Taint Flow Analysis

Primus Security uses two complementary taint engines:

Engine	Scope	Confidence
Single-file Roslyn analyzers	Within one .cs file	0.90
Cross-file taint propagator	Across Controller → Service → Repository	0.80

Both are active by default inside ScanAsync(). No configuration required.

Why taint flow matters

Simple pattern matching flags any SQL string — including hard-coded ones that are perfectly safe. Taint flow checks: "does user-controlled input actually reach this dangerous sink?"

// Pattern-only: FLAGGED (SQL string found)
// Taint flow:   CLEAN (no user input reaches it)
var sql = "SELECT * FROM Products WHERE Category = 'Electronics'";
db.Execute(sql);

// Both flag this — taint proven, confidence 0.90
var category = Request.Query["category"];
var sql = $"SELECT * FROM Products WHERE Category = '{category}'";
db.Execute(sql);

Cross-file taint analysis (Phase 1)

The most common real-world vulnerability pattern spans multiple files:

// UserController.cs — receives HTTP input, calls service
public IActionResult Get(string id)
{
    return Ok(_service.GetUser(id));  // id is tainted here
}

// UserService.cs — passes to repository
public User GetUser(string id) => _repo.Find(id);

// UserRepository.cs — builds raw SQL (SINK)
public User Find(string id) =>
    db.Execute($"SELECT * FROM Users WHERE Id = '{id}'");
    // ↑ SQL injection — single-file analysis misses this entirely

The MethodCallGraph builds a directed call graph across all .cs files in the compilation. The CrossFileTaintPropagator then follows tainted arguments through the graph up to 5 hops (configurable), emitting a CrossFileTaintFlowFinding with the full call chain as evidence.

Cross-file findings use rule IDs prefixed PS-CF- (e.g. PS-CF-001 for cross-file SQL injection) to distinguish them from single-file findings.

What's detected cross-file

Vulnerability	Sink methods
SQL Injection	ExecuteNonQuery, ExecuteSqlRaw, FromSqlRaw, Query (Dapper)
XSS	Html.Raw, HtmlString, response Write
SSRF	HttpClient.GetAsync/PostAsync/SendAsync, new Uri()
Path Traversal	File.ReadAllText, File.Open, Path.Combine
Command Injection	Process.Start, new ProcessStartInfo()
Open Redirect	Controller.Redirect, LocalRedirect
Sensitive Data Logging	ILogger.Log*

Reading cross-file findings

var result = await scanner.ScanAsync("/path/to/project");

var crossFileFindings = result.Findings
    .Where(f => f.RuleId?.StartsWith("PS-CF-") == true)
    .ToList();

foreach (var finding in crossFileFindings)
{
    Console.WriteLine($"[{finding.RuleId}] {finding.Title}");
    Console.WriteLine($"Sink: {finding.FilePath}:{finding.Line}");
    Console.WriteLine($"Chain: {finding.Description}");
    // "Cross-file SQL Injection (80% confidence).
    //  Chain: UserController.cs:24 (Get) → UserService.cs:18 (GetUser)
    //       → UserRepository.cs:31 (Find)"
}

Confidence levels

Engine	Confidence	Reason
Single-file Roslyn	0.90	Exact AST + SemanticModel
Cross-file propagator	0.80	Call graph construction uses approximation

Lower confidence means slightly more potential false positives. Cross-file findings should be reviewed before suppressing.

Single-file taint (per-analyzer)

The 60 Roslyn analyzers each run their own TaintEngine instance per-compilation. Sources, propagation, and sink detection within a single file achieve 0.90 confidence.

Supported sources

Source	Pattern
[FromQuery] / [FromBody] / [FromRoute]	ASP.NET model binding attributes
Request.Query / Request.Form / Request.Headers	HttpContext properties
Action method parameters	MVC route values
Console.ReadLine()	Console input

Supported sinks (sample)

Sink	Rule
SQL concatenation	PS-SEC-001
Process.Start	PS-SEC-006
File.Open / Path.Combine	PS-SEC-007
HttpClient.GetAsync	PS-SEC-009
Html.Raw / HtmlString	PS-SEC-002

For the full list see the rule catalog.

Suppressing taint findings

// Inline — single line
var sql = $"SELECT * FROM Products WHERE Id = {id}"; // primus-nosec

// File-level — .primusignore in project root
**/Migrations/*.cs
**/Seed/*.cs

For persistent cross-run suppression see Suppression Store.

Why taint flow matters

Simple pattern matching flags any SQL string — including hard-coded ones that are perfectly safe. Taint flow checks: "does user-controlled input actually reach this dangerous sink?"

// Pattern-only scanner: FLAGGED (SQL string found)
// Taint flow scanner:   CLEAN (no user input reaches it)
var sql = "SELECT * FROM Products WHERE Category = 'Electronics'";
db.Execute(sql);

// Pattern-only scanner: FLAGGED
// Taint flow scanner:   FLAGGED with HIGH confidence (taint proven to reach sink)
var category = Request.Query["category"];
var sql = $"SELECT * FROM Products WHERE Category = '{category}'";
db.Execute(sql);

Taint flow findings carry Confidence: 0.90 vs 0.70–0.85 for pattern-only.

Usage

using PrimusSaaS.Security.Heuristics;

var analyzer = new TaintFlowAnalyzer();
var code = File.ReadAllText("Controllers/ProductsController.cs");

var findings = analyzer.Analyze(code);

foreach (var finding in findings)
{
    Console.WriteLine(finding.Message);
    // "SQL Injection: 'category' flows from user input (line 12) to dangerous sink (line 15)."

    Console.WriteLine($"Source: {finding.SourceSnippet} (line {finding.SourceLine})");
    Console.WriteLine($"Sink:   {finding.SinkSnippet} (line {finding.SinkLine})");
    Console.WriteLine($"Confidence: {finding.Confidence:P0}");
}

Supported sources

The analyzer recognises these user-controlled input sources in C# / ASP.NET:

Source	Pattern
Request.Query / Request.QueryString / Request.Form	HTTP query/form parameters
RouteData.Values / route values	URL route segments
Controller action parameters (string id, int value)	MVC action method parameters
Console.ReadLine()	Console input
JsonSerializer.Deserialize / ReadFromJsonAsync	Deserialized HTTP body
Environment.GetEnvironmentVariable()	Environment variables
File.ReadAllText / StreamReader	File content

Supported sinks

Sink	Rule	Severity
SQL concatenation / CommandText =	PS-SEC-001	High
SQL via string interpolation ($"SELECT...{var}")	PS-SEC-001	High
Process.Start / ProcessStartInfo	PS-SEC-004	High
File.Open / File.Read / Path.Combine	PS-SEC-007	High
Redirect() / Response.Redirect()	PS-SEC-010	Medium
Logger methods with tainted args	PS-SEC-006	Medium
HttpClient.GetAsync / SendAsync with tainted URL	PS-SEC-009	High

TaintFlowFinding reference

public sealed class TaintFlowFinding
{
    public string VulnerabilityType { get; }  // e.g. "SQL Injection"
    public ThreatSeverity Severity { get; }
    public string RuleId { get; }             // e.g. "PS-SEC-001"
    public double Confidence { get; }         // 0.90 when taint is proven

    public string SourceDescription { get; }  // Human-readable taint source
    public int    SourceLine { get; }
    public string SourceSnippet { get; }

    public int    SinkLine { get; }
    public string SinkSnippet { get; }
    public string TaintedVariable { get; }

    public string Message { get; }
    // "SQL Injection: 'id' flows from user input (line 8) to dangerous sink (line 12)."
}

Integration with the main scanner

The SecurityScanner runs taint flow analysis as part of ScanAsync() automatically — you do not need to call TaintFlowAnalyzer directly in most cases. It appears in findings with the Source = "TaintFlow" property and higher confidence scores than pattern-only findings for the same rule.