fix(s3): harden Snowball archive extraction (#6942)

* fix(s3): harden Snowball extract error boundaries

* fix(s3): close Snowball extract compatibility gaps

* fix(s3): verify Snowball request body completion

* test(s3): reject forged Snowball streaming signatures

* build(deps): pin Snowball archive parser limits

* fix(s3): preserve Snowball trailer and member errors

* docs(architecture): register Snowball tar fork cleanup

* refactor(s3): route Snowball errors through object boundary

* ci(deps): allow pinned tokio-tar source

* ci(e2e): refresh Snowball smoke selection
This commit is contained in:
cxymds
2026-08-31 19:18:09 +08:00
committed by GitHub
parent 35456bcede
commit ff28b79088
12 changed files with 1650 additions and 169 deletions
@@ -17,8 +17,9 @@ mod tests {
use crate::common::{RustFSTestEnvironment, init_logging};
use aws_sdk_s3::error::ProvideErrorMetadata;
use aws_sdk_s3::primitives::ByteStream;
use flate2::{Compression, write::GzEncoder};
use std::error::Error;
use std::io::Cursor;
use std::io::{Cursor, Write};
async fn build_test_archive() -> Result<Vec<u8>, Box<dyn Error + Send + Sync>> {
let mut builder = tokio_tar::Builder::new(Cursor::new(Vec::new()));
@@ -69,12 +70,50 @@ mod tests {
Ok(builder.into_inner().await?.into_inner())
}
fn build_archive_with_parent_dir_entry(victim_bucket: &str) -> Vec<u8> {
let path = format!("../{victim_bucket}/evil-injected.txt");
let data = b"injected-body";
async fn build_archive_with_invalid_checksum() -> Result<Vec<u8>, Box<dyn Error + Send + Sync>> {
let mut archive = build_test_archive().await?;
archive[0] ^= 1;
Ok(archive)
}
async fn build_archive_with_negative_gnu_mtime() -> Result<Vec<u8>, Box<dyn Error + Send + Sync>> {
let mut builder = tokio_tar::Builder::new(Cursor::new(Vec::new()));
let mut header = tokio_tar::Header::new_gnu();
header.set_size(b"negative-mtime-body".len() as u64);
header.set_mode(0o644);
header.as_old_mut().mtime.fill(0xff);
builder
.append_data(&mut header, "negative-mtime.txt", Cursor::new(b"negative-mtime-body".as_slice()))
.await?;
Ok(builder.into_inner().await?.into_inner())
}
fn gzip_member(payload: &[u8]) -> Result<Vec<u8>, Box<dyn Error + Send + Sync>> {
let mut encoder = GzEncoder::new(Vec::new(), Compression::default());
encoder.write_all(payload)?;
Ok(encoder.finish()?)
}
async fn build_concatenated_gzip_archive() -> Result<Vec<u8>, Box<dyn Error + Send + Sync>> {
let archive = build_test_archive().await?;
let split_at = archive.len() / 2;
let mut encoded = gzip_member(&archive[..split_at])?;
encoded.extend(gzip_member(&archive[split_at..])?);
Ok(encoded)
}
async fn build_gzip_archive_with_invalid_crc() -> Result<Vec<u8>, Box<dyn Error + Send + Sync>> {
let mut encoded = gzip_member(&build_test_archive().await?)?;
let crc_offset = encoded.len().checked_sub(8).expect("gzip fixture must contain a trailer");
encoded[crc_offset] ^= 1;
Ok(encoded)
}
fn append_raw_tar_entry_with_type(archive: &mut Vec<u8>, path: &[u8], data: &[u8], entry_type: u8) {
assert!(path.len() <= 100, "raw TAR fixture path must fit in the name field");
let mut header = [0u8; 512];
header[..path.len()].copy_from_slice(path.as_bytes());
header[..path.len()].copy_from_slice(path);
header[100..108].copy_from_slice(b"0000644\0");
header[108..116].copy_from_slice(b"0000000\0");
header[116..124].copy_from_slice(b"0000000\0");
@@ -82,7 +121,7 @@ mod tests {
header[124..136].copy_from_slice(size.as_bytes());
header[136..148].copy_from_slice(b"00000000000\0");
header[148..156].fill(b' ');
header[156] = b'0';
header[156] = entry_type;
header[257..263].copy_from_slice(b"ustar\0");
header[263..265].copy_from_slice(b"00");
@@ -90,11 +129,36 @@ mod tests {
let checksum = format!("{:06o}\0 ", checksum);
header[148..156].copy_from_slice(checksum.as_bytes());
let mut archive = Vec::new();
archive.extend_from_slice(&header);
archive.extend_from_slice(data);
let padding = (512 - (data.len() % 512)) % 512;
archive.extend(std::iter::repeat_n(0, padding));
}
fn append_raw_tar_entry(archive: &mut Vec<u8>, path: &[u8], data: &[u8]) {
append_raw_tar_entry_with_type(archive, path, data, b'0');
}
fn build_archive_with_parent_dir_entry(victim_bucket: &str) -> Vec<u8> {
let path = format!("../{victim_bucket}/evil-injected.txt");
let mut archive = Vec::new();
append_raw_tar_entry(&mut archive, path.as_bytes(), b"injected-body");
archive.extend_from_slice(&[0u8; 1024]);
archive
}
fn build_archive_with_invalid_utf8_entry() -> Vec<u8> {
let mut archive = Vec::new();
append_raw_tar_entry(&mut archive, b"invalid-\xff.txt", b"ignored-body");
append_raw_tar_entry(&mut archive, b"valid.txt", b"valid-body");
archive.extend_from_slice(&[0u8; 1024]);
archive
}
fn build_archive_with_invalid_utf8_symlink() -> Vec<u8> {
let mut archive = Vec::new();
append_raw_tar_entry_with_type(&mut archive, b"invalid-\xff-link", b"", b'2');
append_raw_tar_entry(&mut archive, b"valid.txt", b"valid-body");
archive.extend_from_slice(&[0u8; 1024]);
archive
}
@@ -263,6 +327,113 @@ mod tests {
Ok(())
}
#[tokio::test]
async fn snowball_auto_extract_accepts_negative_gnu_mtime() -> Result<(), Box<dyn Error + Send + Sync>> {
init_logging();
let mut env = RustFSTestEnvironment::new().await?;
env.start_rustfs_server(vec![]).await?;
let client = env.create_s3_client();
let bucket = "snowball-negative-mtime";
client.create_bucket().bucket(bucket).send().await?;
client
.put_object()
.bucket(bucket)
.key("fixture.tar")
.metadata("Snowball-Auto-Extract", "true")
.body(ByteStream::from(build_archive_with_negative_gnu_mtime().await?))
.send()
.await?;
let object = client.get_object().bucket(bucket).key("negative-mtime.txt").send().await?;
assert_eq!(object.body.collect().await?.into_bytes().as_ref(), b"negative-mtime-body");
env.stop_server();
Ok(())
}
#[tokio::test]
async fn snowball_auto_extract_consumes_concatenated_gzip_members() -> Result<(), Box<dyn Error + Send + Sync>> {
init_logging();
let mut env = RustFSTestEnvironment::new().await?;
env.start_rustfs_server(vec![]).await?;
let client = env.create_s3_client();
let bucket = "snowball-concatenated-gzip";
client.create_bucket().bucket(bucket).send().await?;
client
.put_object()
.bucket(bucket)
.key("fixture.tar.gz")
.metadata("Snowball-Auto-Extract", "true")
.body(ByteStream::from(build_concatenated_gzip_archive().await?))
.send()
.await?;
let object = client.get_object().bucket(bucket).key("root.txt").send().await?;
assert_eq!(object.body.collect().await?.into_bytes().as_ref(), b"root payload\n");
env.stop_server();
Ok(())
}
#[tokio::test]
async fn snowball_auto_extract_rejects_gzip_crc_error_when_ignore_errors_enabled() -> Result<(), Box<dyn Error + Send + Sync>>
{
init_logging();
let mut env = RustFSTestEnvironment::new().await?;
env.start_rustfs_server(vec![]).await?;
let client = env.create_s3_client();
let bucket = "snowball-gzip-crc-ignore-errors";
client.create_bucket().bucket(bucket).send().await?;
let err = client
.put_object()
.bucket(bucket)
.key("fixture.tar.gz")
.metadata("Snowball-Auto-Extract", "true")
.metadata("Minio-Snowball-Ignore-Errors", "true")
.body(ByteStream::from(build_gzip_archive_with_invalid_crc().await?))
.send()
.await
.expect_err("gzip integrity failures must remain fatal under ignore-errors");
assert_eq!(err.into_service_error().code(), Some("InvalidArgument"));
env.stop_server();
Ok(())
}
#[tokio::test]
async fn snowball_auto_extract_rejects_mismatched_content_md5() -> Result<(), Box<dyn Error + Send + Sync>> {
init_logging();
let mut env = RustFSTestEnvironment::new().await?;
env.start_rustfs_server(vec![]).await?;
let client = env.create_s3_client();
let bucket = "snowball-content-md5";
client.create_bucket().bucket(bucket).send().await?;
let err = client
.put_object()
.bucket(bucket)
.key("fixture.tar")
.metadata("Snowball-Auto-Extract", "true")
.content_md5("AAAAAAAAAAAAAAAAAAAAAA==")
.body(ByteStream::from(build_test_archive().await?))
.send()
.await
.expect_err("mismatched Content-MD5 must fail after the raw body reaches EOF");
assert_eq!(err.into_service_error().code(), Some("BadDigest"));
env.stop_server();
Ok(())
}
#[tokio::test]
async fn snowball_auto_extract_ignores_invalid_entries_when_requested() -> Result<(), Box<dyn Error + Send + Sync>> {
init_logging();
@@ -299,7 +470,100 @@ mod tests {
}
#[tokio::test]
async fn snowball_auto_extract_rejects_parent_dir_entry_without_cross_bucket_write()
async fn snowball_auto_extract_skips_non_utf8_symlink_without_ignore_errors() -> Result<(), Box<dyn Error + Send + Sync>> {
init_logging();
let mut env = RustFSTestEnvironment::new().await?;
env.start_rustfs_server(vec![]).await?;
let client = env.create_s3_client();
let bucket = "snowball-invalid-utf8-link";
client.create_bucket().bucket(bucket).send().await?;
client
.put_object()
.bucket(bucket)
.key("fixture.tar")
.metadata("Snowball-Auto-Extract", "true")
.body(ByteStream::from(build_archive_with_invalid_utf8_symlink()))
.send()
.await?;
let valid = client.get_object().bucket(bucket).key("valid.txt").send().await?;
assert_eq!(valid.body.collect().await?.into_bytes().as_ref(), b"valid-body");
let listed = client.list_objects_v2().bucket(bucket).send().await?;
let keys: Vec<_> = listed.contents().iter().filter_map(|entry| entry.key()).collect();
assert_eq!(keys, vec!["valid.txt"]);
env.stop_server();
Ok(())
}
#[tokio::test]
async fn snowball_auto_extract_skips_non_utf8_member_without_lossy_key_collision() -> Result<(), Box<dyn Error + Send + Sync>>
{
init_logging();
let mut env = RustFSTestEnvironment::new().await?;
env.start_rustfs_server(vec![]).await?;
let client = env.create_s3_client();
let bucket = "snowball-invalid-utf8";
client.create_bucket().bucket(bucket).send().await?;
client
.put_object()
.bucket(bucket)
.key("fixture.tar")
.metadata("Snowball-Auto-Extract", "true")
.metadata("Minio-Snowball-Ignore-Errors", "true")
.body(ByteStream::from(build_archive_with_invalid_utf8_entry()))
.send()
.await?;
let valid = client.get_object().bucket(bucket).key("valid.txt").send().await?;
assert_eq!(valid.body.collect().await?.into_bytes().as_ref(), b"valid-body");
let listed = client.list_objects_v2().bucket(bucket).send().await?;
let keys: Vec<_> = listed.contents().iter().filter_map(|entry| entry.key()).collect();
assert_eq!(keys, vec!["valid.txt"]);
env.stop_server();
Ok(())
}
#[tokio::test]
async fn snowball_auto_extract_rejects_corrupt_tar_when_ignore_errors_enabled() -> Result<(), Box<dyn Error + Send + Sync>> {
init_logging();
let mut env = RustFSTestEnvironment::new().await?;
env.start_rustfs_server(vec![]).await?;
let client = env.create_s3_client();
let bucket = "snowball-corrupt-ignore-errors";
let archive = build_archive_with_invalid_checksum().await?;
client.create_bucket().bucket(bucket).send().await?;
let err = client
.put_object()
.bucket(bucket)
.key("fixture.tar")
.metadata("Snowball-Auto-Extract", "true")
.metadata("Minio-Snowball-Ignore-Errors", "true")
.body(ByteStream::from(archive))
.send()
.await
.expect_err("corrupt TAR structure must remain fatal under ignore-errors");
assert_eq!(err.into_service_error().code(), Some("InvalidArgument"));
let listed = client.list_objects_v2().bucket(bucket).send().await?;
assert!(listed.contents().is_empty(), "corrupt archive must not produce objects");
env.stop_server();
Ok(())
}
#[tokio::test]
async fn snowball_auto_extract_rejects_parent_dir_entry_even_when_ignore_errors_enabled()
-> Result<(), Box<dyn Error + Send + Sync>> {
init_logging();
@@ -319,6 +583,7 @@ mod tests {
.bucket(attacker_bucket)
.key("fixture.tar")
.metadata("Snowball-Auto-Extract", "true")
.metadata("Minio-Snowball-Ignore-Errors", "true")
.body(ByteStream::from(archive))
.send()
.await