A security check around deletes: probing, blocking, unblocking a false positive

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.Delete.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);

The sample, as it is in the repository.

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

namespace ArxSamples.MongoDb.Operations.Delete.AdvancedCases;

/// <summary>
/// A security check around deletes (<see cref="SecurityGuard"/>, in the project root), in four scenarios: a delete
/// through an allowed list of types passes; an endpoint that forgot the list is probed, and the attacker is blocked
/// after three attempts; and a false positive, an old client sending a harmless type name, is triaged: support unblocks
/// the user and the endpoint maps the old name.
/// </summary>
public sealed class Advanced2_GuardAgainstAttacks : ISample
{
    public string Name => "delete-advanced-2-guard-against-attacks";

    public string Title => "A security check around deletes: probing, blocking, unblocking a false positive";

    /// <summary>The types <c>DELETE /api/{type}/{id}</c> accepts, and their tables.</summary>
    private static readonly Dictionary<string, string> Types = new(StringComparer.OrdinalIgnoreCase)
    {
        ["accounts"] = "account",
        ["contacts"] = "contact",
    };

    public async Task RunAsync(SampleContext dbContext)
    {
        var guard = new SecurityGuard(blockAfter: 3);

        var accounts = new List<ArxEntity>();

        foreach (var number in new[] { "ACC-G301", "ACC-G302", "ACC-G303", "ACC-G304" })
        {
            var account = new ArxEntity("account");
            account["name"] = $"Guarded delete {number}";
            account["account_number"] = number;
            accounts.Add(account);
        }

        var ids = await dbContext.Service.CreateMultipleAsync(accounts);

        await Scenario1_ADeleteThroughTheAllowedTypesAsync(dbContext, guard, ids[0]);

        await Scenario2_ProbingIsBlockedAsync(dbContext, guard, ids[1]);

        await Scenario3_AFalsePositiveIsTriagedAsync(dbContext, guard, ids[2], ids[3]);
    }

    /// <summary>The well-written endpoint: the type is looked up in <see cref="Types"/>, so user input never becomes a table name.</summary>
    private static Task<GuardResult<bool>> DeleteThroughTheListAsync(SampleContext dbContext, SecurityGuard guard, string user, string type, ArxRecordId id)
        => Types.TryGetValue(type, out var table)
            ? guard.RunAsync(user, $"delete {type}", NameSource.Code, () => dbContext.Service.DeleteAsync(table, id))
            : Task.FromResult(new GuardResult<bool>(GuardOutcome.Refused, false, $"404 Not Found: no type '{type}'."));

    /// <summary>A legacy endpoint that uses the type as the table name directly: the names come from the user.</summary>
    private static Task<GuardResult<bool>> DeleteDirectlyAsync(SampleContext dbContext, SecurityGuard guard, string user, string type, ArxRecordId id)
        => guard.RunAsync(user, $"delete {type}", NameSource.UserInput, () => dbContext.Service.DeleteAsync(type, id));

    /// <summary>1. <c>DELETE /api/accounts/{id}</c> through the allowed types: the account is deleted.</summary>
    private static async Task Scenario1_ADeleteThroughTheAllowedTypesAsync(SampleContext dbContext, SecurityGuard guard, ArxRecordId id)
    {
        Step("1. A delete through the allowed types");

        var result = await DeleteThroughTheListAsync(dbContext, guard, "ana", "accounts", id);
        var unknown = await DeleteThroughTheListAsync(dbContext, guard, "ana", "account; DROP TABLE account", id);

        Show($"ana: {result.Outcome}, deleted = {result.Value}; an unknown type: \"{unknown.ForTheUser}\" (never reaches ARX)");

        /* ===========================================================
         * [DIRECT QUERY CHECK]
         * -----------------------------------------------------------
         * db.account.countDocuments({ account_number: "ACC-G301" })
         * -- Expect: 0: deleted.
         * ===========================================================
         */
    }

    /// <summary>2. The legacy endpoint is probed three times: each probe is refused by ARX (ARX1017) and logged; then the user is blocked.</summary>
    private static async Task Scenario2_ProbingIsBlockedAsync(SampleContext dbContext, SecurityGuard guard, ArxRecordId id)
    {
        Step("2. Probing the legacy endpoint gets the user blocked");

        foreach (var probe in new[] { "account WHERE 1=1; --", "account\"; TRUNCATE account; --", "account UNION SELECT 1" })
        {
            await DeleteDirectlyAsync(dbContext, guard, "mallory", probe, id);
        }

        var valid = await DeleteDirectlyAsync(dbContext, guard, "mallory", "account", id);

        Show($"mallory: {guard.Events.Count(x => x.User == "mallory")} events logged, blocked: {guard.IsBlocked("mallory")}; a valid delete: {valid.Outcome}");

        /* ===========================================================
         * [DIRECT QUERY CHECK]
         * -----------------------------------------------------------
         * db.account.countDocuments({ account_number: "ACC-G302" })
         * -- Expect: 1: nothing was deleted, not even by the valid request after the block.
         * ===========================================================
         */
    }

    /// <summary>
    /// 3. A false positive: an old mobile client sends the type as "account (archived)". The parentheses make ARX flag it
    /// as SQL, the client retries, and after three strikes the user is blocked. Support reads the log, sees the same
    /// harmless name three times, unblocks the user, and the endpoint maps the old name to its table.
    /// </summary>
    private static async Task Scenario3_AFalsePositiveIsTriagedAsync(SampleContext dbContext, SecurityGuard guard, ArxRecordId first, ArxRecordId second)
    {
        Step("3. A false positive is triaged: unblock, then map the old name");

        for (var retry = 0; retry < 3; retry++)
        {
            await DeleteDirectlyAsync(dbContext, guard, "sam", "account (archived)", first);
        }

        Show($"sam blocked: {guard.IsBlocked("sam")}; the log shows: {string.Join(" | ", guard.Events.Where(x => x.User == "sam").Select(x => x.Action).Distinct())}");

        // Triage: the same harmless name three times, from a known old client. Unblock, and map the old name.
        guard.Unblock("sam");
        Types["account (archived)"] = "account";

        var afterTheFix = await DeleteThroughTheListAsync(dbContext, guard, "sam", "account (archived)", second);

        Show($"after the fix: {afterTheFix.Outcome}, deleted = {afterTheFix.Value}");

        /* ===========================================================
         * [DIRECT QUERY CHECK]
         * -----------------------------------------------------------
         * db.account.find({ account_number: { $in: ["ACC-G303", "ACC-G304"] } }, { account_number: 1 })
         * -- Expect: 1 document: ACC-G303 (the blocked attempts deleted nothing); ACC-G304 was deleted after the fix.
         * ===========================================================
         */
    }
}

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; }
}

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}");
}

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);
}

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();
}

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);
}

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();
    }
}

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.");
        }
    }
}

  > 1. A delete through the allowed types
      ana: Done, deleted = True; an unknown type: "404 Not Found: no type 'account; DROP TABLE account'." (never reaches ARX)
  > 2. Probing the legacy endpoint gets the user blocked
      mallory: 3 events logged, blocked: True; a valid delete: Blocked
  > 3. A false positive is triaged: unblock, then map the old name
      sam blocked: True; the log shows: delete account (archived)
      after the fix: Done, deleted = True