summaryrefslogtreecommitdiffstats
path: root/rust/vendor/rusticata-macros/src/combinator.rs
diff options
context:
space:
mode:
Diffstat (limited to 'rust/vendor/rusticata-macros/src/combinator.rs')
-rw-r--r--rust/vendor/rusticata-macros/src/combinator.rs231
1 files changed, 231 insertions, 0 deletions
diff --git a/rust/vendor/rusticata-macros/src/combinator.rs b/rust/vendor/rusticata-macros/src/combinator.rs
new file mode 100644
index 0000000..6b6a8b3
--- /dev/null
+++ b/rust/vendor/rusticata-macros/src/combinator.rs
@@ -0,0 +1,231 @@
+//! General purpose combinators
+
+use nom::bytes::streaming::take;
+use nom::combinator::map_parser;
+use nom::error::{make_error, ErrorKind, ParseError};
+use nom::{IResult, Needed, Parser};
+use nom::{InputIter, InputTake};
+use nom::{InputLength, ToUsize};
+
+#[deprecated(since = "3.0.1", note = "please use `be_var_u64` instead")]
+/// Read an entire slice as a big-endian value.
+///
+/// Returns the value as `u64`. This function checks for integer overflows, and returns a
+/// `Result::Err` value if the value is too big.
+pub fn bytes_to_u64(s: &[u8]) -> Result<u64, &'static str> {
+ let mut u: u64 = 0;
+
+ if s.is_empty() {
+ return Err("empty");
+ };
+ if s.len() > 8 {
+ return Err("overflow");
+ }
+ for &c in s {
+ let u1 = u << 8;
+ u = u1 | (c as u64);
+ }
+
+ Ok(u)
+}
+
+/// Read the entire slice as a big endian unsigned integer, up to 8 bytes
+#[inline]
+pub fn be_var_u64<'a, E: ParseError<&'a [u8]>>(input: &'a [u8]) -> IResult<&'a [u8], u64, E> {
+ if input.is_empty() {
+ return Err(nom::Err::Incomplete(Needed::new(1)));
+ }
+ if input.len() > 8 {
+ return Err(nom::Err::Error(make_error(input, ErrorKind::TooLarge)));
+ }
+ let mut res = 0u64;
+ for byte in input {
+ res = (res << 8) + *byte as u64;
+ }
+
+ Ok((&b""[..], res))
+}
+
+/// Read the entire slice as a little endian unsigned integer, up to 8 bytes
+#[inline]
+pub fn le_var_u64<'a, E: ParseError<&'a [u8]>>(input: &'a [u8]) -> IResult<&'a [u8], u64, E> {
+ if input.is_empty() {
+ return Err(nom::Err::Incomplete(Needed::new(1)));
+ }
+ if input.len() > 8 {
+ return Err(nom::Err::Error(make_error(input, ErrorKind::TooLarge)));
+ }
+ let mut res = 0u64;
+ for byte in input.iter().rev() {
+ res = (res << 8) + *byte as u64;
+ }
+
+ Ok((&b""[..], res))
+}
+
+/// Read a slice as a big-endian value.
+#[inline]
+pub fn parse_hex_to_u64<S>(i: &[u8], size: S) -> IResult<&[u8], u64>
+where
+ S: ToUsize + Copy,
+{
+ map_parser(take(size.to_usize()), be_var_u64)(i)
+}
+
+/// Apply combinator, automatically converts between errors if the underlying type supports it
+pub fn upgrade_error<I, O, E1: ParseError<I>, E2: ParseError<I>, F>(
+ mut f: F,
+) -> impl FnMut(I) -> IResult<I, O, E2>
+where
+ F: FnMut(I) -> IResult<I, O, E1>,
+ E2: From<E1>,
+{
+ move |i| f(i).map_err(nom::Err::convert)
+}
+
+/// Create a combinator that returns the provided value, and input unchanged
+pub fn pure<I, O, E: ParseError<I>>(val: O) -> impl Fn(I) -> IResult<I, O, E>
+where
+ O: Clone,
+{
+ move |input: I| Ok((input, val.clone()))
+}
+
+/// Return a closure that takes `len` bytes from input, and applies `parser`.
+pub fn flat_take<I, C, O, E: ParseError<I>, F>(
+ len: C,
+ mut parser: F,
+) -> impl FnMut(I) -> IResult<I, O, E>
+where
+ I: InputTake + InputLength + InputIter,
+ C: ToUsize + Copy,
+ F: Parser<I, O, E>,
+{
+ // Note: this is the same as `map_parser(take(len), parser)`
+ move |input: I| {
+ let (input, o1) = take(len.to_usize())(input)?;
+ let (_, o2) = parser.parse(o1)?;
+ Ok((input, o2))
+ }
+}
+
+/// Take `len` bytes from `input`, and apply `parser`.
+pub fn flat_takec<I, O, E: ParseError<I>, C, F>(input: I, len: C, parser: F) -> IResult<I, O, E>
+where
+ C: ToUsize + Copy,
+ F: Parser<I, O, E>,
+ I: InputTake + InputLength + InputIter,
+ O: InputLength,
+{
+ flat_take(len, parser)(input)
+}
+
+/// Helper macro for nom parsers: run first parser if condition is true, else second parser
+pub fn cond_else<I, O, E: ParseError<I>, C, F, G>(
+ cond: C,
+ mut first: F,
+ mut second: G,
+) -> impl FnMut(I) -> IResult<I, O, E>
+where
+ C: Fn() -> bool,
+ F: Parser<I, O, E>,
+ G: Parser<I, O, E>,
+{
+ move |input: I| {
+ if cond() {
+ first.parse(input)
+ } else {
+ second.parse(input)
+ }
+ }
+}
+
+/// Align input value to the next multiple of n bytes
+/// Valid only if n is a power of 2
+pub const fn align_n2(x: usize, n: usize) -> usize {
+ (x + (n - 1)) & !(n - 1)
+}
+
+/// Align input value to the next multiple of 4 bytes
+pub const fn align32(x: usize) -> usize {
+ (x + 3) & !3
+}
+
+#[cfg(test)]
+mod tests {
+ use super::{align32, be_var_u64, cond_else, flat_take, pure};
+ use nom::bytes::streaming::take;
+ use nom::number::streaming::{be_u16, be_u32, be_u8};
+ use nom::{Err, IResult, Needed};
+
+ #[test]
+ fn test_be_var_u64() {
+ let res: IResult<&[u8], u64> = be_var_u64(b"\x12\x34\x56");
+ let (_, v) = res.expect("be_var_u64 failed");
+ assert_eq!(v, 0x123456);
+ }
+
+ #[test]
+ fn test_flat_take() {
+ let input = &[0x00, 0x01, 0xff];
+ // read first 2 bytes and use correct combinator: OK
+ let res: IResult<&[u8], u16> = flat_take(2u8, be_u16)(input);
+ assert_eq!(res, Ok((&input[2..], 0x0001)));
+ // read 3 bytes and use 2: OK (some input is just lost)
+ let res: IResult<&[u8], u16> = flat_take(3u8, be_u16)(input);
+ assert_eq!(res, Ok((&b""[..], 0x0001)));
+ // read 2 bytes and a combinator requiring more bytes
+ let res: IResult<&[u8], u32> = flat_take(2u8, be_u32)(input);
+ assert_eq!(res, Err(Err::Incomplete(Needed::new(2))));
+ }
+
+ #[test]
+ fn test_flat_take_str() {
+ let input = "abcdef";
+ // read first 2 bytes and use correct combinator: OK
+ let res: IResult<&str, &str> = flat_take(2u8, take(2u8))(input);
+ assert_eq!(res, Ok(("cdef", "ab")));
+ // read 3 bytes and use 2: OK (some input is just lost)
+ let res: IResult<&str, &str> = flat_take(3u8, take(2u8))(input);
+ assert_eq!(res, Ok(("def", "ab")));
+ // read 2 bytes and a use combinator requiring more bytes
+ let res: IResult<&str, &str> = flat_take(2u8, take(4u8))(input);
+ assert_eq!(res, Err(Err::Incomplete(Needed::Unknown)));
+ }
+
+ #[test]
+ fn test_cond_else() {
+ let input = &[0x01][..];
+ let empty = &b""[..];
+ let a = 1;
+ fn parse_u8(i: &[u8]) -> IResult<&[u8], u8> {
+ be_u8(i)
+ }
+ assert_eq!(
+ cond_else(|| a == 1, parse_u8, pure(0x02))(input),
+ Ok((empty, 0x01))
+ );
+ assert_eq!(
+ cond_else(|| a == 1, parse_u8, pure(0x02))(input),
+ Ok((empty, 0x01))
+ );
+ assert_eq!(
+ cond_else(|| a == 2, parse_u8, pure(0x02))(input),
+ Ok((input, 0x02))
+ );
+ assert_eq!(
+ cond_else(|| a == 1, pure(0x02), parse_u8)(input),
+ Ok((input, 0x02))
+ );
+ let res: IResult<&[u8], u8> = cond_else(|| a == 1, parse_u8, parse_u8)(input);
+ assert_eq!(res, Ok((empty, 0x01)));
+ }
+
+ #[test]
+ fn test_align32() {
+ assert_eq!(align32(3), 4);
+ assert_eq!(align32(4), 4);
+ assert_eq!(align32(5), 8);
+ assert_eq!(align32(5usize), 8);
+ }
+}