@forge/isolated-vm
v4.1.2
Published
Access to multiple isolates
Downloads
349
Keywords
Readme
isolated-vm -- Access to multiple isolates in nodejs
isolated-vm
is a library for nodejs which gives you access to v8's Isolate
interface. This
allows you to create JavaScript environments which are completely isolated from each other. You
might find this module useful if you need to run some untrusted code in a secure way. You may also
find this module useful if you need to run some JavaScript simultaneously in multiple threads. You
may find this project very useful if you need to do both at the same time!
REQUIREMENTS
This project requires nodejs LTS version 10.4.0 (or later).
Furthermore, to install this module you will need a compiler installed. If you run into errors while
running npm install isolated-vm
it is likely you don't have a compiler set up, or your compiler is
too old.
- Windows + OS X users should follow the instuctions here: node-gyp
- Ubuntu users should run:
sudo apt-get install python g++ build-essential
- Alpine users should run:
sudo apk add python make g++
- Amazon Linux AMI users should run:
sudo yum install gcc72 gcc72-c++
- Arch Linux users should run:
sudo pacman -S make gcc python
WHO IS USING ISOLATED-VM
Screeps - Screeps is an online JavaScript-based MMO+RPG game. They are using isolated-vm to run arbitrary player-supplied code in secure enviroments which can persistent for several days at a time.
Fly - Fly is a programmable CDN which hosts dynamic endpoints as opposed to just static resources. They are using isolated-vm to run globally distributed applications, where each application may have wildly different traffic patterns.
Algolia - Algolia is a Search as a Service provider. They use
isolated-vm
to power their Custom Crawler product, which allows them to safely execute user-provided code for content extraction.Tripadvisor - Tripadvisor is the world’s largest travel platform. They use
isolated-vm
to server-side render thousands of React pages per second.
API DOCUMENTATION
Since isolates share no resources with each other, most of this API is built to provide primitives
which make marshalling data between many isolates quick and easy. The only way to pass data from one
isolate to another is to first make that data transferable. Primitives (except for Symbol
) are
always transferable. This means if you invoke a function in a different isolate with a number or
string as the argument, it will work fine. If you need to pass more complex information you will
have to first make the data transferable with one of the methods here.
Most methods will provide both a synchronous and an asynchronous version. Calling the synchronous functions will block your thread while the method runs and eventually returns a value. The asynchronous functions will return a Promise while the work runs in a separate thread pool.
There are some rules about which functions may be called from certain contexts:
- Asynchronous functions may be called at any time
- Synchronous functions usually may not be called from an asynchronous function
- You may call a synchronous function from an asynchronous function as long as that function belongs to current isolate
- You may call a synchronous function belonging to the default nodejs isolate at any time
Additionally, some methods will provide an "ignored" version which runs asynchronously but returns no promise. This can be a good option when the calling isolate would ignore the promise anyway, since the ignored versions can skip an extra thread synchronization. Just be careful because this swallows any thrown exceptions which might make problems hard to track down.
It's also worth noting that all asynchronous invocations will run in the order they were queued,
regardless of whether or not you wait on them. So, for instance, you could call several "ignored"
methods in a row and then await
on a final async method to observe some side-effect of the
ignored methods.
Class: Isolate
[transferable]
This is the main reference to an isolate. Every handle to an isolate is transferable, which means
you can give isolates references to each other. An isolate will remain valid as long as someone
holds a handle to the isolate or anything created inside that isolate. Once an isolate is lost the
garbage collector should eventually find it and clean up its memory. Since an isolate and all it
contains can represent quite a large chunk of memory though you may want to explicitly call the
dispose()
method on isolates that you are finished with to get that memory back immediately.
new ivm.Isolate(options)
options
[object]memoryLimit
[number] - Memory limit that this isolate may use, in MB. Note that this is more of a guideline instead of a strict limit. A determined attacker could use 2-3 times this limit before their script is terminated. Against non-hostile code this limit should be pretty close. The default is 128MB and the mimium is 8MB.inspector
[boolean] - Enable v8 inspector support in this isolate. Seeinspector-example.js
in this repository for an example of how to use this.snapshot
[ExternalCopy[ArrayBuffer]] - This is an optional snapshot created fromcreateSnapshot
which will be used to initialize the heap of this isolate.
ivm.Isolate.createSnapshot(scripts, warmup_script)
scripts
[array]code
[string] - Source code to set up this snapshot{ ...ScriptOrigin }
warmup_script
[string] - Optional script to "warmup" the snapshot by triggering code compilation
Isolate snapshots are a very useful feature if you intend to create several isolates running common libraries between them. A snapshot serializes the entire v8 heap including parsed code, global variables, and compiled code. Check out the examples section for tips on using this.
Note: createSnapshot
does not provide the same isolate protection like the rest of
isolated-vm. If the script passed to createSnapshot
uses too much memory the process will crash,
and if it has an infinite loop it will stall the process. Furthermore newer v8 features may simply
fail when attempting to take a snapshot that uses them. It is best to snapshot code that only
defines functions, class, and simple data structures.
**Please note that versions of nodejs 10.4.0 - 10.9.0 may crash while using the snapshot feature.
isolate.compileScript(code)
Promise
isolate.compileScriptSync(code)
code
[string] - The JavaScript code to compile.options
[object]return A
Script
object.
Note that a Script
can only run in the isolate which created it.
isolate.compileModule(code)
Promise
isolate.compileModuleSync(code)
code
[string] - The JavaScript code to compile.options
[object]return A
Module
object.
Note that a Module
can only run in the isolate which created it.
isolate.createContext()
Promise
isolate.createContextSync()
options
[object]inspector
[boolean] - Enable the v8 inspector for this context. The inspector must have been enabled for the isolate as well.
return A
Context
object.
isolate.dispose()
Destroys this isolate and invalidates all references obtained from it.
isolate.getHeapStatistics()
Promise
isolate.getHeapStatisticsSync()
- return [object]
Returns heap statistics from v8. The return value is almost identical to the nodejs function
v8.getHeapStatistics().
This function returns one additional property: externally_allocated_size
which is the total amount
of currently allocated memory which is not included in the v8 heap but counts against this isolate's
memoryLimit
. ArrayBuffer instances over a certain size are externally allocated and will be
counted here.
isolate.cpuTime
[Array]
isolate.wallTime
[Array]
The total CPU and wall time spent in this isolate. CPU time is the amount of time the isolate has spent actively doing work on the CPU. Wall time is the amount of time the isolate has been running, including passive time spent waiting (think "wall" like a clock on the wall). For instance, if an isolate makes a call into another isolate, wall time will continue increasing while CPU time will remain the same.
The return format is [ seconds, nanoseconds ]
, which is the same as the nodejs method
process.hrtime
. To convert this
value to milliseconds you could do something like: (ret[0] + ret[1] / 1e9) * 1000
. Some precision
is lost in this conversion but for most applications it's probably not a big deal.
Note that CPU time may vary drastically if there is contention for the CPU. This could occur if
other processes are trying to do work, or if you have more than require('os').cpus().length
isolates currently doing work in the same nodejs process.
isolate.isDisposed
[boolean]
Flag that indicates whether this isolate has been disposed.
isolate.referenceCount
[number]
Returns the total count of active Reference
instances that belong to this isolate. Note that in
certain cases many Reference
instances in JavaScript will point to the same underlying reference
handle, in which case this number will only reflect the underlying reference handle. This happens
when you transfer a Reference
instance via some method which accepts transferable values. This
will also include underlying reference handles created by isolated-vm like Script
or Context
objects.
Class: Context
[transferable]
A context is a sandboxed execution environment within an isolate. Each context contains its own built-in objects and global space.
context.global
Reference
Reference
to this context's global object. Note that if you call
context.release()
the global reference will be released as well.
context.eval(code, options)
Promise
context.evalIgnored(code, options)
context.evalSync(code, options)
code
[string] - The code to runoptions
[object]timeout
[number] - Maximum amount of time in milliseconds this script is allowed to run before execution is canceled. Default is no timeout.{ ...CachedDataOptions }
{ ...ScriptOrigin }
{ ...TransferOptions }
- return [object]
result
[transferable]
Compiles and executes a script within a context. This will return the last value evaluated, as long
as that value was transferable, otherwise undefined
will be returned.
context.evalClosure(code, arguments, options)
Promise
context.evalClosureIgnored(code, arguments, options)
context.evalClosureSync(code, arguments, options)
code
[string] - The code to runarguments
*[array]` - Arguments to pass to this codeoptions
[object]timeout
[number] - Maximum amount of time in milliseconds this script is allowed to run before execution is canceled. Default is no timeout.{ ...CachedDataOptions }
{ ...ScriptOrigin }
arguments
[object]result
[object]
- return [object]
result
[transferable]
Compiles and runs code as if it were inside a function, similar to the seldom-used new
Function(code)
constructor. You can pass arguments to the function and they will be available as
$0
, $1
, and so on. You can also use return
from the code.
context.release()
Releases this reference to the context. You can call this to free up v8 resources immediately, or you can let the garbage collector handle it when it feels like it. Note that if there are other references to this context it will not be disposed. This only affects this reference to the context.
Class: Script
[transferable]
A script is a compiled chunk of JavaScript which can be executed in any context within a single isolate.
script.release()
Releases the reference to this script, allowing the script data to be garbage collected. Functions
and data created in the isolate by previous invocations to script.run(...)
will still be alive in
their respective contexts-- this only means that you can't invoke script.run(...)
again with this
reference.
script.run(context, options)
Promise
script.runIgnored(context, options)
script.runSync(context, options)
context
Context
- The context in which this script will run.options
[object]release
[boolean] - If truerelease()
will automatically be called on this instance.timeout
[number] - Maximum amount of time in milliseconds this script is allowed to run before execution is canceled. Default is no timeout.{ ...TransferOptions }
- return [transferable]
Runs a given script within a context. This will return the last value evaluated in a given script,
as long as that value was transferable, otherwise undefined
will be returned. For instance if your
script was "let foo = 1; let bar = 2; bar = foo + bar" then the return value will be 3 because that
is the last expression.
Class: Module
[transferable]
A JavaScript module. Note that a Module
can only run in the isolate which created it.
module.dependencySpecifiers
A read-only array of all dependency specifiers the module has.
const code = `import something from './something';`;
const module = await isolate.compileModule(code);
const dependencySpecifiers = module.dependencySpecifiers;
// dependencySpecifiers => ["./something"];
module.namespace
Returns a Reference
containing all exported values.
module.instantiate(context, resolveCallback)
Promise
module.instantiateSync(context, resolveCallback)
context
Context
- The context the module should use.resolveCallback
- This callback is responsible for resolving all direct and indirect dependencies of this module. It accepts two parameters:specifier
andreferrer
. It must return aModule
instance which will be used to satisfy the dependency. The asynchronous version ofinstantiate
may return a promise fromresolveCallback
.
Instantiate the module together with all its dependencies. Calling this more than once on a single module will have no effect.
module.evaluate(options)
Promise
module.evaluateSync(options)
options
[object] - Optional.timeout
[number] - Maximum amount of time in milliseconds this module is allowed to run before execution is canceled. Default is no timeout.
- return [transferable]
Evaluate the module and return the last expression (same as script.run). If evaluate
is called
more than once on the same module the return value from the first invocation will be returned (or
thrown).
Note: nodejs v14.8.0 enabled top-level await by default which has the effect of breaking the return value of this function.
Class: Reference
[transferable]
A instance of Reference
is a pointer to a value stored in any isolate.
reference.typeof
[string]
This is the typeof the referenced value, and is available at any time from any isolate. Note that
this differs from the real typeof
operator in that null
is "null", and Symbols are "object".
reference.copy()
Promise
reference.copySync()
- return JavaScript value of the reference.
Creates a copy of the referenced value and internalizes it into this isolate. This uses the same
copy rules as ExternalCopy
.
reference.deref()
options
[object]release
[boolean] - If truerelease()
will automatically be called on this instance.
- return The value referenced by this handle.
Will attempt to return the actual value or object pointed to by this reference. Note that in order to call this function the reference must be owned by the current isolate, otherwise an error will be thrown.
reference.derefInto()
options
[object]release
[boolean] - If truerelease()
will automatically be called on this instance.
- return [transferable]
Returns an object, which when passed to another isolate will cause that isolate to dereference the handle.
reference.release()
Releases this reference. If you're passing around a lot of references between isolates it's wise to release the references when you are done. Otherwise you may run into issues with isolates running out of memory because other isolates haven't garbage collected recently. After calling this method all attempts to access the reference will throw an error.
reference.delete(property)
Promise
reference.deleteIgnored(property)
reference.deleteSync(property)
property
[transferable] - The property to access on this object.- return
true
orfalse
Delete a property from this reference, as if using delete reference[property]
reference.get(property, options)
Promise
reference.getSync(property, options)
property
[transferable] - The property to access on this object.options
[object]- return A
Reference
object.
Will access a reference as if using reference[property]
and transfer the value out.
reference.set(property, value, options)
Promise
reference.setIgnored(property, value, options)
reference.setSync(property, value, options)
property
[transferable] - The property to set on this object.value
[transferable] - The value to set on this object.options
[object]- return
true
orfalse
Returns a boolean indicating whether or not this operation succeeded. I'm actually not really sure
when false
would be returned, I'm just giving you the result back straight from the v8 API.
reference.apply(receiver, arguments, options)
Promise
reference.applyIgnored(receiver, arguments, options)
reference.applySync(receiver, arguments, options)
reference.applySyncPromise(receiver, arguments, options)
receiver
[transferable] - The value which will bethis
.arguments
[array] - Array of transferables which will be passed to the function.options
[object]timeout
[number] - Maximum amount of time in milliseconds this function is allowed to run before execution is canceled. Default is no timeout.arguments
[object]result
[object]
- return [transferable]
Will attempt to invoke an object as if it were a function. If the return value is transferable it
will be returned to the caller of apply
, otherwise it will return an instance of Reference
. This
behavior can be changed with the result
options.
applySyncPromise
is a special version of applySync
which may only be invoked on functions
belonging to the default isolate AND may only be invoked from a non-default thread. Functions
invoked in this way may return a promise and the invoking isolate will wait for that promise to
resolve before resuming execution. You can use this to implement functions like readFileSync
in a
way that doesn't block the default isolate. Note that the invoking isolate will not respond to any
async functions until this promise is resolved, however synchronous functions will still function
correctly. Misuse of this feature may result in deadlocked isolates, though the default isolate
will never be at risk of a deadlock.
Class: ExternalCopy
[transferable]
Instances of this class represent some value that is stored outside of any v8 isolate. This value can then be quickly copied into any isolate without any extra thread synchronization.
new ivm.ExternalCopy(value, options)
value
- The value to copy.options
[object]transferList
[boolean] - An array ofArrayBuffer
instances to transfer ownership. This behaves in a similar way topostMessage
.transferOut
[boolean] - If true this will release ownership of the given resource from this isolate. This operation completes in constant time since it doesn't have to copy an arbitrarily large object. This only applies to ArrayBuffer and TypedArray instances.
Primitive values can be copied exactly as they are. Date objects will be copied as as Dates. ArrayBuffers, TypedArrays, and DataViews will be copied in an efficient format. SharedArrayBuffers will simply copy a reference to the existing memory and when copied into another isolate the new SharedArrayBuffer will point to the same underlying data. After passing a SharedArrayBuffer to ExternalCopy for the first time isolated-vm will take over management of the underlying memory block, so a "copied" SharedArrayBuffer can outlive the isolate that created the memory originally.
All other objects will be copied in seralized form using the structured clone algorithm.
ExternalCopy
can copy objects with deeply nested transferable objects. For example:
let isolate = new ivm.Isolate;
let context = isolate.createContextSync();
let global = context.global;
let data = new ExternalCopy({ isolate, context, global });
ExternalCopy.totalExternalSize
[number]
This is a static property which will return the total number of bytes that isolated-vm has allocated
outside of v8 due to instances of ExternalCopy
.
externalCopy.copy(options)
options
[object]release
[boolean] - If truerelease()
will automatically be called on this instance.transferIn
[boolean] - If true this will transfer the resource directly into this isolate, invalidating the ExternalCopy handle.
- return - JavaScript value of the external copy.
Internalizes the ExternalCopy data into this isolate.
externalCopy.copyInto(options)
options
[object]release
[boolean] - If truerelease()
will automatically be called on this instance.transferIn
[boolean] - If true this will transfer the resource directly into this isolate, invalidating the ExternalCopy handle.
- return [transferable]
Returns an object, which when passed to another isolate will cause that isolate to internalize a copy of this value.
externalCopy.release()
Releases the reference to this copy. If there are other references to this copy elsewhere the copy will still remain in memory, but this handle will no longer be active. Disposing ExternalCopy instances isn't super important, v8 is a lot better at cleaning these up automatically because there's no inter-isolate dependencies.
Shared Options
Many methods in this library accept common options between them. They are documented here instead of being colocated with each instance.
CachedDataOptions
cachedData
[ExternalCopy[ArrayBuffer]] - This will consume cached compilation data from a previous call to this function.cachedDataRejected
will be set totrue
if the supplied data was rejected by V8.produceCachedData
[boolean] - Produce V8 cache data. Similar to the VM.Script option of the same name. If this is true then the returned object will havecachedData
set to an ExternalCopy handle. Note that this differs from the VM.Script option slightly in thatcachedDataProduced
is never set.
Most functions which compile or run code can produce and consume cached data. You can produce cached
data and use the data in later invocations to drastically speed up parsing of the same script. You
can even save this data to disk and use it in a different process. You can set both cachedData
and
produceCachedData
, in which case new cached data will only be produced if the data supplied was
invalid.
ScriptOrigin
filename
[string] - Filename of this source codecolumnOffset
[number] - Column offset of this source codelineOffset
[number] - Line offset of this source code
You may optionally specify information on compiled code's filename. This is used in various
debugging contexts within v8, including stack traces and the inspector. It is recommended to use a
valid URI scheme, for example: { filename: 'file:///test.js' }
, otherwise some devtools may
malfunction.
TransferOptions
copy
[boolean] - Automatically deep copy valueexternalCopy
[boolean] - Automatically wrap value inExternalCopy
instancereference
[boolean] - Automatically wrap value inReference
instancepromise
[boolean] - Automatically proxy any returned promises between isolates. This can be used in combination with the other transfer options.
Any function which moves data between isolates will accept these transfer options. By default only [transferable] values may pass between isolates. Without specifying one of these options the function may ignore the value, throw, or wrap it in a reference depending on the context.
More advanced situations like transferring ownership of ArrayBuffer
instances will require direct
use of ExternalCopy
or
Reference
.
EXAMPLES
Below is a sample program which shows basic usage of the library.
// Create a new isolate limited to 128MB
const ivm = require('isolated-vm');
const isolate = new ivm.Isolate({ memoryLimit: 128 });
// Create a new context within this isolate. Each context has its own copy of all the builtin
// Objects. So for instance if one context does Object.prototype.foo = 1 this would not affect any
// other contexts.
const context = isolate.createContextSync();
// Get a Reference{} to the global object within the context.
const jail = context.global;
// This make the global object available in the context as `global`. We use `derefInto()` here
// because otherwise `global` would actually be a Reference{} object in the new isolate.
jail.setSync('global', jail.derefInto());
// We will create a basic `log` function for the new isolate to use.
const logCallback = function(...args) {
console.log(...args);
};
context.evalClosureSync(`global.log = function(...args) {
$0.applyIgnored(undefined, args, { arguments: { copy: true } });
}`, [ logCallback ], { arguments: { reference: true } });
// And let's test it out:
context.evalSync('log("hello world")');
// > hello world
// Let's see what happens when we try to blow the isolate's memory
const hostile = isolate.compileScriptSync(`
const storage = [];
const twoMegabytes = 1024 * 1024 * 2;
while (true) {
const array = new Uint8Array(twoMegabytes);
for (let ii = 0; ii < twoMegabytes; ii += 4096) {
array[ii] = 1; // we have to put something in the array to flush to real memory
}
storage.push(array);
log('I\\'ve wasted '+ (storage.length * 2)+ 'MB');
}
`);
// Using the async version of `run` so that calls to `log` will get to the main node isolate
hostile.run(context).catch(err => console.error(err));
// I've wasted 2MB
// I've wasted 4MB
// ...
// I've wasted 130MB
// I've wasted 132MB
// RangeError: Array buffer allocation failed
Another example which shows how calls to the asynchronous methods will execute in separate threads providing you with parallelism. Note that each isolate only "owns" a thread while it is executing. So you could have hundreds of isolates sitting idle and they would not be using a thread.
// A simple function to sum a range of numbers. This can also be expressed as:
// (max * (max - 1) - min * (min - 1)) / 2
// But this is an easy way to show off the async features of the module.
function sum(min, max) {
let sum = 0;
for (let ii = min; ii < max; ++ii) {
sum += ii;
}
return sum;
}
// I chose this number because it's big but also small enough that we don't go past JS's integer
// limit.
let num = Math.pow(2, 27);
// First we execute a single thread run
let start1 = new Date;
let result = sum(0, num);
console.log('Calculated '+ result+ ' in '+ (Date.now() - start1)+ 'ms');
// Now we do the same thing over 4 threads
let start2 = new Date;
let ivm = require('isolated-vm');
let numThreads = 4;
let promises = Array(numThreads).fill().map(async function(_, ii) {
// Set up 4 isolates with the `sum` function from above
let isolate = new ivm.Isolate();
let context = await isolate.createContext();
let script = await isolate.compileScript(sum+ '');
await script.run(context);
let fnReference = await context.global.get('sum');
// Run one slice of the sum loop
let min = Math.floor(num / numThreads * ii);
let max = Math.floor(num / numThreads * (ii + 1));
return await fnReference.apply(undefined, [ min, max ]);
});
Promise.all(promises).then(function(sums) {
let result = sums.reduce((a, b) => a + b, 0);
console.log('Calculated '+ result+ ' in '+ (Date.now() - start2)+ 'ms');
});
// They get the same answer but the async version can do it much faster! Even
// with the overhead of building 4 isolates
// > Calculated 9007199187632128 in 1485ms
// > Calculated 9007199187632128 in 439ms
Included in the repository is an example of how you can write quicksort using a SharedArrayBuffer to sort over multiple threads. See: parallel-sort-example.js.
ALTERNATIVES
The primary goal of isolated-vm is to create a powerful and secure environment for running untrusted JavaScript code. isolated-vm is also a good way to build single-process multithreaded JavaScript applications, though if parallelism is your only goal then there are probably better options out there.
Below is a quick summary of some other options available on nodejs and how they differ from isolated-vm. The table headers are defined as follows:
- Secure: Safely run untrusted code
- Memory Limits: Possible to set memory limits / safe against heap overflow DoS attacks
- Isolated: Is garbage collection, heap, etc isolated from application
- Multithreaded: Run code on many threads from a single process
- Module Support: Is
require
supported out of the box - Inspector Support: Chrome DevTools supported
| Module | Secure | Memory Limits | Isolated | Multithreaded | Module Support | Inspector Support | | ---------------------------------------------------------------------------- | :----: | :-----------: | :------: | :-----------: | :------------: | :---------------: | | vm | | | | | ✅ | ✅ | | worker_threads | | | ✅ | ✅ | ✅ | | | vm2 | ✅ | | | | ✅ | ✅ | | napajs | | | ✅ | ✅ | Partial | | | webworker-threads | | | ✅ | ✅ | | | | tiny-worker | | | ✅ | | ✅ | | | isolated-vm | ✅ | ✅ | ✅ | ✅ | | ✅ |