lz4_flex/block/
decompress.rs

1//! The block decompression algorithm.
2use crate::block::{DecompressError, MINMATCH};
3use crate::fastcpy_unsafe;
4use crate::sink::SliceSink;
5use crate::sink::{PtrSink, Sink};
6#[allow(unused_imports)]
7use alloc::vec::Vec;
8
9/// Copies data to output_ptr by self-referential copy from start and match_length
10#[inline]
11unsafe fn duplicate(
12    output_ptr: &mut *mut u8,
13    output_end: *mut u8,
14    start: *const u8,
15    match_length: usize,
16) {
17    // We cannot simply use memcpy or `extend_from_slice`, because these do not allow
18    // self-referential copies: http://ticki.github.io/img/lz4_runs_encoding_diagram.svg
19
20    // Considering that `wild_copy_match_16` can copy up to `16 - 1` extra bytes.
21    // Defer to `duplicate_overlapping` in case of an overlapping match
22    // OR the if the wild copy would copy beyond the end of the output.
23    if (output_ptr.offset_from(start) as usize) < match_length + 16 - 1
24        || (output_end.offset_from(*output_ptr) as usize) < match_length + 16 - 1
25    {
26        duplicate_overlapping(output_ptr, start, match_length);
27    } else {
28        debug_assert!(
29            output_ptr.add(match_length / 16 * 16 + ((match_length % 16) != 0) as usize * 16)
30                <= output_end
31        );
32        wild_copy_from_src_16(start, *output_ptr, match_length);
33        *output_ptr = output_ptr.add(match_length);
34    }
35}
36
37#[inline]
38fn wild_copy_from_src_16(mut source: *const u8, mut dst_ptr: *mut u8, num_items: usize) {
39    // Note: if the compiler auto-vectorizes this it'll hurt performance!
40    // It's not the case for 16 bytes stepsize, but for 8 bytes.
41    unsafe {
42        let dst_ptr_end = dst_ptr.add(num_items);
43        loop {
44            core::ptr::copy_nonoverlapping(source, dst_ptr, 16);
45            source = source.add(16);
46            dst_ptr = dst_ptr.add(16);
47            if dst_ptr >= dst_ptr_end {
48                break;
49            }
50        }
51    }
52}
53
54/// Copy function, if the data start + match_length overlaps into output_ptr
55#[inline]
56#[cfg_attr(nightly, optimize(size))] // to avoid loop unrolling
57unsafe fn duplicate_overlapping(
58    output_ptr: &mut *mut u8,
59    mut start: *const u8,
60    match_length: usize,
61) {
62    // There is an edge case when output_ptr == start, which causes the decoder to potentially
63    // expose up to match_length bytes of uninitialized data in the decompression buffer.
64    // To prevent that we write a dummy zero to output, which will zero out output in such cases.
65    // This is the same strategy used by the reference C implementation https://github.com/lz4/lz4/pull/772
66    output_ptr.write(0u8);
67    let dst_ptr_end = output_ptr.add(match_length);
68
69    while output_ptr.add(1) < dst_ptr_end {
70        // Note that this loop unrolling is done, so that the compiler doesn't do it in a awful
71        // way.
72        // Without that the compiler will unroll/auto-vectorize the copy with a lot of branches.
73        // This is not what we want, as large overlapping copies are not that common.
74        core::ptr::copy(start, *output_ptr, 1);
75        start = start.add(1);
76        *output_ptr = output_ptr.add(1);
77
78        core::ptr::copy(start, *output_ptr, 1);
79        start = start.add(1);
80        *output_ptr = output_ptr.add(1);
81    }
82
83    if *output_ptr < dst_ptr_end {
84        core::ptr::copy(start, *output_ptr, 1);
85        *output_ptr = output_ptr.add(1);
86    }
87}
88
89#[inline]
90unsafe fn copy_from_dict(
91    output_base: *mut u8,
92    output_ptr: &mut *mut u8,
93    ext_dict: &[u8],
94    offset: usize,
95    match_length: usize,
96) -> usize {
97    // If we're here we know offset > output pos, so we have at least 1 byte to copy from dict
98    debug_assert!(output_ptr.offset_from(output_base) >= 0);
99    debug_assert!(offset > output_ptr.offset_from(output_base) as usize);
100    // If unchecked-decode is not disabled we also know that the offset falls within ext_dict
101    debug_assert!(ext_dict.len() + output_ptr.offset_from(output_base) as usize >= offset);
102
103    let dict_offset = ext_dict.len() + output_ptr.offset_from(output_base) as usize - offset;
104    // Can't copy past ext_dict len, the match may cross dict and output
105    let dict_match_length = match_length.min(ext_dict.len() - dict_offset);
106    // TODO test fastcpy_unsafe
107    core::ptr::copy_nonoverlapping(
108        ext_dict.as_ptr().add(dict_offset),
109        *output_ptr,
110        dict_match_length,
111    );
112    *output_ptr = output_ptr.add(dict_match_length);
113    dict_match_length
114}
115
116/// Read an integer.
117///
118/// In LZ4, we encode small integers in a way that we can have an arbitrary number of bytes. In
119/// particular, we add the bytes repeatedly until we hit a non-0xFF byte. When we do, we add
120/// this byte to our sum and terminate the loop.
121///
122/// # Example
123///
124/// ```notest
125///     255, 255, 255, 4, 2, 3, 4, 6, 7
126/// ```
127///
128/// is encoded to _255 + 255 + 255 + 4 = 769_. The bytes after the first 4 is ignored, because
129/// 4 is the first non-0xFF byte.
130#[inline]
131fn read_integer_ptr(
132    input_ptr: &mut *const u8,
133    _input_ptr_end: *const u8,
134) -> Result<u32, DecompressError> {
135    // We start at zero and count upwards.
136    let mut n: u32 = 0;
137    // If this byte takes value 255 (the maximum value it can take), another byte is read
138    // and added to the sum. This repeats until a byte lower than 255 is read.
139    loop {
140        // We add the next byte until we get a byte which we add to the counting variable.
141
142        #[cfg(not(feature = "unchecked-decode"))]
143        {
144            if *input_ptr >= _input_ptr_end {
145                return Err(DecompressError::ExpectedAnotherByte);
146            }
147        }
148        let extra = unsafe { input_ptr.read() };
149        *input_ptr = unsafe { input_ptr.add(1) };
150        n += extra as u32;
151
152        // We continue if we got 255, break otherwise.
153        if extra != 0xFF {
154            break;
155        }
156    }
157
158    // 255, 255, 255, 8
159    // 111, 111, 111, 101
160
161    Ok(n)
162}
163
164/// Read a little-endian 16-bit integer from the input stream.
165#[inline]
166fn read_u16_ptr(input_ptr: &mut *const u8) -> u16 {
167    let mut num: u16 = 0;
168    unsafe {
169        core::ptr::copy_nonoverlapping(*input_ptr, &mut num as *mut u16 as *mut u8, 2);
170        *input_ptr = input_ptr.add(2);
171    }
172
173    u16::from_le(num)
174}
175
176const FIT_TOKEN_MASK_LITERAL: u8 = 0b00001111;
177const FIT_TOKEN_MASK_MATCH: u8 = 0b11110000;
178
179#[test]
180fn check_token() {
181    assert!(!does_token_fit(15));
182    assert!(does_token_fit(14));
183    assert!(does_token_fit(114));
184    assert!(!does_token_fit(0b11110000));
185    assert!(does_token_fit(0b10110000));
186}
187
188/// The token consists of two parts, the literal length (upper 4 bits) and match_length (lower 4
189/// bits) if the literal length and match_length are both below 15, we don't need to read additional
190/// data, so the token does fit the metadata in a single u8.
191#[inline]
192fn does_token_fit(token: u8) -> bool {
193    !((token & FIT_TOKEN_MASK_LITERAL) == FIT_TOKEN_MASK_LITERAL
194        || (token & FIT_TOKEN_MASK_MATCH) == FIT_TOKEN_MASK_MATCH)
195}
196
197/// Decompress all bytes of `input` into `output`.
198///
199/// Returns the number of bytes written (decompressed) into `output`.
200#[inline]
201pub(crate) fn decompress_internal<const USE_DICT: bool, S: Sink>(
202    input: &[u8],
203    output: &mut S,
204    ext_dict: &[u8],
205) -> Result<usize, DecompressError> {
206    // Prevent segfault for empty input
207    if input.is_empty() {
208        return Err(DecompressError::ExpectedAnotherByte);
209    }
210
211    let ext_dict = if USE_DICT {
212        ext_dict
213    } else {
214        // ensure optimizer knows ext_dict length is 0 if !USE_DICT
215        debug_assert!(ext_dict.is_empty());
216        &[]
217    };
218    let output_base = unsafe { output.base_mut_ptr() };
219    let output_end = unsafe { output_base.add(output.capacity()) };
220    let output_start_pos_ptr = unsafe { output.base_mut_ptr().add(output.pos()) as *mut u8 };
221    let mut output_ptr = output_start_pos_ptr;
222
223    let mut input_ptr = input.as_ptr();
224    let input_ptr_end = unsafe { input.as_ptr().add(input.len()) };
225    let safe_distance_from_end =  (16 /* literal copy */ +  2 /* u16 match offset */ + 1 /* The next token to read (we can skip the check) */).min(input.len()) ;
226    let input_ptr_safe = unsafe { input_ptr_end.sub(safe_distance_from_end) };
227
228    let safe_output_ptr = unsafe {
229        let mut output_num_safe_bytes = output
230            .capacity()
231            .saturating_sub(16 /* literal copy */ + 18 /* match copy */);
232        if USE_DICT {
233            // In the dictionary case the output pointer is moved by the match length in the dictionary.
234            // This may be up to 17 bytes without exiting the loop. So we need to ensure that we have
235            // at least additional 17 bytes of space left in the output buffer in the fast loop.
236            output_num_safe_bytes = output_num_safe_bytes.saturating_sub(17);
237        };
238
239        output_base.add(output_num_safe_bytes)
240    };
241
242    // Exhaust the decoder by reading and decompressing all blocks until the remaining buffer is
243    // empty.
244    loop {
245        // Read the token. The token is the first byte in a block. It is divided into two 4-bit
246        // subtokens, the higher and the lower.
247        // This token contains to 4-bit "fields", a higher and a lower, representing the literals'
248        // length and the back reference's length, respectively.
249        let token = unsafe { input_ptr.read() };
250        input_ptr = unsafe { input_ptr.add(1) };
251
252        // Checking for hot-loop.
253        // In most cases the metadata does fit in a single 1byte token (statistically) and we are in
254        // a safe-distance to the end. This enables some optimized handling.
255        //
256        // Ideally we want to check for safe output pos like: output.pos() <= safe_output_pos; But
257        // that doesn't work when the safe_output_ptr is == output_ptr due to insufficient
258        // capacity. So we use `<` instead of `<=`, which covers that case.
259        if does_token_fit(token)
260            && (input_ptr as usize) <= input_ptr_safe as usize
261            && output_ptr < safe_output_ptr
262        {
263            let literal_length = (token >> 4) as usize;
264            let mut match_length = MINMATCH + (token & 0xF) as usize;
265
266            // output_ptr <= safe_output_ptr should guarantee we have enough space in output
267            debug_assert!(
268                unsafe { output_ptr.add(literal_length + match_length) } <= output_end,
269                "{literal_length} + {match_length} {} wont fit ",
270                literal_length + match_length
271            );
272
273            // Copy the literal
274            // The literal is at max 16 bytes, and the is_safe_distance check assures
275            // that we are far away enough from the end so we can safely copy 16 bytes
276            unsafe {
277                core::ptr::copy_nonoverlapping(input_ptr, output_ptr, 16);
278                input_ptr = input_ptr.add(literal_length);
279                output_ptr = output_ptr.add(literal_length);
280            }
281
282            // input_ptr <= input_ptr_safe should guarantee we have enough space in input
283            debug_assert!(input_ptr_end as usize - input_ptr as usize >= 2);
284            let offset = read_u16_ptr(&mut input_ptr) as usize;
285
286            let output_len = unsafe { output_ptr.offset_from(output_base) as usize };
287            let offset = offset.min(output_len + ext_dict.len());
288
289            // Check if part of the match is in the external dict
290            if USE_DICT && offset > output_len {
291                let copied = unsafe {
292                    copy_from_dict(output_base, &mut output_ptr, ext_dict, offset, match_length)
293                };
294                if copied == match_length {
295                    continue;
296                }
297                // match crosses ext_dict and output
298                match_length -= copied;
299            }
300
301            // Calculate the start of this duplicate segment. At this point offset was already
302            // checked to be in bounds and the external dictionary copy, if any, was
303            // already copied and subtracted from match_length.
304            let start_ptr = unsafe { output_ptr.sub(offset) };
305            debug_assert!(start_ptr >= output_base);
306            debug_assert!(start_ptr < output_end);
307            debug_assert!(unsafe { output_end.offset_from(start_ptr) as usize } >= match_length);
308
309            // In this branch we know that match_length is at most 18 (14 + MINMATCH).
310            // But the blocks can overlap, so make sure they are at least 18 bytes apart
311            // to enable an optimized copy of 18 bytes.
312            if offset >= match_length {
313                unsafe {
314                    // _copy_, not copy_non_overlaping, as it may overlap.
315                    // Compiles to the same assembly on x68_64.
316                    core::ptr::copy(start_ptr, output_ptr, 18);
317                    output_ptr = output_ptr.add(match_length);
318                }
319            } else {
320                unsafe {
321                    duplicate_overlapping(&mut output_ptr, start_ptr, match_length);
322                }
323            }
324
325            continue;
326        }
327
328        // Now, we read the literals section.
329        // Literal Section
330        // If the initial value is 15, it is indicated that another byte will be read and added to
331        // it
332        let mut literal_length = (token >> 4) as usize;
333        if literal_length != 0 {
334            if literal_length == 15 {
335                // The literal_length length took the maximal value, indicating that there is more
336                // than 15 literal_length bytes. We read the extra integer.
337                literal_length += read_integer_ptr(&mut input_ptr, input_ptr_end)? as usize;
338            }
339
340            #[cfg(not(feature = "unchecked-decode"))]
341            {
342                // Check if literal is out of bounds for the input, and if there is enough space on
343                // the output
344                if literal_length > input_ptr_end as usize - input_ptr as usize {
345                    return Err(DecompressError::LiteralOutOfBounds);
346                }
347                if literal_length > unsafe { output_end.offset_from(output_ptr) as usize } {
348                    return Err(DecompressError::OutputTooSmall {
349                        expected: unsafe { output_ptr.offset_from(output_base) as usize }
350                            + literal_length,
351                        actual: output.capacity(),
352                    });
353                }
354            }
355            unsafe {
356                fastcpy_unsafe::slice_copy(input_ptr, output_ptr, literal_length);
357                output_ptr = output_ptr.add(literal_length);
358                input_ptr = input_ptr.add(literal_length);
359            }
360        }
361
362        // If the input stream is emptied, we break out of the loop. This is only the case
363        // in the end of the stream, since the block is intact otherwise.
364        if input_ptr >= input_ptr_end {
365            break;
366        }
367
368        // Read duplicate section
369        #[cfg(not(feature = "unchecked-decode"))]
370        {
371            if (input_ptr_end as usize) - (input_ptr as usize) < 2 {
372                return Err(DecompressError::ExpectedAnotherByte);
373            }
374        }
375        let offset = read_u16_ptr(&mut input_ptr) as usize;
376        // Obtain the initial match length. The match length is the length of the duplicate segment
377        // which will later be copied from data previously decompressed into the output buffer. The
378        // initial length is derived from the second part of the token (the lower nibble), we read
379        // earlier. Since having a match length of less than 4 would mean negative compression
380        // ratio, we start at 4 (MINMATCH).
381
382        // The initial match length can maximally be 19 (MINMATCH + 15). As with the literal length,
383        // this indicates that there are more bytes to read.
384        let mut match_length = MINMATCH + (token & 0xF) as usize;
385        if match_length == MINMATCH + 15 {
386            // The match length took the maximal value, indicating that there is more bytes. We
387            // read the extra integer.
388            match_length += read_integer_ptr(&mut input_ptr, input_ptr_end)? as usize;
389        }
390
391        // We now copy from the already decompressed buffer. This allows us for storing duplicates
392        // by simply referencing the other location.
393        let output_len = unsafe { output_ptr.offset_from(output_base) as usize };
394
395        // We'll do a bounds check except unchecked-decode is enabled.
396        #[cfg(not(feature = "unchecked-decode"))]
397        {
398            if offset > output_len + ext_dict.len() {
399                return Err(DecompressError::OffsetOutOfBounds);
400            }
401            if match_length > unsafe { output_end.offset_from(output_ptr) as usize } {
402                return Err(DecompressError::OutputTooSmall {
403                    expected: output_len + match_length,
404                    actual: output.capacity(),
405                });
406            }
407        }
408
409        if USE_DICT && offset > output_len {
410            let copied = unsafe {
411                copy_from_dict(output_base, &mut output_ptr, ext_dict, offset, match_length)
412            };
413            if copied == match_length {
414                #[cfg(not(feature = "unchecked-decode"))]
415                {
416                    if input_ptr >= input_ptr_end {
417                        return Err(DecompressError::ExpectedAnotherByte);
418                    }
419                }
420
421                continue;
422            }
423            // match crosses ext_dict and output
424            match_length -= copied;
425        }
426
427        // Calculate the start of this duplicate segment. At this point offset was already checked
428        // to be in bounds and the external dictionary copy, if any, was already copied and
429        // subtracted from match_length.
430        let start_ptr = unsafe { output_ptr.sub(offset) };
431        debug_assert!(start_ptr >= output_base);
432        debug_assert!(start_ptr < output_end);
433        debug_assert!(unsafe { output_end.offset_from(start_ptr) as usize } >= match_length);
434        unsafe {
435            duplicate(&mut output_ptr, output_end, start_ptr, match_length);
436        }
437        #[cfg(not(feature = "unchecked-decode"))]
438        {
439            if input_ptr >= input_ptr_end {
440                return Err(DecompressError::ExpectedAnotherByte);
441            }
442        }
443    }
444    unsafe {
445        output.set_pos(output_ptr.offset_from(output_base) as usize);
446        Ok(output_ptr.offset_from(output_start_pos_ptr) as usize)
447    }
448}
449
450/// Decompress all bytes of `input` into `output`.
451/// `output` should be preallocated with a size of of the uncompressed data.
452#[inline]
453pub fn decompress_into(input: &[u8], output: &mut [u8]) -> Result<usize, DecompressError> {
454    decompress_internal::<false, _>(input, &mut SliceSink::new(output, 0), b"")
455}
456
457/// Decompress all bytes of `input` into `output`.
458///
459/// Returns the number of bytes written (decompressed) into `output`.
460#[inline]
461pub fn decompress_into_with_dict(
462    input: &[u8],
463    output: &mut [u8],
464    ext_dict: &[u8],
465) -> Result<usize, DecompressError> {
466    decompress_internal::<true, _>(input, &mut SliceSink::new(output, 0), ext_dict)
467}
468
469/// Decompress all bytes of `input` into a new vec.
470/// The passed parameter `min_uncompressed_size` needs to be equal or larger than the uncompressed size.
471///
472/// # Panics
473/// May panic if the parameter `min_uncompressed_size` is smaller than the
474/// uncompressed data.
475
476#[inline]
477pub fn decompress_with_dict(
478    input: &[u8],
479    min_uncompressed_size: usize,
480    ext_dict: &[u8],
481) -> Result<Vec<u8>, DecompressError> {
482    // Allocate a vector to contain the decompressed stream.
483    let mut vec = Vec::with_capacity(min_uncompressed_size);
484    let decomp_len =
485        decompress_internal::<true, _>(input, &mut PtrSink::from_vec(&mut vec, 0), ext_dict)?;
486    unsafe {
487        vec.set_len(decomp_len);
488    }
489    Ok(vec)
490}
491
492/// Decompress all bytes of `input` into a new vec. The first 4 bytes are the uncompressed size in
493/// little endian. Can be used in conjunction with `compress_prepend_size`
494#[inline]
495pub fn decompress_size_prepended(input: &[u8]) -> Result<Vec<u8>, DecompressError> {
496    let (uncompressed_size, input) = super::uncompressed_size(input)?;
497    decompress(input, uncompressed_size)
498}
499
500/// Decompress all bytes of `input` into a new vec.
501/// The passed parameter `min_uncompressed_size` needs to be equal or larger than the uncompressed size.
502///
503/// # Panics
504/// May panic if the parameter `min_uncompressed_size` is smaller than the
505/// uncompressed data.
506#[inline]
507pub fn decompress(input: &[u8], min_uncompressed_size: usize) -> Result<Vec<u8>, DecompressError> {
508    // Allocate a vector to contain the decompressed stream.
509    let mut vec = Vec::with_capacity(min_uncompressed_size);
510    let decomp_len =
511        decompress_internal::<true, _>(input, &mut PtrSink::from_vec(&mut vec, 0), b"")?;
512    unsafe {
513        vec.set_len(decomp_len);
514    }
515    Ok(vec)
516}
517
518/// Decompress all bytes of `input` into a new vec. The first 4 bytes are the uncompressed size in
519/// little endian. Can be used in conjunction with `compress_prepend_size_with_dict`
520#[inline]
521pub fn decompress_size_prepended_with_dict(
522    input: &[u8],
523    ext_dict: &[u8],
524) -> Result<Vec<u8>, DecompressError> {
525    let (uncompressed_size, input) = super::uncompressed_size(input)?;
526    decompress_with_dict(input, uncompressed_size, ext_dict)
527}
528
529#[cfg(test)]
530mod test {
531    use super::*;
532
533    #[test]
534    fn all_literal() {
535        assert_eq!(decompress(&[0x30, b'a', b'4', b'9'], 3).unwrap(), b"a49");
536    }
537
538    // this error test is only valid with checked-decode.
539    #[cfg(not(feature = "unchecked-decode"))]
540    #[test]
541    fn offset_oob() {
542        decompress(&[0x10, b'a', 2, 0], 4).unwrap_err();
543        decompress(&[0x40, b'a', 1, 0], 4).unwrap_err();
544    }
545}