Hi, Lead Dev of WebVM and CTO of Leaning Technologies here. Happy to answer any questions you might have... potentially in a few hours, it's quite late here :-)
Even better, consider joining our Discord: https://discord.gg/yTNZgySKGa
Hi, Lead Dev of WebVM and CTO of Leaning Technologies here. Happy to answer any questions you might have... potentially in a few hours, it's quite late here :-)
Even better, consider joining our Discord: https://discord.gg/yTNZgySKGa
There are a few others
VMs
https://jamesfriend.com.au/pce-js/ (https://github.com/jsdf/pce)
https://www.pcjs.org/ (lots of hardware and OSes) (https://github.com/jeffpar/pcjs)
Mac OS
https://infinitemac.org/ (https://blog.persistent.info/2023/03/infinitemac-dot-org.htm...)
https://jamesfriend.com.au/projects/basiliskii/BasiliskII-wo...
These are full system emulators, this is more a compatibility layer ("Linux syscall emulator") for emulated userland binaries. I.e. a bit higher level and a bit more usable/flexible as a result.
I think it's also much faster in practice
I'm curious, but what's the catch?
The obvious one is that it is userland-only, you don't get to play around with custom kernels being emulated on the hardware layer.
Beyond the userland-only limitation, syscall emulation is probably incomplete at best[0], and likely doesn't pass the Linux test suite for the subset of calls that are supported.
To get a flavor of the work required when emulating the kernel, you can read up on the work it took Windows developers to emulate Linux syscalls on top of the NT kernel for WSL 1 here: https://learn.microsoft.com/en-us/archive/blogs/wsl/wsl-syst...
WSL devs eventually abandoned that effort in favor of running the actual kernel in a VM -- that's how WSL 2 is implemented.
[0] I'd be surprised if sys_fork actually forks the process using a web worker, for example.
It's a commercial project and not open-source https://news.ycombinator.com/item?id=40944412
It doesn't seem so at all from what I've tried.
What have you tried?
On webvm.io
$ time python3 -c "max(range(10**7))"
real 0m1.558s
On https://copy.sh/v86/?profile=archlinux
# time python3 -c "max(range(10**7))"
real 0m5.283s
Where do the most advanced minecraft computers[1] land on the scale of sophistication compared to all these + WebVM?
What a practical use case of such thing?
(Same answer as a more deeply nested comment below)
The technology is extremely flexibile, off the top of my head:
* Education (Linux, Programming, Security, ...)
* Live docs for arbitrary languages and binary libraries
* Preservation of historical software and games
* Virtualization of legacy Windows enterprise apps.
* Dev environment for Web IDEs
Just a few examples, the list could go on for long
Very interesting, thanks! and good luck on the project.
installing emacs in my web browser?
Making your senior engineers take off their glasses, sigh, and pinch the bridge of their nose.
I don't get you, how about the juniors ?
Nod their heads with approval to mask their confusion.
This isn't WebVM specifically, but one example using WebContainers we've built is https://demo.graphweaver.com/
This runs a NodeJS server and frontend in your browser so you can see what Graphweaver can do without having to install or run anything more than a browser on your computer. The example has an SQLite database joined to a REST API. Graphweaver serves a GraphQL API with the result.
Just a curiosity to play a bit
Networking provided by Tailscale. Clever! Works around some potential issues.
Using Tailscale is the most effective solution for in-browser networking, we wrote at length about the topic some time ago: https://labs.leaningtech.com/blog/webvm-virtual-machine-with...
I am really confused by one part of your blog post: it says that you had to embed Lwip as applications don't send/receive IP packets... but you also describe this as x86 virtualization running specifically-Linux, and Linux absolutely has its own TCP stack, so I don't understand what Lwip is actually accomplishing here... I would have thought you'd essentially be implemented as a paravirtualized network adapter, but then the article mentions system calls? I guess I just fundamentally don't understand at what level of the stack you have this implemented... did you replace the Linux kernel and reimplement its entire system call layer?
Yes: https://labs.leaningtech.com/cheerpx#cheerpx-versus-v86jslin...
Also, a little below:
CheerpX currently focuses on user mode (ring 3), and does not fully emulate the kernel (ring 0). We do, however, implement a subset of the Linux system call interface, which is enough to run most applications.
Have you looked at gvisor to help implement some of the outstanding system calls given it’s a kernel written in go?
Can we use custom derp/headscale setups? Would be interesting even for use internally then, hell rolling custom versions with preloaded software could also be pretty cool all connected to a corporate headscale/derp.
Per their docs, yes: https://github.com/leaningtech/webvm/blob/main/docs/Tailscal...
Could be cool to see an implementation using Chrome's experimental Direct Sockets API [0]. The API requires using a form of new web packaging known as Isolated Web Apps which have a strict CSP [1] though, so it might not be possible to use it for this since I'd assume it's dynamically executing a lot of code.
[0]: https://github.com/WICG/direct-sockets/blob/main/docs/explai...
[1]: https://github.com/WICG/isolated-web-apps/blob/main/README.m...
can i run a webserver that can talk with the internet and others can call
I suspect you’d have trouble opening ports on the client’s machine. Maybe you could use a dedicated http proxy the client talks to
A custom service worker might be able to proxy for you without needing Tail Scale.
Tailscale funnel should work. I'll double-check when I'm back home if someone else hasn't.
This feels way slower than https://bellard.org/jslinux/
I tried compiling mandelbrot.c (single threaded one) from the benchmarks game and execution on jslinux took 12 seconds, while webvm completed it in 1.2 sec. The host takes 0.03 sec with the same gcc flags.
It may be faster for this particular benchmark, yet it feels slow and sluggish for interactive use.
Hard to compete with Bellard :)
Looks like it’s broken on mobile
Worked on stock iOS in Safari! Must've taken them ages to debug...
Works for me in Firefox Focus on Android
Can we port Adobe Flash to WebVM to avoid having to pay for "CheerpX for Flash"?
Have you tried Ruffle? It's also capable of running Flash in the browser.
Why do you want to scam them out of their hard work?
Looks like the underlying project is not open source:
"WebVM is based on a CheerpX build that we host. We encourage users to deploy their own versions of WebVM, but we don’t currently allow self-hosting of the CheerpX engine itself. This public build of CheerpX is provided as-is and is free to use for technological exploration, testing and non-commercial uses. If you want to build a product on top of CheerpX/WebVM, please get in touch:"
https://labs.leaningtech.com/blog/mini-webvm-your-linux-box-...
That’s a shame. I was excited at first, but my interest in this project is now gone.
Agreed, even the ability to self host the closed source WASM binary would have been acceptable.
Can it be made to work offline and keep state locally between reloads?
Working offline is not truly possible since that would require downloading the full 2GB disk image, which is something that is deliberately avoided by the system architecture.
State is already locally maintained via IndexedDB, no data is ever sent back or saved on our servers.
For many people that's not an issues, so maybe you need to add it someday.
Love the buzzwords.
It's a Linux command line environment running in a browser. Slowly.
Have to agree, that description is more confusing than anything. I guess JS code that a web browser executes is considered "serverless" now.
Lol I was so confused
Totally unusable on mobile unfortunately (at least Firefox and Chrome on Android). Might be a limitation of how the keyboard works, but every command gets duplicated (at least) when I enter it.
Using Brave browser on iPhoneXR works nicely.
It doesn't work in my Chrome on Android. The cursor is blinking, nothing happens.
If a browser is not supported, the site should display a specific error message.
Works fine for me on Android
root password?
"password", per https://github.com/leaningtech/webvm/blob/main/dockerfiles/d...
Note that "CheerpX enables you to run existing 32-bit x86 native binaries". For some reason support for wasm64 (in browsers) has been stagnated for years, which is a pity.
Most WASM, WebGL and WebGPU features take ages in browsers, measured in decades.
$ uptime
Error: /proc must be mounted
Cool, but very often my keystrokes aren't showing up, making this pretty unusable.
Also, for the use case of local dev environment for Web IDEs, this doesn't look fast enough.
node --version
runs fast after the first time, but executing any JavaScript is very, very slow: user@:~$ time node -e "console.log('Hello')"
Hello
real 0m7.693s
user 0m0.012s
sys 0m0.000s
It even emulates fork!
Sweet, could i run a nodejs http api directly from browser ?
Now if only I could install chrome via apt and run it inside of this linux machine.
Can this run NFS or other network filesystems?
First thing I tried was `apt-get install emacs` and it hung on downloading. Editor preferences aside, I assume any nonstandard package will have the same problem. Interesting idea, but apart from trivial glue you may not need more than a container or unikernel for anyway, I'm struggling to understand a practical use case.
Hi, this looks really amazing, will give it a try later thanks
bash: man: command not found
Where are files located, server, client, browser local storage?
How do you import a .exe compiled elsewhere, curl, scp?
This is a Linux environment, but it can run Windows apps? How does that work?
Let me fix the title: "WebVM is a wasm VM running in browser"
I first used Slackware Linux off floppies in the mid 90s, yet every time I see linux in the browser I am gobsmacked.
I’ve said for a while that a browser is basically its own operating system. Now it can host operating systems!
ungodly slow on apple silicon
server-less - like the server that runs the browser that the VM runs inside.
/sarcasm
Now just needs some kind of hack for networking through a browser extension and this would be so awesome.
Hi thanks for taking part! I have probably very silly questions but why?
What are the use cases you Invision?
Last but not least, an IRC server/channel in lieu of discord? I'm stuck in my ways!
Nice work too, I hope this doesn't come across as flippant.
EDIT: Side note, I opened vi and escape is.. well maybe a browser issue but 'cant exit vi' .. Never thought id say that again
Answer to the second question is that no one uses IRC.
The trend of using discord over forums does suck though, it’s really hard to search old discussions.
I use IRC.. Discord is fine, I just prefer to have both.
What are some good places for a technical / programming hangouts on IRC? Can you point me at some materials please?
https://libera.chat/
A channel listing will give you _many_ results.
I remember back in the day I wouldn't even check the listings, just joining whatever #topic or #hobby or #software and finding it filled with people almost always.
Also see https://oftc.net/, where a bunch of older-school development projects hang out. (That's where the Xen Project moved to after freenode imploded, before eventually switching to Matrix.)
IRC still has 270k users in 2024 via https://netsplit.de/
It's not like it used to be in the 90s but that's still something.
Discord has around 200 million monthly active users. At least according to one albeit dubious source: https://backlinko.com/discord-users#discord-monthly-active-u...
Seems reasonable to go with the growing platform with 1000x more users.
How many of those are bots. 270k is basically no one. IRC is dead
IRC <3 Still daily driving it with some friends. I wouldn’t be surprised if my Discord chat history was unavailable in a decade, so IRC is a nice option to run on the side. There’s value in simplicity, and I admit the risk of sounding like a tech hipster.
IRC has no chat history either, right. I get the simplicity of IRC but searchable history is a bonus for Discord. As long as the service is available, searching is kinda possible. With irc you have to find out which bot provides history, which is then usually split over multiple files
I agree! I am a nobody and forever will be a proud nobody!
Another option is a community hosted Mattermost (it's like Discord/Slack/IRC). It's pretty easy to run, all the chat are in PostgreSQL so it's "easy" to produce archive logs (like IRC has). Or do any other magic.
The target audience for web-based linux vm is much more likely to use IRC
May it is a browser issue. I was able to exit with ':q<Enter>'
Right, I can go to insert mode, but to leave is, tricky. I found going to replace mode and escape worked in FF. I'm sure its a browser issue, I also have vimium so probably a nightmare of hooks!
Have you tried with <Control-Left Square Bracket>? aka Ctrl-[, or ^[, as seen sometimes in the terminal. This works for me in most terminals as an alternative to Escape when using vi/vim.
In mobile (at least in mine, firefox) the input become buferred autocomplete and when accepted via return, the typed word gets duplicated before entered. Workaround is to accept via space and delete as necessary
IRC is nice for people who use it, but it has a huuuuge barrier to entry for new users. The very first thing you have to do is choose a client out of a list of 100 clients, half of which haven't been maintained since 2003. Then you need to set up a bouncer somewhere so that you can see messages that were posted when you were away. Those two things will filter out 99.99% of new people.
Or, you could use Discord (or Matrix); people sign up for an account and that's it.
Our open-source project switched from IRC to Matrix (after having been on IRC for over a decade before that), and got a noticeable uptick in engagement.
Unfortunately, Matrix still has some long-standing UI quirks, which would make me reluctant to use it in a context where I want people to pay me money; particularly if some of those are non-technical users.
"huuuuge barrier to entry for new users"
1. go to https://web.libera.chat/
2. click Start (as a Guest account)
3. Discover you can't join a lot of channels because you're not registered; or worse, that your messages sent to the channel are silently discarded
And you still have the problem that if you're in the wrong time zone and happen to close the window, you may miss responses.
There are advantages to having high barriers to entry for a community; Ham radio is essentially a global chat room with an entrance exam, for instance. But there are costs too.
Not OP, but they said the following in an answer to similar question: https://news.ycombinator.com/item?id=40940707
"The technology is extremely flexibile, off the top of my head:
* Education (Linux, Programming, Security, ...)
* Live docs for arbitrary languages and binary libraries
* Preservation of historical software and games
* Virtualization of legacy Windows enterprise apps.
* Dev environment for Web IDEs
Just a few examples, the list could go on for long"
Hi and thanks for the effort. Is there a roadmap for supporting GUI applications?
Yes, we are working on it right now. The immediate priority is booting a complete desktop environment. Next steps will include 3D graphics.
What are some use cases for this?
The technology is extremely flexibile, off the top of my head:
* Education (Linux, Programming, Security, ...)
* Live docs for arbitrary languages and binary libraries
* Preservation of historical software and games
* Virtualization of legacy Windows enterprise apps.
* Dev environment for Web IDEs
Just a few examples, the list could go on for long
You want it to be 3D!! Loading this is a multiplayer social VR sandboxed WebXR environment would be so amazing one day! I wonder if it could all run off-thread in a Worker and one day maybe use webGPU to update textures async! I hope computers can be fun and care free like a game one day and AI companions will let people learn and do anything! Thanks for working to make this!
I tested in Firefox on Linux. It works well except:
1. Copy and paste does not work and this could be bad for basically every editing task (programming etc).
2. The screen freezes sometimes and keypresses don't appear. Usually pressing backspace unfreezes the screen. Firefox about:performance doesn't show any CPU usage in the tab. Linux's htop (on my machine) doesn't show anything strange with Firefox.
And a question: I saw that new files persist after closing the tab and opening it again. I didn't investigate my localstorage etc with Developer Tools. Did you use that or some other browser feature? How truly persistent are those files?
1. Support for traditional copy/paste shortcuts is problematic, since Ctrl-C is interpreted as the SIGINT signal. Copying via the mouse menu should work. Pasting is supported via Ctrl+Shift+V.
2. We had these reported from several users and we will investigate as soon as we can: https://github.com/leaningtech/webvm/issues/102
To answer you question: filesystem chunks are cached/persisted using a local IndexedDB. It provides a fairly strong level of persistence, although the user can wipe it out of course.
Thanks for the reply.
Copy&paste via menu works (I didn't think about it) and Ctrl+Shift+V works too. Ctrl+Shift+C opens Firefox's Developer Tools. It's difficult to find a reasonable shortcut.
I don't believe I've ever seen Ctrl+insert and shift+insert makes to anything _other_ than copy and paste.
Could it be possible to run containers within this environment?
I ask because I personally think we are headed towards a future where web browsers are just sandboxed container environments where we download and run containers for each app...
Eventually, even calling them "web" browsers might feel a little odd, almost the way in which we "dial" a phone number today, even though we now mostly key it in on a touch screen.
Well, the environment itself is a container. Supporting Linux "namespaces" used to implement containers is also a possibility.
As described in the project README we provide a GitHub Action workflow to convert Dockerfiles into bootable images for WebVM: https://github.com/leaningtech/webvm
Hello there, I havent tried webvm yet but looks similar to v86. There were lot of network related challenges working with v86, for eg. You cant do curl, DNS resolution does not work out of the box. How did you address these challenges in webvm. I think it is very interesting and exciting what you guys are building.
We support networking via Tailscale, we wrote a detailed blog post some time ago: https://labs.leaningtech.com/blog/webvm-virtual-machine-with...
Hey Alessandro - I'm a huge fan of WebVM + y'all's other work at Leaning Tech.
Curious - is your implementation of how you wrangled lwIP for the networking piece open source? I had previously read the article you linked elsewhere, re: networking, but would love to see the specifics of your approach there.
All the components of the network stack are open source: namely the Tailscale go client and the lwip networking stack. Some (not fully trivial) glue code is also required, which is not currently published. We do plan to cleanup the stack and release it as one of our upcoming internship projects.
Are you associated with Tail Scale? Is Tail Scale like a proxy? Are other options available - is Tail Scale open source too?
edit: Saw your other reply!
https://labs.leaningtech.com/blog/webvm-virtual-machine-with...