|From the Red Hat bugzilla:
libcurl is vulnerable to a data injection attack for certain protocols
through control characters embedded or percent-encoded in URLs.
When parsing URLs, libcurl's parser is very laxed and liberal and only
parses as little as possible and lets as much as possible through as long as it can figure out what to do.
In the specific process when libcurl extracts the file path part from a
given URL, it didn't always verify the data or escape control characters
properly before it passed the file path on to the protocol-specific code
that then would use it for its protocol business.
This passing through of control characters could be exploited by someone who would be able to pass in a handicrafted URL to libcurl. Lots of libcurl using applications let users enter URLs in one form or another and not all of these check the input carefully to prevent malicious ones.
A malicious user might pass in %0d%0a to get treated as CR LF by libcurl, and by using this fact a user can trick for example a POP3 client to delete a message instead of getting it or trick an SMTP server to send an unintended message.
This vulnerability can be used to fool libcurl with the following protocols: IMAP, POP3 and SMTP.
This flaw only affects curl versions 7.20.0 up to and including 7.23.1 It is corrected in 7.24.0