Bugfix #1702 - Unicode escape parser for Strand. (#1888)

lib/src/sql/strand.rs

@@ -2,16 +2,15 @@ use crate::sql::error::Error::Parser;
 use crate::sql::error::IResult;
 use crate::sql::escape::escape_str;
 use nom::branch::alt;
-use nom::bytes::complete::escaped_transform;
-use nom::bytes::complete::is_not;
-use nom::bytes::complete::take_while_m_n;
+use nom::bytes::complete::{escaped_transform, is_not, tag, take, take_while_m_n};
 use nom::character::complete::char;
 use nom::combinator::value;
+use nom::sequence::preceded;
 use nom::Err::Failure;
 use serde::{Deserialize, Serialize};
 use std::fmt::{self, Display, Formatter};
-use std::ops;
 use std::ops::Deref;
+use std::ops::{self, RangeInclusive};
 use std::str;
 
 pub(crate) const TOKEN: &str = "$surrealdb::private::sql::Strand";
@@ -22,7 +21,8 @@ const SINGLE_ESC: &str = r#"\'"#;
 const DOUBLE: char = '"';
 const DOUBLE_ESC: &str = r#"\""#;
 
-const SURROGATES: [u32; 2] = [55296, 57343];
+const LEADING_SURROGATES: RangeInclusive<u16> = 0xD800..=0xDBFF;
+const TRAILING_SURROGATES: RangeInclusive<u16> = 0xDC00..=0xDFFF;
 
 #[derive(Clone, Debug, Default, Eq, PartialEq, PartialOrd, Serialize, Deserialize, Hash)]
 #[serde(rename = "$surrealdb::private::sql::Strand")]
@@ -111,7 +111,7 @@ fn strand_single(i: &str) -> IResult<&str, String> {
 		is_not(SINGLE_ESC),
 		'\\',
 		alt((
-			strand_unicode,
+			char_unicode,
 			value('\u{5c}', char('\\')),
 			value('\u{27}', char('\'')),
 			value('\u{2f}', char('/')),
@@ -132,7 +132,7 @@ fn strand_double(i: &str) -> IResult<&str, String> {
 		is_not(DOUBLE_ESC),
 		'\\',
 		alt((
-			strand_unicode,
+			char_unicode,
 			value('\u{5c}', char('\\')),
 			value('\u{22}', char('\"')),
 			value('\u{2f}', char('/')),
@@ -147,30 +147,59 @@ fn strand_double(i: &str) -> IResult<&str, String> {
 	Ok((i, v))
 }
 
-fn strand_unicode(i: &str) -> IResult<&str, char> {
-	// Read the \u character
-	let (i, _) = char('u')(i)?;
-	// Let's read the next 6 ascii hexadecimal characters
+fn char_unicode(i: &str) -> IResult<&str, char> {
+	preceded(char('u'), alt((char_unicode_bracketed, char_unicode_bare)))(i)
+}
+
+// \uABCD or \uDBFF\uDFFF (surrogate pair)
+fn char_unicode_bare(i: &str) -> IResult<&str, char> {
+	// Take exactly 4 bytes
+	let (i, v) = take(4usize)(i)?;
+	// Parse them as hex, where an error indicates invalid hex digits
+	let v: u16 = u16::from_str_radix(v, 16).map_err(|_| Failure(Parser(i)))?;
+
+	if LEADING_SURROGATES.contains(&v) {
+		let leading = v;
+
+		// Read the next \u.
+		let (i, _) = tag("\\u")(i)?;
+		// Take exactly 4 more bytes
+		let (i, v) = take(4usize)(i)?;
+		// Parse them as hex, where an error indicates invalid hex digits
+		let trailing = u16::from_str_radix(v, 16).map_err(|_| Failure(Parser(i)))?;
+		if !TRAILING_SURROGATES.contains(&trailing) {
+			return Err(Failure(Parser(i)));
+		}
+		// Compute the codepoint.
+		// https://datacadamia.com/data/type/text/surrogate#from_surrogate_to_character_code
+		let codepoint = 0x10000
+			+ ((leading as u32 - *LEADING_SURROGATES.start() as u32) << 10)
+			+ trailing as u32
+			- *TRAILING_SURROGATES.start() as u32;
+		// Convert to char
+		let v = char::from_u32(codepoint).ok_or(Failure(Parser(i)))?;
+		// Return the char
+		Ok((i, v))
+	} else {
+		// We can convert this to char or error in the case of invalid Unicode character
+		let v = char::from_u32(v as u32).ok_or(Failure(Parser(i)))?;
+		// Return the char
+		Ok((i, v))
+	}
+}
+
+// \u{10ffff}
+fn char_unicode_bracketed(i: &str) -> IResult<&str, char> {
+	// Read the { character
+	let (i, _) = char('{')(i)?;
+	// Let's up to 6 ascii hexadecimal characters
 	let (i, v) = take_while_m_n(1, 6, |c: char| c.is_ascii_hexdigit())(i)?;
-	// We can convert this to u32 as we only have 6 chars
-	let v = match u32::from_str_radix(v, 16) {
-		// We found an invalid unicode sequence
-		Err(_) => return Err(Failure(Parser(i))),
-		// The unicode sequence was valid
-		Ok(v) => match v {
-			// This is a surrogate, so convert to a space
-			v if v >= SURROGATES[0] && v <= SURROGATES[1] => 32,
-			// This is a valid UTF-8 / UTF-16 character
-			_ => v,
-		},
-	};
-	// We can convert this to char as we know it is valid
-	let v = match std::char::from_u32(v) {
-		// We found an invalid unicode sequence
-		None => return Err(Failure(Parser(i))),
-		// The unicode sequence was valid
-		Some(v) => v,
-	};
+	// We can convert this to u32 as the max is 0xffffff
+	let v = u32::from_str_radix(v, 16).unwrap();
+	// We can convert this to char or error in the case of invalid Unicode character
+	let v = char::from_u32(v).ok_or(Failure(Parser(i)))?;
+	// Read the } character
+	let (i, _) = char('}')(i)?;
 	// Return the char
 	Ok((i, v))
 }
@@ -239,4 +268,32 @@ mod tests {
 		assert_eq!("'te\"st\n\tand\u{08}some\u{05d9}'", format!("{}", out));
 		assert_eq!(out, Strand::from("te\"st\n\tand\u{08}some\u{05d9}"));
 	}
+
+	#[test]
+	fn strand_fuzz_escape() {
+		for n in (0..=char::MAX as u32).step_by(101) {
+			if let Some(c) = char::from_u32(n) {
+				let expected = format!("a{c}b");
+
+				let utf32 = format!("\"a\\u{{{n:x}}}b\"");
+				let (rest, s) = strand(&utf32).unwrap();
+				assert_eq!(rest, "");
+				assert_eq!(s.as_str(), &expected);
+
+				let mut utf16 = String::with_capacity(16);
+				utf16 += "\"a";
+				let mut buf = [0; 2];
+				for &mut n in c.encode_utf16(&mut buf) {
+					utf16 += &format!("\\u{n:04x}");
+				}
+				utf16 += "b\"";
+				let (rest, s) = strand(&utf16).unwrap();
+				assert_eq!(rest, "");
+				assert_eq!(s.as_str(), &expected);
+			}
+		}
+
+		// Unpaired surrogate.
+		assert!(strand("\"\\u{DBFF}\"").is_err());
+	}
 }