Properties of Interpreters or the Browser Environment that allow Privilege Escalation

Below is a list of known attack vectors. We discuss the EcmaScript 3 language, quirks of existing interpreters, and browser specific extensions that could allow privilege escalation so that we can come up with tests for a safe JavaScript rewriter or verifier.

Attack Vectors at the EcmaScript/JavaScript level

• GlobalObjectPoisoning -- Global object poisoning
• EvalArbitraryCodeExecution -- eval and the Function constructor allow arbitrary code execution
• ArgumentsMaskedByVar -- function arguments array masked by var arguments on Opera
• CrossScopeParameterModification -- arguments array allows modification of parameters
• ArgumentsExposesCaller -- arguments Array and function object expose caller
• FunctionMemberCrossScopeParameterAccess -- function object's arguments array expose arguments while call in progress
• TypeofInconsistent -- typeof inconsistent for regular expressions
• InaccessibleLocalVariables -- Inaccessible local variables
• CatchBlocksScopeBleed -- catch blocks may cause global assignment, or local scope creep
• GlobalScopeViaThis -- Global scope reachable via this from functions not invoked as methods
• DeleteUnmasksGlobals -- Delete can unmask globals
• FunctionConstructor -- Function constructor accessible via the 'constructor' property
• ObjectEvalArbitraryCodeExecution -- Object.eval allows execution of unsanitized code on Firefox.
• ObjectWatch -- Object.watch allows stealing and poisoning of otherwise restricted data
• ObjectToSourceLeaksPrivates -- Object.toSource and uneval allow access to private fields
• FunctionMethodsLeakGlobalScope -- Function.call or Function.apply can leak window with certain this-values.
• ConditionalCompilationComments -- Conditional compilation may allow disabling of runtime checks.
• StringObfuscationIsEasy -- Approaches that rely on detecting code for other languages in string literals is easy to defeat
• ParentCircumventsScoping -- The javascript1.2 feature __parent__ circumvents normal scoping.
• JsControlFormatChars -- [:Cf:] can be used hide code in string or comments.
• InconsistentlyReservedKeywords -- Different reserved keyword set can cause parser ambiguity
• ErrorExposesParameterValues -- The stack property of Error includes parameter values.
• HiddenControlFlowHazard -- Seemingly safe Caja data computations may result in a control-flow transfer to a potential adversary.
• RegexpsLeakMatchGlobally -- Any regular expression can match against the last string passed to any other
• EvalBreaksClosureEncapsulation -- Eval extensions allow reaching into the scope chain of closures
• PostIncrementAndDecrementCanReturnNonNumber -- Incorrect implementations of postincrement and postdecrement can cause confusion as to which property is being accessed
• MisOptimizations -- Some interpreters try to optimize javascript before execution subtly changing the semantics of builtin operators (PostIncrementAndDecrementCanReturnNonNumber is a specific example)
• CompoundAssignmentsCanReturnNonNumber -- The type of assignment expressions may not be correct.
• FinallySkipped -- An exception that is thrown not inside a try/catch caught skips finally blocks.

Attack Vectors at the Browser Environment, DOM, HTML, or CSS levels

• ScriptInHtml -- HTML Tags in Javascript Strings can allow Unsanitized Script Execution
• SetTimeoutArbitraryCodeExecution -- setTimeout & setInterval allow arbitrary code execution
• DomNodeAllowArbitraryCodeExecution -- ActiveXObject, document.createElement, document allow arbitrary code execution
• InnerHtmlYieldsCdata -- script, style, xmp and listing elements' innerHTML cannot be safely inserted into another element's innerHTML
• DomAllowsXsrf -- document object allows arbitrary XSRF with the user's credentials
• DomAllowsKeylogging -- DOM access allows keylogging
• XsrfViaXxe -- XMLHttpRequest and DOMParser parsing allow arbitrary XSRF via XXE
• CssAllowsArbitraryCodeExecution -- Some CSS properties allows execution of unsanitized javascript?
• CssImportsAllowUnsanitizedCodeExecution -- @import can import unsanitized CSS which can execute unsanitized javascript
• NullCharEscapes -- Null characters in URL can disguise protocols such as javascript:
• ConfusedHtmlParsers -- Differences in the way HTML parsers parse malformed HTML can hide unsanitized scripts
• EventHandlersEvalWithDom -- The scope that event handlers are executed in may expose DOM properties as globals
• DocTypesCanInjectUnsanitizedContent -- DOCTYPEs can define entities which can inject unsanitized script or markup.
• EventChecksCircumventableByInfLoops -- Invariants preserved by event handlers can be circumvented by causing the browser to turn off javascript.
• IdAndNameMasking -- Members of HtmlCollection, HTMLFormElement, etc. masked by ids&names
• UrlFetchingSideChannel -- Side-channels from unproxied connections leak information across closed networks
• HistoryMining -- CSS can be used to determine whether a user has visited a URL.
• RedirectWithoutUserAction -- JS and HTML both allow redirection with user interaction.
• PhishingViaCrossSiteHttpAuth -- An attacker can display an HTTP authorization dialog that looks like it may have come from another site.