we do not listen to deletion hooks anymore, because there is no guarantee that they
will be heard - requires that something fetches the CommentsManager first.
Instead, in the user deletion routine the clean up method will be called directly. Same way
as it happens for files, group memberships, config values.
register CommentsManager service, allow override, document in config.sample.php
don't insert autoincrement ids in tests, because of dislikes from oracle and pgsql
specify timezone in null date
only accepts strings for ID parameter that can be converted to int
replace forgotten hardcoded IDs in tests
react on deleted users
react on file deletion
Postgresql compatibility
lastInsertId needs *PREFIX* with the table name
do not listen for file deletion, because it is not reliable (trashbin, external storages)
add runtime cache for comments
We already support the `config_is_read_only` for the config file itself. However not for the whole directory (which is a bug).
This unifies the check in the checkServer routine with the one in base.php. Now one can enable a read only config folder so that ownCloud is not allowed to overwrite it's own source code.
To test this set the whole config folder to read only, clear your session, refresh, see it fails, add the new code, refresh, see it works. Also verify that setup still works fine. (obviously setup does not work with a read only config Also verify that setup still works fine. (obviously setup does not work with a read only config))
Fixes https://github.com/owncloud/core/issues/14455
Added in isSharable() in incoming remote share.
Added in isSharable() in regular incoming share.
Added in FileInfo to make sure the proper attributes are returned to the
clients.
When `DirectorySlash off` is set then Apache will not lookup folders anymore. This is required for example when we use the rewrite directives on an existing path such as `/core/search`. By default Apache would load `/core/search/` instead `/core/search` so the redirect would fail here.
This leads however to the problem that URLs such as `localhost/owncloud` would not load anymore while `localhost/owncloud/` would. This has caused problems such as https://github.com/owncloud/core/pull/21015
With this change we add the `DirectorySlash off` directive only when the `.htaccess` is writable to the dynamic part of it. This would also make `localhost/owncloud` work again as it would trigger the 404 directive which triggers the redirect in base.php.
mod_rewrite as used by the front controller may require a `RewriteBase` in case the installation is done using an alias. Since we cannot enforce a writable `.htaccess` file this will move the `front_controller_active` environment variable into the main .htaccess file. If administrators decide to have this one not writable they can still enable this feature by setting the `front_controller_active` environment variable within the Apache config.
This changeset allows ownCloud to run with pretty URLs, they will be used if mod_rewrite and mod_env are available. This means basically that the `index.php` in the URL is not shown to the user anymore.
Also the not deprecated functions to generate URLs have been modified to support this behaviour, old functions such as `filePath` will still behave as before for compatibility reasons.
Examples:
http://localhost/owncloud/index.php/s/AIDyKbxiRZWAAjP => http://localhost/owncloud/s/AIDyKbxiRZWAAjPhttp://localhost/owncloud/index.php/apps/files/ => http://localhost/owncloud/apps/files/
Due to the way our CSS and JS is structured the .htaccess uses some hacks for the final result but could be worse... And I was just annoyed by all that users crying for the removal of `index.php` ;-)
This PR implements the base foundation of the code signing and integrity check. In this PR implemented is the signing and verification logic, as well as commands to sign single apps or the core repository.
Furthermore, there is a basic implementation to display problems with the code integrity on the update screen.
Code signing basically happens the following way:
- There is a ownCloud Root Certificate authority stored `resources/codesigning/root.crt` (in this PR I also ship the private key which we obviously need to change before a release 😉). This certificate is not intended to be used for signing directly and only is used to sign new certificates.
- Using the `integrity:sign-core` and `integrity:sign-app` commands developers can sign either the core release or a single app. The core release needs to be signed with a certificate that has a CN of `core`, apps need to be signed with a certificate that either has a CN of `core` (shipped apps!) or the AppID.
- The command generates a signature.json file of the following format:
```json
{
"hashes": {
"/filename.php": "2401fed2eea6f2c1027c482a633e8e25cd46701f811e2d2c10dc213fd95fa60e350bccbbebdccc73a042b1a2799f673fbabadc783284cc288e4f1a1eacb74e3d",
"/lib/base.php": "55548cc16b457cd74241990cc9d3b72b6335f2e5f45eee95171da024087d114fcbc2effc3d5818a6d5d55f2ae960ab39fd0414d0c542b72a3b9e08eb21206dd9"
},
"certificate": "-----BEGIN CERTIFICATE-----MIIBvTCCASagAwIBAgIUPvawyqJwCwYazcv7iz16TWxfeUMwDQYJKoZIhvcNAQEF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\niG+UxI/MNcWV0uJg9S63LF8=\n-----END CERTIFICATE-----",
"signature": "U29tZVNpZ25lZERhdGFFeGFtcGxl"
}
```
`hashes` is an array of all files in the folder with their corresponding SHA512 hashes (this is actually quite cheap to calculate), the `certificate` is the certificate used for signing. It has to be issued by the ownCloud Root Authority and it's CN needs to be permitted to perform the required action. The `signature` is then a signature of the `hashes` which can be verified using the `certificate`.
Steps to do in other PRs, this is already a quite huge one:
- Add nag screen in case the code check fails to ensure that administrators are aware of this.
- Add code verification also to OCC upgrade and unify display code more.
- Add enforced code verification to apps shipped from the appstore with a level of "official"
- Add enfocrced code verification to apps shipped from the appstore that were already signed in a previous release
- Add some developer documentation on how devs can request their own certificate
- Check when installing ownCloud
- Add support for CRLs to allow revoking certificates
**Note:** The upgrade checks are only run when the instance has a defined release channel of `stable` (defined in `version.php`). If you want to test this, you need to change the channel thus and then generate the core signature:
```
➜ master git:(add-integrity-checker) ✗ ./occ integrity:sign-core --privateKey=resources/codesigning/core.key --certificate=resources/codesigning/core.crt
Successfully signed "core"
```
Then increase the version and you should see something like the following:
![2015-11-04_12-02-57](https://cloud.githubusercontent.com/assets/878997/10936336/6adb1d14-82ec-11e5-8f06-9a74801c9abf.png)
As you can see a failed code check will not prevent the further update. It will instead just be a notice to the admin. In a next step we will add some nag screen.
For packaging stable releases this requires the following additional steps as a last action before zipping:
1. Run `./occ integrity:sign-core` once
2. Run `./occ integrity:sign-app` _for each_ app. However, this can be simply automated using a simple foreach on the apps folder.
Adding group Db to federation tests and ldap tests
Add group DB to Test_UrlGenerator
Adding group DB to trashbin and versions tests
Adding group DB to Test_Util_CheckServer for pg
In case the update server may deliver malicious content this would allow an adversary to inject arbitrary HTML into the response. So very bad stuff.
While signing the response would be better and something we can also do in the future (considering the code signing work), this is already a good first start.