Since 4.12 the use of typehints and return types has caused issues with
values fetched directly from config without validation. This has lead to
upgrade woes in a minor version (#10721) with no immediate recourse
other than manual system intervention.
To use types, we should ensure types, leaving a stable API that won't
error on a bad value - or should give a thoughtful and directive error
message if so.
Issue #10721 summary:
SessionMiddleware runs before FlushMiddleware
SessionMiddleware causes a PHP fatal error passing `null` to a `string`
parameter.
`null` comes from config, because default string value doesn't exist. We
need flush for this - but system execution never makes it that far.
A potential account hijacking may happen if an attacker has physical access to
victim's computer to perform session fixation. Also possible if the targeted application contains an XSS vulnerability.
Requires the victim to click the password reset link sent to their email.
If all the above happens, attackers may reset the password before the actual user does that.
A potential account hijacking may happen if an attacker has physical access to
victim's computer to perform session fixation. Also possible if the targeted application contains an XSS vulnerability.
Requires the victim to click the password reset link sent to their email.
If all the above happens, attackers may reset the password before the actual user does that.
- Replace session name lookup with function to also check secure cookies
- Added timeout which defaults to 0 (same as PHP)
- Removed php7 style of session_start from PR
- moved session_start into headers sent block to prevent warnings.
If a session already exists, and Session::start() isn’t called until
after a large enough block of content is output, then headers_sent()
will be false. The previous code prevented the session from being
started in this case. That might makes sense for the creation of a new
session, but it prevent legitimate access to an existing session.
This mostly manifested when running debugging tools such as showqueries,
which may output content before the session is started.
Fixes regression from 3.x, where sessions where lazy started as required:
Either because an existing session identifier was sent through with the request,
or because new session data needed to be persisted as part of the request execution.
Without this lazy starting, *every* request will get a session,
which makes all those responses uncacheable by HTTP layers.
Note that 4.x also changed the $data vs. $changedData payloads:
In 3.x, they both contained key/value pairs.
In 4.x, $data contains key/value, while $changedData contains key/boolean to declare isChanged.
While this reduces duplication in the class, it also surfaced a bug which was latent in 3.x:
When an existing session is lazily resumed via start(), $data is set back to an empty array.
In 3.x, any changed data before this point was *also* retained in $changedData,
ensuring it gets merged into existing $_SESSION data.
In 4.x, this clears out data - hence the need for a more complex merge logic.
Since isset($this->data) is no longer an accurate indicator of a started session,
we introduce a separate $this->started flag.
Note that I've chosen not to make lazy an opt-in (e.g. via start($request, $lazy=false)).
We already have a distinction between lazy starting via init(), and force starting via start().
BUG Fix up test regressions
FIX director references to request object
API Move all middlewares to common namespace
API Implement RequestHandlerMiddlewareAdapter
ENHANCEMENT Improve IP address parsing
Fix up PHPDoc / psr2 linting
BUG Fix property parsing in TrustedProxyMiddleware
BUG Fix Director::is_https()
NEW: Pass HTTPRequest to session
NEW: Pass HTTPReuqest optionally to Director statics
The session handler now expects to operate on a specific
HTTPRequest object.
