[toraobo]

> Due to the use of static keys, an authenticated attacker can trick the server into deserializing maliciously crafted ViewState data.

Many years ago I was shocked that ASP.net will deserialize arbitrary (potentially unsafe) objects from the client and relies on signatures to ensure that parameters sent by the client were in fact round-tripped via HTTP POST from the same server.

How is __VIEWSTATE not a horrible idea???

[watermelon0]

I assume the idea behind __VIEWSTATE is that you don't have to save anything on the server side, which makes the whole architecture easier (no need for storage layer, sharing the secret between machines allows for load balancing), and you can support as many clients as possible, since you don't have to worry about the storage size.

__VIEWSTATE should be as secure as JSON Web Tokens, unless you manage to leak the secret (or as in this case, you use a shared one for all customers.)

I don't know why exactly did they choose to (de)serialize executable code, instead of using an XML or similar format, but similar choices were made (and later changed) with frameworks/libraries for other languages, so they are not the only one.

[danielheath]

There have been vulnerabilities in java and rails caused by deserializing arbitrary classes with given constructor parameters.

An attacker just has to find one that (e.g.) accepts a url and a filepath in its constructor and saves the file to that location, and you've (usually) got code execution (by eg overwriting something that gets run frequently).

[to11mtm]

I could have some of the specifics wrong here, especially since it's been forever...

But the problem is there are some things that needed to be serialized for things like .NET remoting or firing stuff across appdomains. Things like Delegates.

Now, I -think- the intent, back in the day that ASP.NET was created, to use Code Access Security/Security Transparent Code (CAS/STC) to safely put a system like this together. But CAS/STC is pretty easy to do 'wrong' and a lot of people, rather than setting up the correct LinkDemands for individual items like Database, File IO, ETC on the specific endpoints, would just toss 'AllowPartiallyTrustedCallersAttribute' on the assembly and call it a day. This renders it CAS/STC almost worthless, to the point Microsoft half-killed this concept in Net 4.0 and in Core it's just about gone.

[e12e]

Handling user input is tricky, and so is (de) serialization (I seem to recall most big frameworks from ruby, via php and python through Java - all doing some form of unsafe marshaling from both xml and json).

For encrypted and signed data on the client, you get to add "crypto is hard" on top.

You get things like clear signed data - that makes it easier to de-couple verifying signatures from marshaling contained data - perhaps without checking the signature...

You get encrypted data that isn't using a proper authenticated encryption construct - and again open the door for manipulated data.

Even with proper encryption, you get key revocation wrong (left out), and there's maybe no expiry on chiper texts either - so you happily accept data encrypted and signed by a compromised or expired key.

And there's key management, like in this case, re-using a key that should be sigle client/single session.

Oh,and re-play attacks due to missing serial numbers/nonce on transactions/messages...

But yeah, sure. Like another comment mentions - in theory encrypted data managed by the client is sound.. Unfortunately in practice theory isn't always right.

[geocar]

Seaside[1] and (IIRC) arc[2] both do this too.

My understanding was that __ViewState was intended to be encrypted in production by setting something in machine.config[3].

[1]: https://en.wikipedia.org/wiki/Seaside_(software)

[2]: http://arclanguage.org/

[3]: https://docs.microsoft.com/en-us/previous-versions/dotnet/ar...

[needusername]

Hell no. Seaside captures the state server side, stores it under an id and sends that id to the client. The client sends the id of the state snapshot back to the server.

[Quekid5]

AFAICT there's nothing inherently insecure about server-encrypted data on the client. You have to secure that encryption, but fundamentally it Should Be Fine(TM).

Now, deserializing arbitrary objects which might contain code?... that's crazy town, but it was a different time.

[mjevans]

This is an area where 'fine' rapidly becomes problematic.

Remember that the dataset is often going to be small, and frequently will contain known or easily guessed strings. Uncompressed, often at predicable locations.

There are whole classes of cryptographic vulnerabilities which might result from either not compressing (compression should in theory would normalize entropy over the stream) or using a compression algorithm that results in a predictable dictionary, length, or other value in the same location (if junk padding isn't used properly).

Also, sending the state from client to server might open replay attacks and all sorts of other horrid situations.

Security depends on doing everything correctly all the time; this context IMO just feels open to too many plausible and unknown (potentially introduced in the future) vulnerabilities.

[Quekid5]

> Remember that the dataset is often going to be small, and frequently will contain known or easily guessed strings.

This is why modern crypto is designed to be resistant to prefix attacks, rainbow attacks, why CBC is considered a good idea, etc. soo....

Yeah, you're being paranoid about the wrong things, I think.

EDIT: FWIW, I was being a bit cheeky with the Fine(TM) thing, but if you get the "process" bits about encrypting the thing you send to the browser (and will be returned to you) right, you should really be in a position about as good as if the encryption itself had been compromised.

[altmind]

I remember that JSF(java faces) used the same __VIEWSTATE approach to preserve the state, attaching the state to every request. It may also be a great source of security vulnerabilities.

Luckily neither .net WebForms, nor JSF are popular nowadays.

[adossi]

>Luckily neither .net WebForms, nor JSF are popular nowadays.

Oh boy do I have news for you!

[sebazzz]

I'm suprised this attack is even possible. Disabling of the viewstate MAC validation is disallowed, since a while [1]. However, MAC validation is apparently circumvented because the ViewStateUserKey is known.

[1]: https://devblogs.microsoft.com/aspnet/farewell-enableviewsta...