MongoDB · Create

A security check around creates: log, block, handle false positives

A security check around every create (SecurityGuard, in the project root), in four scenarios: a normal request passes through; an attack is logged and the user sees a plain message; repeated attacks block the user; and a false positive, a harmless file header flagged as SQL, is triaged and fixed without blocking anyone.

Set up for this page

Besides the sample itself, it uses SampleContext.cs (the dbContext every sample is given); SampleOutput.cs (Step and Show, which print); SampleContext.Application.cs (dbContext.Service, the application's ARX service); SampleCatalog.cs (the tables' ids, keys, lookups and display names); SampleContext.Direct.cs (dbContext.Direct); MongoDirectHelper.cs (the checks behind dbContext.Direct); SecurityGuard.cs (the security log and strikes of the attack samples), in the files below.

dbContext.Direct is the samples' own checking code on the plain driver, not part of ARX: leave it out of your application.

Before scenario 1 the sample runs this, so the service and the rows the scenarios use are there:

Set up
var guard = new SecurityGuard(blockAfter: 3);

1 A normal request passes through

A normal request goes through the guard untouched.

Already in scope: service (ArxMongoDbService), guard (SecurityGuard)

C#
var result = await CreateContactAsync(service, guard, "ana", new Dictionary<string, object>
{
    ["first_name"] = "Ana",
    ["last_name"] = "Guard-C1",
    ["email"] = "[email protected]",
});

Console.WriteLine($"ana: {result.Outcome}, contact {result.Value}");
Used above

Defined once in this sample.

private static Task<GuardResult<ArxRecordId>> CreateContactAsync(ArxMongoDbService service, SecurityGuard guard, string user, Dictionary<string, object> posted)
{
    var contact = new ArxEntity("contact");

    foreach (var field in posted)
    {
        contact[field.Key] = field.Value;
    }

    return guard.RunAsync(user, "create contact", NameSource.UserInput, () => service.CreateAsync(contact));
}

Check it in the database

javascript

Run it in mongosh connected to the database, after the code has run.

db.contact.find({ last_name: "Guard-C1" }, { first_name: 1, email: 1 })

Expect: 1 document: Ana, [email protected].

2 An attack is logged; the user sees a plain message

A field name carrying SQL: ARX refuses it (ARX1017), the guard logs it and the user sees a plain message.

Already in scope: service (ArxMongoDbService), guard (SecurityGuard)

C#
var result = await CreateContactAsync(service, guard, "mallory", new Dictionary<string, object>
{
    ["first_name"] = "Mallory",
    ["last_name"] = "Guard-C2",
    ["notes\") VALUES ('x'); DROP TABLE contact; --"] = "x",
});

Console.WriteLine($"mallory sees: \"{result.ForTheUser}\"");
var logged = guard.Events[^1];
Console.WriteLine($"security log: {logged.Kind}, {logged.User}, {logged.Action}");
Used above

Defined once in this sample.

private static Task<GuardResult<ArxRecordId>> CreateContactAsync(ArxMongoDbService service, SecurityGuard guard, string user, Dictionary<string, object> posted)
{
    var contact = new ArxEntity("contact");

    foreach (var field in posted)
    {
        contact[field.Key] = field.Value;
    }

    return guard.RunAsync(user, "create contact", NameSource.UserInput, () => service.CreateAsync(contact));
}

Check it in the database

javascript

Run it in mongosh connected to the database, after the code has run.

db.contact.countDocuments({ last_name: "Guard-C2" })

Expect: 0: nothing created.

3 Repeated attacks block the user

Two more attempts make three strikes: the user is blocked, and even a valid request isn't attempted.

Already in scope: service (ArxMongoDbService), guard (SecurityGuard)

C#
foreach (var probe in new[] { "last_name; DROP TABLE account", "email UNION SELECT 1" })
{
    await CreateContactAsync(service, guard, "mallory", new Dictionary<string, object> { ["first_name"] = "Mallory", [probe] = "x" });
}

var valid = await CreateContactAsync(service, guard, "mallory", new Dictionary<string, object>
{
    ["first_name"] = "Mallory",
    ["last_name"] = "Guard-C3",
});

Console.WriteLine($"mallory blocked: {guard.IsBlocked("mallory")}; a valid request: {valid.Outcome}, \"{valid.ForTheUser}\"");
Used above

Defined once in this sample.

private static Task<GuardResult<ArxRecordId>> CreateContactAsync(ArxMongoDbService service, SecurityGuard guard, string user, Dictionary<string, object> posted)
{
    var contact = new ArxEntity("contact");

    foreach (var field in posted)
    {
        contact[field.Key] = field.Value;
    }

    return guard.RunAsync(user, "create contact", NameSource.UserInput, () => service.CreateAsync(contact));
}

Check it in the database

javascript

Run it in mongosh connected to the database, after the code has run.

db.contact.countDocuments({ first_name: "Mallory" })

Expect: 0: no request from mallory created anything, the last one wasn't even attempted.

4 A false positive: a harmless file header flagged as SQL

A false positive: a supplier file's header "E-mail (work)" is used as an attribute name. The parentheses make ARX flag it as SQL, though it's harmless. One strike doesn't block anyone; triage reads the log, and the fix is the mapping that should have been there: file headers mapped to known attributes, so user input never becomes a name.

Already in scope: service (ArxMongoDbService), guard (SecurityGuard)

C#
var header = new[] { "first_name", "last_name", "E-mail (work)" };
var row = new object[] { "Sam", "Guard-C4", "[email protected]" };

var asIs = header.Zip(row).ToDictionary(x => x.First, x => x.Second);
var refused = await CreateContactAsync(service, guard, "sam", asIs);

Console.WriteLine($"sam: {refused.Outcome}; blocked: {guard.IsBlocked("sam")} (one strike blocks nobody)");
Console.WriteLine($"triage reads: {guard.Events[^1].Detail}");

// The fix: known headers mapped to attributes; anything else is ignored (or reported to the user), never a name.
var headerMap = new Dictionary<string, string>(StringComparer.OrdinalIgnoreCase)
{
    ["first_name"] = "first_name",
    ["last_name"] = "last_name",
    ["E-mail (work)"] = "email",
    ["E-mail"] = "email",
};

var mapped = header.Zip(row).Where(x => headerMap.ContainsKey(x.First)).ToDictionary(x => headerMap[x.First], x => x.Second);
var created = await CreateContactAsync(service, guard, "sam", mapped);

Console.WriteLine($"after the fix: {created.Outcome}, contact {created.Value}");
Used above

Defined once in this sample.

private static Task<GuardResult<ArxRecordId>> CreateContactAsync(ArxMongoDbService service, SecurityGuard guard, string user, Dictionary<string, object> posted)
{
    var contact = new ArxEntity("contact");

    foreach (var field in posted)
    {
        contact[field.Key] = field.Value;
    }

    return guard.RunAsync(user, "create contact", NameSource.UserInput, () => service.CreateAsync(contact));
}

Check it in the database

javascript

Run it in mongosh connected to the database, after the code has run.

db.contact.find({ last_name: "Guard-C4" }, { first_name: 1, email: 1 })

Expect: 1 document: Sam, [email protected] (created once, after the fix).

Good to know

ARX1017 is only raised for a name that is already invalid, so a false positive never stops a call that would otherwise have worked: without the check it would fail with ARX1003. What can be wrong is the label: a harmless name (E-mail (work), Price (EUR), notes; pasted into a setting) reported as an attack. Handle it with:

  • The source of the name: NameSource.Code for names from your code or configuration, so an ARX1017 there is logged as a bug and blocks nobody; NameSource.UserInput for names built from what a user sent.
  • Strikes, not one event: a single ARX1017 never blocks anyone.
  • Triage: the log has ARX's evidence (contains '('); if it's harmless, Unblock the user and fix what let the name through.
  • The real fix: map user input to names from a fixed list (headers to attributes, request types to tables), so user input never becomes a name, and neither attacks nor false positives can happen.

See also

Run the complete sample

The code above comes from one program, built in a fresh project and run against the database. To run it yourself, here is every file and step.

The program, its files and how to run it

Run it locally

  1. .NET SDK: .NET SDK 10.0.100 or later 10.0.x (target net10.0).
  2. Create the project:
    Terminal
    dotnet new console -n ArxSample -f net10.0
    cd ArxSample
  3. Install the packages:
    Terminal
    dotnet add package ARX.SDK.MongoDb --version 1.0.0-rc.1
  4. Database: a MongoDB 7.0, a replica set (tested with 7.0.45, one node) database with the sample schema.
  5. Configuration: set ARX_SAMPLES_MONGODB to its connection string. Nothing is hard-coded.
  6. Source code: replace Program.cs and add the 8 other files below, with the file names shown.
  7. Run it:
    Terminal
    dotnet run
  8. Expected output: below.

The files

Program.cs

Runs the sample. Generated for this page; it reads the connection string from the environment and first remakes the sample database from sample-schema.json.

using ArxSamples.MongoDb;
using ArxSamples.MongoDb.Operations.Create.AdvancedCases;

// The connection string comes from the environment, never from the code (see "Configure the connection").
var connectionString = Environment.GetEnvironmentVariable("ARX_SAMPLES_MONGODB")
    ?? throw new InvalidOperationException("Set the ARX_SAMPLES_MONGODB environment variable to your MongoDB connection string.");

var dbContext = new SampleContext(connectionString);

// Every run starts from the sample data: this deletes and makes again what sample-schema.json lists
// (save the file in the project folder, see Get started).
await dbContext.Direct.ResetAsync();

await new Advanced2_GuardAgainstAttacks().RunAsync(dbContext);
Advanced2_GuardAgainstAttacks.cs

The sample, as it is in the repository.

using ARX.SDK.Services;
using static ArxSamples.MongoDb.SampleOutput;

namespace ArxSamples.MongoDb.Operations.Create.AdvancedCases;

/// <summary>
/// A security check around every create (<see cref="SecurityGuard"/>, in the project root), in four scenarios: a normal
/// request passes through; an attack is logged and the user sees a plain message; repeated attacks block the user; and
/// a false positive, a harmless file header flagged as SQL, is triaged and fixed without blocking anyone.
/// </summary>
/// <remarks>
/// <para>ARX1017 is only raised for a name that is already invalid, so a false positive never stops a call that would
/// otherwise have worked: without the check it would fail with ARX1003. What can be wrong is the label: a harmless name
/// (<c>E-mail (work)</c>, <c>Price (EUR)</c>, <c>notes;</c> pasted into a setting) reported as an attack. Handle it with:</para>
/// <list type="bullet">
/// <item><b>The source of the name:</b> <c>NameSource.Code</c> for names from your code or configuration, so an ARX1017
/// there is logged as a bug and blocks nobody; <c>NameSource.UserInput</c> for names built from what a user sent.</item>
/// <item><b>Strikes, not one event:</b> a single ARX1017 never blocks anyone.</item>
/// <item><b>Triage:</b> the log has ARX's evidence (<c>contains '('</c>); if it's harmless, <c>Unblock</c> the user and
/// fix what let the name through.</item>
/// <item><b>The real fix:</b> map user input to names from a fixed list (headers to attributes, request types to
/// tables), so user input never becomes a name, and neither attacks nor false positives can happen.</item>
/// </list>
/// </remarks>
public sealed class Advanced2_GuardAgainstAttacks : ISample
{
    public string Name => "create-advanced-2-guard-against-attacks";

    public string Title => "A security check around creates: log, block, handle false positives";

    public async Task RunAsync(SampleContext dbContext)
    {
        // One guard for the application (register it as a singleton). Three strikes block a user.
        var guard = new SecurityGuard(blockAfter: 3);

        await Scenario1_ANormalRequestPassesThroughAsync(dbContext, guard);

        await Scenario2_AnAttackIsLoggedAsync(dbContext, guard);

        await Scenario3_RepeatedAttacksBlockTheUserAsync(dbContext, guard);

        await Scenario4_AFalsePositiveAsync(dbContext, guard);
    }

    /// <summary>The "create a contact" endpoint: the posted fields become attributes, so the names come from the user.</summary>
    private static Task<GuardResult<ArxRecordId>> CreateContactAsync(SampleContext dbContext, SecurityGuard guard, string user, Dictionary<string, object> posted)
    {
        var contact = new ArxEntity("contact");

        foreach (var field in posted)
        {
            contact[field.Key] = field.Value;
        }

        return guard.RunAsync(user, "create contact", NameSource.UserInput, () => dbContext.Service.CreateAsync(contact));
    }

    /// <summary>1. A normal request goes through the guard untouched.</summary>
    private static async Task Scenario1_ANormalRequestPassesThroughAsync(SampleContext dbContext, SecurityGuard guard)
    {
        Step("1. A normal request passes through");

        var result = await CreateContactAsync(dbContext, guard, "ana", new Dictionary<string, object>
        {
            ["first_name"] = "Ana",
            ["last_name"] = "Guard-C1",
            ["email"] = "[email protected]",
        });

        Show($"ana: {result.Outcome}, contact {result.Value}");

        /* ===========================================================
         * [DIRECT QUERY CHECK]
         * -----------------------------------------------------------
         * db.contact.find({ last_name: "Guard-C1" }, { first_name: 1, email: 1 })
         * -- Expect: 1 document: Ana, [email protected].
         * ===========================================================
         */
    }

    /// <summary>2. A field name carrying SQL: ARX refuses it (ARX1017), the guard logs it and the user sees a plain message.</summary>
    private static async Task Scenario2_AnAttackIsLoggedAsync(SampleContext dbContext, SecurityGuard guard)
    {
        Step("2. An attack is logged; the user sees a plain message");

        var result = await CreateContactAsync(dbContext, guard, "mallory", new Dictionary<string, object>
        {
            ["first_name"] = "Mallory",
            ["last_name"] = "Guard-C2",
            ["notes\") VALUES ('x'); DROP TABLE contact; --"] = "x",
        });

        Show($"mallory sees: \"{result.ForTheUser}\"");
        var logged = guard.Events[^1];
        Show($"security log: {logged.Kind}, {logged.User}, {logged.Action}");

        /* ===========================================================
         * [DIRECT QUERY CHECK]
         * -----------------------------------------------------------
         * db.contact.countDocuments({ last_name: "Guard-C2" })
         * -- Expect: 0: nothing created.
         * ===========================================================
         */
    }

    /// <summary>3. Two more attempts make three strikes: the user is blocked, and even a valid request isn't attempted.</summary>
    private static async Task Scenario3_RepeatedAttacksBlockTheUserAsync(SampleContext dbContext, SecurityGuard guard)
    {
        Step("3. Repeated attacks block the user");

        foreach (var probe in new[] { "last_name; DROP TABLE account", "email UNION SELECT 1" })
        {
            await CreateContactAsync(dbContext, guard, "mallory", new Dictionary<string, object> { ["first_name"] = "Mallory", [probe] = "x" });
        }

        var valid = await CreateContactAsync(dbContext, guard, "mallory", new Dictionary<string, object>
        {
            ["first_name"] = "Mallory",
            ["last_name"] = "Guard-C3",
        });

        Show($"mallory blocked: {guard.IsBlocked("mallory")}; a valid request: {valid.Outcome}, \"{valid.ForTheUser}\"");

        /* ===========================================================
         * [DIRECT QUERY CHECK]
         * -----------------------------------------------------------
         * db.contact.countDocuments({ first_name: "Mallory" })
         * -- Expect: 0: no request from mallory created anything, the last one wasn't even attempted.
         * ===========================================================
         */
    }

    /// <summary>
    /// 4. A false positive: a supplier file's header "E-mail (work)" is used as an attribute name. The parentheses make
    /// ARX flag it as SQL, though it's harmless. One strike doesn't block anyone; triage reads the log, and the fix is
    /// the mapping that should have been there: file headers mapped to known attributes, so user input never becomes a name.
    /// </summary>
    private static async Task Scenario4_AFalsePositiveAsync(SampleContext dbContext, SecurityGuard guard)
    {
        Step("4. A false positive: a harmless file header flagged as SQL");

        var header = new[] { "first_name", "last_name", "E-mail (work)" };
        var row = new object[] { "Sam", "Guard-C4", "[email protected]" };

        var asIs = header.Zip(row).ToDictionary(x => x.First, x => x.Second);
        var refused = await CreateContactAsync(dbContext, guard, "sam", asIs);

        Show($"sam: {refused.Outcome}; blocked: {guard.IsBlocked("sam")} (one strike blocks nobody)");
        Show($"triage reads: {guard.Events[^1].Detail}");

        // The fix: known headers mapped to attributes; anything else is ignored (or reported to the user), never a name.
        var headerMap = new Dictionary<string, string>(StringComparer.OrdinalIgnoreCase)
        {
            ["first_name"] = "first_name",
            ["last_name"] = "last_name",
            ["E-mail (work)"] = "email",
            ["E-mail"] = "email",
        };

        var mapped = header.Zip(row).Where(x => headerMap.ContainsKey(x.First)).ToDictionary(x => headerMap[x.First], x => x.Second);
        var created = await CreateContactAsync(dbContext, guard, "sam", mapped);

        Show($"after the fix: {created.Outcome}, contact {created.Value}");

        /* ===========================================================
         * [DIRECT QUERY CHECK]
         * -----------------------------------------------------------
         * db.contact.find({ last_name: "Guard-C4" }, { first_name: 1, email: 1 })
         * -- Expect: 1 document: Sam, [email protected] (created once, after the fix).
         * ===========================================================
         */
    }
}
SampleContext.cs

What every sample is given: the connection string (and the ISample interface the samples implement).

namespace ArxSamples.MongoDb;

/// <summary>One documentation sample: a scenario a real application has, shown end to end.</summary>
public interface ISample
{
    /// <summary>The name to run it alone: <c>dotnet run -- &lt;name&gt;</c>.</summary>
    string Name { get; }

    /// <summary>One line on what it shows.</summary>
    string Title { get; }

    Task RunAsync(SampleContext dbContext);
}

/// <summary>
/// What a sample is given. Every sample gets the connection string; the application's service and catalog
/// (SampleContext.Application.cs) and the plain MongoDB driver read-back helper (SampleContext.Direct.cs) live in their own
/// files, so a sample needs only the files it uses.
/// </summary>
public sealed partial class SampleContext
{
    public SampleContext(string connectionString)
    {
        ConnectionString = connectionString;
    }

    public string ConnectionString { get; }
}
SampleOutput.cs

How the samples print: Step for each scenario, Show for what it did, ShowError for an error caught on purpose.

using ARX.SDK;

namespace ArxSamples.MongoDb;

/// <summary>How the samples print: a numbered step for each scenario, then what it did.</summary>
public static class SampleOutput
{
    /// <summary>The start of a scenario: <c>Step("1. By the id Create returned")</c>.</summary>
    public static void Step(string text) => Console.WriteLine($"  > {text}");

    /// <summary>What the scenario did.</summary>
    public static void Show(string text) => Console.WriteLine($"      {text}");

    /// <summary>An ARX error the scenario caught on purpose.</summary>
    public static void ShowError(ArxException exception)
        => Console.WriteLine($"      caught {exception.GetType().Name}: {exception.Message}");
}
SampleContext.Application.cs

The application's ARX service and its schema catalog, created once and shared by the samples.

using ARX.SDK.MongoDb.Services;
using ARX.SDK.Services;

namespace ArxSamples.MongoDb;

/// <summary>The application's ARX service and its schema catalog, shared by the real and advanced samples.</summary>
public sealed partial class SampleContext
{
    private ArxSchemaCatalog? _catalog;
    private ArxMongoDbService? _service;

    /// <summary>The application's schema catalog: collection ids, keys, lookups and selection sets (SampleCatalog.cs).</summary>
    public ArxSchemaCatalog Catalog => _catalog ??= SampleCatalog.Build();

    /// <summary>
    /// One service for the whole application. It shares one MongoDB client per connection string (a client is a
    /// connection pool), so make it once and keep it.
    /// </summary>
    public ArxMongoDbService Service => _service ??= new ArxMongoDbService(ConnectionString, Catalog);
}
SampleCatalog.cs

The application's schema catalog: every table's id column and keys, and its selection sets.

using ARX.SDK.Services;

namespace ArxSamples.MongoDb;

/// <summary>An account's industry. Stored in account.industry as the enum's number.</summary>
public enum Industry
{
    Retail = 1,
    Manufacturing = 2,
    Healthcare = 3,
    Finance = 4,
    Technology = 5
}

/// <summary>
/// The application's schema catalog: every collection's id and keys, the lookups between collections, and every
/// picklist (selection set), defined once.
/// </summary>
/// <remarks>
/// <para>
/// <b>Ids.</b> On MongoDB the declared id is the document's <c>_id</c>, named in your code by the column you declare
/// (<c>account["account_id"]</c>). A generated string id is an ObjectId ARX makes before the insert and gives back as its
/// 24-character text; a generated Guid is one ARX makes; a supplied id is yours (a country code, a SKU). MongoDB has no
/// sequences, so a generated number isn't offered, and every document has an <c>_id</c>, so there is no NoIdColumn.
/// </para>
/// <para>
/// <b>Keys.</b> Business keys (an account number; a region and a SKU together) let Update, Delete and Retrieve find
/// documents by them. MongoDB enforces them with unique indexes: the sample database has one for each.
/// </para>
/// <para>
/// <b>Lookups.</b> MongoDB has no foreign keys, so the catalog says which field refers to which collection
/// (<c>.Lookup("contact", "account_id", "account")</c>). ARX stores the referenced id as it is stored there (an ObjectId)
/// and reads the field back as an <see cref="ArxEntityReference"/> with the referenced record's display name.
/// </para>
/// <para>
/// <b>Selection sets.</b> A field that only accepts a fixed list of codes: an account's status, a contact's preferred
/// channels (several codes, stored as an array). ARX checks every <see cref="ArxSelectionSetValue"/> against the catalog
/// before anything is sent and stores the codes themselves.
/// </para>
/// </remarks>
public static class SampleCatalog
{
    public static ArxSchemaCatalog Build() => ArxSchemaCatalog.Create()
        .Table("account", ArxIdColumn.Generated<string>("account_id"))
        .Key("account", "account_number")
        .DisplayName("account", "name")
        .Table("contact", ArxIdColumn.Generated<string>("contact_id"))
        .Key("contact", "email")
        .Lookup("contact", "account_id", "account")
        .DisplayName("contact", new[] { "first_name", "last_name" }, " ")
        .Table("country", ArxIdColumn.Supplied<string>("code"))
        .DisplayName("country", "name")
        .Table("product", ArxIdColumn.Supplied<string>("sku"))
        .DisplayName("product", "name")
        .Table("price_list", ArxIdColumn.Generated<string>("price_id"))
        .Key("price_list", "region", "sku")
        .Lookup("price_list", "region", "country")
        .Table("sales_order", ArxIdColumn.Generated<Guid>("order_id"))
        .Key("sales_order", "order_number")
        .Lookup("sales_order", "account_id", "account")
        .Table("activity_log", ArxIdColumn.Generated<string>("log_id"))
        .AddEnum<Industry>("Industry")
        .Add("AccountStatus", ArxSelectionSetKind.String, ("ACTIVE", "Active"), ("ON_HOLD", "On hold"), ("CLOSED", "Closed"))
        .Add("ContactChannels", ArxSelectionSetKind.String, ArxSelectionSetMultiplicity.Array, ("E", "Email"), ("P", "Phone"), ("S", "SMS"))
        .Build();
}
SampleContext.Direct.cs

The plain MongoDB.Driver read-back helper, for the samples that show what MongoDB stored.

namespace ArxSamples.MongoDb;

/// <summary>The plain MongoDB driver outside the SDK, for the samples that show what MongoDB stored.</summary>
public sealed partial class SampleContext
{
    private MongoDirectHelper? _direct;

    /// <summary>The plain MongoDB driver outside the SDK, used only to reset the database and to show what was stored.</summary>
    public MongoDirectHelper Direct => _direct ??= new MongoDirectHelper(ConnectionString);
}
MongoDirectHelper.cs

Plain MongoDB.Driver, not part of ARX: reads rows back so a sample can show what MongoDB stored.

using MongoDB.Bson;
using MongoDB.Driver;

namespace ArxSamples.MongoDb;

/// <summary>
/// The plain MongoDB driver, outside the SDK: it resets the sample database from Schema/sample-schema.json and shows
/// what was stored. Samples never write through it.
/// </summary>
public sealed class MongoDirectHelper
{
    private readonly IMongoDatabase _database;

    public MongoDirectHelper(string connectionString)
    {
        var url = MongoUrl.Create(connectionString);
        _database = new MongoClient(url).GetDatabase(url.DatabaseName ?? "arx_samples");
    }

    /// <summary>
    /// The sample schema: Schema/sample-schema.json next to the program (the samples' own project), or sample-schema.json
    /// in the project folder (a sample copied from its page, as Get started says).
    /// </summary>
    private static string SchemaPath()
    {
        var candidates = new[]
        {
            Path.Combine(AppContext.BaseDirectory, "Schema", "sample-schema.json"),
            Path.Combine(Directory.GetCurrentDirectory(), "sample-schema.json"),
            Path.Combine(Directory.GetCurrentDirectory(), "Schema", "sample-schema.json")
        };

        return candidates.FirstOrDefault(File.Exists)
            ?? throw new FileNotFoundException("sample-schema.json isn't in the project folder: download it from the Get started page and save it next to Program.cs.");
    }

    /// <summary>Drops the sample database and makes its collections, indexes and validators again.</summary>
    public async Task ResetAsync()
    {
        var schema = BsonDocument.Parse(await File.ReadAllTextAsync(SchemaPath()));

        foreach (var collection in schema["collections"].AsBsonArray.Select(x => x.AsBsonDocument))
        {
            var name = collection["name"].AsString;

            // Only the collections the schema lists are dropped and made again: anything else in the database is left alone.
            await _database.DropCollectionAsync(name);
            var options = new CreateCollectionOptions<BsonDocument>();

            if (collection.TryGetValue("validator", out var validator))
            {
                options.Validator = validator.AsBsonDocument;
            }

            await _database.CreateCollectionAsync(name, options);

            foreach (var index in collection.GetValue("indexes", new BsonArray()).AsBsonArray.Select(x => x.AsBsonDocument))
            {
                var keys = index["keys"].AsBsonDocument;
                var indexOptions = new CreateIndexOptions<BsonDocument> { Unique = index.GetValue("unique", false).AsBoolean };

                if (index.GetValue("partial", false).AsBoolean)
                {
                    indexOptions.PartialFilterExpression = new BsonDocument(keys.Names.Select(x => new BsonElement(x, new BsonDocument("$exists", true))));
                }

                await _database.GetCollection<BsonDocument>(name).Indexes.CreateOneAsync(new CreateIndexModel<BsonDocument>(keys, indexOptions));
            }
        }
    }

    /// <summary>The stored documents a filter finds, as MongoDB's JSON, one line each.</summary>
    public async Task<IReadOnlyList<string>> FindAsync(string collection, string filterJson)
    {
        var documents = await _database.GetCollection<BsonDocument>(collection).Find(BsonDocument.Parse(filterJson)).ToListAsync();
        return documents.Select(x => x.ToJson()).ToList();
    }
}
SecurityGuard.cs

A security check around ARX calls: suspected injection logged as a security event, strikes, blocking.

using ARX.SDK;

namespace ArxSamples.MongoDb;

/// <summary>Where the names (tables, attributes, match columns) in a call come from. It decides what an ARX1017 means.</summary>
public enum NameSource
{
    /// <summary>A name built from what a user sent (a form field, a file header, a query string): ARX1017 is a suspected attack.</summary>
    UserInput,

    /// <summary>A name from your code or configuration: ARX1017 is a bug in your code, not an attack. Nobody is blocked.</summary>
    Code,
}

/// <summary>What the guard did with a call.</summary>
public enum GuardOutcome
{
    /// <summary>The call ran.</summary>
    Done,

    /// <summary>ARX refused the call with ARX1017; the event is logged and the user got a plain message.</summary>
    Refused,

    /// <summary>The user was already blocked: the call wasn't attempted.</summary>
    Blocked,
}

/// <summary>One entry of the security log: a suspected attack, or a bug when the names came from code.</summary>
public sealed record SecurityEvent(DateTimeOffset At, string User, string Action, string Kind, string Detail);

/// <summary>The outcome, the call's result when it ran, and the message to show the user (never the detection details).</summary>
public sealed record GuardResult<T>(GuardOutcome Outcome, T? Value, string ForTheUser);

/// <summary>
/// A security check around ARX calls: the pattern to copy into an application, kept small so it can be read in one
/// go. It lets every ARX error through except <c>ARX1017 SuspectedInjection</c>, which it turns into:
/// <list type="bullet">
/// <item>an entry in the security log (<see cref="Events"/>), with the user, the action and ARX's message (the name is
/// escaped and cut to 64 characters by ARX, so it is safe to log);</item>
/// <item>a plain message for the user that doesn't say what was detected;</item>
/// <item>a strike against the user. After <c>blockAfter</c> strikes the user is blocked and their calls aren't attempted.
/// One strike never blocks anyone, so a single false positive only costs one refused request;</item>
/// <item>no strike when the names came from your own code (<see cref="NameSource.Code"/>): that's a bug to fix, logged
/// as one.</item>
/// </list>
/// A false positive (a harmless name such as a file header <c>E-mail (work)</c> flagged as SQL) is handled by triage:
/// read the event, fix the mapping that let the name through, and <see cref="Unblock"/> the user if needed.
/// In a real application the log goes to your SIEM or security monitoring and the strikes live in a shared store
/// (Redis, the database), not in memory.
/// </summary>
public sealed class SecurityGuard
{
    private const string ATTACK = "suspected attack";
    private const string BUG = "bug";

    private readonly object _gate = new();
    private readonly int _blockAfter;
    private readonly Dictionary<string, int> _strikes = new(StringComparer.Ordinal);
    private readonly List<SecurityEvent> _events = new();

    public SecurityGuard(int blockAfter = 3)
    {
        _blockAfter = blockAfter;
    }

    /// <summary>The security log, oldest first.</summary>
    public IReadOnlyList<SecurityEvent> Events
    {
        get
        {
            lock (_gate)
            {
                return _events.ToArray();
            }
        }
    }

    public bool IsBlocked(string user)
    {
        lock (_gate)
        {
            return _strikes.TryGetValue(user, out var strikes) && strikes >= _blockAfter;
        }
    }

    /// <summary>After triage found the events were a false positive: clears the user's strikes (the log is kept).</summary>
    public void Unblock(string user)
    {
        lock (_gate)
        {
            _strikes.Remove(user);
        }
    }

    /// <summary>Runs <paramref name="call"/> for <paramref name="user"/> unless they are blocked, and handles ARX1017 as described above.</summary>
    public async Task<GuardResult<T>> RunAsync<T>(string user, string action, NameSource names, Func<Task<T>> call)
    {
        if (IsBlocked(user))
        {
            return new GuardResult<T>(GuardOutcome.Blocked, default, "Your access is suspended. Contact support.");
        }

        try
        {
            return new GuardResult<T>(GuardOutcome.Done, await call(), "Done.");
        }
        catch (ArxValidationException error) when (error.Code == ArxErrorCode.SuspectedInjection)
        {
            lock (_gate)
            {
                if (names == NameSource.Code)
                {
                    _events.Add(new SecurityEvent(DateTimeOffset.UtcNow, user, action, BUG, error.Message));
                    return new GuardResult<T>(GuardOutcome.Refused, default, "Something went wrong on our side. It has been reported.");
                }

                _strikes[user] = _strikes.GetValueOrDefault(user) + 1;
                _events.Add(new SecurityEvent(DateTimeOffset.UtcNow, user, action, ATTACK, error.Message));
            }

            return new GuardResult<T>(GuardOutcome.Refused, default, "The request couldn't be processed.");
        }
    }
}

Expected output

What dotnet run printed when the sample was verified, on a freshly created sample schema. Ids differ when the tables already hold rows.

Output
  > 1. A normal request passes through
      ana: Done, contact 6af533afbd188482979011b2
  > 2. An attack is logged; the user sees a plain message
      mallory sees: "The request couldn't be processed."
      security log: suspected attack, mallory, create contact
  > 3. Repeated attacks block the user
      mallory blocked: True; a valid request: Blocked, "Your access is suspended. Contact support."
  > 4. A false positive: a harmless file header flagged as SQL
      sam: Refused; blocked: False (one strike blocks nobody)
      triage reads: ARX1017 SuspectedInjection: Attribute name 'E-mail (work)' looks like an SQL injection attempt (contains '('). ARX refused it before sending anything to the database. Names must come from your code or a fixed list, never from user input.
      after the fix: Done, contact 6af533b0bd188482979011b3