- Reported
-
- Issued
-
- Package
-
url
(crates.io)
- Type
-
Vulnerability
- Categories
-
- Keywords
-
#stack-overflow
#recursion
#blob
- Aliases
-
- References
-
- Patched
-
- Affected Functions
- Version
url::Url::origin
-
url::quirks::origin
-
Description
In url 2.5.8 and earlier, Url::origin() handled a blob: URL by parsing its path as another URL and calling itself on the result. It did so regardless of the inner URL's scheme, blob included. Each blob: prefix cost one stack frame, with no limit on nesting depth. Url::parse() accepts a URL with 10,000 blob: prefixes followed by https://example.com/ (50,020 bytes in total); calling Url::origin() on it overflowed the default 2 MiB stack of a spawned thread.
A stack overflow aborts the whole process, and catch_unwind() does not catch it. One such URL crashes any program that calls Url::origin() on untrusted input. url::quirks::origin() also calls Url::origin(). For example, Servo's URL.origin DOM getter calls url::quirks::origin(), so a script on a page can trigger the overflow.
url 2.5.9 follows the URL Standard: Url::origin() takes the origin of the inner URL only when its scheme is http, https, or file, returns an opaque origin otherwise, and never calls itself. A blob: URL with another inner scheme that previously produced a tuple origin, such as blob:ws://... or blob:blob:https://..., now returns an opaque origin.
Advisory available under CC0-1.0
license.