liblzma: Add EROFS LZMA encoder and decoder.

Right now this is just a planned extra-compact format for use
in the EROFS file system in Linux. At this point it's possible
that the format will either change or be abandoned and removed
completely.

The special thing about the encoder is that it uses the
output-size-limited encoding added in the previous commit.
EROFS uses fixed-sized blocks (e.g. 4 KiB) to hold compressed
data so the compressors must be able to create valid streams
that fill the given block size.

5 files changed, 367 insertions, 0 deletions

diff --git a/src/liblzma/api/lzma/container.h b/src/liblzma/api/lzma/container.h
index 9fbf4df0..581f3507 100644
--- a/src/liblzma/api/lzma/container.h
+++ b/src/liblzma/api/lzma/container.h
@@ -444,6 +444,55 @@ extern LZMA_API(lzma_ret) lzma_stream_buffer_encode(
 		lzma_nothrow lzma_attr_warn_unused_result;
 
 
+/**
+ * \brief       EROFS LZMA encoder
+ *
+ * The EROFS LZMA format is a raw LZMA stream whose first byte (always 0x00)
+ * has been replaced with bitwise-negation of the LZMA properties (lc/lp/pb).
+ * This encoding ensures that the first byte of EROFS LZMA stream is never
+ * 0x00. There is no end of payload marker and thus the uncompressed size
+ * must be stored separately. For the best error detection the dictionary
+ * size should be stored separately as well but alternatively one may use
+ * the uncompressed size as the dictionary size when decoding.
+ *
+ * With the EROFS LZMA encoder, lzma_code() behaves slightly unusually.
+ * The action argument must be LZMA_FINISH and the return value cannot be
+ * LZMA_OK. Thus the encoding is always done with a single lzma_code() after
+ * the initialization. The benefit of the combination of initialization
+ * function and lzma_code() is that memory allocations can be re-used for
+ * better performance.
+ *
+ * lzma_code() will try to encode as much input as is possible to fit into
+ * the given output buffer. If not all input can be encoded, the stream will
+ * be finished without encoding all the input. The caller must check both
+ * input and output buffer usage after lzma_code() (total_in and total_out
+ * in lzma_stream can be convenient). Often lzma_code() can fill the output
+ * buffer completely if there is a lot of input, but sometimes a few bytes
+ * may remain unused because the next LZMA symbol would require more space.
+ *
+ * lzma_stream.avail_out must be at least 6. Otherwise LZMA_PROG_ERROR
+ * will be returned.
+ *
+ * The LZMA dictionary should be reasonably low to speed up the encoder
+ * re-initialization. A good value is bigger than the resulting
+ * uncompressed size of most of the output chunks. For example, if output
+ * size is 4 KiB, dictionary size of 32 KiB or 64 KiB is good. If the
+ * data compresses extremely well, even 128 KiB may be useful.
+ *
+ * \return      - LZMA_STREAM_END: All good. Check the amounts of input used
+ *                and output produced. Store the amount of input used
+ *                (uncompressed size) as it needs to be known to decompress
+ *                the data.
+ *              - LZMA_OPTIONS_ERROR
+ *              - LZMA_MEM_ERROR
+ *              - LZMA_PROG_ERROR: In addition to the generic reasons for this
+ *                error code, this may also be returned if there isn't enough
+ *                output space (6 bytes) to create a valid EROFS LZMA stream.
+ */
+extern LZMA_API(lzma_ret) lzma_erofs_encoder(
+		lzma_stream *strm, const lzma_options_lzma *options);
+
+
 /************
  * Decoding *
  ************/
@@ -630,3 +679,30 @@ extern LZMA_API(lzma_ret) lzma_stream_buffer_decode(
 		const uint8_t *in, size_t *in_pos, size_t in_size,
 		uint8_t *out, size_t *out_pos, size_t out_size)
 		lzma_nothrow lzma_attr_warn_unused_result;
+
+
+/**
+ * \brief       EROFS LZMA decoder
+ *
+ * See lzma_erofs_decoder() for more information.
+ *
+ * The lzma_code() usage with this decoder is completely normal.
+ * The special behavior of lzma_code() applies to lzma_erofs_encoder() only.
+ *
+ * \param       strm        Pointer to properly prepared lzma_stream
+ * \param       uncomp_size Uncompressed size of the EROFS LZMA stream.
+ *                          The caller must somehow know this. Note that
+ *                          while the EROFS LZMA decoder in XZ Embedded needs
+ *                          also the compressed size, the implementation in
+ *                          liblzma doesn't need to know the compressed size.
+ * \param       dict_size   LZMA dictionary size that was used when
+ *                          compressing the data. It is OK to use a bigger
+ *                          value too but liblzma will then allocate more
+ *                          memory than would actually be required and error
+ *                          detection will be slightly worse. (Note that with
+ *                          the implementation in XZ Embedded it doesn't
+ *                          affect the memory usage if one specifies bigger
+ *                          dictionary than actually required.)
+ */
+extern LZMA_API(lzma_ret) lzma_erofs_decoder(
+		lzma_stream *strm, uint64_t uncomp_size, uint32_t dict_size);
diff --git a/src/liblzma/common/Makefile.inc b/src/liblzma/common/Makefile.inc
index 0408f9a4..8205eb7f 100644
--- a/src/liblzma/common/Makefile.inc
+++ b/src/liblzma/common/Makefile.inc
@@ -36,6 +36,7 @@ liblzma_la_SOURCES += \
 	common/easy_buffer_encoder.c \
 	common/easy_encoder.c \
 	common/easy_encoder_memusage.c \
+	common/erofs_encoder.c \
 	common/filter_buffer_encoder.c \
 	common/filter_encoder.c \
 	common/filter_encoder.h \
@@ -65,6 +66,7 @@ liblzma_la_SOURCES += \
 	common/block_decoder.h \
 	common/block_header_decoder.c \
 	common/easy_decoder_memusage.c \
+	common/erofs_decoder.c \
 	common/file_info.c \
 	common/filter_buffer_decoder.c \
 	common/filter_decoder.c \
diff --git a/src/liblzma/common/erofs_decoder.c b/src/liblzma/common/erofs_decoder.c
new file mode 100644
index 00000000..ef584373
--- /dev/null
+++ b/src/liblzma/common/erofs_decoder.c
@@ -0,0 +1,148 @@
+///////////////////////////////////////////////////////////////////////////////
+//
+/// \file       erofs_decoder.c
+/// \brief      Decode EROFS LZMA format
+//
+//  Author:     Lasse Collin
+//
+//  This file has been put into the public domain.
+//  You can do whatever you want with this file.
+//
+///////////////////////////////////////////////////////////////////////////////
+
+#include "lzma_decoder.h"
+#include "lz_decoder.h"
+
+
+typedef struct {
+	/// LZMA1 decoder
+	lzma_next_coder lzma;
+
+	/// Uncompressed size of the stream as given by the application
+	lzma_vli uncomp_size;
+
+	/// LZMA dictionary size as given by the application
+	uint32_t dict_size;
+
+	/// True once the first byte of the EROFS LZMA stream
+	/// has been processed.
+	bool props_decoded;
+} lzma_erofs_coder;
+
+
+static lzma_ret
+erofs_decode(void *coder_ptr, const lzma_allocator *allocator,
+		const uint8_t *restrict in, size_t *restrict in_pos,
+		size_t in_size, uint8_t *restrict out,
+		size_t *restrict out_pos, size_t out_size, lzma_action action)
+{
+	lzma_erofs_coder *coder = coder_ptr;
+
+	if (!coder->props_decoded) {
+		// There must be at least one byte of input to decode
+		// the properties byte.
+		if (*in_pos >= in_size)
+			return LZMA_OK;
+
+		lzma_options_lzma options = {
+			.preset_dict = NULL,
+			.preset_dict_size = 0,
+		};
+
+		// The properties are stored as bitwise-negation
+		// of the typical encoding.
+		if (lzma_lzma_lclppb_decode(&options, ~in[*in_pos]))
+			return LZMA_OPTIONS_ERROR;
+
+		++*in_pos;
+
+		// Initialize the decoder.
+		options.dict_size = coder->dict_size;
+		lzma_filter_info filters[2] = {
+			{
+				.init = &lzma_lzma_decoder_init,
+				.options = &options,
+			}, {
+				.init = NULL,
+			}
+		};
+
+		return_if_error(lzma_next_filter_init(&coder->lzma,
+				allocator, filters));
+
+		// Use a hack to set the uncompressed size.
+		lzma_lz_decoder_uncompressed(coder->lzma.coder,
+				coder->uncomp_size);
+
+		// Pass one dummy 0x00 byte to the LZMA decoder since that
+		// is what it expects the first byte to be.
+		const uint8_t dummy_in = 0;
+		size_t dummy_in_pos = 0;
+		if (coder->lzma.code(coder->lzma.coder, allocator,
+				&dummy_in, &dummy_in_pos, 1,
+				out, out_pos, out_size, LZMA_RUN) != LZMA_OK)
+			return LZMA_PROG_ERROR;
+
+		assert(dummy_in_pos == 1);
+		coder->props_decoded = true;
+	}
+
+	// The rest is normal LZMA decoding.
+	return coder->lzma.code(coder->lzma.coder, allocator,
+				in, in_pos, in_size,
+				out, out_pos, out_size, action);
+}
+
+
+static void
+erofs_decoder_end(void *coder_ptr, const lzma_allocator *allocator)
+{
+	lzma_erofs_coder *coder = coder_ptr;
+	lzma_next_end(&coder->lzma, allocator);
+	lzma_free(coder, allocator);
+	return;
+}
+
+
+static lzma_ret
+erofs_decoder_init(lzma_next_coder *next, const lzma_allocator *allocator,
+		uint64_t uncomp_size, uint32_t dict_size)
+{
+	lzma_next_coder_init(&erofs_decoder_init, next, allocator);
+
+	lzma_erofs_coder *coder = next->coder;
+
+	if (coder == NULL) {
+		coder = lzma_alloc(sizeof(lzma_erofs_coder), allocator);
+		if (coder == NULL)
+			return LZMA_MEM_ERROR;
+
+		next->coder = coder;
+		next->code = &erofs_decode;
+		next->end = &erofs_decoder_end;
+
+		coder->lzma = LZMA_NEXT_CODER_INIT;
+	}
+
+	if (uncomp_size > LZMA_VLI_MAX)
+		return LZMA_OPTIONS_ERROR;
+
+	coder->uncomp_size = uncomp_size;
+	coder->dict_size = dict_size;
+
+	coder->props_decoded = false;
+
+	return LZMA_OK;
+}
+
+
+extern LZMA_API(lzma_ret)
+lzma_erofs_decoder(lzma_stream *strm, uint64_t uncomp_size, uint32_t dict_size)
+{
+	lzma_next_strm_init(erofs_decoder_init, strm, uncomp_size, dict_size);
+
+	strm->internal->supported_actions[LZMA_RUN] = true;
+	strm->internal->supported_actions[LZMA_FINISH] = true;
+
+	return LZMA_OK;
+}
diff --git a/src/liblzma/common/erofs_encoder.c b/src/liblzma/common/erofs_encoder.c
new file mode 100644
index 00000000..4cdd08f1
--- /dev/null
+++ b/src/liblzma/common/erofs_encoder.c
@@ -0,0 +1,139 @@
+///////////////////////////////////////////////////////////////////////////////
+//
+/// \file       erofs_encoder.c
+/// \brief      Encode into EROFS LZMA format
+//
+//  Author:     Lasse Collin
+//
+//  This file has been put into the public domain.
+//  You can do whatever you want with this file.
+//
+///////////////////////////////////////////////////////////////////////////////
+
+#include "lzma_encoder.h"
+
+
+typedef struct {
+	/// LZMA1 encoder
+	lzma_next_coder lzma;
+
+	/// LZMA properties byte (lc/lp/pb)
+	uint8_t props;
+} lzma_erofs_coder;
+
+
+static lzma_ret
+erofs_encode(void *coder_ptr, const lzma_allocator *allocator,
+		const uint8_t *restrict in, size_t *restrict in_pos,
+		size_t in_size, uint8_t *restrict out,
+		size_t *restrict out_pos, size_t out_size, lzma_action action)
+{
+	lzma_erofs_coder *coder = coder_ptr;
+
+	// Remember *out_pos so that we can overwrite the first byte with
+	// the LZMA properties byte.
+	const size_t out_start = *out_pos;
+
+	// Remember *in_pos so that we can set it based on how many
+	// uncompressed bytes were actually encoded.
+	const size_t in_start = *in_pos;
+
+	// Set the output size limit based on the available output space.
+	// We know that the encoder supports set_out_limit() so
+	// LZMA_OPTIONS_ERROR isn't possible. LZMA_BUF_ERROR is possible
+	// but lzma_code() has an assertion to not allow it to be returned
+	// from here and I don't want to change that for now, so
+	// LZMA_BUF_ERROR becomes LZMA_PROG_ERROR.
+	uint64_t uncomp_size;
+	if (coder->lzma.set_out_limit(coder->lzma.coder,
+			&uncomp_size, out_size - *out_pos) != LZMA_OK)
+		return LZMA_PROG_ERROR;
+
+	// set_out_limit fails if this isn't true.
+	assert(out_size - *out_pos >= 6);
+
+	// Encode as much as possible.
+	const lzma_ret ret = coder->lzma.code(coder->lzma.coder, allocator,
+			in, in_pos, in_size, out, out_pos, out_size, action);
+
+	if (ret != LZMA_STREAM_END) {
+		if (ret == LZMA_OK) {
+			assert(0);
+			return LZMA_PROG_ERROR;
+		}
+
+		return ret;
+	}
+
+	// The first output byte is bitwise-negation of the properties byte.
+	// We know that there is space for this byte because set_out_limit
+	// and the actual encoding succeeded.
+	out[out_start] = (uint8_t)(~coder->props);
+
+	// The LZMA encoder likely read more input than it was able to encode.
+	// Set *in_pos based on uncomp_size.
+	assert(uncomp_size <= in_size - in_start);
+	*in_pos = in_start + (size_t)(uncomp_size);
+
+	return ret;
+}
+
+
+static void
+erofs_encoder_end(void *coder_ptr, const lzma_allocator *allocator)
+{
+	lzma_erofs_coder *coder = coder_ptr;
+	lzma_next_end(&coder->lzma, allocator);
+	lzma_free(coder, allocator);
+	return;
+}
+
+
+static lzma_ret
+erofs_encoder_init(lzma_next_coder *next, const lzma_allocator *allocator,
+		const lzma_options_lzma *options)
+{
+	lzma_next_coder_init(&erofs_encoder_init, next, allocator);
+
+	lzma_erofs_coder *coder = next->coder;
+
+	if (coder == NULL) {
+		coder = lzma_alloc(sizeof(lzma_erofs_coder), allocator);
+		if (coder == NULL)
+			return LZMA_MEM_ERROR;
+
+		next->coder = coder;
+		next->code = &erofs_encode;
+		next->end = &erofs_encoder_end;
+
+		coder->lzma = LZMA_NEXT_CODER_INIT;
+	}
+
+	// Encode the properties byte. Bitwise-negation of it will be the
+	// first output byte.
+	return_if_error(lzma_lzma_lclppb_encode(options, &coder->props));
+
+	// Initialize the LZMA encoder.
+	const lzma_filter_info filters[2] = {
+		{
+			.init = &lzma_lzma_encoder_init,
+			.options = (void *)(options),
+		}, {
+			.init = NULL,
+		}
+	};
+
+	return lzma_next_filter_init(&coder->lzma, allocator, filters);
+}
+
+
+extern LZMA_API(lzma_ret)
+lzma_erofs_encoder(lzma_stream *strm, const lzma_options_lzma *options)
+{
+	lzma_next_strm_init(erofs_encoder_init, strm, options);
+
+	strm->internal->supported_actions[LZMA_FINISH] = true;
+
+	return LZMA_OK;
+
+}
diff --git a/src/liblzma/liblzma.map b/src/liblzma/liblzma.map
index bad8633c..251ef022 100644
--- a/src/liblzma/liblzma.map
+++ b/src/liblzma/liblzma.map
@@ -106,6 +106,8 @@ global:
 
 XZ_5.3.1alpha {
 global:
+	lzma_erofs_decoder;
+	lzma_erofs_encoder;
 	lzma_file_info_decoder;
 
 local: