| 1 |
#![allow(unused_imports)] |
| 2 |
|
| 3 |
use alloc::vec::Vec; |
| 4 |
use alloc::{format, vec}; |
| 5 |
|
| 6 |
use crate::bitstream::BitStreamReader; |
| 7 |
use crate::constants::{ |
| 8 |
DEFLATE_BLOCKTYPE_DYNAMIC_HUFFMAN, DEFLATE_BLOCKTYPE_RESERVED, DEFLATE_BLOCKTYPE_STATIC, |
| 9 |
DEFLATE_BLOCKTYPE_UNCOMPRESSED, DEFLATE_MAX_CODEWORD_LENGTH, |
| 10 |
DEFLATE_MAX_LITLEN_CODEWORD_LENGTH, DEFLATE_MAX_NUM_SYMS, DEFLATE_MAX_OFFSET_CODEWORD_LENGTH, |
| 11 |
DEFLATE_MAX_PRE_CODEWORD_LEN, DEFLATE_NUM_LITLEN_SYMS, DEFLATE_NUM_OFFSET_SYMS, |
| 12 |
DEFLATE_NUM_PRECODE_SYMS, DEFLATE_PRECODE_LENS_PERMUTATION, DELFATE_MAX_LENS_OVERRUN, |
| 13 |
FASTCOPY_BYTES, FASTLOOP_MAX_BYTES_WRITTEN, HUFFDEC_END_OF_BLOCK, HUFFDEC_EXCEPTIONAL, |
| 14 |
HUFFDEC_LITERAL, HUFFDEC_SUITABLE_POINTER, LITLEN_DECODE_BITS, LITLEN_DECODE_RESULTS, |
| 15 |
LITLEN_ENOUGH, LITLEN_TABLE_BITS, OFFSET_DECODE_RESULTS, OFFSET_ENOUGH, OFFSET_TABLEBITS, |
| 16 |
PRECODE_DECODE_RESULTS, PRECODE_ENOUGH, PRECODE_TABLE_BITS |
| 17 |
}; |
| 18 |
use crate::errors::{DecodeErrorStatus, InflateDecodeErrors}; |
| 19 |
#[cfg(feature = "gzip")] |
| 20 |
use crate::gzip_constants::{ |
| 21 |
GZIP_CM_DEFLATE, GZIP_FCOMMENT, GZIP_FEXTRA, GZIP_FHCRC, GZIP_FNAME, GZIP_FOOTER_SIZE, |
| 22 |
GZIP_FRESERVED, GZIP_ID1, GZIP_ID2 |
| 23 |
}; |
| 24 |
use crate::utils::{copy_rep_matches, fixed_copy_within, make_decode_table_entry}; |
| 25 |
|
| 26 |
struct DeflateHeaderTables |
| 27 |
{ |
| 28 |
litlen_decode_table: [u32; LITLEN_ENOUGH], |
| 29 |
offset_decode_table: [u32; OFFSET_ENOUGH] |
| 30 |
} |
| 31 |
|
| 32 |
impl Default for DeflateHeaderTables |
| 33 |
{ |
| 34 |
fn default() -> Self |
| 35 |
{ |
| 36 |
DeflateHeaderTables { |
| 37 |
litlen_decode_table: [0; LITLEN_ENOUGH], |
| 38 |
offset_decode_table: [0; OFFSET_ENOUGH] |
| 39 |
} |
| 40 |
} |
| 41 |
} |
| 42 |
|
| 43 |
/// Options that can influence decompression |
| 44 |
/// in Deflate/Zlib/Gzip |
| 45 |
/// |
| 46 |
/// To use them, pass a customized options to |
| 47 |
/// the deflate decoder. |
| 48 |
#[derive(Copy, Clone)] |
| 49 |
pub struct DeflateOptions |
| 50 |
{ |
| 51 |
limit: usize, |
| 52 |
confirm_checksum: bool, |
| 53 |
size_hint: usize |
| 54 |
} |
| 55 |
|
| 56 |
impl Default for DeflateOptions |
| 57 |
{ |
| 58 |
fn default() -> Self |
| 59 |
{ |
| 60 |
DeflateOptions { |
| 61 |
limit: 1 << 30, |
| 62 |
confirm_checksum: true, |
| 63 |
size_hint: 37000 |
| 64 |
} |
| 65 |
} |
| 66 |
} |
| 67 |
|
| 68 |
impl DeflateOptions |
| 69 |
{ |
| 70 |
/// Get deflate/zlib limit option |
| 71 |
/// |
| 72 |
/// The decoder won't extend the inbuilt limit and will |
| 73 |
/// return an error if the limit is exceeded |
| 74 |
/// |
| 75 |
/// # Returns |
| 76 |
/// The currently set limit of the instance |
| 77 |
/// # Note |
| 78 |
/// This is provided as a best effort, correctly quiting |
| 79 |
/// is detrimental to speed and hence this should not be relied too much. |
| 80 |
pub const fn get_limit(&self) -> usize |
| 81 |
{ |
| 82 |
self.limit |
| 83 |
} |
| 84 |
/// Set a limit to the internal vector |
| 85 |
/// used to store decoded zlib/deflate output. |
| 86 |
/// |
| 87 |
/// # Arguments |
| 88 |
/// limit: The new decompressor limit |
| 89 |
/// # Returns |
| 90 |
/// A modified version of DeflateDecoder |
| 91 |
/// |
| 92 |
/// # Note |
| 93 |
/// This is provided as a best effort, correctly quiting |
| 94 |
/// is detrimental to speed and hence this should not be relied too much |
| 95 |
#[must_use] |
| 96 |
pub fn set_limit(mut self, limit: usize) -> Self |
| 97 |
{ |
| 98 |
self.limit = limit; |
| 99 |
self |
| 100 |
} |
| 101 |
|
| 102 |
/// Get whether the decoder will confirm a checksum |
| 103 |
/// after decoding |
| 104 |
pub const fn get_confirm_checksum(&self) -> bool |
| 105 |
{ |
| 106 |
self.confirm_checksum |
| 107 |
} |
| 108 |
/// Set whether the decoder should confirm a checksum |
| 109 |
/// after decoding |
| 110 |
/// |
| 111 |
/// Note, you should definitely confirm your checksum, use |
| 112 |
/// this with caution, otherwise data returned may be corrupt |
| 113 |
/// |
| 114 |
/// # Arguments |
| 115 |
/// - yes: When true, the decoder will confirm checksum |
| 116 |
/// when false, the decoder will skip checksum verification |
| 117 |
/// # Notes |
| 118 |
/// This does not have an influence for deflate decoding as |
| 119 |
/// it does not have a checksum |
| 120 |
pub fn set_confirm_checksum(mut self, yes: bool) -> Self |
| 121 |
{ |
| 122 |
self.confirm_checksum = yes; |
| 123 |
self |
| 124 |
} |
| 125 |
|
| 126 |
/// Get the default set size hint for the decompressor |
| 127 |
/// |
| 128 |
/// The decompressor initializes the internal storage for decompressed bytes |
| 129 |
/// with this size and will reallocate the vec if the decompressed size becomes bigger |
| 130 |
/// than this, but when the user currently knows how big the output will be, can be used |
| 131 |
/// to prevent unnecessary re-allocations |
| 132 |
pub const fn get_size_hint(&self) -> usize |
| 133 |
{ |
| 134 |
self.size_hint |
| 135 |
} |
| 136 |
/// Set the size hint for the decompressor |
| 137 |
/// |
| 138 |
/// This can be used to prevent multiple re-allocations |
| 139 |
#[must_use] |
| 140 |
pub const fn set_size_hint(mut self, hint: usize) -> Self |
| 141 |
{ |
| 142 |
self.size_hint = hint; |
| 143 |
self |
| 144 |
} |
| 145 |
} |
| 146 |
|
| 147 |
/// A deflate decoder instance. |
| 148 |
/// |
| 149 |
/// The decoder manages output buffer as opposed to requiring the caller to provide a pre-allocated buffer |
| 150 |
/// it tracks number of bytes written and on successfully reaching the |
| 151 |
/// end of the block, will return a vector with exactly |
| 152 |
/// the number of decompressed bytes. |
| 153 |
/// |
| 154 |
/// This means that it may use up huge amounts of memory if not checked, but |
| 155 |
/// there are [options] that can prevent that |
| 156 |
/// |
| 157 |
/// [options]: DeflateOptions |
| 158 |
pub struct DeflateDecoder<'a> |
| 159 |
{ |
| 160 |
data: &'a [u8], |
| 161 |
position: usize, |
| 162 |
stream: BitStreamReader<'a>, |
| 163 |
is_last_block: bool, |
| 164 |
static_codes_loaded: bool, |
| 165 |
deflate_header_tables: DeflateHeaderTables, |
| 166 |
options: DeflateOptions |
| 167 |
} |
| 168 |
|
| 169 |
impl<'a> DeflateDecoder<'a> |
| 170 |
{ |
| 171 |
/// Create a new decompressor that will read compressed |
| 172 |
/// data from `data` and return a new vector containing new data |
| 173 |
/// |
| 174 |
/// # Arguments |
| 175 |
/// - `data`: The compressed data. Data can be of any type |
| 176 |
/// gzip,zlib or raw deflate. |
| 177 |
/// |
| 178 |
/// # Returns |
| 179 |
/// A decoder instance which will pull compressed data from `data` to inflate the output output |
| 180 |
/// |
| 181 |
/// # Note |
| 182 |
/// |
| 183 |
/// The default output size limit is **1 GiB.** |
| 184 |
/// this is to protect the end user against ddos attacks as deflate does not specify it's |
| 185 |
/// output size upfront |
| 186 |
/// |
| 187 |
/// The checksum will be verified depending on the called function. |
| 188 |
/// this only works for zlib and gzip since deflate does not have a checksum |
| 189 |
/// |
| 190 |
/// These defaults can be overridden via [new_with_options()](Self::new_with_options). |
| 191 |
pub fn new(data: &'a [u8]) -> DeflateDecoder<'a> |
| 192 |
{ |
| 193 |
let options = DeflateOptions::default(); |
| 194 |
|
| 195 |
Self::new_with_options(data, options) |
| 196 |
} |
| 197 |
/// Create new decoder with specified options |
| 198 |
/// |
| 199 |
/// This can be used to fine tune the decoder to the user's |
| 200 |
/// needs. |
| 201 |
/// |
| 202 |
/// |
| 203 |
/// # Arguments |
| 204 |
/// - `data`: The compressed data. Data can be of any format i.e |
| 205 |
/// gzip, zlib or raw deflate. |
| 206 |
/// - `options` : A set of user defined options which tune how the decompressor |
| 207 |
/// |
| 208 |
/// # Returns |
| 209 |
/// A decoder instance which will pull compressed data from `data` to inflate output |
| 210 |
/// |
| 211 |
/// # Example |
| 212 |
/// ```no_run |
| 213 |
/// use zune_inflate::{DeflateDecoder, DeflateOptions}; |
| 214 |
/// let data = [37]; |
| 215 |
/// let options = DeflateOptions::default() |
| 216 |
/// .set_confirm_checksum(true) // confirm the checksum for zlib and gzip |
| 217 |
/// .set_limit(1000); // how big I think the input will be |
| 218 |
/// let mut decoder = DeflateDecoder::new_with_options(&data,options); |
| 219 |
/// // do some stuff and then call decode |
| 220 |
/// let data = decoder.decode_zlib(); |
| 221 |
/// |
| 222 |
/// ``` |
| 223 |
pub fn new_with_options(data: &'a [u8], options: DeflateOptions) -> DeflateDecoder<'a> |
| 224 |
{ |
| 225 |
// create stream |
| 226 |
DeflateDecoder { |
| 227 |
data, |
| 228 |
position: 0, |
| 229 |
stream: BitStreamReader::new(data), |
| 230 |
is_last_block: false, |
| 231 |
static_codes_loaded: false, |
| 232 |
deflate_header_tables: DeflateHeaderTables::default(), |
| 233 |
options |
| 234 |
} |
| 235 |
} |
| 236 |
/// Decode zlib-encoded data returning the uncompressed in a `Vec<u8>` |
| 237 |
/// or an error if something went wrong. |
| 238 |
/// |
| 239 |
/// Bytes consumed will be from the data passed when the |
| 240 |
/// `new` method was called. |
| 241 |
/// |
| 242 |
/// # Arguments |
| 243 |
/// - None |
| 244 |
/// # Returns |
| 245 |
/// Result type containing the decoded data. |
| 246 |
/// |
| 247 |
/// - `Ok(Vec<u8>)`: Decoded vector containing the uncompressed bytes |
| 248 |
/// - `Err(InflateDecodeErrors)`: Error that occurred during decoding |
| 249 |
/// |
| 250 |
/// It's possible to recover bytes even after an error occurred, bytes up |
| 251 |
/// to when error was encountered are stored in [InflateDecodeErrors] |
| 252 |
/// |
| 253 |
/// |
| 254 |
/// # Note |
| 255 |
/// This needs the `zlib` feature enabled to be available otherwise it's a |
| 256 |
/// compile time error |
| 257 |
/// |
| 258 |
/// [InflateDecodeErrors]:crate::errors::InflateDecodeErrors |
| 259 |
/// |
| 260 |
#[cfg(feature = "zlib")] |
| 261 |
pub fn decode_zlib(&mut self) -> Result<Vec<u8>, InflateDecodeErrors> |
| 262 |
{ |
| 263 |
use crate::utils::calc_adler_hash; |
| 264 |
|
| 265 |
if self.data.len() |
| 266 |
< 2 /* zlib header */ |
| 267 |
+ 4 |
| 268 |
/* Deflate */ |
| 269 |
{ |
| 270 |
return Err(InflateDecodeErrors::new_with_error( |
| 271 |
DecodeErrorStatus::InsufficientData |
| 272 |
)); |
| 273 |
} |
| 274 |
|
| 275 |
// Zlib flags |
| 276 |
// See https://www.ietf.org/rfc/rfc1950.txt for |
| 277 |
// the RFC |
| 278 |
let cmf = self.data[0]; |
| 279 |
let flg = self.data[1]; |
| 280 |
|
| 281 |
let cm = cmf & 0xF; |
| 282 |
let cinfo = cmf >> 4; |
| 283 |
|
| 284 |
// let fcheck = flg & 0xF; |
| 285 |
// let fdict = (flg >> 4) & 1; |
| 286 |
// let flevel = flg >> 5; |
| 287 |
|
| 288 |
// confirm we have the right deflate methods |
| 289 |
if cm != 8 |
| 290 |
{ |
| 291 |
if cm == 15 |
| 292 |
{ |
| 293 |
return Err(InflateDecodeErrors::new_with_error(DecodeErrorStatus::Generic( |
| 294 |
"CM of 15 is preserved by the standard,currently don't know how to handle it" |
| 295 |
))); |
| 296 |
} |
| 297 |
return Err(InflateDecodeErrors::new_with_error( |
| 298 |
DecodeErrorStatus::GenericStr(format!("Unknown zlib compression method {cm}")) |
| 299 |
)); |
| 300 |
} |
| 301 |
if cinfo > 7 |
| 302 |
{ |
| 303 |
return Err(InflateDecodeErrors::new_with_error( |
| 304 |
DecodeErrorStatus::GenericStr(format!( |
| 305 |
"Unknown cinfo `{cinfo}` greater than 7, not allowed" |
| 306 |
)) |
| 307 |
)); |
| 308 |
} |
| 309 |
let flag_checks = (u16::from(cmf) * 256) + u16::from(flg); |
| 310 |
|
| 311 |
if flag_checks % 31 != 0 |
| 312 |
{ |
| 313 |
return Err(InflateDecodeErrors::new_with_error( |
| 314 |
DecodeErrorStatus::Generic("FCHECK integrity not preserved") |
| 315 |
)); |
| 316 |
} |
| 317 |
|
| 318 |
self.position = 2; |
| 319 |
|
| 320 |
let data = self.decode_deflate()?; |
| 321 |
|
| 322 |
if self.options.confirm_checksum |
| 323 |
{ |
| 324 |
// Get number of consumed bytes from the input |
| 325 |
let out_pos = self.stream.get_position() + self.position + self.stream.over_read; |
| 326 |
|
| 327 |
// read adler |
| 328 |
if let Some(adler) = self.data.get(out_pos..out_pos + 4) |
| 329 |
{ |
| 330 |
let adler_bits: [u8; 4] = adler.try_into().unwrap(); |
| 331 |
|
| 332 |
let adler32_expected = u32::from_be_bytes(adler_bits); |
| 333 |
|
| 334 |
let adler32_found = calc_adler_hash(&data); |
| 335 |
|
| 336 |
if adler32_expected != adler32_found |
| 337 |
{ |
| 338 |
let err_msg = |
| 339 |
DecodeErrorStatus::MismatchedAdler(adler32_expected, adler32_found); |
| 340 |
let err = InflateDecodeErrors::new(err_msg, data); |
| 341 |
|
| 342 |
return Err(err); |
| 343 |
} |
| 344 |
} |
| 345 |
else |
| 346 |
{ |
| 347 |
let err = InflateDecodeErrors::new(DecodeErrorStatus::InsufficientData, data); |
| 348 |
|
| 349 |
return Err(err); |
| 350 |
} |
| 351 |
} |
| 352 |
|
| 353 |
Ok(data) |
| 354 |
} |
| 355 |
|
| 356 |
/// Decode a gzip encoded data and return the uncompressed data in a |
| 357 |
/// `Vec<u8>` or an error if something went wrong |
| 358 |
/// |
| 359 |
/// Bytes consumed will be from the data passed when the |
| 360 |
/// `new` method was called. |
| 361 |
/// |
| 362 |
/// # Arguments |
| 363 |
/// - None |
| 364 |
/// # Returns |
| 365 |
/// Result type containing the decoded data. |
| 366 |
/// |
| 367 |
/// - `Ok(Vec<u8>)`: Decoded vector containing the uncompressed bytes |
| 368 |
/// - `Err(InflateDecodeErrors)`: Error that occurred during decoding |
| 369 |
/// |
| 370 |
/// It's possible to recover bytes even after an error occurred, bytes up |
| 371 |
/// to when error was encountered are stored in [InflateDecodeErrors] |
| 372 |
/// |
| 373 |
/// # Note |
| 374 |
/// This needs the `gzip` feature enabled to be available, otherwise it's a |
| 375 |
/// compile time error |
| 376 |
/// |
| 377 |
/// [InflateDecodeErrors]:crate::errors::InflateDecodeErrors |
| 378 |
/// |
| 379 |
#[cfg(feature = "gzip")] |
| 380 |
pub fn decode_gzip(&mut self) -> Result<Vec<u8>, InflateDecodeErrors> |
| 381 |
{ |
| 382 |
if self.data.len() < 18 |
| 383 |
{ |
| 384 |
return Err(InflateDecodeErrors::new_with_error( |
| 385 |
DecodeErrorStatus::InsufficientData |
| 386 |
)); |
| 387 |
} |
| 388 |
|
| 389 |
if self.data[self.position] != GZIP_ID1 |
| 390 |
{ |
| 391 |
return Err(InflateDecodeErrors::new_with_error( |
| 392 |
DecodeErrorStatus::CorruptData |
| 393 |
)); |
| 394 |
} |
| 395 |
self.position += 1; |
| 396 |
if self.data[self.position] != GZIP_ID2 |
| 397 |
{ |
| 398 |
return Err(InflateDecodeErrors::new_with_error( |
| 399 |
DecodeErrorStatus::CorruptData |
| 400 |
)); |
| 401 |
} |
| 402 |
self.position += 1; |
| 403 |
|
| 404 |
if self.data[self.position] != GZIP_CM_DEFLATE |
| 405 |
{ |
| 406 |
return Err(InflateDecodeErrors::new_with_error( |
| 407 |
DecodeErrorStatus::CorruptData |
| 408 |
)); |
| 409 |
} |
| 410 |
self.position += 1; |
| 411 |
|
| 412 |
let flg = self.data[self.position]; |
| 413 |
self.position += 1; |
| 414 |
|
| 415 |
// skip mtime |
| 416 |
self.position += 4; |
| 417 |
// skip xfl |
| 418 |
self.position += 1; |
| 419 |
// skip os |
| 420 |
self.position += 1; |
| 421 |
|
| 422 |
if (flg & GZIP_FRESERVED) != 0 |
| 423 |
{ |
| 424 |
return Err(InflateDecodeErrors::new_with_error( |
| 425 |
DecodeErrorStatus::CorruptData |
| 426 |
)); |
| 427 |
} |
| 428 |
// extra field |
| 429 |
if (flg & GZIP_FEXTRA) != 0 |
| 430 |
{ |
| 431 |
let len_bytes = self.data[self.position..self.position + 2] |
| 432 |
.try_into() |
| 433 |
.unwrap(); |
| 434 |
let xlen = usize::from(u16::from_le_bytes(len_bytes)); |
| 435 |
|
| 436 |
self.position += 2; |
| 437 |
|
| 438 |
if self.data.len().saturating_sub(self.position) < xlen + GZIP_FOOTER_SIZE |
| 439 |
{ |
| 440 |
return Err(InflateDecodeErrors::new_with_error( |
| 441 |
DecodeErrorStatus::CorruptData |
| 442 |
)); |
| 443 |
} |
| 444 |
self.position += xlen; |
| 445 |
} |
| 446 |
// original file name zero terminated |
| 447 |
if (flg & GZIP_FNAME) != 0 |
| 448 |
{ |
| 449 |
loop |
| 450 |
{ |
| 451 |
if let Some(byte) = self.data.get(self.position) |
| 452 |
{ |
| 453 |
self.position += 1; |
| 454 |
|
| 455 |
if *byte == 0 |
| 456 |
{ |
| 457 |
break; |
| 458 |
} |
| 459 |
} |
| 460 |
else |
| 461 |
{ |
| 462 |
return Err(InflateDecodeErrors::new_with_error( |
| 463 |
DecodeErrorStatus::InsufficientData |
| 464 |
)); |
| 465 |
} |
| 466 |
} |
| 467 |
} |
| 468 |
// File comment zero terminated |
| 469 |
if (flg & GZIP_FCOMMENT) != 0 |
| 470 |
{ |
| 471 |
loop |
| 472 |
{ |
| 473 |
if let Some(byte) = self.data.get(self.position) |
| 474 |
{ |
| 475 |
self.position += 1; |
| 476 |
|
| 477 |
if *byte == 0 |
| 478 |
{ |
| 479 |
break; |
| 480 |
} |
| 481 |
} |
| 482 |
else |
| 483 |
{ |
| 484 |
return Err(InflateDecodeErrors::new_with_error( |
| 485 |
DecodeErrorStatus::InsufficientData |
| 486 |
)); |
| 487 |
} |
| 488 |
} |
| 489 |
} |
| 490 |
// crc16 for gzip header |
| 491 |
if (flg & GZIP_FHCRC) != 0 |
| 492 |
{ |
| 493 |
self.position += 2; |
| 494 |
} |
| 495 |
|
| 496 |
if self.position + GZIP_FOOTER_SIZE > self.data.len() |
| 497 |
{ |
| 498 |
return Err(InflateDecodeErrors::new_with_error( |
| 499 |
DecodeErrorStatus::InsufficientData |
| 500 |
)); |
| 501 |
} |
| 502 |
|
| 503 |
let data = self.decode_deflate()?; |
| 504 |
|
| 505 |
let mut out_pos = self.stream.get_position() + self.position + self.stream.over_read; |
| 506 |
|
| 507 |
if self.options.confirm_checksum |
| 508 |
{ |
| 509 |
// Get number of consumed bytes from the input |
| 510 |
|
| 511 |
if let Some(crc) = self.data.get(out_pos..out_pos + 4) |
| 512 |
{ |
| 513 |
let crc_bits: [u8; 4] = crc.try_into().unwrap(); |
| 514 |
|
| 515 |
let crc32_expected = u32::from_le_bytes(crc_bits); |
| 516 |
|
| 517 |
let crc32_found = !crate::crc::crc32(&data, !0); |
| 518 |
|
| 519 |
if crc32_expected != crc32_found |
| 520 |
{ |
| 521 |
let err_msg = DecodeErrorStatus::MismatchedCRC(crc32_expected, crc32_found); |
| 522 |
let err = InflateDecodeErrors::new(err_msg, data); |
| 523 |
|
| 524 |
return Err(err); |
| 525 |
} |
| 526 |
} |
| 527 |
else |
| 528 |
{ |
| 529 |
let err = InflateDecodeErrors::new(DecodeErrorStatus::InsufficientData, data); |
| 530 |
|
| 531 |
return Err(err); |
| 532 |
} |
| 533 |
} |
| 534 |
//checksum |
| 535 |
out_pos += 4; |
| 536 |
|
| 537 |
if let Some(val) = self.data.get(out_pos..out_pos + 4) |
| 538 |
{ |
| 539 |
let actual_bytes: [u8; 4] = val.try_into().unwrap(); |
| 540 |
let ac = u32::from_le_bytes(actual_bytes) as usize; |
| 541 |
|
| 542 |
if data.len() != ac |
| 543 |
{ |
| 544 |
let err = DecodeErrorStatus::Generic("ISIZE does not match actual bytes"); |
| 545 |
|
| 546 |
let err = InflateDecodeErrors::new(err, data); |
| 547 |
|
| 548 |
return Err(err); |
| 549 |
} |
| 550 |
} |
| 551 |
else |
| 552 |
{ |
| 553 |
let err = InflateDecodeErrors::new(DecodeErrorStatus::InsufficientData, data); |
| 554 |
|
| 555 |
return Err(err); |
| 556 |
} |
| 557 |
|
| 558 |
Ok(data) |
| 559 |
} |
| 560 |
/// Decode a deflate stream returning the data as `Vec<u8>` or an error |
| 561 |
/// indicating what went wrong. |
| 562 |
/// # Arguments |
| 563 |
/// - None |
| 564 |
/// # Returns |
| 565 |
/// Result type containing the decoded data. |
| 566 |
/// |
| 567 |
/// - `Ok(Vec<u8>)`: Decoded vector containing the uncompressed bytes |
| 568 |
/// - `Err(InflateDecodeErrors)`: Error that occurred during decoding |
| 569 |
/// |
| 570 |
/// It's possible to recover bytes even after an error occurred, bytes up |
| 571 |
/// to when error was encountered are stored in [InflateDecodeErrors] |
| 572 |
/// |
| 573 |
/// |
| 574 |
/// # Example |
| 575 |
/// ```no_run |
| 576 |
/// let data = [42]; // answer to life, the universe and everything |
| 577 |
/// |
| 578 |
/// let mut decoder = zune_inflate::DeflateDecoder::new(&data); |
| 579 |
/// let bytes = decoder.decode_deflate().unwrap(); |
| 580 |
/// ``` |
| 581 |
/// |
| 582 |
/// [InflateDecodeErrors]:crate::errors::InflateDecodeErrors |
| 583 |
pub fn decode_deflate(&mut self) -> Result<Vec<u8>, InflateDecodeErrors> |
| 584 |
{ |
| 585 |
self.start_deflate_block() |
| 586 |
} |
| 587 |
/// Main inner loop for decompressing deflate data |
| 588 |
#[allow(unused_assignments)] |
| 589 |
fn start_deflate_block(&mut self) -> Result<Vec<u8>, InflateDecodeErrors> |
| 590 |
{ |
| 591 |
// start deflate decode |
| 592 |
// re-read the stream so that we can remove code read by zlib |
| 593 |
self.stream = BitStreamReader::new(&self.data[self.position..]); |
| 594 |
|
| 595 |
self.stream.refill(); |
| 596 |
|
| 597 |
// Output space for our decoded bytes. |
| 598 |
let mut out_block = vec![0; self.options.size_hint]; |
| 599 |
// bits used |
| 600 |
|
| 601 |
let mut src_offset = 0; |
| 602 |
let mut dest_offset = 0; |
| 603 |
|
| 604 |
loop |
| 605 |
{ |
| 606 |
self.stream.refill(); |
| 607 |
|
| 608 |
self.is_last_block = self.stream.get_bits(1) == 1; |
| 609 |
let block_type = self.stream.get_bits(2); |
| 610 |
|
| 611 |
if block_type == DEFLATE_BLOCKTYPE_UNCOMPRESSED |
| 612 |
{ |
| 613 |
/* |
| 614 |
* Uncompressed block: copy 'len' bytes literally from the input |
| 615 |
* buffer to the output buffer. |
| 616 |
*/ |
| 617 |
/* |
| 618 |
* The RFC says that |
| 619 |
* skip any remaining bits in current partially |
| 620 |
* processed byte |
| 621 |
* read LEN and NLEN (see next section) |
| 622 |
* copy LEN bytes of data to output |
| 623 |
*/ |
| 624 |
|
| 625 |
if self.stream.over_read > usize::from(self.stream.get_bits_left() >> 3) |
| 626 |
{ |
| 627 |
out_block.truncate(dest_offset); |
| 628 |
|
| 629 |
let err_msg = DecodeErrorStatus::Generic("over-read stream"); |
| 630 |
let error = InflateDecodeErrors::new(err_msg, out_block); |
| 631 |
|
| 632 |
return Err(error); |
| 633 |
} |
| 634 |
let partial_bits = self.stream.get_bits_left() & 7; |
| 635 |
|
| 636 |
self.stream.drop_bits(partial_bits); |
| 637 |
|
| 638 |
let len = self.stream.get_bits(16) as u16; |
| 639 |
let nlen = self.stream.get_bits(16) as u16; |
| 640 |
|
| 641 |
// copy to deflate |
| 642 |
if len != !nlen |
| 643 |
{ |
| 644 |
out_block.truncate(dest_offset); |
| 645 |
|
| 646 |
let err_msg = DecodeErrorStatus::Generic("Len and nlen do not match"); |
| 647 |
let error = InflateDecodeErrors::new(err_msg, out_block); |
| 648 |
|
| 649 |
return Err(error); |
| 650 |
} |
| 651 |
let len = len as usize; |
| 652 |
|
| 653 |
let start = self.stream.get_position() + self.position + self.stream.over_read; |
| 654 |
|
| 655 |
// ensure there is enough space for a fast copy |
| 656 |
if dest_offset + len + FASTCOPY_BYTES > out_block.len() |
| 657 |
{ |
| 658 |
// and if there is not, resize |
| 659 |
let new_len = out_block.len() + RESIZE_BY + len; |
| 660 |
|
| 661 |
out_block.resize(new_len, 0); |
| 662 |
} |
| 663 |
|
| 664 |
if self.data.get((start + len).saturating_sub(1)).is_none() |
| 665 |
{ |
| 666 |
out_block.truncate(dest_offset); |
| 667 |
|
| 668 |
let err_msg = DecodeErrorStatus::CorruptData; |
| 669 |
let error = InflateDecodeErrors::new(err_msg, out_block); |
| 670 |
|
| 671 |
return Err(error); |
| 672 |
} |
| 673 |
if dest_offset > self.options.limit |
| 674 |
{ |
| 675 |
out_block.truncate(dest_offset); |
| 676 |
|
| 677 |
let err_msg = |
| 678 |
DecodeErrorStatus::OutputLimitExceeded(self.options.limit, out_block.len()); |
| 679 |
let error = InflateDecodeErrors::new(err_msg, out_block); |
| 680 |
|
| 681 |
return Err(error); |
| 682 |
} |
| 683 |
|
| 684 |
out_block[dest_offset..dest_offset + len] |
| 685 |
.copy_from_slice(&self.data[start..start + len]); |
| 686 |
|
| 687 |
dest_offset += len; |
| 688 |
|
| 689 |
// get the new position to write. |
| 690 |
self.stream.position = |
| 691 |
len + (self.stream.position - usize::from(self.stream.bits_left >> 3)); |
| 692 |
|
| 693 |
self.stream.reset(); |
| 694 |
|
| 695 |
if self.is_last_block |
| 696 |
{ |
| 697 |
break; |
| 698 |
} |
| 699 |
|
| 700 |
continue; |
| 701 |
} |
| 702 |
else if block_type == DEFLATE_BLOCKTYPE_RESERVED |
| 703 |
{ |
| 704 |
out_block.truncate(dest_offset); |
| 705 |
|
| 706 |
let err_msg = DecodeErrorStatus::Generic("Reserved block type 0b11 encountered"); |
| 707 |
let error = InflateDecodeErrors::new(err_msg, out_block); |
| 708 |
|
| 709 |
return Err(error); |
| 710 |
} |
| 711 |
|
| 712 |
// build decode tables for static and dynamic tables |
| 713 |
match self.build_decode_table(block_type) |
| 714 |
{ |
| 715 |
Ok(_) => (), |
| 716 |
Err(value) => |
| 717 |
{ |
| 718 |
out_block.truncate(dest_offset); |
| 719 |
|
| 720 |
let err_msg = value; |
| 721 |
let error = InflateDecodeErrors::new(err_msg, out_block); |
| 722 |
|
| 723 |
return Err(error); |
| 724 |
} |
| 725 |
}; |
| 726 |
|
| 727 |
// Tables are mutated into the struct, so at this point we know the tables |
| 728 |
// are loaded, take a reference to them |
| 729 |
let litlen_decode_table = &self.deflate_header_tables.litlen_decode_table; |
| 730 |
let offset_decode_table = &self.deflate_header_tables.offset_decode_table; |
| 731 |
|
| 732 |
/* |
| 733 |
* This is the "fast loop" for decoding literals and matches. It does |
| 734 |
* bounds checks on in_next and out_next in the loop conditions so that |
| 735 |
* additional bounds checks aren't needed inside the loop body. |
| 736 |
* |
| 737 |
* To reduce latency, the bit-buffer is refilled and the next litlen |
| 738 |
* decode table entry is preloaded before each loop iteration. |
| 739 |
*/ |
| 740 |
let (mut literal, mut length, mut offset, mut entry) = (0, 0, 0, 0); |
| 741 |
|
| 742 |
let mut saved_bitbuf; |
| 743 |
|
| 744 |
'decode: loop |
| 745 |
{ |
| 746 |
let close_src = 3 * FASTCOPY_BYTES < self.stream.remaining_bytes(); |
| 747 |
|
| 748 |
if close_src |
| 749 |
{ |
| 750 |
self.stream.refill_inner_loop(); |
| 751 |
|
| 752 |
let lit_mask = self.stream.peek_bits::<LITLEN_DECODE_BITS>(); |
| 753 |
|
| 754 |
entry = litlen_decode_table[lit_mask]; |
| 755 |
|
| 756 |
'sequence: loop |
| 757 |
{ |
| 758 |
// Resize the output vector here to ensure we can always have |
| 759 |
// enough space for sloppy copies |
| 760 |
if dest_offset + FASTLOOP_MAX_BYTES_WRITTEN > out_block.len() |
| 761 |
{ |
| 762 |
let curr_len = out_block.len(); |
| 763 |
out_block.resize(curr_len + FASTLOOP_MAX_BYTES_WRITTEN + RESIZE_BY, 0) |
| 764 |
} |
| 765 |
// At this point entry contains the next value of the litlen |
| 766 |
// This will always be the case so meaning all our exit paths need |
| 767 |
// to load in the next entry. |
| 768 |
|
| 769 |
// recheck after every sequence |
| 770 |
// when we hit continue, we need to recheck this |
| 771 |
// as we are trying to emulate a do while |
| 772 |
let new_check = self.stream.src.len() < self.stream.position + 8; |
| 773 |
|
| 774 |
if new_check |
| 775 |
{ |
| 776 |
break 'sequence; |
| 777 |
} |
| 778 |
|
| 779 |
self.stream.refill_inner_loop(); |
| 780 |
/* |
| 781 |
* Consume the bits for the litlen decode table entry. Save the |
| 782 |
* original bit-buf for later, in case the extra match length |
| 783 |
* bits need to be extracted from it. |
| 784 |
*/ |
| 785 |
saved_bitbuf = self.stream.buffer; |
| 786 |
|
| 787 |
self.stream.drop_bits((entry & 0xFF) as u8); |
| 788 |
|
| 789 |
/* |
| 790 |
* Begin by checking for a "fast" literal, i.e. a literal that |
| 791 |
* doesn't need a subtable. |
| 792 |
*/ |
| 793 |
if (entry & HUFFDEC_LITERAL) != 0 |
| 794 |
{ |
| 795 |
/* |
| 796 |
* On 64-bit platforms, we decode up to 2 extra fast |
| 797 |
* literals in addition to the primary item, as this |
| 798 |
* increases performance and still leaves enough bits |
| 799 |
* remaining for what follows. We could actually do 3, |
| 800 |
* assuming LITLEN_TABLEBITS=11, but that actually |
| 801 |
* decreases performance slightly (perhaps by messing |
| 802 |
* with the branch prediction of the conditional refill |
| 803 |
* that happens later while decoding the match offset). |
| 804 |
*/ |
| 805 |
|
| 806 |
literal = entry >> 16; |
| 807 |
|
| 808 |
let new_pos = self.stream.peek_bits::<LITLEN_DECODE_BITS>(); |
| 809 |
|
| 810 |
entry = litlen_decode_table[new_pos]; |
| 811 |
saved_bitbuf = self.stream.buffer; |
| 812 |
|
| 813 |
self.stream.drop_bits(entry as u8); |
| 814 |
|
| 815 |
let out: &mut [u8; 2] = out_block |
| 816 |
.get_mut(dest_offset..dest_offset + 2) |
| 817 |
.unwrap() |
| 818 |
.try_into() |
| 819 |
.unwrap(); |
| 820 |
|
| 821 |
out[0] = literal as u8; |
| 822 |
dest_offset += 1; |
| 823 |
|
| 824 |
if (entry & HUFFDEC_LITERAL) != 0 |
| 825 |
{ |
| 826 |
/* |
| 827 |
* Another fast literal, but this one is in lieu of the |
| 828 |
* primary item, so it doesn't count as one of the extras. |
| 829 |
*/ |
| 830 |
|
| 831 |
// load in the next entry. |
| 832 |
literal = entry >> 16; |
| 833 |
|
| 834 |
let new_pos = self.stream.peek_bits::<LITLEN_DECODE_BITS>(); |
| 835 |
|
| 836 |
entry = litlen_decode_table[new_pos]; |
| 837 |
|
| 838 |
out[1] = literal as u8; |
| 839 |
dest_offset += 1; |
| 840 |
|
| 841 |
continue; |
| 842 |
} |
| 843 |
} |
| 844 |
/* |
| 845 |
* It's not a literal entry, so it can be a length entry, a |
| 846 |
* subtable pointer entry, or an end-of-block entry. Detect the |
| 847 |
* two unlikely cases by testing the HUFFDEC_EXCEPTIONAL flag. |
| 848 |
*/ |
| 849 |
if (entry & HUFFDEC_EXCEPTIONAL) != 0 |
| 850 |
{ |
| 851 |
// Subtable pointer or end of block entry |
| 852 |
if (entry & HUFFDEC_END_OF_BLOCK) != 0 |
| 853 |
{ |
| 854 |
// block done |
| 855 |
break 'decode; |
| 856 |
} |
| 857 |
/* |
| 858 |
* A subtable is required. Load and consume the |
| 859 |
* subtable entry. The subtable entry can be of any |
| 860 |
* type: literal, length, or end-of-block. |
| 861 |
*/ |
| 862 |
let entry_position = ((entry >> 8) & 0x3F) as usize; |
| 863 |
let mut pos = (entry >> 16) as usize; |
| 864 |
|
| 865 |
saved_bitbuf = self.stream.buffer; |
| 866 |
|
| 867 |
pos += self.stream.peek_var_bits(entry_position); |
| 868 |
entry = litlen_decode_table[pos.min(LITLEN_ENOUGH - 1)]; |
| 869 |
|
| 870 |
self.stream.drop_bits(entry as u8); |
| 871 |
|
| 872 |
if (entry & HUFFDEC_LITERAL) != 0 |
| 873 |
{ |
| 874 |
// decode a literal that required a sub table |
| 875 |
let new_pos = self.stream.peek_bits::<LITLEN_DECODE_BITS>(); |
| 876 |
|
| 877 |
literal = entry >> 16; |
| 878 |
entry = litlen_decode_table[new_pos]; |
| 879 |
|
| 880 |
*out_block.get_mut(dest_offset).unwrap_or(&mut 0) = |
| 881 |
(literal & 0xFF) as u8; |
| 882 |
|
| 883 |
dest_offset += 1; |
| 884 |
|
| 885 |
continue; |
| 886 |
} |
| 887 |
|
| 888 |
if (entry & HUFFDEC_END_OF_BLOCK) != 0 |
| 889 |
{ |
| 890 |
break 'decode; |
| 891 |
} |
| 892 |
} |
| 893 |
|
| 894 |
// At this point,we dropped at most 22 bits(LITLEN_DECODE is 11 and we |
| 895 |
// can do it twice), we now just have 34 bits min remaining. |
| 896 |
|
| 897 |
/* |
| 898 |
* Decode the match length: the length base value associated |
| 899 |
* with the litlen symbol (which we extract from the decode |
| 900 |
* table entry), plus the extra length bits. We don't need to |
| 901 |
* consume the extra length bits here, as they were included in |
| 902 |
* the bits consumed by the entry earlier. We also don't need |
| 903 |
* to check for too-long matches here, as this is inside the |
| 904 |
* fast loop where it's already been verified that the output |
| 905 |
* buffer has enough space remaining to copy a max-length match. |
| 906 |
*/ |
| 907 |
let entry_dup = entry; |
| 908 |
|
| 909 |
entry = offset_decode_table[self.stream.peek_bits::<OFFSET_TABLEBITS>()]; |
| 910 |
length = (entry_dup >> 16) as usize; |
| 911 |
|
| 912 |
let mask = (1 << entry_dup as u8) - 1; |
| 913 |
|
| 914 |
length += (saved_bitbuf & mask) as usize >> ((entry_dup >> 8) as u8); |
| 915 |
|
| 916 |
// offset requires a subtable |
| 917 |
if (entry & HUFFDEC_EXCEPTIONAL) != 0 |
| 918 |
{ |
| 919 |
self.stream.drop_bits(OFFSET_TABLEBITS as u8); |
| 920 |
let extra = self.stream.peek_var_bits(((entry >> 8) & 0x3F) as usize); |
| 921 |
entry = offset_decode_table[((entry >> 16) as usize + extra) & 511]; |
| 922 |
// refill to handle some weird edge case where we have |
| 923 |
// less bits than needed for reading the lit-len |
| 924 |
} |
| 925 |
saved_bitbuf = self.stream.buffer; |
| 926 |
|
| 927 |
self.stream.drop_bits((entry & 0xFF) as u8); |
| 928 |
|
| 929 |
let mask = (1 << entry as u8) - 1; |
| 930 |
|
| 931 |
offset = (entry >> 16) as usize; |
| 932 |
offset += (saved_bitbuf & mask) as usize >> (((entry >> 8) & 0xFF) as u8); |
| 933 |
|
| 934 |
if offset > dest_offset |
| 935 |
{ |
| 936 |
out_block.truncate(dest_offset); |
| 937 |
|
| 938 |
let err_msg = DecodeErrorStatus::CorruptData; |
| 939 |
let error = InflateDecodeErrors::new(err_msg, out_block); |
| 940 |
|
| 941 |
return Err(error); |
| 942 |
} |
| 943 |
|
| 944 |
src_offset = dest_offset - offset; |
| 945 |
|
| 946 |
if self.stream.bits_left < 11 |
| 947 |
{ |
| 948 |
self.stream.refill_inner_loop(); |
| 949 |
} |
| 950 |
// Copy some bytes unconditionally |
| 951 |
// This makes us copy smaller match lengths quicker because we don't need |
| 952 |
// a loop + don't send too much pressure to the Memory unit. |
| 953 |
fixed_copy_within::<FASTCOPY_BYTES>( |
| 954 |
&mut out_block, |
| 955 |
src_offset, |
| 956 |
dest_offset |
| 957 |
); |
| 958 |
|
| 959 |
entry = litlen_decode_table[self.stream.peek_bits::<LITLEN_DECODE_BITS>()]; |
| 960 |
|
| 961 |
let mut current_position = dest_offset; |
| 962 |
|
| 963 |
dest_offset += length; |
| 964 |
|
| 965 |
if offset == 1 |
| 966 |
{ |
| 967 |
// RLE fill with a single byte |
| 968 |
let byte_to_repeat = out_block[src_offset]; |
| 969 |
out_block[current_position..dest_offset].fill(byte_to_repeat); |
| 970 |
} |
| 971 |
else if offset <= FASTCOPY_BYTES |
| 972 |
&& current_position + offset < dest_offset |
| 973 |
{ |
| 974 |
// The second conditional ensures we only come |
| 975 |
// here if the first copy didn't succeed to copy just enough bytes for a rep |
| 976 |
// match to be valid, i.e we want this path to be taken the least amount |
| 977 |
// of times possible |
| 978 |
|
| 979 |
// the unconditional copy above copied some bytes |
| 980 |
// don't let it go into waste |
| 981 |
// Increment the position we are in by the number of correct bytes |
| 982 |
// currently copied |
| 983 |
let mut src_position = src_offset + offset; |
| 984 |
let mut dest_position = current_position + offset; |
| 985 |
|
| 986 |
// loop copying offset bytes in place |
| 987 |
// notice this loop does fixed copies but increments in offset bytes :) |
| 988 |
// that is intentional. |
| 989 |
loop |
| 990 |
{ |
| 991 |
fixed_copy_within::<FASTCOPY_BYTES>( |
| 992 |
&mut out_block, |
| 993 |
src_position, |
| 994 |
dest_position |
| 995 |
); |
| 996 |
|
| 997 |
src_position += offset; |
| 998 |
dest_position += offset; |
| 999 |
|
| 1000 |
if dest_position > dest_offset |
| 1001 |
{ |
| 1002 |
break; |
| 1003 |
} |
| 1004 |
} |
| 1005 |
} |
| 1006 |
else if length > FASTCOPY_BYTES |
| 1007 |
{ |
| 1008 |
current_position += FASTCOPY_BYTES; |
| 1009 |
// fast non-overlapping copy |
| 1010 |
// |
| 1011 |
// We have enough space to write the ML+FAST_COPY bytes ahead |
| 1012 |
// so we know this won't come to shoot us in the foot. |
| 1013 |
// |
| 1014 |
// An optimization is to copy FAST_COPY_BITS per invocation |
| 1015 |
// Currently FASTCOPY_BYTES is 16, this fits in nicely as we |
| 1016 |
// it's a single SIMD instruction on a lot of things, i.e x86,Arm and even |
| 1017 |
// wasm. |
| 1018 |
|
| 1019 |
// current position of the match |
| 1020 |
let mut dest_src_offset = src_offset + FASTCOPY_BYTES; |
| 1021 |
|
| 1022 |
// Number of bytes we are to copy |
| 1023 |
// copy in batches of FAST_BYTES |
| 1024 |
'match_lengths: loop |
| 1025 |
{ |
| 1026 |
// Safety: We resized out_block hence we know it can handle |
| 1027 |
// sloppy copies without it being out of bounds |
| 1028 |
// |
| 1029 |
// Reason: This is a latency critical loop, even branches start |
| 1030 |
// to matter |
| 1031 |
fixed_copy_within::<FASTCOPY_BYTES>( |
| 1032 |
&mut out_block, |
| 1033 |
dest_src_offset, |
| 1034 |
current_position |
| 1035 |
); |
| 1036 |
|
| 1037 |
dest_src_offset += FASTCOPY_BYTES; |
| 1038 |
current_position += FASTCOPY_BYTES; |
| 1039 |
|
| 1040 |
if current_position > dest_offset |
| 1041 |
{ |
| 1042 |
break 'match_lengths; |
| 1043 |
} |
| 1044 |
} |
| 1045 |
} |
| 1046 |
|
| 1047 |
if dest_offset > self.options.limit |
| 1048 |
{ |
| 1049 |
out_block.truncate(dest_offset); |
| 1050 |
|
| 1051 |
let err_msg = DecodeErrorStatus::OutputLimitExceeded( |
| 1052 |
self.options.limit, |
| 1053 |
dest_offset |
| 1054 |
); |
| 1055 |
let error = InflateDecodeErrors::new(err_msg, out_block); |
| 1056 |
|
| 1057 |
return Err(error); |
| 1058 |
} |
| 1059 |
|
| 1060 |
if self.stream.src.len() < self.stream.position + 8 |
| 1061 |
{ |
| 1062 |
// close to input end, move to the slower one |
| 1063 |
break 'sequence; |
| 1064 |
} |
| 1065 |
} |
| 1066 |
} |
| 1067 |
// generic loop that does things a bit slower but it's okay since it doesn't |
| 1068 |
// deal with a lot of things |
| 1069 |
// We can afford to be more careful here, checking that we do |
| 1070 |
// not drop non-existent bits etc etc as we do not have the |
| 1071 |
// assurances of the fast loop bits above. |
| 1072 |
loop |
| 1073 |
{ |
| 1074 |
self.stream.refill(); |
| 1075 |
|
| 1076 |
if self.stream.over_read > usize::from(self.stream.bits_left >> 3) |
| 1077 |
{ |
| 1078 |
out_block.truncate(dest_offset); |
| 1079 |
|
| 1080 |
let err_msg = DecodeErrorStatus::CorruptData; |
| 1081 |
let error = InflateDecodeErrors::new(err_msg, out_block); |
| 1082 |
|
| 1083 |
return Err(error); |
| 1084 |
} |
| 1085 |
|
| 1086 |
let literal_mask = self.stream.peek_bits::<LITLEN_DECODE_BITS>(); |
| 1087 |
|
| 1088 |
entry = litlen_decode_table[literal_mask]; |
| 1089 |
|
| 1090 |
saved_bitbuf = self.stream.buffer; |
| 1091 |
|
| 1092 |
self.stream.drop_bits((entry & 0xFF) as u8); |
| 1093 |
|
| 1094 |
if (entry & HUFFDEC_SUITABLE_POINTER) != 0 |
| 1095 |
{ |
| 1096 |
let extra = self.stream.peek_var_bits(((entry >> 8) & 0x3F) as usize); |
| 1097 |
|
| 1098 |
entry = litlen_decode_table[(entry >> 16) as usize + extra]; |
| 1099 |
saved_bitbuf = self.stream.buffer; |
| 1100 |
|
| 1101 |
self.stream.drop_bits((entry & 0xFF) as u8); |
| 1102 |
} |
| 1103 |
|
| 1104 |
length = (entry >> 16) as usize; |
| 1105 |
|
| 1106 |
if (entry & HUFFDEC_LITERAL) != 0 |
| 1107 |
{ |
| 1108 |
resize_and_push(&mut out_block, dest_offset, length as u8); |
| 1109 |
|
| 1110 |
dest_offset += 1; |
| 1111 |
|
| 1112 |
continue; |
| 1113 |
} |
| 1114 |
|
| 1115 |
if (entry & HUFFDEC_END_OF_BLOCK) != 0 |
| 1116 |
{ |
| 1117 |
break 'decode; |
| 1118 |
} |
| 1119 |
|
| 1120 |
let mask = (1 << entry as u8) - 1; |
| 1121 |
|
| 1122 |
length += (saved_bitbuf & mask) as usize >> ((entry >> 8) as u8); |
| 1123 |
|
| 1124 |
self.stream.refill(); |
| 1125 |
|
| 1126 |
entry = offset_decode_table[self.stream.peek_bits::<OFFSET_TABLEBITS>()]; |
| 1127 |
|
| 1128 |
if (entry & HUFFDEC_EXCEPTIONAL) != 0 |
| 1129 |
{ |
| 1130 |
// offset requires a subtable |
| 1131 |
self.stream.drop_bits(OFFSET_TABLEBITS as u8); |
| 1132 |
|
| 1133 |
let extra = self.stream.peek_var_bits(((entry >> 8) & 0x3F) as usize); |
| 1134 |
|
| 1135 |
entry = offset_decode_table[((entry >> 16) as usize + extra) & 511]; |
| 1136 |
} |
| 1137 |
|
| 1138 |
// ensure there is enough space for a fast copy |
| 1139 |
if dest_offset + length + FASTCOPY_BYTES > out_block.len() |
| 1140 |
{ |
| 1141 |
let new_len = out_block.len() + RESIZE_BY + length; |
| 1142 |
out_block.resize(new_len, 0); |
| 1143 |
} |
| 1144 |
saved_bitbuf = self.stream.buffer; |
| 1145 |
|
| 1146 |
let mask = (1 << (entry & 0xFF) as u8) - 1; |
| 1147 |
|
| 1148 |
offset = (entry >> 16) as usize; |
| 1149 |
offset += (saved_bitbuf & mask) as usize >> ((entry >> 8) as u8); |
| 1150 |
|
| 1151 |
if offset > dest_offset |
| 1152 |
{ |
| 1153 |
out_block.truncate(dest_offset); |
| 1154 |
|
| 1155 |
let err_msg = DecodeErrorStatus::CorruptData; |
| 1156 |
let error = InflateDecodeErrors::new(err_msg, out_block); |
| 1157 |
|
| 1158 |
return Err(error); |
| 1159 |
} |
| 1160 |
|
| 1161 |
src_offset = dest_offset - offset; |
| 1162 |
|
| 1163 |
self.stream.drop_bits(entry as u8); |
| 1164 |
|
| 1165 |
let (dest_src, dest_ptr) = out_block.split_at_mut(dest_offset); |
| 1166 |
|
| 1167 |
if src_offset + length + FASTCOPY_BYTES > dest_offset |
| 1168 |
{ |
| 1169 |
// overlapping copy |
| 1170 |
// do a simple rep match |
| 1171 |
copy_rep_matches(&mut out_block, src_offset, dest_offset, length); |
| 1172 |
} |
| 1173 |
else |
| 1174 |
{ |
| 1175 |
dest_ptr[0..length] |
| 1176 |
.copy_from_slice(&dest_src[src_offset..src_offset + length]); |
| 1177 |
} |
| 1178 |
|
| 1179 |
dest_offset += length; |
| 1180 |
|
| 1181 |
if dest_offset > self.options.limit |
| 1182 |
{ |
| 1183 |
out_block.truncate(dest_offset); |
| 1184 |
|
| 1185 |
let err_msg = |
| 1186 |
DecodeErrorStatus::OutputLimitExceeded(self.options.limit, dest_offset); |
| 1187 |
let error = InflateDecodeErrors::new(err_msg, out_block); |
| 1188 |
|
| 1189 |
return Err(error); |
| 1190 |
} |
| 1191 |
} |
| 1192 |
} |
| 1193 |
/* |
| 1194 |
* If any of the implicit appended zero bytes were consumed (not just |
| 1195 |
* refilled) before hitting end of stream, then the data is bad. |
| 1196 |
*/ |
| 1197 |
if self.stream.over_read > usize::from(self.stream.bits_left >> 3) |
| 1198 |
{ |
| 1199 |
out_block.truncate(dest_offset); |
| 1200 |
|
| 1201 |
let err_msg = DecodeErrorStatus::CorruptData; |
| 1202 |
let error = InflateDecodeErrors::new(err_msg, out_block); |
| 1203 |
|
| 1204 |
return Err(error); |
| 1205 |
} |
| 1206 |
|
| 1207 |
if self.is_last_block |
| 1208 |
{ |
| 1209 |
break; |
| 1210 |
} |
| 1211 |
} |
| 1212 |
|
| 1213 |
// decompression. DONE |
| 1214 |
// Truncate data to match the number of actual |
| 1215 |
// bytes written. |
| 1216 |
out_block.truncate(dest_offset); |
| 1217 |
|
| 1218 |
Ok(out_block) |
| 1219 |
} |
| 1220 |
|
| 1221 |
/// Build decode tables for static and dynamic |
| 1222 |
/// huffman blocks. |
| 1223 |
fn build_decode_table(&mut self, block_type: u64) -> Result<(), DecodeErrorStatus> |
| 1224 |
{ |
| 1225 |
const COUNT: usize = |
| 1226 |
DEFLATE_NUM_LITLEN_SYMS + DEFLATE_NUM_OFFSET_SYMS + DELFATE_MAX_LENS_OVERRUN; |
| 1227 |
|
| 1228 |
let mut lens = [0_u8; COUNT]; |
| 1229 |
let mut precode_lens = [0; DEFLATE_NUM_PRECODE_SYMS]; |
| 1230 |
let mut precode_decode_table = [0_u32; PRECODE_ENOUGH]; |
| 1231 |
let mut litlen_decode_table = [0_u32; LITLEN_ENOUGH]; |
| 1232 |
let mut offset_decode_table = [0; OFFSET_ENOUGH]; |
| 1233 |
|
| 1234 |
let mut num_litlen_syms = 0; |
| 1235 |
let mut num_offset_syms = 0; |
| 1236 |
|
| 1237 |
if block_type == DEFLATE_BLOCKTYPE_DYNAMIC_HUFFMAN |
| 1238 |
{ |
| 1239 |
const SINGLE_PRECODE: usize = 3; |
| 1240 |
|
| 1241 |
self.static_codes_loaded = false; |
| 1242 |
|
| 1243 |
// Dynamic Huffman block |
| 1244 |
// Read codeword lengths |
| 1245 |
if !self.stream.has(5 + 5 + 4) |
| 1246 |
{ |
| 1247 |
return Err(DecodeErrorStatus::InsufficientData); |
| 1248 |
} |
| 1249 |
|
| 1250 |
num_litlen_syms = 257 + (self.stream.get_bits(5)) as usize; |
| 1251 |
num_offset_syms = 1 + (self.stream.get_bits(5)) as usize; |
| 1252 |
|
| 1253 |
let num_explicit_precode_lens = 4 + (self.stream.get_bits(4)) as usize; |
| 1254 |
|
| 1255 |
self.stream.refill(); |
| 1256 |
|
| 1257 |
if !self.stream.has(3) |
| 1258 |
{ |
| 1259 |
return Err(DecodeErrorStatus::InsufficientData); |
| 1260 |
} |
| 1261 |
|
| 1262 |
let first_precode = self.stream.get_bits(3) as u8; |
| 1263 |
let expected = (SINGLE_PRECODE * num_explicit_precode_lens.saturating_sub(1)) as u8; |
| 1264 |
|
| 1265 |
precode_lens[usize::from(DEFLATE_PRECODE_LENS_PERMUTATION[0])] = first_precode; |
| 1266 |
|
| 1267 |
self.stream.refill(); |
| 1268 |
|
| 1269 |
if !self.stream.has(expected) |
| 1270 |
{ |
| 1271 |
return Err(DecodeErrorStatus::InsufficientData); |
| 1272 |
} |
| 1273 |
|
| 1274 |
for i in DEFLATE_PRECODE_LENS_PERMUTATION[1..] |
| 1275 |
.iter() |
| 1276 |
.take(num_explicit_precode_lens - 1) |
| 1277 |
{ |
| 1278 |
let bits = self.stream.get_bits(3) as u8; |
| 1279 |
|
| 1280 |
precode_lens[usize::from(*i)] = bits; |
| 1281 |
} |
| 1282 |
|
| 1283 |
self.build_decode_table_inner( |
| 1284 |
&precode_lens, |
| 1285 |
&PRECODE_DECODE_RESULTS, |
| 1286 |
&mut precode_decode_table, |
| 1287 |
PRECODE_TABLE_BITS, |
| 1288 |
DEFLATE_NUM_PRECODE_SYMS, |
| 1289 |
DEFLATE_MAX_CODEWORD_LENGTH |
| 1290 |
)?; |
| 1291 |
|
| 1292 |
/* Decode the litlen and offset codeword lengths. */ |
| 1293 |
|
| 1294 |
let mut i = 0; |
| 1295 |
|
| 1296 |
loop |
| 1297 |
{ |
| 1298 |
if i >= num_litlen_syms + num_offset_syms |
| 1299 |
{ |
| 1300 |
// confirm here since with a continue loop stuff |
| 1301 |
// breaks |
| 1302 |
break; |
| 1303 |
} |
| 1304 |
|
| 1305 |
let rep_val: u8; |
| 1306 |
let rep_count: u64; |
| 1307 |
|
| 1308 |
if !self.stream.has(DEFLATE_MAX_PRE_CODEWORD_LEN + 7) |
| 1309 |
{ |
| 1310 |
self.stream.refill(); |
| 1311 |
} |
| 1312 |
// decode next pre-code symbol |
| 1313 |
let entry_pos = self |
| 1314 |
.stream |
| 1315 |
.peek_bits::<{ DEFLATE_MAX_PRE_CODEWORD_LEN as usize }>(); |
| 1316 |
|
| 1317 |
let entry = precode_decode_table[entry_pos]; |
| 1318 |
let presym = entry >> 16; |
| 1319 |
|
| 1320 |
if !self.stream.has(entry as u8) |
| 1321 |
{ |
| 1322 |
return Err(DecodeErrorStatus::InsufficientData); |
| 1323 |
} |
| 1324 |
|
| 1325 |
self.stream.drop_bits(entry as u8); |
| 1326 |
|
| 1327 |
if presym < 16 |
| 1328 |
{ |
| 1329 |
// explicit codeword length |
| 1330 |
lens[i] = presym as u8; |
| 1331 |
i += 1; |
| 1332 |
continue; |
| 1333 |
} |
| 1334 |
|
| 1335 |
/* Run-length encoded codeword lengths */ |
| 1336 |
|
| 1337 |
/* |
| 1338 |
* Note: we don't need verify that the repeat count |
| 1339 |
* doesn't overflow the number of elements, since we've |
| 1340 |
* sized the lens array to have enough extra space to |
| 1341 |
* allow for the worst-case overrun (138 zeroes when |
| 1342 |
* only 1 length was remaining). |
| 1343 |
* |
| 1344 |
* In the case of the small repeat counts (presyms 16 |
| 1345 |
* and 17), it is fastest to always write the maximum |
| 1346 |
* number of entries. That gets rid of branches that |
| 1347 |
* would otherwise be required. |
| 1348 |
* |
| 1349 |
* It is not just because of the numerical order that |
| 1350 |
* our checks go in the order 'presym < 16', 'presym == |
| 1351 |
* 16', and 'presym == 17'. For typical data this is |
| 1352 |
* ordered from most frequent to least frequent case. |
| 1353 |
*/ |
| 1354 |
if presym == 16 |
| 1355 |
{ |
| 1356 |
if i == 0 |
| 1357 |
{ |
| 1358 |
return Err(DecodeErrorStatus::CorruptData); |
| 1359 |
} |
| 1360 |
|
| 1361 |
if !self.stream.has(2) |
| 1362 |
{ |
| 1363 |
return Err(DecodeErrorStatus::InsufficientData); |
| 1364 |
} |
| 1365 |
|
| 1366 |
// repeat previous length three to 6 times |
| 1367 |
rep_val = lens[i - 1]; |
| 1368 |
rep_count = 3 + self.stream.get_bits(2); |
| 1369 |
lens[i..i + 6].fill(rep_val); |
| 1370 |
i += rep_count as usize; |
| 1371 |
} |
| 1372 |
else if presym == 17 |
| 1373 |
{ |
| 1374 |
if !self.stream.has(3) |
| 1375 |
{ |
| 1376 |
return Err(DecodeErrorStatus::InsufficientData); |
| 1377 |
} |
| 1378 |
/* Repeat zero 3 - 10 times. */ |
| 1379 |
rep_count = 3 + self.stream.get_bits(3); |
| 1380 |
lens[i..i + 10].fill(0); |
| 1381 |
i += rep_count as usize; |
| 1382 |
} |
| 1383 |
else |
| 1384 |
{ |
| 1385 |
if !self.stream.has(7) |
| 1386 |
{ |
| 1387 |
return Err(DecodeErrorStatus::InsufficientData); |
| 1388 |
} |
| 1389 |
// repeat zero 11-138 times. |
| 1390 |
rep_count = 11 + self.stream.get_bits(7); |
| 1391 |
lens[i..i + rep_count as usize].fill(0); |
| 1392 |
i += rep_count as usize; |
| 1393 |
} |
| 1394 |
|
| 1395 |
if i >= num_litlen_syms + num_offset_syms |
| 1396 |
{ |
| 1397 |
break; |
| 1398 |
} |
| 1399 |
} |
| 1400 |
} |
| 1401 |
else if block_type == DEFLATE_BLOCKTYPE_STATIC |
| 1402 |
{ |
| 1403 |
if self.static_codes_loaded |
| 1404 |
{ |
| 1405 |
return Ok(()); |
| 1406 |
} |
| 1407 |
|
| 1408 |
self.static_codes_loaded = true; |
| 1409 |
|
| 1410 |
lens[000..144].fill(8); |
| 1411 |
lens[144..256].fill(9); |
| 1412 |
lens[256..280].fill(7); |
| 1413 |
lens[280..288].fill(8); |
| 1414 |
lens[288..].fill(5); |
| 1415 |
|
| 1416 |
num_litlen_syms = 288; |
| 1417 |
num_offset_syms = 32; |
| 1418 |
} |
| 1419 |
// build offset decode table |
| 1420 |
self.build_decode_table_inner( |
| 1421 |
&lens[num_litlen_syms..], |
| 1422 |
&OFFSET_DECODE_RESULTS, |
| 1423 |
&mut offset_decode_table, |
| 1424 |
OFFSET_TABLEBITS, |
| 1425 |
num_offset_syms, |
| 1426 |
DEFLATE_MAX_OFFSET_CODEWORD_LENGTH |
| 1427 |
)?; |
| 1428 |
|
| 1429 |
self.build_decode_table_inner( |
| 1430 |
&lens, |
| 1431 |
&LITLEN_DECODE_RESULTS, |
| 1432 |
&mut litlen_decode_table, |
| 1433 |
LITLEN_TABLE_BITS, |
| 1434 |
num_litlen_syms, |
| 1435 |
DEFLATE_MAX_LITLEN_CODEWORD_LENGTH |
| 1436 |
)?; |
| 1437 |
|
| 1438 |
self.deflate_header_tables.offset_decode_table = offset_decode_table; |
| 1439 |
self.deflate_header_tables.litlen_decode_table = litlen_decode_table; |
| 1440 |
|
| 1441 |
Ok(()) |
| 1442 |
} |
| 1443 |
/// Build the decode table for the precode |
| 1444 |
#[allow(clippy::needless_range_loop)] |
| 1445 |
fn build_decode_table_inner( |
| 1446 |
&mut self, lens: &[u8], decode_results: &[u32], decode_table: &mut [u32], |
| 1447 |
table_bits: usize, num_syms: usize, mut max_codeword_len: usize |
| 1448 |
) -> Result<(), DecodeErrorStatus> |
| 1449 |
{ |
| 1450 |
const BITS: u32 = usize::BITS - 1; |
| 1451 |
|
| 1452 |
let mut len_counts: [u32; DEFLATE_MAX_CODEWORD_LENGTH + 1] = |
| 1453 |
[0; DEFLATE_MAX_CODEWORD_LENGTH + 1]; |
| 1454 |
let mut offsets: [u32; DEFLATE_MAX_CODEWORD_LENGTH + 1] = |
| 1455 |
[0; DEFLATE_MAX_CODEWORD_LENGTH + 1]; |
| 1456 |
let mut sorted_syms: [u16; DEFLATE_MAX_NUM_SYMS] = [0; DEFLATE_MAX_NUM_SYMS]; |
| 1457 |
|
| 1458 |
let mut i; |
| 1459 |
|
| 1460 |
// count how many codewords have each length, including 0. |
| 1461 |
for sym in 0..num_syms |
| 1462 |
{ |
| 1463 |
len_counts[usize::from(lens[sym])] += 1; |
| 1464 |
} |
| 1465 |
|
| 1466 |
/* |
| 1467 |
* Determine the actual maximum codeword length that was used, and |
| 1468 |
* decrease table_bits to it if allowed. |
| 1469 |
*/ |
| 1470 |
while max_codeword_len > 1 && len_counts[max_codeword_len] == 0 |
| 1471 |
{ |
| 1472 |
max_codeword_len -= 1; |
| 1473 |
} |
| 1474 |
/* |
| 1475 |
* Sort the symbols primarily by increasing codeword length and |
| 1476 |
* A temporary array of length @num_syms. |
| 1477 |
* secondarily by increasing symbol value; or equivalently by their |
| 1478 |
* codewords in lexicographic order, since a canonical code is assumed. |
| 1479 |
* |
| 1480 |
* For efficiency, also compute 'codespace_used' in the same pass over |
| 1481 |
* 'len_counts[]' used to build 'offsets[]' for sorting. |
| 1482 |
*/ |
| 1483 |
offsets[0] = 0; |
| 1484 |
offsets[1] = len_counts[0]; |
| 1485 |
|
| 1486 |
let mut codespace_used = 0_u32; |
| 1487 |
|
| 1488 |
for len in 1..max_codeword_len |
| 1489 |
{ |
| 1490 |
offsets[len + 1] = offsets[len] + len_counts[len]; |
| 1491 |
codespace_used = (codespace_used << 1) + len_counts[len]; |
| 1492 |
} |
| 1493 |
codespace_used = (codespace_used << 1) + len_counts[max_codeword_len]; |
| 1494 |
|
| 1495 |
for sym in 0..num_syms |
| 1496 |
{ |
| 1497 |
let pos = usize::from(lens[sym]); |
| 1498 |
sorted_syms[offsets[pos] as usize] = sym as u16; |
| 1499 |
offsets[pos] += 1; |
| 1500 |
} |
| 1501 |
i = (offsets[0]) as usize; |
| 1502 |
|
| 1503 |
/* |
| 1504 |
* Check whether the lengths form a complete code (exactly fills the |
| 1505 |
* codespace), an incomplete code (doesn't fill the codespace), or an |
| 1506 |
* overfull code (overflows the codespace). A codeword of length 'n' |
| 1507 |
* uses proportion '1/(2^n)' of the codespace. An overfull code is |
| 1508 |
* nonsensical, so is considered invalid. An incomplete code is |
| 1509 |
* considered valid only in two specific cases; see below. |
| 1510 |
*/ |
| 1511 |
|
| 1512 |
// Overfull code |
| 1513 |
if codespace_used > 1 << max_codeword_len |
| 1514 |
{ |
| 1515 |
return Err(DecodeErrorStatus::Generic("Overflown code")); |
| 1516 |
} |
| 1517 |
// incomplete code |
| 1518 |
if codespace_used < 1 << max_codeword_len |
| 1519 |
{ |
| 1520 |
let entry = if codespace_used == 0 |
| 1521 |
{ |
| 1522 |
/* |
| 1523 |
* An empty code is allowed. This can happen for the |
| 1524 |
* offset code in DEFLATE, since a dynamic Huffman block |
| 1525 |
* need not contain any matches. |
| 1526 |
*/ |
| 1527 |
|
| 1528 |
/* sym=0, len=1 (arbitrary) */ |
| 1529 |
make_decode_table_entry(decode_results, 0, 1) |
| 1530 |
} |
| 1531 |
else |
| 1532 |
{ |
| 1533 |
/* |
| 1534 |
* Allow codes with a single used symbol, with codeword |
| 1535 |
* length 1. The DEFLATE RFC is unclear regarding this |
| 1536 |
* case. What zlib's decompressor does is permit this |
| 1537 |
* for the litlen and offset codes and assume the |
| 1538 |
* codeword is '0' rather than '1'. We do the same |
| 1539 |
* except we allow this for precodes too, since there's |
| 1540 |
* no convincing reason to treat the codes differently. |
| 1541 |
* We also assign both codewords '0' and '1' to the |
| 1542 |
* symbol to avoid having to handle '1' specially. |
| 1543 |
*/ |
| 1544 |
if codespace_used != 1 << (max_codeword_len - 1) || len_counts[1] != 1 |
| 1545 |
{ |
| 1546 |
return Err(DecodeErrorStatus::Generic( |
| 1547 |
"Cannot work with empty pre-code table" |
| 1548 |
)); |
| 1549 |
} |
| 1550 |
make_decode_table_entry(decode_results, usize::from(sorted_syms[i]), 1) |
| 1551 |
}; |
| 1552 |
/* |
| 1553 |
* Note: the decode table still must be fully initialized, in |
| 1554 |
* case the stream is malformed and contains bits from the part |
| 1555 |
* of the codespace the incomplete code doesn't use. |
| 1556 |
*/ |
| 1557 |
decode_table.fill(entry); |
| 1558 |
return Ok(()); |
| 1559 |
} |
| 1560 |
|
| 1561 |
/* |
| 1562 |
* The lengths form a complete code. Now, enumerate the codewords in |
| 1563 |
* lexicographic order and fill the decode table entries for each one. |
| 1564 |
* |
| 1565 |
* First, process all codewords with len <= table_bits. Each one gets |
| 1566 |
* '2^(table_bits-len)' direct entries in the table. |
| 1567 |
* |
| 1568 |
* Since DEFLATE uses bit-reversed codewords, these entries aren't |
| 1569 |
* consecutive but rather are spaced '2^len' entries apart. This makes |
| 1570 |
* filling them naively somewhat awkward and inefficient, since strided |
| 1571 |
* stores are less cache-friendly and preclude the use of word or |
| 1572 |
* vector-at-a-time stores to fill multiple entries per instruction. |
| 1573 |
* |
| 1574 |
* To optimize this, we incrementally double the table size. When |
| 1575 |
* processing codewords with length 'len', the table is treated as |
| 1576 |
* having only '2^len' entries, so each codeword uses just one entry. |
| 1577 |
* Then, each time 'len' is incremented, the table size is doubled and |
| 1578 |
* the first half is copied to the second half. This significantly |
| 1579 |
* improves performance over naively doing strided stores. |
| 1580 |
* |
| 1581 |
* Note that some entries copied for each table doubling may not have |
| 1582 |
* been initialized yet, but it doesn't matter since they're guaranteed |
| 1583 |
* to be initialized later (because the Huffman code is complete). |
| 1584 |
*/ |
| 1585 |
let mut codeword = 0; |
| 1586 |
let mut len = 1; |
| 1587 |
let mut count = len_counts[1]; |
| 1588 |
|
| 1589 |
while count == 0 |
| 1590 |
{ |
| 1591 |
len += 1; |
| 1592 |
|
| 1593 |
if len >= len_counts.len() |
| 1594 |
{ |
| 1595 |
break; |
| 1596 |
} |
| 1597 |
count = len_counts[len]; |
| 1598 |
} |
| 1599 |
|
| 1600 |
let mut curr_table_end = 1 << len; |
| 1601 |
|
| 1602 |
while len <= table_bits |
| 1603 |
{ |
| 1604 |
// Process all count codewords with length len |
| 1605 |
loop |
| 1606 |
{ |
| 1607 |
let entry = make_decode_table_entry( |
| 1608 |
decode_results, |
| 1609 |
usize::from(sorted_syms[i]), |
| 1610 |
len as u32 |
| 1611 |
); |
| 1612 |
i += 1; |
| 1613 |
// fill first entry for current codeword |
| 1614 |
decode_table[codeword] = entry; |
| 1615 |
|
| 1616 |
if codeword == curr_table_end - 1 |
| 1617 |
{ |
| 1618 |
// last codeword (all 1's) |
| 1619 |
for _ in len..table_bits |
| 1620 |
{ |
| 1621 |
decode_table.copy_within(0..curr_table_end, curr_table_end); |
| 1622 |
|
| 1623 |
curr_table_end <<= 1; |
| 1624 |
} |
| 1625 |
return Ok(()); |
| 1626 |
} |
| 1627 |
/* |
| 1628 |
* To advance to the lexicographically next codeword in |
| 1629 |
* the canonical code, the codeword must be incremented, |
| 1630 |
* then 0's must be appended to the codeword as needed |
| 1631 |
* to match the next codeword's length. |
| 1632 |
* |
| 1633 |
* Since the codeword is bit-reversed, appending 0's is |
| 1634 |
* a no-op. However, incrementing it is nontrivial. To |
| 1635 |
* do so efficiently, use the 'bsr' instruction to find |
| 1636 |
* the last (highest order) 0 bit in the codeword, set |
| 1637 |
* it, and clear any later (higher order) 1 bits. But |
| 1638 |
* 'bsr' actually finds the highest order 1 bit, so to |
| 1639 |
* use it first flip all bits in the codeword by XOR' ing |
| 1640 |
* it with (1U << len) - 1 == cur_table_end - 1. |
| 1641 |
*/ |
| 1642 |
|
| 1643 |
let adv = BITS - (codeword ^ (curr_table_end - 1)).leading_zeros(); |
| 1644 |
let bit = 1 << adv; |
| 1645 |
|
| 1646 |
codeword &= bit - 1; |
| 1647 |
codeword |= bit; |
| 1648 |
count -= 1; |
| 1649 |
|
| 1650 |
if count == 0 |
| 1651 |
{ |
| 1652 |
break; |
| 1653 |
} |
| 1654 |
} |
| 1655 |
// advance to the next codeword length |
| 1656 |
loop |
| 1657 |
{ |
| 1658 |
len += 1; |
| 1659 |
|
| 1660 |
if len <= table_bits |
| 1661 |
{ |
| 1662 |
// dest is decode_table[curr_table_end] |
| 1663 |
// source is decode_table(start of table); |
| 1664 |
// size is curr_table; |
| 1665 |
|
| 1666 |
decode_table.copy_within(0..curr_table_end, curr_table_end); |
| 1667 |
|
| 1668 |
//decode_table.copy_within(range, curr_table_end); |
| 1669 |
curr_table_end <<= 1; |
| 1670 |
} |
| 1671 |
count = len_counts[len]; |
| 1672 |
|
| 1673 |
if count != 0 |
| 1674 |
{ |
| 1675 |
break; |
| 1676 |
} |
| 1677 |
} |
| 1678 |
} |
| 1679 |
// process codewords with len > table_bits. |
| 1680 |
// Require sub-tables |
| 1681 |
curr_table_end = 1 << table_bits; |
| 1682 |
|
| 1683 |
let mut subtable_prefix = usize::MAX; |
| 1684 |
let mut subtable_start = 0; |
| 1685 |
let mut subtable_bits; |
| 1686 |
|
| 1687 |
loop |
| 1688 |
{ |
| 1689 |
/* |
| 1690 |
* Start a new sub-table if the first 'table_bits' bits of the |
| 1691 |
* codeword don't match the prefix of the current subtable. |
| 1692 |
*/ |
| 1693 |
if codeword & ((1_usize << table_bits) - 1) != subtable_prefix |
| 1694 |
{ |
| 1695 |
subtable_prefix = codeword & ((1 << table_bits) - 1); |
| 1696 |
subtable_start = curr_table_end; |
| 1697 |
|
| 1698 |
/* |
| 1699 |
* Calculate the subtable length. If the codeword has |
| 1700 |
* length 'table_bits + n', then the subtable needs |
| 1701 |
* '2^n' entries. But it may need more; if fewer than |
| 1702 |
* '2^n' codewords of length 'table_bits + n' remain, |
| 1703 |
* then the length will need to be incremented to bring |
| 1704 |
* in longer codewords until the subtable can be |
| 1705 |
* completely filled. Note that because the Huffman |
| 1706 |
* code is complete, it will always be possible to fill |
| 1707 |
* the sub-table eventually. |
| 1708 |
*/ |
| 1709 |
subtable_bits = len - table_bits; |
| 1710 |
codespace_used = count; |
| 1711 |
|
| 1712 |
while codespace_used < (1 << subtable_bits) |
| 1713 |
{ |
| 1714 |
subtable_bits += 1; |
| 1715 |
|
| 1716 |
if subtable_bits + table_bits > 15 |
| 1717 |
{ |
| 1718 |
return Err(DecodeErrorStatus::CorruptData); |
| 1719 |
} |
| 1720 |
|
| 1721 |
codespace_used = (codespace_used << 1) + len_counts[table_bits + subtable_bits]; |
| 1722 |
} |
| 1723 |
|
| 1724 |
/* |
| 1725 |
* Create the entry that points from the main table to |
| 1726 |
* the subtable. |
| 1727 |
*/ |
| 1728 |
decode_table[subtable_prefix] = (subtable_start as u32) << 16 |
| 1729 |
| HUFFDEC_EXCEPTIONAL |
| 1730 |
| HUFFDEC_SUITABLE_POINTER |
| 1731 |
| (subtable_bits as u32) << 8 |
| 1732 |
| table_bits as u32; |
| 1733 |
|
| 1734 |
curr_table_end = subtable_start + (1 << subtable_bits); |
| 1735 |
} |
| 1736 |
|
| 1737 |
/* Fill the sub-table entries for the current codeword. */ |
| 1738 |
|
| 1739 |
let stride = 1 << (len - table_bits); |
| 1740 |
|
| 1741 |
let mut j = subtable_start + (codeword >> table_bits); |
| 1742 |
|
| 1743 |
let entry = make_decode_table_entry( |
| 1744 |
decode_results, |
| 1745 |
sorted_syms[i] as usize, |
| 1746 |
(len - table_bits) as u32 |
| 1747 |
); |
| 1748 |
i += 1; |
| 1749 |
|
| 1750 |
while j < curr_table_end |
| 1751 |
{ |
| 1752 |
decode_table[j] = entry; |
| 1753 |
j += stride; |
| 1754 |
} |
| 1755 |
//advance to the next codeword |
| 1756 |
if codeword == (1 << len) - 1 |
| 1757 |
{ |
| 1758 |
// last codeword |
| 1759 |
return Ok(()); |
| 1760 |
} |
| 1761 |
|
| 1762 |
let adv = BITS - (codeword ^ ((1 << len) - 1)).leading_zeros(); |
| 1763 |
let bit = 1 << adv; |
| 1764 |
|
| 1765 |
codeword &= bit - 1; |
| 1766 |
codeword |= bit; |
| 1767 |
count -= 1; |
| 1768 |
|
| 1769 |
while count == 0 |
| 1770 |
{ |
| 1771 |
len += 1; |
| 1772 |
count = len_counts[len]; |
| 1773 |
} |
| 1774 |
} |
| 1775 |
} |
| 1776 |
} |
| 1777 |
|
| 1778 |
const RESIZE_BY: usize = 1024 * 4; // 4 kb |
| 1779 |
|
| 1780 |
/// Resize vector if its current space wont |
| 1781 |
/// be able to store a new byte and then push an element to that new space |
| 1782 |
#[inline(always)] |
| 1783 |
fn resize_and_push(buf: &mut Vec<u8>, position: usize, elm: u8) |
| 1784 |
{ |
| 1785 |
if buf.len() <= position |
| 1786 |
{ |
| 1787 |
let new_len = buf.len() + RESIZE_BY; |
| 1788 |
buf.resize(new_len, 0); |
| 1789 |
} |
| 1790 |
buf[position] = elm; |
| 1791 |
} |