Please don't open a public issue. Report it privately instead: on the repository's "Security" tab, choose "Report a vulnerability", or go straight to the form. Tell me what you found, how to reproduce it, and what an attacker could do with it. The input that triggers it is the most useful thing you can send.
You'll hear back within a week. Once there's a fix, it ships with an advisory crediting you, unless you'd rather not be named.
Parsec reads HTML that nobody vouches for. Anything that lets that HTML, or a selector, do more than produce a tree is in scope. For example:
- input that makes the parser read or write memory it shouldn't, through
unsafecode or otherwise; - input that crashes it: a panic, a stack overflow, an abort;
- input that hangs it, or makes it take time or memory far out of proportion to its size, which is a denial of service for anyone parsing pages from the web;
- a selector or DOM edit that does any of the above.
Output that differs from what a browser builds, with nothing worse, is a bug rather than a security problem. Please open an ordinary issue for it.
| Version | Fixes |
|---|---|
| 0.1.x | yes |
| 0.0.0 | no: it's an empty placeholder on crates.io |
Fixes go into the newest minor release, as a patch release, on every registry at once. Until Parsec reaches 1.0, older minor releases don't get backports, so when 0.2 comes out, 0.1 stops getting them.
A release is a signed tag, and on its GitHub release page the crates
(parsec-html-VERSION.crate and the five it's built from), their
SHA256SUMS and the provenance bundle (parsec-vVERSION.intoto.jsonl).
Each can be checked on its own.
The tag is signed with a maintainer's SSH key, listed in
.github/allowed_signers. Take that file from
main rather than from the tag, as the release workflow does:
git fetch origin tag v0.1.0
git show origin/main:.github/allowed_signers > allowed_signers
git -c gpg.ssh.allowedSignersFile=allowed_signers tag -v v0.1.0The crates were built by the release workflow from that tag, and GitHub signed a record of each:
gh attestation verify parsec-html-0.1.0.crate --repo parsec-html/parsec \
--signer-workflow parsec-html/parsec/.github/workflows/release.ymlThe C libraries (parsec-capi-VERSION-TARGET.tar.gz or .zip, one
per platform, listed in parsec-capi-SHA256SUMS) are built and attested
the same way, by their own workflow:
gh attestation verify parsec-capi-0.1.0-x86_64-unknown-linux-gnu.tar.gz \
--repo parsec-html/parsec \
--signer-workflow parsec-html/parsec/.github/workflows/release-capi.ymlThey're attached only once the tag's signature has checked out. Unlike the crates, they aren't yet checked to rebuild byte for byte.
crates.io has the same bytes as the release: the release workflow
fails if a published crate's checksum isn't the one in SHA256SUMS. To
check one yourself:
curl -s https://crates.io/api/v1/crates/parsec-html/0.1.0 | jq -r .version.checksumPyPI and npm packages are published by trusted publishing from
release-python.yml and release-npm.yml, with the attestations PyPI
shows on each file and npm's provenance statement, and only once the
tag's signature has checked out:
npm audit signatures # in a project that depends on parsec-htmlA rebuild gives the same file, byte for byte. At the tag, run
git checkout v0.1.0
scripts/package-twice.shIt packages every crate twice from fresh checkouts, fails if they differ,
and prints their SHA-256s, which must match the release's SHA256SUMS.
It needs cargo 1.90 or newer, which packages a workspace's crates against
each other before they're published; past that, the version of Rust
doesn't change the packages. CI runs the
same script on every change, so a change that would break this fails there
first.