Skip to content

Commit

Permalink
detect/transforms: in place modifications of buffers
Browse files Browse the repository at this point in the history
As is the case when chaining multiple transforms.
Avoids using memcpy in these cases.

Add tests for these cases.

Ticket: 7409
  • Loading branch information
catenacyber authored and victorjulien committed Dec 3, 2024
1 parent e9173f3 commit b58b886
Show file tree
Hide file tree
Showing 7 changed files with 99 additions and 38 deletions.
37 changes: 20 additions & 17 deletions rust/src/detect/transforms/compress_whitespace.rs
Original file line number Diff line number Diff line change
Expand Up @@ -35,21 +35,16 @@ unsafe extern "C" fn compress_whitespace_setup(

fn compress_whitespace_transform_do(input: &[u8], output: &mut [u8]) -> u32 {
let mut nb = 0;
// seems faster than writing one byte at a time via
// for (i, o) in input.iter().filter(|c| !matches!(*c, b'\t' | b'\n' | b'\x0B' | b'\x0C' | b'\r' | b' ')).zip(output)
for subslice in
input.split_inclusive(|c| matches!(*c, b'\t' | b'\n' | b'\x0B' | b'\x0C' | b'\r' | b' '))
{
// a subslice of length 1 not at the beginning is a space following another space
if nb == 0
|| subslice.len() > 1
|| !matches!(
subslice[0],
b'\t' | b'\n' | b'\x0B' | b'\x0C' | b'\r' | b' '
)
{
output[nb..nb + subslice.len()].copy_from_slice(subslice);
nb += subslice.len();
let mut space = false;
for c in input {
if !matches!(*c, b'\t' | b'\n' | b'\x0B' | b'\x0C' | b'\r' | b' ') {
output[nb] = *c;
nb += 1;
space = false;
} else if !space {
output[nb] = *c;
nb += 1;
space = true;
}
}
return nb as u32;
Expand Down Expand Up @@ -142,14 +137,22 @@ mod tests {
exp.len() as u32
);
assert_eq!(&out[..exp.len()], exp);
let buf = b"I \t J";
let mut buf = Vec::new();
buf.extend_from_slice(b"I \t J");
let mut out = vec![0; buf.len()];
let exp = b"I J";
assert_eq!(
compress_whitespace_transform_do(buf, &mut out),
compress_whitespace_transform_do(&buf, &mut out),
exp.len() as u32
);
assert_eq!(&out[..exp.len()], exp);
// test in place
let still_buf = unsafe { std::slice::from_raw_parts(buf.as_ptr(), buf.len()) };
assert_eq!(
compress_whitespace_transform_do(still_buf, &mut buf),
exp.len() as u32
);
assert_eq!(&still_buf[..exp.len()], b"I J");
}

#[test]
Expand Down
28 changes: 21 additions & 7 deletions rust/src/detect/transforms/dotprefix.rs
Original file line number Diff line number Diff line change
Expand Up @@ -34,24 +34,32 @@ unsafe extern "C" fn dot_prefix_setup(
}

fn dot_prefix_transform_do(input: &[u8], output: &mut [u8]) {
if std::ptr::eq(output.as_ptr(), input.as_ptr()) {
output.copy_within(0..input.len(), 1);
} else {
output[1..].copy_from_slice(input);
}
output[0] = b'.';
output[1..].copy_from_slice(input);
}

#[no_mangle]
unsafe extern "C" fn dot_prefix_transform(buffer: *mut c_void, _ctx: *mut c_void) {
let input = InspectionBufferPtr(buffer);
let input_len = InspectionBufferLength(buffer);
if input.is_null() || input_len == 0 {
if input_len == 0 {
return;
}
let input = build_slice!(input, input_len as usize);

let output = InspectionBufferCheckAndExpand(buffer, input_len + 1);
if output.is_null() {
// allocation failure
return;
}
// get input after possible realloc
let input = InspectionBufferPtr(buffer);
if input.is_null() {
// allocation failure
return;
}
let input = build_slice!(input, input_len as usize);
let output = std::slice::from_raw_parts_mut(output, (input_len + 1) as usize);

dot_prefix_transform_do(input, output);
Expand Down Expand Up @@ -89,9 +97,15 @@ mod tests {
let mut out = vec![0; b"example.com".len() + 1];
dot_prefix_transform_do(buf, &mut out);
assert_eq!(out, b".example.com");
let buf = b"hello.example.com";
let mut buf = Vec::with_capacity(b"hello.example.com".len() + 1);
buf.extend_from_slice(b"hello.example.com");
let mut out = vec![0; b"hello.example.com".len() + 1];
dot_prefix_transform_do(buf, &mut out);
dot_prefix_transform_do(&buf, &mut out);
assert_eq!(out, b".hello.example.com");
// test in place
let still_buf = unsafe { std::slice::from_raw_parts(buf.as_ptr(), buf.len()) };
buf.push(b'.');
dot_prefix_transform_do(still_buf, &mut buf);
assert_eq!(&buf, b".hello.example.com");
}
}
10 changes: 8 additions & 2 deletions rust/src/detect/transforms/hash.rs
Original file line number Diff line number Diff line change
Expand Up @@ -231,9 +231,15 @@ mod tests {

#[test]
fn test_sha256_transform() {
let buf = b" A B C D ";
let mut buf = Vec::with_capacity(SC_SHA256_LEN);
buf.extend_from_slice(b" A B C D ");
let mut out = vec![0; SC_SHA256_LEN];
sha256_transform_do(buf, &mut out);
sha256_transform_do(&buf, &mut out);
assert_eq!(out, b"\xd6\xbf\x7d\x8d\x69\x53\x02\x4d\x0d\x84\x5c\x99\x9b\xae\x93\xcc\xac\x68\xea\xab\x9a\xc9\x77\xd0\xfd\x30\x6a\xf5\x9a\x3d\xe4\x3a");
// test in place
let still_buf = unsafe { std::slice::from_raw_parts(buf.as_ptr(), buf.len()) };
buf.resize(SC_SHA256_LEN, 0);
sha256_transform_do(still_buf, &mut buf);
assert_eq!(&buf, b"\xd6\xbf\x7d\x8d\x69\x53\x02\x4d\x0d\x84\x5c\x99\x9b\xae\x93\xcc\xac\x68\xea\xab\x9a\xc9\x77\xd0\xfd\x30\x6a\xf5\x9a\x3d\xe4\x3a");
}
}
20 changes: 19 additions & 1 deletion rust/src/detect/transforms/http_headers.rs
Original file line number Diff line number Diff line change
Expand Up @@ -103,11 +103,18 @@ unsafe extern "C" fn strip_pseudo_setup(

fn strip_pseudo_transform_do(input: &[u8], output: &mut [u8]) -> u32 {
let mut nb = 0;
let mut inb = 0;
let same = std::ptr::eq(output.as_ptr(), input.as_ptr());
for subslice in input.split_inclusive(|c| *c == b'\n') {
if !subslice.is_empty() && subslice[0] != b':' {
output[nb..nb + subslice.len()].copy_from_slice(subslice);
if same {
output.copy_within(inb..inb + subslice.len(), nb);
} else {
output[nb..nb + subslice.len()].copy_from_slice(subslice);
}
nb += subslice.len();
}
inb += subslice.len();
}
return nb as u32;
}
Expand Down Expand Up @@ -183,5 +190,16 @@ mod tests {
let mut out = vec![0; buf.len()];
let nb = strip_pseudo_transform_do(buf, &mut out);
assert_eq!(&out[..nb as usize], b"header2:Value2");
let mut buf = Vec::new();
buf.extend_from_slice(
b"Header1: Value1\n:method:get\nheader2:Value2\n:scheme:https\nheader3:Value3\n",
);
// test in place
let still_buf = unsafe { std::slice::from_raw_parts(buf.as_ptr(), buf.len()) };
let nb = strip_pseudo_transform_do(still_buf, &mut buf);
assert_eq!(
&still_buf[..nb as usize],
b"Header1: Value1\nheader2:Value2\nheader3:Value3\n"
);
}
}
24 changes: 17 additions & 7 deletions rust/src/detect/transforms/strip_whitespace.rs
Original file line number Diff line number Diff line change
Expand Up @@ -35,13 +35,15 @@ unsafe extern "C" fn strip_whitespace_setup(

fn strip_whitespace_transform_do(input: &[u8], output: &mut [u8]) -> u32 {
let mut nb = 0;
// seems faster than writing one byte at a time via
// for (i, o) in input.iter().filter(|c| !matches!(*c, b'\t' | b'\n' | b'\x0B' | b'\x0C' | b'\r' | b' ')).zip(output)
for subslice in input.split(|c| matches!(*c, b'\t' | b'\n' | b'\x0B' | b'\x0C' | b'\r' | b' '))
for (i, o) in input
.iter()
.filter(|c| !matches!(*c, b'\t' | b'\n' | b'\x0B' | b'\x0C' | b'\r' | b' '))
.zip(output)
{
output[nb..nb + subslice.len()].copy_from_slice(subslice);
nb += subslice.len();
*o = *i;
nb += 1;
}
// do not use faster copy_from_slice because input and output may overlap (point to the same data)
return nb as u32;
}

Expand Down Expand Up @@ -124,13 +126,21 @@ mod tests {
);
assert_eq!(&out[..exp.len()], exp);

let buf = b"I \t J";
let mut buf = Vec::new();
buf.extend_from_slice(b"I \t J");
let mut out = vec![0; buf.len()];
let exp = b"IJ";
assert_eq!(
strip_whitespace_transform_do(buf, &mut out),
strip_whitespace_transform_do(&buf, &mut out),
exp.len() as u32
);
assert_eq!(&out[..exp.len()], exp);
// test in place
let still_buf = unsafe { std::slice::from_raw_parts(buf.as_ptr(), buf.len()) };
assert_eq!(
strip_whitespace_transform_do(still_buf, &mut buf),
exp.len() as u32
);
assert_eq!(&still_buf[..exp.len()], b"IJ");
}
}
9 changes: 7 additions & 2 deletions rust/src/detect/transforms/urldecode.rs
Original file line number Diff line number Diff line change
Expand Up @@ -135,9 +135,14 @@ mod tests {

#[test]
fn test_url_decode_transform() {
let buf = b"Suricata%20is+%27%61wesome%21%27%25%30%30%ZZ%4";
let mut buf = Vec::new();
buf.extend_from_slice(b"Suricata%20is+%27%61wesome%21%27%25%30%30%ZZ%4");
let mut out = vec![0; buf.len()];
let nb = url_decode_transform_do(buf, &mut out);
let nb = url_decode_transform_do(&buf, &mut out);
assert_eq!(&out[..nb as usize], b"Suricata is 'awesome!'%00%ZZ%4");
// test in place
let still_buf = unsafe { std::slice::from_raw_parts(buf.as_ptr(), buf.len()) };
let nb = url_decode_transform_do(still_buf, &mut buf);
assert_eq!(&still_buf[..nb as usize], b"Suricata is 'awesome!'%00%ZZ%4");
}
}
9 changes: 7 additions & 2 deletions rust/src/detect/transforms/xor.rs
Original file line number Diff line number Diff line change
Expand Up @@ -143,10 +143,15 @@ mod tests {

#[test]
fn test_xor_transform() {
let buf = b"example.com";
let mut buf = Vec::new();
buf.extend_from_slice(b"example.com");
let mut out = vec![0; buf.len()];
let ctx = xor_parse_do("0a0DC8ff").unwrap();
xor_transform_do(buf, &mut out, &ctx);
xor_transform_do(&buf, &mut out, &ctx);
assert_eq!(out, b"ou\xa9\x92za\xad\xd1ib\xa5");
// test in place
let still_buf = unsafe { std::slice::from_raw_parts(buf.as_ptr(), buf.len()) };
xor_transform_do(still_buf, &mut buf, &ctx);
assert_eq!(&still_buf, b"ou\xa9\x92za\xad\xd1ib\xa5");
}
}

0 comments on commit b58b886

Please sign in to comment.