Haijian Zhang

"""Canonical, compiled JavaScript regex patterns shared across JS extractors.

All suspicious-API patterns and the few feature-only patterns live here so each
expression is compiled exactly once per Python process and so any pattern that
was previously duplicated across `js_features.py` and `js_suspicious_apis.py`
now resolves to a single shared compiled object.

JavaScript is case-sensitive at runtime, but every suspicious-API pattern matches
either a literal-case identifier (`\\beval\\s*\\(`, `String\\.fromCharCode`, etc.)
or a string-quoted token (`"powershell"`). Compiling them with `re.IGNORECASE`
matches the historical behaviour of `JSSuspiciousAPIsExtractor` and is safe for
the patterns that historically came from `JSFeaturesExtractor` — those literals
are spelled in real-world JS exactly as written.

`scan_source()` is the entry point used by extractors: it walks the source once
per pattern using the pre-compiled regexes and returns a flat
`{name: {"count": N, "lines": [unique_line_numbers_sorted]}}` dict. Both
`JSFeaturesExtractor` and `JSSuspiciousAPIsExtractor` consume the same dict so
the per-pattern × per-line loops they used to run independently collapse to a
single shared scan.
"""

import bisect
import re
from typing import Dict, Iterable, List, Mapping

_FLAGS = re.IGNORECASE

# Canonical compiled patterns, keyed by their human-readable name. The name is
# also the value emitted into `redb_js_suspicious_apis.api_name`.
PATTERNS = {
    # ---- code execution ----
    "eval": re.compile(r"\beval\s*\(", _FLAGS),
    "Function constructor": re.compile(r"\bnew\s+Function\s*\(", _FLAGS),
    "execScript": re.compile(r"\bexecScript\s*\(", _FLAGS),
    "document.write": re.compile(r"\bdocument\.write(?:ln)?\s*\(", _FLAGS),
    "innerHTML assignment": re.compile(r"\.innerHTML\s*=", _FLAGS),
    "outerHTML assignment": re.compile(r"\.outerHTML\s*=", _FLAGS),
    "insertAdjacentHTML": re.compile(r"\.insertAdjacentHTML\s*\(", _FLAGS),
    # ---- network ----
    "XMLHttpRequest": re.compile(r"\bnew\s+XMLHttpRequest\b", _FLAGS),
    "fetch": re.compile(r"\bfetch\s*\(", _FLAGS),
    "WebSocket": re.compile(r"\bnew\s+WebSocket\s*\(", _FLAGS),
    "navigator.sendBeacon": re.compile(r"\bnavigator\.sendBeacon\s*\(", _FLAGS),
    "ActiveXObject XMLHTTP": re.compile(
        r"ActiveXObject\s*\(\s*[\"\'](?:MSXML2\.XMLHTTP|Microsoft\.XMLHTTP)", _FLAGS
    ),
    "require network module": re.compile(
        r"require\s*\(\s*[\"\'](?:http|https|net|dgram)[\"\']", _FLAGS
    ),
    "axios": re.compile(r"\baxios\b", _FLAGS),
    # ---- filesystem ----
    "require fs": re.compile(r"require\s*\(\s*[\"\']fs[\"\']", _FLAGS),
    "require path": re.compile(r"require\s*\(\s*[\"\']path[\"\']", _FLAGS),
    "FileSystemObject": re.compile(r"Scripting\.FileSystemObject", _FLAGS),
    "ADODB.Stream": re.compile(r"ADODB\.Stream", _FLAGS),
    "Shell.Application": re.compile(r"Shell\.Application", _FLAGS),
    "WScript.CreateObject": re.compile(r"WScript\.CreateObject", _FLAGS),
    # ---- process ----
    "require child_process": re.compile(r"require\s*\(\s*[\"\']child_process[\"\']", _FLAGS),
    "child_process exec": re.compile(r"child_process\.(?:exec|spawn|execFile|fork)\s*\(", _FLAGS),
    "WScript.Shell": re.compile(r"WScript\.Shell", _FLAGS),
    "WScript.Shell.Run": re.compile(r"\.Run\s*\(", _FLAGS),
    "WScript.Shell.Exec": re.compile(r"\.Exec\s*\(", _FLAGS),
    "ShellExecute": re.compile(r"\bShellExecute\b", _FLAGS),
    "PowerShell reference": re.compile(r"[\"\']powershell[\"\']", _FLAGS),
    "cmd.exe reference": re.compile(r"[\"\']cmd\.exe[\"\']", _FLAGS),
    "require os": re.compile(r"require\s*\(\s*[\"\']os[\"\']", _FLAGS),
    # ---- registry ----
    "RegRead": re.compile(r"\.RegRead\s*\(", _FLAGS),
    "RegWrite": re.compile(r"\.RegWrite\s*\(", _FLAGS),
    "RegDelete": re.compile(r"\.RegDelete\s*\(", _FLAGS),
    "StdRegProv": re.compile(r"StdRegProv", _FLAGS),
    # ---- crypto / encoding ----
    "atob": re.compile(r"\batob\s*\(", _FLAGS),
    "btoa": re.compile(r"\bbtoa\s*\(", _FLAGS),
    "String.fromCharCode": re.compile(r"String\.fromCharCode\s*\(", _FLAGS),
    "unescape": re.compile(r"\bunescape\s*\(", _FLAGS),
    "decodeURIComponent": re.compile(r"\bdecodeURIComponent\s*\(", _FLAGS),
    "Buffer.from": re.compile(r"Buffer\.from\s*\(", _FLAGS),
    "crypto module": re.compile(r"crypto\.create(?:Cipher|Decipher|Hash|Hmac)", _FLAGS),
    # ---- DOM manipulation ----
    "document.forms": re.compile(r"document\.forms", _FLAGS),
    "document.cookie": re.compile(r"document\.cookie", _FLAGS),
    "querySelector sensitive input": re.compile(
        r"document\.querySelector\s*\([^)]*(?:password|credit|card|cvv|ssn)", _FLAGS
    ),
    "submit event listener": re.compile(r"addEventListener\s*\(\s*[\"\']submit", _FLAGS),
    "createElement script/iframe": re.compile(
        r"\.createElement\s*\(\s*[\"\'](?:script|iframe)", _FLAGS
    ),
    "dynamic script src": re.compile(r"\.src\s*=\s*[\"\'](?:https?://|//)", _FLAGS),
}

# Pattern name -> category (one of code_execution / network / filesystem /
# process / registry / crypto_encoding / dom_manipulation).
CATEGORIES = {
    "eval": "code_execution",
    "Function constructor": "code_execution",
    "execScript": "code_execution",
    "document.write": "code_execution",
    "innerHTML assignment": "code_execution",
    "outerHTML assignment": "code_execution",
    "insertAdjacentHTML": "code_execution",
    "XMLHttpRequest": "network",
    "fetch": "network",
    "WebSocket": "network",
    "navigator.sendBeacon": "network",
    "ActiveXObject XMLHTTP": "network",
    "require network module": "network",
    "axios": "network",
    "require fs": "filesystem",
    "require path": "filesystem",
    "FileSystemObject": "filesystem",
    "ADODB.Stream": "filesystem",
    "Shell.Application": "filesystem",
    "WScript.CreateObject": "filesystem",
    "require child_process": "process",
    "child_process exec": "process",
    "WScript.Shell": "process",
    "WScript.Shell.Run": "process",
    "WScript.Shell.Exec": "process",
    "ShellExecute": "process",
    "PowerShell reference": "process",
    "cmd.exe reference": "process",
    "require os": "process",
    "RegRead": "registry",
    "RegWrite": "registry",
    "RegDelete": "registry",
    "StdRegProv": "registry",
    "atob": "crypto_encoding",
    "btoa": "crypto_encoding",
    "String.fromCharCode": "crypto_encoding",
    "unescape": "crypto_encoding",
    "decodeURIComponent": "crypto_encoding",
    "Buffer.from": "crypto_encoding",
    "crypto module": "crypto_encoding",
    "document.forms": "dom_manipulation",
    "document.cookie": "dom_manipulation",
    "querySelector sensitive input": "dom_manipulation",
    "submit event listener": "dom_manipulation",
    "createElement script/iframe": "dom_manipulation",
    "dynamic script src": "dom_manipulation",
}

# Patterns consumed only by JSFeaturesExtractor (no category, never surfaced as
# a suspicious-API row). Kept here so every JS regex is compiled in one place.
FEATURE_PATTERNS = {
    "hex_escape": re.compile(r"\\x[0-9a-fA-F]{2}"),
    "unicode_escape": re.compile(r"\\u[0-9a-fA-F]{4}"),
    "base64_string": re.compile(r"[A-Za-z0-9+/]{40,}={0,2}"),
    # decodeURI matches BOTH decodeURI and decodeURIComponent. The latter is also
    # a suspicious-API pattern in PATTERNS; this broader form is what the
    # `decodeuri_count` feature column has historically counted.
    "decodeURI": re.compile(r"\b(?:decodeURI|decodeURIComponent)\s*\(", _FLAGS),
    "settimeout_setinterval": re.compile(r"\b(?:setTimeout|setInterval)\s*\(", _FLAGS),
    "function_decl": re.compile(r"\bfunction\s+\w+\s*\(|\bfunction\s*\("),
    "var_decl": re.compile(r"\b(?:var|let|const)\s+"),
    "string_concat": re.compile(r"[\"\'][\s]*\+[\s]*[\"\']"),
    "comment": re.compile(r"//.*?$|/\*[\s\S]*?\*/", re.MULTILINE),
    "long_string": re.compile(r"[\"\']([^\"\']{256,})[\"\']"),
    "array_function_call": re.compile(r"\[(?:0x[0-9a-f]+|[\d]+)\]\s*\(", _FLAGS),
}

# Patterns consumed only by JSStringsExtractor for encoded-string discovery.
# Scoped to *hidden* strings only — patterns whose decoded form is not visible
# to a substring search over the raw text. Plain long literals are not
# extracted here because they're already preserved in code_text_content and
# scraped by the IOC pipeline over text_raw / text_normalized.
#
# Distinct from FEATURE_PATTERNS even where the names rhyme:
#   FEATURE_PATTERNS["hex_escape"] / ["unicode_escape"]   -> single escape
#   STRING_PATTERNS["hex_escape_seq"] / ["unicode_escape_seq"] -> 4+ / 3+ in a row
#   FEATURE_PATTERNS["base64_string"]                     -> bare base64 token
#   STRING_PATTERNS["base64_quoted"]                      -> base64 inside JS quotes
# These do not share match objects with the suspicious-API or feature scans, so
# they are not folded into JSContext.scan; the strings extractor walks them
# itself (one finditer per pattern, with shared line-offset bisect in #4b).
STRING_PATTERNS = {
    "hex_escape_seq": re.compile(r"(?:\\x[0-9a-fA-F]{2}){4,}"),
    "unicode_escape_seq": re.compile(r"(?:\\u[0-9a-fA-F]{4}){3,}"),
    "charcode_call": re.compile(r"String\.fromCharCode\s*\(\s*([\d,\s]+)\s*\)"),
    "base64_quoted": re.compile(r"[\"\']([A-Za-z0-9+/]{40,}={0,2})[\"\']"),
    "concat_chain": re.compile(r"(?:[\"\'][^\"\']+[\"\']\s*\+\s*){3,}[\"\'][^\"\']+[\"\']"),
}


def line_offsets(source: str) -> List[int]:
    """Sorted list of byte offsets for every newline in `source`, plus a final
    sentinel of len(source). Used to translate match offsets into 1-indexed
    line numbers via bisect.
    """
    offsets = [-1]  # so that bisect_right of offset 0 returns line 1
    push = offsets.append
    idx = source.find("\n")
    while idx != -1:
        push(idx)
        idx = source.find("\n", idx + 1)
    return offsets


def _scan_one(
    pattern: "re.Pattern[str]", source: str, offsets: List[int]
) -> Dict[str, object]:
    """Run a single compiled pattern over `source` and return count + unique lines."""
    count = 0
    seen_lines: "set[int]" = set()
    for m in pattern.finditer(source):
        count += 1
        seen_lines.add(bisect.bisect_right(offsets, m.start()))
    if not count:
        return None  # type: ignore[return-value]
    return {"count": count, "lines": sorted(seen_lines)}


def scan_source(
    source: str,
    patterns: Iterable[Mapping[str, "re.Pattern[str]"]] = (PATTERNS, FEATURE_PATTERNS),
) -> Dict[str, Dict[str, object]]:
    """Scan `source` against every compiled pattern in `patterns`.

    Returns a dict keyed by pattern name. Each entry has:
        "count": total number of matches in the source
        "lines": sorted list of unique 1-indexed line numbers where the pattern
                 matched (deduplicated — multiple matches on the same line
                 collapse to one entry, preserving the historical
                 line-set semantics of JSSuspiciousAPIsExtractor)
    Patterns with zero matches are absent from the dict; callers should default
    to {"count": 0, "lines": []}.
    """
    if not source:
        return {}
    offsets = line_offsets(source)
    results: Dict[str, Dict[str, object]] = {}
    for table in patterns:
        for name, pat in table.items():
            entry = _scan_one(pat, source, offsets)
            if entry is not None:
                results[name] = entry
    return results