RCE Vulnerability in React and Next.js

A similar bug could be introduced in the implementation of other RPC systems too. It's not entirely specific to this design.

(I contribute to React but not really on RSC.)

Building a private, out of date repo doesn't seem great either.

If I had a dollar for every time a serious vulnerability that started like this was discovered in the last 30 years...

> My guess is this won't be the last time we see security fallout from that design choice. Not because React is sloppy, but because it’s trying to solve a problem category that traditionally requires explicitness, not magic.

Now I'm worried, but I don't use React. So I will have to ask: how does SvelteKit fares in this aspect?

The vast majority of developers do not update their frameworks to the latest version so this is something that will linger on for years. Particularly if you're on Next something-like-12 and there's breaking changes in order to go to 16 + patch.

OTOH this is great news for bad actors and pentesters.

  console.log("jsjs")

}

i've been thinking basically this for so long, i'm kinda happy to be validated about this lol

Intentionally? That's a scary feature

I like to think of Server Components as componentized BFF ("backend for frontend") layer. Each piece of UI has some associated "API" with it (whether REST endpoints, GraphQL, RPC, or what have you). Server Components let you express the dependency between the "backend piece" and the "frontend piece" as an import, instead of as a `fetch` (client calling server) or a <script> (server calling client). You can still have an API layer of course, but this gives you a syntactical way to express that there's a piece of backend that prepares data for this piece of frontend.

This resolves tensions between evolving both sides: the each piece of backend always prepares the exact data the corresponding piece of frontend needs because they're literally bound by a function call (or rather JSX). This also lets you load data as granularly as you want without blocking (very nice when you have a low-latency data layer).

Of course you can still have a traditional REST API if you want. But you can also have UI-specific server computation in the middle. There's inherent tension between the data needed to display the UI (a view model) and the way the data is stored (database model); RSC gives you a place to put UI-specific logic that should execute on the backend but keeps composability benefits of components.

I really do want a good frontend framework that lets me expressively build and render dynamic frontend components, but it feels like 99% of React's development in the last few years has been just been creating churn and making that core frontend experience worse and worse. Hooks solve challenges around sharing component meta-functionality but then end up far worse for all other non-trivial cases, and it seems like RSC & concurrency just break things and add constraints instead of improving any part of my existing experience.

I guess this is cool if you're building mega-projects, but it makes React actively painful to use for anything smaller. I still use it every day, but as soon as I find a good off-ramp for my product (something similar, but simpler) I will take it. Moving towards Preact & signals currently seems like the best option for existing projects so far as I can tell.

I agree that the developer experience provided by the compiler model used in Svelte and React is much nicer to work with

IMO, a big part of it is the lack of competition (in approach) exacerbated by the inability to provide alternatives due to technical/syntactical limitations of JavaScript itself.

Vue, Svelte, Angular, Ripple - anything other than React-y JSX based frameworks require custom compilers, custom file-types and custom LSPs/extensions to work with.

React/JSX frameworks have preferential treatment with pre-processors essentially baking in a crude compile time macro for JSX transformations.

Rust solved this by having a macro system that facilitated language expansion without external pre-processors - e.g. Yew and Leptos implement Vue-like and React-like patterns, including support for JSX and HTML templating natively inside standard .rs files, with standard testing tools and standard LSP support;

https://github.com/leptos-rs/leptos/blob/main/examples/count...

https://github.com/yewstack/yew/blob/master/examples/counter...

So either the ECMAScript folks figure out a way to have standardized runtime & compilable userland language extensions (e.g. macros) or WASM paves the way for languages better suited to the task to take over.

Neither of these cases are likely, however, so the web world is likely destined to remain unergonomic, overly complex and slow - at least for the next 5 - 10 years.

RSC is their solution to not being able to figure out how to make SSR faster and an attempt to reduce client-side bloat (which also failed)

https://www.arrow-js.com/docs/

It makes it easy to have a central JSON-like state object representing what's on the page, then have components watch that for changes and re-render. That avoids the opaqueness of Redux and promise chains, which can be difficult to examine and debug (unless we add browser extensions for that stuff, which feels like a code smell).

I've also heard heard good things about Astro, which can wrap components written in other frameworks (like React) so that a total rewrite can be avoided:

https://docs.astro.build/en/guides/imports/

I'm way outside my wheelhouse on this as a backend developer, so if anyone knows the actual names of the frameworks I'm trying to remember (hah), please let us know.

IMHO React creates far more problems than it solves:

  - Virtual DOM: just use Facebook's vast budget to fix the browser's DOM so it renders 1000 fps using the GPU, memoization, caching, etc and then add the HTML parsing cruft over that
  - Redux: doesn't actually solve state transfer between backend and frontend like, say, Firebase
  - JSX: do we really need this when Javascript has template literals now?
  - Routing: so much work to make permalinks when file-based URLs already worked fine 30 years ago and the browser was the V in MVC
  - Components: steep learning curve (but why?) and they didn't even bother to implement hooks for class components, instead putting that work onto users, and don't tell us that's hard when packages like react-universal-hooks and react-hookable-component do it
  - Endless browser console warnings about render changing state and other errata: just design a unidirectional data flow that detects infinite loops so that this scenario isn't possible

What's really going on here, and what I would like to work on if I ever win the internet lottery (unlikely now with the arrival of AI since app sales will soon plummet along with website traffic) is a distributed logic flow. In other words, a framework where developers write a single thread of execution that doesn't care if it's running on backend or frontend, that handles all state synchronization, preferably favoring a deterministic fork/join runtime like Go over async behavior with promise chains. It would work a bit like a conflict-free replicated data type (CRDT) or software transactional memory (STM) but with full atomicity/consistency/isolation/durability (ACID) compliance. So we could finally get back to writing what looks like backend code in Node.js, PHP/Laravel, whatever, but have it run in the browser too so that users can lose their internet connection and merge conflicts "just work" when they go back online.

Somewhat ironically, I thought that was how Node.js worked before I learned it, where maybe we could wrap portions of the code to have @backend {} or @frontend {} annotations that told it where to run. I never dreamed that it would go through so much handwaving to even allow module imports in the browser!

But instead, it seems that framework maintainers that reached any level of success just pulled up the ladder behind them, doing little or nothing to advance the status quo. Never donating to groups working from first principles. Never rocking the boat by criticizing established norms. Just joining all of the other yes men to spread that gospel of "I've got mine" to the highest financial and political levels.

So much of this feels like having to send developers to the end of the earth to cater to the runtime that I question if it's even programming anymore. It would be like having people write the low-level RTF codewords in MS word rather than just typing documents via WYSIWYG. We seem to have all lost our collective minds ..the emperor has no clothes.

It works out because it keeps a workforce of React Developers on their feet, learning about the new features, rather than doing other stuff. It's like SaSS for developers, only instead of paying a monthly subscription in cash, you have to pay a monthly subscription in man-hours.

We still strongly recommend everyone to upgrade their Next, React, and other React meta-frameworks (peer)dependencies immediately.

and Deno Deploy/Subhosting: https://deno.com/blog/react-server-functions-rce

> Experimental React Flight bindings for DOM using Webpack.

> Use it at your own risk.

311,955 weekly downloads though :-|

[1]: https://www.npmjs.com/package/react-server-dom-webpack

I’m interested in learning more about the history here.

However, if I am reading this correctly, your PoC falls in the category described here: https://react2shell.com/

> Anything that requires the developer to have explicitly exposed dangerous functionality to the client is not a valid PoC. Common examples we've seen in supposed "PoCs" are vm#runInThisContext, child_process#exec, and fs#writeFile.

> This would only be exploitable if you had consciously chosen to let clients invoke these, which would be dangerous no matter what. The genuine vulnerability does not have this constraint. In Next.js, the list of server functions is managed for you, and does not contain these.

Context: This is from Lachlan Davidson, the reporter of the vulnerability

So I don't think this mechanism is exactly correct, can you demo it with an actual nextjs project, instead of your mock server?

Delete this distraction to genuine blue teamers and stop shitting up the information landscape with this utter hogwash.

This is why infosec is dead.

https://react2shell.com/

https://github.com/ejpir/CVE-2025-55182-poc/issues/1#issueco...

What do you like to work with now?

It's totally fine to say you don't understand why they have benefits, but it really irks me when people exclaim they have no value or exist just for complexity's sake. There's no system for web development that provides the developer with more grounded flexibility than RSCs. I wrote a blog post about this[0].

To answer your question, htmx solves this by leaning on the server immensely. It doesn't provide a complete client-side framework when you need it. RSCs allow both the server and the client to co-exist, simply composing between the two while maintaining the full power of each.

[0] https://saewitz.com/server-components-give-you-optionality

I mean it's a lot of complexity but ideally you shouldn't bring it in unless you actually need it. These solutions do solve real problems. The only issue is people try to use it everywhere. I don't use RSC, standard SPAs are fine for my projects and simpler

> Even in systems that prevent server functions from being declared in client code (such as "use server" in React Server Components), experienced developers can be caught out. We prefer a design that emphasises the public nature of remote functions rather than the fact that they run on the server, and avoids any confusion around lexical scope. [0]

[0] https://github.com/sveltejs/kit/discussions/13897

> In remote procedure call systems, client-side stub code must be generated and linked into a client before a remote procedure call can be done. This code may be either statically linked into the client or linked in at run-time via dynamic linking with libraries available locally or over a network file system. In either the case of static or dynamic linking, the specific code to handle an RPC must be available to the client machine in compiled form... Dynamic stub loading is used only when code for a needed stub is not already available. The argument and return types specified in the remote interfaces are made available using the same mechanism. Loading arbitrary classes into clients or servers presents a potential security problem;

https://pdos.csail.mit.edu/archive/6.824-2009/papers/waldo-r...

To be fair, they also haven't released (even experimental) RSC support yet, so maybe they lucked out on timing here.

I have built electron apps for fun that utilize React Server Components entirely offline.

With SSR, those round trips to the server could be down to single-digit milliseconds assuming your frontend server is in the same datacenter as your backend. Plus you send HTML that has actual content to be rendered right away.

A truly functional pageload can go from seconds to milliseconds, and you're transferring less data over the wire. Better all around at the expense of running a React Server instead of a static file host.

SSR can be a game-changer in domains like e-commerce. But completely useless for some other use case.

RSC advantages are a bit more complex to explain, because even a simple portfolio website would benefit from it. Contrary to the common belief created by long-term ReactJS dev, RSC simplifies a lot of the logic. Adapting existing code to RSC can be quite a mess and RSC is a big change of mindset for anybody used to ReactJS.