1. Key concepts
FileSender 2.0 is built around the following concepts:
A transfer consists of one or more files sent to one (zero?) or more recipients
guest use (guest invitations)
2. privacy footprint
- full details of a transfer are logged in the audit log
- once a transfer expires, depending on the config settings and the preferences a user indicated an audit report in html or pdf can be sent to a user by email.
- after this, the audit log for that transfer can be deleted depending on config settings
- all activity is also logged to the statistics log but without any information identifying the particular end user or the details of what was transferred. What is logged? (overview!)
- relevant config directives
- common scenarios: “best privacy”
3. Maintenance mode
Version 2.0 supports a maintenance mode. This allows to interrupt the service for database upgrade or even server restart without breaking uploads. When switched on:
- all pages are replaced with the maintenance page
- webservice returns specific exception to all requests
- clients display a popup explaining what happens
- clients pause uploads and put all requests they were about to make in a stack
- clients starts to query the server on a regular basis to see if maintenance ended (server responding with no exception status)
- when server exits maintenance mode clients restart uploading and run stacked requests and remove maintenance popup
How to switch on maintenance mode
Do <fill out here how to do this>
Where to change the maintenance messages
Change the following language tags in your localised language file(s): $lang[‘undergoing_maintenance’] = ‘This application is under maintenance’; $lang[‘maintenance_autoresume’] = ‘Your operations will automatically resume when maintenance ends.’;
Sizing your installation — How much space do I need
Just get something that can easily expand, preferably in a matter of hours :)
4. Authentication and user account handling
FileSender has no user database and has no concept of user accounts.
What happens when a login session expires
There are two expiry timers on a user logon session. One controlled by FileSender on its side, in its simplesamlphp or shibboleth configuration. The other is controlled by the IdP the user uses to authenticate against.
The <…> is set in the SAML2 message received by
- Any uploads started during an active logon session get their own “upload session” authentication token. This allows the upload to finish even if the user’s logon session should expire.
- When a login session is expired and a user tries do do something, a message pops up informing the user of the expired session and inviting to re-logon
5. How uploads are handled
Any size with HTML5 chunked upload
fall-back for non-html5 browsers
This uses an html post. User can still select multiple files, when uploading a hidden iframe takes care of the upload. Progress event through polling the server. How often polling is done is defined by legacy_upload_progress_refresh_period.
TeraSender high speed upload module
Use of temporary files
<filesender>/tmp used for following tmp files:
- instance.secret which is the seed value used in the secure generation of random values used for fileUID, download link etc.
- used by dompdf to store tmp files when generating audit report
- NOT used for tmp storage of chunks: those are written directly in actual file on disk
xsrf token and instance.secret if instance.secret disappears a user session might break. question: do we need to implement auto-refresh of the instance.secret?
Difference html5 and non-html5 browsers
- no automatic transfer restart (uses html5 localstorage)
- no chunked upload (=limit of 2 GB per file)
Detection of stalled and corrupted chunks
We open file from zero, seek to chunk offset and write the data received in the chunk. So checks on file size of total file less meaningfull. You could only write last chunk and file size would still be good. Advantage: can restart any chunk.
Checks to ensure file integrity
- No hashing yet (slows things down too much)
- For each chunk: is received size identical to what client sent as “sent chunk length”? Client sends this in a header (not content-length header!)
- Check chunk size against max. chunksize set in config: can be smaller but not bigger than this
- At end of file upload we check the entire file size?
- check whether filesize on server is filesize client
- on wish list: chunk map or file integrity. For big files this map would be big. For each chunk sent client side we keep chunk offset and chunk data length. At end of upload, hash that, server does same during upload and check if all are equal. Means keeping data both sides, can become huge. Can store in flat file, put it in tmp directory “transferid.map”, one line per chunk. When transfer is labeled as “failed” after no data has come for multiple days, just remove the file. Writing entry in map file can only be done once chunk is successfully written to disk. At end of upload read map to check if all expected offsets have been uploaded. This is not really a map but more like an “uploaded offset list”. As long as we can detect holes in this list we can detect missing chunks.
- observation: no hash is browser limitation. But we have API. Researchers are the most likely to use API with either command line client or their own code and most likely to send the big files that you would like to integrity check. Can implement hashing server-side and make available via API in preparation for browser fucntionality supporting client-side hashing.
- reason we don’t do file integrity check now: with sending blob we pass on a pointer to the blob to the browser’s http client. We don’t have access to the data. If we want to hash it, we need to read it, it’s an intensive process and memory consuming. Workers time chunk size. Slows down upload considerably. File reader also doesn’t really support binary, so get string encoded. You have to create a byte array from that on which you can hash.
6. Language and internationalisation
- UTF8, multiple languages
- Uses browser preference for automatic language selection.
- describe how language is selected (use browser indication, use default language, hard coded default back to EN_AU
- describe what happens with language tags that are not translated
- User can select language in UI as well, if this is enabled in config
Can use multiple languages in emails as well (ask Etienne on implementation status and implemented algorithm)
default email language: use what sender is using in UI user can select other language but applies to _all_ emails (?)
Overriding the default language files
- describe how to modify language
- local overrides (directory in config directory
Creating and changing default language files
- improve language file that ships with filesender: send to filesender team as patch?
- add language
- which script to use to see which tags are not translated
7. UI Customisation
header footer logo
skinning using style sheets
See the directory
Each template uses “Foundation”. There are template overrides.
When you want to override css or add script, like for Fonts: create skin directory in www/skin and put it there. CSS must be named “styles.css”. In skin: skin/script.js and styles.css are immediately interpreted by FileSender. Can have other scripts, but must include them in your files or tweak templates/header.php to include them. Can copy it in config/templates/header.php and tweak there. Start of page with HTML headers etc. Can add scripts and styles you want there.
html template engine
8. Email bounce handling and SPF
examples of sendmail and postfix
Add possibility to forward feedback with non-detectable type but identified related target (recipient, guest) to person of choice.
Related config parameter is “relay_unknown_feedbacks” (string, defaults to “sender”) :
“sender” : relay to recipient’s transfer owner or guest owner
“admin” : relay to admin emails
“support” : relay to support_email (which itself must then be set to a valid email address)
“email@example.com” : choosen email address
Received feedback is forwarded as a message/rfc822 attachment.
###Email address validation: uses the PHP built-in facility for checking email address validity which these days works well. The PHPfunction used is filter_var with the filter FILTER_VALIDATE_EMAIL.
9. Multi-tenant hosting
Database initialisation and updates
Describe how the database is initialised for version 2.0 (from classes definiation) and how it’s updated automatically!
Workarounds implemented to work around various client quirks
mac_unzip_link (to work around Mac unsigned 32 bit default unzip client)
recipient Id includes transfer Id
- increase logging (log levels)
- firebug (!!)
Storage block concept
Can create storage blocks from different chunks of storage with filesender without lvm Use multiple mount points.
Revisit lateron: progress updates in web workers (for slow uploads with terasender on)
Log file analysis
In debug, why do I see error reports about simplesaml classes not being found?
Class SimpleSAML_Auth_Simple is needed
PHP runs autoloaders
FileSender’s autoloader look for SimpleSAML_Auth_Simple class in its classes sub-folder according the the layout we defined
It does not find it
He reports it in case this is a problem
PHP moves onto next autoloader (SimpleSamlPHP’s)
SimpleSamlPHP looks for the class by its own logic
It finds it
PHP is happy
PHP cannot magically consider a class to be part of a software from its name, not reliable enough, so it tries to resolve it using its default classpaths and provided autoloaders.
Securing your FileSender instance
- security mechanisms
- cross site scripting protection: csrf cookies (security-token)
- sp-session (authentication)
- chunk-upload-security token
- fileUID random generated,
- download URL protection
- API: api access key
- secure webserver config
Unique ID generation
Using random_uid_generation (6 calls to mt_rand , build X-string, put dashes.
Unique IDs were generated before without collision checking. Now we check for that until we get a real unique one.
Requirements for use of the Selenium Tests
- A SauceLabs. After is has been created a API key can be requested
- A running installation of FileSender
- Optional: When the installation is running locally or behind a NAT-solution, Sauce Connect needs to be set up first
- A terminal window with PHP command line
- Firefox: For writing tests more easily
###Requesting API key Log in to saucelabs.com. Go to Dashboard, and click on your name that you have filled in during sign-up in the bottom right corner. Click User Settings. Under Access Key click Show to see your access key. Copy and paste it into an empty text file. We’ll be needing this later.
###Setting up Sauce Connect When the FileSender installation isn’t accessible on a public domain, you are still able to run it on Sauce Labs using a tunnel. To do this download the Sauce Connect Binary, found at: https://wiki.saucelabs.com/display/DOCS/Setting+Up+Sauce+Connect Unzip the file, then go to the location where the file has been exstracted using a terminal window. Then enter the following command:
`bin/sc --user <User Name Sauce Labs> --api-key <Api key Sauce Labs>`
For Windows enter the following command instead:
`bin/sc.exe --user <User Name Sauce Labs> --api-key <Api key Sauce Labs>`
###Setting up API key Go to the API project, and open the config.php file. Add / replace the following values:
$config['sauce_username'] = <User Name Sauce Labs>
$config['sauce_access_key'] = <Api key Sauce Labs>
###Other config settings The following configuration options need to be set in the config.php, these will be overwritten temporarily during the tests to check if the changes made by them still work
The values from these fields van be copied from the default values configuration located in ‘/includes/ConfigDefaults.php’ or set to a desired value.
Open a terminal window in the project
To run the selenium tests, enter the following command:
`vendor/phpunit/phpunit/phpunit --testsuite="Selenium Test Suite"`
The tests will appear one by one on the Sauce Labs dashboard. When a tests fails, it will be colored red. Click on the test to find out more on why it failed
Setting up Firefox plugin
To simplify the writing of tests, we have put an ‘selenium-ide.xpi’ file in the Docker Repository. This can be found in the /plugins/ folder
This is a Firefox plug-in and can be installed by dragging it into a new Firefox instance. If Firefox rejects the plug-in, it could be because of a setting in the configuration of Firefox. This can be solved by going to about:config and setting xpinstall.signatures.required to false
Using the Firefox plug-in
Start the Firefox plugin ( by pressing the button in the Plug-in bar in the top right corner). When the plug-in is booted, a window will appear. Press the record button in the top right cornor of this new window. After that go to a tab in Firefox where the current FileSender installation is running and perform the steps you want to test. While you are doing this, the same steps will appear in the plugin window. When you are finished performing the steps, go back to the plug-in window and click on File ‘Export Test Case As’ Php / Codebridge Tests Anonymous. De plug-in will ask where to put the new test. You can save it directly in the selenium_test/anonymous_tests folder in the FileSender API project. This test will be run when the Test Suite is activated.
In the previous segment we discussed setting up an ‘Anonymous Test’. This means that the user wasn’t logged in when the test started. To start a test form the dashboard of a test user, we can create a ‘User Test’. This can be created by first logging in in the Firefox window before creating the tests. The test can then be exported by going to File ‘Export Test Case As’ Php / Codebridge Tests User. The exported file can then be saved to the selenium_tests/user_tests folder. To configure under which user the tests should be running, you can modify the credentials in the doLogin function in selenium_tests/SeleniumUserTest.php
Unfortunatly the test that are written by the Firefox Plug-in don’t always work. The tests that are written by this plugin don’t take into account the loading times between the actions the tests perform. To account for this, time-outs need to be set-up. Triggers can also be set so the test will wait for a certain element to become available. This as already been done in some of the current tests. There is also documentation available on-line on how to write Selenium Tests in Php.
- Modern browsers supporting the WebCryptoAPI
- File selection process remains the same
- Encryption checkbox is checked, a password is chosen ( or generated ). The password distribution needs to be done by the users since we will not even receive it on the API
- The encryption process encrypts each chunk as it is send to the API. Each chunk increases a little in size compared to the original. The WebCryptoAPI adds a checksum, no larger then 16 bytes and a IV (a salt, 16 bytes) is added to each chunk.
- The API saves the file(chunks) like usual and stores a encryption flag in the transfer
- The download process is the encryption process in reverse. The file is downloaded as a whole, chunked, then decrypted with the password provided and the IV stored in each chunk
- Once each chunk is decrypted the chunks are merged into a single blob wich is then handed to the browser as a file download
- RAM size is a limiting factor to decrypt each chunk and hand the download over as a single file since this happens in memory, in the future this can be enhanced and made streamable, however at the moment of writing only the latest dev version of chrome supports this feature