1
0
Fork 0
mirror of https://github.com/YunoHost-Apps/trustyhash_ynh.git synced 2024-09-04 01:15:56 +02:00
trustyhash_ynh/doc/DISCLAIMER.md
2022-05-05 18:28:08 +02:00

286 lines
10 KiB
Markdown

## Trust
This app does the hash calculation in the browser using the
[WebCryptoAPI](https://developer.mozilla.org/en-US/docs/Web/API/SubtleCrypto/digest).
This means we can trust the hash calculation under the following assumptions:
- The integrity of the application has been preserved when it executes in the
browser.
- The browser and any extensions can be trusted.
### Integrity of the TrustyHash itself
Because the application itself is a single HTML file which can be saved locally,
there are various means for verifying integrity.
#### Trusted Hash Utility
The most reliable way to verify is to compute the hash of the HTML file with a
trusted hash utility on the local system, and compare against the values
published below.
#### Code Audit
Someone familiar with JavaScript can spend a few minutes reading the concise
source to be assured that the program does what it claims to.
#### No Hash Utility, Can't Audit Code?
Hmm, no trusted hash utility, can't audit the code... you just can't give up
and use your copy of TrustyHash without trusting it! While you could try some
half-measure like getting some kind of consensus on the hash value of
TrustyHash from untrusted hash utilities on the web, maybe other copies of
TrustyHash found elsewhere... ultimately if you really need to trust
TrustyHash, you've got to be a bit more rigorous.
Without knowing a thing about JavaScript until this moment, you can create a
very small, simple program in about 5 minutes, that while not as nice as
TrustyHash perhaps, will get the job of hashing a local file done. As long as
you can follow along as the following code is explained, and you can be pretty
confident the code is not doing anything fishy, you can use this to verify
TrustyHash itself. I'll show you the whole program up-front before I explain
it - see, 5 minutes, no more!
```
<!doctype html>
<html>
<body>
<script>
var fileinput = document.createElement('input')
fileinput.type = 'file'
fileinput.onchange = function(){
var reader = new FileReader()
reader.readAsArrayBuffer(this.files[0]);
reader.onload = function(){
crypto.subtle.digest("SHA-256", this.result)
.then(function(buffer) {
var hexCodes = []
var view = new DataView(buffer)
for (var i = 0; i < view.byteLength; i += 1) {
var stringValue = view.getUint8(i).toString(16)
var paddedValue = ('0' + stringValue).slice(-2)
hexCodes.push(paddedValue)}
alert(hexCodes.join(""))})}}
document.body.appendChild(fileinput)
</script>
</body>
</html>
```
JavaScript programmers may take offense with the lack of conventional
formatting above, but I'm trying to making this easy to re-type for someone who
shouldn't need to be concerned with formatting conventions.
Now a more-or-less line-by-line explanation:
```
<!doctype html>
<html>
<body>
<input type="file">
<script>
```
These lines declare an HTML document with a file input and a script. HTML
technically requires a `<head>` element, but I'm trying to save you a bit of
typing. Now let's get into the JavaScript itself:
```
document.querySelector('input').onchange = function(){
```
This says, when a file is selected via the file input...
```
var reader = new FileReader()
```
...create a reader to read the file.
```
reader.readAsArrayBuffer(this.files[0])
```
Read the file into memory.
```
reader.onload = function(){
```
When the reader finishes...
```
crypto.subtle.digest("SHA-256", this.result)
```
Hash the buffer with SHA-256.
```
.then(function(buffer) {
```
When hashing finishes, we have an unprintable object, called a buffer...
```
var hexCodes = []
```
...that we want to turn in to printable hex codes, which is just a way to
represent a number using only the characters 0 to 9 and 'a' to 'f'. Hex codes
are the standard representation of SHA-256 hash values.
```
var view = new DataView(buffer)
```
We create a view so we can read the buffer in chunks.
```
for (var i = 0; i < view.byteLength; i += 1) {
```
With each byte-size chunk...
```
var stringValue = view.getUint8(i).toString(16)
```
...convert each chunk to a number and get a string, which is just something we
can print. The string will be one or two hex digits.
```
var paddedValue = ('0' + stringValue).slice(-2)
```
To correctly print the string as a hex code, we need to add a zero to the front
in case the number is less than two hex digits, keeping the last two digits.
```
hexCodes.push(paddedValue)}
```
Keep the string we just created before moving on to the next chunk.
```
alert(hexCodes.join(""))})}}
```
Join all the strings created from each chunk together and pop it up on the
screen. That's it!
```
</script>
</body>
</html>
```
Oh, and we need these lines to formally close the HTML document.
If you followed all that, put this code into a file called
`TrustyHashLite.html` and open it up in your browser. I recommend re-typing,
rather than copy-pasting, since there are a bunch of sneaky ways someone could
trick you into copy-pasting something besides what you see on a web page. If
creating HTML files by hand is a bit confusing, you can save [the file I
created for
you](https://raw.githubusercontent.com/sprin/TrustyHash/master/TrustyHashLite.html)
as long as you promise you will make sure the code matches the above after you
have saved it. One way to do this is to open the file in a browser, right-click
and select "View Page Source".
Open the `TrustyHashLite.html` file in your browser, click the file input
button, select the `TrustyHash.html` you saved earlier. If the printed hex code
matches the published hash values, congratulations, you just wrote a program
that computes SHA-256 hashes *and* used it to validate TrustyHash!
### Hash Values
TODO: Publish hash value for 1.0.0
### Integrity of the Browser
In order to trust the results of TrustyHash, we need to trust the browser that
it runs in. Is the implementation of WebCryptoAPI to be trusted? Are extensions
able to modify the result the user sees?
If one is able to see the source of the browser and deterministic, reproducible
builds are possible, then we can start to form a strong basis of trust. Closed
source browsers must be excluded - the vendor is not able to assert a strong
claim of *what* they are distributing. At best, they may be able to publish
complete specifications for all functionality, but users still must trust the
vendor ultimately to actually implement the specifications as claimed. The
point is moot since no closed-source browser vendor publishes complete
specifications anyway.
Currently, open-source browsers are little better off. Deterministic builds are
still a work-in-progress for all popular open-source browsers
([Tor Browser](https://blog.torproject.org/category/tags/deterministic-builds),
[Firefox](https://bugzilla.mozilla.org/show_bug.cgi?id=885777),
[Chromium](https://bugs.chromium.org/p/chromium/issues/detail?id=314403).
Without deterministic builds, we must still trust the vendor ultimately to
build and distribute what they say they are building. If we trust the vendor
when they say, "deterministic builds are hard, we are working on it", and we
can trust them to secure their build environment, then we can take the signed
hash values they publish to represent the objects built from the published
sources.
So then there are three realistic ways we might have a trusted browser on our
systems:
- A verified open-source browser was bundled with our trusted operating system
distro.
- We installed an open-source browser from a trusted package manager that
handled checking verification for us.
- We downloaded the open-source browser directly from the vendor and checked
the signatures/hashes ourselves.
The third possible way, which is only feasible for a tiny fraction of extremely
diligent users, is to build the browser from source, rebuilding whenever
security updates are pushed to users.
Since the majority of browser users do not use an operating system that bundles
a verified open-source browser nor supplies a package manager which can
download and verify an open-source browser for them, this leaves manually
verifying. Because no operating system makes it easy or obvious to verify
signed downloads and awareness of the importance of verification is very low,
we have to conclude that the majority of browser users have very little basis
for trusting their browser. Similar arguments can be made for the operating
system as a whole.
So where does that leave us? Is running any program inside a browser with any
degre of trust hopeless for the vast majority of users? I would say that we may
be forced to accept some uncertainty that a program such as TrustyHash will
produce the correct results in an untrusted browser. If we accept this
uncertainty, we can use TrustyHash to bootstrap trust for a new browser or even
operating system. This, I think, is the real value of TrustyHash - to bootstrap
trust on a system by providing the best possible effort at producing trusted
hash values in an accessible way.
## Deployment
The entire application is packaged in a single, brief HTML file. Simply deploy
the file under the web server root directory.
## Why only SHA-256?
SHA-256 remains the de facto standard for verifying files via hash in 2016.
Here are some popular projects have standardized on SHA-256 for verifying
release materials:
- [Tor Browser](https://www.torproject.org/docs/verifying-signatures.html#BuildVerification)
- [OpenBSD](http://man.openbsd.org/signify)
- [FreeBSD](https://www.freebsd.org/releases/10.2R/signatures.html)
- [Centos](http://mirror.centos.org/centos/7/isos/x86_64/sha256sum.txt)
- [Fedora](https://getfedora.org/verify)
In the interests of standardization and keeping things simple, only SHA-256
will be shown. A possible addition to this project is to allow the user to
select other hash algorithms, with SHA-256 remaining the default.
## Limitations
When the application is retrieved on an HTTPS connection, the application
cannot fetch HTTP URLs due to restrictions against [mixed active
content](https://developer.mozilla.org/en-US/docs/Security/Mixed_content#Mixed_active_content]).
A workaround for this is to save the page locally and open the local copy in
the browser, as recommended anyway.