This commit changes the way Rocket parses form items. In particular, it now
(liberally) validates form strings, returning a Bad Request on malformed inputs
and Unprocessable Entity on bad parses.
The 'FormItems' iterator was modified to accomodate this. The iterator is now
initialized using 'from': 'FormItems::from(form_string)'. The iterator can be
queried to check for a complete parse using either 'completed()' or
'exhausted()', the latter of which will consume valid keys/values and return
true only if the entire string was consumed.
The 'FromForm' trait now takes a mutable borrow to a 'FormItems' iterator.
The 'Form' and 'FormForm' implementation for 'Form' were modified to use the new
iterfaces and check for 'exhausted' after a parse, returning a Bad Request error
if the iterator cannot be exhausted.
Resolves#46.
This commit changes the routing algorithm. In particular, it enforces
precise matching of formats. With this change, a route with a specified
format only matches requests that have the same format specified. A
route with no format specified matches any request's format. This is
contrast to the previous behavior, where a route without a specified
format would match requests regardless of their format or whether one
was specified.
This commit also changes the following:
* The return type of the 'content_type' method of 'Request' is now
'Option<ContentType>'.
* The 'ContentType' request guard forwards when the request has no
specified ContentType.
* The 'add_header' and 'replace_header' methods take the header
argument generically.
Closes#120.
* The `unmanaged_state` lint emits a warning when a `State<T>` request
guard is used without an accompanying `manage` call for `T`.
* The `unmounted_route` lint emits a warning when a route declared via
a Rocket attribute is not mounted via a call to `mount`.
There is one known shortcoming of these lints at present: _any_ call to
`manage` or `mount` marks state/routes as managed/mounted. This can be
an issue when an application uses more than one `Rocket` instance, with
different calls to `mount` and `manage` in each. The lints should
perform their analyses on a per-instance basis.
A few interesting notes on this breakage:
* `Cookie` how has a lifetime. It should be `'static'` everywhere.
* The `SetCookie` header is no longer reexported.
* Instead, `Cookie` implements `Into<Header>` for Set-Cookie.
In #134, @tunz discovered that Rocket does not properly prevent path traversal
or local file inclusion attacks. The issue is caused by a failure to check for
some dangerous characters after decoding. In this case, the path separator '/'
was left as-is after decoding. As such, an attacker could construct a path with
containing any number of `..%2f..` sequences to traverse the file system.
This commit resolves the issue by ensuring that the decoded segment does not
contains any `/` characters. It further hardens the `FromSegments`
implementation by checking for additional risky characters: ':', '>', '<' as the
last character, and '\' on Windows. This is in addition to the already present
checks for '.' and '*' as the first character.
The behavior for a failing check has also changed. Previously, Rocket would skip
segments that contained illegal characters. In this commit, the implementation
instead return an error.
The `Error` type of the `PathBuf::FromSegment` implementations was changed to a
new `SegmentError` type that indicates the condition that failed.
Closes#134.
This commit also includes the following changes:
* `FromRequest` for `SocketAddr` implemented: extracts remote address.
* All built-in `FromRequest` implementations are documented.
* Request preprocessing overrides remote IP with value from X-Real-IP header.
* `MockRequest` allows setting the remote address with `remote()`.
Resolves#38.
This commit improves and changes the templating library in the following ways:
* Templates are now registered/loaded at initialization.
* No synchronization is required to read templates.
* All templates are properly loaded (fixes#122).
* Tera templates are given the proper name: `index`, not `index.html.tera`.
* Rendering tests added for both templating engines.
There is one breaking change:
* Tera templates are given the proper name: `index`, not `index.html.tera`.
