1 | //! Provides the [Engine] abstraction and out of the box implementations. |
2 | #[cfg (any(feature = "alloc" , test))] |
3 | use crate::chunked_encoder; |
4 | use crate::{ |
5 | encode::{encode_with_padding, EncodeSliceError}, |
6 | encoded_len, DecodeError, DecodeSliceError, |
7 | }; |
8 | #[cfg (any(feature = "alloc" , test))] |
9 | use alloc::vec::Vec; |
10 | |
11 | #[cfg (any(feature = "alloc" , test))] |
12 | use alloc::{string::String, vec}; |
13 | |
14 | pub mod general_purpose; |
15 | |
16 | #[cfg (test)] |
17 | mod naive; |
18 | |
19 | #[cfg (test)] |
20 | mod tests; |
21 | |
22 | pub use general_purpose::{GeneralPurpose, GeneralPurposeConfig}; |
23 | |
24 | /// An `Engine` provides low-level encoding and decoding operations that all other higher-level parts of the API use. Users of the library will generally not need to implement this. |
25 | /// |
26 | /// Different implementations offer different characteristics. The library currently ships with |
27 | /// [GeneralPurpose] that offers good speed and works on any CPU, with more choices |
28 | /// coming later, like a constant-time one when side channel resistance is called for, and vendor-specific vectorized ones for more speed. |
29 | /// |
30 | /// See [general_purpose::STANDARD_NO_PAD] if you just want standard base64. Otherwise, when possible, it's |
31 | /// recommended to store the engine in a `const` so that references to it won't pose any lifetime |
32 | /// issues, and to avoid repeating the cost of engine setup. |
33 | /// |
34 | /// Since almost nobody will need to implement `Engine`, docs for internal methods are hidden. |
35 | // When adding an implementation of Engine, include them in the engine test suite: |
36 | // - add an implementation of [engine::tests::EngineWrapper] |
37 | // - add the implementation to the `all_engines` macro |
38 | // All tests run on all engines listed in the macro. |
39 | pub trait Engine: Send + Sync { |
40 | /// The config type used by this engine |
41 | type Config: Config; |
42 | /// The decode estimate used by this engine |
43 | type DecodeEstimate: DecodeEstimate; |
44 | |
45 | /// This is not meant to be called directly; it is only for `Engine` implementors. |
46 | /// See the other `encode*` functions on this trait. |
47 | /// |
48 | /// Encode the `input` bytes into the `output` buffer based on the mapping in `encode_table`. |
49 | /// |
50 | /// `output` will be long enough to hold the encoded data. |
51 | /// |
52 | /// Returns the number of bytes written. |
53 | /// |
54 | /// No padding should be written; that is handled separately. |
55 | /// |
56 | /// Must not write any bytes into the output slice other than the encoded data. |
57 | #[doc (hidden)] |
58 | fn internal_encode(&self, input: &[u8], output: &mut [u8]) -> usize; |
59 | |
60 | /// This is not meant to be called directly; it is only for `Engine` implementors. |
61 | /// |
62 | /// As an optimization to prevent the decoded length from being calculated twice, it is |
63 | /// sometimes helpful to have a conservative estimate of the decoded size before doing the |
64 | /// decoding, so this calculation is done separately and passed to [Engine::decode()] as needed. |
65 | #[doc (hidden)] |
66 | fn internal_decoded_len_estimate(&self, input_len: usize) -> Self::DecodeEstimate; |
67 | |
68 | /// This is not meant to be called directly; it is only for `Engine` implementors. |
69 | /// See the other `decode*` functions on this trait. |
70 | /// |
71 | /// Decode `input` base64 bytes into the `output` buffer. |
72 | /// |
73 | /// `decode_estimate` is the result of [Engine::internal_decoded_len_estimate()], which is passed in to avoid |
74 | /// calculating it again (expensive on short inputs).` |
75 | /// |
76 | /// Each complete 4-byte chunk of encoded data decodes to 3 bytes of decoded data, but this |
77 | /// function must also handle the final possibly partial chunk. |
78 | /// If the input length is not a multiple of 4, or uses padding bytes to reach a multiple of 4, |
79 | /// the trailing 2 or 3 bytes must decode to 1 or 2 bytes, respectively, as per the |
80 | /// [RFC](https://tools.ietf.org/html/rfc4648#section-3.5). |
81 | /// |
82 | /// Decoding must not write any bytes into the output slice other than the decoded data. |
83 | /// |
84 | /// Non-canonical trailing bits in the final tokens or non-canonical padding must be reported as |
85 | /// errors unless the engine is configured otherwise. |
86 | /// |
87 | /// # Panics |
88 | /// |
89 | /// Panics if `output` is too small. |
90 | #[doc (hidden)] |
91 | fn internal_decode( |
92 | &self, |
93 | input: &[u8], |
94 | output: &mut [u8], |
95 | decode_estimate: Self::DecodeEstimate, |
96 | ) -> Result<DecodeMetadata, DecodeError>; |
97 | |
98 | /// Returns the config for this engine. |
99 | fn config(&self) -> &Self::Config; |
100 | |
101 | /// Encode arbitrary octets as base64 using the provided `Engine`. |
102 | /// Returns a `String`. |
103 | /// |
104 | /// # Example |
105 | /// |
106 | /// ```rust |
107 | /// use base64::{Engine as _, engine::{self, general_purpose}, alphabet}; |
108 | /// |
109 | /// let b64 = general_purpose::STANDARD.encode(b"hello world~" ); |
110 | /// println!("{}" , b64); |
111 | /// |
112 | /// const CUSTOM_ENGINE: engine::GeneralPurpose = |
113 | /// engine::GeneralPurpose::new(&alphabet::URL_SAFE, general_purpose::NO_PAD); |
114 | /// |
115 | /// let b64_url = CUSTOM_ENGINE.encode(b"hello internet~" ); |
116 | #[cfg (any(feature = "alloc" , test))] |
117 | #[inline ] |
118 | fn encode<T: AsRef<[u8]>>(&self, input: T) -> String { |
119 | fn inner<E>(engine: &E, input_bytes: &[u8]) -> String |
120 | where |
121 | E: Engine + ?Sized, |
122 | { |
123 | let encoded_size = encoded_len(input_bytes.len(), engine.config().encode_padding()) |
124 | .expect("integer overflow when calculating buffer size" ); |
125 | |
126 | let mut buf = vec![0; encoded_size]; |
127 | |
128 | encode_with_padding(input_bytes, &mut buf[..], engine, encoded_size); |
129 | |
130 | String::from_utf8(buf).expect("Invalid UTF8" ) |
131 | } |
132 | |
133 | inner(self, input.as_ref()) |
134 | } |
135 | |
136 | /// Encode arbitrary octets as base64 into a supplied `String`. |
137 | /// Writes into the supplied `String`, which may allocate if its internal buffer isn't big enough. |
138 | /// |
139 | /// # Example |
140 | /// |
141 | /// ```rust |
142 | /// use base64::{Engine as _, engine::{self, general_purpose}, alphabet}; |
143 | /// const CUSTOM_ENGINE: engine::GeneralPurpose = |
144 | /// engine::GeneralPurpose::new(&alphabet::URL_SAFE, general_purpose::NO_PAD); |
145 | /// |
146 | /// fn main() { |
147 | /// let mut buf = String::new(); |
148 | /// general_purpose::STANDARD.encode_string(b"hello world~" , &mut buf); |
149 | /// println!("{}" , buf); |
150 | /// |
151 | /// buf.clear(); |
152 | /// CUSTOM_ENGINE.encode_string(b"hello internet~" , &mut buf); |
153 | /// println!("{}" , buf); |
154 | /// } |
155 | /// ``` |
156 | #[cfg (any(feature = "alloc" , test))] |
157 | #[inline ] |
158 | fn encode_string<T: AsRef<[u8]>>(&self, input: T, output_buf: &mut String) { |
159 | fn inner<E>(engine: &E, input_bytes: &[u8], output_buf: &mut String) |
160 | where |
161 | E: Engine + ?Sized, |
162 | { |
163 | let mut sink = chunked_encoder::StringSink::new(output_buf); |
164 | |
165 | chunked_encoder::ChunkedEncoder::new(engine) |
166 | .encode(input_bytes, &mut sink) |
167 | .expect("Writing to a String shouldn't fail" ); |
168 | } |
169 | |
170 | inner(self, input.as_ref(), output_buf) |
171 | } |
172 | |
173 | /// Encode arbitrary octets as base64 into a supplied slice. |
174 | /// Writes into the supplied output buffer. |
175 | /// |
176 | /// This is useful if you wish to avoid allocation entirely (e.g. encoding into a stack-resident |
177 | /// or statically-allocated buffer). |
178 | /// |
179 | /// # Example |
180 | /// |
181 | #[cfg_attr (feature = "alloc" , doc = "```" )] |
182 | #[cfg_attr (not(feature = "alloc" ), doc = "```ignore" )] |
183 | /// use base64::{Engine as _, engine::general_purpose}; |
184 | /// let s = b"hello internet!" ; |
185 | /// let mut buf = Vec::new(); |
186 | /// // make sure we'll have a slice big enough for base64 + padding |
187 | /// buf.resize(s.len() * 4 / 3 + 4, 0); |
188 | /// |
189 | /// let bytes_written = general_purpose::STANDARD.encode_slice(s, &mut buf).unwrap(); |
190 | /// |
191 | /// // shorten our vec down to just what was written |
192 | /// buf.truncate(bytes_written); |
193 | /// |
194 | /// assert_eq!(s, general_purpose::STANDARD.decode(&buf).unwrap().as_slice()); |
195 | /// ``` |
196 | #[inline ] |
197 | fn encode_slice<T: AsRef<[u8]>>( |
198 | &self, |
199 | input: T, |
200 | output_buf: &mut [u8], |
201 | ) -> Result<usize, EncodeSliceError> { |
202 | fn inner<E>( |
203 | engine: &E, |
204 | input_bytes: &[u8], |
205 | output_buf: &mut [u8], |
206 | ) -> Result<usize, EncodeSliceError> |
207 | where |
208 | E: Engine + ?Sized, |
209 | { |
210 | let encoded_size = encoded_len(input_bytes.len(), engine.config().encode_padding()) |
211 | .expect("usize overflow when calculating buffer size" ); |
212 | |
213 | if output_buf.len() < encoded_size { |
214 | return Err(EncodeSliceError::OutputSliceTooSmall); |
215 | } |
216 | |
217 | let b64_output = &mut output_buf[0..encoded_size]; |
218 | |
219 | encode_with_padding(input_bytes, b64_output, engine, encoded_size); |
220 | |
221 | Ok(encoded_size) |
222 | } |
223 | |
224 | inner(self, input.as_ref(), output_buf) |
225 | } |
226 | |
227 | /// Decode the input into a new `Vec`. |
228 | /// |
229 | /// # Example |
230 | /// |
231 | /// ```rust |
232 | /// use base64::{Engine as _, alphabet, engine::{self, general_purpose}}; |
233 | /// |
234 | /// let bytes = general_purpose::STANDARD |
235 | /// .decode("aGVsbG8gd29ybGR+Cg==" ).unwrap(); |
236 | /// println!("{:?}" , bytes); |
237 | /// |
238 | /// // custom engine setup |
239 | /// let bytes_url = engine::GeneralPurpose::new( |
240 | /// &alphabet::URL_SAFE, |
241 | /// general_purpose::NO_PAD) |
242 | /// .decode("aGVsbG8gaW50ZXJuZXR-Cg" ).unwrap(); |
243 | /// println!("{:?}" , bytes_url); |
244 | /// ``` |
245 | #[cfg (any(feature = "alloc" , test))] |
246 | #[inline ] |
247 | fn decode<T: AsRef<[u8]>>(&self, input: T) -> Result<Vec<u8>, DecodeError> { |
248 | fn inner<E>(engine: &E, input_bytes: &[u8]) -> Result<Vec<u8>, DecodeError> |
249 | where |
250 | E: Engine + ?Sized, |
251 | { |
252 | let estimate = engine.internal_decoded_len_estimate(input_bytes.len()); |
253 | let mut buffer = vec![0; estimate.decoded_len_estimate()]; |
254 | |
255 | let bytes_written = engine |
256 | .internal_decode(input_bytes, &mut buffer, estimate)? |
257 | .decoded_len; |
258 | |
259 | buffer.truncate(bytes_written); |
260 | |
261 | Ok(buffer) |
262 | } |
263 | |
264 | inner(self, input.as_ref()) |
265 | } |
266 | |
267 | /// Decode the `input` into the supplied `buffer`. |
268 | /// |
269 | /// Writes into the supplied `Vec`, which may allocate if its internal buffer isn't big enough. |
270 | /// Returns a `Result` containing an empty tuple, aka `()`. |
271 | /// |
272 | /// # Example |
273 | /// |
274 | /// ```rust |
275 | /// use base64::{Engine as _, alphabet, engine::{self, general_purpose}}; |
276 | /// const CUSTOM_ENGINE: engine::GeneralPurpose = |
277 | /// engine::GeneralPurpose::new(&alphabet::URL_SAFE, general_purpose::PAD); |
278 | /// |
279 | /// fn main() { |
280 | /// use base64::Engine; |
281 | /// let mut buffer = Vec::<u8>::new(); |
282 | /// // with the default engine |
283 | /// general_purpose::STANDARD |
284 | /// .decode_vec("aGVsbG8gd29ybGR+Cg==" , &mut buffer,).unwrap(); |
285 | /// println!("{:?}" , buffer); |
286 | /// |
287 | /// buffer.clear(); |
288 | /// |
289 | /// // with a custom engine |
290 | /// CUSTOM_ENGINE.decode_vec( |
291 | /// "aGVsbG8gaW50ZXJuZXR-Cg==" , |
292 | /// &mut buffer, |
293 | /// ).unwrap(); |
294 | /// println!("{:?}" , buffer); |
295 | /// } |
296 | /// ``` |
297 | #[cfg (any(feature = "alloc" , test))] |
298 | #[inline ] |
299 | fn decode_vec<T: AsRef<[u8]>>( |
300 | &self, |
301 | input: T, |
302 | buffer: &mut Vec<u8>, |
303 | ) -> Result<(), DecodeError> { |
304 | fn inner<E>(engine: &E, input_bytes: &[u8], buffer: &mut Vec<u8>) -> Result<(), DecodeError> |
305 | where |
306 | E: Engine + ?Sized, |
307 | { |
308 | let starting_output_len = buffer.len(); |
309 | let estimate = engine.internal_decoded_len_estimate(input_bytes.len()); |
310 | |
311 | let total_len_estimate = estimate |
312 | .decoded_len_estimate() |
313 | .checked_add(starting_output_len) |
314 | .expect("Overflow when calculating output buffer length" ); |
315 | |
316 | buffer.resize(total_len_estimate, 0); |
317 | |
318 | let buffer_slice = &mut buffer.as_mut_slice()[starting_output_len..]; |
319 | |
320 | let bytes_written = engine |
321 | .internal_decode(input_bytes, buffer_slice, estimate)? |
322 | .decoded_len; |
323 | |
324 | buffer.truncate(starting_output_len + bytes_written); |
325 | |
326 | Ok(()) |
327 | } |
328 | |
329 | inner(self, input.as_ref(), buffer) |
330 | } |
331 | |
332 | /// Decode the input into the provided output slice. |
333 | /// |
334 | /// Returns the number of bytes written to the slice, or an error if `output` is smaller than |
335 | /// the estimated decoded length. |
336 | /// |
337 | /// This will not write any bytes past exactly what is decoded (no stray garbage bytes at the end). |
338 | /// |
339 | /// See [crate::decoded_len_estimate] for calculating buffer sizes. |
340 | /// |
341 | /// See [Engine::decode_slice_unchecked] for a version that panics instead of returning an error |
342 | /// if the output buffer is too small. |
343 | #[inline ] |
344 | fn decode_slice<T: AsRef<[u8]>>( |
345 | &self, |
346 | input: T, |
347 | output: &mut [u8], |
348 | ) -> Result<usize, DecodeSliceError> { |
349 | fn inner<E>( |
350 | engine: &E, |
351 | input_bytes: &[u8], |
352 | output: &mut [u8], |
353 | ) -> Result<usize, DecodeSliceError> |
354 | where |
355 | E: Engine + ?Sized, |
356 | { |
357 | let estimate = engine.internal_decoded_len_estimate(input_bytes.len()); |
358 | |
359 | if output.len() < estimate.decoded_len_estimate() { |
360 | return Err(DecodeSliceError::OutputSliceTooSmall); |
361 | } |
362 | |
363 | engine |
364 | .internal_decode(input_bytes, output, estimate) |
365 | .map_err(|e| e.into()) |
366 | .map(|dm| dm.decoded_len) |
367 | } |
368 | |
369 | inner(self, input.as_ref(), output) |
370 | } |
371 | |
372 | /// Decode the input into the provided output slice. |
373 | /// |
374 | /// Returns the number of bytes written to the slice. |
375 | /// |
376 | /// This will not write any bytes past exactly what is decoded (no stray garbage bytes at the end). |
377 | /// |
378 | /// See [crate::decoded_len_estimate] for calculating buffer sizes. |
379 | /// |
380 | /// See [Engine::decode_slice] for a version that returns an error instead of panicking if the output |
381 | /// buffer is too small. |
382 | /// |
383 | /// # Panics |
384 | /// |
385 | /// Panics if the provided output buffer is too small for the decoded data. |
386 | #[inline ] |
387 | fn decode_slice_unchecked<T: AsRef<[u8]>>( |
388 | &self, |
389 | input: T, |
390 | output: &mut [u8], |
391 | ) -> Result<usize, DecodeError> { |
392 | fn inner<E>(engine: &E, input_bytes: &[u8], output: &mut [u8]) -> Result<usize, DecodeError> |
393 | where |
394 | E: Engine + ?Sized, |
395 | { |
396 | engine |
397 | .internal_decode( |
398 | input_bytes, |
399 | output, |
400 | engine.internal_decoded_len_estimate(input_bytes.len()), |
401 | ) |
402 | .map(|dm| dm.decoded_len) |
403 | } |
404 | |
405 | inner(self, input.as_ref(), output) |
406 | } |
407 | } |
408 | |
409 | /// The minimal level of configuration that engines must support. |
410 | pub trait Config { |
411 | /// Returns `true` if padding should be added after the encoded output. |
412 | /// |
413 | /// Padding is added outside the engine's encode() since the engine may be used |
414 | /// to encode only a chunk of the overall output, so it can't always know when |
415 | /// the output is "done" and would therefore need padding (if configured). |
416 | // It could be provided as a separate parameter when encoding, but that feels like |
417 | // leaking an implementation detail to the user, and it's hopefully more convenient |
418 | // to have to only pass one thing (the engine) to any part of the API. |
419 | fn encode_padding(&self) -> bool; |
420 | } |
421 | |
422 | /// The decode estimate used by an engine implementation. Users do not need to interact with this; |
423 | /// it is only for engine implementors. |
424 | /// |
425 | /// Implementors may store relevant data here when constructing this to avoid having to calculate |
426 | /// them again during actual decoding. |
427 | pub trait DecodeEstimate { |
428 | /// Returns a conservative (err on the side of too big) estimate of the decoded length to use |
429 | /// for pre-allocating buffers, etc. |
430 | /// |
431 | /// The estimate must be no larger than the next largest complete triple of decoded bytes. |
432 | /// That is, the final quad of tokens to decode may be assumed to be complete with no padding. |
433 | fn decoded_len_estimate(&self) -> usize; |
434 | } |
435 | |
436 | /// Controls how pad bytes are handled when decoding. |
437 | /// |
438 | /// Each [Engine] must support at least the behavior indicated by |
439 | /// [DecodePaddingMode::RequireCanonical], and may support other modes. |
440 | #[derive (Clone, Copy, Debug, PartialEq, Eq)] |
441 | pub enum DecodePaddingMode { |
442 | /// Canonical padding is allowed, but any fewer padding bytes than that is also allowed. |
443 | Indifferent, |
444 | /// Padding must be canonical (0, 1, or 2 `=` as needed to produce a 4 byte suffix). |
445 | RequireCanonical, |
446 | /// Padding must be absent -- for when you want predictable padding, without any wasted bytes. |
447 | RequireNone, |
448 | } |
449 | |
450 | /// Metadata about the result of a decode operation |
451 | #[derive (PartialEq, Eq, Debug)] |
452 | pub struct DecodeMetadata { |
453 | /// Number of decoded bytes output |
454 | pub(crate) decoded_len: usize, |
455 | /// Offset of the first padding byte in the input, if any |
456 | pub(crate) padding_offset: Option<usize>, |
457 | } |
458 | |
459 | impl DecodeMetadata { |
460 | pub(crate) fn new(decoded_bytes: usize, padding_index: Option<usize>) -> Self { |
461 | Self { |
462 | decoded_len: decoded_bytes, |
463 | padding_offset: padding_index, |
464 | } |
465 | } |
466 | } |
467 | |