport2,
+    operand: BinExport2.Operand,
+    expression_tree: List[List[int]],
+    tree_index: int,
+):
+    expression_index = operand.expression_index[tree_index]
+    expression = be2.expression[expression_index]
+    children_tree_indexes: List[int] = expression_tree[tree_index]
+
+    if expression.type == BinExport2.Expression.OPERATOR:
+        if len(children_tree_indexes) == 1 and expression.symbol == ",":
+            # Due to the above pruning of empty LSL or LSR expressions,
+            # the parents might need to be fixed up.
+            #
+            # Specifically, if the pruned node was part of a comma list with two children,
+            # now there's only a single child, which renders as an extra comma,
+            # so we replace references to the comma node with the immediate child.
+            #
+            # A more correct way of doing this might be to walk up the parents and do fixups,
+            # but I'm not quite sure how to do this yet. Just do two passes right now.
+            child = children_tree_indexes[0]
+
+            for tree_node in expression_tree:
+                tree_node.index
+                if tree_index in tree_node:
+                    tree_node[tree_node.index(tree_index)] = child
+
+            return
+
+    for child_tree_index in children_tree_indexes:
+        _prune_expression_tree_empty_commas(be2, operand, expression_tree, child_tree_index)
+
+
+# internal to `build_expression_tree`
+# this is unstable: it is subject to change, so don't rely on it!
+def _prune_expression_tree(
+    be2: BinExport2,
+    operand: BinExport2.Operand,
+    expression_tree: List[List[int]],
+):
+    _prune_expression_tree_empty_shifts(be2, operand, expression_tree, 0)
+    _prune_expression_tree_empty_commas(be2, operand, expression_tree, 0)
+
+
+# this is unstable: it is subject to change, so don't rely on it!
+def _build_expression_tree(
+    be2: BinExport2,
+    operand: BinExport2.Operand,
+) -> List[List[int]]:
+    # The reconstructed expression tree layout, linking parent nodes to their children.
+    #
+    # There is one list of integers for each expression in the operand.
+    # These integers are indexes of other expressions in the same operand,
+    # which are the children of that expression.
+    #
+    # So:
+    #
+    #   [ [1, 3], [2], [], [4], [5], []]
+    #
+    # means the first expression has two children, at index 1 and 3,
+    # and the tree looks like:
+    #
+    #        0
+    #       / \
+    #      1   3
+    #      |   |
+    #      2   4
+    #          |
+    #          5
+    #
+    # Remember, these are the indices into the entries in operand.expression_index.
+    if len(operand.expression_index) == 0:
+        # Ghidra bug where empty operands (no expressions) may
+        # exist (see https://github.com/NationalSecurityAgency/ghidra/issues/6817)
+        return []
+
+    tree: List[List[int]] = []
+    for i, expression_index in enumerate(operand.expression_index):
+        children = []
+
+        # scan all subsequent expressions, looking for those that have parent_index == current.expression_index
+        for j, candidate_index in enumerate(operand.expression_index[i + 1 :]):
+            candidate = be2.expression[candidate_index]
+
+            if candidate.parent_index == expression_index:
+                children.append(i + j + 1)
+
+        tree.append(children)
+
+    _prune_expression_tree(be2, operand, tree)
+    _prune_expression_tree(be2, operand, tree)
+
+    return tree
+
+
+def _fill_operand_expression_list(
+    be2: BinExport2,
+    operand: BinExport2.Operand,
+    expression_tree: List[List[int]],
+    tree_index: int,
+    expression_list: List[BinExport2.Expression],
+):
+    """
+    Walk the given expression tree and collect the expression nodes in-order.
+    """
+    expression_index = operand.expression_index[tree_index]
+    expression = be2.expression[expression_index]
+    children_tree_indexes: List[int] = expression_tree[tree_index]
+
+    if expression.type == BinExport2.Expression.REGISTER:
+        assert len(children_tree_indexes) == 0
+        expression_list.append(expression)
+        return
+
+    elif expression.type == BinExport2.Expression.SYMBOL:
+        assert len(children_tree_indexes) <= 1
+        expression_list.append(expression)
+
+        if len(children_tree_indexes) == 0:
+            return
+        elif len(children_tree_indexes) == 1:
+            # like: v
+            # from: mov v0.D[0x1], x9
+            #           |
+            #           0
+            #           .
+            #           |
+            #           D
+            child_index = children_tree_indexes[0]
+            _fill_operand_expression_list(be2, operand, expression_tree, child_index, expression_list)
+            return
+        else:
+            raise NotImplementedError(len(children_tree_indexes))
+
+    elif expression.type == BinExport2.Expression.IMMEDIATE_INT:
+        assert len(children_tree_indexes) == 0
+        expression_list.append(expression)
+        return
+
+    elif expression.type == BinExport2.Expression.SIZE_PREFIX:
+        # like: b4
+        #
+        # We might want to use this occasionally, such as to disambiguate the
+        # size of MOVs into/out of memory. But I'm not sure when/where we need that yet.
+        #
+        # IDA spams this size prefix hint *everywhere*, so we can't rely on the exporter
+        # to provide it only when necessary.
+        assert len(children_tree_indexes) == 1
+        child_index = children_tree_indexes[0]
+        _fill_operand_expression_list(be2, operand, expression_tree, child_index, expression_list)
+        return
+
+    elif expression.type == BinExport2.Expression.OPERATOR:
+        if len(children_tree_indexes) == 1:
+            # prefix operator, like "ds:"
+            expression_list.append(expression)
+            child_index = children_tree_indexes[0]
+            _fill_operand_expression_list(be2, operand, expression_tree, child_index, expression_list)
+            return
+
+        elif len(children_tree_indexes) == 2:
+            # infix operator: like "+" in "ebp+10"
+            child_a = children_tree_indexes[0]
+            child_b = children_tree_indexes[1]
+            _fill_operand_expression_list(be2, operand, expression_tree, child_a, expression_list)
+            expression_list.append(expression)
+            _fill_operand_expression_list(be2, operand, expression_tree, child_b, expression_list)
+            return
+
+        elif len(children_tree_indexes) == 3:
+            # infix operator: like "+" in "ebp+ecx+10"
+            child_a = children_tree_indexes[0]
+            child_b = children_tree_indexes[1]
+            child_c = children_tree_indexes[2]
+            _fill_operand_expression_list(be2, operand, expression_tree, child_a, expression_list)
+            expression_list.append(expression)
+            _fill_operand_expression_list(be2, operand, expression_tree, child_b, expression_list)
+            expression_list.append(expression)
+            _fill_operand_expression_list(be2, operand, expression_tree, child_c, expression_list)
+            return
+
+        else:
+            raise NotImplementedError(len(children_tree_indexes))
+
+    elif expression.type == BinExport2.Expression.DEREFERENCE:
+        assert len(children_tree_indexes) == 1
+        expression_list.append(expression)
+
+        child_index = children_tree_indexes[0]
+        _fill_operand_expression_list(be2, operand, expression_tree, child_index, expression_list)
+        return
+
+    elif expression.type == BinExport2.Expression.IMMEDIATE_FLOAT:
+        raise NotImplementedError(expression.type)
+
+    else:
+        raise NotImplementedError(expression.type)
+
+
+def get_operand_expressions(be2: BinExport2, op: BinExport2.Operand) -> List[BinExport2.Expression]:
+    tree = _build_expression_tree(be2, op)
+
+    expressions: List[BinExport2.Expression] = []
+    _fill_operand_expression_list(be2, op, tree, 0, expressions)
+
+    return expressions
+
+
+def get_operand_register_expression(be2: BinExport2, operand: BinExport2.Operand) -> Optional[BinExport2.Expression]:
+    if len(operand.expression_index) == 1:
+        expression: BinExport2.Expression = be2.expression[operand.expression_index[0]]
+        if expression.type == BinExport2.Expression.REGISTER:
+            return expression
+    return None
+
+
+def get_operand_immediate_expression(be2: BinExport2, operand: BinExport2.Operand) -> Optional[BinExport2.Expression]:
+    if len(operand.expression_index) == 1:
+        # - type: IMMEDIATE_INT
+        #   immediate: 20588728364
+        #   parent_index: 0
+        expression: BinExport2.Expression = be2.expression[operand.expression_index[0]]
+        if expression.type == BinExport2.Expression.IMMEDIATE_INT:
+            return expression
+
+    elif len(operand.expression_index) == 2:
+        # from IDA, which provides a size hint for every operand,
+        # we get the following pattern for immediate constants:
+        #
+        # - type: SIZE_PREFIX
+        #   symbol: "b8"
+        # - type: IMMEDIATE_INT
+        #   immediate: 20588728364
+        #   parent_index: 0
+        expression0: BinExport2.Expression = be2.expression[operand.expression_index[0]]
+        expression1: BinExport2.Expression = be2.expression[operand.expression_index[1]]
+
+        if expression0.type == BinExport2.Expression.SIZE_PREFIX:
+            if expression1.type == BinExport2.Expression.IMMEDIATE_INT:
+                return expression1
+
+    return None
+
+
+def get_instruction_mnemonic(be2: BinExport2, instruction: BinExport2.Instruction) -> str:
+    return be2.mnemonic[instruction.mnemonic_index].name.lower()
+
+
+def get_instruction_operands(be2: BinExport2, instruction: BinExport2.Instruction) -> List[BinExport2.Operand]:
+    return [be2.operand[operand_index] for operand_index in instruction.operand_index]
+
+
+def split_with_delimiters(s: str, delimiters: Tuple[str, ...]) -> Iterator[str]:
+    """
+    Splits a string by any of the provided delimiter characters,
+    including the delimiters in the results.
+
+    Args:
+        string: The string to split.
+        delimiters: A string containing the characters to use as delimiters.
+    """
+    start = 0
+    for i, char in enumerate(s):
+        if char in delimiters:
+            yield s[start:i]
+            yield char
+            start = i + 1
+
+    if start < len(s):
+        yield s[start:]
+
+
+BinExport2OperandPattern = Union[str, Tuple[str, ...]]
+
+
+@dataclass
+class BinExport2InstructionPattern:
+    """
+    This describes a way to match disassembled instructions, with mnemonics and operands.
+
+    You can specify constraints on the instruction, via:
+      - the mnemonics, like "mov",
+      - number of operands, and
+      - format of each operand, "[reg, reg, #int]".
+
+    During matching, you can also capture a single element, to see its concrete value.
+    For example, given the pattern:
+
+        mov reg0, #int0  ; capture int0
+
+    and the instruction:
+
+        mov eax, 1
+
+    Then the capture will contain the immediate integer 1.
+
+    This matcher uses the BinExport2 data layout under the hood.
+    """
+
+    mnemonics: Tuple[str, ...]
+    operands: Tuple[Union[str, BinExport2OperandPattern], ...]
+    capture: Optional[str]
+
+    @classmethod
+    def from_str(cls, query: str):
+        """
+        Parse a pattern string into a Pattern instance.
+        The supported syntax is like this:
+
+            br      reg
+            br      reg                          ; capture reg
+            br      reg(stack)                   ; capture reg
+            br      reg(not-stack)               ; capture reg
+            mov     reg0, reg1                   ; capture reg0
+            adrp    reg, #int                    ; capture #int
+            add     reg, reg, #int               ; capture #int
+            ldr     reg0, [reg1]                 ; capture reg1
+            ldr|str reg, [reg, #int]             ; capture #int
+            ldr|str reg, [reg(stack), #int]      ; capture #int
+            ldr|str reg, [reg(not-stack), #int]  ; capture #int
+            ldr|str reg, [reg, #int]!            ; capture #int
+            ldr|str reg, [reg], #int             ; capture #int
+            ldp|stp reg, reg, [reg, #int]        ; capture #int
+            ldp|stp reg, reg, [reg, #int]!       ; capture #int
+            ldp|stp reg, reg, [reg], #int        ; capture #int
+        """
+        #
+        # The implementation of the parser here is obviously ugly.
+        # Its handwritten and probably fragile. But since we don't
+        # expect this to be widely used, its probably ok.
+        # Don't hesitate to rewrite this if it becomes more important.
+        #
+        # Note that this doesn't have to be very performant.
+        # We expect these patterns to be parsed once upfront and then reused
+        # (globally at the module level?) rather than within any loop.
+        #
+
+        pattern, _, comment = query.strip().partition(";")
+
+        # we don't support fs: yet
+        assert ":" not in pattern
+
+        # from "capture #int" to "#int"
+        if comment:
+            comment = comment.strip()
+            assert comment.startswith("capture ")
+            capture = comment[len("capture ") :]
+        else:
+            capture = None
+
+        # from "ldr|str ..." to ["ldr", "str"]
+        pattern = pattern.strip()
+        mnemonic, _, rest = pattern.partition(" ")
+        mnemonics = mnemonic.split("|")
+
+        operands: List[Union[str, Tuple[str, ...]]] = []
+        while rest:
+            rest = rest.strip()
+            if not rest.startswith("["):
+                # If its not a dereference, which looks like `[op, op, op, ...]`,
+                # then its a simple operand, which we can split by the next comma.
+                operand, _, rest = rest.partition(", ")
+                rest = rest.strip()
+                operands.append(operand)
+
+            else:
+                # This looks like a dereference, something like `[op, op, op, ...]`.
+                # Since these can't be nested, look for the next ] and then parse backwards.
+                deref_end = rest.index("]")
+                try:
+                    deref_end = rest.index(", ", deref_end)
+                    deref_end += len(", ")
+                except ValueError:
+                    deref = rest
+                    rest = ""
+                else:
+                    deref = rest[:deref_end]
+                    rest = rest[deref_end:]
+                    rest = rest.strip()
+                    deref = deref.rstrip(" ")
+                    deref = deref.rstrip(",")
+
+                # like: [reg, #int]!
+                has_postindex_writeback = deref.endswith("!")
+
+                deref = deref.rstrip("!")
+                deref = deref.rstrip("]")
+                deref = deref.lstrip("[")
+
+                parts = tuple(split_with_delimiters(deref, (",", "+", "*")))
+                parts = tuple(s.strip() for s in parts)
+
+                # emit operands in this order to match
+                # how BinExport2 expressions are flatted
+                # by get_operand_expressions
+                if has_postindex_writeback:
+                    operands.append(("!", "[") + parts)
+                else:
+                    operands.append(("[",) + parts)
+
+        for operand in operands:  # type: ignore
+            # Try to ensure we've parsed the operands correctly.
+            # This is just sanity checking.
+            for o in (operand,) if isinstance(operand, str) else operand:
+                # operands can look like:
+                #  - reg
+                #  - reg0
+                #  - reg(stack)
+                #  - reg0(stack)
+                #  - reg(not-stack)
+                #  - reg0(not-stack)
+                #  - #int
+                #  - #int0
+                # and a limited set of supported operators.
+                # use an inline regex so that its easy to read. not perf critical.
+                assert re.match(r"^(reg|#int)[0-9]?(\(stack\)|\(not-stack\))?$", o) or o in ("[", ",", "!", "+", "*")
+
+        return cls(tuple(mnemonics), tuple(operands), capture)
+
+    @dataclass
+    class MatchResult:
+        operand_index: int
+        expression_index: int
+        expression: BinExport2.Expression
+
+    def match(
+        self, mnemonic: str, operand_expressions: List[List[BinExport2.Expression]]
+    ) -> Optional["BinExport2InstructionPattern.MatchResult"]:
+        """
+        Match the given BinExport2 data against this pattern.
+
+        The BinExport2 expression tree must have been flattened, such as with
+        capa.features.extractors.binexport2.helpers.get_operand_expressions.
+
+        If there's a match, the captured Expression instance is returned.
+        Otherwise, you get None back.
+        """
+        if mnemonic not in self.mnemonics:
+            return None
+
+        if len(self.operands) != len(operand_expressions):
+            return None
+
+        captured = None
+
+        for operand_index, found_expressions in enumerate(operand_expressions):
+            wanted_expressions = self.operands[operand_index]
+
+            # from `"reg"` to `("reg", )`
+            if isinstance(wanted_expressions, str):
+                wanted_expressions = (wanted_expressions,)
+            assert isinstance(wanted_expressions, tuple)
+
+            if len(wanted_expressions) != len(found_expressions):
+                return None
+
+            for expression_index, (wanted_expression, found_expression) in enumerate(
+                zip(wanted_expressions, found_expressions)
+            ):
+                if wanted_expression.startswith("reg"):
+                    if found_expression.type != BinExport2.Expression.REGISTER:
+                        return None
+
+                    if wanted_expression.endswith(")"):
+                        if wanted_expression.endswith("(not-stack)"):
+                            # intel 64: rsp, esp, sp,
+                            # intel 32: ebp, ebp, bp
+                            # arm: sp
+                            register_name = found_expression.symbol.lower()
+                            if register_name in ("rsp", "esp", "sp", "rbp", "ebp", "bp"):
+                                return None
+
+                        elif wanted_expression.endswith("(stack)"):
+                            register_name = found_expression.symbol.lower()
+                            if register_name not in ("rsp", "esp", "sp", "rbp", "ebp", "bp"):
+                                return None
+
+                        else:
+                            raise ValueError("unexpected expression suffix", wanted_expression)
+
+                    if self.capture == wanted_expression:
+                        captured = BinExport2InstructionPattern.MatchResult(
+                            operand_index, expression_index, found_expression
+                        )
+
+                elif wanted_expression.startswith("#int"):
+                    if found_expression.type != BinExport2.Expression.IMMEDIATE_INT:
+                        return None
+
+                    if self.capture == wanted_expression:
+                        captured = BinExport2InstructionPattern.MatchResult(
+                            operand_index, expression_index, found_expression
+                        )
+
+                elif wanted_expression == "[":
+                    if found_expression.type != BinExport2.Expression.DEREFERENCE:
+                        return None
+
+                elif wanted_expression in (",", "!", "+", "*"):
+                    if found_expression.type != BinExport2.Expression.OPERATOR:
+                        return None
+
+                    if found_expression.symbol != wanted_expression:
+                        return None
+
+                else:
+                    raise ValueError(found_expression)
+
+        if captured:
+            return captured
+        else:
+            # There were no captures, so
+            # return arbitrary non-None expression
+            return BinExport2InstructionPattern.MatchResult(operand_index, expression_index, found_expression)
+
+
+class BinExport2InstructionPatternMatcher:
+    """Index and match a collection of instruction patterns."""
+
+    def __init__(self, queries: List[BinExport2InstructionPattern]):
+        self.queries = queries
+        # shard the patterns by (mnemonic, #operands)
+        self._index: Dict[Tuple[str, int], List[BinExport2InstructionPattern]] = defaultdict(list)
+
+        for query in queries:
+            for mnemonic in query.mnemonics:
+                self._index[(mnemonic.lower(), len(query.operands))].append(query)
+
+    @classmethod
+    def from_str(cls, patterns: str):
+        return cls(
+            [
+                BinExport2InstructionPattern.from_str(line)
+                for line in filter(
+                    lambda line: not line.startswith("#"), (line.strip() for line in patterns.split("\n"))
+                )
+            ]
+        )
+
+    def match(
+        self, mnemonic: str, operand_expressions: List[List[BinExport2.Expression]]
+    ) -> Optional[BinExport2InstructionPattern.MatchResult]:
+        queries = self._index.get((mnemonic.lower(), len(operand_expressions)), [])
+        for query in queries:
+            captured = query.match(mnemonic.lower(), operand_expressions)
+            if captured:
+                return captured
+
+        return None
+
+    def match_with_be2(
+        self, be2: BinExport2, instruction_index: int
+    ) -> Optional[BinExport2InstructionPattern.MatchResult]:
+        instruction: BinExport2.Instruction = be2.instruction[instruction_index]
+        mnemonic: str = get_instruction_mnemonic(be2, instruction)
+
+        if (mnemonic.lower(), len(instruction.operand_index)) not in self._index:
+            # verify that we might have a hit before we realize the operand expression list
+            return None
+
+        operands = []
+        for operand_index in instruction.operand_index:
+            operands.append(get_operand_expressions(be2, be2.operand[operand_index]))
+
+        return self.match(mnemonic, operands)

capa/features/extractors/binexport2/insn.py

@@ -0,0 +1,254 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+import logging
+from typing import List, Tuple, Iterator
+
+import capa.features.extractors.helpers
+import capa.features.extractors.strings
+import capa.features.extractors.binexport2.helpers
+import capa.features.extractors.binexport2.arch.arm.insn
+import capa.features.extractors.binexport2.arch.intel.insn
+from capa.features.insn import API, Mnemonic
+from capa.features.common import Bytes, String, Feature, Characteristic
+from capa.features.address import Address, AbsoluteVirtualAddress
+from capa.features.extractors.binexport2 import (
+    AddressSpace,
+    AnalysisContext,
+    BinExport2Index,
+    FunctionContext,
+    ReadMemoryError,
+    BinExport2Analysis,
+    InstructionContext,
+)
+from capa.features.extractors.base_extractor import BBHandle, InsnHandle, FunctionHandle
+from capa.features.extractors.binexport2.helpers import HAS_ARCH_ARM, HAS_ARCH_INTEL
+from capa.features.extractors.binexport2.binexport2_pb2 import BinExport2
+
+logger = logging.getLogger(__name__)
+
+
+def extract_insn_api_features(fh: FunctionHandle, _bbh: BBHandle, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
+    fhi: FunctionContext = fh.inner
+    ii: InstructionContext = ih.inner
+
+    be2: BinExport2 = fhi.ctx.be2
+    be2_index: BinExport2Index = fhi.ctx.idx
+    be2_analysis: BinExport2Analysis = fhi.ctx.analysis
+    insn: BinExport2.Instruction = be2.instruction[ii.instruction_index]
+
+    for addr in insn.call_target:
+        addr = be2_analysis.thunks.get(addr, addr)
+
+        if addr not in be2_index.vertex_index_by_address:
+            # disassembler did not define function at address
+            logger.debug("0x%x is not a vertex", addr)
+            continue
+
+        vertex_idx: int = be2_index.vertex_index_by_address[addr]
+        vertex: BinExport2.CallGraph.Vertex = be2.call_graph.vertex[vertex_idx]
+
+        if not capa.features.extractors.binexport2.helpers.is_vertex_type(
+            vertex, BinExport2.CallGraph.Vertex.Type.IMPORTED
+        ):
+            continue
+
+        if not vertex.HasField("mangled_name"):
+            logger.debug("vertex %d does not have mangled_name", vertex_idx)
+            continue
+
+        api_name: str = vertex.mangled_name
+        for name in capa.features.extractors.helpers.generate_symbols("", api_name):
+            yield API(name), ih.address
+
+
+def extract_insn_number_features(
+    fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
+    fhi: FunctionContext = fh.inner
+
+    if fhi.arch & HAS_ARCH_INTEL:
+        yield from capa.features.extractors.binexport2.arch.intel.insn.extract_insn_number_features(fh, bbh, ih)
+    elif fhi.arch & HAS_ARCH_ARM:
+        yield from capa.features.extractors.binexport2.arch.arm.insn.extract_insn_number_features(fh, bbh, ih)
+
+
+def extract_insn_bytes_features(fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
+    fhi: FunctionContext = fh.inner
+    ii: InstructionContext = ih.inner
+
+    ctx: AnalysisContext = fhi.ctx
+    be2: BinExport2 = ctx.be2
+    idx: BinExport2Index = ctx.idx
+    address_space: AddressSpace = ctx.address_space
+
+    instruction_index: int = ii.instruction_index
+
+    if instruction_index in idx.string_reference_index_by_source_instruction_index:
+        # disassembler already identified string reference from instruction
+        return
+
+    reference_addresses: List[int] = []
+
+    if instruction_index in idx.data_reference_index_by_source_instruction_index:
+        for data_reference_index in idx.data_reference_index_by_source_instruction_index[instruction_index]:
+            data_reference: BinExport2.DataReference = be2.data_reference[data_reference_index]
+            data_reference_address: int = data_reference.address
+
+            if data_reference_address in idx.insn_address_by_index:
+                # appears to be code
+                continue
+
+            reference_addresses.append(data_reference_address)
+
+    for reference_address in reference_addresses:
+        try:
+            # if at end of segment then there might be an overrun here.
+            buf: bytes = address_space.read_memory(reference_address, 0x100)
+        except ReadMemoryError:
+            logger.debug("failed to read memory: 0x%x", reference_address)
+            continue
+
+        if capa.features.extractors.helpers.all_zeros(buf):
+            continue
+
+        is_string: bool = False
+
+        # note: we *always* break after the first iteration
+        for s in capa.features.extractors.strings.extract_ascii_strings(buf):
+            if s.offset != 0:
+                break
+
+            yield String(s.s), ih.address
+            is_string = True
+            break
+
+        # note: we *always* break after the first iteration
+        for s in capa.features.extractors.strings.extract_unicode_strings(buf):
+            if s.offset != 0:
+                break
+
+            yield String(s.s), ih.address
+            is_string = True
+            break
+
+        if not is_string:
+            yield Bytes(buf), ih.address
+
+
+def extract_insn_string_features(
+    fh: FunctionHandle, _bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
+    fhi: FunctionContext = fh.inner
+    ii: InstructionContext = ih.inner
+
+    be2: BinExport2 = fhi.ctx.be2
+    idx: BinExport2Index = fhi.ctx.idx
+
+    instruction_index: int = ii.instruction_index
+
+    if instruction_index in idx.string_reference_index_by_source_instruction_index:
+        for string_reference_index in idx.string_reference_index_by_source_instruction_index[instruction_index]:
+            string_reference: BinExport2.Reference = be2.string_reference[string_reference_index]
+            string_index: int = string_reference.string_table_index
+            string: str = be2.string_table[string_index]
+            yield String(string), ih.address
+
+
+def extract_insn_offset_features(
+    fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
+    fhi: FunctionContext = fh.inner
+
+    if fhi.arch & HAS_ARCH_INTEL:
+        yield from capa.features.extractors.binexport2.arch.intel.insn.extract_insn_offset_features(fh, bbh, ih)
+    elif fhi.arch & HAS_ARCH_ARM:
+        yield from capa.features.extractors.binexport2.arch.arm.insn.extract_insn_offset_features(fh, bbh, ih)
+
+
+def extract_insn_nzxor_characteristic_features(
+    fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
+    fhi: FunctionContext = fh.inner
+
+    if fhi.arch & HAS_ARCH_INTEL:
+        yield from capa.features.extractors.binexport2.arch.intel.insn.extract_insn_nzxor_characteristic_features(
+            fh, bbh, ih
+        )
+    elif fhi.arch & HAS_ARCH_ARM:
+        yield from capa.features.extractors.binexport2.arch.arm.insn.extract_insn_nzxor_characteristic_features(
+            fh, bbh, ih
+        )
+
+
+def extract_insn_mnemonic_features(
+    fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
+    fhi: FunctionContext = fh.inner
+    ii: InstructionContext = ih.inner
+
+    be2: BinExport2 = fhi.ctx.be2
+
+    instruction: BinExport2.Instruction = be2.instruction[ii.instruction_index]
+    mnemonic: BinExport2.Mnemonic = be2.mnemonic[instruction.mnemonic_index]
+    mnemonic_name: str = mnemonic.name.lower()
+    yield Mnemonic(mnemonic_name), ih.address
+
+
+def extract_function_calls_from(fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
+    """extract functions calls from features
+
+    most relevant at the function scope;
+    however, its most efficient to extract at the instruction scope.
+    """
+    fhi: FunctionContext = fh.inner
+    ii: InstructionContext = ih.inner
+
+    be2: BinExport2 = fhi.ctx.be2
+
+    instruction: BinExport2.Instruction = be2.instruction[ii.instruction_index]
+    for call_target_address in instruction.call_target:
+        addr: AbsoluteVirtualAddress = AbsoluteVirtualAddress(call_target_address)
+        yield Characteristic("calls from"), addr
+
+        if fh.address == addr:
+            yield Characteristic("recursive call"), addr
+
+
+def extract_function_indirect_call_characteristic_features(
+    fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
+    fhi: FunctionContext = fh.inner
+
+    if fhi.arch & HAS_ARCH_INTEL:
+        yield from capa.features.extractors.binexport2.arch.intel.insn.extract_function_indirect_call_characteristic_features(
+            fh, bbh, ih
+        )
+    elif fhi.arch & HAS_ARCH_ARM:
+        yield from capa.features.extractors.binexport2.arch.arm.insn.extract_function_indirect_call_characteristic_features(
+            fh, bbh, ih
+        )
+
+
+def extract_features(f: FunctionHandle, bbh: BBHandle, insn: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
+    """extract instruction features"""
+    for inst_handler in INSTRUCTION_HANDLERS:
+        for feature, ea in inst_handler(f, bbh, insn):
+            yield feature, ea
+
+
+INSTRUCTION_HANDLERS = (
+    extract_insn_api_features,
+    extract_insn_number_features,
+    extract_insn_bytes_features,
+    extract_insn_string_features,
+    extract_insn_offset_features,
+    extract_insn_nzxor_characteristic_features,
+    extract_insn_mnemonic_features,
+    extract_function_calls_from,
+    extract_function_indirect_call_characteristic_features,
+)

capa/features/extractors/binja/file.py

@@ -5,8 +5,6 @@
 # Unless required by applicable law or agreed to in writing, software distributed under the License
 #  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and limitations under the License.
-
-import struct
 from typing import Tuple, Iterator
 
 from binaryninja import Segment, BinaryView, SymbolType, SymbolBinding
@@ -20,56 +18,24 @@ from capa.features.address import NO_ADDRESS, Address, FileOffsetAddress, Absolu
 from capa.features.extractors.binja.helpers import read_c_string, unmangle_c_name
 
 
-def check_segment_for_pe(bv: BinaryView, seg: Segment) -> Iterator[Tuple[int, int]]:
+def check_segment_for_pe(bv: BinaryView, seg: Segment) -> Iterator[Tuple[Feature, Address]]:
-    """check segment for embedded PE
+    """check segment for embedded PE"""
-
+    start = 0
-    adapted for binja from:
+    if bv.view_type == "PE" and seg.start == bv.start:
-    https://github.com/vivisect/vivisect/blob/7be4037b1cecc4551b397f840405a1fc606f9b53/PE/carve.py#L19
+        # If this is the first segment of the binary, skip the first bytes.
-    """
+        # Otherwise, there will always be a matched PE at the start of the binaryview.
-    mz_xor = [
-        (
-            capa.features.extractors.helpers.xor_static(b"MZ", i),
-            capa.features.extractors.helpers.xor_static(b"PE", i),
-            i,
-        )
-        for i in range(256)
-    ]
-
-    todo = []
-    # If this is the first segment of the binary, skip the first bytes. Otherwise, there will always be a matched
-    # PE at the start of the binaryview.
-    start = seg.start
-    if bv.view_type == "PE" and start == bv.start:
         start += 1
 
-    for mzx, pex, i in mz_xor:
+    buf = bv.read(seg.start, seg.length)
-        for off, _ in bv.find_all_data(start, seg.end, mzx):
-            todo.append((off, mzx, pex, i))
 
-    while len(todo):
+    for offset, _ in capa.features.extractors.helpers.carve_pe(buf, start):
-        off, mzx, pex, i = todo.pop()
+        yield Characteristic("embedded pe"), FileOffsetAddress(seg.start + offset)
-
-        # The MZ header has one field we will check e_lfanew is at 0x3c
-        e_lfanew = off + 0x3C
-
-        if seg.end < (e_lfanew + 4):
-            continue
-
-        newoff = struct.unpack("<I", capa.features.extractors.helpers.xor_static(bv.read(e_lfanew, 4), i))[0]
-
-        peoff = off + newoff
-        if seg.end < (peoff + 2):
-            continue
-
-        if bv.read(peoff, 2) == pex:
-            yield off, i
 
 
 def extract_file_embedded_pe(bv: BinaryView) -> Iterator[Tuple[Feature, Address]]:
     """extract embedded PE features"""
     for seg in bv.segments:
-        for ea, _ in check_segment_for_pe(bv, seg):
+        yield from check_segment_for_pe(bv, seg)
-            yield Characteristic("embedded pe"), FileOffsetAddress(ea)
 
 
 def extract_file_export_names(bv: BinaryView) -> Iterator[Tuple[Feature, Address]]:

capa/features/extractors/binja/find_binja_api.py

@@ -5,31 +5,175 @@
 # Unless required by applicable law or agreed to in writing, software distributed under the License
 #  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and limitations under the License.
+import os
+import sys
+import logging
 import subprocess
+import importlib.util
+from typing import Optional
 from pathlib import Path
 
+logger = logging.getLogger(__name__)
+
+
 # When the script gets executed as a standalone executable (via PyInstaller), `import binaryninja` does not work because
 # we have excluded the binaryninja module in `pyinstaller.spec`. The trick here is to call the system Python and try
 # to find out the path of the binaryninja module that has been installed.
 # Note, including the binaryninja module in the `pyinstaller.spec` would not work, since the binaryninja module tries to
 # find the binaryninja core e.g., `libbinaryninjacore.dylib`, using a relative path. And this does not work when the
 # binaryninja module is extracted by the PyInstaller.
-code = r"""
+CODE = r"""
 from pathlib import Path
 from importlib import util
 spec = util.find_spec('binaryninja')
 if spec is not None:
     if len(spec.submodule_search_locations) > 0:
-            path = Path(spec.submodule_search_locations[0])
+        path = Path(spec.submodule_search_locations[0])
-            # encode the path with utf8 then convert to hex, make sure it can be read and restored properly
+        # encode the path with utf8 then convert to hex, make sure it can be read and restored properly
-            print(str(path.parent).encode('utf8').hex())
+        print(str(path.parent).encode('utf8').hex())
 """
 
 
-def find_binja_path() -> Path:
+def find_binaryninja_path_via_subprocess() -> Optional[Path]:
-    raw_output = subprocess.check_output(["python", "-c", code]).decode("ascii").strip()
+    raw_output = subprocess.check_output(["python", "-c", CODE]).decode("ascii").strip()
-    return Path(bytes.fromhex(raw_output).decode("utf8"))
+    output = bytes.fromhex(raw_output).decode("utf8")
+    if not output.strip():
+        return None
+    return Path(output)
+
+
+def get_desktop_entry(name: str) -> Optional[Path]:
+    """
+    Find the path for the given XDG Desktop Entry name.
+
+    Like:
+
+        >> get_desktop_entry("com.vector35.binaryninja.desktop")
+        Path("~/.local/share/applications/com.vector35.binaryninja.desktop")
+    """
+    assert sys.platform in ("linux", "linux2")
+    assert name.endswith(".desktop")
+
+    data_dirs = os.environ.get("XDG_DATA_DIRS", "/usr/share") + f":{Path.home()}/.local/share"
+    for data_dir in data_dirs.split(":"):
+        applications = Path(data_dir) / "applications"
+        for application in applications.glob("*.desktop"):
+            if application.name == name:
+                return application
+
+    return None
+
+
+def get_binaryninja_path(desktop_entry: Path) -> Optional[Path]:
+    # from: Exec=/home/wballenthin/software/binaryninja/binaryninja %u
+    # to:        /home/wballenthin/software/binaryninja/
+    for line in desktop_entry.read_text(encoding="utf-8").splitlines():
+        if not line.startswith("Exec="):
+            continue
+
+        if not line.endswith("binaryninja %u"):
+            continue
+
+        binaryninja_path = Path(line[len("Exec=") : -len("binaryninja %u")])
+        if not binaryninja_path.exists():
+            return None
+
+        return binaryninja_path
+
+    return None
+
+
+def validate_binaryninja_path(binaryninja_path: Path) -> bool:
+    if not binaryninja_path:
+        return False
+
+    module_path = binaryninja_path / "python"
+    if not module_path.is_dir():
+        return False
+
+    if not (module_path / "binaryninja" / "__init__.py").is_file():
+        return False
+
+    return True
+
+
+def find_binaryninja() -> Optional[Path]:
+    binaryninja_path = find_binaryninja_path_via_subprocess()
+    if not binaryninja_path or not validate_binaryninja_path(binaryninja_path):
+        if sys.platform == "linux" or sys.platform == "linux2":
+            # ok
+            logger.debug("detected OS: linux")
+        elif sys.platform == "darwin":
+            logger.warning("unsupported platform to find Binary Ninja: %s", sys.platform)
+            return False
+        elif sys.platform == "win32":
+            logger.warning("unsupported platform to find Binary Ninja: %s", sys.platform)
+            return False
+        else:
+            logger.warning("unsupported platform to find Binary Ninja: %s", sys.platform)
+            return False
+
+        desktop_entry = get_desktop_entry("com.vector35.binaryninja.desktop")
+        if not desktop_entry:
+            logger.debug("failed to find Binary Ninja application")
+            return None
+        logger.debug("found Binary Ninja application: %s", desktop_entry)
+
+        binaryninja_path = get_binaryninja_path(desktop_entry)
+        if not binaryninja_path:
+            logger.debug("failed to determine Binary Ninja installation path")
+            return None
+
+        if not validate_binaryninja_path(binaryninja_path):
+            logger.debug("failed to validate Binary Ninja installation")
+            return None
+
+    logger.debug("found Binary Ninja installation: %s", binaryninja_path)
+
+    return binaryninja_path / "python"
+
+
+def is_binaryninja_installed() -> bool:
+    """Is the binaryninja module ready to import?"""
+    try:
+        return importlib.util.find_spec("binaryninja") is not None
+    except ModuleNotFoundError:
+        return False
+
+
+def has_binaryninja() -> bool:
+    if is_binaryninja_installed():
+        logger.debug("found installed Binary Ninja API")
+        return True
+
+    logger.debug("Binary Ninja API not installed, searching...")
+
+    binaryninja_path = find_binaryninja()
+    if not binaryninja_path:
+        logger.debug("failed to find Binary Ninja installation")
+
+    logger.debug("found Binary Ninja API: %s", binaryninja_path)
+    return binaryninja_path is not None
+
+
+def load_binaryninja() -> bool:
+    try:
+        import binaryninja
+
+        return True
+    except ImportError:
+        binaryninja_path = find_binaryninja()
+        if not binaryninja_path:
+            return False
+
+        sys.path.append(binaryninja_path.absolute().as_posix())
+        try:
+            import binaryninja  # noqa: F401 unused import
+
+            return True
+        except ImportError:
+            return False
 
 
 if __name__ == "__main__":
-    print(find_binja_path())
+    print(find_binaryninja_path_via_subprocess())

capa/features/extractors/cape/call.py

@@ -9,6 +9,7 @@
 import logging
 from typing import Tuple, Iterator
 
+import capa.features.extractors.helpers
 from capa.helpers import assert_never
 from capa.features.insn import API, Number
 from capa.features.common import String, Feature
@@ -50,7 +51,8 @@ def extract_call_features(ph: ProcessHandle, th: ThreadHandle, ch: CallHandle) -
         else:
             assert_never(value)
 
-    yield API(call.api), ch.address
+    for name in capa.features.extractors.helpers.generate_symbols("", call.api):
+        yield API(name), ch.address
 
 
 def extract_features(ph: ProcessHandle, th: ThreadHandle, ch: CallHandle) -> Iterator[Tuple[Feature, Address]]:

capa/features/extractors/common.py

@@ -75,7 +75,7 @@ def extract_format(buf: bytes) -> Iterator[Tuple[Feature, Address]]:
         #  1. handling a file format (e.g. macho)
         #
         # for (1), this logic will need to be updated as the format is implemented.
-        logger.debug("unsupported file format: %s", binascii.hexlify(buf[:4]).decode("ascii"))
+        logger.debug("unknown file format: %s", buf[:4].hex())
         return
 
 

capa/features/extractors/drakvuf/call.py

@@ -9,6 +9,7 @@
 import logging
 from typing import Tuple, Iterator
 
+import capa.features.extractors.helpers
 from capa.features.insn import API, Number
 from capa.features.common import String, Feature
 from capa.features.address import Address
@@ -44,7 +45,8 @@ def extract_call_features(ph: ProcessHandle, th: ThreadHandle, ch: CallHandle) -
             # but yielding the entire string would be helpful for an analyst looking at the verbose output
             yield String(arg_value), ch.address
 
-    yield API(call.name), ch.address
+    for name in capa.features.extractors.helpers.generate_symbols("", call.name):
+        yield API(name), ch.address
 
 
 def extract_features(ph: ProcessHandle, th: ThreadHandle, ch: CallHandle) -> Iterator[Tuple[Feature, Address]]:

capa/features/extractors/elf.py

@@ -310,6 +310,9 @@ class ELF:
         98: "TPC",
         99: "SNP1K",
         100: "ST200",
+        # https://www.sco.com/developers/gabi/latest/ch4.eheader.html
+        183: "aarch64",
+        243: "riscv",
     }
 
     @property

capa/features/extractors/elffile.py

@@ -158,6 +158,10 @@ def extract_file_arch(elf: ELFFile, **kwargs):
         yield Arch("i386"), NO_ADDRESS
     elif arch == "x64":
         yield Arch("amd64"), NO_ADDRESS
+    elif arch == "ARM":
+        yield Arch("arm"), NO_ADDRESS
+    elif arch == "AArch64":
+        yield Arch("aarch64"), NO_ADDRESS
     else:
         logger.warning("unsupported architecture: %s", arch)
 

capa/features/extractors/helpers.py

@@ -63,6 +63,7 @@ def generate_symbols(dll: str, symbol: str, include_dll=False) -> Iterator[str]:
     # trim extensions observed in dynamic traces
     dll = dll[0:-4] if dll.endswith(".dll") else dll
     dll = dll[0:-4] if dll.endswith(".drv") else dll
+    dll = dll[0:-3] if dll.endswith(".so") else dll
 
     if include_dll or is_ordinal(symbol):
         # ws2_32.#1

capa/features/extractors/ida/extractor.py

@@ -8,7 +8,6 @@
 from typing import List, Tuple, Iterator
 
 import idaapi
-import ida_nalt
 
 import capa.ida.helpers
 import capa.features.extractors.elf
@@ -32,7 +31,9 @@ class IdaFeatureExtractor(StaticFeatureExtractor):
     def __init__(self):
         super().__init__(
             hashes=SampleHashes(
-                md5=ida_nalt.retrieve_input_file_md5(), sha1="(unknown)", sha256=ida_nalt.retrieve_input_file_sha256()
+                md5=capa.ida.helpers.retrieve_input_file_md5(),
+                sha1="(unknown)",
+                sha256=capa.ida.helpers.retrieve_input_file_sha256(),
             )
         )
         self.global_features: List[Tuple[Feature, Address]] = []

capa/features/extractors/ida/file.py

@@ -14,6 +14,7 @@ import idaapi
 import idautils
 import ida_entry
 
+import capa.ida.helpers
 import capa.features.extractors.common
 import capa.features.extractors.helpers
 import capa.features.extractors.strings
@@ -177,17 +178,17 @@ def extract_file_function_names() -> Iterator[Tuple[Feature, Address]]:
 
 
 def extract_file_format() -> Iterator[Tuple[Feature, Address]]:
-    file_info = idaapi.get_inf_structure()
+    filetype = capa.ida.helpers.get_filetype()
 
-    if file_info.filetype in (idaapi.f_PE, idaapi.f_COFF):
+    if filetype in (idaapi.f_PE, idaapi.f_COFF):
         yield Format(FORMAT_PE), NO_ADDRESS
-    elif file_info.filetype == idaapi.f_ELF:
+    elif filetype == idaapi.f_ELF:
         yield Format(FORMAT_ELF), NO_ADDRESS
-    elif file_info.filetype == idaapi.f_BIN:
+    elif filetype == idaapi.f_BIN:
         # no file type to return when processing a binary file, but we want to continue processing
         return
     else:
-        raise NotImplementedError(f"unexpected file format: {file_info.filetype}")
+        raise NotImplementedError(f"unexpected file format: {filetype}")
 
 
 def extract_features() -> Iterator[Tuple[Feature, Address]]:

capa/features/extractors/ida/global_.py

@@ -9,7 +9,6 @@ import logging
 import contextlib
 from typing import Tuple, Iterator
 
-import idaapi
 import ida_loader
 
 import capa.ida.helpers
@@ -48,12 +47,12 @@ def extract_os() -> Iterator[Tuple[Feature, Address]]:
 
 
 def extract_arch() -> Iterator[Tuple[Feature, Address]]:
-    info: idaapi.idainfo = idaapi.get_inf_structure()
+    procname = capa.ida.helpers.get_processor_name()
-    if info.procname == "metapc" and info.is_64bit():
+    if procname == "metapc" and capa.ida.helpers.is_64bit():
         yield Arch(ARCH_AMD64), NO_ADDRESS
-    elif info.procname == "metapc" and info.is_32bit():
+    elif procname == "metapc" and capa.ida.helpers.is_32bit():
         yield Arch(ARCH_I386), NO_ADDRESS
-    elif info.procname == "metapc":
+    elif procname == "metapc":
         logger.debug("unsupported architecture: non-32-bit nor non-64-bit intel")
         return
     else:
@@ -61,5 +60,5 @@ def extract_arch() -> Iterator[Tuple[Feature, Address]]:
         #  1. handling a new architecture (e.g. aarch64)
         #
         # for (1), this logic will need to be updated as the format is implemented.
-        logger.debug("unsupported architecture: %s", info.procname)
+        logger.debug("unsupported architecture: %s", procname)
         return

capa/features/extractors/ida/helpers.py

@@ -21,28 +21,49 @@ from capa.features.extractors.base_extractor import FunctionHandle
 IDA_NALT_ENCODING = ida_nalt.get_default_encoding_idx(ida_nalt.BPU_1B)  # use one byte-per-character encoding
 
 
-def find_byte_sequence(start: int, end: int, seq: bytes) -> Iterator[int]:
+if hasattr(ida_bytes, "parse_binpat_str"):
-    """yield all ea of a given byte sequence
+    # TODO (mr): use find_bytes
+    # https://github.com/mandiant/capa/issues/2339
+    def find_byte_sequence(start: int, end: int, seq: bytes) -> Iterator[int]:
+        """yield all ea of a given byte sequence
 
-    args:
+        args:
-        start: min virtual address
+            start: min virtual address
-        end: max virtual address
+            end: max virtual address
-        seq: bytes to search e.g. b"\x01\x03"
+            seq: bytes to search e.g. b"\x01\x03"
-    """
+        """
-    patterns = ida_bytes.compiled_binpat_vec_t()
+        patterns = ida_bytes.compiled_binpat_vec_t()
 
-    seqstr = " ".join([f"{b:02x}" for b in seq])
+        seqstr = " ".join([f"{b:02x}" for b in seq])
-    err = ida_bytes.parse_binpat_str(patterns, 0, seqstr, 16, IDA_NALT_ENCODING)
+        err = ida_bytes.parse_binpat_str(patterns, 0, seqstr, 16, IDA_NALT_ENCODING)
 
-    if err:
+        if err:
-        return
+            return
 
-    while True:
+        while True:
-        ea = ida_bytes.bin_search(start, end, patterns, ida_bytes.BIN_SEARCH_FORWARD)
+            ea, _ = ida_bytes.bin_search(start, end, patterns, ida_bytes.BIN_SEARCH_FORWARD)
-        if ea == idaapi.BADADDR:
+            if ea == idaapi.BADADDR:
-            break
+                break
-        start = ea + 1
+            start = ea + 1
-        yield ea
+            yield ea
+
+else:
+    # for IDA 7.5 and older; using deprecated find_binary instead of bin_search
+    def find_byte_sequence(start: int, end: int, seq: bytes) -> Iterator[int]:
+        """yield all ea of a given byte sequence
+
+        args:
+            start: min virtual address
+            end: max virtual address
+            seq: bytes to search e.g. b"\x01\x03"
+        """
+        seqstr = " ".join([f"{b:02x}" for b in seq])
+        while True:
+            ea = idaapi.find_binary(start, end, seqstr, 0, idaapi.SEARCH_DOWN)
+            if ea == idaapi.BADADDR:
+                break
+            start = ea + 1
+            yield ea
 
 
 def get_functions(

capa/features/extractors/ida/idalib.py

@@ -0,0 +1,117 @@
+# Copyright (C) 2024 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+import os
+import sys
+import json
+import logging
+import importlib.util
+from typing import Optional
+from pathlib import Path
+
+logger = logging.getLogger(__name__)
+
+
+def is_idalib_installed() -> bool:
+    try:
+        return importlib.util.find_spec("idapro") is not None
+    except ModuleNotFoundError:
+        return False
+
+
+def get_idalib_user_config_path() -> Optional[Path]:
+    """Get the path to the user's config file based on platform following IDA's user directories."""
+    # derived from `py-activate-idalib.py` from IDA v9.0 Beta 4
+
+    if sys.platform == "win32":
+        # On Windows, use the %APPDATA%\Hex-Rays\IDA Pro directory
+        config_dir = Path(os.getenv("APPDATA")) / "Hex-Rays" / "IDA Pro"
+    else:
+        # On macOS and Linux, use ~/.idapro
+        config_dir = Path.home() / ".idapro"
+
+    # Return the full path to the config file (now in JSON format)
+    user_config_path = config_dir / "ida-config.json"
+    if not user_config_path.exists():
+        return None
+    return user_config_path
+
+
+def find_idalib() -> Optional[Path]:
+    config_path = get_idalib_user_config_path()
+    if not config_path:
+        logger.error("IDA Pro user configuration does not exist, please make sure you've installed idalib properly.")
+        return None
+
+    config = json.loads(config_path.read_text(encoding="utf-8"))
+
+    try:
+        ida_install_dir = Path(config["Paths"]["ida-install-dir"])
+    except KeyError:
+        logger.error(
+            "IDA Pro user configuration does not contain location of IDA Pro installation, please make sure you've installed idalib properly."
+        )
+        return None
+
+    if not ida_install_dir.exists():
+        return None
+
+    libname = {
+        "win32": "idalib.dll",
+        "linux": "libidalib.so",
+        "linux2": "libidalib.so",
+        "darwin": "libidalib.dylib",
+    }[sys.platform]
+
+    if not (ida_install_dir / "ida.hlp").is_file():
+        return None
+
+    if not (ida_install_dir / libname).is_file():
+        return None
+
+    idalib_path = ida_install_dir / "idalib" / "python"
+    if not idalib_path.exists():
+        return None
+
+    if not (idalib_path / "idapro" / "__init__.py").is_file():
+        return None
+
+    return idalib_path
+
+
+def has_idalib() -> bool:
+    if is_idalib_installed():
+        logger.debug("found installed IDA idalib API")
+        return True
+
+    logger.debug("IDA idalib API not installed, searching...")
+
+    idalib_path = find_idalib()
+    if not idalib_path:
+        logger.debug("failed to find IDA idalib installation")
+
+    logger.debug("found IDA idalib API: %s", idalib_path)
+    return idalib_path is not None
+
+
+def load_idalib() -> bool:
+    try:
+        import idapro
+
+        return True
+    except ImportError:
+        idalib_path = find_idalib()
+        if not idalib_path:
+            return False
+
+        sys.path.append(idalib_path.absolute().as_posix())
+        try:
+            import idapro  # noqa: F401 unused import
+
+            return True
+        except ImportError:
+            return False

capa/features/extractors/pefile.py

@@ -130,7 +130,13 @@ def extract_file_arch(pe, **kwargs):
     elif pe.FILE_HEADER.Machine == pefile.MACHINE_TYPE["IMAGE_FILE_MACHINE_AMD64"]:
         yield Arch(ARCH_AMD64), NO_ADDRESS
     else:
-        logger.warning("unsupported architecture: %s", pefile.MACHINE_TYPE[pe.FILE_HEADER.Machine])
+        try:
+            logger.warning(
+                "unsupported architecture: %s",
+                pefile.MACHINE_TYPE[pe.FILE_HEADER.Machine],
+            )
+        except KeyError:
+            logger.warning("unknown architecture: %s", pe.FILE_HEADER.Machine)
 
 
 def extract_file_features(pe, buf):

capa/features/extractors/vmray/__init__.py

@@ -0,0 +1,198 @@
+# Copyright (C) 2024 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+import logging
+from typing import Dict, List, Tuple, Optional
+from pathlib import Path
+from zipfile import ZipFile
+from collections import defaultdict
+from dataclasses import dataclass
+
+from capa.exceptions import UnsupportedFormatError
+from capa.features.extractors.vmray.models import File, Flog, SummaryV2, StaticData, FunctionCall, xml_to_dict
+
+logger = logging.getLogger(__name__)
+
+DEFAULT_ARCHIVE_PASSWORD = b"infected"
+
+SUPPORTED_FLOG_VERSIONS = ("2",)
+
+
+@dataclass
+class VMRayMonitorThread:
+    tid: int  # thread ID assigned by OS
+    monitor_id: int  # unique ID assigned to thread by VMRay
+    process_monitor_id: int  # unqiue ID assigned to containing process by VMRay
+
+
+@dataclass
+class VMRayMonitorProcess:
+    pid: int  # process ID assigned by OS
+    ppid: int  # parent process ID assigned by OS
+    monitor_id: int  # unique ID assigned to process by VMRay
+    image_name: str
+
+
+class VMRayAnalysis:
+    def __init__(self, zipfile_path: Path):
+        self.zipfile = ZipFile(zipfile_path, "r")
+
+        # summary_v2.json is the entry point to the entire VMRay archive and
+        # we use its data to find everything else that we need for capa
+        self.sv2 = SummaryV2.model_validate_json(
+            self.zipfile.read("logs/summary_v2.json", pwd=DEFAULT_ARCHIVE_PASSWORD)
+        )
+        self.file_type: str = self.sv2.analysis_metadata.sample_type
+
+        # flog.xml contains all of the call information that VMRay captured during execution
+        flog_xml = self.zipfile.read("logs/flog.xml", pwd=DEFAULT_ARCHIVE_PASSWORD)
+        flog_dict = xml_to_dict(flog_xml)
+        self.flog = Flog.model_validate(flog_dict)
+
+        if self.flog.analysis.log_version not in SUPPORTED_FLOG_VERSIONS:
+            raise UnsupportedFormatError(
+                "VMRay feature extractor does not support flog version %s" % self.flog.analysis.log_version
+            )
+
+        self.exports: Dict[int, str] = {}
+        self.imports: Dict[int, Tuple[str, str]] = {}
+        self.sections: Dict[int, str] = {}
+        self.monitor_processes: Dict[int, VMRayMonitorProcess] = {}
+        self.monitor_threads: Dict[int, VMRayMonitorThread] = {}
+
+        # map monitor thread IDs to their associated monitor process ID
+        self.monitor_threads_by_monitor_process: Dict[int, List[int]] = defaultdict(list)
+
+        # map function calls to their associated monitor thread ID mapped to its associated monitor process ID
+        self.monitor_process_calls: Dict[int, Dict[int, List[FunctionCall]]] = defaultdict(lambda: defaultdict(list))
+
+        self.base_address: int
+
+        self.sample_file_name: Optional[str] = None
+        self.sample_file_analysis: Optional[File] = None
+        self.sample_file_static_data: Optional[StaticData] = None
+
+        self._find_sample_file()
+
+        # VMRay analysis archives in various shapes and sizes and file type does not definitively tell us what data
+        # we can expect to find in the archive, so to be explicit we check for the various pieces that we need at
+        # minimum to run capa analysis
+        if self.sample_file_name is None or self.sample_file_analysis is None:
+            raise UnsupportedFormatError("VMRay archive does not contain sample file (file_type: %s)" % self.file_type)
+
+        if not self.sample_file_static_data:
+            raise UnsupportedFormatError("VMRay archive does not contain static data (file_type: %s)" % self.file_type)
+
+        if not self.sample_file_static_data.pe and not self.sample_file_static_data.elf:
+            raise UnsupportedFormatError(
+                "VMRay feature extractor only supports PE and ELF at this time (file_type: %s)" % self.file_type
+            )
+
+        # VMRay does not store static strings for the sample file so we must use the source file
+        # stored in the archive
+        sample_sha256: str = self.sample_file_analysis.hash_values.sha256.lower()
+        sample_file_path: str = f"internal/static_analyses/{sample_sha256}/objects/files/{sample_sha256}"
+
+        logger.debug("file_type: %s, file_path: %s", self.file_type, sample_file_path)
+
+        self.sample_file_buf: bytes = self.zipfile.read(sample_file_path, pwd=DEFAULT_ARCHIVE_PASSWORD)
+
+        # do not change order, it matters
+        self._compute_base_address()
+        self._compute_imports()
+        self._compute_exports()
+        self._compute_sections()
+        self._compute_monitor_processes()
+        self._compute_monitor_threads()
+        self._compute_monitor_process_calls()
+
+    def _find_sample_file(self):
+        for file_name, file_analysis in self.sv2.files.items():
+            if file_analysis.is_sample:
+                # target the sample submitted for analysis
+                self.sample_file_name = file_name
+                self.sample_file_analysis = file_analysis
+
+                if file_analysis.ref_static_data:
+                    # like "path": ["static_data","static_data_0"] where "static_data_0" is the summary_v2 static data
+                    # key for the file's static data
+                    self.sample_file_static_data = self.sv2.static_data[file_analysis.ref_static_data.path[1]]
+
+                break
+
+    def _compute_base_address(self):
+        assert self.sample_file_static_data is not None
+        if self.sample_file_static_data.pe:
+            self.base_address = self.sample_file_static_data.pe.basic_info.image_base
+
+    def _compute_exports(self):
+        assert self.sample_file_static_data is not None
+        if self.sample_file_static_data.pe:
+            for export in self.sample_file_static_data.pe.exports:
+                self.exports[export.address] = export.api.name
+
+    def _compute_imports(self):
+        assert self.sample_file_static_data is not None
+        if self.sample_file_static_data.pe:
+            for module in self.sample_file_static_data.pe.imports:
+                for api in module.apis:
+                    self.imports[api.address] = (module.dll, api.api.name)
+
+    def _compute_sections(self):
+        assert self.sample_file_static_data is not None
+        if self.sample_file_static_data.pe:
+            for pefile_section in self.sample_file_static_data.pe.sections:
+                self.sections[pefile_section.virtual_address] = pefile_section.name
+        elif self.sample_file_static_data.elf:
+            for elffile_section in self.sample_file_static_data.elf.sections:
+                self.sections[elffile_section.header.sh_addr] = elffile_section.header.sh_name
+
+    def _compute_monitor_processes(self):
+        for process in self.sv2.processes.values():
+            # we expect monitor IDs to be unique
+            assert process.monitor_id not in self.monitor_processes
+
+            ppid: int = (
+                self.sv2.processes[process.ref_parent_process.path[1]].os_pid if process.ref_parent_process else 0
+            )
+            self.monitor_processes[process.monitor_id] = VMRayMonitorProcess(
+                process.os_pid, ppid, process.monitor_id, process.image_name
+            )
+
+        # not all processes are recorded in SummaryV2.json, get missing data from flog.xml, see #2394
+        for monitor_process in self.flog.analysis.monitor_processes:
+            vmray_monitor_process: VMRayMonitorProcess = VMRayMonitorProcess(
+                monitor_process.os_pid,
+                monitor_process.os_parent_pid,
+                monitor_process.process_id,
+                monitor_process.image_name,
+            )
+
+            if monitor_process.process_id not in self.monitor_processes:
+                self.monitor_processes[monitor_process.process_id] = vmray_monitor_process
+            else:
+                # we expect monitor processes recorded in both SummaryV2.json and flog.xml to equal
+                assert self.monitor_processes[monitor_process.process_id] == vmray_monitor_process
+
+    def _compute_monitor_threads(self):
+        for monitor_thread in self.flog.analysis.monitor_threads:
+            # we expect monitor IDs to be unique
+            assert monitor_thread.thread_id not in self.monitor_threads
+
+            self.monitor_threads[monitor_thread.thread_id] = VMRayMonitorThread(
+                monitor_thread.os_tid, monitor_thread.thread_id, monitor_thread.process_id
+            )
+
+            # we expect each monitor thread ID to be unique for its associated monitor process ID e.g. monitor
+            # thread ID 10 should not be captured twice for monitor process ID 1
+            assert monitor_thread.thread_id not in self.monitor_threads_by_monitor_process[monitor_thread.thread_id]
+
+            self.monitor_threads_by_monitor_process[monitor_thread.process_id].append(monitor_thread.thread_id)
+
+    def _compute_monitor_process_calls(self):
+        for function_call in self.flog.analysis.function_calls:
+            self.monitor_process_calls[function_call.process_id][function_call.thread_id].append(function_call)

capa/features/extractors/vmray/call.py

@@ -0,0 +1,59 @@
+# Copyright (C) 2024 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+import logging
+from typing import Tuple, Iterator
+
+import capa.features.extractors.helpers
+from capa.features.insn import API, Number
+from capa.features.common import String, Feature
+from capa.features.address import Address
+from capa.features.extractors.vmray.models import PARAM_TYPE_INT, PARAM_TYPE_STR, Param, FunctionCall, hexint
+from capa.features.extractors.base_extractor import CallHandle, ThreadHandle, ProcessHandle
+
+logger = logging.getLogger(__name__)
+
+
+def get_call_param_features(param: Param, ch: CallHandle) -> Iterator[Tuple[Feature, Address]]:
+    if param.deref is not None:
+        # pointer types contain a special "deref" member that stores the deref'd value
+        # so we check for this first and ignore Param.value as this always contains the
+        # deref'd pointer value
+        if param.deref.value is not None:
+            if param.deref.type_ in PARAM_TYPE_INT:
+                yield Number(hexint(param.deref.value)), ch.address
+            elif param.deref.type_ in PARAM_TYPE_STR:
+                # TODO(mr-tz): remove FPS like " \\x01\\x02\\x03\\x04\\x05\\x06\\x07\\x08\\x09\\x0a\\x0b\\x0c\\x0d\\x0e\\x0f\\x10\\x11\\x12\\x13\\x14\\x15\\x16\\x17\\x18\\x19\\x1a\\x1b\\x1c\\x1d\\x1e\..."
+                # https://github.com/mandiant/capa/issues/2432
+
+                # parsing the data up to here results in double-escaped backslashes, remove those here
+                yield String(param.deref.value.replace("\\\\", "\\")), ch.address
+            else:
+                logger.debug("skipping deref param type %s", param.deref.type_)
+    elif param.value is not None:
+        if param.type_ in PARAM_TYPE_INT:
+            yield Number(hexint(param.value)), ch.address
+
+
+def extract_call_features(ph: ProcessHandle, th: ThreadHandle, ch: CallHandle) -> Iterator[Tuple[Feature, Address]]:
+    call: FunctionCall = ch.inner
+
+    if call.params_in:
+        for param in call.params_in.params:
+            yield from get_call_param_features(param, ch)
+
+    for name in capa.features.extractors.helpers.generate_symbols("", call.name):
+        yield API(name), ch.address
+
+
+def extract_features(ph: ProcessHandle, th: ThreadHandle, ch: CallHandle) -> Iterator[Tuple[Feature, Address]]:
+    for handler in CALL_HANDLERS:
+        for feature, addr in handler(ph, th, ch):
+            yield feature, addr
+
+
+CALL_HANDLERS = (extract_call_features,)

capa/features/extractors/vmray/extractor.py

@@ -0,0 +1,133 @@
+# Copyright (C) 2024 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+
+
+from typing import List, Tuple, Iterator
+from pathlib import Path
+
+import capa.helpers
+import capa.features.extractors.vmray.call
+import capa.features.extractors.vmray.file
+import capa.features.extractors.vmray.global_
+from capa.features.common import Feature, Characteristic
+from capa.features.address import (
+    NO_ADDRESS,
+    Address,
+    ThreadAddress,
+    ProcessAddress,
+    DynamicCallAddress,
+    AbsoluteVirtualAddress,
+)
+from capa.features.extractors.vmray import VMRayAnalysis, VMRayMonitorThread, VMRayMonitorProcess
+from capa.features.extractors.vmray.models import PARAM_TYPE_STR, ParamList, FunctionCall
+from capa.features.extractors.base_extractor import (
+    CallHandle,
+    SampleHashes,
+    ThreadHandle,
+    ProcessHandle,
+    DynamicFeatureExtractor,
+)
+
+
+def get_formatted_params(params: ParamList) -> List[str]:
+    params_list: List[str] = []
+
+    for param in params:
+        if param.deref and param.deref.value is not None:
+            deref_value: str = f'"{param.deref.value}"' if param.deref.type_ in PARAM_TYPE_STR else param.deref.value
+            params_list.append(f"{param.name}: {deref_value}")
+        else:
+            value: str = "" if param.value is None else param.value
+            params_list.append(f"{param.name}: {value}")
+
+    return params_list
+
+
+class VMRayExtractor(DynamicFeatureExtractor):
+    def __init__(self, analysis: VMRayAnalysis):
+        assert analysis.sample_file_analysis is not None
+
+        super().__init__(
+            hashes=SampleHashes(
+                md5=analysis.sample_file_analysis.hash_values.md5.lower(),
+                sha1=analysis.sample_file_analysis.hash_values.sha1.lower(),
+                sha256=analysis.sample_file_analysis.hash_values.sha256.lower(),
+            )
+        )
+
+        self.analysis = analysis
+
+        # pre-compute these because we'll yield them at *every* scope.
+        self.global_features = list(capa.features.extractors.vmray.global_.extract_features(self.analysis))
+
+    def get_base_address(self) -> Address:
+        # value according to the PE header, the actual trace may use a different imagebase
+        return AbsoluteVirtualAddress(self.analysis.base_address)
+
+    def extract_file_features(self) -> Iterator[Tuple[Feature, Address]]:
+        yield from capa.features.extractors.vmray.file.extract_features(self.analysis)
+
+    def extract_global_features(self) -> Iterator[Tuple[Feature, Address]]:
+        yield from self.global_features
+
+    def get_processes(self) -> Iterator[ProcessHandle]:
+        for monitor_process in self.analysis.monitor_processes.values():
+            address: ProcessAddress = ProcessAddress(pid=monitor_process.pid, ppid=monitor_process.ppid)
+            yield ProcessHandle(address, inner=monitor_process)
+
+    def extract_process_features(self, ph: ProcessHandle) -> Iterator[Tuple[Feature, Address]]:
+        # we have not identified process-specific features for VMRay yet
+        yield from []
+
+    def get_process_name(self, ph) -> str:
+        monitor_process: VMRayMonitorProcess = ph.inner
+        return monitor_process.image_name
+
+    def get_threads(self, ph: ProcessHandle) -> Iterator[ThreadHandle]:
+        for monitor_thread_id in self.analysis.monitor_threads_by_monitor_process[ph.inner.monitor_id]:
+            monitor_thread: VMRayMonitorThread = self.analysis.monitor_threads[monitor_thread_id]
+
+            address: ThreadAddress = ThreadAddress(process=ph.address, tid=monitor_thread.tid)
+            yield ThreadHandle(address=address, inner=monitor_thread)
+
+    def extract_thread_features(self, ph: ProcessHandle, th: ThreadHandle) -> Iterator[Tuple[Feature, Address]]:
+        if False:
+            # force this routine to be a generator,
+            # but we don't actually have any elements to generate.
+            yield Characteristic("never"), NO_ADDRESS
+        return
+
+    def get_calls(self, ph: ProcessHandle, th: ThreadHandle) -> Iterator[CallHandle]:
+        for function_call in self.analysis.monitor_process_calls[ph.inner.monitor_id][th.inner.monitor_id]:
+            addr = DynamicCallAddress(thread=th.address, id=function_call.fncall_id)
+            yield CallHandle(address=addr, inner=function_call)
+
+    def extract_call_features(
+        self, ph: ProcessHandle, th: ThreadHandle, ch: CallHandle
+    ) -> Iterator[Tuple[Feature, Address]]:
+        yield from capa.features.extractors.vmray.call.extract_features(ph, th, ch)
+
+    def get_call_name(self, ph, th, ch) -> str:
+        call: FunctionCall = ch.inner
+        call_formatted: str = call.name
+
+        # format input parameters
+        if call.params_in:
+            call_formatted += f"({', '.join(get_formatted_params(call.params_in.params))})"
+        else:
+            call_formatted += "()"
+
+        # format output parameters
+        if call.params_out:
+            call_formatted += f" -> {', '.join(get_formatted_params(call.params_out.params))}"
+
+        return call_formatted
+
+    @classmethod
+    def from_zipfile(cls, zipfile_path: Path):
+        return cls(VMRayAnalysis(zipfile_path))

capa/features/extractors/vmray/file.py

@@ -0,0 +1,82 @@
+# Copyright (C) 2024 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+import logging
+from typing import Tuple, Iterator
+
+import capa.features.extractors.common
+from capa.features.file import Export, Import, Section
+from capa.features.common import String, Feature
+from capa.features.address import NO_ADDRESS, Address, AbsoluteVirtualAddress
+from capa.features.extractors.vmray import VMRayAnalysis
+from capa.features.extractors.helpers import generate_symbols
+
+logger = logging.getLogger(__name__)
+
+
+def extract_export_names(analysis: VMRayAnalysis) -> Iterator[Tuple[Feature, Address]]:
+    for addr, name in analysis.exports.items():
+        yield Export(name), AbsoluteVirtualAddress(addr)
+
+
+def extract_import_names(analysis: VMRayAnalysis) -> Iterator[Tuple[Feature, Address]]:
+    for addr, (module, api) in analysis.imports.items():
+        for symbol in generate_symbols(module, api, include_dll=True):
+            yield Import(symbol), AbsoluteVirtualAddress(addr)
+
+
+def extract_section_names(analysis: VMRayAnalysis) -> Iterator[Tuple[Feature, Address]]:
+    for addr, name in analysis.sections.items():
+        yield Section(name), AbsoluteVirtualAddress(addr)
+
+
+def extract_referenced_filenames(analysis: VMRayAnalysis) -> Iterator[Tuple[Feature, Address]]:
+    for filename in analysis.sv2.filenames.values():
+        yield String(filename.filename), NO_ADDRESS
+
+
+def extract_referenced_mutex_names(analysis: VMRayAnalysis) -> Iterator[Tuple[Feature, Address]]:
+    for mutex in analysis.sv2.mutexes.values():
+        yield String(mutex.name), NO_ADDRESS
+
+
+def extract_referenced_domain_names(analysis: VMRayAnalysis) -> Iterator[Tuple[Feature, Address]]:
+    for domain in analysis.sv2.domains.values():
+        yield String(domain.domain), NO_ADDRESS
+
+
+def extract_referenced_ip_addresses(analysis: VMRayAnalysis) -> Iterator[Tuple[Feature, Address]]:
+    for ip_address in analysis.sv2.ip_addresses.values():
+        yield String(ip_address.ip_address), NO_ADDRESS
+
+
+def extract_referenced_registry_key_names(analysis: VMRayAnalysis) -> Iterator[Tuple[Feature, Address]]:
+    for registry_record in analysis.sv2.registry_records.values():
+        yield String(registry_record.reg_key_name), NO_ADDRESS
+
+
+def extract_file_strings(analysis: VMRayAnalysis) -> Iterator[Tuple[Feature, Address]]:
+    yield from capa.features.extractors.common.extract_file_strings(analysis.sample_file_buf)
+
+
+def extract_features(analysis: VMRayAnalysis) -> Iterator[Tuple[Feature, Address]]:
+    for handler in FILE_HANDLERS:
+        for feature, addr in handler(analysis):
+            yield feature, addr
+
+
+FILE_HANDLERS = (
+    extract_import_names,
+    extract_export_names,
+    extract_section_names,
+    extract_referenced_filenames,
+    extract_referenced_mutex_names,
+    extract_referenced_domain_names,
+    extract_referenced_ip_addresses,
+    extract_referenced_registry_key_names,
+    extract_file_strings,
+)

capa/features/extractors/vmray/global_.py

@@ -0,0 +1,72 @@
+# Copyright (C) 2024 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+
+import logging
+from typing import Tuple, Iterator
+
+from capa.features.common import (
+    OS,
+    OS_LINUX,
+    ARCH_I386,
+    FORMAT_PE,
+    ARCH_AMD64,
+    FORMAT_ELF,
+    OS_WINDOWS,
+    Arch,
+    Format,
+    Feature,
+)
+from capa.features.address import NO_ADDRESS, Address
+from capa.features.extractors.vmray import VMRayAnalysis
+
+logger = logging.getLogger(__name__)
+
+
+def extract_arch(analysis: VMRayAnalysis) -> Iterator[Tuple[Feature, Address]]:
+    file_type: str = analysis.file_type
+
+    if "x86-32" in file_type:
+        yield Arch(ARCH_I386), NO_ADDRESS
+    elif "x86-64" in file_type:
+        yield Arch(ARCH_AMD64), NO_ADDRESS
+    else:
+        raise ValueError("unrecognized arch from the VMRay report: %s" % file_type)
+
+
+def extract_format(analysis: VMRayAnalysis) -> Iterator[Tuple[Feature, Address]]:
+    assert analysis.sample_file_static_data is not None
+    if analysis.sample_file_static_data.pe:
+        yield Format(FORMAT_PE), NO_ADDRESS
+    elif analysis.sample_file_static_data.elf:
+        yield Format(FORMAT_ELF), NO_ADDRESS
+    else:
+        raise ValueError("unrecognized file format from the VMRay report: %s" % analysis.file_type)
+
+
+def extract_os(analysis: VMRayAnalysis) -> Iterator[Tuple[Feature, Address]]:
+    file_type: str = analysis.file_type
+
+    if "windows" in file_type.lower():
+        yield OS(OS_WINDOWS), NO_ADDRESS
+    elif "linux" in file_type.lower():
+        yield OS(OS_LINUX), NO_ADDRESS
+    else:
+        raise ValueError("unrecognized OS from the VMRay report: %s" % file_type)
+
+
+def extract_features(analysis: VMRayAnalysis) -> Iterator[Tuple[Feature, Address]]:
+    for global_handler in GLOBAL_HANDLER:
+        for feature, addr in global_handler(analysis):
+            yield feature, addr
+
+
+GLOBAL_HANDLER = (
+    extract_format,
+    extract_os,
+    extract_arch,
+)

capa/features/extractors/vmray/models.py

@@ -0,0 +1,368 @@
+# Copyright (C) 2024 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+
+from typing import Dict, List, Union, Optional
+
+import xmltodict
+from pydantic import Field, BaseModel
+from typing_extensions import Annotated
+from pydantic.functional_validators import BeforeValidator
+
+"""
+# possible param types, included for documentation
+PARAM_TYPE = (
+    "signed_8bit",
+    "unsigned_8bit",
+    "signed_16bit",
+    "unsigned_16bit",
+    "signed_32bit",
+    "unsigned_32bit",
+    "signed_64bit",
+    "unsigned_64bit",
+    "double",
+    "void_ptr",
+    "bool",
+    "unknown",
+    "ptr",
+    "void",
+    "str",
+    "array",
+    "container",
+    "bindata",
+    "undefined_type",
+)
+"""
+
+PARAM_TYPE_PTR = ("void_ptr", "ptr")
+PARAM_TYPE_STR = ("str",)
+PARAM_TYPE_INT = (
+    "signed_8bit",
+    "unsigned_8bit",
+    "signed_16bit",
+    "unsigned_16bit",
+    "signed_32bit",
+    "unsigned_32bit",
+    "signed_64bit",
+    "unsigned_64bit",
+    "double",
+    "bool",
+    "unknown",
+)
+
+
+def xml_to_dict(xml):
+    return xmltodict.parse(xml, attr_prefix="")
+
+
+def hexint(value: Union[str, int]) -> int:
+    if isinstance(value, str):
+        return int(value, 16) if value.startswith("0x") else int(value, 10)
+    else:
+        return value
+
+
+def validate_hex_int(value: Union[str, int]) -> int:
+    return hexint(value)
+
+
+# convert the input value to a Python int type before inner validation (int) is called
+HexInt = Annotated[int, BeforeValidator(validate_hex_int)]
+
+
+# models flog.xml file, certain fields left as comments for documentation purposes
+class ParamDeref(BaseModel):
+    type_: str = Field(alias="type")
+    value: Optional[str] = None
+
+
+class Param(BaseModel):
+    name: str
+    type_: str = Field(alias="type")
+    value: Optional[str] = None
+    deref: Optional[ParamDeref] = None
+
+
+def validate_ensure_is_list(value: Union[List[Param], Param]) -> List[Param]:
+    if isinstance(value, list):
+        return value
+    else:
+        return [value]
+
+
+# params may be stored as a list of Param or a single Param so we convert
+# the input value to Python list type before the inner validation (List[Param])
+# is called
+ParamList = Annotated[List[Param], BeforeValidator(validate_ensure_is_list)]
+
+
+class Params(BaseModel):
+    params: ParamList = Field(alias="param")
+
+
+def validate_call_name(value: str) -> str:
+    if value.startswith("sys_"):
+        # VMRay appears to log kernel function calls ("sys_*") for Linux so we remove that
+        # here to enable capa matching
+        return value[4:]
+    else:
+        return value
+
+
+# function call names may need to be reformatted to remove data, etc. so we reformat
+# before calling the inner validation (str)
+CallName = Annotated[str, BeforeValidator(validate_call_name)]
+
+
+class FunctionCall(BaseModel):
+    # ts: HexInt
+    fncall_id: HexInt
+    process_id: HexInt
+    thread_id: HexInt
+    name: CallName
+    # addr: HexInt
+    # from_addr: HexInt = Field(alias="from")
+    params_in: Optional[Params] = Field(alias="in", default=None)
+    params_out: Optional[Params] = Field(alias="out", default=None)
+
+
+class FunctionReturn(BaseModel):
+    ts: HexInt
+    fncall_id: HexInt
+    addr: HexInt
+    from_addr: HexInt = Field(alias="from")
+
+
+class MonitorProcess(BaseModel):
+    ts: HexInt
+    process_id: int
+    image_name: str
+    filename: str
+    # page_root: HexInt
+    os_pid: HexInt
+    # os_integrity_level: HexInt
+    # os_privileges: HexInt
+    monitor_reason: str
+    parent_id: int
+    os_parent_pid: HexInt
+    # cmd_line: str
+    # cur_dir: str
+    # os_username: str
+    # bitness: int
+    # os_groups: str
+
+
+class MonitorThread(BaseModel):
+    ts: HexInt
+    thread_id: int
+    process_id: int
+    os_tid: HexInt
+
+
+# handle if there's only single entries, but the model expects a list
+MonitorProcessList = Annotated[List[MonitorProcess], BeforeValidator(validate_ensure_is_list)]
+MonitorThreadList = Annotated[List[MonitorThread], BeforeValidator(validate_ensure_is_list)]
+FunctionCallList = Annotated[List[FunctionCall], BeforeValidator(validate_ensure_is_list)]
+
+
+class Analysis(BaseModel):
+    log_version: str  # tested 2
+    analyzer_version: str  # tested 2024.2.1
+    # analysis_date: str
+
+    monitor_processes: MonitorProcessList = Field(alias="monitor_process", default=[])
+    monitor_threads: MonitorThreadList = Field(alias="monitor_thread", default=[])
+    function_calls: FunctionCallList = Field(alias="fncall", default=[])
+    # function_returns: List[FunctionReturn] = Field(alias="fnret", default=[])
+
+
+class Flog(BaseModel):
+    analysis: Analysis
+
+
+# models for summary_v2.json file, certain fields left as comments for documentation purposes
+class GenericReference(BaseModel):
+    path: List[str]
+    source: str
+
+
+class StaticDataReference(GenericReference): ...
+
+
+class PEFileBasicInfo(BaseModel):
+    # compile_time: str
+    # file_type: str
+    image_base: int
+    # machine_type: str
+    # size_of_code: int
+    # size_of_initialized_data: int
+    # size_of_uninitialized_data: int
+    # subsystem: str
+    # entry_point: int
+    # imphash: Optional[str] = None
+
+
+class API(BaseModel):
+    name: str
+    ordinal: Optional[int] = None
+
+
+class PEFileExport(BaseModel):
+    address: int
+    api: API
+
+
+class PEFileImport(BaseModel):
+    address: int
+    api: API
+    # thunk_offset: int
+    # hint: Optional[int] = None
+    # thunk_rva: int
+
+
+class PEFileImportModule(BaseModel):
+    dll: str
+    apis: List[PEFileImport]
+
+
+class PEFileSection(BaseModel):
+    # entropy: float
+    # flags: List[str] = []
+    name: str
+    # raw_data_offset: int
+    # raw_data_size: int
+    virtual_address: int
+    # virtual_size: int
+
+
+class PEFile(BaseModel):
+    basic_info: PEFileBasicInfo
+    exports: List[PEFileExport] = []
+    imports: List[PEFileImportModule] = []
+    sections: List[PEFileSection] = []
+
+
+class ElfFileSectionHeader(BaseModel):
+    sh_name: str
+    sh_addr: int
+
+
+class ElfFileSection(BaseModel):
+    header: ElfFileSectionHeader
+
+
+"""
+class ElfFileHeader(BaseModel):
+    file_class: str
+    endianness: str
+    file_type: str
+    architecture: str
+    architecture_human_str: str
+    entry_point: int
+"""
+
+
+class ElfFile(BaseModel):
+    # file_header: ElfFileHeader
+    sections: List[ElfFileSection]
+
+
+class StaticData(BaseModel):
+    pe: Optional[PEFile] = None
+    elf: Optional[ElfFile] = None
+
+
+class FileHashes(BaseModel):
+    md5: str
+    sha1: str
+    sha256: str
+    # ssdeep: str
+
+
+class File(BaseModel):
+    # categories: List[str]
+    hash_values: FileHashes
+    # is_artifact: bool
+    # is_ioc: bool
+    is_sample: bool
+    # size: int
+    # is_truncated: bool
+    # mime_type: Optional[str] = None
+    # operations: List[str] = []
+    # ref_filenames: List[GenericReference] = []
+    # ref_gfncalls: List[GenericReference] = []
+    ref_static_data: Optional[StaticDataReference] = None
+    # ref_vti_matches: List[GenericReference] = []
+    # verdict: str
+
+
+class Process(BaseModel):
+    # bitness: int
+    # is_artifact: bool
+    # is_ioc: bool
+    monitor_id: int
+    # monitor_reason: str
+    os_pid: int
+    filename: str
+    image_name: str
+    ref_parent_process: Optional[GenericReference] = None
+
+
+class Filename(BaseModel):
+    filename: str
+    # is_artifact: bool
+    # is_ioc: bool
+    # verdict: str
+
+
+class Mutex(BaseModel):
+    name: str
+    # is_artifact: bool
+    # is_ioc: bool
+    # verdict: str
+
+
+class Registry(BaseModel):
+    reg_key_name: str
+    # reg_key_value_type: Optional[str] = None
+    # is_artifact: bool
+    # is_ioc: bool
+    # verdict: str
+
+
+class Domain(BaseModel):
+    domain: str
+    # is_artifact: bool
+    # is_ioc: bool
+    # verdict: str
+
+
+class IPAddress(BaseModel):
+    ip_address: str
+    # is_artifact: bool
+    # is_ioc: bool
+    # verdict: str
+
+
+class AnalysisMetadata(BaseModel):
+    sample_type: str
+    submission_filename: str
+
+
+class SummaryV2(BaseModel):
+    analysis_metadata: AnalysisMetadata
+
+    static_data: Dict[str, StaticData] = {}
+
+    # recorded artifacts
+    files: Dict[str, File] = {}
+    processes: Dict[str, Process] = {}
+    filenames: Dict[str, Filename] = {}
+    mutexes: Dict[str, Mutex] = {}
+    domains: Dict[str, Domain] = {}
+    ip_addresses: Dict[str, IPAddress] = {}
+    registry_records: Dict[str, Registry] = {}

capa/features/insn.py

@@ -100,9 +100,10 @@ class Mnemonic(Feature):
 
 
 # max number of operands to consider for a given instruction.
-# since we only support Intel and .NET, we can assume this is 3
+# for Intel and .NET, this is 3
 # which covers cases up to e.g. "vinserti128 ymm0,ymm0,ymm5,1"
-MAX_OPERAND_COUNT = 4
+# for ARM/aarch64, we assume 4
+MAX_OPERAND_COUNT = 5
 MAX_OPERAND_INDEX = MAX_OPERAND_COUNT - 1
 
 

capa/ghidra/capa_explorer.py

@@ -372,6 +372,10 @@ if __name__ == "__main__":
         from capa.exceptions import UnsupportedRuntimeError
 
         raise UnsupportedRuntimeError("This version of capa can only be used with Python 3.8+")
+    elif sys.version_info < (3, 10):
+        from warnings import warn
+
+        warn("This is the last capa version supporting Python 3.8 and 3.9.", DeprecationWarning, stacklevel=2)
     exit_code = main()
     if exit_code != 0:
         popup("capa explorer encountered errors during analysis. Please check the console output for more information.")  # type: ignore [name-defined] # noqa: F821

capa/ghidra/capa_ghidra.py

@@ -164,4 +164,8 @@ if __name__ == "__main__":
         from capa.exceptions import UnsupportedRuntimeError
 
         raise UnsupportedRuntimeError("This version of capa can only be used with Python 3.8+")
+    elif sys.version_info < (3, 10):
+        from warnings import warn
+
+        warn("This is the last capa version supporting Python 3.8 and 3.9.", DeprecationWarning, stacklevel=2)
     sys.exit(main())

capa/helpers.py

@@ -5,17 +5,35 @@
 # Unless required by applicable law or agreed to in writing, software distributed under the License
 #  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and limitations under the License.
+import io
+import os
 import sys
 import gzip
-import inspect
+import ctypes
 import logging
+import tempfile
 import contextlib
 import importlib.util
-from typing import Dict, Union, BinaryIO, Iterator, NoReturn
+from typing import Dict, List, Union, BinaryIO, Iterator, NoReturn
 from pathlib import Path
+from zipfile import ZipFile
+from datetime import datetime
 
-import tqdm
 import msgspec.json
+from rich.console import Console
+from rich.progress import (
+    Task,
+    Text,
+    Progress,
+    BarColumn,
+    TextColumn,
+    SpinnerColumn,
+    ProgressColumn,
+    TimeElapsedColumn,
+    MofNCompleteColumn,
+    TaskProgressColumn,
+    TimeRemainingColumn,
+)
 
 from capa.exceptions import UnsupportedFormatError
 from capa.features.common import (
@@ -23,24 +41,32 @@ from capa.features.common import (
     FORMAT_CAPE,
     FORMAT_SC32,
     FORMAT_SC64,
+    FORMAT_VMRAY,
     FORMAT_DOTNET,
     FORMAT_FREEZE,
     FORMAT_DRAKVUF,
     FORMAT_UNKNOWN,
+    FORMAT_BINEXPORT2,
     Format,
 )
 
 EXTENSIONS_SHELLCODE_32 = ("sc32", "raw32")
 EXTENSIONS_SHELLCODE_64 = ("sc64", "raw64")
-# CAPE extensions: .json, .json_, .json.gz
+# CAPE (.json, .json_, .json.gz)
-# DRAKVUF Sandbox extensions: .log, .log.gz
+# DRAKVUF (.log, .log.gz)
-EXTENSIONS_DYNAMIC = ("json", "json_", "json.gz", "log", ".log.gz")
+# VMRay (.zip)
+EXTENSIONS_DYNAMIC = ("json", "json_", "json.gz", "log", ".log.gz", ".zip")
+EXTENSIONS_BINEXPORT2 = ("BinExport", "BinExport2")
 EXTENSIONS_ELF = "elf_"
 EXTENSIONS_FREEZE = "frz"
 
 logger = logging.getLogger("capa")
 
 
+# shared console used to redirect logging to stderr
+log_console: Console = Console(stderr=True)
+
+
 def hex(n: int) -> str:
     """render the given number using upper case hex, like: 0x123ABC"""
     if n < 0:
@@ -74,6 +100,59 @@ def assert_never(value) -> NoReturn:
     assert False, f"Unhandled value: {value} ({type(value).__name__})"  # noqa: B011
 
 
+@contextlib.contextmanager
+def stdout_redirector(stream):
+    """
+    Redirect stdout at the C runtime level,
+     which lets us handle native libraries that spam stdout.
+
+    *But*, this only works on Linux! Otherwise will silently still write to stdout.
+    So, try to upstream the fix when possible.
+
+    Via: https://eli.thegreenplace.net/2015/redirecting-all-kinds-of-stdout-in-python/
+    """
+    if sys.platform not in ("linux", "linux2"):
+        logger.warning("Unable to capture STDOUT on non-Linux (begin)")
+        yield
+        logger.warning("Unable to capture STDOUT on non-Linux (end)")
+        return
+
+    # libc is only on Linux
+    LIBC = ctypes.CDLL(None)
+    C_STDOUT = ctypes.c_void_p.in_dll(LIBC, "stdout")
+
+    # The original fd stdout points to. Usually 1 on POSIX systems.
+    original_stdout_fd = sys.stdout.fileno()
+
+    def _redirect_stdout(to_fd):
+        """Redirect stdout to the given file descriptor."""
+        # Flush the C-level buffer stdout
+        LIBC.fflush(C_STDOUT)
+        # Flush and close sys.stdout - also closes the file descriptor (fd)
+        sys.stdout.close()
+        # Make original_stdout_fd point to the same file as to_fd
+        os.dup2(to_fd, original_stdout_fd)
+        # Create a new sys.stdout that points to the redirected fd
+        sys.stdout = io.TextIOWrapper(os.fdopen(original_stdout_fd, "wb"))
+
+    # Save a copy of the original stdout fd in saved_stdout_fd
+    saved_stdout_fd = os.dup(original_stdout_fd)
+    try:
+        # Create a temporary file and redirect stdout to it
+        tfile = tempfile.TemporaryFile(mode="w+b")
+        _redirect_stdout(tfile.fileno())
+        # Yield to caller, then redirect stdout back to the saved fd
+        yield
+        _redirect_stdout(saved_stdout_fd)
+        # Copy contents of temporary file to the given stream
+        tfile.flush()
+        tfile.seek(0, io.SEEK_SET)
+        stream.write(tfile.read())
+    finally:
+        tfile.close()
+        os.close(saved_stdout_fd)
+
+
 def load_json_from_path(json_path: Path):
     with gzip.open(json_path, "r") as compressed_report:
         try:
@@ -123,16 +202,20 @@ def get_format_from_report(sample: Path) -> str:
         line = load_one_jsonl_from_path(sample)
         if "Plugin" in line:
             return FORMAT_DRAKVUF
-        return FORMAT_UNKNOWN
+    elif sample.name.endswith(".zip"):
-
+        with ZipFile(sample, "r") as zipfile:
-    report = load_json_from_path(sample)
+            namelist: List[str] = zipfile.namelist()
-    if "CAPE" in report:
+            if "logs/summary_v2.json" in namelist and "logs/flog.xml" in namelist:
-        return FORMAT_CAPE
+                # assume VMRay zipfile at a minimum has these files
-
+                return FORMAT_VMRAY
-    if "target" in report and "info" in report and "behavior" in report:
+    elif sample.name.endswith(("json", "json_", "json.gz")):
-        # CAPE report that's missing the "CAPE" key,
+        report = load_json_from_path(sample)
-        # which is not going to be much use, but its correct.
+        if "CAPE" in report:
-        return FORMAT_CAPE
+            return FORMAT_CAPE
+        if "target" in report and "info" in report and "behavior" in report:
+            # CAPE report that's missing the "CAPE" key,
+            # which is not going to be much use, but its correct.
+            return FORMAT_CAPE
 
     return FORMAT_UNKNOWN
 
@@ -147,6 +230,8 @@ def get_format_from_extension(sample: Path) -> str:
         format_ = get_format_from_report(sample)
     elif sample.name.endswith(EXTENSIONS_FREEZE):
         format_ = FORMAT_FREEZE
+    elif sample.name.endswith(EXTENSIONS_BINEXPORT2):
+        format_ = FORMAT_BINEXPORT2
     return format_
 
 
@@ -178,45 +263,13 @@ def get_format(sample: Path) -> str:
     return FORMAT_UNKNOWN
 
 
-@contextlib.contextmanager
-def redirecting_print_to_tqdm(disable_progress):
-    """
-    tqdm (progress bar) expects to have fairly tight control over console output.
-    so calls to `print()` will break the progress bar and make things look bad.
-    so, this context manager temporarily replaces the `print` implementation
-    with one that is compatible with tqdm.
-    via: https://stackoverflow.com/a/42424890/87207
-    """
-    old_print = print  # noqa: T202 [reserved word print used]
-
-    def new_print(*args, **kwargs):
-        # If tqdm.tqdm.write raises error, use builtin print
-        if disable_progress:
-            old_print(*args, **kwargs)
-        else:
-            try:
-                tqdm.tqdm.write(*args, **kwargs)
-            except Exception:
-                old_print(*args, **kwargs)
-
-    try:
-        # Globally replace print with new_print.
-        # Verified this works manually on Python 3.11:
-        #     >>> import inspect
-        #     >>> inspect.builtins
-        #     <module 'builtins' (built-in)>
-        inspect.builtins.print = new_print  # type: ignore
-        yield
-    finally:
-        inspect.builtins.print = old_print  # type: ignore
-
-
 def log_unsupported_format_error():
     logger.error("-" * 80)
     logger.error(" Input file does not appear to be a supported file.")
     logger.error(" ")
     logger.error(" See all supported file formats via capa's help output (-h).")
-    logger.error(" If you don't know the input file type, you can try using the `file` utility to guess it.")
+    logger.error(" If you don't know the input file type,")
+    logger.error(" you can try using the `file` utility to guess it.")
     logger.error("-" * 80)
 
 
@@ -242,6 +295,17 @@ def log_unsupported_drakvuf_report_error(error: str):
     logger.error("-" * 80)
 
 
+def log_unsupported_vmray_report_error(error: str):
+    logger.error("-" * 80)
+    logger.error(" Input file is not a valid VMRay analysis archive: %s", error)
+    logger.error(" ")
+    logger.error(
+        " capa only supports analyzing VMRay dynamic analysis archives containing summary_v2.json and flog.xml log files."
+    )
+    logger.error(" Please make sure you have downloaded a dynamic analysis archive from VMRay.")
+    logger.error("-" * 80)
+
+
 def log_empty_sandbox_report_error(error: str, sandbox_name: str):
     logger.error("-" * 80)
     logger.error(" %s report is empty or only contains little useful data: %s", sandbox_name, error)
@@ -254,9 +318,8 @@ def log_unsupported_os_error():
     logger.error("-" * 80)
     logger.error(" Input file does not appear to target a supported OS.")
     logger.error(" ")
-    logger.error(
+    logger.error(" capa currently only analyzes executables for some operating systems")
-        " capa currently only supports analyzing executables for some operating systems (including Windows and Linux)."
+    logger.error(" (including Windows, Linux, and Android).")
-    )
     logger.error("-" * 80)
 
 
@@ -274,9 +337,8 @@ def log_unsupported_runtime_error():
     logger.error(" ")
     logger.error(" capa supports running under Python 3.8 and higher.")
     logger.error(" ")
-    logger.error(
+    logger.error(" If you're seeing this message on the command line,")
-        " If you're seeing this message on the command line, please ensure you're running a supported Python version."
+    logger.error(" please ensure you're running a supported Python version.")
-    )
     logger.error("-" * 80)
 
 
@@ -291,3 +353,110 @@ def is_running_standalone() -> bool:
     # so we keep this in a common area.
     # generally, other library code should not use this function.
     return hasattr(sys, "frozen") and hasattr(sys, "_MEIPASS")
+
+
+def is_dev_environment() -> bool:
+    if is_running_standalone():
+        return False
+
+    if "site-packages" in __file__:
+        # running from a site-packages installation
+        return False
+
+    capa_root = Path(__file__).resolve().parent.parent
+    git_dir = capa_root / ".git"
+
+    if not git_dir.is_dir():
+        # .git directory doesn't exist
+        return False
+
+    return True
+
+
+def is_cache_newer_than_rule_code(cache_dir: Path) -> bool:
+    """
+    basic check to prevent issues if the rules cache is older than relevant rules code
+
+    args:
+      cache_dir: the cache directory containing cache files
+
+    returns:
+      True if latest cache file is newer than relevant rule cache code
+    """
+
+    # retrieve the latest modified cache file
+    cache_files = list(cache_dir.glob("*.cache"))
+    if not cache_files:
+        logger.debug("no rule cache files found")
+        return False
+
+    latest_cache_file = max(cache_files, key=os.path.getmtime)
+    cache_timestamp = os.path.getmtime(latest_cache_file)
+
+    # these are the relevant rules code files that could conflict with using an outdated cache
+    # delayed import due to circular dependencies
+    import capa.rules
+    import capa.rules.cache
+
+    latest_rule_code_file = max([Path(capa.rules.__file__), Path(capa.rules.cache.__file__)], key=os.path.getmtime)
+    rule_code_timestamp = os.path.getmtime(latest_rule_code_file)
+
+    if rule_code_timestamp > cache_timestamp:
+
+        def ts_to_str(ts):
+            return datetime.fromtimestamp(ts).strftime("%Y-%m-%d %H:%M:%S")
+
+        logger.warning(
+            "latest rule code file %s (%s) is newer than the latest rule cache file %s (%s)",
+            latest_rule_code_file,
+            ts_to_str(rule_code_timestamp),
+            latest_cache_file,
+            ts_to_str(cache_timestamp),
+        )
+        return False
+
+    return True
+
+
+class RateColumn(ProgressColumn):
+    """Renders speed column in progress bar."""
+
+    def render(self, task: "Task") -> Text:
+        speed = f"{task.speed:>.1f}" if task.speed else "00.0"
+        unit = task.fields.get("unit", "it")
+        return Text.from_markup(f"[progress.data.speed]{speed} {unit}/s")
+
+
+class PostfixColumn(ProgressColumn):
+    """Renders a postfix column in progress bar."""
+
+    def render(self, task: "Task") -> Text:
+        return Text(task.fields.get("postfix", ""))
+
+
+class MofNCompleteColumnWithUnit(MofNCompleteColumn):
+    """Renders completed/total count column with a unit."""
+
+    def render(self, task: "Task") -> Text:
+        ret = super().render(task)
+        unit = task.fields.get("unit")
+        return ret.append(f" {unit}") if unit else ret
+
+
+class CapaProgressBar(Progress):
+    @classmethod
+    def get_default_columns(cls):
+        return (
+            SpinnerColumn(),
+            TextColumn("[progress.description]{task.description}"),
+            TaskProgressColumn(),
+            BarColumn(),
+            MofNCompleteColumnWithUnit(),
+            "•",
+            TimeElapsedColumn(),
+            "<",
+            TimeRemainingColumn(),
+            "•",
+            RateColumn(),
+            PostfixColumn(),
+        )

capa/ida/helpers.py

@@ -13,6 +13,8 @@ from pathlib import Path
 
 import idc
 import idaapi
+import ida_ida
+import ida_nalt
 import idautils
 import ida_bytes
 import ida_loader
@@ -45,6 +47,51 @@ NETNODE_RESULTS = "results"
 NETNODE_RULES_CACHE_ID = "rules-cache-id"
 
 
+# wrappers for IDA Pro (IDAPython) 7, 8 and 9 compability
+version = float(idaapi.get_kernel_version())
+if version < 9.0:
+
+    def get_filetype() -> "ida_ida.filetype_t":
+        return idaapi.get_inf_structure().filetype
+
+    def get_processor_name() -> str:
+        return idaapi.get_inf_structure().procname
+
+    def is_32bit() -> bool:
+        info: idaapi.idainfo = idaapi.get_inf_structure()
+        return info.is_32bit()
+
+    def is_64bit() -> bool:
+        info: idaapi.idainfo = idaapi.get_inf_structure()
+        return info.is_64bit()
+
+    def retrieve_input_file_md5() -> str:
+        return ida_nalt.retrieve_input_file_md5()
+
+    def retrieve_input_file_sha256() -> str:
+        return ida_nalt.retrieve_input_file_sha256()
+
+else:
+
+    def get_filetype() -> "ida_ida.filetype_t":
+        return ida_ida.inf_get_filetype()
+
+    def get_processor_name() -> str:
+        return idc.get_processor_name()
+
+    def is_32bit() -> bool:
+        return idaapi.inf_is_32bit_exactly()
+
+    def is_64bit() -> bool:
+        return idaapi.inf_is_64bit()
+
+    def retrieve_input_file_md5() -> str:
+        return ida_nalt.retrieve_input_file_md5().hex()
+
+    def retrieve_input_file_sha256() -> str:
+        return ida_nalt.retrieve_input_file_sha256().hex()
+
+
 def inform_user_ida_ui(message):
     # this isn't a logger, this is IDA's logging facility
     idaapi.info(f"{message}. Please refer to IDA Output window for more information.")  # noqa: G004
@@ -52,17 +99,16 @@ def inform_user_ida_ui(message):
 
 def is_supported_ida_version():
     version = float(idaapi.get_kernel_version())
-    if version < 7.4 or version >= 9:
+    if version < 7.4 or version >= 10:
         warning_msg = "This plugin does not support your IDA Pro version"
         logger.warning(warning_msg)
-        logger.warning("Your IDA Pro version is: %s. Supported versions are: IDA >= 7.4 and IDA < 9.0.", version)
+        logger.warning("Your IDA Pro version is: %s. Supported versions are: IDA >= 7.4 and IDA < 10.0.", version)
         return False
     return True
 
 
 def is_supported_file_type():
-    file_info = idaapi.get_inf_structure()
+    if get_filetype() not in SUPPORTED_FILE_TYPES:
-    if file_info.filetype not in SUPPORTED_FILE_TYPES:
         logger.error("-" * 80)
         logger.error(" Input file does not appear to be a supported file type.")
         logger.error(" ")
@@ -76,8 +122,7 @@ def is_supported_file_type():
 
 
 def is_supported_arch_type():
-    file_info = idaapi.get_inf_structure()
+    if get_processor_name() not in SUPPORTED_ARCH_TYPES or not any((is_32bit(), is_64bit())):
-    if file_info.procname not in SUPPORTED_ARCH_TYPES or not any((file_info.is_32bit(), file_info.is_64bit())):
         logger.error("-" * 80)
         logger.error(" Input file does not appear to target a supported architecture.")
         logger.error(" ")
@@ -125,10 +170,10 @@ def collect_metadata(rules: List[Path]):
     md5 = get_file_md5()
     sha256 = get_file_sha256()
 
-    info: idaapi.idainfo = idaapi.get_inf_structure()
+    procname = get_processor_name()
-    if info.procname == "metapc" and info.is_64bit():
+    if procname == "metapc" and is_64bit():
         arch = "x86_64"
-    elif info.procname == "metapc" and info.is_32bit():
+    elif procname == "metapc" and is_32bit():
         arch = "x86"
     else:
         arch = "unknown arch"

capa/loader.py

@@ -5,7 +5,8 @@
 # Unless required by applicable law or agreed to in writing, software distributed under the License
 #  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and limitations under the License.
-import sys
+import io
+import os
 import logging
 import datetime
 import contextlib
@@ -44,8 +45,10 @@ from capa.features.common import (
     FORMAT_CAPE,
     FORMAT_SC32,
     FORMAT_SC64,
+    FORMAT_VMRAY,
     FORMAT_DOTNET,
     FORMAT_DRAKVUF,
+    FORMAT_BINEXPORT2,
 )
 from capa.features.address import Address
 from capa.features.extractors.base_extractor import (
@@ -63,7 +66,10 @@ BACKEND_BINJA = "binja"
 BACKEND_PEFILE = "pefile"
 BACKEND_CAPE = "cape"
 BACKEND_DRAKVUF = "drakvuf"
+BACKEND_VMRAY = "vmray"
 BACKEND_FREEZE = "freeze"
+BACKEND_BINEXPORT2 = "binexport2"
+BACKEND_IDA = "ida"
 
 
 class CorruptFile(ValueError):
@@ -165,6 +171,7 @@ def get_workspace(path: Path, input_format: str, sigpaths: List[Path]):
         # to do a subclass check via isinstance.
         if type(e) is Exception and "Couldn't convert rva" in e.args[0]:
             raise CorruptFile(e.args[0]) from e
+        raise
 
     viv_utils.flirt.register_flirt_signature_analyzers(vw, [str(s) for s in sigpaths])
 
@@ -218,6 +225,11 @@ def get_extractor(
         report = capa.helpers.load_jsonl_from_path(input_path)
         return capa.features.extractors.drakvuf.extractor.DrakvufExtractor.from_report(report)
 
+    elif backend == BACKEND_VMRAY:
+        import capa.features.extractors.vmray.extractor
+
+        return capa.features.extractors.vmray.extractor.VMRayExtractor.from_zipfile(input_path)
+
     elif backend == BACKEND_DOTNET:
         import capa.features.extractors.dnfile.extractor
 
@@ -227,24 +239,15 @@ def get_extractor(
         return capa.features.extractors.dnfile.extractor.DnfileFeatureExtractor(input_path)
 
     elif backend == BACKEND_BINJA:
-        import capa.helpers
+        import capa.features.extractors.binja.find_binja_api as finder
-        from capa.features.extractors.binja.find_binja_api import find_binja_path
 
-        # When we are running as a standalone executable, we cannot directly import binaryninja
+        if not finder.has_binaryninja():
-        # We need to fist find the binja API installation path and add it into sys.path
+            raise RuntimeError("cannot find Binary Ninja API module.")
-        if capa.helpers.is_running_standalone():
-            bn_api = find_binja_path()
-            if bn_api.exists():
-                sys.path.append(str(bn_api))
 
-        try:
+        if not finder.load_binaryninja():
-            import binaryninja
+            raise RuntimeError("failed to load Binary Ninja API module.")
-            from binaryninja import BinaryView
+
-        except ImportError:
+        import binaryninja
-            raise RuntimeError(
-                "Cannot import binaryninja module. Please install the Binary Ninja Python API first: "
-                + "https://docs.binary.ninja/dev/batch.html#install-the-api)."
-            )
 
         import capa.features.extractors.binja.extractor
 
@@ -259,7 +262,7 @@ def get_extractor(
                 raise UnsupportedOSError()
 
         with console.status("analyzing program...", spinner="dots"):
-            bv: BinaryView = binaryninja.load(str(input_path))
+            bv: binaryninja.BinaryView = binaryninja.load(str(input_path))
             if bv is None:
                 raise RuntimeError(f"Binary Ninja cannot open file {input_path}")
 
@@ -301,10 +304,70 @@ def get_extractor(
     elif backend == BACKEND_FREEZE:
         return frz.load(input_path.read_bytes())
 
+    elif backend == BACKEND_BINEXPORT2:
+        import capa.features.extractors.binexport2
+        import capa.features.extractors.binexport2.extractor
+
+        be2 = capa.features.extractors.binexport2.get_binexport2(input_path)
+        assert sample_path is not None
+        buf = sample_path.read_bytes()
+
+        return capa.features.extractors.binexport2.extractor.BinExport2FeatureExtractor(be2, buf)
+
+    elif backend == BACKEND_IDA:
+        import capa.features.extractors.ida.idalib as idalib
+
+        if not idalib.has_idalib():
+            raise RuntimeError("cannot find IDA idalib module.")
+
+        if not idalib.load_idalib():
+            raise RuntimeError("failed to load IDA idalib module.")
+
+        import idapro
+        import ida_auto
+
+        import capa.features.extractors.ida.extractor
+
+        logger.debug("idalib: opening database...")
+        # idalib writes to stdout (ugh), so we have to capture that
+        # so as not to screw up structured output.
+        with capa.helpers.stdout_redirector(io.BytesIO()):
+            with console.status("analyzing program...", spinner="dots"):
+                if idapro.open_database(str(input_path), run_auto_analysis=True):
+                    raise RuntimeError("failed to analyze input file")
+
+            logger.debug("idalib: waiting for analysis...")
+            ida_auto.auto_wait()
+            logger.debug("idalib: opened database.")
+
+        return capa.features.extractors.ida.extractor.IdaFeatureExtractor()
+
     else:
         raise ValueError("unexpected backend: " + backend)
 
 
+def _get_binexport2_file_extractors(input_file: Path) -> List[FeatureExtractor]:
+    # I'm not sure this is where this logic should live, but it works for now.
+    # we'll keep this a "private" routine until we're sure.
+    import capa.features.extractors.binexport2
+
+    be2 = capa.features.extractors.binexport2.get_binexport2(input_file)
+    sample_path = capa.features.extractors.binexport2.get_sample_from_binexport2(
+        input_file, be2, [Path(os.environ.get("CAPA_SAMPLES_DIR", "."))]
+    )
+
+    with sample_path.open("rb") as f:
+        taste = f.read()
+
+    if taste.startswith(capa.features.extractors.common.MATCH_PE):
+        return get_file_extractors(sample_path, FORMAT_PE)
+    elif taste.startswith(capa.features.extractors.common.MATCH_ELF):
+        return get_file_extractors(sample_path, FORMAT_ELF)
+    else:
+        logger.warning("unsupported format")
+        return []
+
+
 def get_file_extractors(input_file: Path, input_format: str) -> List[FeatureExtractor]:
     file_extractors: List[FeatureExtractor] = []
 
@@ -342,6 +405,14 @@ def get_file_extractors(input_file: Path, input_format: str) -> List[FeatureExtr
         report = capa.helpers.load_jsonl_from_path(input_file)
         file_extractors.append(capa.features.extractors.drakvuf.extractor.DrakvufExtractor.from_report(report))
 
+    elif input_format == FORMAT_VMRAY:
+        import capa.features.extractors.vmray.extractor
+
+        file_extractors.append(capa.features.extractors.vmray.extractor.VMRayExtractor.from_zipfile(input_file))
+
+    elif input_format == FORMAT_BINEXPORT2:
+        file_extractors = _get_binexport2_file_extractors(input_file)
+
     return file_extractors
 
 

capa/main.py

@@ -22,6 +22,7 @@ from pathlib import Path
 
 import colorama
 from pefile import PEFormatError
+from rich.logging import RichHandler
 from elftools.common.exceptions import ELFError
 
 import capa.perf
@@ -43,13 +44,16 @@ import capa.features.extractors.common
 from capa.rules import RuleSet
 from capa.engine import MatchResults
 from capa.loader import (
+    BACKEND_IDA,
     BACKEND_VIV,
     BACKEND_CAPE,
     BACKEND_BINJA,
+    BACKEND_VMRAY,
     BACKEND_DOTNET,
     BACKEND_FREEZE,
     BACKEND_PEFILE,
     BACKEND_DRAKVUF,
+    BACKEND_BINEXPORT2,
 )
 from capa.helpers import (
     get_file_taste,
@@ -59,6 +63,7 @@ from capa.helpers import (
     log_unsupported_format_error,
     log_empty_sandbox_report_error,
     log_unsupported_cape_report_error,
+    log_unsupported_vmray_report_error,
     log_unsupported_drakvuf_report_error,
 )
 from capa.exceptions import (
@@ -80,12 +85,14 @@ from capa.features.common import (
     FORMAT_CAPE,
     FORMAT_SC32,
     FORMAT_SC64,
+    FORMAT_VMRAY,
     FORMAT_DOTNET,
     FORMAT_FREEZE,
     FORMAT_RESULT,
     FORMAT_DRAKVUF,
     STATIC_FORMATS,
     DYNAMIC_FORMATS,
+    FORMAT_BINEXPORT2,
 )
 from capa.capabilities.common import find_capabilities, has_file_limitation, find_file_capabilities
 from capa.features.extractors.base_extractor import (
@@ -190,12 +197,13 @@ def simple_message_exception_handler(exctype, value: BaseException, traceback: T
     """
 
     if exctype is KeyboardInterrupt:
-        print("KeyboardInterrupt detected, program terminated")
+        print("KeyboardInterrupt detected, program terminated", file=sys.stderr)
     else:
         print(
             f"Unexpected exception raised: {exctype}. Please run capa in debug mode (-d/--debug) "
             + "to see the stack trace. Please also report your issue on the capa GitHub page so we "
-            + "can improve the code! (https://github.com/mandiant/capa/issues)"
+            + "can improve the code! (https://github.com/mandiant/capa/issues)",
+            file=sys.stderr,
         )
 
 
@@ -259,7 +267,9 @@ def install_common_args(parser, wanted=None):
             (FORMAT_SC64, "64-bit shellcode"),
             (FORMAT_CAPE, "CAPE sandbox report"),
             (FORMAT_DRAKVUF, "DRAKVUF sandbox report"),
+            (FORMAT_VMRAY, "VMRay sandbox report"),
             (FORMAT_FREEZE, "features previously frozen by capa"),
+            (FORMAT_BINEXPORT2, "BinExport2"),
         ]
         format_help = ", ".join([f"{f[0]}: {f[1]}" for f in formats])
 
@@ -275,12 +285,15 @@ def install_common_args(parser, wanted=None):
         backends = [
             (BACKEND_AUTO, "(default) detect appropriate backend automatically"),
             (BACKEND_VIV, "vivisect"),
+            (BACKEND_IDA, "IDA via idalib"),
             (BACKEND_PEFILE, "pefile (file features only)"),
             (BACKEND_BINJA, "Binary Ninja"),
             (BACKEND_DOTNET, ".NET"),
+            (BACKEND_BINEXPORT2, "BinExport2"),
             (BACKEND_FREEZE, "capa freeze"),
             (BACKEND_CAPE, "CAPE"),
             (BACKEND_DRAKVUF, "DRAKVUF"),
+            (BACKEND_VMRAY, "VMRay"),
         ]
         backend_help = ", ".join([f"{f[0]}: {f[1]}" for f in backends])
         parser.add_argument(
@@ -393,15 +406,23 @@ def handle_common_args(args):
       ShouldExitError: if the program is invoked incorrectly and should exit.
     """
     if args.quiet:
-        logging.basicConfig(level=logging.WARNING)
         logging.getLogger().setLevel(logging.WARNING)
     elif args.debug:
-        logging.basicConfig(level=logging.DEBUG)
         logging.getLogger().setLevel(logging.DEBUG)
     else:
-        logging.basicConfig(level=logging.INFO)
         logging.getLogger().setLevel(logging.INFO)
 
+    # use [/] after the logger name to reset any styling,
+    # and prevent the color from carrying over to the message
+    logformat = "[dim]%(name)s[/]: %(message)s"
+
+    # set markup=True to allow the use of Rich's markup syntax in log messages
+    rich_handler = RichHandler(markup=True, show_time=False, show_path=True, console=capa.helpers.log_console)
+    rich_handler.setFormatter(logging.Formatter(logformat))
+
+    # use RichHandler for root logger
+    logging.getLogger().addHandler(rich_handler)
+
     # disable vivisect-related logging, it's verbose and not relevant for capa users
     set_vivisect_log_level(logging.CRITICAL)
 
@@ -445,8 +466,12 @@ def handle_common_args(args):
         if args.rules == [RULES_PATH_DEFAULT_STRING]:
             logger.debug("-" * 80)
             logger.debug(" Using default embedded rules.")
-            logger.debug(" To provide your own rules, use the form `capa.exe -r ./path/to/rules/  /path/to/mal.exe`.")
+            logger.debug(" To provide your own rules, use the form:")
+            logger.debug("")
+            logger.debug("     `capa.exe -r ./path/to/rules/  /path/to/mal.exe`.")
+            logger.debug("")
             logger.debug(" You can see the current default rule set here:")
+            logger.debug("")
             logger.debug("     https://github.com/mandiant/capa-rules")
             logger.debug("-" * 80)
 
@@ -552,12 +577,18 @@ def get_backend_from_cli(args, input_format: str) -> str:
     if input_format == FORMAT_DRAKVUF:
         return BACKEND_DRAKVUF
 
+    elif input_format == FORMAT_VMRAY:
+        return BACKEND_VMRAY
+
     elif input_format == FORMAT_DOTNET:
         return BACKEND_DOTNET
 
     elif input_format == FORMAT_FREEZE:
         return BACKEND_FREEZE
 
+    elif input_format == FORMAT_BINEXPORT2:
+        return BACKEND_BINEXPORT2
+
     else:
         return BACKEND_VIV
 
@@ -576,8 +607,15 @@ def get_sample_path_from_cli(args, backend: str) -> Optional[Path]:
     raises:
       ShouldExitError: if the program is invoked incorrectly and should exit.
     """
-    if backend in (BACKEND_CAPE, BACKEND_DRAKVUF):
+    if backend in (BACKEND_CAPE, BACKEND_DRAKVUF, BACKEND_VMRAY):
         return None
+    elif backend == BACKEND_BINEXPORT2:
+        import capa.features.extractors.binexport2
+
+        be2 = capa.features.extractors.binexport2.get_binexport2(args.input_file)
+        return capa.features.extractors.binexport2.get_sample_from_binexport2(
+            args.input_file, be2, [Path(os.environ.get("CAPA_SAMPLES_DIR", "."))]
+        )
     else:
         return args.input_file
 
@@ -612,13 +650,22 @@ def get_rules_from_cli(args) -> RuleSet:
     raises:
       ShouldExitError: if the program is invoked incorrectly and should exit.
     """
+    enable_cache: bool = True
     try:
         if capa.helpers.is_running_standalone() and args.is_default_rules:
             cache_dir = get_default_root() / "cache"
         else:
             cache_dir = capa.rules.cache.get_default_cache_directory()
 
-        rules = capa.rules.get_rules(args.rules, cache_dir=cache_dir)
+        if capa.helpers.is_dev_environment():
+            # using the rules cache during development may result in unexpected errors, see #1898
+            enable_cache = capa.helpers.is_cache_newer_than_rule_code(cache_dir)
+            if not enable_cache:
+                logger.debug("not using cache. delete the cache file manually to use rule caching again")
+            else:
+                logger.debug("cache can be used, no potentially outdated cache files found")
+
+        rules = capa.rules.get_rules(args.rules, cache_dir=cache_dir, enable_cache=enable_cache)
     except (IOError, capa.rules.InvalidRule, capa.rules.InvalidRuleSet) as e:
         logger.error("%s", str(e))
         logger.error(
@@ -681,6 +728,8 @@ def get_file_extractors_from_cli(args, input_format: str) -> List[FeatureExtract
             log_unsupported_cape_report_error(str(e))
         elif input_format == FORMAT_DRAKVUF:
             log_unsupported_drakvuf_report_error(str(e))
+        elif input_format == FORMAT_VMRAY:
+            log_unsupported_vmray_report_error(str(e))
         else:
             log_unsupported_format_error()
         raise ShouldExitError(E_INVALID_FILE_TYPE) from e
@@ -783,6 +832,9 @@ def get_extractor_from_cli(args, input_format: str, backend: str) -> FeatureExtr
     sample_path = get_sample_path_from_cli(args, backend)
     extractor_filters = get_extractor_filters_from_cli(args, input_format)
 
+    logger.debug("format:  %s", input_format)
+    logger.debug("backend: %s", backend)
+
     try:
         extractor = capa.loader.get_extractor(
             args.input_file,
@@ -800,6 +852,8 @@ def get_extractor_from_cli(args, input_format: str, backend: str) -> FeatureExtr
             log_unsupported_cape_report_error(str(e))
         elif input_format == FORMAT_DRAKVUF:
             log_unsupported_drakvuf_report_error(str(e))
+        elif input_format == FORMAT_VMRAY:
+            log_unsupported_vmray_report_error(str(e))
         else:
             log_unsupported_format_error()
         raise ShouldExitError(E_INVALID_FILE_TYPE) from e
@@ -849,6 +903,10 @@ def apply_extractor_filters(extractor: FeatureExtractor, extractor_filters: Filt
 def main(argv: Optional[List[str]] = None):
     if sys.version_info < (3, 8):
         raise UnsupportedRuntimeError("This version of capa can only be used with Python 3.8+")
+    elif sys.version_info < (3, 10):
+        from warnings import warn
+
+        warn("This is the last capa version supporting Python 3.8 and 3.9.", DeprecationWarning, stacklevel=2)
 
     if argv is None:
         argv = sys.argv[1:]
@@ -860,6 +918,9 @@ def main(argv: Optional[List[str]] = None):
         You can see the rule set here:
           https://github.com/mandiant/capa-rules
 
+        You can load capa JSON output to capa Explorer Web:
+          https://github.com/mandiant/capa/explorer
+
         To provide your own rule set, use the `-r` flag:
           capa  --rules /path/to/rules  suspicious.exe
           capa  -r      /path/to/rules  suspicious.exe

capa/render/default.py

@@ -6,18 +6,43 @@
 #  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and limitations under the License.
 
+import io
 import collections
+import urllib.parse
 
-import tabulate
+import rich
+import rich.table
+import rich.console
+from rich.console import Console
 
 import capa.render.utils as rutils
 import capa.render.result_document as rd
 import capa.features.freeze.features as frzf
 from capa.rules import RuleSet
 from capa.engine import MatchResults
-from capa.render.utils import StringIO
 
-tabulate.PRESERVE_WHITESPACE = True
+
+def bold_markup(s) -> str:
+    """
+    Generate Rich markup in a bold style.
+
+    The resulting string should be passed to a Rich renderable
+    and/or printed via Rich or the markup will be visible to the user.
+    """
+    return f"[cyan]{s}[/cyan]"
+
+
+def link_markup(s: str, href: str) -> str:
+    """
+    Generate Rich markup for a clickable hyperlink.
+    This works in many modern terminals.
+    When it doesn't work, the fallback is just to show the link name (s),
+     as if it was not a link.
+
+    The resulting string should be passed to a Rich renderable
+    and/or printed via Rich or the markup will be visible to the user.
+    """
+    return f"[link={href}]{s}[/link]"
 
 
 def width(s: str, character_count: int) -> str:
@@ -28,11 +53,16 @@ def width(s: str, character_count: int) -> str:
         return s
 
 
-def render_meta(doc: rd.ResultDocument, ostream: StringIO):
+def render_sample_link(hash: str) -> str:
+    url = "https://www.virustotal.com/gui/file/" + hash
+    return link_markup(hash, url)
+
+
+def render_meta(doc: rd.ResultDocument, console: Console):
     rows = [
-        (width("md5", 22), width(doc.meta.sample.md5, 82)),
+        ("md5", render_sample_link(doc.meta.sample.md5)),
-        ("sha1", doc.meta.sample.sha1),
+        ("sha1", render_sample_link(doc.meta.sample.sha1)),
-        ("sha256", doc.meta.sample.sha256),
+        ("sha256", render_sample_link(doc.meta.sample.sha256)),
         ("analysis", doc.meta.flavor.value),
         ("os", doc.meta.analysis.os),
         ("format", doc.meta.analysis.format),
@@ -40,8 +70,14 @@ def render_meta(doc: rd.ResultDocument, ostream: StringIO):
         ("path", doc.meta.sample.path),
     ]
 
-    ostream.write(tabulate.tabulate(rows, tablefmt="mixed_outline"))
+    table = rich.table.Table(show_header=False, min_width=100)
-    ostream.write("\n")
+    table.add_column()
+    table.add_column()
+
+    for row in rows:
+        table.add_row(*row)
+
+    console.print(table)
 
 
 def find_subrule_matches(doc: rd.ResultDocument):
@@ -71,7 +107,12 @@ def find_subrule_matches(doc: rd.ResultDocument):
     return matches
 
 
-def render_capabilities(doc: rd.ResultDocument, ostream: StringIO):
+def render_rule_name(name: str) -> str:
+    url = f"https://mandiant.github.io/capa/rules/{urllib.parse.quote(name)}/"
+    return bold_markup(link_markup(name, url))
+
+
+def render_capabilities(doc: rd.ResultDocument, console: Console):
     """
     example::
 
@@ -95,25 +136,30 @@ def render_capabilities(doc: rd.ResultDocument, ostream: StringIO):
 
         count = len(rule.matches)
         if count == 1:
-            capability = rutils.bold(rule.meta.name)
+            capability = render_rule_name(rule.meta.name)
         else:
-            capability = f"{rutils.bold(rule.meta.name)} ({count} matches)"
+            capability = render_rule_name(rule.meta.name) + f" ({count} matches)"
         rows.append((capability, rule.meta.namespace))
 
     if rows:
-        ostream.write(
+        table = rich.table.Table(min_width=100)
-            tabulate.tabulate(
+        table.add_column(width("Capability", 20))
-                rows,
+        table.add_column("Namespace")
-                headers=[width("Capability", 50), width("Namespace", 50)],
+
-                tablefmt="mixed_outline",
+        for row in rows:
-            )
+            table.add_row(*row)
-        )
+
-        ostream.write("\n")
+        console.print(table)
     else:
-        ostream.writeln(rutils.bold("no capabilities found"))
+        console.print(bold_markup("no capabilities found"))
 
 
-def render_attack(doc: rd.ResultDocument, ostream: StringIO):
+def render_attack_link(id: str) -> str:
+    url = f"https://attack.mitre.org/techniques/{id.replace('.', '/')}/"
+    return rf"\[{link_markup(id, url)}]"
+
+
+def render_attack(doc: rd.ResultDocument, console: Console):
     """
     example::
 
@@ -132,35 +178,36 @@ def render_attack(doc: rd.ResultDocument, ostream: StringIO):
     tactics = collections.defaultdict(set)
     for rule in rutils.capability_rules(doc):
         for attack in rule.meta.attack:
-            tactics[attack.tactic].add((attack.technique, attack.subtechnique, attack.id))
+            tactics[attack.tactic].add((attack.technique, attack.subtechnique, attack.id.strip("[").strip("]")))
 
     rows = []
     for tactic, techniques in sorted(tactics.items()):
         inner_rows = []
         for technique, subtechnique, id in sorted(techniques):
             if not subtechnique:
-                inner_rows.append(f"{rutils.bold(technique)} {id}")
+                # example: File and Directory Discovery [T1083]
+                inner_rows.append(f"{bold_markup(technique)} {render_attack_link(id)}")
             else:
-                inner_rows.append(f"{rutils.bold(technique)}::{subtechnique} {id}")
+                # example: Code Discovery::Enumerate PE Sections [T1084.001]
-        rows.append(
+                inner_rows.append(f"{bold_markup(technique)}::{subtechnique} {render_attack_link(id)}")
-            (
+
-                rutils.bold(tactic.upper()),
+        tactic = bold_markup(tactic.upper())
-                "\n".join(inner_rows),
+        technique = "\n".join(inner_rows)
-            )
+
-        )
+        rows.append((tactic, technique))
 
     if rows:
-        ostream.write(
+        table = rich.table.Table(min_width=100)
-            tabulate.tabulate(
+        table.add_column(width("ATT&CK Tactic", 20))
-                rows,
+        table.add_column("ATT&CK Technique")
-                headers=[width("ATT&CK Tactic", 20), width("ATT&CK Technique", 80)],
+
-                tablefmt="mixed_grid",
+        for row in rows:
-            )
+            table.add_row(*row)
-        )
+
-        ostream.write("\n")
+        console.print(table)
 
 
-def render_maec(doc: rd.ResultDocument, ostream: StringIO):
+def render_maec(doc: rd.ResultDocument, console: Console):
     """
     example::
 
@@ -193,20 +240,37 @@ def render_maec(doc: rd.ResultDocument, ostream: StringIO):
     for category in sorted(maec_categories):
         values = maec_table.get(category, set())
         if values:
-            rows.append((rutils.bold(category.replace("_", "-")), "\n".join(sorted(values))))
+            rows.append((bold_markup(category.replace("_", "-")), "\n".join(sorted(values))))
 
     if rows:
-        ostream.write(
+        table = rich.table.Table(min_width=100)
-            tabulate.tabulate(
+        table.add_column(width("MAEC Category", 20))
-                rows,
+        table.add_column("MAEC Value")
-                headers=[width("MAEC Category", 25), width("MAEC Value", 75)],
+
-                tablefmt="mixed_grid",
+        for row in rows:
-            )
+            table.add_row(*row)
-        )
+
-        ostream.write("\n")
+        console.print(table)
 
 
-def render_mbc(doc: rd.ResultDocument, ostream: StringIO):
+def render_mbc_link(id: str, objective: str, behavior: str) -> str:
+    if id[0] in {"B", "T", "E", "F"}:
+        # behavior
+        base_url = "https://github.com/MBCProject/mbc-markdown/blob/main"
+    elif id[0] == "C":
+        # micro-behavior
+        base_url = "https://github.com/MBCProject/mbc-markdown/blob/main/micro-behaviors"
+    else:
+        raise ValueError("unexpected MBC prefix")
+
+    objective_fragment = objective.lower().replace(" ", "-")
+    behavior_fragment = behavior.lower().replace(" ", "-")
+
+    url = f"{base_url}/{objective_fragment}/{behavior_fragment}.md"
+    return rf"\[{link_markup(id, url)}]"
+
+
+def render_mbc(doc: rd.ResultDocument, console: Console):
     """
     example::
 
@@ -223,48 +287,48 @@ def render_mbc(doc: rd.ResultDocument, ostream: StringIO):
     objectives = collections.defaultdict(set)
     for rule in rutils.capability_rules(doc):
         for mbc in rule.meta.mbc:
-            objectives[mbc.objective].add((mbc.behavior, mbc.method, mbc.id))
+            objectives[mbc.objective].add((mbc.behavior, mbc.method, mbc.id.strip("[").strip("]")))
 
     rows = []
     for objective, behaviors in sorted(objectives.items()):
         inner_rows = []
-        for behavior, method, id in sorted(behaviors):
+        for technique, subtechnique, id in sorted(behaviors):
-            if not method:
+            if not subtechnique:
-                inner_rows.append(f"{rutils.bold(behavior)} [{id}]")
+                # example: File and Directory Discovery [T1083]
+                inner_rows.append(f"{bold_markup(technique)} {render_mbc_link(id, objective, technique)}")
             else:
-                inner_rows.append(f"{rutils.bold(behavior)}::{method} [{id}]")
+                # example: Code Discovery::Enumerate PE Sections [T1084.001]
-        rows.append(
+                inner_rows.append(
-            (
+                    f"{bold_markup(technique)}::{subtechnique} {render_mbc_link(id, objective, technique)}"
-                rutils.bold(objective.upper()),
+                )
-                "\n".join(inner_rows),
+
-            )
+        objective = bold_markup(objective.upper())
-        )
+        technique = "\n".join(inner_rows)
+
+        rows.append((objective, technique))
 
     if rows:
-        ostream.write(
+        table = rich.table.Table(min_width=100)
-            tabulate.tabulate(
+        table.add_column(width("MBC Objective", 20))
-                rows,
+        table.add_column("MBC Behavior")
-                headers=[width("MBC Objective", 25), width("MBC Behavior", 75)],
+
-                tablefmt="mixed_grid",
+        for row in rows:
-            )
+            table.add_row(*row)
-        )
+
-        ostream.write("\n")
+        console.print(table)
 
 
 def render_default(doc: rd.ResultDocument):
-    ostream = rutils.StringIO()
+    f = io.StringIO()
+    console = rich.console.Console()
 
-    render_meta(doc, ostream)
+    render_meta(doc, console)
-    ostream.write("\n")
+    render_attack(doc, console)
-    render_attack(doc, ostream)
+    render_maec(doc, console)
-    ostream.write("\n")
+    render_mbc(doc, console)
-    render_maec(doc, ostream)
+    render_capabilities(doc, console)
-    ostream.write("\n")
-    render_mbc(doc, ostream)
-    ostream.write("\n")
-    render_capabilities(doc, ostream)
 
-    return ostream.getvalue()
+    return f.getvalue()
 
 
 def render(meta, rules: RuleSet, capabilities: MatchResults) -> str:

capa/render/result_document.py

@@ -22,7 +22,7 @@ import capa.features.address
 import capa.features.freeze.features as frzf
 from capa.rules import RuleSet
 from capa.engine import MatchResults
-from capa.helpers import assert_never
+from capa.helpers import assert_never, load_json_from_path
 
 
 class FrozenModel(BaseModel):
@@ -668,4 +668,5 @@ class ResultDocument(FrozenModel):
 
     @classmethod
     def from_file(cls, path: Path) -> "ResultDocument":
-        return cls.model_validate_json(path.read_text(encoding="utf-8"))
+        report = load_json_from_path(path)
+        return cls.model_validate(report)

capa/render/utils.py

@@ -9,28 +9,29 @@
 import io
 from typing import Dict, List, Tuple, Union, Iterator, Optional
 
-import termcolor
+import rich.console
+from rich.progress import Text
 
 import capa.render.result_document as rd
 
 
-def bold(s: str) -> str:
+def bold(s: str) -> Text:
     """draw attention to the given string"""
-    return termcolor.colored(s, "cyan")
+    return Text.from_markup(f"[cyan]{s}")
 
 
-def bold2(s: str) -> str:
+def bold2(s: str) -> Text:
     """draw attention to the given string, within a `bold` section"""
-    return termcolor.colored(s, "green")
+    return Text.from_markup(f"[green]{s}")
 
 
-def mute(s: str) -> str:
+def mute(s: str) -> Text:
     """draw attention away from the given string"""
-    return termcolor.colored(s, "dark_grey")
+    return Text.from_markup(f"[dim]{s}")
 
 
-def warn(s: str) -> str:
+def warn(s: str) -> Text:
-    return termcolor.colored(s, "yellow")
+    return Text.from_markup(f"[yellow]{s}")
 
 
 def format_parts_id(data: Union[rd.AttackSpec, rd.MBCSpec]):
@@ -85,3 +86,17 @@ class StringIO(io.StringIO):
     def writeln(self, s):
         self.write(s)
         self.write("\n")
+
+
+class Console(rich.console.Console):
+    def writeln(self, *args, **kwargs) -> None:
+        """
+        prints the text with a new line at the end.
+        """
+        return self.print(*args, **kwargs)
+
+    def write(self, *args, **kwargs) -> None:
+        """
+        prints the text without a new line at the end.
+        """
+        return self.print(*args, **kwargs, end="")

capa/render/verbose.py

@@ -25,7 +25,8 @@ See the License for the specific language governing permissions and limitations
 
 from typing import cast
 
-import tabulate
+from rich.text import Text
+from rich.table import Table
 
 import capa.rules
 import capa.helpers
@@ -34,6 +35,7 @@ import capa.features.freeze as frz
 import capa.render.result_document as rd
 from capa.rules import RuleSet
 from capa.engine import MatchResults
+from capa.render.utils import Console
 
 
 def format_address(address: frz.Address) -> str:
@@ -140,7 +142,7 @@ def render_call(layout: rd.DynamicLayout, addr: frz.Address) -> str:
     )
 
 
-def render_static_meta(ostream, meta: rd.StaticMetadata):
+def render_static_meta(console: Console, meta: rd.StaticMetadata):
     """
     like:
 
@@ -161,12 +163,16 @@ def render_static_meta(ostream, meta: rd.StaticMetadata):
         total feature count  1918
     """
 
+    grid = Table.grid(padding=(0, 2))
+    grid.add_column(style="dim")
+    grid.add_column()
+
     rows = [
         ("md5", meta.sample.md5),
         ("sha1", meta.sample.sha1),
         ("sha256", meta.sample.sha256),
         ("path", meta.sample.path),
-        ("timestamp", meta.timestamp),
+        ("timestamp", str(meta.timestamp)),
         ("capa version", meta.version),
         ("os", meta.analysis.os),
         ("format", meta.analysis.format),
@@ -175,18 +181,21 @@ def render_static_meta(ostream, meta: rd.StaticMetadata):
         ("extractor", meta.analysis.extractor),
         ("base address", format_address(meta.analysis.base_address)),
         ("rules", "\n".join(meta.analysis.rules)),
-        ("function count", len(meta.analysis.feature_counts.functions)),
+        ("function count", str(len(meta.analysis.feature_counts.functions))),
-        ("library function count", len(meta.analysis.library_functions)),
+        ("library function count", str(len(meta.analysis.library_functions))),
         (
             "total feature count",
-            meta.analysis.feature_counts.file + sum(f.count for f in meta.analysis.feature_counts.functions),
+            str(meta.analysis.feature_counts.file + sum(f.count for f in meta.analysis.feature_counts.functions)),
         ),
     ]
 
-    ostream.writeln(tabulate.tabulate(rows, tablefmt="plain"))
+    for row in rows:
+        grid.add_row(*row)
+
+    console.print(grid)
 
 
-def render_dynamic_meta(ostream, meta: rd.DynamicMetadata):
+def render_dynamic_meta(console: Console, meta: rd.DynamicMetadata):
     """
     like:
 
@@ -205,12 +214,16 @@ def render_dynamic_meta(ostream, meta: rd.DynamicMetadata):
         total feature count  1918
     """
 
+    table = Table.grid(padding=(0, 2))
+    table.add_column(style="dim")
+    table.add_column()
+
     rows = [
         ("md5", meta.sample.md5),
         ("sha1", meta.sample.sha1),
         ("sha256", meta.sample.sha256),
         ("path", meta.sample.path),
-        ("timestamp", meta.timestamp),
+        ("timestamp", str(meta.timestamp)),
         ("capa version", meta.version),
         ("os", meta.analysis.os),
         ("format", meta.analysis.format),
@@ -218,26 +231,29 @@ def render_dynamic_meta(ostream, meta: rd.DynamicMetadata):
         ("analysis", meta.flavor.value),
         ("extractor", meta.analysis.extractor),
         ("rules", "\n".join(meta.analysis.rules)),
-        ("process count", len(meta.analysis.feature_counts.processes)),
+        ("process count", str(len(meta.analysis.feature_counts.processes))),
         (
             "total feature count",
-            meta.analysis.feature_counts.file + sum(p.count for p in meta.analysis.feature_counts.processes),
+            str(meta.analysis.feature_counts.file + sum(p.count for p in meta.analysis.feature_counts.processes)),
         ),
     ]
 
-    ostream.writeln(tabulate.tabulate(rows, tablefmt="plain"))
+    for row in rows:
+        table.add_row(*row)
+
+    console.print(table)
 
 
-def render_meta(osstream, doc: rd.ResultDocument):
+def render_meta(console: Console, doc: rd.ResultDocument):
     if doc.meta.flavor == rd.Flavor.STATIC:
-        render_static_meta(osstream, cast(rd.StaticMetadata, doc.meta))
+        render_static_meta(console, cast(rd.StaticMetadata, doc.meta))
     elif doc.meta.flavor == rd.Flavor.DYNAMIC:
-        render_dynamic_meta(osstream, cast(rd.DynamicMetadata, doc.meta))
+        render_dynamic_meta(console, cast(rd.DynamicMetadata, doc.meta))
     else:
         raise ValueError("invalid meta analysis")
 
 
-def render_rules(ostream, doc: rd.ResultDocument):
+def render_rules(console: Console, doc: rd.ResultDocument):
     """
     like:
 
@@ -254,11 +270,15 @@ def render_rules(ostream, doc: rd.ResultDocument):
         if count == 1:
             capability = rutils.bold(rule.meta.name)
         else:
-            capability = f"{rutils.bold(rule.meta.name)} ({count} matches)"
+            capability = Text.assemble(rutils.bold(rule.meta.name), f" ({count} matches)")
 
-        ostream.writeln(capability)
+        console.print(capability)
         had_match = True
 
+        table = Table.grid(padding=(0, 2))
+        table.add_column(style="dim")
+        table.add_column()
+
         rows = []
 
         ns = rule.meta.namespace
@@ -310,23 +330,26 @@ def render_rules(ostream, doc: rd.ResultDocument):
 
             rows.append(("matches", "\n".join(lines)))
 
-        ostream.writeln(tabulate.tabulate(rows, tablefmt="plain"))
+        for row in rows:
-        ostream.write("\n")
+            table.add_row(*row)
+
+        console.print(table)
+        console.print()
 
     if not had_match:
-        ostream.writeln(rutils.bold("no capabilities found"))
+        console.print(rutils.bold("no capabilities found"))
 
 
 def render_verbose(doc: rd.ResultDocument):
-    ostream = rutils.StringIO()
+    console = Console(highlight=False)
 
-    render_meta(ostream, doc)
+    with console.capture() as capture:
-    ostream.write("\n")
+        render_meta(console, doc)
+        console.print()
+        render_rules(console, doc)
+        console.print()
 
-    render_rules(ostream, doc)
+    return capture.get()
-    ostream.write("\n")
-
-    return ostream.getvalue()
 
 
 def render(meta, rules: RuleSet, capabilities: MatchResults) -> str:

capa/render/vverbose.py

@@ -9,7 +9,8 @@ import logging
 import textwrap
 from typing import Dict, Iterable, Optional
 
-import tabulate
+from rich.text import Text
+from rich.table import Table
 
 import capa.rules
 import capa.helpers
@@ -22,6 +23,7 @@ import capa.render.result_document as rd
 import capa.features.freeze.features as frzf
 from capa.rules import RuleSet
 from capa.engine import MatchResults
+from capa.render.utils import Console
 
 logger = logging.getLogger(__name__)
 
@@ -45,7 +47,7 @@ def hanging_indent(s: str, indent: int) -> str:
     return textwrap.indent(s, prefix=prefix)[len(prefix) :]
 
 
-def render_locations(ostream, layout: rd.Layout, locations: Iterable[frz.Address], indent: int):
+def render_locations(console: Console, layout: rd.Layout, locations: Iterable[frz.Address], indent: int):
     import capa.render.verbose as v
 
     # it's possible to have an empty locations array here,
@@ -56,7 +58,7 @@ def render_locations(ostream, layout: rd.Layout, locations: Iterable[frz.Address
     if len(locations) == 0:
         return
 
-    ostream.write(" @ ")
+    console.write(" @ ")
     location0 = locations[0]
 
     if len(locations) == 1:
@@ -64,58 +66,58 @@ def render_locations(ostream, layout: rd.Layout, locations: Iterable[frz.Address
 
         if location.type == frz.AddressType.CALL:
             assert isinstance(layout, rd.DynamicLayout)
-            ostream.write(hanging_indent(v.render_call(layout, location), indent + 1))
+            console.write(hanging_indent(v.render_call(layout, location), indent + 1))
         else:
-            ostream.write(v.format_address(locations[0]))
+            console.write(v.format_address(locations[0]))
 
     elif location0.type == frz.AddressType.CALL and len(locations) > 1:
         location = locations[0]
 
         assert isinstance(layout, rd.DynamicLayout)
         s = f"{v.render_call(layout, location)}\nand {(len(locations) - 1)} more..."
-        ostream.write(hanging_indent(s, indent + 1))
+        console.write(hanging_indent(s, indent + 1))
 
     elif len(locations) > 4:
         # don't display too many locations, because it becomes very noisy.
         # probably only the first handful of locations will be useful for inspection.
-        ostream.write(", ".join(map(v.format_address, locations[0:4])))
+        console.write(", ".join(map(v.format_address, locations[0:4])))
-        ostream.write(f", and {(len(locations) - 4)} more...")
+        console.write(f", and {(len(locations) - 4)} more...")
 
     elif len(locations) > 1:
-        ostream.write(", ".join(map(v.format_address, locations)))
+        console.write(", ".join(map(v.format_address, locations)))
 
     else:
         raise RuntimeError("unreachable")
 
 
-def render_statement(ostream, layout: rd.Layout, match: rd.Match, statement: rd.Statement, indent: int):
+def render_statement(console: Console, layout: rd.Layout, match: rd.Match, statement: rd.Statement, indent: int):
-    ostream.write("  " * indent)
+    console.write("  " * indent)
 
     if isinstance(statement, rd.SubscopeStatement):
         # emit `basic block:`
         # rather than `subscope:`
-        ostream.write(statement.scope)
+        console.write(statement.scope)
 
-        ostream.write(":")
+        console.write(":")
         if statement.description:
-            ostream.write(f" = {statement.description}")
+            console.write(f" = {statement.description}")
-        ostream.writeln("")
+        console.writeln()
 
     elif isinstance(statement, (rd.CompoundStatement)):
         # emit `and:`  `or:`  `optional:`  `not:`
-        ostream.write(statement.type)
+        console.write(statement.type)
 
-        ostream.write(":")
+        console.write(":")
         if statement.description:
-            ostream.write(f" = {statement.description}")
+            console.write(f" = {statement.description}")
-        ostream.writeln("")
+        console.writeln()
 
     elif isinstance(statement, rd.SomeStatement):
-        ostream.write(f"{statement.count} or more:")
+        console.write(f"{statement.count} or more:")
 
         if statement.description:
-            ostream.write(f" = {statement.description}")
+            console.write(f" = {statement.description}")
-        ostream.writeln("")
+        console.writeln()
 
     elif isinstance(statement, rd.RangeStatement):
         # `range` is a weird node, its almost a hybrid of statement+feature.
@@ -133,25 +135,25 @@ def render_statement(ostream, layout: rd.Layout, match: rd.Match, statement: rd.
             value = rutils.bold2(value)
 
             if child.description:
-                ostream.write(f"count({child.type}({value} = {child.description})): ")
+                console.write(f"count({child.type}({value} = {child.description})): ")
             else:
-                ostream.write(f"count({child.type}({value})): ")
+                console.write(f"count({child.type}({value})): ")
         else:
-            ostream.write(f"count({child.type}): ")
+            console.write(f"count({child.type}): ")
 
         if statement.max == statement.min:
-            ostream.write(f"{statement.min}")
+            console.write(f"{statement.min}")
         elif statement.min == 0:
-            ostream.write(f"{statement.max} or fewer")
+            console.write(f"{statement.max} or fewer")
         elif statement.max == (1 << 64 - 1):
-            ostream.write(f"{statement.min} or more")
+            console.write(f"{statement.min} or more")
         else:
-            ostream.write(f"between {statement.min} and {statement.max}")
+            console.write(f"between {statement.min} and {statement.max}")
 
         if statement.description:
-            ostream.write(f" = {statement.description}")
+            console.write(f" = {statement.description}")
-        render_locations(ostream, layout, match.locations, indent)
+        render_locations(console, layout, match.locations, indent)
-        ostream.writeln("")
+        console.writeln()
 
     else:
         raise RuntimeError("unexpected match statement type: " + str(statement))
@@ -162,9 +164,9 @@ def render_string_value(s: str) -> str:
 
 
 def render_feature(
-    ostream, layout: rd.Layout, rule: rd.RuleMatches, match: rd.Match, feature: frzf.Feature, indent: int
+    console: Console, layout: rd.Layout, rule: rd.RuleMatches, match: rd.Match, feature: frzf.Feature, indent: int
 ):
-    ostream.write("  " * indent)
+    console.write("  " * indent)
 
     key = feature.type
     value: Optional[str]
@@ -205,14 +207,14 @@ def render_feature(
         elif isinstance(feature, frzf.OperandOffsetFeature):
             key = f"operand[{feature.index}].offset"
 
-        ostream.write(f"{key}: ")
+        console.write(f"{key}: ")
 
         if value:
-            ostream.write(rutils.bold2(value))
+            console.write(rutils.bold2(value))
 
             if feature.description:
-                ostream.write(capa.rules.DESCRIPTION_SEPARATOR)
+                console.write(capa.rules.DESCRIPTION_SEPARATOR)
-                ostream.write(feature.description)
+                console.write(feature.description)
 
         if isinstance(feature, (frzf.OSFeature, frzf.ArchFeature, frzf.FormatFeature)):
             # don't show the location of these global features
@@ -224,35 +226,32 @@ def render_feature(
         elif isinstance(feature, (frzf.OSFeature, frzf.ArchFeature, frzf.FormatFeature)):
             pass
         else:
-            render_locations(ostream, layout, match.locations, indent)
+            render_locations(console, layout, match.locations, indent)
-        ostream.write("\n")
+        console.writeln()
     else:
         # like:
         #  regex: /blah/ = SOME_CONSTANT
         #    - "foo blah baz" @ 0x401000
         #    - "aaa blah bbb" @ 0x402000, 0x403400
-        ostream.write(key)
+        console.writeln(f"{key}: {value}")
-        ostream.write(": ")
-        ostream.write(value)
-        ostream.write("\n")
 
         for capture, locations in sorted(match.captures.items()):
-            ostream.write("  " * (indent + 1))
+            console.write("  " * (indent + 1))
-            ostream.write("- ")
+            console.write("- ")
-            ostream.write(rutils.bold2(render_string_value(capture)))
+            console.write(rutils.bold2(render_string_value(capture)))
             if isinstance(layout, rd.DynamicLayout) and rule.meta.scopes.dynamic == capa.rules.Scope.CALL:
                 # like above, don't re-render calls when in call scope.
                 pass
             else:
-                render_locations(ostream, layout, locations, indent=indent)
+                render_locations(console, layout, locations, indent=indent)
-            ostream.write("\n")
+            console.writeln()
 
 
-def render_node(ostream, layout: rd.Layout, rule: rd.RuleMatches, match: rd.Match, node: rd.Node, indent: int):
+def render_node(console: Console, layout: rd.Layout, rule: rd.RuleMatches, match: rd.Match, node: rd.Node, indent: int):
     if isinstance(node, rd.StatementNode):
-        render_statement(ostream, layout, match, node.statement, indent=indent)
+        render_statement(console, layout, match, node.statement, indent=indent)
     elif isinstance(node, rd.FeatureNode):
-        render_feature(ostream, layout, rule, match, node.feature, indent=indent)
+        render_feature(console, layout, rule, match, node.feature, indent=indent)
     else:
         raise RuntimeError("unexpected node type: " + str(node))
 
@@ -265,7 +264,9 @@ MODE_SUCCESS = "success"
 MODE_FAILURE = "failure"
 
 
-def render_match(ostream, layout: rd.Layout, rule: rd.RuleMatches, match: rd.Match, indent=0, mode=MODE_SUCCESS):
+def render_match(
+    console: Console, layout: rd.Layout, rule: rd.RuleMatches, match: rd.Match, indent=0, mode=MODE_SUCCESS
+):
     child_mode = mode
     if mode == MODE_SUCCESS:
         # display only nodes that evaluated successfully.
@@ -297,13 +298,13 @@ def render_match(ostream, layout: rd.Layout, rule: rd.RuleMatches, match: rd.Mat
     else:
         raise RuntimeError("unexpected mode: " + mode)
 
-    render_node(ostream, layout, rule, match, match.node, indent=indent)
+    render_node(console, layout, rule, match, match.node, indent=indent)
 
     for child in match.children:
-        render_match(ostream, layout, rule, child, indent=indent + 1, mode=child_mode)
+        render_match(console, layout, rule, child, indent=indent + 1, mode=child_mode)
 
 
-def render_rules(ostream, doc: rd.ResultDocument):
+def render_rules(console: Console, doc: rd.ResultDocument):
     """
     like:
 
@@ -350,13 +351,13 @@ def render_rules(ostream, doc: rd.ResultDocument):
         if count == 1:
             if rule.meta.lib:
                 lib_info = " (library rule)"
-            capability = f"{rutils.bold(rule.meta.name)}{lib_info}"
+            capability = Text.assemble(rutils.bold(rule.meta.name), f"{lib_info}")
         else:
             if rule.meta.lib:
                 lib_info = ", only showing first match of library rule"
-            capability = f"{rutils.bold(rule.meta.name)} ({count} matches{lib_info})"
+            capability = Text.assemble(rutils.bold(rule.meta.name), f" ({count} matches{lib_info})")
 
-        ostream.writeln(capability)
+        console.writeln(capability)
         had_match = True
 
         rows = []
@@ -402,7 +403,14 @@ def render_rules(ostream, doc: rd.ResultDocument):
         if rule.meta.description:
             rows.append(("description", rule.meta.description))
 
-        ostream.writeln(tabulate.tabulate(rows, tablefmt="plain"))
+        grid = Table.grid(padding=(0, 2))
+        grid.add_column(style="dim")
+        grid.add_column()
+
+        for row in rows:
+            grid.add_row(*row)
+
+        console.writeln(grid)
 
         if capa.rules.Scope.FILE in rule.meta.scopes:
             matches = doc.rules[rule.meta.name].matches
@@ -413,61 +421,58 @@ def render_rules(ostream, doc: rd.ResultDocument):
                 # so, lets be explicit about our assumptions and raise an exception if they fail.
                 raise RuntimeError(f"unexpected file scope match count: {len(matches)}")
             _, first_match = matches[0]
-            render_match(ostream, doc.meta.analysis.layout, rule, first_match, indent=0)
+            render_match(console, doc.meta.analysis.layout, rule, first_match, indent=0)
         else:
             for location, match in sorted(doc.rules[rule.meta.name].matches):
                 if doc.meta.flavor == rd.Flavor.STATIC:
                     assert rule.meta.scopes.static is not None
-                    ostream.write(rule.meta.scopes.static.value)
+                    console.write(rule.meta.scopes.static.value + " @ ")
-                    ostream.write(" @ ")
+                    console.write(capa.render.verbose.format_address(location))
-                    ostream.write(capa.render.verbose.format_address(location))
 
                     if rule.meta.scopes.static == capa.rules.Scope.BASIC_BLOCK:
                         func = frz.Address.from_capa(functions_by_bb[location.to_capa()])
-                        ostream.write(f" in function {capa.render.verbose.format_address(func)}")
+                        console.write(f" in function {capa.render.verbose.format_address(func)}")
 
                 elif doc.meta.flavor == rd.Flavor.DYNAMIC:
                     assert rule.meta.scopes.dynamic is not None
                     assert isinstance(doc.meta.analysis.layout, rd.DynamicLayout)
 
-                    ostream.write(rule.meta.scopes.dynamic.value)
+                    console.write(rule.meta.scopes.dynamic.value + " @ ")
-
-                    ostream.write(" @ ")
 
                     if rule.meta.scopes.dynamic == capa.rules.Scope.PROCESS:
-                        ostream.write(v.render_process(doc.meta.analysis.layout, location))
+                        console.write(v.render_process(doc.meta.analysis.layout, location))
                     elif rule.meta.scopes.dynamic == capa.rules.Scope.THREAD:
-                        ostream.write(v.render_thread(doc.meta.analysis.layout, location))
+                        console.write(v.render_thread(doc.meta.analysis.layout, location))
                     elif rule.meta.scopes.dynamic == capa.rules.Scope.CALL:
-                        ostream.write(hanging_indent(v.render_call(doc.meta.analysis.layout, location), indent=1))
+                        console.write(hanging_indent(v.render_call(doc.meta.analysis.layout, location), indent=1))
                     else:
                         capa.helpers.assert_never(rule.meta.scopes.dynamic)
 
                 else:
                     capa.helpers.assert_never(doc.meta.flavor)
 
-                ostream.write("\n")
+                console.writeln()
-                render_match(ostream, doc.meta.analysis.layout, rule, match, indent=1)
+                render_match(console, doc.meta.analysis.layout, rule, match, indent=1)
                 if rule.meta.lib:
                     # only show first match
                     break
 
-        ostream.write("\n")
+        console.writeln()
 
     if not had_match:
-        ostream.writeln(rutils.bold("no capabilities found"))
+        console.writeln(rutils.bold("no capabilities found"))
 
 
 def render_vverbose(doc: rd.ResultDocument):
-    ostream = rutils.StringIO()
+    console = Console(highlight=False)
 
-    capa.render.verbose.render_meta(ostream, doc)
+    with console.capture() as capture:
-    ostream.write("\n")
+        capa.render.verbose.render_meta(console, doc)
+        console.writeln()
+        render_rules(console, doc)
+        console.writeln()
 
-    render_rules(ostream, doc)
+    return capture.get()
-    ostream.write("\n")
-
-    return ostream.getvalue()
 
 
 def render(meta, rules: RuleSet, capabilities: MatchResults) -> str:

capa/rules/__init__.py

@@ -575,6 +575,15 @@ def trim_dll_part(api: str) -> str:
     return api
 
 
+def unique(sequence):
+    """deduplicate the items in the given sequence, returning a list with the same order.
+
+    via: https://stackoverflow.com/a/58666031
+    """
+    seen = set()
+    return [x for x in sequence if not (x in seen or seen.add(x))]  # type: ignore [func-returns-value]
+
+
 def build_statements(d, scopes: Scopes):
     if len(d.keys()) > 2:
         raise InvalidRule("too many statements")
@@ -582,21 +591,21 @@ def build_statements(d, scopes: Scopes):
     key = list(d.keys())[0]
     description = pop_statement_description_entry(d[key])
     if key == "and":
-        return ceng.And([build_statements(dd, scopes) for dd in d[key]], description=description)
+        return ceng.And(unique(build_statements(dd, scopes) for dd in d[key]), description=description)
     elif key == "or":
-        return ceng.Or([build_statements(dd, scopes) for dd in d[key]], description=description)
+        return ceng.Or(unique(build_statements(dd, scopes) for dd in d[key]), description=description)
     elif key == "not":
         if len(d[key]) != 1:
             raise InvalidRule("not statement must have exactly one child statement")
         return ceng.Not(build_statements(d[key][0], scopes), description=description)
     elif key.endswith(" or more"):
         count = int(key[: -len("or more")])
-        return ceng.Some(count, [build_statements(dd, scopes) for dd in d[key]], description=description)
+        return ceng.Some(count, unique(build_statements(dd, scopes) for dd in d[key]), description=description)
     elif key == "optional":
         # `optional` is an alias for `0 or more`
         # which is useful for documenting behaviors,
         # like with `write file`, we might say that `WriteFile` is optionally found alongside `CreateFileA`.
-        return ceng.Some(0, [build_statements(dd, scopes) for dd in d[key]], description=description)
+        return ceng.Some(0, unique(build_statements(dd, scopes) for dd in d[key]), description=description)
 
     elif key == "process":
         if Scope.FILE not in scopes:
@@ -672,7 +681,7 @@ def build_statements(d, scopes: Scopes):
             #       - arch: i386
             #       - mnemonic: cmp
             #
-            statements = ceng.And([build_statements(dd, Scopes(static=Scope.INSTRUCTION)) for dd in d[key]])
+            statements = ceng.And(unique(build_statements(dd, Scopes(static=Scope.INSTRUCTION)) for dd in d[key]))
 
         return ceng.Subscope(Scope.INSTRUCTION, statements, description=description)
 
@@ -2121,12 +2130,14 @@ def get_rules(
     rule_paths: List[RulePath],
     cache_dir=None,
     on_load_rule: Callable[[RulePath, int, int], None] = on_load_rule_default,
+    enable_cache: bool = True,
 ) -> RuleSet:
     """
     args:
       rule_paths: list of paths to rules files or directories containing rules files
       cache_dir: directory to use for caching rules, or will use the default detected cache directory if None
       on_load_rule: callback to invoke before a rule is loaded, use for progress or cancellation
+      enable_cache: enable loading of a cached ruleset (default: True)
     """
     if cache_dir is None:
         cache_dir = capa.rules.cache.get_default_cache_directory()
@@ -2138,9 +2149,10 @@ def get_rules(
     # rule_file_paths[i] corresponds to rule_contents[i].
     rule_contents = [file_path.read_bytes() for file_path in rule_file_paths]
 
-    ruleset = capa.rules.cache.load_cached_ruleset(cache_dir, rule_contents)
+    if enable_cache:
-    if ruleset is not None:
+        ruleset = capa.rules.cache.load_cached_ruleset(cache_dir, rule_contents)
-        return ruleset
+        if ruleset is not None:
+            return ruleset
 
     rules: List[Rule] = []
 

capa/version.py

@@ -5,7 +5,7 @@
 # Unless required by applicable law or agreed to in writing, software distributed under the License
 #  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and limitations under the License.
-__version__ = "7.2.0"
+__version__ = "7.4.0"
 
 
 def get_major_version():

doc/release.md

@@ -26,7 +26,9 @@
 
     ### Bug Fixes
 
-    ### capa explorer IDA Pro plugin
+    ### capa Explorer Web
+
+    ### capa Explorer IDA Pro plugin
 
     ### Development
 
@@ -42,5 +44,6 @@
   - [ ] [publish to PyPI](https://pypi.org/project/flare-capa)
   - [ ] [create tag in capa rules](https://github.com/mandiant/capa-rules/tags)
   - [ ] [create release in capa rules](https://github.com/mandiant/capa-rules/releases)
+- [ ] Update [homepage](https://github.com/mandiant/capa/blob/master/web/public/index.html)
 - [ ] [Spread the word](https://twitter.com)
 - [ ] Update internal service

pyproject.toml

@@ -20,7 +20,7 @@ authors = [
 description = "The FLARE team's open-source tool to identify capabilities in executable files."
 readme = {file = "README.md", content-type = "text/markdown"}
 license = {file = "LICENSE.txt"}
-requires-python = ">=3.8"
+requires-python = ">=3.8.1"
 keywords = ["malware analysis", "reverse engineering", "capability detection", "software behaviors", "capa", "FLARE"]
 classifiers = [
     "Development Status :: 5 - Production/Stable",
@@ -65,12 +65,8 @@ dependencies = [
     # or minor otherwise).
     # As specific constraints are identified, please provide
     # comments and context.
-    "tqdm>=4",
     "pyyaml>=6",
-    "tabulate>=0.9",
     "colorama>=0.4",
-    "termcolor>=2",
-    "wcwidth>=0.2",
     "ida-settings>=2",
     "ruamel.yaml>=0.18",
     "pefile>=2023.2.7",
@@ -80,6 +76,9 @@ dependencies = [
     "humanize>=4",
     "protobuf>=5",
     "msgspec>=0.18.6",
+    "xmltodict>=0.13.0",
+    # for library detection (in development)
+    "nltk>=3",
 
     # ---------------------------------------
     # Dependencies that we develop
@@ -124,8 +123,8 @@ dev = [
     "pytest-sugar==1.0.0",
     "pytest-instafail==0.5.0",
     "pytest-cov==5.0.0",
-    "flake8==7.1.0",
+    "flake8==7.1.1",
-    "flake8-bugbear==24.4.26",
+    "flake8-bugbear==24.8.19",
     "flake8-encodings==0.5.1",
     "flake8-comprehensions==3.15.0",
     "flake8-logging-format==0.9.0",
@@ -135,22 +134,20 @@ dev = [
     "flake8-simplify==0.21.0",
     "flake8-use-pathlib==0.3.0",
     "flake8-copyright==0.2.4",
-    "ruff==0.5.6",
+    "ruff==0.6.4",
     "black==24.8.0",
     "isort==5.13.2",
-    "mypy==1.11.1",
+    "mypy==1.11.2",
     "mypy-protobuf==3.6.0",
-    "PyGithub==2.3.0",
+    "PyGithub==2.4.0",
     # type stubs for mypy
     "types-backports==0.1.3",
     "types-colorama==0.4.15.11",
     "types-PyYAML==6.0.8",
-    "types-tabulate==0.9.0.20240106",
+    "types-psutil==6.0.0.20240901",
-    "types-termcolor==1.1.4",
-    "types-psutil==6.0.0.20240621",
     "types_requests==2.32.0.20240712",
-    "types-protobuf==5.27.0.20240626",
+    "types-protobuf==5.28.0.20240924",
-    "deptry==0.17.0"
+    "deptry==0.20.0"
 ]
 build = [
     # Dev and build dependencies are not relaxed because
@@ -158,8 +155,8 @@ build = [
     # These dependencies are not used in production environments
     # and should not conflict with other libraries/tooling.
     "pyinstaller==6.10.0",
-    "setuptools==70.0.0",
+    "setuptools==75.1.0",
-    "build==1.2.1"
+    "build==1.2.2"
 ]
 scripts = [
     "jschema_to_python==1.2.3",
@@ -172,7 +169,8 @@ scripts = [
 [tool.deptry]
 extend_exclude = [
     "sigs",
-    "tests"
+    "tests",
+    "web",
 ]
 
 # dependencies marked as first party, to inform deptry that they are local
@@ -181,6 +179,9 @@ known_first_party = [
     "binaryninja",
     "flirt",
     "ghidra",
+    "idapro",
+    "ida_ida",
+    "ida_auto",
     "ida_bytes",
     "ida_entry",
     "ida_funcs",
@@ -231,10 +232,7 @@ DEP002 = [
     "types-protobuf",
     "types-psutil",
     "types-PyYAML",
-    "types-tabulate",
-    "types-termcolor",
     "types_requests",
-    "wcwidth"
 ]
 
 # dependencies imported but missing from definitions

requirements.txt

@@ -8,11 +8,11 @@
 # Kept up to date by dependabot.
 annotated-types==0.7.0
 colorama==0.4.6
-cxxfilt==0.2.2
+cxxfilt==0.3.0
 dncil==1.0.2
 dnfile==0.15.0
 funcy==2.0
-humanize==4.9.0
+humanize==4.10.0
 ida-netnode==3.0
 ida-settings==2.1.0
 intervaltree==3.1.0
@@ -20,27 +20,28 @@ markdown-it-py==3.0.0
 mdurl==0.1.2
 msgpack==1.0.8
 networkx==3.1
-pefile==2023.2.7
+pefile==2024.8.26
 pip==24.2
-protobuf==5.27.3
+protobuf==5.28.2
-pyasn1==0.4.8
+pyasn1==0.5.1
-pyasn1-modules==0.2.8
+pyasn1-modules==0.3.0
 pycparser==2.22
-pydantic==2.7.3
+pydantic==2.9.2
-pydantic-core==2.18.4
+# pydantic pins pydantic-core, 
+# but dependabot updates these separately (which is broken) and is annoying,
+# so we rely on pydantic to pull in the right version of pydantic-core.
+# pydantic-core==2.23.4
+xmltodict==0.13.0
 pyelftools==0.31
 pygments==2.18.0
 python-flirt==0.8.10
-pyyaml==6.0.1
+pyyaml==6.0.2
-rich==13.7.1
+rich==13.9.2
 ruamel-yaml==0.18.6
 ruamel-yaml-clib==0.2.8
-setuptools==70.0.0
+setuptools==75.1.0
 six==1.16.0
 sortedcontainers==2.4.0
-tabulate==0.9.0
-termcolor==2.4.0
-tqdm==4.66.5
 viv-utils==0.7.11
-vivisect==1.1.1
+vivisect==1.2.1
-wcwidth==0.2.13
+msgspec==0.18.6

scripts/capa2sarif.py

@@ -84,8 +84,7 @@ def main() -> int:
     args = _parse_args()
 
     try:
-        with Path(args.capa_output).open() as capa_output:
+        json_data = json.loads(Path(args.capa_output).read_text(encoding="utf-8"))
-            json_data = json.load(capa_output)
     except ValueError:
         logger.error("Input data was not valid JSON, input should be a capa json output file.")
         return -1

scripts/compare-backends.py

@@ -0,0 +1,316 @@
+# Copyright (C) 2024 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+
+import sys
+import json
+import time
+import logging
+import argparse
+import contextlib
+import statistics
+import subprocess
+import multiprocessing
+from typing import Set, Dict, List, Optional
+from pathlib import Path
+from collections import Counter
+from dataclasses import dataclass
+from multiprocessing import Pool
+
+import rich
+import rich.box
+import rich.table
+
+import capa.main
+
+logger = logging.getLogger("capa.compare-backends")
+
+BACKENDS = ("vivisect", "ida", "binja")
+
+
+@dataclass
+class CapaInvocation:
+    path: Path
+    backend: str
+    duration: float
+    returncode: int
+    stdout: Optional[str]
+    stderr: Optional[str]
+    err: Optional[str]
+
+
+def invoke_capa(file: Path, backend: str) -> CapaInvocation:
+    stdout = None
+    stderr = None
+    err = None
+    returncode: int
+    try:
+        logger.debug("run capa: %s: %s", backend, file.name)
+        t1 = time.time()
+        child = subprocess.run(
+            ["python", "-m", "capa.main", "--json", "--backend=" + backend, str(file)],
+            capture_output=True,
+            check=True,
+            text=True,
+            encoding="utf-8",
+        )
+        returncode = child.returncode
+        stdout = child.stdout
+        stderr = child.stderr
+    except subprocess.CalledProcessError as e:
+        returncode = e.returncode
+        stdout = e.stdout
+        stderr = e.stderr
+
+        logger.debug("%s:%s: error", backend, file.name)
+        err = str(e)
+    else:
+        pass
+    finally:
+        t2 = time.time()
+
+    return CapaInvocation(
+        path=file,
+        backend=backend,
+        duration=t2 - t1,
+        returncode=returncode,
+        stdout=stdout,
+        stderr=stderr,
+        err=err,
+    )
+
+
+def wrapper_invoke_capa(args):
+    file, backend = args
+    return invoke_capa(file, backend)
+
+
+def collect(args):
+    results_path = args.results_path
+    if not results_path.is_file():
+        default_doc = {backend: {} for backend in BACKENDS}  # type: ignore
+        results_path.write_text(json.dumps(default_doc), encoding="utf-8")
+
+    testfiles = Path(__file__).parent.parent / "tests" / "data"
+
+    for file in sorted(p for p in testfiles.glob("*")):
+        # remove leftover analysis files
+        # because IDA doesn't cleanup after itself, currently.
+        if file.suffix in (".til", ".id0", ".id1", ".id2", ".nam", ".viv"):
+            logger.debug("removing: %s", file)
+            with contextlib.suppress(IOError):
+                file.unlink()
+
+    doc = json.loads(results_path.read_text(encoding="utf-8"))
+
+    plan = []
+    for file in sorted(p for p in testfiles.glob("*")):
+        if not file.is_file():
+            continue
+
+        if file.is_dir():
+            continue
+
+        if file.name.startswith("."):
+            continue
+
+        if file.suffix not in (".exe_", ".dll_", ".elf_", ""):
+            continue
+
+        logger.debug("%s", file.name)
+        key = str(file)
+
+        for backend in BACKENDS:
+
+            if (backend, file.name) in {
+                ("binja", "0953cc3b77ed2974b09e3a00708f88de931d681e2d0cb64afbaf714610beabe6.exe_")
+            }:
+                # this file takes 38GB+ and 20hrs+
+                # https://github.com/Vector35/binaryninja-api/issues/5951
+                continue
+
+            if key in doc[backend]:
+                if not args.retry_failures:
+                    continue
+
+                if not doc[backend][key]["err"]:
+                    # didn't previously fail, don't repeat work
+                    continue
+
+                else:
+                    # want to retry this previous failure
+                    pass
+
+            plan.append((file, backend))
+
+    pool_size = multiprocessing.cpu_count() // 2
+    logger.info("work pool size: %d", pool_size)
+    with Pool(processes=pool_size) as pool:
+        for i, result in enumerate(pool.imap_unordered(wrapper_invoke_capa, plan)):
+            doc[result.backend][str(result.path)] = {
+                "path": str(result.path),
+                "returncode": result.returncode,
+                "stdout": result.stdout,
+                "stderr": result.stderr,
+                "err": result.err,
+                "duration": result.duration,
+            }
+
+            if i % 8 == 0:
+                logger.info("syncing output database")
+                results_path.write_text(json.dumps(doc))
+
+            logger.info(
+                "%.1f\t%s %s %s",
+                result.duration,
+                "(err)" if result.err else "     ",
+                result.backend.ljust(8),
+                result.path.name,
+            )
+
+    results_path.write_text(json.dumps(doc))
+    return
+
+
+def report(args):
+    doc = json.loads(args.results_path.read_text(encoding="utf-8"))
+
+    samples = set()
+    for backend in BACKENDS:
+        samples.update(doc[backend].keys())
+
+    failures_by_backend: Dict[str, Set[str]] = {backend: set() for backend in BACKENDS}
+    durations_by_backend: Dict[str, List[float]] = {backend: [] for backend in BACKENDS}
+
+    console = rich.get_console()
+    for key in sorted(samples):
+        sample = Path(key).name
+        console.print(sample, style="bold")
+
+        seen_rules: Counter[str] = Counter()
+
+        rules_by_backend: Dict[str, Set[str]] = {backend: set() for backend in BACKENDS}
+
+        for backend in BACKENDS:
+            if key not in doc[backend]:
+                continue
+
+            entry = doc[backend][key]
+            duration = entry["duration"]
+
+            if not entry["err"]:
+                matches = json.loads(entry["stdout"])["rules"].keys()
+                seen_rules.update(matches)
+                rules_by_backend[backend].update(matches)
+                durations_by_backend[backend].append(duration)
+
+                console.print(f"  {backend: >8}: {duration: >6.1f}s   {len(matches): >3d} matches")
+
+            else:
+                failures_by_backend[backend].add(sample)
+                console.print(f"  {backend: >8}: {duration: >6.1f}s   (error)")
+
+        if not seen_rules:
+            console.print()
+            continue
+
+        t = rich.table.Table(box=rich.box.SIMPLE, header_style="default")
+        t.add_column("viv")
+        t.add_column("ida")
+        t.add_column("bn")
+        t.add_column("rule")
+
+        for rule, _ in seen_rules.most_common():
+            t.add_row(
+                "x" if rule in rules_by_backend["vivisect"] else " ",
+                "x" if rule in rules_by_backend["ida"] else " ",
+                "x" if rule in rules_by_backend["binja"] else " ",
+                rule,
+            )
+
+        console.print(t)
+
+    for backend in BACKENDS:
+        console.print(f"failures for {backend}:", style="bold")
+        for failure in sorted(failures_by_backend[backend]):
+            console.print(f"  - {failure}")
+
+        if not failures_by_backend[backend]:
+            console.print("  (none)", style="green")
+    console.print()
+
+    console.print("durations:", style="bold")
+    console.print("  (10-quantiles, in seconds)", style="grey37")
+    for backend in BACKENDS:
+        q = statistics.quantiles(durations_by_backend[backend], n=10)
+        console.print(f"  {backend: <8}: ", end="")
+        for i in range(9):
+            if i in (4, 8):
+                style = "bold"
+            else:
+                style = "default"
+            console.print(f"{q[i]: >6.1f}", style=style, end=" ")
+        console.print()
+    console.print("                ^-- 10% of samples took less than this                  ^", style="grey37")
+    console.print("                    10% of samples took more than this -----------------+", style="grey37")
+
+    console.print()
+    for backend in BACKENDS:
+        total = sum(durations_by_backend[backend])
+        successes = len(durations_by_backend[backend])
+        avg = statistics.mean(durations_by_backend[backend])
+        console.print(
+            f"  {backend: <8}: {total: >7.0f} seconds across {successes: >4d} successful runs, {avg: >4.1f} average"
+        )
+    console.print()
+
+    console.print("slowest samples:", style="bold")
+    for backend in BACKENDS:
+        console.print(backend)
+        for duration, path in sorted(
+            ((d["duration"], Path(d["path"]).name) for d in doc[backend].values()), reverse=True
+        )[:5]:
+            console.print(f"  - {duration: >6.1f} {path}")
+
+    return
+
+
+def main(argv=None):
+    if argv is None:
+        argv = sys.argv[1:]
+
+    default_samples_path = Path(__file__).resolve().parent.parent / "tests" / "data"
+
+    parser = argparse.ArgumentParser(description="Compare analysis backends.")
+    capa.main.install_common_args(
+        parser,
+        wanted=set(),
+    )
+
+    subparsers = parser.add_subparsers()
+    collect_parser = subparsers.add_parser("collect")
+    collect_parser.add_argument("results_path", type=Path, help="Path to output JSON file")
+    collect_parser.add_argument("--samples", type=Path, default=default_samples_path, help="Path to samples")
+    collect_parser.add_argument("--retry-failures", action="store_true", help="Retry previous failures")
+    collect_parser.set_defaults(func=collect)
+
+    report_parser = subparsers.add_parser("report")
+    report_parser.add_argument("results_path", type=Path, help="Path to JSON file")
+    report_parser.set_defaults(func=report)
+
+    args = parser.parse_args(args=argv)
+
+    try:
+        capa.main.handle_common_args(args)
+    except capa.main.ShouldExitError as e:
+        return e.status_code
+
+    args.func(args)
+
+
+if __name__ == "__main__":
+    sys.exit(main())

scripts/detect-backends.py

@@ -0,0 +1,106 @@
+# Copyright (C) 2024 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+
+import sys
+import logging
+import argparse
+import importlib.util
+
+import rich
+import rich.table
+
+import capa.main
+from capa.features.extractors.ida.idalib import find_idalib, load_idalib, is_idalib_installed
+from capa.features.extractors.binja.find_binja_api import find_binaryninja, load_binaryninja, is_binaryninja_installed
+
+logger = logging.getLogger(__name__)
+
+
+def is_vivisect_installed() -> bool:
+    try:
+        return importlib.util.find_spec("vivisect") is not None
+    except ModuleNotFoundError:
+        return False
+
+
+def load_vivisect() -> bool:
+    try:
+        import vivisect  # noqa: F401 unused import
+
+        return True
+    except ImportError:
+        return False
+
+
+def main(argv=None):
+    if argv is None:
+        argv = sys.argv[1:]
+
+    parser = argparse.ArgumentParser(description="Detect analysis backends.")
+    capa.main.install_common_args(parser, wanted=set())
+    args = parser.parse_args(args=argv)
+
+    try:
+        capa.main.handle_common_args(args)
+    except capa.main.ShouldExitError as e:
+        return e.status_code
+
+    if args.debug:
+        logging.getLogger("capa").setLevel(logging.DEBUG)
+        logging.getLogger("viv_utils").setLevel(logging.DEBUG)
+    else:
+        logging.getLogger("capa").setLevel(logging.ERROR)
+        logging.getLogger("viv_utils").setLevel(logging.ERROR)
+
+    table = rich.table.Table()
+    table.add_column("backend")
+    table.add_column("already installed?")
+    table.add_column("found?")
+    table.add_column("loads?")
+
+    if True:
+        row = ["vivisect"]
+        if is_vivisect_installed():
+            row.append("True")
+            row.append("-")
+        else:
+            row.append("False")
+            row.append("False")
+
+        row.append(str(load_vivisect()))
+        table.add_row(*row)
+
+    if True:
+        row = ["Binary Ninja"]
+        if is_binaryninja_installed():
+            row.append("True")
+            row.append("-")
+        else:
+            row.append("False")
+            row.append(str(find_binaryninja() is not None))
+
+        row.append(str(load_binaryninja()))
+        table.add_row(*row)
+
+    if True:
+        row = ["IDA idalib"]
+        if is_idalib_installed():
+            row.append("True")
+            row.append("-")
+        else:
+            row.append("False")
+            row.append(str(find_idalib() is not None))
+
+        row.append(str(load_idalib()))
+        table.add_row(*row)
+
+    rich.print(table)
+
+
+if __name__ == "__main__":
+    sys.exit(main())