Repository navigation
Security
- Fix a denial of service with crafted multipart content (GHSA-wvq4-f5wr-7hj4). Adjacent boundaries copied the rest of the stream for each of them, and the number of parts, the nesting depth and the cost of header parsing were not bounded: a few KB of input could use hundreds of MB of memory and seconds of CPU. Upgrading is recommended for any application parsing untrusted multipart content.
- New limits: 1000 parts per content, nested parts included, and 16 nesting levels, configurable with
StreamedPart::setMaxParts()andStreamedPart::setMaxDepth()(nullfor no limit). Above them, aLimitExceededException(a\LogicException) is thrown. At most 100 options are parsed per header.
Fixes
- Raw UTF-8 in headers, such as non-ASCII field and file names sent by browsers, is no longer replaced with
?(#62) - Binary parts with a base64 or quoted-printable encoding and no charset are no longer corrupted by charset detection
- An unsupported charset no longer throws a
ValueErroron PHP 8 - Semicolons inside quoted header options no longer split them (#64)
- Header option names are looked up case-insensitively,
filename*takes precedence overfilename, and RFC 2231 continuations are supported - Underscores in Q encoded-words are decoded as spaces on PHP < 8.3
Behaviour changes
- Header values are always returned as valid UTF-8
- Non-text parts without a declared charset are returned as is, without charset detection
\"is unescaped in quoted header optionsgetHeaderValue()andgetMimeType()are trimmed- Malformed headers, common in emails, are parsed as before
Credits
Thanks to @aran112000 for #62, #64 and for reporting GHSA-wvq4-f5wr-7hj4.