Password protect the JSONRPC interface

[ghost]

The JSONRPC interface is currently completely unprotected, I believe it should be a priority to add at least some form of password protection.

Scans for the JSONRPC interface of Ethereum wallets have already started:
https://www.bleepingcomputer.com/news/security/theres-some-intense-web-scans-going-on-for-bitcoin-and-ethereum-wallets/

[SoundButton]

Even if a hacker got access to your wallet somehow it's still encrypted with strong password i hope.

[ghost]

The JSONRPC interface is not about your personal wallet at home. It is mostly used by web servers for remotely executing commands, like handling a wallet via a web interface or accepting web payments.

While the electrum daemon is running, someone on a different virtual host of the web server could easily access your wallet via the local RPC port.

Currently, there is no security/authentication, giving someone access to the RPC port full access to the wallet.

[taviso]

Hello, I'm not a bitcoin user, a colleague pointed me at this bug report because localhost RPC servers drive me crazy 😛.

I installed Electrum to look, and I'm confused why this isn't being treated as a critical and urgent vulnerability? If this bug wasn't already open for months, I would have reported this as a vulnerability, but maybe I misunderstand something.

The JSON RPC server is enabled by default, it does use a random port but a website can simply scan for the right port in seconds.

I made you a demo. It's very basic, but you get the idea.

If you did set a password, some misdirection is required, but it's still game over, no?

Here is how I reproduced:

• Install Electrum 3.0.3 on Windows.
• Create a new wallet, all default settings. I left the wallet password blank - the default setting.
• Visit in Chrome.
• Wait a few seconds while it guesses the port, then an alert() appears with: seed: {"id": 0.7398595146147573, "result": "pony south strike horror throw acquire able afford pen lunch monster runway", "jsonrpc": "2.0"}

(Note: i dont use bitcoin, you can steal my empty wallet if you like)

[devrandom]

Looks like electrum unconditionally returns a positive CORS pre-flight check. Looks like a serious issue.

[mithrandi]

The code doing this dates back to the original implementation:

electrum/gui/jsonrpc.py

Lines 32 to 42 in 2c67de8

	class RequestHandler(SimpleJSONRPCRequestHandler):
	def do_OPTIONS(self):
	self.send_response(200)
	self.end_headers()
	def end_headers(self):
	self.send_header("Access-Control-Allow-Headers",
	"Origin, X-Requested-With, Content-Type, Accept")
	self.send_header("Access-Control-Allow-Origin", "*")
	SimpleJSONRPCRequestHandler.end_headers(self)

I suspect this should just be removed, but there may be a reason it's there?

[ecdsa]

thanks for pointing this out. this was introduced when the jsonrpc interface was merged with the daemon 4682d95

[taviso]

Thanks for investigating so quickly, in my opinion this merits a new release and an announcement, what do you think?

I think just scanning for people in the background of a website is really easy, and seems likely someone will try that. Even with encrypted wallets, you can still change options, change destination addresses, deanonymize users via listaddresses and so on.

Maybe there is a way to do more, like setting requests_dir to the wallet directory? I noticed if you set {method:"gui" param=[{uri: <bitcoinuri>}] it will silently change any address you've already entered - that seems kinda scary.

[ecdsa]

indeed we will do a new release

[ecdsa]

A new release (3.0.4) is available on the website and on Google Play, that disables CORS.
We may enable it again after we add password protection.

[kirelagin]

TBH simply disabling CORS doesn’t seem to be a proper solution, given that there is plenty of software other than browsers that runs locally.

I’d expect this RPC server to at least be disabled by default and, even better, password protected, given that, I suspect, the vast majority of users never use it and don’t even know that it exists.

[SomberNight]

@kirelagin The application itself uses the RPC server to work. Password protection is currently being implemented.

[kirelagin]

@SomberNight Oh, I see.

In that case you’ll still need a way to securely agree on a password betwen the daemon and the GUI app. I mean, if you simply write the password to a file for the GUI to read, rouge local apps will also get it from there :(.

[SomberNight]

@kirelagin Yes, I know. Though the same apps might as well go for the wallet file directly, right?

[kirelagin]

@SomberNight Yep, you are right, there is nothing you can do about that, so that’s the problem of the user who doesn’t set a password.
Though, as @taviso pointed out, signing transactions is not the only issue, being able to modify GUI input fields via this API is also a serious one.

[h43z]

Is it correct that without explicitly loading the wallet with electrum daemon load_wallet one cannot run commands against the RPC? So a "normal" desktop user of electrum cannot be attacked, right?

[fyookball]

good work guys. We ported the patch to E.C. thanks.

[kirelagin]

@h43z It is not correct. The RPC daemon is started automatically as soon as you launch the GUI app.

[taviso]

@h43z No, there is a load_wallet rpc, method=daemon, params=[{subcommand: “load_wallet“}].

[h43z]

got ya guys. One can load the the wallet via rpc... 🤦‍♂️

[SomberNight]

To sum up the parts that are probably the most interesting,
on an internet-connected device, when you open a website that exploits this, if Electrum (pre-3.0.4) is running at that time,

• the xpub of any wallet can be read
• the seed words/private keys can be read directly IF your wallet has no password set
  • if there is a password, the ciphertext can not be retrieved, only online bruteforce is possible (making a new json-rpc request for every password-retry)

[mautematico]

@loshchil it is irresponsible itself if a user decides not to encrypt their wallets.
Also, users don't need to be familiar with github or anything else. They just need to keep their software up-to-date.

However, I agree that we need a proper way to let users know there are critical bugs.
Are you able to come up with a solution and propose it in a PR?

[loshchil]

@mautematico

They just need to keep their software up-to-date.

The current situation we are in is a good example that the above-mentioned view is hard to defend if user's security is taken into account. More specifically, a few moments after the vulnerability appeared on github, a ton of responsible users who used the most recent version so far have become targets to trivial exploits (i.e., hundreds or thousands of hackers are qualified to implement them) which could allow to hack wallets with or without brute-forcing their passwords.

I appreciate your proposal to come with a solution and a PR. However, I will limit my contribution to this reply here.

My proposal would be to
i) Make Electrum regularly check whether a more recent version is available.
ii) Modify Electrum main window title/caption "Electrum x.y.z" to "Electrum x.y.z TEXT",
where TEXT depends on the situation we are in. For instance, it can be empty, "(is not safe to use)",
"(outdated but does not have known vulnerabilities)", "(outdated and has known vulnerabilities)"
iii) Update Electrum.org to have a section where known vulnerabilities are described version-wise and a set of best practices to deal with them is presented.

[ghost]

@attritionorg, invalid reference, because that bug i submitted is not by any means related to this issue, if you had properly read and understood the bug report you would understand that, but youre probably too busy sending people boxes of feces to care

[msadar]

@ecdsa

what if I never access to the previous version and deleted the wallet software? which means keeping as cold storage?

[leonklingele]

@msadar you were only susceptible to an attack while Electrum was running and your wallet unlocked.
Nevertheless, as it is unknown whether the bug has been exploited, it might be worth to create a new cold wallet / Electrum wallet, encrypt it, and transfer your funds to it.

[AngelTs]

It is an excellent example why you should use only the ported version:
because the ported version (even with bug(s)) in most cases is used for a very short time (set cd/usb stick, do stuff, then remove it), while installed version is very probably to works all time!

[zatricky]

@ecdsa re CVE id, is there a place where we will know it has been issued? Or will you be posting it here when it has been issued?

[ecdsa]

@zatricky you can check https://docs.google.com/spreadsheets/d/1PlDOsZ4Q36JU4Dz9zyBB2F3814dScppCRCe1muCT7JI/edit#gid=1009122160

[attritionorg]

That sheet is for assignments requested via DWF, a CNA that handles some open source software. If you request via the form on MITRE's web site, you won't be able to look it up until it is published, or the requesting party receives the assignment and posts it here.

[petterreinholdtsen]

Hi. Do you need help getting a CVE assigned, or is the process already in progress? I could ask the Debian security team for one, if it is useful.

[ecdsa]

@petterreinholdtsen I filled the form on MITRE, and my request is listed on the document linked above. Is there something else I need to do? I am not familiar with the process.

[petterreinholdtsen]

I'm not sure either, but have no better suggestion. I asked on #debian-security and it seem to be the recommended approach based on the feedback there.

[carnil]

The assigned CVE for this issue looks to be CVE-2018-1000022

[attritionorg]

@carnil have a reference for that? CVE-2018-6353 was assigned and opened:

http://cve.mitre.org/cgi-bin/cvename.cgi?name=2018-6353

[mithrandi]

CVE-2018-6353 is for #3678, a different issue.

[carnil]

@attritionorg: it was apparently assigned by DWF project. I got that confirmation by MITRE, since I stumpled over the CVE when I was investigating CVE-2018-6353, which mentioned "a different vulnerability than CVE-2018-1000022". I queried about this MITRE, who confirmed that DWF has assigned the CVE (but not yet published back?).

As @mithrandi stated the CVE-2018-6353 is for #3678, whereas CVE-2018-1000022 is for #3374

[mautematico]

Someone could please delete file posted by @Ruethairat?