| 1 |
use crate::{engine::Engine, DecodeError, PAD_BYTE}; |
| 2 |
use std::{cmp, fmt, io}; |
| 3 |
|
| 4 |
// This should be large, but it has to fit on the stack. |
| 5 |
pub(crate) const BUF_SIZE: usize = 1024; |
| 6 |
|
| 7 |
// 4 bytes of base64 data encode 3 bytes of raw data (modulo padding). |
| 8 |
const BASE64_CHUNK_SIZE: usize = 4; |
| 9 |
const DECODED_CHUNK_SIZE: usize = 3; |
| 10 |
|
| 11 |
/// A `Read` implementation that decodes base64 data read from an underlying reader. |
| 12 |
/// |
| 13 |
/// # Examples |
| 14 |
/// |
| 15 |
/// ``` |
| 16 |
/// use std::io::Read; |
| 17 |
/// use std::io::Cursor; |
| 18 |
/// use base64::engine::general_purpose; |
| 19 |
/// |
| 20 |
/// // use a cursor as the simplest possible `Read` -- in real code this is probably a file, etc. |
| 21 |
/// let mut wrapped_reader = Cursor::new(b"YXNkZg=="); |
| 22 |
/// let mut decoder = base64::read::DecoderReader::new( |
| 23 |
/// &mut wrapped_reader, |
| 24 |
/// &general_purpose::STANDARD); |
| 25 |
/// |
| 26 |
/// // handle errors as you normally would |
| 27 |
/// let mut result = Vec::new(); |
| 28 |
/// decoder.read_to_end(&mut result).unwrap(); |
| 29 |
/// |
| 30 |
/// assert_eq!(b"asdf", &result[..]); |
| 31 |
/// |
| 32 |
/// ``` |
| 33 |
pub struct DecoderReader<'e, E: Engine, R: io::Read> { |
| 34 |
engine: &'e E, |
| 35 |
/// Where b64 data is read from |
| 36 |
inner: R, |
| 37 |
|
| 38 |
// Holds b64 data read from the delegate reader. |
| 39 |
b64_buffer: [u8; BUF_SIZE], |
| 40 |
// The start of the pending buffered data in b64_buffer. |
| 41 |
b64_offset: usize, |
| 42 |
// The amount of buffered b64 data. |
| 43 |
b64_len: usize, |
| 44 |
// Since the caller may provide us with a buffer of size 1 or 2 that's too small to copy a |
| 45 |
// decoded chunk in to, we have to be able to hang on to a few decoded bytes. |
| 46 |
// Technically we only need to hold 2 bytes but then we'd need a separate temporary buffer to |
| 47 |
// decode 3 bytes into and then juggle copying one byte into the provided read buf and the rest |
| 48 |
// into here, which seems like a lot of complexity for 1 extra byte of storage. |
| 49 |
decoded_buffer: [u8; DECODED_CHUNK_SIZE], |
| 50 |
// index of start of decoded data |
| 51 |
decoded_offset: usize, |
| 52 |
// length of decoded data |
| 53 |
decoded_len: usize, |
| 54 |
// used to provide accurate offsets in errors |
| 55 |
total_b64_decoded: usize, |
| 56 |
// offset of previously seen padding, if any |
| 57 |
padding_offset: Option<usize>, |
| 58 |
} |
| 59 |
|
| 60 |
impl<'e, E: Engine, R: io::Read> fmt::Debug for DecoderReader<'e, E, R> { |
| 61 |
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result { |
| 62 |
f.debug_struct("DecoderReader") |
| 63 |
.field("b64_offset", &self.b64_offset) |
| 64 |
.field("b64_len", &self.b64_len) |
| 65 |
.field("decoded_buffer", &self.decoded_buffer) |
| 66 |
.field("decoded_offset", &self.decoded_offset) |
| 67 |
.field("decoded_len", &self.decoded_len) |
| 68 |
.field("total_b64_decoded", &self.total_b64_decoded) |
| 69 |
.field("padding_offset", &self.padding_offset) |
| 70 |
.finish() |
| 71 |
} |
| 72 |
} |
| 73 |
|
| 74 |
impl<'e, E: Engine, R: io::Read> DecoderReader<'e, E, R> { |
| 75 |
/// Create a new decoder that will read from the provided reader `r`. |
| 76 |
pub fn new(reader: R, engine: &'e E) -> Self { |
| 77 |
DecoderReader { |
| 78 |
engine, |
| 79 |
inner: reader, |
| 80 |
b64_buffer: [0; BUF_SIZE], |
| 81 |
b64_offset: 0, |
| 82 |
b64_len: 0, |
| 83 |
decoded_buffer: [0; DECODED_CHUNK_SIZE], |
| 84 |
decoded_offset: 0, |
| 85 |
decoded_len: 0, |
| 86 |
total_b64_decoded: 0, |
| 87 |
padding_offset: None, |
| 88 |
} |
| 89 |
} |
| 90 |
|
| 91 |
/// Write as much as possible of the decoded buffer into the target buffer. |
| 92 |
/// Must only be called when there is something to write and space to write into. |
| 93 |
/// Returns a Result with the number of (decoded) bytes copied. |
| 94 |
fn flush_decoded_buf(&mut self, buf: &mut [u8]) -> io::Result<usize> { |
| 95 |
debug_assert!(self.decoded_len > 0); |
| 96 |
debug_assert!(!buf.is_empty()); |
| 97 |
|
| 98 |
let copy_len = cmp::min(self.decoded_len, buf.len()); |
| 99 |
debug_assert!(copy_len > 0); |
| 100 |
debug_assert!(copy_len <= self.decoded_len); |
| 101 |
|
| 102 |
buf[..copy_len].copy_from_slice( |
| 103 |
&self.decoded_buffer[self.decoded_offset..self.decoded_offset + copy_len], |
| 104 |
); |
| 105 |
|
| 106 |
self.decoded_offset += copy_len; |
| 107 |
self.decoded_len -= copy_len; |
| 108 |
|
| 109 |
debug_assert!(self.decoded_len < DECODED_CHUNK_SIZE); |
| 110 |
|
| 111 |
Ok(copy_len) |
| 112 |
} |
| 113 |
|
| 114 |
/// Read into the remaining space in the buffer after the current contents. |
| 115 |
/// Must only be called when there is space to read into in the buffer. |
| 116 |
/// Returns the number of bytes read. |
| 117 |
fn read_from_delegate(&mut self) -> io::Result<usize> { |
| 118 |
debug_assert!(self.b64_offset + self.b64_len < BUF_SIZE); |
| 119 |
|
| 120 |
let read = self |
| 121 |
.inner |
| 122 |
.read(&mut self.b64_buffer[self.b64_offset + self.b64_len..])?; |
| 123 |
self.b64_len += read; |
| 124 |
|
| 125 |
debug_assert!(self.b64_offset + self.b64_len <= BUF_SIZE); |
| 126 |
|
| 127 |
Ok(read) |
| 128 |
} |
| 129 |
|
| 130 |
/// Decode the requested number of bytes from the b64 buffer into the provided buffer. It's the |
| 131 |
/// caller's responsibility to choose the number of b64 bytes to decode correctly. |
| 132 |
/// |
| 133 |
/// Returns a Result with the number of decoded bytes written to `buf`. |
| 134 |
fn decode_to_buf(&mut self, b64_len_to_decode: usize, buf: &mut [u8]) -> io::Result<usize> { |
| 135 |
debug_assert!(self.b64_len >= b64_len_to_decode); |
| 136 |
debug_assert!(self.b64_offset + self.b64_len <= BUF_SIZE); |
| 137 |
debug_assert!(!buf.is_empty()); |
| 138 |
|
| 139 |
let b64_to_decode = &self.b64_buffer[self.b64_offset..self.b64_offset + b64_len_to_decode]; |
| 140 |
let decode_metadata = self |
| 141 |
.engine |
| 142 |
.internal_decode( |
| 143 |
b64_to_decode, |
| 144 |
buf, |
| 145 |
self.engine.internal_decoded_len_estimate(b64_len_to_decode), |
| 146 |
) |
| 147 |
.map_err(|e| match e { |
| 148 |
DecodeError::InvalidByte(offset, byte) => { |
| 149 |
// This can be incorrect, but not in a way that probably matters to anyone: |
| 150 |
// if there was padding handled in a previous decode, and we are now getting |
| 151 |
// InvalidByte due to more padding, we should arguably report InvalidByte with |
| 152 |
// PAD_BYTE at the original padding position (`self.padding_offset`), but we |
| 153 |
// don't have a good way to tie those two cases together, so instead we |
| 154 |
// just report the invalid byte as if the previous padding, and its possibly |
| 155 |
// related downgrade to a now invalid byte, didn't happen. |
| 156 |
DecodeError::InvalidByte(self.total_b64_decoded + offset, byte) |
| 157 |
} |
| 158 |
DecodeError::InvalidLength => DecodeError::InvalidLength, |
| 159 |
DecodeError::InvalidLastSymbol(offset, byte) => { |
| 160 |
DecodeError::InvalidLastSymbol(self.total_b64_decoded + offset, byte) |
| 161 |
} |
| 162 |
DecodeError::InvalidPadding => DecodeError::InvalidPadding, |
| 163 |
}) |
| 164 |
.map_err(|e| io::Error::new(io::ErrorKind::InvalidData, e))?; |
| 165 |
|
| 166 |
if let Some(offset) = self.padding_offset { |
| 167 |
// we've already seen padding |
| 168 |
if decode_metadata.decoded_len > 0 { |
| 169 |
// we read more after already finding padding; report error at first padding byte |
| 170 |
return Err(io::Error::new( |
| 171 |
io::ErrorKind::InvalidData, |
| 172 |
DecodeError::InvalidByte(offset, PAD_BYTE), |
| 173 |
)); |
| 174 |
} |
| 175 |
} |
| 176 |
|
| 177 |
self.padding_offset = self.padding_offset.or(decode_metadata |
| 178 |
.padding_offset |
| 179 |
.map(|offset| self.total_b64_decoded + offset)); |
| 180 |
self.total_b64_decoded += b64_len_to_decode; |
| 181 |
self.b64_offset += b64_len_to_decode; |
| 182 |
self.b64_len -= b64_len_to_decode; |
| 183 |
|
| 184 |
debug_assert!(self.b64_offset + self.b64_len <= BUF_SIZE); |
| 185 |
|
| 186 |
Ok(decode_metadata.decoded_len) |
| 187 |
} |
| 188 |
|
| 189 |
/// Unwraps this `DecoderReader`, returning the base reader which it reads base64 encoded |
| 190 |
/// input from. |
| 191 |
/// |
| 192 |
/// Because `DecoderReader` performs internal buffering, the state of the inner reader is |
| 193 |
/// unspecified. This function is mainly provided because the inner reader type may provide |
| 194 |
/// additional functionality beyond the `Read` implementation which may still be useful. |
| 195 |
pub fn into_inner(self) -> R { |
| 196 |
self.inner |
| 197 |
} |
| 198 |
} |
| 199 |
|
| 200 |
impl<'e, E: Engine, R: io::Read> io::Read for DecoderReader<'e, E, R> { |
| 201 |
/// Decode input from the wrapped reader. |
| 202 |
/// |
| 203 |
/// Under non-error circumstances, this returns `Ok` with the value being the number of bytes |
| 204 |
/// written in `buf`. |
| 205 |
/// |
| 206 |
/// Where possible, this function buffers base64 to minimize the number of read() calls to the |
| 207 |
/// delegate reader. |
| 208 |
/// |
| 209 |
/// # Errors |
| 210 |
/// |
| 211 |
/// Any errors emitted by the delegate reader are returned. Decoding errors due to invalid |
| 212 |
/// base64 are also possible, and will have `io::ErrorKind::InvalidData`. |
| 213 |
fn read(&mut self, buf: &mut [u8]) -> io::Result<usize> { |
| 214 |
if buf.is_empty() { |
| 215 |
return Ok(0); |
| 216 |
} |
| 217 |
|
| 218 |
// offset == BUF_SIZE when we copied it all last time |
| 219 |
debug_assert!(self.b64_offset <= BUF_SIZE); |
| 220 |
debug_assert!(self.b64_offset + self.b64_len <= BUF_SIZE); |
| 221 |
debug_assert!(if self.b64_offset == BUF_SIZE { |
| 222 |
self.b64_len == 0 |
| 223 |
} else { |
| 224 |
self.b64_len <= BUF_SIZE |
| 225 |
}); |
| 226 |
|
| 227 |
debug_assert!(if self.decoded_len == 0 { |
| 228 |
// can be = when we were able to copy the complete chunk |
| 229 |
self.decoded_offset <= DECODED_CHUNK_SIZE |
| 230 |
} else { |
| 231 |
self.decoded_offset < DECODED_CHUNK_SIZE |
| 232 |
}); |
| 233 |
|
| 234 |
// We shouldn't ever decode into decoded_buffer when we can't immediately write at least one |
| 235 |
// byte into the provided buf, so the effective length should only be 3 momentarily between |
| 236 |
// when we decode and when we copy into the target buffer. |
| 237 |
debug_assert!(self.decoded_len < DECODED_CHUNK_SIZE); |
| 238 |
debug_assert!(self.decoded_len + self.decoded_offset <= DECODED_CHUNK_SIZE); |
| 239 |
|
| 240 |
if self.decoded_len > 0 { |
| 241 |
// we have a few leftover decoded bytes; flush that rather than pull in more b64 |
| 242 |
self.flush_decoded_buf(buf) |
| 243 |
} else { |
| 244 |
let mut at_eof = false; |
| 245 |
while self.b64_len < BASE64_CHUNK_SIZE { |
| 246 |
// Copy any bytes we have to the start of the buffer. |
| 247 |
self.b64_buffer |
| 248 |
.copy_within(self.b64_offset..self.b64_offset + self.b64_len, 0); |
| 249 |
self.b64_offset = 0; |
| 250 |
|
| 251 |
// then fill in more data |
| 252 |
let read = self.read_from_delegate()?; |
| 253 |
if read == 0 { |
| 254 |
// we never read into an empty buf, so 0 => we've hit EOF |
| 255 |
at_eof = true; |
| 256 |
break; |
| 257 |
} |
| 258 |
} |
| 259 |
|
| 260 |
if self.b64_len == 0 { |
| 261 |
debug_assert!(at_eof); |
| 262 |
// we must be at EOF, and we have no data left to decode |
| 263 |
return Ok(0); |
| 264 |
}; |
| 265 |
|
| 266 |
debug_assert!(if at_eof { |
| 267 |
// if we are at eof, we may not have a complete chunk |
| 268 |
self.b64_len > 0 |
| 269 |
} else { |
| 270 |
// otherwise, we must have at least one chunk |
| 271 |
self.b64_len >= BASE64_CHUNK_SIZE |
| 272 |
}); |
| 273 |
|
| 274 |
debug_assert_eq!(0, self.decoded_len); |
| 275 |
|
| 276 |
if buf.len() < DECODED_CHUNK_SIZE { |
| 277 |
// caller requested an annoyingly short read |
| 278 |
// have to write to a tmp buf first to avoid double mutable borrow |
| 279 |
let mut decoded_chunk = [0_u8; DECODED_CHUNK_SIZE]; |
| 280 |
// if we are at eof, could have less than BASE64_CHUNK_SIZE, in which case we have |
| 281 |
// to assume that these last few tokens are, in fact, valid (i.e. must be 2-4 b64 |
| 282 |
// tokens, not 1, since 1 token can't decode to 1 byte). |
| 283 |
let to_decode = cmp::min(self.b64_len, BASE64_CHUNK_SIZE); |
| 284 |
|
| 285 |
let decoded = self.decode_to_buf(to_decode, &mut decoded_chunk[..])?; |
| 286 |
self.decoded_buffer[..decoded].copy_from_slice(&decoded_chunk[..decoded]); |
| 287 |
|
| 288 |
self.decoded_offset = 0; |
| 289 |
self.decoded_len = decoded; |
| 290 |
|
| 291 |
// can be less than 3 on last block due to padding |
| 292 |
debug_assert!(decoded <= 3); |
| 293 |
|
| 294 |
self.flush_decoded_buf(buf) |
| 295 |
} else { |
| 296 |
let b64_bytes_that_can_decode_into_buf = (buf.len() / DECODED_CHUNK_SIZE) |
| 297 |
.checked_mul(BASE64_CHUNK_SIZE) |
| 298 |
.expect("too many chunks"); |
| 299 |
debug_assert!(b64_bytes_that_can_decode_into_buf >= BASE64_CHUNK_SIZE); |
| 300 |
|
| 301 |
let b64_bytes_available_to_decode = if at_eof { |
| 302 |
self.b64_len |
| 303 |
} else { |
| 304 |
// only use complete chunks |
| 305 |
self.b64_len - self.b64_len % 4 |
| 306 |
}; |
| 307 |
|
| 308 |
let actual_decode_len = cmp::min( |
| 309 |
b64_bytes_that_can_decode_into_buf, |
| 310 |
b64_bytes_available_to_decode, |
| 311 |
); |
| 312 |
self.decode_to_buf(actual_decode_len, buf) |
| 313 |
} |
| 314 |
} |
| 315 |
} |
| 316 |
} |