From 0c060ead5035261d952a2503e9158b20db62e53b Mon Sep 17 00:00:00 2001 From: jrparks <7059873+jrparks@users.noreply.github.com> Date: Wed, 11 Nov 2026 16:00:27 -0600 Subject: [PATCH] sqlite3-parser: don't assume input is UTF-8 (CVE-2025-47736) Parsing non-UTF-8 input hit from_utf8_unchecked in from_bytes and keyword lookup, which can panic on invalid input. Use from_utf8_lossy for identifiers (as upstream gwenn/lemon-rs 14f422a does) and treat invalid UTF-8 as a non-keyword. Adds regression tests. --- vendored/sqlite3-parser/src/dialect/mod.rs | 28 ++++++++++++++++++---- 1 file changed, 24 insertions(+), 4 deletions(-) diff --git a/vendored/sqlite3-parser/src/dialect/mod.rs b/vendored/sqlite3-parser/src/dialect/mod.rs index e0266f10e4..2afe4ddc58 100644 --- a/vendored/sqlite3-parser/src/dialect/mod.rs +++ b/vendored/sqlite3-parser/src/dialect/mod.rs @@ -57,17 +57,19 @@ impl TokenType { } } +// Input bytes are not validated as UTF-8 before parsing, so never assume they are. +// Upstream gwenn/lemon-rs 14f422a (CVE-2025-47736). fn from_bytes(bytes: &[u8]) -> String { - unsafe { str::from_utf8_unchecked(bytes).to_owned() } + String::from_utf8_lossy(bytes).to_string() } include!(concat!(env!("OUT_DIR"), "/keywords.rs")); pub(crate) const MAX_KEYWORD_LEN: usize = 17; pub fn keyword_token(word: &[u8]) -> Option { - KEYWORDS - .get(UncasedStr::new(unsafe { str::from_utf8_unchecked(word) })) - .cloned() + // Invalid UTF-8 can never spell a keyword. + let word = str::from_utf8(word).ok()?; + KEYWORDS.get(UncasedStr::new(word)).cloned() } pub(crate) fn is_identifier(name: &str) -> bool { @@ -403,3 +405,21 @@ impl TokenType { } } } + +#[cfg(test)] +mod invalid_utf8_tests { + use super::{from_bytes, keyword_token}; + + // CVE-2025-47736: non-UTF-8 input must not panic the parser. + #[test] + fn invalid_utf8_is_not_a_keyword() { + assert_eq!(keyword_token(&[0xff, 0xfe, b'x']), None); + assert_eq!(keyword_token(b"select"), keyword_token(b"SELECT")); + assert!(keyword_token(b"select").is_some()); + } + + #[test] + fn invalid_utf8_bytes_are_replaced() { + assert_eq!(from_bytes(&[b'a', 0xff, b'b']), "a\u{fffd}b"); + } +}