* Add typehints
* Add return types
* Opcode opts from phpstorm
* Made strict
* Fixed tests: No need to test bogus values anymore strict typing fixes
this
Signed-off-by: Roeland Jago Douma <roeland@famdouma.nl>
This adds a phan plugin which checks for SQL injections on code using our QueryBuilder, while it isn't perfect it should already catch most potential issues.
As always, static analysis will sometimes have false positives and this is also here the case. So in some cases the analyzer just doesn't know if something is potential user input or not, thus I had to add some `@suppress SqlInjectionChecker` in front of those potential injections.
The Phan plugin hasn't the most awesome code but it works and I also added a file with test cases.
Signed-off-by: Lukas Reschke <lukas@statuscode.ch>
(Possibly) fixes#3470
When updating the main file /files_external/rootcerts.crt we should not
read from /files_external/rootcerts.crt at the same time.
For 2 reasons: writing to a file and reading from it at the same time
can have non deterministic results
And we don't want all the certificates to appear 2 times in there.
This isn't caught by our standard file locking (that does not allow this
actually) because it is in a non locked path....
Signed-off-by: Roeland Jago Douma <roeland@famdouma.nl>
Currently, when disabling the brute force protection no new brute force attempts are logged. However, the ones logged within the last 24 hours will still be used for throttling.
This is quite an unexpected behaviour and caused some support issues. With this change when the brute force protection is disabled also the existing attempts within the last 24 hours will be disregarded.
Signed-off-by: Lukas Reschke <lukas@statuscode.ch>
This allows adding rate limiting via annotations to controllers, as one example:
```
@UserRateThrottle(limit=5, period=100)
@AnonRateThrottle(limit=1, period=100)
```
Would mean that logged-in users can access the page 5 times within 100 seconds, and anonymous users 1 time within 100 seconds. If only an AnonRateThrottle is specified that one will also be applied to logged-in users.
Signed-off-by: Lukas Reschke <lukas@statuscode.ch>
With 10.0.1 CSPv3 is broken in Safari if it doesn't run from a local IP. Awesome.
=> Let's remove this for Safari and keep chrome and Firefox in the whitelist.
Signed-off-by: Lukas Reschke <lukas@statuscode.ch>
CSP nonces are a feature available with CSP v2. Basically instead of saying "JS resources from the same domain are ok to be served" we now say "Ressources from everywhere are allowed as long as they add a `nonce` attribute to the script tag with the right nonce.
At the moment the nonce is basically just a `<?php p(base64_encode($_['requesttoken'])) ?>`, we have to decode the requesttoken since `:` is not an allowed value in the nonce. So if somebody does on their own include JS files (instead of using the `addScript` public API, they now must also include that attribute.)
IE does currently not implement CSP v2, thus there is a whitelist included that delivers the new CSP v2 policy to newer browsers. Check http://caniuse.com/#feat=contentsecuritypolicy2 for the current browser support list. An alternative approach would be to just add `'unsafe-inline'` as well as `'unsafe-inline'` is ignored by CSPv2 when a nonce is set. But this would make this security feature unusable at all in IE. Not worth it at the moment IMO.
Implementing this offers the following advantages:
1. **Security:** As we host resources from the same domain by design we don't have to worry about 'self' anymore being in the whitelist
2. **Performance:** We can move oc.js again to inline JS. This makes the loading way quicker as we don't have to load on every load of a new web page a blocking dynamically non-cached JavaScript file.
If you want to toy with CSP see also https://csp-evaluator.withgoogle.com/
Signed-off-by: Lukas Reschke <lukas@statuscode.ch>
Class Throttler implements the bruteforce protection for security actions in
Nextcloud.
It is working by logging invalid login attempts to the database and slowing
down all login attempts from the same subnet. The max delay is 30 seconds and
the starting delay are 200 milliseconds. (after the first failed login)
