We don't need a library that requires as much boilerplate code as
writing the AuthGuard ourselves, especially since the token validation
was already custom code by us.
The previous name PublicAuthToken was a bit misleading, since PublicAuth
could also be interpreted as being used for the public frontend in
contrast to the API. The old name before that (AuthToken) wasn't better
since it wasn't clear what type of auth is meant. I know, this is the
second renaming of the same module in less than a month. However, I
would say the name ApiToken seems rather reasonable and understandable.
Signed-off-by: Erik Michelson <github@erik.michelson.eu>
This adds support for keeping the uploads attached to a note when
deleting the same note. This is done by a simple checkbox that can be
clicked in the DeletionModal.
To do this, some parts of the note deletion had to be refactored,
especially in the case of the history page. Both the note deletion and
history removal methods used the same modal, which isn't applicable now
anymore. Additionally, there was a bug that the modal checked for
ownership in the frontend before allowing the note deletion. However, in
the context of the history page, the ownership couldn't be evaluated
since the backend API didn't include that information. This is now fixed
as well.
Signed-off-by: Erik Michelson <github@erik.michelson.eu>
For non-OIDC compliant OAuth2 providers it was only possible to define
the authorize, token and userinfo URLs but not the end_session_endpoint.
This commit adds that functionality.
Signed-off-by: Erik Michelson <github@erik.michelson.eu>
Error messages for manual OIDC attributes such as overriding the scope
resulted in wrong error messages when misconfigured.
Signed-off-by: Erik Michelson <github@erik.michelson.eu>
When the OIDC login flow for a new user is cancelled, the oidcIdToken
session variable should be cleared as well.
Signed-off-by: Erik Michelson <github@erik.michelson.eu>
The userinfo response endpoint from the OIDC provider should not be
trusted to return what we expect. Fields could be undefined. In that
case HedgeDoc would have written "undefined" into the fields for
profile picture or email address.
This fix checks for fields being undefined and returns a default value
in that case.
Signed-off-by: Erik Michelson <github@erik.michelson.eu>
The provided types by ldapauth-fork are re-exported from ldapjs. ldapjs
is unmaintained by now but since their last update, the
ConnectionOptions type seems to not contain the mandatory parameter
`url` anymore. Therefore this typecast is needed.
Signed-off-by: Erik Michelson <github@erik.michelson.eu>
The API documentation belongs strictly to the API itself.
Due to the usage of version-prefixed API endpoints, there is no conflict
with existing or future endpoints.
The reason behind this is that we already have enough exceptions in the
routing (default everything to react-frontend, exceptions for backend)
and it is hard to keep it synchronized throughout all relevant places.
This came to attention as the dev setup didn't proxy the API docs to the
backend.
Signed-off-by: Erik Michelson <github@erik.michelson.eu>
Thanks to all HedgeDoc team members for the time discussing,
helping with weird Nest issues, providing feedback
and suggestions!
Co-authored-by: Philip Molares <philip.molares@udo.edu>
Signed-off-by: Philip Molares <philip.molares@udo.edu>
Signed-off-by: Erik Michelson <github@erik.michelson.eu>
OWASP [1] recommends for password hashing the following algorithms in
descending order: argon2id, scrypt, bcrypt. They state that bcrypt may
be used in legacy systems or when required due to legal regulations.
We're however not building any legacy application. Even HedgeDoc 1.x
utilizes a more modern algorithm by using scrypt.
While bcrypt is not insecure per se, our implementation had a major
security flaw, leading to invalid passwords being accepted in certain
cases. The bcrypt nodejs package - and the OWASP cheatsheet as well -
point out, that the maximum input length of passwords is limited to 72
bytes with bcrypt. When some user has a password longer than 72 bytes in
use, only the first 72 bytes are required to log in successfully.
Depending on the encoding (which could be UTF-8 or UTF-16 depending on
different circumstances) this could in worst-case be at 36 characters,
which is not very unusual for a password. See also [2].
This commit changes the used algorithm to argon2id. Argon2id has been in
use for several years now and seems to be a well-designed password
hashing function that even won the 2015 Password Hashing Competition.
Argon2 does not have any real-world max input length for passwords (it
is at 4 GiB).
The node-rs/argon2 implementation seems to be well maintained, widely
used (more than 150k downloads per week) and is published with
provenance, proving that the npm package was built on GitHub actions
using the source code in the repository. The implementation is written
in Rust, so it should be safe against memory leakages etc.
[1]: https://cheatsheetseries.owasp.org/cheatsheets/Password_Storage_Che
at_Sheet.html#password-hashing-algorithms
[2]: https://security.stackexchange.com/a/39851
Signed-off-by: Erik Michelson <github@erik.michelson.eu>
When the frontend is notified about metadata updates, it refreshes the
data and therefore refreshes information like the timestamp of the last
revision save in the sidebar.
This commit adds such a notification from the backend to all clients on
each revision save, so that the "last saved at" value in the frontend is
correct.
Signed-off-by: Erik Michelson <github@erik.michelson.eu>
When creating a new note or adding a new alias to one,
it is checked that the new name
is neither forbidden nor already in use.
Co-authored-by: David Mehren <git@herrmehren.de>
Signed-off-by: Erik Michelson <github@erik.michelson.eu>
Previous versions of HedgeDoc suffered from the problem
that changing the media backend required manipulation of
the media links in all created notes. We discussed in
#3704 that it's favourable to have an endpoint that
redirects to the image's original URL. When changing the
media backend, the link stays the same but just the
redirect changes.
Signed-off-by: Erik Michelson <github@erik.michelson.eu>
The frontend now doesn't try to reconnect, when the disconnection happened because of a lack of permissions
Signed-off-by: Philip Molares <philip.molares@udo.edu>
We need to use .ts only if we run inside ts-node
or other tools that use it. In all other cases, we need to
refer to the .js migration files.
Signed-off-by: David Mehren <git@herrmehren.de>
Previously, an undefined result in fetchUsernameForSessionId
was handled the same way as an error, rejecting the promise.
This fixes the behavior, only rejecting the promise if an error
is returned from the session store and properly returning
undefined if the session store returns that.
Signed-off-by: David Mehren <git@herrmehren.de>
