xref: /freebsd/contrib/libarchive/libarchive/archive_read_support_format_rar5.c (revision 185becb1e1bd2657c156f78aeb52edac05ba5fb5)

/*-
* Copyright (c) 2018 Grzegorz Antoniak (http://antoniak.org)
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
*    notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
*    notice, this list of conditions and the following disclaimer in the
*    documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR(S) ``AS IS'' AND ANY EXPRESS OR
* IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
* OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
* IN NO EVENT SHALL THE AUTHOR(S) BE LIABLE FOR ANY DIRECT, INDIRECT,
* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
* THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/

#include "archive_platform.h"
#include "archive_endian.h"

#ifdef HAVE_ERRNO_H
#include <errno.h>
#endif
#include <time.h>
#ifdef HAVE_ZLIB_H
#include <zlib.h> /* crc32 */
#endif
#ifdef HAVE_LIMITS_H
#include <limits.h>
#endif

#include "archive.h"
#ifndef HAVE_ZLIB_H
#include "archive_crc32.h"
#endif

#include "archive_entry.h"
#include "archive_entry_locale.h"
#include "archive_integer.h"
#include "archive_ppmd7_private.h"
#include "archive_entry_private.h"
#include "archive_time_private.h"

#ifdef HAVE_BLAKE2_H
#include <blake2.h>
#else
#include "archive_blake2.h"
#endif

/*#define CHECK_CRC_ON_SOLID_SKIP*/
/*#define DONT_FAIL_ON_CRC_ERROR*/
/*#define DEBUG*/

#define rar5_min(a, b) (((a) > (b)) ? (b) : (a))
#define rar5_max(a, b) (((a) > (b)) ? (a) : (b))
#define rar5_countof(X) ((const ssize_t) (sizeof(X) / sizeof(*X)))

#if defined DEBUG
#define DEBUG_CODE if(1)
#define LOG(...) do { printf("rar5: " __VA_ARGS__); puts(""); } while(0)
#else
#define DEBUG_CODE if(0)
#endif

/* Real RAR5 magic number is:
 *
 * 0x52, 0x61, 0x72, 0x21, 0x1a, 0x07, 0x01, 0x00
 * "Rar!→•☺·\x00"
 *
 * Retrieved with `rar5_signature()` by XOR'ing it with 0xA1, because I don't
 * want to put this magic sequence in each binary that uses libarchive, so
 * applications that scan through the file for this marker won't trigger on
 * this "false" one.
 *
 * The array itself is decrypted in `rar5_init` function. */

static const unsigned char rar5_signature_xor[] = {
	243, 192, 211, 128, 187, 166, 160, 161
};
static const size_t g_unpack_window_size = 0x20000;

/* These could have been static const's, but they aren't, because of
 * Visual Studio. */
#define MAX_NAME_IN_CHARS 2048
#define MAX_NAME_IN_BYTES (4 * MAX_NAME_IN_CHARS)

struct file_header {
	ssize_t bytes_remaining;
	ssize_t unpacked_size;
	int64_t last_offset;         /* Used in sanity checks. */
	int64_t last_size;           /* Used in sanity checks. */

	uint8_t solid : 1;           /* Is this a solid stream? */
	uint8_t service : 1;         /* Is this file a service data? */
	uint8_t eof : 1;             /* Did we finish unpacking the file? */
	uint8_t dir : 1;             /* Is this file entry a directory? */

	/* Optional time fields. */
	int64_t e_mtime;
	int64_t e_ctime;
	int64_t e_atime;
	uint32_t e_mtime_ns;
	uint32_t e_ctime_ns;
	uint32_t e_atime_ns;

	/* Optional hash fields. */
	uint32_t stored_crc32;
	uint32_t calculated_crc32;
	uint8_t blake2sp[32];
	blake2sp_state b2state;
	char has_blake2;

	/* Optional redir fields */
	uint64_t redir_type;
	uint64_t redir_flags;

	ssize_t solid_window_size; /* Used in file format check. */
};

enum EXTRA {
	EX_CRYPT = 0x01,
	EX_HASH = 0x02,
	EX_HTIME = 0x03,
	EX_VERSION = 0x04,
	EX_REDIR = 0x05,
	EX_UOWNER = 0x06,
	EX_SUBDATA = 0x07
};

#define REDIR_SYMLINK_IS_DIR	1

enum REDIR_TYPE {
	REDIR_TYPE_NONE = 0,
	REDIR_TYPE_UNIXSYMLINK = 1,
	REDIR_TYPE_WINSYMLINK = 2,
	REDIR_TYPE_JUNCTION = 3,
	REDIR_TYPE_HARDLINK = 4,
	REDIR_TYPE_FILECOPY = 5,
};

#define	OWNER_USER_NAME		0x01
#define	OWNER_GROUP_NAME	0x02
#define	OWNER_USER_UID		0x04
#define	OWNER_GROUP_GID		0x08
#define	OWNER_MAXNAMELEN	256

enum FILTER_TYPE {
	FILTER_DELTA = 0,   /* Generic pattern. */
	FILTER_E8    = 1,   /* Intel x86 code. */
	FILTER_E8E9  = 2,   /* Intel x86 code. */
	FILTER_ARM   = 3,   /* ARM code. */
	FILTER_AUDIO = 4,   /* Audio filter, not used in RARv5. */
	FILTER_RGB   = 5,   /* Color palette, not used in RARv5. */
	FILTER_ITANIUM = 6, /* Intel's Itanium, not used in RARv5. */
	FILTER_PPM   = 7,   /* Predictive pattern matching, not used in
			       RARv5. */
	FILTER_NONE  = 8,
};

struct filter_info {
	int type;
	int channels;

	int64_t block_start;
	ssize_t block_length;
};

struct data_ready {
	char used;
	const uint8_t* buf;
	size_t size;
	int64_t offset;
};

struct cdeque {
	uint16_t beg_pos;
	uint16_t end_pos;
	uint16_t cap_mask;
	uint16_t size;
	size_t* arr;
};

struct decode_table {
	uint32_t size;
	int32_t decode_len[16];
	uint32_t decode_pos[16];
	uint32_t quick_bits;
	uint8_t quick_len[1 << 10];
	uint16_t quick_num[1 << 10];
	uint16_t decode_num[306];
};

struct comp_state {
	/* Flag used to specify if unpacker needs to reinitialize the
	   uncompression context. */
	uint8_t initialized : 1;

	/* Flag used when applying filters. */
	uint8_t all_filters_applied : 1;

	/* Flag used to skip file context reinitialization, used when unpacker
	   is skipping through different multivolume archives. */
	uint8_t switch_multivolume : 1;

	/* Flag used to specify if unpacker has processed the whole data block
	   or just a part of it. */
	uint8_t block_parsing_finished : 1;

	/* Flag used to indicate that a previous file using this buffer was
	   encrypted, meaning no data in the buffer can be trusted */
	uint8_t data_encrypted : 1;

	signed int notused : 3;

	int flags;                   /* Uncompression flags. */
	int method;                  /* Uncompression algorithm method. */
	int version;                 /* Uncompression algorithm version. */
	ssize_t window_size;         /* Size of window_buf. */
	uint8_t* window_buf;         /* Circular buffer used during
	                                decompression. */
	uint8_t* filtered_buf;       /* Buffer used when applying filters. */
	const uint8_t* block_buf;    /* Buffer used when merging blocks. */
	ssize_t window_mask;         /* Convenience field; window_size - 1. */
	int64_t write_ptr;           /* This amount of data has been unpacked
					in the window buffer. */
	int64_t last_write_ptr;      /* This amount of data has been stored in
	                                the output file. */
	int64_t last_unstore_ptr;    /* Counter of bytes extracted during
	                                unstoring. This is separate from
	                                last_write_ptr because of how SERVICE
	                                base blocks are handled during skipping
	                                in solid multiarchive archives. */
	int64_t solid_offset;        /* Additional offset inside the window
	                                buffer, used in unpacking solid
	                                archives. */
	ssize_t cur_block_size;      /* Size of current data block. */
	int last_len;                /* Flag used in lzss decompression. */

	/* Decode tables used during lzss uncompression. */

#define HUFF_BC 20
	struct decode_table bd;      /* huffman bit lengths */
#define HUFF_NC 306
	struct decode_table ld;      /* literals */
#define HUFF_DC 64
	struct decode_table dd;      /* distances */
#define HUFF_LDC 16
	struct decode_table ldd;     /* lower bits of distances */
#define HUFF_RC 44
	struct decode_table rd;      /* repeating distances */
#define HUFF_TABLE_SIZE (HUFF_NC + HUFF_DC + HUFF_RC + HUFF_LDC)

	/* Circular deque for storing filters. */
	struct cdeque filters;
	int64_t last_block_start;    /* Used for sanity checking. */
	ssize_t last_block_length;   /* Used for sanity checking. */

	/* Distance cache used during lzss uncompression. */
	int dist_cache[4];

	/* Data buffer stack. */
	struct data_ready dready[2];
};

/* Bit reader state. */
struct bit_reader {
	int8_t bit_addr;    /* Current bit pointer inside current byte. */
	int in_addr;        /* Current byte pointer. */
};

/* RARv5 block header structure. Use bf_* functions to get values from
 * block_flags_u8 field. I.e. bf_byte_count, etc. */
struct compressed_block_header {
	/* block_flags_u8 contain fields encoded in little-endian bitfield:
	 *
	 * - table present flag (shr 7, and 1),
	 * - last block flag    (shr 6, and 1),
	 * - byte_count         (shr 3, and 7),
	 * - bit_size           (shr 0, and 7).
	 */
	uint8_t block_flags_u8;
	uint8_t block_cksum;
};

/* RARv5 main header structure. */
struct main_header {
	/* Does the archive contain solid streams? */
	uint8_t solid : 1;

	/* If this a multi-file archive? */
	uint8_t volume : 1;
	uint8_t endarc : 1;
	uint8_t notused : 5;

	unsigned int vol_no;
};

struct generic_header {
	uint8_t split_after : 1;
	uint8_t split_before : 1;
	uint8_t padding : 6;
	int size;
	int last_header_id;
};

struct multivolume {
	unsigned int expected_vol_no;
	uint8_t* push_buf;
};

/* Main context structure. */
struct rar5 {
	int header_initialized;

	/* Set to 1 if current file is positioned AFTER the magic value
	 * of the archive file. This is used in header reading functions. */
	int skipped_magic;

	/* Set to not zero if we're in skip mode (either by calling
	 * rar5_data_skip function or when skipping over solid streams).
	 * Set to 0 when in * extraction mode. This is used during checksum
	 * calculation functions. */
	int skip_mode;

	/* Set to not zero if we're in block merging mode (i.e. when switching
	 * to another file in multivolume archive, last block from 1st archive
	 * needs to be merged with 1st block from 2nd archive). This flag
	 * guards against recursive use of the merging function, which doesn't
	 * support recursive calls. */
	int merge_mode;

	/* An offset to QuickOpen list. This is not supported by this unpacker,
	 * because we're focusing on streaming interface. QuickOpen is designed
	 * to make things quicker for non-stream interfaces, so it's not our
	 * use case. */
	uint64_t qlist_offset;

	/* An offset to additional Recovery data. This is not supported by this
	 * unpacker. Recovery data are additional Reed-Solomon codes that could
	 * be used to calculate bytes that are missing in archive or are
	 * corrupted. */
	uint64_t rr_offset;

	/* Various context variables grouped to different structures. */
	struct generic_header generic;
	struct main_header main;
	struct comp_state cstate;
	struct file_header file;
	struct bit_reader bits;
	struct multivolume vol;

	/* The header of currently processed RARv5 block. Used in main
	 * decompression logic loop. */
	struct compressed_block_header last_block_hdr;

	/*
	 * Custom field to denote that this archive contains encrypted entries
	 */
	int has_encrypted_entries;
	int headers_are_encrypted;
};

/* Forward function declarations. */

static void rar5_signature(char *buf);
static int verify_global_checksums(struct archive_read* a);
static int rar5_read_data_skip(struct archive_read *a);
static int push_data_ready(struct archive_read* a, struct rar5 *rar5,
	const uint8_t* buf, size_t size, int64_t offset);
static void clear_data_ready_stack(struct rar5 *rar5);
static void rar5_deinit(struct rar5 *rar5);

/* CDE_xxx = Circular Double Ended (Queue) return values. */
enum CDE_RETURN_VALUES {
	CDE_OK, CDE_ALLOC, CDE_PARAM, CDE_OUT_OF_BOUNDS,
};

/* Clears the contents of this circular deque. */
static void cdeque_clear(struct cdeque* d) {
	d->size = 0;
	d->beg_pos = 0;
	d->end_pos = 0;
}

/* Creates a new circular deque object. Capacity must be power of 2: 8, 16, 32,
 * 64, 256, etc. When the user will add another item above current capacity,
 * the circular deque will overwrite the oldest entry. */
static int cdeque_init(struct cdeque* d, int max_capacity_power_of_2) {
	if(d == NULL || max_capacity_power_of_2 == 0)
		return CDE_PARAM;

	d->cap_mask = max_capacity_power_of_2 - 1;
	d->arr = NULL;

	if((max_capacity_power_of_2 & d->cap_mask) != 0)
		return CDE_PARAM;

	cdeque_clear(d);
	d->arr = malloc(sizeof(void*) * max_capacity_power_of_2);

	return d->arr ? CDE_OK : CDE_ALLOC;
}

/* Return the current size (not capacity) of circular deque `d`. */
static size_t cdeque_size(struct cdeque* d) {
	return d->size;
}

/* Returns the first element of current circular deque. Note that this function
 * doesn't perform any bounds checking. If you need bounds checking, use
 * `cdeque_front()` function instead. */
static void cdeque_front_fast(struct cdeque* d, void** value) {
	*value = (void*) d->arr[d->beg_pos];
}

/* Returns the first element of current circular deque. This function
 * performs bounds checking. */
static int cdeque_front(struct cdeque* d, void** value) {
	if(d->size > 0) {
		cdeque_front_fast(d, value);
		return CDE_OK;
	} else
		return CDE_OUT_OF_BOUNDS;
}

/* Pushes a new element into the end of this circular deque object. */
static int cdeque_push_back(struct cdeque* d, void* item) {
	if(d == NULL)
		return CDE_PARAM;

	if(d->size == d->cap_mask + 1)
		return CDE_OUT_OF_BOUNDS;

	d->arr[d->end_pos] = (size_t) item;
	d->end_pos = (d->end_pos + 1) & d->cap_mask;
	d->size++;

	return CDE_OK;
}

/* Pops a front element of this circular deque object and returns its value.
 * This function doesn't perform any bounds checking. */
static void cdeque_pop_front_fast(struct cdeque* d, void** value) {
	*value = (void*) d->arr[d->beg_pos];
	d->beg_pos = (d->beg_pos + 1) & d->cap_mask;
	d->size--;
}

/* Pops a front element of this circular deque object and returns its value.
 * This function performs bounds checking. */
static int cdeque_pop_front(struct cdeque* d, void** value) {
	if(!d || !value)
		return CDE_PARAM;

	if(d->size == 0)
		return CDE_OUT_OF_BOUNDS;

	cdeque_pop_front_fast(d, value);
	return CDE_OK;
}

/* Convenience function to cast filter_info** to void **. */
static void** cdeque_filter_p(struct filter_info** f) {
	return (void**) (size_t) f;
}

/* Convenience function to cast filter_info* to void *. */
static void* cdeque_filter(struct filter_info* f) {
	return (void**) (size_t) f;
}

/* Destroys this circular deque object. Deallocates the memory of the
 * collection buffer, but doesn't deallocate the memory of any pointer passed
 * to this deque as a value. */
static void cdeque_free(struct cdeque* d) {
	if(!d)
		return;

	if(!d->arr)
		return;

	free(d->arr);

	d->arr = NULL;
	d->beg_pos = -1;
	d->end_pos = -1;
	d->cap_mask = 0;
}

static inline
uint8_t bf_bit_size(const struct compressed_block_header* hdr) {
	return hdr->block_flags_u8 & 7;
}

static inline
uint8_t bf_byte_count(const struct compressed_block_header* hdr) {
	return (hdr->block_flags_u8 >> 3) & 7;
}

static inline
uint8_t bf_is_table_present(const struct compressed_block_header* hdr) {
	return (hdr->block_flags_u8 >> 7) & 1;
}

static inline
uint8_t bf_is_last_block(const struct compressed_block_header* hdr) {
	return (hdr->block_flags_u8 >> 6) & 1;
}

/* Convenience functions used by filter implementations. */
static void circular_memcpy(uint8_t* dst, uint8_t* window, const ssize_t mask,
    int64_t start, int64_t end)
{
	if((start & mask) > (end & mask)) {
		ssize_t len1 = mask + 1 - (start & mask);
		ssize_t len2 = end & mask;

		memcpy(dst, &window[start & mask], len1);
		memcpy(dst + len1, window, len2);
	} else {
		memcpy(dst, &window[start & mask], (size_t) (end - start));
	}
}

static uint32_t read_filter_data(struct rar5 *rar5, uint32_t offset) {
	uint8_t linear_buf[4];
	circular_memcpy(linear_buf, rar5->cstate.window_buf,
	    rar5->cstate.window_mask, offset, offset + 4);
	return archive_le32dec(linear_buf);
}

static void write_filter_data(struct rar5 *rar5, uint32_t offset,
    uint32_t value)
{
	archive_le32enc(&rar5->cstate.filtered_buf[offset], value);
}

/* Allocates a new filter descriptor and adds it to the filter array. */
static struct filter_info* add_new_filter(struct rar5 *rar5) {
	struct filter_info* f = calloc(1, sizeof(*f));

	if(!f) {
		return NULL;
	}

	if (CDE_OK != cdeque_push_back(&rar5->cstate.filters, cdeque_filter(f))) {
		free(f);
		return NULL;
	}

	return f;
}

static int run_delta_filter(struct rar5 *rar5, struct filter_info* flt) {
	int i;
	ssize_t dest_pos, src_pos = 0;

	for(i = 0; i < flt->channels; i++) {
		uint8_t prev_byte = 0;
		for(dest_pos = i;
				dest_pos < flt->block_length;
				dest_pos += flt->channels)
		{
			uint8_t byte;

			byte = rar5->cstate.window_buf[
			    (rar5->cstate.solid_offset + flt->block_start +
			    src_pos) & rar5->cstate.window_mask];

			prev_byte -= byte;
			rar5->cstate.filtered_buf[dest_pos] = prev_byte;
			src_pos++;
		}
	}

	return ARCHIVE_OK;
}

static int run_e8e9_filter(struct rar5 *rar5, struct filter_info* flt,
		int extended)
{
	const uint32_t file_size = 0x1000000;
	ssize_t i;

	circular_memcpy(rar5->cstate.filtered_buf,
	    rar5->cstate.window_buf, rar5->cstate.window_mask,
	    rar5->cstate.solid_offset + flt->block_start,
	    rar5->cstate.solid_offset + flt->block_start + flt->block_length);

	for(i = 0; i < flt->block_length - 4;) {
		uint8_t b = rar5->cstate.window_buf[
		    (rar5->cstate.solid_offset + flt->block_start +
		    i++) & rar5->cstate.window_mask];

		/*
		 * 0xE8 = x86's call <relative_addr_uint32> (function call)
		 * 0xE9 = x86's jmp <relative_addr_uint32> (unconditional jump)
		 */
		if(b == 0xE8 || (extended && b == 0xE9)) {

			uint32_t addr;
			uint32_t offset = (i + flt->block_start) % file_size;

			addr = read_filter_data(rar5,
			    (uint32_t)(rar5->cstate.solid_offset +
			    flt->block_start + i) & rar5->cstate.window_mask);

			if(addr & 0x80000000) {
				if(((addr + offset) & 0x80000000) == 0) {
					write_filter_data(rar5, (uint32_t)i,
					    addr + file_size);
				}
			} else {
				if((addr - file_size) & 0x80000000) {
					uint32_t naddr = addr - offset;
					write_filter_data(rar5, (uint32_t)i,
					    naddr);
				}
			}

			i += 4;
		}
	}

	return ARCHIVE_OK;
}

static int run_arm_filter(struct rar5 *rar5, struct filter_info* flt) {
	ssize_t i = 0;
	uint32_t offset;

	circular_memcpy(rar5->cstate.filtered_buf,
	    rar5->cstate.window_buf, rar5->cstate.window_mask,
	    rar5->cstate.solid_offset + flt->block_start,
	    rar5->cstate.solid_offset + flt->block_start + flt->block_length);

	for(i = 0; i < flt->block_length - 3; i += 4) {
		uint8_t* b = &rar5->cstate.window_buf[
		    (rar5->cstate.solid_offset +
		    flt->block_start + i + 3) & rar5->cstate.window_mask];

		if(*b == 0xEB) {
			/* 0xEB = ARM's BL (branch + link) instruction. */
			offset = read_filter_data(rar5,
			    (rar5->cstate.solid_offset + flt->block_start + i) &
			     (uint32_t)rar5->cstate.window_mask) & 0x00ffffff;

			offset -= (uint32_t) ((i + flt->block_start) / 4);
			offset = (offset & 0x00ffffff) | 0xeb000000;
			write_filter_data(rar5, (uint32_t)i, offset);
		}
	}

	return ARCHIVE_OK;
}

static int run_filter(struct archive_read* a, struct filter_info* flt) {
	struct rar5 *rar5 = a->format->data;
	int ret;

	clear_data_ready_stack(rar5);
	free(rar5->cstate.filtered_buf);

	rar5->cstate.filtered_buf = malloc(flt->block_length);
	if(!rar5->cstate.filtered_buf) {
		archive_set_error(&a->archive, ENOMEM,
		    "Can't allocate memory for filter data");
		return ARCHIVE_FATAL;
	}

	switch(flt->type) {
		case FILTER_DELTA:
			ret = run_delta_filter(rar5, flt);
			break;

		case FILTER_E8:
			/* fallthrough */
		case FILTER_E8E9:
			ret = run_e8e9_filter(rar5, flt,
			    flt->type == FILTER_E8E9);
			break;

		case FILTER_ARM:
			ret = run_arm_filter(rar5, flt);
			break;

		default:
			archive_set_error(&a->archive,
			    ARCHIVE_ERRNO_FILE_FORMAT,
			    "Unsupported filter type: 0x%x",
			    (unsigned int)flt->type);
			return ARCHIVE_FAILED;
	}

	if(ret != ARCHIVE_OK) {
		/* Filter has failed. */
		return ret;
	}

	if(ARCHIVE_OK != push_data_ready(a, rar5, rar5->cstate.filtered_buf,
	    flt->block_length, rar5->cstate.last_write_ptr))
	{
		archive_set_error(&a->archive, ARCHIVE_ERRNO_PROGRAMMER,
		    "Stack overflow when submitting unpacked data");

		return ARCHIVE_FATAL;
	}

	rar5->cstate.last_write_ptr += flt->block_length;
	return ARCHIVE_OK;
}

/* The `push_data` function submits the selected data range to the user.
 * Next call of `use_data` will use the pointer, size and offset arguments
 * that are specified here. These arguments are pushed to the FIFO stack here,
 * and popped from the stack by the `use_data` function. */
static void push_data(struct archive_read* a, struct rar5 *rar5,
    const uint8_t* buf, int64_t idx_begin, int64_t idx_end)
{
	const ssize_t wmask = rar5->cstate.window_mask;
	const ssize_t solid_write_ptr = (rar5->cstate.solid_offset +
	    rar5->cstate.last_write_ptr) & wmask;

	idx_begin += rar5->cstate.solid_offset;
	idx_end += rar5->cstate.solid_offset;

	/* Check if our unpacked data is wrapped inside the window circular
	 * buffer.  If it's not wrapped, it can be copied out by using
	 * a single memcpy, but when it's wrapped, we need to copy the first
	 * part with one memcpy, and the second part with another memcpy. */

	if((idx_begin & wmask) > (idx_end & wmask)) {
		/* The data is wrapped (begin offset sis bigger than end
		 * offset). */
		const ssize_t frag1_size = rar5->cstate.window_size -
		    (idx_begin & wmask);
		const ssize_t frag2_size = idx_end & wmask;

		/* Copy the first part of the buffer first. */
		push_data_ready(a, rar5, buf + solid_write_ptr, frag1_size,
		    rar5->cstate.last_write_ptr);

		/* Copy the second part of the buffer. */
		push_data_ready(a, rar5, buf, frag2_size,
		    rar5->cstate.last_write_ptr + frag1_size);

		rar5->cstate.last_write_ptr += frag1_size + frag2_size;
	} else {
		/* Data is not wrapped, so we can just use one call to copy the
		 * data. */
		push_data_ready(a, rar5,
		    buf + solid_write_ptr, (idx_end - idx_begin) & wmask,
		    rar5->cstate.last_write_ptr);

		rar5->cstate.last_write_ptr += idx_end - idx_begin;
	}
}

/* Convenience function that submits the data to the user. It uses the
 * unpack window buffer as a source location. */
static void push_window_data(struct archive_read* a, struct rar5 *rar5,
    int64_t idx_begin, int64_t idx_end)
{
	push_data(a, rar5, rar5->cstate.window_buf, idx_begin, idx_end);
}

static int apply_filters(struct archive_read* a) {
	struct rar5 *rar5 = a->format->data;
	struct filter_info* flt;
	int ret;

	rar5->cstate.all_filters_applied = 0;

	/* Get the first filter that can be applied to our data. The data
	 * needs to be fully unpacked before the filter can be run. */
	if(CDE_OK == cdeque_front(&rar5->cstate.filters,
	    cdeque_filter_p(&flt))) {
		/* Check if our unpacked data fully covers this filter's
		 * range. */
		if(rar5->cstate.write_ptr > flt->block_start &&
		    rar5->cstate.write_ptr >= flt->block_start +
		    flt->block_length) {
			/* Check if we have some data pending to be written
			 * right before the filter's start offset. */
			if(rar5->cstate.last_write_ptr == flt->block_start) {
				/* Run the filter specified by descriptor
				 * `flt`. */
				ret = run_filter(a, flt);
				if(ret != ARCHIVE_OK) {
					/* Filter failure, return error. */
					return ret;
				}

				/* Filter descriptor won't be needed anymore
				 * after it's used, * so remove it from the
				 * filter list and free its memory. */
				(void) cdeque_pop_front(&rar5->cstate.filters,
				    cdeque_filter_p(&flt));

				free(flt);
			} else {
				/* We can't run filters yet, dump the memory
				 * right before the filter. */
				push_window_data(a, rar5,
				    rar5->cstate.last_write_ptr,
				    flt->block_start);
			}

			/* Return 'filter applied or not needed' state to the
			 * caller. */
			return ARCHIVE_RETRY;
		}
	}

	rar5->cstate.all_filters_applied = 1;
	return ARCHIVE_OK;
}

static void dist_cache_push(struct rar5 *rar5, int value) {
	int* q = rar5->cstate.dist_cache;

	q[3] = q[2];
	q[2] = q[1];
	q[1] = q[0];
	q[0] = value;
}

static int dist_cache_touch(struct rar5 *rar5, int idx) {
	int* q = rar5->cstate.dist_cache;
	int i, dist = q[idx];

	for(i = idx; i > 0; i--)
		q[i] = q[i - 1];

	q[0] = dist;
	return dist;
}

static void free_filters(struct rar5 *rar5) {
	struct cdeque* d = &rar5->cstate.filters;

	/* Free any remaining filters. All filters should be naturally
	 * consumed by the unpacking function, so remaining filters after
	 * unpacking normally mean that unpacking wasn't successful.
	 * But still of course we shouldn't leak memory in such case. */

	/* cdeque_size() is a fast operation, so we can use it as a loop
	 * expression. */
	while(cdeque_size(d) > 0) {
		struct filter_info* f = NULL;

		/* Pop_front will also decrease the collection's size. */
		if (CDE_OK == cdeque_pop_front(d, cdeque_filter_p(&f)))
			free(f);
	}

	cdeque_clear(d);

	/* Also clear out the variables needed for sanity checking. */
	rar5->cstate.last_block_start = 0;
	rar5->cstate.last_block_length = 0;
}

static void reset_file_context(struct rar5 *rar5) {
	memset(&rar5->file, 0, sizeof(rar5->file));
	blake2sp_init(&rar5->file.b2state, 32);

	if(rar5->main.solid) {
		rar5->cstate.solid_offset += rar5->cstate.write_ptr;
	} else {
		rar5->cstate.solid_offset = 0;
	}

	rar5->cstate.write_ptr = 0;
	rar5->cstate.last_write_ptr = 0;
	rar5->cstate.last_unstore_ptr = 0;

	rar5->file.redir_type = REDIR_TYPE_NONE;
	rar5->file.redir_flags = 0;

	free_filters(rar5);
}

static inline int get_archive_read(struct archive* a,
    struct archive_read** ar)
{
	*ar = (struct archive_read*) a;
	archive_check_magic(a, ARCHIVE_READ_MAGIC, ARCHIVE_STATE_NEW,
	    "archive_read_support_format_rar5");

	return ARCHIVE_OK;
}

static int read_ahead(struct archive_read* a, size_t how_many,
    const uint8_t** ptr)
{
	if(!ptr)
		return 0;

	*ptr = __archive_read_ahead(a, how_many, NULL);
	if(*ptr == NULL) {
		return 0;
	}

	return 1;
}

static int consume(struct archive_read* a, int64_t how_many) {
	int ret;

	ret = how_many == __archive_read_consume(a, how_many)
		? ARCHIVE_OK
		: ARCHIVE_FATAL;

	return ret;
}

/**
 * Read a RAR5 variable sized numeric value. This value will be stored in
 * `pvalue`. The `pvalue_len` argument points to a variable that will receive
 * the byte count that was consumed in order to decode the `pvalue` value, plus
 * one.
 *
 * pvalue_len is optional and can be NULL.
 *
 * NOTE: if `pvalue_len` is NOT NULL, the caller needs to manually consume
 * the number of bytes that `pvalue_len` value contains. If the `pvalue_len`
 * is NULL, this consuming operation is done automatically.
 *
 * Returns 1 if *pvalue was successfully read.
 * Returns 0 if there was an error. In this case, *pvalue contains an
 *           invalid value.
 */

static int read_var(struct archive_read* a, uint64_t* pvalue,
    uint64_t* pvalue_len)
{
	uint64_t multiplier;
	uint64_t result = 0;
	size_t i;
	const uint8_t* p;

	/* We will read maximum of 10 bytes. We don't have to handle the
	 * situation to read the RAR5 variable-sized value stored at the end of
	 * the file, because such situation will never happen. */
	if(!read_ahead(a, 10, &p))
		return 0;

	for(multiplier = 1, i = 0; i < 10; i++, multiplier *= 128) {
		uint64_t val;
		uint8_t b;

		b = p[i];

		/* Strip the MSB from the input byte and add the resulting
		 * number to the `result`. */
		if(archive_ckd_mul_u64(&val, b & 0x7F, multiplier) ||
		   archive_ckd_add_u64(&result, result, val)) {
			/* Integer overflow occurred. */
			return 0;
		}

		/* MSB set to 1 means we need to continue decoding process.
		 * MSB set to 0 means we're done.
		 *
		 * This conditional checks for the second case. */
		if((b & 0x80) == 0) {
			if(pvalue) {
				*pvalue = result;
			}

			/* If the caller has passed the `pvalue_len` pointer,
			 * store the number of consumed bytes in it and do NOT
			 * consume those bytes, since the caller has all the
			 * information it needs to perform */
			if(pvalue_len) {
				*pvalue_len = 1 + i;
			} else {
				/* If the caller did not provide the
				 * `pvalue_len` pointer, it will not have the
				 * possibility to advance the file pointer,
				 * because it will not know how many bytes it
				 * needs to consume. This is why we handle
				 * such situation here automatically. */
				if(ARCHIVE_OK != consume(a, 1 + i)) {
					return 0;
				}
			}

			/* End of decoding process, return success. */
			return 1;
		}
	}

	/* All continuation bits were set. This is an error. */
	return 0;
}

static int read_var_sized(struct archive_read* a, size_t* pvalue,
    size_t* pvalue_len)
{
	uint64_t v;
	uint64_t v_size = 0;

	const int ret = pvalue_len ? read_var(a, &v, &v_size)
				   : read_var(a, &v, NULL);

	if(ret == 1 && pvalue) {
		*pvalue = (size_t) v;
	}

	if(pvalue_len) {
		/* Possible data truncation should be safe. */
		*pvalue_len = (size_t) v_size;
	}

	return ret;
}

static int read_bits_32(struct archive_read* a, struct rar5 *rar5,
	const uint8_t* p, uint32_t* value)
{
	if(rar5->bits.in_addr >= rar5->cstate.cur_block_size) {
		archive_set_error(&a->archive,
			ARCHIVE_ERRNO_PROGRAMMER,
			"Premature end of stream during extraction of data (#1)");
		return ARCHIVE_FATAL;
	}

	uint32_t bits = archive_be32dec(p + rar5->bits.in_addr);
	bits <<= rar5->bits.bit_addr;
	bits |= p[rar5->bits.in_addr + 4] >> (8 - rar5->bits.bit_addr);
	*value = bits;
	return ARCHIVE_OK;
}

static int read_bits_16(struct archive_read* a, struct rar5 *rar5,
	const uint8_t* p, uint16_t* value)
{
	if(rar5->bits.in_addr >= rar5->cstate.cur_block_size) {
		archive_set_error(&a->archive,
			ARCHIVE_ERRNO_PROGRAMMER,
			"Premature end of stream during extraction of data (#2)");
		return ARCHIVE_FATAL;
	}

	uint32_t bits = archive_be24dec(p + (unsigned)rar5->bits.in_addr);
	bits >>= (8 - rar5->bits.bit_addr);
	*value = bits & 0xffff;
	return ARCHIVE_OK;
}

static void skip_bits(struct rar5 *rar5, int bits) {
	const int new_bits = rar5->bits.bit_addr + bits;
	rar5->bits.in_addr += new_bits >> 3;
	rar5->bits.bit_addr = new_bits & 7;
}

/* n = up to 16 */
static int read_consume_bits(struct archive_read* a, struct rar5 *rar5,
	const uint8_t* p, int n, int* value)
{
	uint16_t v;
	int ret, num;

	if(n == 0 || n > 16) {
		/* This is a programmer error and should never happen
		 * in runtime. */
		return ARCHIVE_FATAL;
	}

	ret = read_bits_16(a, rar5, p, &v);
	if(ret != ARCHIVE_OK)
		return ret;

	num = (int) v;
	num >>= 16 - n;

	skip_bits(rar5, n);

	if(value)
		*value = num;

	return ARCHIVE_OK;
}

static char read_u32(struct archive_read* a, uint32_t* pvalue) {
	const uint8_t* p;
	if(!read_ahead(a, 4, &p))
		return 0;

	*pvalue = archive_le32dec(p);
	return ARCHIVE_OK == consume(a, 4);
}

static char read_u64(struct archive_read* a, uint64_t* pvalue) {
	const uint8_t* p;
	if(!read_ahead(a, 8, &p))
		return 0;

	*pvalue = archive_le64dec(p);
	return ARCHIVE_OK == consume(a, 8);
}

static int bid_standard(struct archive_read* a) {
	const uint8_t* p;
	char signature[sizeof(rar5_signature_xor)];

	rar5_signature(signature);

	if(!read_ahead(a, sizeof(rar5_signature_xor), &p))
		return -1;

	if(!memcmp(signature, p, sizeof(rar5_signature_xor)))
		return 30;

	return -1;
}

static int bid_sfx(struct archive_read *a)
{
	const char *p;

	if ((p = __archive_read_ahead(a, 7, NULL)) == NULL)
		return -1;

	if ((p[0] == 'M' && p[1] == 'Z') || memcmp(p, "\x7F\x45LF", 4) == 0) {
		/* This is a PE file */
		char signature[sizeof(rar5_signature_xor)];
		ssize_t offset = 0x10000;
		ssize_t window = 4096;
		ssize_t bytes_avail;

		rar5_signature(signature);

		while (offset + window <= (1024 * 512)) {
			const char *buff = __archive_read_ahead(a, offset + window, &bytes_avail);
			if (buff == NULL) {
				/* Remaining bytes are less than window. */
				window >>= 1;
				if (window < 0x40)
					return 0;
				continue;
			}
			p = buff + offset;
			while (p + 8 < buff + bytes_avail) {
				if (memcmp(p, signature, sizeof(signature)) == 0)
					return 30;
				p += 0x10;
			}
			offset = p - buff;
		}
	}

	return 0;
}

static int rar5_bid(struct archive_read* a, int best_bid) {
	int my_bid;

	if(best_bid > 30)
		return -1;

	my_bid = bid_standard(a);
	if(my_bid > -1) {
		return my_bid;
	}
	my_bid = bid_sfx(a);
	if (my_bid > -1) {
		return my_bid;
	}

	return -1;
}

static int rar5_options(struct archive_read *a, const char *key,
    const char *val) {
	(void) a;
	(void) key;
	(void) val;

	/* No options supported in this version. Return the ARCHIVE_WARN code
	 * to signal the options supervisor that the unpacker didn't handle
	 * setting this option. */

	return ARCHIVE_WARN;
}

static void init_header(struct archive_read* a) {
	a->archive.archive_format = ARCHIVE_FORMAT_RAR_V5;
	a->archive.archive_format_name = "RAR5";
}

static void init_window_mask(struct rar5 *rar5) {
	if (rar5->cstate.window_size)
		rar5->cstate.window_mask = rar5->cstate.window_size - 1;
	else
		rar5->cstate.window_mask = 0;
}

enum HEADER_FLAGS {
	HFL_EXTRA_DATA = 0x0001,
	HFL_DATA = 0x0002,
	HFL_SKIP_IF_UNKNOWN = 0x0004,
	HFL_SPLIT_BEFORE = 0x0008,
	HFL_SPLIT_AFTER = 0x0010,
	HFL_CHILD = 0x0020,
	HFL_INHERITED = 0x0040
};

static int process_main_locator_extra_block(struct archive_read* a,
    struct rar5 *rar5)
{
	uint64_t locator_flags;

	enum LOCATOR_FLAGS {
		QLIST = 0x01, RECOVERY = 0x02,
	};

	if(!read_var(a, &locator_flags, NULL)) {
		return ARCHIVE_EOF;
	}

	if(locator_flags & QLIST) {
		if(!read_var(a, &rar5->qlist_offset, NULL)) {
			return ARCHIVE_EOF;
		}

		/* qlist is not used */
	}

	if(locator_flags & RECOVERY) {
		if(!read_var(a, &rar5->rr_offset, NULL)) {
			return ARCHIVE_EOF;
		}

		/* rr is not used */
	}

	return ARCHIVE_OK;
}

static int parse_file_extra_hash(struct archive_read* a, struct rar5 *rar5,
    int64_t* extra_data_size)
{
	size_t hash_type = 0;
	size_t value_len;

	enum HASH_TYPE {
		BLAKE2sp = 0x00
	};

	if(!read_var_sized(a, &hash_type, &value_len))
		return ARCHIVE_EOF;

	*extra_data_size -= value_len;
	if(ARCHIVE_OK != consume(a, value_len)) {
		return ARCHIVE_EOF;
	}

	/* The file uses BLAKE2sp checksum algorithm instead of plain old
	 * CRC32. */
	if(hash_type == BLAKE2sp) {
		const uint8_t* p;
		const int hash_size = sizeof(rar5->file.blake2sp);

		if(!read_ahead(a, hash_size, &p))
			return ARCHIVE_EOF;

		rar5->file.has_blake2 = 1;
		memcpy(&rar5->file.blake2sp, p, hash_size);

		if(ARCHIVE_OK != consume(a, hash_size)) {
			return ARCHIVE_EOF;
		}

		*extra_data_size -= hash_size;
	} else {
		archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
		    "Unsupported hash type (0x%jx)", (uintmax_t)hash_type);
		return ARCHIVE_FATAL;
	}

	return ARCHIVE_OK;
}

static int parse_htime_item(struct archive_read* a, char unix_time,
    int64_t* sec, uint32_t* nsec, int64_t* extra_data_size)
{
	if(unix_time) {
		uint32_t time_val;
		if(!read_u32(a, &time_val))
			return ARCHIVE_EOF;

		*extra_data_size -= 4;
		*sec = (int64_t) time_val;
	} else {
		uint64_t windows_time;
		if(!read_u64(a, &windows_time))
			return ARCHIVE_EOF;

		ntfs_to_unix(windows_time, sec, nsec);
		*extra_data_size -= 8;
	}

	return ARCHIVE_OK;
}

static int parse_file_extra_version(struct archive_read* a,
    struct archive_entry* e, int64_t* extra_data_size)
{
	size_t flags = 0;
	size_t version = 0;
	size_t value_len = 0;
	struct archive_string version_string;
	struct archive_string name_utf8_string;
	const char* cur_filename;

	/* Flags are ignored. */
	if(!read_var_sized(a, &flags, &value_len))
		return ARCHIVE_EOF;

	*extra_data_size -= value_len;
	if(ARCHIVE_OK != consume(a, value_len))
		return ARCHIVE_EOF;

	if(!read_var_sized(a, &version, &value_len))
		return ARCHIVE_EOF;

	*extra_data_size -= value_len;
	if(ARCHIVE_OK != consume(a, value_len))
		return ARCHIVE_EOF;

	/* extra_data_size should be zero here. */

	cur_filename = archive_entry_pathname_utf8(e);
	if(cur_filename == NULL) {
		archive_set_error(&a->archive, ARCHIVE_ERRNO_PROGRAMMER,
		    "Version entry without file name");
		return ARCHIVE_FATAL;
	}

	archive_string_init(&version_string);
	archive_string_init(&name_utf8_string);

	/* Prepare a ;123 suffix for the filename, where '123' is the version
	 * value of this file. */
	archive_string_sprintf(&version_string, ";%zu", version);

	/* Build the new filename. */
	archive_strcat(&name_utf8_string, cur_filename);
	archive_strcat(&name_utf8_string, version_string.s);

	/* Apply the new filename into this file's context. */
	archive_entry_update_pathname_utf8(e, name_utf8_string.s);

	/* Free buffers. */
	archive_string_free(&version_string);
	archive_string_free(&name_utf8_string);
	return ARCHIVE_OK;
}

static int parse_file_extra_htime(struct archive_read* a,
    struct archive_entry* e, struct rar5 *rar5, int64_t* extra_data_size)
{
	char unix_time, has_unix_ns, has_mtime, has_ctime, has_atime;
	size_t flags = 0;
	size_t value_len;

	enum HTIME_FLAGS {
		IS_UNIX       = 0x01,
		HAS_MTIME     = 0x02,
		HAS_CTIME     = 0x04,
		HAS_ATIME     = 0x08,
		HAS_UNIX_NS   = 0x10,
	};

	if(!read_var_sized(a, &flags, &value_len))
		return ARCHIVE_EOF;

	*extra_data_size -= value_len;
	if(ARCHIVE_OK != consume(a, value_len)) {
		return ARCHIVE_EOF;
	}

	unix_time = flags & IS_UNIX;
	has_unix_ns = unix_time && (flags & HAS_UNIX_NS);
	has_mtime = flags & HAS_MTIME;
	has_atime = flags & HAS_ATIME;
	has_ctime = flags & HAS_CTIME;
	rar5->file.e_atime_ns = rar5->file.e_ctime_ns = rar5->file.e_mtime_ns = 0;

	if(has_mtime) {
		parse_htime_item(a, unix_time, &rar5->file.e_mtime,
		    &rar5->file.e_mtime_ns, extra_data_size);
	}

	if(has_ctime) {
		parse_htime_item(a, unix_time, &rar5->file.e_ctime,
		    &rar5->file.e_ctime_ns, extra_data_size);
	}

	if(has_atime) {
		parse_htime_item(a, unix_time, &rar5->file.e_atime,
		    &rar5->file.e_atime_ns, extra_data_size);
	}

	if(has_mtime && has_unix_ns) {
		if(!read_u32(a, &rar5->file.e_mtime_ns))
			return ARCHIVE_EOF;

		*extra_data_size -= 4;
	}

	if(has_ctime && has_unix_ns) {
		if(!read_u32(a, &rar5->file.e_ctime_ns))
			return ARCHIVE_EOF;

		*extra_data_size -= 4;
	}

	if(has_atime && has_unix_ns) {
		if(!read_u32(a, &rar5->file.e_atime_ns))
			return ARCHIVE_EOF;

		*extra_data_size -= 4;
	}

	/* The seconds and nanoseconds are either together, or separated in two
	 * fields so we parse them, then set the archive_entry's times. */
	if(has_mtime) {
		archive_entry_set_mtime(e, rar5->file.e_mtime, rar5->file.e_mtime_ns);
	}

	if(has_ctime) {
		archive_entry_set_ctime(e, rar5->file.e_ctime, rar5->file.e_ctime_ns);
	}

	if(has_atime) {
		archive_entry_set_atime(e, rar5->file.e_atime, rar5->file.e_atime_ns);
	}

	return ARCHIVE_OK;
}

static int parse_file_extra_redir(struct archive_read* a,
    struct archive_entry* e, struct rar5 *rar5, int64_t* extra_data_size)
{
	uint64_t value_size = 0;
	size_t target_size = 0;
	char target_utf8_buf[MAX_NAME_IN_BYTES];
	const uint8_t* p;

	if(!read_var(a, &rar5->file.redir_type, &value_size))
		return ARCHIVE_EOF;
	if(ARCHIVE_OK != consume(a, (int64_t)value_size))
		return ARCHIVE_EOF;
	*extra_data_size -= value_size;

	if(!read_var(a, &rar5->file.redir_flags, &value_size))
		return ARCHIVE_EOF;
	if(ARCHIVE_OK != consume(a, (int64_t)value_size))
		return ARCHIVE_EOF;
	*extra_data_size -= value_size;

	if(!read_var_sized(a, &target_size, NULL))
		return ARCHIVE_EOF;
	*extra_data_size -= target_size + 1;

	if(target_size > (MAX_NAME_IN_CHARS - 1)) {
		archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
		    "Link target is too long");
		return ARCHIVE_FATAL;
	}

	if(target_size == 0) {
		archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
		    "No link target specified");
		return ARCHIVE_FATAL;
	}

	if(!read_ahead(a, target_size, &p))
		return ARCHIVE_EOF;

	memcpy(target_utf8_buf, p, target_size);
	target_utf8_buf[target_size] = 0;

	if(ARCHIVE_OK != consume(a, (int64_t)target_size))
		return ARCHIVE_EOF;

	switch(rar5->file.redir_type) {
		case REDIR_TYPE_UNIXSYMLINK:
		case REDIR_TYPE_WINSYMLINK:
			archive_entry_set_filetype(e, AE_IFLNK);
			archive_entry_update_symlink_utf8(e, target_utf8_buf);
			if (rar5->file.redir_flags & REDIR_SYMLINK_IS_DIR) {
				archive_entry_set_symlink_type(e,
					AE_SYMLINK_TYPE_DIRECTORY);
			} else {
				archive_entry_set_symlink_type(e,
				AE_SYMLINK_TYPE_FILE);
			}
			break;

		case REDIR_TYPE_HARDLINK:
			archive_entry_set_filetype(e, AE_IFREG);
			archive_entry_update_hardlink_utf8(e, target_utf8_buf);
			break;

		default:
			/* Unknown redir type, skip it. */
			break;
	}
	return ARCHIVE_OK;
}

static int parse_file_extra_owner(struct archive_read* a,
    struct archive_entry* e, int64_t* extra_data_size)
{
	uint64_t flags = 0;
	uint64_t value_size = 0;
	uint64_t id = 0;
	size_t name_len = 0;
	size_t name_size = 0;
	char namebuf[OWNER_MAXNAMELEN];
	const uint8_t* p;

	if(!read_var(a, &flags, &value_size))
		return ARCHIVE_EOF;
	if(ARCHIVE_OK != consume(a, (int64_t)value_size))
		return ARCHIVE_EOF;
	*extra_data_size -= value_size;

	if ((flags & OWNER_USER_NAME) != 0) {
		if(!read_var_sized(a, &name_size, NULL))
			return ARCHIVE_EOF;
		*extra_data_size -= name_size + 1;

		if(!read_ahead(a, name_size, &p))
			return ARCHIVE_EOF;

		if (name_size >= OWNER_MAXNAMELEN) {
			name_len = OWNER_MAXNAMELEN - 1;
		} else {
			name_len = name_size;
		}

		memcpy(namebuf, p, name_len);
		namebuf[name_len] = 0;
		if(ARCHIVE_OK != consume(a, (int64_t)name_size))
			return ARCHIVE_EOF;

		archive_entry_set_uname(e, namebuf);
	}
	if ((flags & OWNER_GROUP_NAME) != 0) {
		if(!read_var_sized(a, &name_size, NULL))
			return ARCHIVE_EOF;
		*extra_data_size -= name_size + 1;

		if(!read_ahead(a, name_size, &p))
			return ARCHIVE_EOF;

		if (name_size >= OWNER_MAXNAMELEN) {
			name_len = OWNER_MAXNAMELEN - 1;
		} else {
			name_len = name_size;
		}

		memcpy(namebuf, p, name_len);
		namebuf[name_len] = 0;
		if(ARCHIVE_OK != consume(a, (int64_t)name_size))
			return ARCHIVE_EOF;

		archive_entry_set_gname(e, namebuf);
	}
	if ((flags & OWNER_USER_UID) != 0) {
		if(!read_var(a, &id, &value_size))
			return ARCHIVE_EOF;
		if(ARCHIVE_OK != consume(a, (int64_t)value_size))
			return ARCHIVE_EOF;
		*extra_data_size -= value_size;

		archive_entry_set_uid(e, (la_int64_t)id);
	}
	if ((flags & OWNER_GROUP_GID) != 0) {
		if(!read_var(a, &id, &value_size))
			return ARCHIVE_EOF;
		if(ARCHIVE_OK != consume(a, (int64_t)value_size))
			return ARCHIVE_EOF;
		*extra_data_size -= value_size;

		archive_entry_set_gid(e, (la_int64_t)id);
	}
	return ARCHIVE_OK;
}

static int process_head_file_extra(struct archive_read* a,
    struct archive_entry* e, struct rar5 *rar5, int64_t extra_data_size)
{
	uint64_t extra_field_size;
	uint64_t extra_field_id = 0;
	uint64_t var_size;

	while(extra_data_size > 0) {
		/* Make sure we won't fail if the file declares only unsupported
		attributes. */
		int ret = ARCHIVE_OK;

		if(!read_var(a, &extra_field_size, &var_size))
			return ARCHIVE_EOF;

		extra_data_size -= var_size;
		if(ARCHIVE_OK != consume(a, var_size)) {
			return ARCHIVE_EOF;
		}

		if(!read_var(a, &extra_field_id, &var_size))
			return ARCHIVE_EOF;

		extra_field_size -= var_size;
		extra_data_size -= var_size;
		if(ARCHIVE_OK != consume(a, var_size)) {
			return ARCHIVE_EOF;
		}

		switch(extra_field_id) {
			case EX_HASH:
				ret = parse_file_extra_hash(a, rar5,
				    &extra_data_size);
				break;
			case EX_HTIME:
				ret = parse_file_extra_htime(a, e, rar5,
				    &extra_data_size);
				break;
			case EX_REDIR:
				ret = parse_file_extra_redir(a, e, rar5,
				    &extra_data_size);
				break;
			case EX_UOWNER:
				ret = parse_file_extra_owner(a, e,
				    &extra_data_size);
				break;
			case EX_VERSION:
				ret = parse_file_extra_version(a, e,
				    &extra_data_size);
				break;
			case EX_CRYPT:
				/* Mark the entry as encrypted */
				archive_entry_set_is_data_encrypted(e, 1);
				rar5->has_encrypted_entries = 1;
				rar5->cstate.data_encrypted = 1;
				/* fallthrough */
			case EX_SUBDATA:
				/* fallthrough */
			default:
				/* Skip unsupported entry. */
				extra_data_size -= extra_field_size;
				if (ARCHIVE_OK != consume(a, extra_field_size)) {
					return ARCHIVE_EOF;
				}

				/* Don't fail on unsupported attribute -- we've handled it
				   by skipping over it. */
				ret = ARCHIVE_OK;
		}

		if (ret != ARCHIVE_OK) {
			/* Forward any errors signalled by the attribute parsing
			   functions. */
			return ret;
		}
	}

	if (extra_data_size != 0) {
		/* We didn't skip everything, or we skipped too much; either way,
		   there's an error in this parsing function. */

		archive_set_error(&a->archive, ARCHIVE_ERRNO_PROGRAMMER,
				"unsupported structure of file header extra data");
		return ARCHIVE_FATAL;
	}

	return ARCHIVE_OK;
}

static int file_entry_sanity_checks(struct archive_read* a,
	size_t block_flags, uint8_t is_dir, uint64_t unpacked_size,
	size_t packed_size)
{
	if (is_dir) {
		const int declares_data_size =
			(int) (unpacked_size != 0 || packed_size != 0);

		/* FILE entries for directories still declare HFL_DATA in block flags,
		   even though attaching data to such blocks doesn't make much sense. */
		if (declares_data_size) {
			archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
				"directory entries cannot have any data");
			return ARCHIVE_FATAL;
		}
	} else {
		const int declares_hfl_data = (int) ((block_flags & HFL_DATA) != 0);
		if (!declares_hfl_data) {
			archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
					"no data found in file/service block");
			return ARCHIVE_FATAL;
		}
	}

	return ARCHIVE_OK;
}

static int process_head_file(struct archive_read* a, struct rar5 *rar5,
    struct archive_entry* entry, size_t block_flags)
{
	int64_t extra_data_size = 0;
	size_t data_size = 0;
	size_t file_flags = 0;
	size_t file_attr = 0;
	size_t compression_info = 0;
	size_t host_os = 0;
	size_t name_size = 0;
	uint64_t unpacked_size, window_size;
	uint32_t mtime = 0, crc = 0;
	int c_method = 0, c_version = 0;
	char name_utf8_buf[MAX_NAME_IN_BYTES];
	const uint8_t* p;
	int sanity_ret;

	enum FILE_FLAGS {
		DIRECTORY = 0x0001, UTIME = 0x0002, CRC32 = 0x0004,
		UNKNOWN_UNPACKED_SIZE = 0x0008,
	};

	enum FILE_ATTRS {
		ATTR_READONLY = 0x1, ATTR_HIDDEN = 0x2, ATTR_SYSTEM = 0x4,
		ATTR_DIRECTORY = 0x10,
	};

	enum COMP_INFO_FLAGS {
		SOLID = 0x0040,
	};

	enum HOST_OS {
		HOST_WINDOWS = 0,
		HOST_UNIX = 1,
	};

	archive_entry_clear(entry);

	/* Do not reset file context if we're switching archives. */
	if(!rar5->cstate.switch_multivolume) {
		reset_file_context(rar5);
	}

	if(block_flags & HFL_EXTRA_DATA) {
		uint64_t edata_size = 0;
		if(!read_var(a, &edata_size, NULL))
			return ARCHIVE_EOF;

		/* Intentional type cast from unsigned to signed. */
		extra_data_size = (int64_t) edata_size;
	}

	if(block_flags & HFL_DATA) {
		if(!read_var_sized(a, &data_size, NULL))
			return ARCHIVE_EOF;

		if(data_size > SSIZE_MAX) {
			archive_set_error(&a->archive,
			    ARCHIVE_ERRNO_FILE_FORMAT,
			    "File data size is too large");
			return ARCHIVE_FATAL;
		}

		rar5->file.bytes_remaining = data_size;
	} else {
		rar5->file.bytes_remaining = 0;
	}

	if(!read_var_sized(a, &file_flags, NULL))
		return ARCHIVE_EOF;

	if(!read_var(a, &unpacked_size, NULL))
		return ARCHIVE_EOF;

	if(file_flags & UNKNOWN_UNPACKED_SIZE) {
		archive_set_error(&a->archive, ARCHIVE_ERRNO_PROGRAMMER,
		    "Files with unknown unpacked size are not supported");
		return ARCHIVE_FATAL;
	}

	rar5->file.dir = (uint8_t) ((file_flags & DIRECTORY) > 0);

	sanity_ret = file_entry_sanity_checks(a, block_flags, rar5->file.dir,
		unpacked_size, data_size);

	if (sanity_ret != ARCHIVE_OK) {
		return sanity_ret;
	}

	if(!read_var_sized(a, &file_attr, NULL))
		return ARCHIVE_EOF;

	if(file_flags & UTIME) {
		if(!read_u32(a, &mtime))
			return ARCHIVE_EOF;
	}

	if(file_flags & CRC32) {
		if(!read_u32(a, &crc))
			return ARCHIVE_EOF;
	}

	if(!read_var_sized(a, &compression_info, NULL))
		return ARCHIVE_EOF;

	c_method = (int) (compression_info >> 7) & 0x7;
	c_version = (int) (compression_info & 0x3f);

	/* RAR5 seems to limit the dictionary size to 64MB. */
	window_size = (rar5->file.dir > 0) ?
		0 :
		g_unpack_window_size << ((compression_info >> 10) & 15);
	rar5->cstate.method = c_method;
	rar5->cstate.version = c_version + 50;
	rar5->file.solid = (compression_info & SOLID) > 0;

	/* Archives which declare solid files without initializing the window
	 * buffer first are invalid, unless previous data was encrypted, in
	 * which case we may never have had the chance */

	if(rar5->file.solid > 0 && rar5->cstate.data_encrypted == 0 &&
	    rar5->cstate.window_buf == NULL) {
		archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
				  "Declared solid file, but no window buffer "
				  "initialized yet");
		return ARCHIVE_FATAL;
	}

	/* Check if window_size is a sane value. Also, if the file is not
	 * declared as a directory, disallow window_size == 0. */
	if(window_size > (64 * 1024 * 1024) ||
	    (rar5->file.dir == 0 && window_size == 0))
	{
		archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
		    "Declared dictionary size is not supported");
		return ARCHIVE_FATAL;
	}

	if(rar5->file.solid > 0) {
		/* Re-check if current window size is the same as previous
		 * window size (for solid files only). */
		if(rar5->file.solid_window_size > 0 &&
		    rar5->file.solid_window_size != (ssize_t) window_size)
		{
			archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
			    "Window size for this solid file doesn't match "
			    "the window size used in previous solid file");
			return ARCHIVE_FATAL;
		}
	}
	else
		rar5->cstate.data_encrypted = 0; /* Reset for new buffer */

	if(rar5->cstate.window_size < (ssize_t) window_size &&
	    rar5->cstate.window_buf)
	{
		/* The `data_ready` stack contains pointers to the `window_buf` or
		 * `filtered_buf` buffers.  Since we're about to reallocate the first
		 * buffer, some of those pointers could become invalid. Therefore, we
		 * need to dispose of all entries from the stack before attempting the
		 * realloc. */
		clear_data_ready_stack(rar5);

		/* If window_buf has been allocated before, reallocate it, so
		 * that its size will match new window_size. */

		uint8_t* new_window_buf =
			realloc(rar5->cstate.window_buf, (size_t) window_size);

		if(!new_window_buf) {
			archive_set_error(&a->archive, ARCHIVE_ERRNO_PROGRAMMER,
				"Not enough memory when trying to realloc the window "
				"buffer");
			return ARCHIVE_FATAL;
		}

		rar5->cstate.window_buf = new_window_buf;
	}

	/* Values up to 64M should fit into ssize_t on every
	 * architecture. */
	rar5->cstate.window_size = (ssize_t) window_size;

	if(rar5->file.solid > 0 && rar5->file.solid_window_size == 0) {
		/* Solid files have to have the same window_size across
		   whole archive. Remember the window_size parameter
		   for first solid file found. */
		rar5->file.solid_window_size = rar5->cstate.window_size;
	}

	init_window_mask(rar5);

	rar5->file.service = 0;

	if(!read_var_sized(a, &host_os, NULL))
		return ARCHIVE_EOF;

	if(host_os == HOST_WINDOWS) {
		/* Host OS is Windows */

		__LA_MODE_T mode;

		if(file_attr & ATTR_DIRECTORY) {
			if (file_attr & ATTR_READONLY) {
				mode = 0555 | AE_IFDIR;
			} else {
				mode = 0755 | AE_IFDIR;
			}
		} else {
			if (file_attr & ATTR_READONLY) {
				mode = 0444 | AE_IFREG;
			} else {
				mode = 0644 | AE_IFREG;
			}
		}

		archive_entry_set_mode(entry, mode);

		if (file_attr & (ATTR_READONLY | ATTR_HIDDEN | ATTR_SYSTEM)) {
			char buf[sizeof(",rdonly,hidden,system")];
			const char *fflags[3] = { "", "", "" };
			const char **flag = fflags;

			if (file_attr & ATTR_READONLY)
				*flag++ = ",rdonly";
			if (file_attr & ATTR_HIDDEN)
				*flag++ = ",hidden";
			if (file_attr & ATTR_SYSTEM)
				*flag++ = ",system";

			snprintf(buf, sizeof(buf), "%s%s%s",
			    fflags[0], fflags[1], fflags[2]);
			archive_entry_copy_fflags_text(entry, buf + 1);
		}
	} else if(host_os == HOST_UNIX) {
		/* Host OS is Unix */
		archive_entry_set_mode(entry, (__LA_MODE_T) file_attr);
	} else {
		/* Unknown host OS */
		archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
				"Unsupported Host OS: 0x%jx",
				(uintmax_t)host_os);

		return ARCHIVE_FATAL;
	}

	if(!read_var_sized(a, &name_size, NULL))
		return ARCHIVE_EOF;

	if(name_size > (MAX_NAME_IN_CHARS - 1)) {
		archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
				"Filename is too long");

		return ARCHIVE_FATAL;
	}

	if(name_size == 0) {
		archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
				"No filename specified");

		return ARCHIVE_FATAL;
	}

	if(!read_ahead(a, name_size, &p))
		return ARCHIVE_EOF;

	memcpy(name_utf8_buf, p, name_size);
	name_utf8_buf[name_size] = 0;
	if(ARCHIVE_OK != consume(a, name_size)) {
		return ARCHIVE_EOF;
	}

	archive_entry_update_pathname_utf8(entry, name_utf8_buf);

	if(extra_data_size > 0) {
		int ret = process_head_file_extra(a, entry, rar5,
		    extra_data_size);

		/*
		 * TODO: rewrite or remove useless sanity check
		 *       as extra_data_size is not passed as a pointer
		 *
		if(extra_data_size < 0) {
			archive_set_error(&a->archive, ARCHIVE_ERRNO_PROGRAMMER,
			    "File extra data size is not zero");
			return ARCHIVE_FATAL;
		}
		 */

		if(ret != ARCHIVE_OK)
			return ret;
	}

	if((file_flags & UNKNOWN_UNPACKED_SIZE) == 0) {
		rar5->file.unpacked_size = (ssize_t) unpacked_size;
		if(rar5->file.redir_type == REDIR_TYPE_NONE)
			archive_entry_set_size(entry, unpacked_size);
	}

	if(file_flags & UTIME) {
		archive_entry_set_mtime(entry, (time_t) mtime, 0);
	}

	if(file_flags & CRC32) {
		rar5->file.stored_crc32 = crc;
	}

	if(!rar5->cstate.switch_multivolume) {
		/* Do not reinitialize unpacking state if we're switching
		 * archives. */
		rar5->cstate.block_parsing_finished = 1;
		rar5->cstate.all_filters_applied = 1;
		rar5->cstate.initialized = 0;
	}

	if(rar5->generic.split_before > 0) {
		/* If now we're standing on a header that has a 'split before'
		 * mark, it means we're standing on a 'continuation' file
		 * header. Signal the caller that if it wants to move to
		 * another file, it must call rar5_read_header() function
		 * again. */

		return ARCHIVE_RETRY;
	} else {
		return ARCHIVE_OK;
	}
}

static int process_head_service(struct archive_read* a, struct rar5 *rar5,
    struct archive_entry* entry, size_t block_flags)
{
	/* Process this SERVICE block the same way as FILE blocks. */
	int ret = process_head_file(a, rar5, entry, block_flags);
	if(ret != ARCHIVE_OK)
		return ret;

	rar5->file.service = 1;

	/* But skip the data part automatically. It's no use for the user
	 * anyway.  It contains only service data, not even needed to
	 * properly unpack the file. */
	ret = rar5_read_data_skip(a);
	if(ret != ARCHIVE_OK)
		return ret;

	/* After skipping, try parsing another block automatically. */
	return ARCHIVE_RETRY;
}

static int process_head_main(struct archive_read* a, struct rar5 *rar5,
    struct archive_entry* entry, size_t block_flags)
{
	int ret;
	uint64_t extra_data_size = 0;
	size_t extra_field_size = 0;
	size_t extra_field_id = 0;
	size_t archive_flags = 0;

	enum MAIN_FLAGS {
		VOLUME = 0x0001,         /* multi-volume archive */
		VOLUME_NUMBER = 0x0002,  /* volume number, first vol doesn't
					  * have it */
		SOLID = 0x0004,          /* solid archive */
		PROTECT = 0x0008,        /* contains Recovery info */
		LOCK = 0x0010,           /* readonly flag, not used */
	};

	enum MAIN_EXTRA {
		// Just one attribute here.
		LOCATOR = 0x01,
	};

	(void) entry;

	if(block_flags & HFL_EXTRA_DATA) {
		if(!read_var(a, &extra_data_size, NULL))
			return ARCHIVE_EOF;
	} else {
		extra_data_size = 0;
	}

	if(!read_var_sized(a, &archive_flags, NULL)) {
		return ARCHIVE_EOF;
	}

	rar5->main.volume = (archive_flags & VOLUME) > 0;
	rar5->main.solid = (archive_flags & SOLID) > 0;

	if(archive_flags & VOLUME_NUMBER) {
		size_t v = 0;
		if(!read_var_sized(a, &v, NULL)) {
			return ARCHIVE_EOF;
		}

		if (v > UINT_MAX) {
			archive_set_error(&a->archive,
			    ARCHIVE_ERRNO_FILE_FORMAT,
			    "Invalid volume number");
			return ARCHIVE_FATAL;
		}

		rar5->main.vol_no = (unsigned int) v;
	} else {
		rar5->main.vol_no = 0;
	}

	if(rar5->vol.expected_vol_no > 0 &&
		rar5->main.vol_no != rar5->vol.expected_vol_no)
	{
		/* Returning EOF instead of FATAL because of strange
		 * libarchive behavior. When opening multiple files via
		 * archive_read_open_filenames(), after reading up the whole
		 * last file, the __archive_read_ahead function wraps up to
		 * the first archive instead of returning EOF. */
		return ARCHIVE_EOF;
	}

	if(extra_data_size == 0) {
		/* Early return. */
		return ARCHIVE_OK;
	}

	if(!read_var_sized(a, &extra_field_size, NULL)) {
		return ARCHIVE_EOF;
	}

	if(!read_var_sized(a, &extra_field_id, NULL)) {
		return ARCHIVE_EOF;
	}

	if(extra_field_size == 0) {
		archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
		    "Invalid extra field size");
		return ARCHIVE_FATAL;
	}

	switch(extra_field_id) {
		case LOCATOR:
			ret = process_main_locator_extra_block(a, rar5);
			if(ret != ARCHIVE_OK) {
				/* Error while parsing main locator extra
				 * block. */
				return ret;
			}

			break;
		default:
			archive_set_error(&a->archive,
			    ARCHIVE_ERRNO_FILE_FORMAT,
			    "Unsupported extra type (0x%jx)",
			    (uintmax_t)extra_field_id);
			return ARCHIVE_FATAL;
	}

	return ARCHIVE_OK;
}

static int skip_unprocessed_bytes(struct archive_read* a) {
	struct rar5 *rar5 = a->format->data;
	int ret;

	if(rar5->file.bytes_remaining) {
		/* Use different skipping method in block merging mode than in
		 * normal mode. If merge mode is active, rar5_read_data_skip
		 * can't be used, because it could allow recursive use of
		 * merge_block() * function, and this function doesn't support
		 * recursive use. */
		if(rar5->merge_mode) {
			/* Discard whole merged block. This is valid in solid
			 * mode as well, because the code will discard blocks
			 * only if those blocks are safe to discard (i.e.
			 * they're not FILE blocks).  */
			ret = consume(a, rar5->file.bytes_remaining);
			if(ret != ARCHIVE_OK) {
				return ret;
			}
			rar5->file.bytes_remaining = 0;
		} else {
			/* If we're not in merge mode, use safe skipping code.
			 * This will ensure we'll handle solid archives
			 * properly. */
			ret = rar5_read_data_skip(a);
			if(ret != ARCHIVE_OK) {
				return ret;
			}
		}
	}

	return ARCHIVE_OK;
}

static int scan_for_signature(struct archive_read* a);

/* Base block processing function. A 'base block' is a RARv5 header block
 * that tells the reader what kind of data is stored inside the block.
 *
 * From the birds-eye view a RAR file looks file this:
 *
 * <magic><base_block_1><base_block_2>...<base_block_n>
 *
 * There are a few types of base blocks. Those types are specified inside
 * the 'switch' statement in this function. For example purposes, I'll write
 * how a standard RARv5 file could look like here:
 *
 * <magic><MAIN><FILE><FILE><FILE><SERVICE><ENDARC>
 *
 * The structure above could describe an archive file with 3 files in it,
 * one service "QuickOpen" block (that is ignored by this parser), and an
 * end of file base block marker.
 *
 * If the file is stored in multiple archive files ("multiarchive"), it might
 * look like this:
 *
 * .part01.rar: <magic><MAIN><FILE><ENDARC>
 * .part02.rar: <magic><MAIN><FILE><ENDARC>
 * .part03.rar: <magic><MAIN><FILE><ENDARC>
 *
 * This example could describe 3 RAR files that contain ONE archived file.
 * Or it could describe 3 RAR files that contain 3 different files. Or 3
 * RAR files than contain 2 files. It all depends what metadata is stored in
 * the headers of <FILE> blocks.
 *
 * Each <FILE> block contains info about its size, the name of the file it's
 * storing inside, and whether this FILE block is a continuation block of
 * previous archive ('split before'), and is this FILE block should be
 * continued in another archive ('split after'). By parsing the 'split before'
 * and 'split after' flags, we're able to tell if multiple <FILE> base blocks
 * are describing one file, or multiple files (with the same filename, for
 * example).
 *
 * One thing to note is that if we're parsing the first <FILE> block, and
 * we see 'split after' flag, then we need to jump over to another <FILE>
 * block to be able to decompress rest of the data. To do this, we need
 * to skip the <ENDARC> block, then switch to another file, then skip the
 * <magic> block, <MAIN> block, and then we're standing on the proper
 * <FILE> block.
 */

/*
 * A header that carries no file data (HEAD_MAIN, or an unknown block
 * flagged HFL_SKIP_IF_UNKNOWN) may leave bytes in its body that the
 * sub-parser did not read. Skip them before returning ARCHIVE_RETRY,
 * otherwise rar5_read_header() re-parses the same block region O(N)
 * times instead of O(1), letting a crafted RAR5 file stall the reader
 * (GHSA-9h2c-464f-j3hj).
 *
 * Safe because read_ahead(a, hdr_size, &p) pre-loaded the whole block
 * into one contiguous buffer with no compaction until we return, so
 * body_start stays valid and (cur - body_start) is the exact number of
 * body bytes consumed so far.
 */
static void
rar5_skip_remaining_block(struct archive_read* a,
    const uint8_t* body_start, size_t raw_hdr_size)
{
	const uint8_t* cur;

	if(read_ahead(a, 1, &cur)) {
		size_t body_used = (size_t)(cur - body_start);

		if(body_used < raw_hdr_size)
			(void)consume(a, raw_hdr_size - body_used);
	}
}

static int process_base_block(struct archive_read* a,
    struct archive_entry* entry)
{
	const size_t SMALLEST_RAR5_BLOCK_SIZE = 3;

	struct rar5 *rar5 = a->format->data;
	uint32_t hdr_crc, computed_crc;
	size_t raw_hdr_size = 0, hdr_size_len, hdr_size;
	size_t header_id = 0;
	size_t header_flags = 0;
	const uint8_t* p;
	const uint8_t* body_start;
	int ret;

	enum HEADER_TYPE {
		HEAD_MARK    = 0x00, HEAD_MAIN  = 0x01, HEAD_FILE   = 0x02,
		HEAD_SERVICE = 0x03, HEAD_CRYPT = 0x04, HEAD_ENDARC = 0x05,
		HEAD_UNKNOWN = 0xff,
	};

	/* Skip any unprocessed data for this file. */
	ret = skip_unprocessed_bytes(a);
	if(ret != ARCHIVE_OK)
		return ret;

	/* Read the expected CRC32 checksum. */
	if(!read_u32(a, &hdr_crc)) {
		return ARCHIVE_EOF;
	}

	/* Read header size. */
	if(!read_var_sized(a, &raw_hdr_size, &hdr_size_len)) {
		return ARCHIVE_EOF;
	}

	hdr_size = raw_hdr_size + hdr_size_len;

	/* Sanity check, maximum header size for RAR5 is 2MB. */
	if(hdr_size > (2 * 1024 * 1024)) {
		archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
		    "Base block header is too large");

		return ARCHIVE_FATAL;
	}

	/* Additional sanity checks to weed out invalid files. */
	if(raw_hdr_size == 0 || hdr_size_len == 0 ||
		hdr_size < SMALLEST_RAR5_BLOCK_SIZE)
	{
		archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
		    "Too small block encountered (%zu bytes)",
		    raw_hdr_size);

		return ARCHIVE_FATAL;
	}

	/* Read the whole header data into memory, maximum memory use here is
	 * 2MB. */
	if(!read_ahead(a, hdr_size, &p)) {
		return ARCHIVE_EOF;
	}

	/* Verify the CRC32 of the header data. */
	computed_crc = (uint32_t) crc32(0, p, (int) hdr_size);
	if(computed_crc != hdr_crc) {
#ifndef DONT_FAIL_ON_CRC_ERROR
		archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
		    "Header CRC error");

		return ARCHIVE_FATAL;
#endif
	}

	/* Remember the first byte of the block body so we can later skip
	 * any bytes the sub-parser leaves unconsumed. */
	body_start = p + hdr_size_len;

	/* If the checksum is OK, we proceed with parsing. */
	if(ARCHIVE_OK != consume(a, hdr_size_len)) {
		return ARCHIVE_EOF;
	}

	if(!read_var_sized(a, &header_id, NULL))
		return ARCHIVE_EOF;

	if(!read_var_sized(a, &header_flags, NULL))
		return ARCHIVE_EOF;

	rar5->generic.split_after = (header_flags & HFL_SPLIT_AFTER) > 0;
	rar5->generic.split_before = (header_flags & HFL_SPLIT_BEFORE) > 0;
	rar5->generic.size = (int)hdr_size;
	rar5->generic.last_header_id = (int)header_id;
	rar5->main.endarc = 0;

	/* Those are possible header ids in RARv5. */
	switch(header_id) {
		case HEAD_MAIN:
			ret = process_head_main(a, rar5, entry, header_flags);

			/* Main header doesn't have any files in it, so it's
			 * pointless to return to the caller. Retry to next
			 * header, which should be HEAD_FILE/HEAD_SERVICE. */
			if(ret == ARCHIVE_OK) {
				rar5_skip_remaining_block(a, body_start,
				    raw_hdr_size);
				return ARCHIVE_RETRY;
			}

			return ret;
		case HEAD_SERVICE:
			ret = process_head_service(a, rar5, entry, header_flags);
			return ret;
		case HEAD_FILE:
			ret = process_head_file(a, rar5, entry, header_flags);
			return ret;
		case HEAD_CRYPT:
			archive_entry_set_is_metadata_encrypted(entry, 1);
			archive_entry_set_is_data_encrypted(entry, 1);
			rar5->has_encrypted_entries = 1;
			rar5->headers_are_encrypted = 1;
			archive_set_error(&a->archive,
			    ARCHIVE_ERRNO_FILE_FORMAT,
			    "Encryption is not supported");
			return ARCHIVE_FATAL;
		case HEAD_ENDARC:
			rar5->main.endarc = 1;

			/* After encountering an end of file marker, we need
			 * to take into consideration if this archive is
			 * continued in another file (i.e. is it part01.rar:
			 * is there a part02.rar?) */
			if(rar5->main.volume) {
				/* In case there is part02.rar, position the
				 * read pointer in a proper place, so we can
				 * resume parsing. */
				ret = scan_for_signature(a);
				if(ret == ARCHIVE_FATAL) {
					return ARCHIVE_EOF;
				} else {
					if(rar5->vol.expected_vol_no ==
					    UINT_MAX) {
						archive_set_error(&a->archive,
						    ARCHIVE_ERRNO_FILE_FORMAT,
						    "Header error");
							return ARCHIVE_FATAL;
					}

					rar5->vol.expected_vol_no =
					    rar5->main.vol_no + 1;
					return ARCHIVE_OK;
				}
			} else {
				return ARCHIVE_EOF;
			}
		case HEAD_MARK:
			return ARCHIVE_EOF;
		default:
			if((header_flags & HFL_SKIP_IF_UNKNOWN) == 0) {
				archive_set_error(&a->archive,
				    ARCHIVE_ERRNO_FILE_FORMAT,
				    "Header type error");
				return ARCHIVE_FATAL;
			} else {
				/* If the block is marked as 'skip if unknown',
				 * do as the flag says: skip the block
				 * instead on failing on it. */
				rar5_skip_remaining_block(a, body_start,
				    raw_hdr_size);
				return ARCHIVE_RETRY;
			}
	}

#if !defined WIN32
	// Not reached.
	archive_set_error(&a->archive, ARCHIVE_ERRNO_PROGRAMMER,
	    "Internal unpacker error");
	return ARCHIVE_FATAL;
#endif
}

static int skip_base_block(struct archive_read* a) {
	struct rar5 *rar5 = a->format->data;
	int ret;

	/* Create a new local archive_entry structure that will be operated on
	 * by header reader; operations on this archive_entry will be discarded.
	 */
	struct archive_entry* entry = archive_entry_new();
	if (entry == NULL)
		return ARCHIVE_FATAL;

	ret = process_base_block(a, entry);

	/* Discard operations on this archive_entry structure. */
	archive_entry_free(entry);
	if(ret == ARCHIVE_FATAL)
		return ret;

	if(rar5->generic.last_header_id == 2 && rar5->generic.split_before > 0)
		return ARCHIVE_OK;

	if(ret == ARCHIVE_OK)
		return ARCHIVE_RETRY;
	else
		return ret;
}

static int try_skip_sfx(struct archive_read *a)
{
	const char *p;

	if ((p = __archive_read_ahead(a, 7, NULL)) == NULL)
		return ARCHIVE_EOF;

	if ((p[0] == 'M' && p[1] == 'Z') || memcmp(p, "\x7F\x45LF", 4) == 0)
	{
		char signature[sizeof(rar5_signature_xor)];
		const void *h;
		const char *q;
		size_t skip, total = 0;
		ssize_t bytes, window = 4096;

		rar5_signature(signature);

		while (total + window <= (1024 * 512)) {
			h = __archive_read_ahead(a, window, &bytes);
			if (h == NULL) {
				/* Remaining bytes are less than window. */
				window >>= 1;
				if (window < 0x40)
					goto fatal;
				continue;
			}
			if (bytes < 0x40)
				goto fatal;
			p = h;
			q = p + bytes;

			/*
			 * Scan ahead until we find something that looks
			 * like the RAR header.
			 */
			while (p + 8 < q) {
				if (memcmp(p, signature, sizeof(signature)) == 0) {
					skip = p - (const char *)h;
					__archive_read_consume(a, skip);
					return (ARCHIVE_OK);
				}
				p += 0x10;
			}
			skip = p - (const char *)h;
			__archive_read_consume(a, skip);
			total += skip;
		}
	}

	return ARCHIVE_OK;
fatal:
	archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
			"Couldn't find out RAR header");
	return (ARCHIVE_FATAL);
}

static int rar5_read_header(struct archive_read *a,
    struct archive_entry *entry)
{
	struct rar5 *rar5 = a->format->data;
	int ret;

	/*
	 * It should be sufficient to call archive_read_next_header() for
	 * a reader to determine if an entry is encrypted or not.
	 */
	if (rar5->has_encrypted_entries == ARCHIVE_READ_FORMAT_ENCRYPTION_DONT_KNOW) {
		rar5->has_encrypted_entries = 0;
	}

	if(rar5->header_initialized == 0) {
		init_header(a);
		if ((ret = try_skip_sfx(a)) < ARCHIVE_WARN)
			return ret;
		rar5->header_initialized = 1;
	}

	if(rar5->skipped_magic == 0) {
		if(ARCHIVE_OK != consume(a, sizeof(rar5_signature_xor))) {
			return ARCHIVE_EOF;
		}

		rar5->skipped_magic = 1;
	}

	do {
		ret = process_base_block(a, entry);
	} while(ret == ARCHIVE_RETRY ||
			(rar5->main.endarc > 0 && ret == ARCHIVE_OK));

	return ret;
}

static int init_unpack(struct rar5 *rar5) {
	rar5->file.calculated_crc32 = 0;
	init_window_mask(rar5);

	free(rar5->cstate.window_buf);
	free(rar5->cstate.filtered_buf);

	rar5->cstate.window_buf = NULL;
	rar5->cstate.filtered_buf = NULL;

	if(rar5->cstate.window_size > 0) {
		rar5->cstate.window_buf = calloc(1, rar5->cstate.window_size);
		if(rar5->cstate.window_buf == NULL)
			return ARCHIVE_FATAL;
		rar5->cstate.filtered_buf = calloc(1,
		    rar5->cstate.window_size);
		if(rar5->cstate.filtered_buf == NULL)
			return ARCHIVE_FATAL;
	}

	clear_data_ready_stack(rar5);

	rar5->cstate.write_ptr = 0;
	rar5->cstate.last_write_ptr = 0;

	memset(&rar5->cstate.bd, 0, sizeof(rar5->cstate.bd));
	memset(&rar5->cstate.ld, 0, sizeof(rar5->cstate.ld));
	memset(&rar5->cstate.dd, 0, sizeof(rar5->cstate.dd));
	memset(&rar5->cstate.ldd, 0, sizeof(rar5->cstate.ldd));
	memset(&rar5->cstate.rd, 0, sizeof(rar5->cstate.rd));
	return ARCHIVE_OK;
}

static void update_crc(struct rar5 *rar5, const uint8_t* p, size_t to_read) {
    int verify_crc;

	if(rar5->skip_mode) {
#if defined CHECK_CRC_ON_SOLID_SKIP
		verify_crc = 1;
#else
		verify_crc = 0;
#endif
	} else
		verify_crc = 1;

	if(verify_crc) {
		/* Don't update CRC32 if the file doesn't have the
		 * `stored_crc32` info filled in. */
		if(rar5->file.stored_crc32 > 0) {
			rar5->file.calculated_crc32 =
				crc32(rar5->file.calculated_crc32, p,
				    (unsigned int)to_read);
		}

		/* Check if the file uses an optional BLAKE2sp checksum
		 * algorithm. */
		if(rar5->file.has_blake2 > 0) {
			/* Return value of the `update` function is always 0,
			 * so we can explicitly ignore it here. */
			(void) blake2sp_update(&rar5->file.b2state, p,
			    to_read);
		}
	}
}

static int create_decode_tables(uint8_t* bit_length,
    struct decode_table* table, int size)
{
	int code, upper_limit = 0, i, lc[16];
	uint32_t decode_pos_clone[rar5_countof(table->decode_pos)];
	ssize_t cur_len, quick_data_size;

	memset(&lc, 0, sizeof(lc));
	memset(table->decode_num, 0, sizeof(table->decode_num));
	table->size = size;
	table->quick_bits = size == HUFF_NC ? 10 : 7;

	for(i = 0; i < size; i++) {
		lc[bit_length[i] & 15]++;
	}

	lc[0] = 0;
	table->decode_pos[0] = 0;
	table->decode_len[0] = 0;

	for(i = 1; i < 16; i++) {
		upper_limit += lc[i];

		table->decode_len[i] = upper_limit << (16 - i);
		table->decode_pos[i] = table->decode_pos[i - 1] + lc[i - 1];

		upper_limit <<= 1;
	}

	/* Verify the code-length distribution is not over-subscribed.
	 * After the loop above, upper_limit == sum(lc[i] * 2^(16-i)).
	 * For a valid prefix-free code this must be <= 2^16 = 65536.
	 * An over-subscribed table (> 65536) cannot produce a valid
	 * decode table and must be rejected. */
	if(upper_limit > 65536) {
		return ARCHIVE_FAILED;
	}

	memcpy(decode_pos_clone, table->decode_pos, sizeof(decode_pos_clone));

	for(i = 0; i < size; i++) {
		uint8_t clen = bit_length[i] & 15;
		if(clen > 0) {
			int last_pos = decode_pos_clone[clen];
			table->decode_num[last_pos] = i;
			decode_pos_clone[clen]++;
		}
	}

	quick_data_size = (int64_t)1 << table->quick_bits;
	cur_len = 1;
	for(code = 0; code < quick_data_size; code++) {
		int bit_field = code << (16 - table->quick_bits);
		int dist, pos;

		while(cur_len < rar5_countof(table->decode_len) &&
				bit_field >= table->decode_len[cur_len]) {
			cur_len++;
		}

		table->quick_len[code] = (uint8_t) cur_len;

		dist = bit_field - table->decode_len[cur_len - 1];
		dist >>= (16 - cur_len);

		pos = table->decode_pos[cur_len & 15] + dist;
		if(cur_len < rar5_countof(table->decode_pos) && pos < size) {
			table->quick_num[code] = table->decode_num[pos];
		} else {
			table->quick_num[code] = 0;
		}
	}

	return ARCHIVE_OK;
}

static int decode_number(struct archive_read* a, struct decode_table* table,
    const uint8_t* p, uint16_t* num)
{
	struct rar5 *rar5 = a->format->data;
	int i, bits, dist, ret;
	uint16_t bitfield;
	uint32_t pos;

	if(ARCHIVE_OK != (ret = read_bits_16(a, rar5, p, &bitfield))) {
		return ret;
	}

	bitfield &= 0xfffe;

	if(bitfield < table->decode_len[table->quick_bits]) {
		int code = bitfield >> (16 - table->quick_bits);
		skip_bits(rar5, table->quick_len[code]);
		*num = table->quick_num[code];
		return ARCHIVE_OK;
	}

	bits = 15;

	for(i = table->quick_bits + 1; i < 15; i++) {
		if(bitfield < table->decode_len[i]) {
			bits = i;
			break;
		}
	}

	skip_bits(rar5, bits);

	dist = bitfield - table->decode_len[bits - 1];
	dist >>= (16 - bits);
	pos = table->decode_pos[bits] + dist;

	if(pos >= table->size)
		pos = 0;

	*num = table->decode_num[pos];
	return ARCHIVE_OK;
}

/* Reads and parses Huffman tables from the beginning of the block. */
static int parse_tables(struct archive_read* a, struct rar5 *rar5,
    const uint8_t* p)
{
	int ret, value, i, w, idx = 0;
	uint8_t bit_length[HUFF_BC],
		table[HUFF_TABLE_SIZE],
		nibble_mask = 0xF0,
		nibble_shift = 4;

	enum { ESCAPE = 15 };

	/* The data for table generation is compressed using a simple RLE-like
	 * algorithm when storing zeroes, so we need to unpack it first. */
	for(w = 0, i = 0; w < HUFF_BC;) {
		if(i >= rar5->cstate.cur_block_size) {
			/* Truncated data, can't continue. */
			archive_set_error(&a->archive,
			    ARCHIVE_ERRNO_FILE_FORMAT,
			    "Truncated data in huffman tables");
			return ARCHIVE_FAILED;
		}

		value = (p[i] & nibble_mask) >> nibble_shift;

		if(nibble_mask == 0x0F)
			++i;

		nibble_mask ^= 0xFF;
		nibble_shift ^= 4;

		/* Values smaller than 15 is data, so we write it directly.
		 * Value 15 is a flag telling us that we need to unpack more
		 * bytes. */
		if(value == ESCAPE) {
			value = (p[i] & nibble_mask) >> nibble_shift;
			if(nibble_mask == 0x0F)
				++i;
			nibble_mask ^= 0xFF;
			nibble_shift ^= 4;

			if(value == 0) {
				/* We sometimes need to write the actual value
				 * of 15, so this case handles that. */
				bit_length[w++] = ESCAPE;
			} else {
				int k;

				/* Fill zeroes. */
				for(k = 0; (k < value + 2) && (w < HUFF_BC);
				    k++) {
					bit_length[w++] = 0;
				}
			}
		} else {
			bit_length[w++] = value;
		}
	}

	rar5->bits.in_addr = i;
	rar5->bits.bit_addr = nibble_shift ^ 4;

	ret = create_decode_tables(bit_length, &rar5->cstate.bd, HUFF_BC);
	if(ret != ARCHIVE_OK) {
		archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
		    "Decoding huffman tables failed");
		return ARCHIVE_FAILED;
	}

	for(i = 0; i < HUFF_TABLE_SIZE;) {
		uint16_t num;

		ret = decode_number(a, &rar5->cstate.bd, p, &num);
		if(ret != ARCHIVE_OK) {
			archive_set_error(&a->archive,
			    ARCHIVE_ERRNO_FILE_FORMAT,
			    "Decoding huffman tables failed");
			return ARCHIVE_FAILED;
		}

		if(num < 16) {
			/* 0..15: store directly */
			table[i] = (uint8_t) num;
			i++;
		} else if(num < 18) {
			/* 16..17: repeat previous code */
			uint16_t n;

			if(ARCHIVE_OK != (ret = read_bits_16(a, rar5, p, &n)))
				return ret;

			if(num == 16) {
				n >>= 13;
				n += 3;
				skip_bits(rar5, 3);
			} else {
				n >>= 9;
				n += 11;
				skip_bits(rar5, 7);
			}

			if(i > 0) {
				while(n-- > 0 && i < HUFF_TABLE_SIZE) {
					table[i] = table[i - 1];
					i++;
				}
			} else {
				archive_set_error(&a->archive,
				    ARCHIVE_ERRNO_FILE_FORMAT,
				    "Unexpected error when decoding "
				    "huffman tables");
				return ARCHIVE_FAILED;
			}
		} else {
			/* other codes: fill with zeroes `n` times */
			uint16_t n;

			if(ARCHIVE_OK != (ret = read_bits_16(a, rar5, p, &n)))
				return ret;

			if(num == 18) {
				n >>= 13;
				n += 3;
				skip_bits(rar5, 3);
			} else {
				n >>= 9;
				n += 11;
				skip_bits(rar5, 7);
			}

			while(n-- > 0 && i < HUFF_TABLE_SIZE)
				table[i++] = 0;
		}
	}

	ret = create_decode_tables(&table[idx], &rar5->cstate.ld, HUFF_NC);
	if(ret != ARCHIVE_OK) {
		archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
		     "Failed to create literal table");
		return ARCHIVE_FAILED;
	}

	idx += HUFF_NC;

	ret = create_decode_tables(&table[idx], &rar5->cstate.dd, HUFF_DC);
	if(ret != ARCHIVE_OK) {
		archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
		    "Failed to create distance table");
		return ARCHIVE_FAILED;
	}

	idx += HUFF_DC;

	ret = create_decode_tables(&table[idx], &rar5->cstate.ldd, HUFF_LDC);
	if(ret != ARCHIVE_OK) {
		archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
		    "Failed to create lower bits of distances table");
		return ARCHIVE_FAILED;
	}

	idx += HUFF_LDC;

	ret = create_decode_tables(&table[idx], &rar5->cstate.rd, HUFF_RC);
	if(ret != ARCHIVE_OK) {
		archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
		    "Failed to create repeating distances table");
		return ARCHIVE_FAILED;
	}

	return ARCHIVE_OK;
}

/* Parses the block header, verifies its CRC byte, and saves the header
 * fields inside the `hdr` pointer. */
static int parse_block_header(struct archive_read* a, const uint8_t* p,
    ssize_t* block_size, struct compressed_block_header* hdr)
{
	uint8_t calculated_cksum;
	memcpy(hdr, p, sizeof(struct compressed_block_header));

	if(bf_byte_count(hdr) > 2) {
		archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
		    "Unsupported block header size (was %d, max is 2)",
		    bf_byte_count(hdr));
		return ARCHIVE_FAILED;
	}

	/* This should probably use bit reader interface in order to be more
	 * future-proof. */
	*block_size = 0;
	switch(bf_byte_count(hdr)) {
		/* 1-byte block size */
		case 0:
			*block_size = *(const uint8_t*) &p[2];
			break;

		/* 2-byte block size */
		case 1:
			*block_size = archive_le16dec(&p[2]);
			break;

		/* 3-byte block size */
		case 2:
			*block_size = archive_le32dec(&p[2]);
			*block_size &= 0x00FFFFFF;
			break;

		/* Other block sizes are not supported. This case is not
		 * reached, because we have an 'if' guard before the switch
		 * that makes sure of it. */
		default:
			return ARCHIVE_FATAL;
	}

	/* Verify the block header checksum. 0x5A is a magic value and is
	 * always * constant. */
	calculated_cksum = 0x5A
	    ^ (uint8_t) hdr->block_flags_u8
	    ^ (uint8_t) *block_size
	    ^ (uint8_t) (*block_size >> 8)
	    ^ (uint8_t) (*block_size >> 16);

	if(calculated_cksum != hdr->block_cksum) {
#ifndef DONT_FAIL_ON_CRC_ERROR
		archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
		    "Block checksum error: got 0x%x, expected 0x%x",
		    hdr->block_cksum, calculated_cksum);

		return ARCHIVE_FAILED;
#endif
	}

	return ARCHIVE_OK;
}

/* Convenience function used during filter processing. */
static int parse_filter_data(struct archive_read* a, struct rar5 *rar5,
	const uint8_t* p, uint32_t* filter_data)
{
	int i, bytes, ret;
	uint32_t data = 0;

	if(ARCHIVE_OK != (ret = read_consume_bits(a, rar5, p, 2, &bytes)))
		return ret;

	bytes++;

	for(i = 0; i < bytes; i++) {
		uint16_t byte;

		if(ARCHIVE_OK != (ret = read_bits_16(a, rar5, p, &byte))) {
			return ret;
		}

		/* Cast to uint32_t will ensure the shift operation will not
		 * produce undefined result. */
		data += ((uint32_t) byte >> 8) << (i * 8);
		skip_bits(rar5, 8);
	}

	*filter_data = data;
	return ARCHIVE_OK;
}

/* Function is used during sanity checking. */
static int is_valid_filter_block_start(struct rar5 *rar5,
    uint32_t start)
{
	const int64_t block_start = (ssize_t) start + rar5->cstate.write_ptr;
	const int64_t last_bs = rar5->cstate.last_block_start;
	const ssize_t last_bl = rar5->cstate.last_block_length;

	if(last_bs == 0 || last_bl == 0) {
		/* We didn't have any filters yet, so accept this offset. */
		return 1;
	}

	if(block_start >= last_bs + last_bl) {
		/* Current offset is bigger than last block's end offset, so
		 * accept current offset. */
		return 1;
	}

	/* Any other case is not a normal situation and we should fail. */
	return 0;
}

/* The function will create a new filter, read its parameters from the input
 * stream and add it to the filter collection. */
static int parse_filter(struct archive_read* ar, const uint8_t* p) {
	struct rar5 *rar5 = ar->format->data;
	uint32_t block_start, block_length;
	uint16_t filter_type;
	struct filter_info* filt = NULL;
	int ret;

	/* Read the parameters from the input stream. */
	if(ARCHIVE_OK != (ret = parse_filter_data(ar, rar5, p, &block_start)))
		return ret;

	if(ARCHIVE_OK != (ret = parse_filter_data(ar, rar5, p, &block_length)))
		return ret;

	if(ARCHIVE_OK != (ret = read_bits_16(ar, rar5, p, &filter_type)))
		return ret;

	filter_type >>= 13;
	skip_bits(rar5, 3);

	/* Perform some sanity checks on this filter parameters. */

	if(block_length < 4 ||
	    block_length > 0x400000 ||
	    !is_valid_filter_block_start(rar5, block_start) ||
	    (rar5->cstate.window_size > 0 &&
	     (ssize_t)block_length > rar5->cstate.window_size >> 1))
	{
		archive_set_error(&ar->archive, ARCHIVE_ERRNO_FILE_FORMAT,
		    "Invalid filter encountered");
		return ARCHIVE_FAILED;
	}

	/* Allocate a new filter. */
	filt = add_new_filter(rar5);
	if(filt == NULL) {
		archive_set_error(&ar->archive, ENOMEM,
		    "Can't allocate memory for a filter descriptor");
		return ARCHIVE_FATAL;
	}

	filt->type = filter_type;
	filt->block_start = rar5->cstate.write_ptr + block_start;
	filt->block_length = block_length;

	rar5->cstate.last_block_start = filt->block_start;
	rar5->cstate.last_block_length = filt->block_length;

	/* Read some more data in case this is a DELTA filter. Other filter
	 * types don't require any additional data over what was already
	 * read. */
	if(filter_type == FILTER_DELTA) {
		int channels;

		if(ARCHIVE_OK != (ret = read_consume_bits(ar, rar5, p, 5, &channels)))
			return ret;

		filt->channels = channels + 1;
	}

	return ARCHIVE_OK;
}

static int decode_code_length(struct archive_read* a, struct rar5 *rar5,
	const uint8_t* p, uint16_t code)
{
	int lbits, length = 2;

	if(code < 8) {
		lbits = 0;
		length += code;
	} else {
		lbits = code / 4 - 1;
		length += (4 | (code & 3)) << lbits;
	}

	if(lbits > 0) {
		int add;

		if(ARCHIVE_OK != read_consume_bits(a, rar5, p, lbits, &add))
			return -1;

		length += add;
	}

	return length;
}

static int copy_string(struct archive_read* a, int len, int dist) {
	struct rar5 *rar5 = a->format->data;
	const ssize_t cmask = rar5->cstate.window_mask;
	const uint64_t write_ptr = rar5->cstate.write_ptr +
	    rar5->cstate.solid_offset;
	int i;

	if (rar5->cstate.window_buf == NULL)
		return ARCHIVE_FATAL;

	if (rar5->cstate.write_ptr > rar5->file.unpacked_size ||
	    len > rar5->file.unpacked_size - rar5->cstate.write_ptr) {
		archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
		    "Uncompressed data exceeds declared size");
		return rar5->main.solid ? ARCHIVE_FATAL : ARCHIVE_FAILED;
	}

	/* The unpacker spends most of the time in this function. It would be
	 * a good idea to introduce some optimizations here.
	 *
	 * Just remember that this loop treats buffers that overlap differently
	 * than buffers that do not overlap. This is why a simple memcpy(3)
	 * call will not be enough. */

	for(i = 0; i < len; i++) {
		const ssize_t write_idx = (write_ptr + i) & cmask;
		const ssize_t read_idx = (write_ptr + i - dist) & cmask;
		rar5->cstate.window_buf[write_idx] =
		    rar5->cstate.window_buf[read_idx];
	}

	rar5->cstate.write_ptr += len;
	return ARCHIVE_OK;
}

static int do_uncompress_block(struct archive_read* a, const uint8_t* p) {
	struct rar5 *rar5 = a->format->data;
	uint16_t num;
	int ret;

	const uint64_t cmask = rar5->cstate.window_mask;
	const struct compressed_block_header* hdr = &rar5->last_block_hdr;
	const uint8_t bit_size = 1 + bf_bit_size(hdr);

	while(1) {
		if(rar5->cstate.write_ptr - rar5->cstate.last_write_ptr >
		    (rar5->cstate.window_size >> 1)) {
			/* Don't allow growing data by more than half of the
			 * window size at a time. In such case, break the loop;
			 *  next call to this function will continue processing
			 *  from this moment. */
			break;
		}

		if(rar5->bits.in_addr > rar5->cstate.cur_block_size - 1 ||
		    (rar5->bits.in_addr == rar5->cstate.cur_block_size - 1 &&
		    rar5->bits.bit_addr >= bit_size))
		{
			/* If the program counter is here, it means the
			 * function has finished processing the block. */
			rar5->cstate.block_parsing_finished = 1;
			break;
		}

		/* Decode the next literal. */
		if(ARCHIVE_OK != decode_number(a, &rar5->cstate.ld, p, &num)) {
			return ARCHIVE_EOF;
		}

		/* Num holds a decompression literal, or 'command code'.
		 *
		 * - Values lower than 256 are just bytes. Those codes
		 *   can be stored in the output buffer directly.
		 *
		 * - Code 256 defines a new filter, which is later used to
		 *   transform the data block accordingly to the filter type.
		 *   The data block needs to be fully uncompressed first.
		 *
		 * - Code bigger than 257 and smaller than 262 define
		 *   a repetition pattern that should be copied from
		 *   an already uncompressed chunk of data.
		 */

		if(num < 256) {
			/* Directly store the byte. */
			int64_t write_idx;

			/* A literal write emits one byte; copy_string() checks len. */
			if(rar5->cstate.write_ptr >= rar5->file.unpacked_size) {
				archive_set_error(&a->archive,
				    ARCHIVE_ERRNO_FILE_FORMAT,
				    "Uncompressed data exceeds declared size");
				return rar5->main.solid ? ARCHIVE_FATAL : ARCHIVE_FAILED;
			}

			write_idx = rar5->cstate.solid_offset +
			    rar5->cstate.write_ptr++;

			rar5->cstate.window_buf[write_idx & cmask] =
			    (uint8_t) num;
			continue;
		} else if(num >= 262) {
			uint16_t dist_slot;
			int len = decode_code_length(a, rar5, p, num - 262),
				dbits,
				dist = 1;

			if(len == -1) {
				archive_set_error(&a->archive,
				    ARCHIVE_ERRNO_PROGRAMMER,
				    "Failed to decode the code length");

				return rar5->main.solid ? ARCHIVE_FATAL : ARCHIVE_FAILED;
			}

			if(ARCHIVE_OK != decode_number(a, &rar5->cstate.dd, p,
			    &dist_slot))
			{
				archive_set_error(&a->archive,
				    ARCHIVE_ERRNO_PROGRAMMER,
				    "Failed to decode the distance slot");

				return rar5->main.solid ? ARCHIVE_FATAL : ARCHIVE_FAILED;
			}

			if(dist_slot < 4) {
				dbits = 0;
				dist += dist_slot;
			} else {
				dbits = dist_slot / 2 - 1;

				/* Cast to uint32_t will make sure the shift
				 * left operation won't produce undefined
				 * result. Then, the uint32_t type will
				 * be implicitly casted to int. */
				dist += (uint32_t) (2 |
				    (dist_slot & 1)) << dbits;
			}

			if(dbits > 0) {
				if(dbits >= 4) {
					uint32_t add = 0;
					uint16_t low_dist;

					if(dbits > 4) {
						if(ARCHIVE_OK != (ret = read_bits_32(
						    a, rar5, p, &add))) {
							/* Return EOF if we
							 * can't read more
							 * data. */
							return ret;
						}

						skip_bits(rar5, dbits - 4);
						add = (add >> (
						    36 - dbits)) << 4;
						dist += add;
					}

					if(ARCHIVE_OK != decode_number(a,
					    &rar5->cstate.ldd, p, &low_dist))
					{
						archive_set_error(&a->archive,
						    ARCHIVE_ERRNO_PROGRAMMER,
						    "Failed to decode the "
						    "distance slot");

						return rar5->main.solid ? ARCHIVE_FATAL : ARCHIVE_FAILED;
					}

					if(dist >= INT_MAX - low_dist - 1) {
						/* This only happens in
						 * invalid archives. */
						archive_set_error(&a->archive,
						    ARCHIVE_ERRNO_FILE_FORMAT,
						    "Distance pointer "
						    "overflow");
						return rar5->main.solid ? ARCHIVE_FATAL : ARCHIVE_FAILED;
					}

					dist += low_dist;
				} else {
					/* dbits is one of [0,1,2,3] */
					int add;

					if(ARCHIVE_OK != (ret = read_consume_bits(a, rar5,
					     p, dbits, &add))) {
						/* Return EOF if we can't read
						 * more data. */
						return ret;
					}

					dist += add;
				}
			}

			if(dist > 0x100) {
				len++;

				if(dist > 0x2000) {
					len++;

					if(dist > 0x40000) {
						len++;
					}
				}
			}

			dist_cache_push(rar5, dist);
			rar5->cstate.last_len = len;

			ret = copy_string(a, len, dist);
			if(ret != ARCHIVE_OK)
				return ret;

			continue;
		} else if(num == 256) {
			/* Create a filter. */
			ret = parse_filter(a, p);
			if(ret != ARCHIVE_OK)
				return ret;

			continue;
		} else if(num == 257) {
			if(rar5->cstate.last_len != 0) {
				ret = copy_string(a,
				    rar5->cstate.last_len,
				    rar5->cstate.dist_cache[0]);
				if(ret != ARCHIVE_OK)
					return ret;
			}

			continue;
		} else {
			/* num < 262 */
			const int idx = num - 258;
			const int dist = dist_cache_touch(rar5, idx);

			uint16_t len_slot;
			int len;

			if(ARCHIVE_OK != decode_number(a, &rar5->cstate.rd, p,
			    &len_slot)) {
				return rar5->main.solid ? ARCHIVE_FATAL : ARCHIVE_FAILED;
			}

			len = decode_code_length(a, rar5, p, len_slot);
			if (len == -1) {
				return rar5->main.solid ? ARCHIVE_FATAL : ARCHIVE_FAILED;
			}

			rar5->cstate.last_len = len;

			ret = copy_string(a, len, dist);
			if(ret != ARCHIVE_OK)
				return ret;

			continue;
		}
	}

	return ARCHIVE_OK;
}

/* Binary search for the RARv5 signature. */
static int scan_for_signature(struct archive_read* a) {
	const uint8_t* p;
	const int chunk_size = 512;
	ssize_t i;
	char signature[sizeof(rar5_signature_xor)];

	/* If we're here, it means we're on an 'unknown territory' data.
	 * There's no indication what kind of data we're reading here.
	 * It could be some text comment, any kind of binary data,
	 * digital sign, dragons, etc.
	 *
	 * We want to find a valid RARv5 magic header inside this unknown
	 * data. */

	/* Is it possible in libarchive to just skip everything until the
	 * end of the file? If so, it would be a better approach than the
	 * current implementation of this function. */

	rar5_signature(signature);

	while(1) {
		if(!read_ahead(a, chunk_size, &p))
			return ARCHIVE_EOF;

		for(i = 0; i < chunk_size - (int)sizeof(rar5_signature_xor);
		    i++) {
			if(memcmp(&p[i], signature,
			    sizeof(rar5_signature_xor)) == 0) {
				/* Consume the number of bytes we've used to
				 * search for the signature, as well as the
				 * number of bytes used by the signature
				 * itself. After this we should be standing
				 * on a valid base block header. */
				(void) consume(a,
				    i + sizeof(rar5_signature_xor));
				return ARCHIVE_OK;
			}
		}

		consume(a, chunk_size);
	}

	return ARCHIVE_FATAL;
}

/* This function will switch the multivolume archive file to another file,
 * i.e. from part03 to part 04. */
static int advance_multivolume(struct archive_read* a) {
	struct rar5 *rar5 = a->format->data;
	int lret;

	/* A small state machine that will skip unnecessary data, needed to
	 * switch from one multivolume to another. Such skipping is needed if
	 * we want to be an stream-oriented (instead of file-oriented)
	 * unpacker.
	 *
	 * The state machine starts with `rar5->main.endarc` == 0. It also
	 * assumes that current stream pointer points to some base block
	 * header.
	 *
	 * The `endarc` field is being set when the base block parsing
	 * function encounters the 'end of archive' marker.
	 */

	while(1) {
		if(rar5->main.endarc == 1) {
			int looping = 1;

			rar5->main.endarc = 0;

			while(looping) {
				lret = skip_base_block(a);
				switch(lret) {
					case ARCHIVE_RETRY:
						/* Continue looping. */
						break;
					case ARCHIVE_OK:
						/* Break loop. */
						looping = 0;
						break;
					default:
						/* Forward any errors to the
						 * caller. */
						return lret;
				}
			}

			break;
		} else {
			/* Skip current base block. In order to properly skip
			 * it, we really need to simply parse it and discard
			 * the results. */

			lret = skip_base_block(a);
			if(lret == ARCHIVE_FATAL || lret == ARCHIVE_FAILED)
				return lret;

			/* The `skip_base_block` function tells us if we
			 * should continue with skipping, or we should stop
			 * skipping. We're trying to skip everything up to
			 * a base FILE block. */

			if(lret != ARCHIVE_RETRY) {
				/* If there was an error during skipping, or we
				 * have just skipped a FILE base block... */

				if(rar5->main.endarc == 0) {
					return lret;
				} else {
					continue;
				}
			}
		}
	}

	return ARCHIVE_OK;
}

/* Merges the partial block from the first multivolume archive file, and
 * partial block from the second multivolume archive file. The result is
 * a chunk of memory containing the whole block, and the stream pointer
 * is advanced to the next block in the second multivolume archive file. */
static int merge_block(struct archive_read* a, ssize_t block_size,
    const uint8_t** p)
{
	struct rar5 *rar5 = a->format->data;
	ssize_t cur_block_size, partial_offset = 0;
	const uint8_t* lp;
	int ret;

	if(rar5->merge_mode) {
		archive_set_error(&a->archive, ARCHIVE_ERRNO_PROGRAMMER,
		    "Recursive merge is not allowed");

		return ARCHIVE_FATAL;
	}

	/* Set a flag that we're in the switching mode. */
	rar5->cstate.switch_multivolume = 1;

	/* Reallocate the memory which will hold the whole block. */
	if(rar5->vol.push_buf)
		free((void*) rar5->vol.push_buf);

	/* Increasing the allocation block by 8 is due to bit reading functions,
	 * which are using additional 2 or 4 bytes. Allocating the block size
	 * by exact value would make bit reader perform reads from invalid
	 * memory block when reading the last byte from the buffer. */
	rar5->vol.push_buf = malloc(block_size + 8);
	if(!rar5->vol.push_buf) {
		archive_set_error(&a->archive, ENOMEM,
		    "Can't allocate memory for a merge block buffer");
		rar5->cstate.switch_multivolume = 0;
		return ARCHIVE_FATAL;
	}

	/* Valgrind complains if the extension block for bit reader is not
	 * initialized, so initialize it. */
	memset(&rar5->vol.push_buf[block_size], 0, 8);

	/* A single block can span across multiple multivolume archive files,
	 * so we use a loop here. This loop will consume enough multivolume
	 * archive files until the whole block is read. */

	while(1) {
		/* Get the size of current block chunk in this multivolume
		 * archive file and read it. */
		cur_block_size = rar5_min(rar5->file.bytes_remaining,
		    block_size - partial_offset);

		if(cur_block_size < 1) {
			/* bytes_remaining is less than 1 at the wrong point in
			 * the merge loop, indicating corrupt volume
			 * accounting. */
			archive_set_error(&a->archive,
			    ARCHIVE_ERRNO_FILE_FORMAT,
			    "Encountered invalid block size during block merge");
			rar5->cstate.switch_multivolume = 0;
			return ARCHIVE_FATAL;
		}

		if(!read_ahead(a, cur_block_size, &lp)) {
			rar5->cstate.switch_multivolume = 0;
			return ARCHIVE_EOF;
		}

		/* Sanity check; there should never be a situation where this
		 * function reads more data than the block's size. */
		if(partial_offset + cur_block_size > block_size) {
			archive_set_error(&a->archive,
			    ARCHIVE_ERRNO_PROGRAMMER,
			    "Consumed too much data when merging blocks");
			rar5->cstate.switch_multivolume = 0;
			return ARCHIVE_FATAL;
		}

		/* Merge previous block chunk with current block chunk,
		 * or create first block chunk if this is our first
		 * iteration. */
		memcpy(&rar5->vol.push_buf[partial_offset], lp, cur_block_size);

		/* Advance the stream read pointer by this block chunk size. */
		if(ARCHIVE_OK != consume(a, cur_block_size)) {
			/* Data was copied but stream pointer didn't advance;
			 * stream position is unrecoverable. */
			rar5->cstate.switch_multivolume = 0;
			return ARCHIVE_FATAL;
		}

		/* Update the pointers. `partial_offset` contains information
		 * about the sum of merged block chunks. */
		partial_offset += cur_block_size;
		rar5->file.bytes_remaining -= cur_block_size;

		/* If `partial_offset` is the same as `block_size`, this means
		 * we've merged all block chunks and we have a valid full
		 * block. */
		if(partial_offset == block_size) {
			break;
		}

		/* If we don't have any bytes to read, this means we should
		 * switch to another multivolume archive file. */
		if(rar5->file.bytes_remaining == 0) {
			rar5->merge_mode++;
			ret = advance_multivolume(a);
			rar5->merge_mode--;
			if(ret != ARCHIVE_OK) {
				rar5->cstate.switch_multivolume = 0;
				return ret;
			}
		}
	}

	*p = rar5->vol.push_buf;

	/* If we're here, we can resume unpacking by processing the block
	 * pointed to by the `*p` memory pointer. */

	return ARCHIVE_OK;
}

static int process_block(struct archive_read* a) {
	struct rar5 *rar5 = a->format->data;
	const uint8_t* p;
	int ret;

	/* If we don't have any data to be processed, this most probably means
	 * we need to switch to the next volume. */
	if(rar5->main.volume && rar5->file.bytes_remaining == 0) {
		ret = advance_multivolume(a);
		if(ret != ARCHIVE_OK)
			return ret;
	}

	if(rar5->cstate.block_parsing_finished) {
		ssize_t block_size;
		ssize_t to_skip;
		ssize_t cur_block_size;

		/* The header size won't be bigger than 6 bytes. */
		if(!read_ahead(a, 6, &p)) {
			/* Failed to prefetch data block header. */
			return ARCHIVE_EOF;
		}

		/*
		 * Read block_size by parsing block header. Validate the header
		 * by calculating CRC byte stored inside the header. Size of
		 * the header is not constant (block size can be stored either
		 * in 1 or 2 bytes), that's why block size is left out from the
		 * `compressed_block_header` structure and returned by
		 * `parse_block_header` as the second argument. */

		ret = parse_block_header(a, p, &block_size,
		    &rar5->last_block_hdr);
		if(ret != ARCHIVE_OK) {
			return ret;
		}

		/* Skip block header. Next data is huffman tables,
		 * if present. */
		to_skip = sizeof(struct compressed_block_header) +
			bf_byte_count(&rar5->last_block_hdr) + 1;

		/* If the block header's to_skip value exceeds the declared
		 * remaining data, the archive is malformed. */
		if(to_skip > rar5->file.bytes_remaining) {
			archive_set_error(&a->archive,
			    ARCHIVE_ERRNO_FILE_FORMAT,
			    "Block header size exceeds remaining file data");
			return ARCHIVE_FATAL;
		}

		if(ARCHIVE_OK != consume(a, to_skip))
			return ARCHIVE_EOF;

		rar5->file.bytes_remaining -= to_skip;

		/* The block size gives information about the whole block size,
		 * but the block could be stored in split form when using
		 * multi-volume archives. In this case, the block size will be
		 * bigger than the actual data stored in this file. Remaining
		 * part of the data will be in another file. */

		cur_block_size =
			rar5_min(rar5->file.bytes_remaining, block_size);

		if(block_size > rar5->file.bytes_remaining) {
			/* If current blocks' size is bigger than our data
			 * size, this means we have a multivolume archive.
			 * In this case, skip all base headers until the end
			 * of the file, proceed to next "partXXX.rar" volume,
			 * find its signature, skip all headers up to the first
			 * FILE base header, and continue from there.
			 *
			 * Note that `merge_block` will update the `rar`
			 * context structure quite extensively. */

			ret = merge_block(a, block_size, &p);
			if(ret != ARCHIVE_OK) {
				return ret;
			}

			cur_block_size = block_size;

			/* Current stream pointer should be now directly
			 * *after* the block that spanned through multiple
			 * archive files. `p` pointer should have the data of
			 * the *whole* block (merged from partial blocks
			 * stored in multiple archives files). */
		} else {
			rar5->cstate.switch_multivolume = 0;

			/* Read the whole block size into memory. This can take
			 * up to  8 megabytes of memory in theoretical cases.
			 * Might be worth to optimize this and use a standard
			 * chunk of 4kb's. */
			if(!read_ahead(a, 4 + cur_block_size, &p)) {
				/* Failed to prefetch block data. */
				return ARCHIVE_EOF;
			}
		}

		rar5->cstate.block_buf = p;
		rar5->cstate.cur_block_size = cur_block_size;
		rar5->cstate.block_parsing_finished = 0;

		rar5->bits.in_addr = 0;
		rar5->bits.bit_addr = 0;

		if(bf_is_table_present(&rar5->last_block_hdr)) {
			/* Load Huffman tables. */
			ret = parse_tables(a, rar5, p);
			if(ret != ARCHIVE_OK) {
				/* Error during decompression of Huffman
				 * tables. */
				return ret;
			}
		}
	} else {
		/* Block parsing not finished, reuse previous memory buffer. */
		p = rar5->cstate.block_buf;
	}

	/* Uncompress the block, or a part of it, depending on how many bytes
	 * will be generated by uncompressing the block.
	 *
	 * In case too many bytes will be generated, calling this function
	 * again will resume the uncompression operation. */
	ret = do_uncompress_block(a, p);
	if(ret != ARCHIVE_OK) {
		return ret;
	}

	if(rar5->cstate.block_parsing_finished &&
	    rar5->cstate.switch_multivolume == 0 &&
	    rar5->cstate.cur_block_size > 0)
	{
		/* If we're processing a normal block, consume the whole
		 * block. We can do this because we've already read the whole
		 * block to memory. */
		if(ARCHIVE_OK != consume(a, rar5->cstate.cur_block_size))
			return ARCHIVE_FATAL;

		rar5->file.bytes_remaining -= rar5->cstate.cur_block_size;
	} else if(rar5->cstate.switch_multivolume) {
		/* Don't consume the block if we're doing multivolume
		 * processing. The volume switching function will consume
		 * the proper count of bytes instead. */
		rar5->cstate.switch_multivolume = 0;
	}

	return ARCHIVE_OK;
}

/* Pops the `buf`, `size` and `offset` from the "data ready" stack.
 *
 * Returns ARCHIVE_OK when those arguments can be used, ARCHIVE_RETRY
 * when there is no data on the stack. */
static int use_data(struct rar5 *rar5, const void** buf, size_t* size,
    int64_t* offset)
{
	int i;

	for(i = 0; i < rar5_countof(rar5->cstate.dready); i++) {
		struct data_ready *d = &rar5->cstate.dready[i];

		if(d->used) {
			if(buf)    *buf = d->buf;
			if(size)   *size = d->size;
			if(offset) *offset = d->offset;

			d->used = 0;
			return ARCHIVE_OK;
		}
	}

	return ARCHIVE_RETRY;
}

static void clear_data_ready_stack(struct rar5 *rar5) {
	memset(&rar5->cstate.dready, 0, sizeof(rar5->cstate.dready));
}

/* Pushes the `buf`, `size` and `offset` arguments to the rar5->cstate.dready
 * FIFO stack. Those values will be popped from this stack by the `use_data`
 * function. */
static int push_data_ready(struct archive_read* a, struct rar5 *rar5,
    const uint8_t* buf, size_t size, int64_t offset)
{
	int i;

	/* Don't push if we're in skip mode. This is needed because solid
	 * streams need full processing even if we're skipping data. After
	 * fully processing the stream, we need to discard the generated bytes,
	 * because we're interested only in the side effect: building up the
	 * internal window circular buffer. This window buffer will be used
	 * later during unpacking of requested data. */
	if(rar5->skip_mode)
		return ARCHIVE_OK;

	/* Sanity check. */
	if(offset != rar5->file.last_offset + rar5->file.last_size) {
		archive_set_error(&a->archive, ARCHIVE_ERRNO_PROGRAMMER,
		    "Sanity check error: output stream is not continuous");
		return ARCHIVE_FATAL;
	}

	for(i = 0; i < rar5_countof(rar5->cstate.dready); i++) {
		struct data_ready* d = &rar5->cstate.dready[i];
		if(!d->used) {
			d->used = 1;
			d->buf = buf;
			d->size = size;
			d->offset = offset;

			/* These fields are used only in sanity checking. */
			rar5->file.last_offset = offset;
			rar5->file.last_size = size;

			/* Calculate the checksum of this new block before
			 * submitting data to libarchive's engine. */
			update_crc(rar5, d->buf, d->size);

			return ARCHIVE_OK;
		}
	}

	/* Program counter will reach this code if the
	 * `rar5->cstate.data_ready` stack will be filled up so that no new
	 * entries will be allowed. The code shouldn't allow such situation to
	 * occur. So we treat this case as an internal error. */

	archive_set_error(&a->archive, ARCHIVE_ERRNO_PROGRAMMER,
	    "Premature end of data_ready stack");
	return ARCHIVE_FATAL;
}

/* This function uncompresses the data that is stored in the <FILE> base
 * block.
 *
 * The FILE base block looks like this:
 *
 * <header><huffman tables><block_1><block_2>...<block_n>
 *
 * The <header> is a block header, that is parsed in parse_block_header().
 * It's a "compressed_block_header" structure, containing metadata needed
 * to know when we should stop looking for more <block_n> blocks.
 *
 * <huffman tables> contain data needed to set up the huffman tables, needed
 * for the actual decompression.
 *
 * Each <block_n> consists of series of literals:
 *
 * <literal><literal><literal>...<literal>
 *
 * Those literals generate the uncompression data. They operate on a circular
 * buffer, sometimes writing raw data into it, sometimes referencing
 * some previous data inside this buffer, and sometimes declaring a filter
 * that will need to be executed on the data stored in the circular buffer.
 * It all depends on the literal that is used.
 *
 * Sometimes blocks produce output data, sometimes they don't. For example, for
 * some huge files that use lots of filters, sometimes a block is filled with
 * only filter declaration literals. Such blocks won't produce any data in the
 * circular buffer.
 *
 * Sometimes blocks will produce 4 bytes of data, and sometimes 1 megabyte,
 * because a literal can reference previously decompressed data. For example,
 * there can be a literal that says: 'append a byte 0xFE here', and after
 * it another literal can say 'append 1 megabyte of data from circular buffer
 * offset 0x12345'. This is how RAR format handles compressing repeated
 * patterns.
 *
 * The RAR compressor creates those literals and the actual efficiency of
 * compression depends on what those literals are. The literals can also
 * be seen as a kind of a non-turing-complete virtual machine that simply
 * tells the decompressor what it should do.
 * */

static int do_uncompress_file(struct archive_read* a) {
	struct rar5 *rar5 = a->format->data;
	int ret;
	int64_t max_end_pos;

	if(!rar5->cstate.initialized) {
		/* Don't perform full context reinitialization if we're
		 * processing a solid archive. */
		if(!rar5->main.solid || !rar5->cstate.window_buf) {
			if((ret = init_unpack(rar5)) != ARCHIVE_OK)
				return ret;
		}

		rar5->cstate.initialized = 1;
	}

	/* Don't allow extraction if window_size is invalid. */
	if(rar5->cstate.window_size == 0) {
		archive_set_error(&a->archive,
			ARCHIVE_ERRNO_FILE_FORMAT,
			"Invalid window size declaration in this file");

		/* This should never happen in valid files. */
		return ARCHIVE_FAILED;
	}

	if(rar5->cstate.all_filters_applied == 1) {
		/* We use while(1) here, but standard case allows for just 1
		 * iteration. The loop will iterate if process_block() didn't
		 * generate any data at all. This can happen if the block
		 * contains only filter definitions (this is common in big
		 * files). */
		while(1) {
			ret = process_block(a);
			if(ret != ARCHIVE_OK)
				return ret;

			if(rar5->cstate.last_write_ptr ==
			    rar5->cstate.write_ptr) {
				/* The block didn't generate any new data,
				 * so just process a new block if this one
				 * wasn't the last block in the file. */
				if (bf_is_last_block(&rar5->last_block_hdr)) {
					return ARCHIVE_EOF;
				}

				continue;
			}

			/* The block has generated some new data, so break
			 * the loop. */
			break;
		}
	}

	/* Try to run filters. If filters won't be applied, it means that
	 * insufficient data was generated. */
	ret = apply_filters(a);
	if(ret == ARCHIVE_RETRY) {
		return ARCHIVE_OK;
	} else if(ret != ARCHIVE_OK) {
		return ret;
	}

	if(cdeque_size(&rar5->cstate.filters) > 0) {
		/* Check if we can write something before hitting first
		 * filter. */
		struct filter_info* flt;

		/* Get the block_start offset from the first filter. */
		if(CDE_OK != cdeque_front(&rar5->cstate.filters,
		    cdeque_filter_p(&flt)))
		{
			archive_set_error(&a->archive,
			    ARCHIVE_ERRNO_PROGRAMMER,
			    "Can't read first filter");
			return ARCHIVE_FATAL;
		}

		max_end_pos = rar5_min(flt->block_start,
		    rar5->cstate.write_ptr);
	} else {
		/* There are no filters defined, or all filters were applied.
		 * This means we can just store the data without any
		 * postprocessing. */
		max_end_pos = rar5->cstate.write_ptr;
	}

	if(max_end_pos == rar5->cstate.last_write_ptr) {
		/* We can't write anything yet. The block uncompression
		 * function did not generate enough data, and no filter can be
		 * applied. At the same time we don't have any data that can be
		 *  stored without filter postprocessing. This means we need to
		 *  wait for more data to be generated, so we can apply the
		 * filters.
		 *
		 * Signal the caller that we need more data to be able to do
		 * anything.
		 */
		return ARCHIVE_RETRY;
	} else {
		/* We can write the data before hitting the first filter.
		 * So let's do it. The push_window_data() function will
		 * effectively return the selected data block to the user
		 * application. */
		push_window_data(a, rar5, rar5->cstate.last_write_ptr,
		    max_end_pos);
		rar5->cstate.last_write_ptr = max_end_pos;
	}

	return ARCHIVE_OK;
}

static int uncompress_file(struct archive_read* a) {
	int ret;

	while(1) {
		/* Sometimes the uncompression function will return a
		 * 'retry' signal. If this will happen, we have to retry
		 * the function. */
		ret = do_uncompress_file(a);
		if(ret != ARCHIVE_RETRY)
			return ret;
	}
}


static int do_unstore_file(struct archive_read* a,
    struct rar5 *rar5, const void** buf, size_t* size, int64_t* offset)
{
	size_t to_read;
	const uint8_t* p;

	if(rar5->file.bytes_remaining == 0 && rar5->main.volume > 0 &&
	    rar5->generic.split_after > 0)
	{
		int ret;

		rar5->cstate.switch_multivolume = 1;
		ret = advance_multivolume(a);
		rar5->cstate.switch_multivolume = 0;

		if(ret != ARCHIVE_OK) {
			/* Failed to advance to next multivolume archive
			 * file. */
			return ret;
		}
	}

	to_read = rar5_min(rar5->file.bytes_remaining, 64 * 1024);
	if(to_read == 0) {
		return ARCHIVE_EOF;
	}

	if(!read_ahead(a, to_read, &p)) {
		archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
		    "I/O error when unstoring file");
		return ARCHIVE_FATAL;
	}

	if(ARCHIVE_OK != consume(a, to_read)) {
		return ARCHIVE_EOF;
	}

	if(buf)    *buf = p;
	if(size)   *size = to_read;
	if(offset) *offset = rar5->cstate.last_unstore_ptr;

	rar5->file.bytes_remaining -= to_read;
	rar5->cstate.last_unstore_ptr += to_read;

	update_crc(rar5, p, to_read);
	return ARCHIVE_OK;
}

static int do_unpack(struct archive_read* a, struct rar5 *rar5,
    const void** buf, size_t* size, int64_t* offset)
{
	enum COMPRESSION_METHOD {
		STORE = 0, FASTEST = 1, FAST = 2, NORMAL = 3, GOOD = 4,
		BEST = 5
	};

	if(rar5->file.service > 0) {
		return do_unstore_file(a, rar5, buf, size, offset);
	} else {
		switch(rar5->cstate.method) {
			case STORE:
				return do_unstore_file(a, rar5, buf, size,
				    offset);
			case FASTEST:
				/* fallthrough */
			case FAST:
				/* fallthrough */
			case NORMAL:
				/* fallthrough */
			case GOOD:
				/* fallthrough */
			case BEST:
				/* No data is returned here. But because a sparse-file aware
				 * caller (like archive_read_data_into_fd) may treat zero-size
				 * as a sparse file block, we need to update the offset
				 * accordingly. At this point the decoder doesn't have any
				 * pending uncompressed data blocks, so the current position in
				 * the output file should be last_write_ptr. */
				if (offset) *offset = rar5->cstate.last_write_ptr;
				return uncompress_file(a);
			default:
				archive_set_error(&a->archive,
				    ARCHIVE_ERRNO_FILE_FORMAT,
				    "Compression method not supported: 0x%x",
				    (unsigned int)rar5->cstate.method);

				return ARCHIVE_FATAL;
		}
	}

#if !defined WIN32
	/* Not reached. */
	return ARCHIVE_OK;
#endif
}

static int verify_checksums(struct archive_read* a) {
	struct rar5 *rar5 = a->format->data;
	int verify_crc;

	/* Check checksums only when actually unpacking the data. There's no
	 * need to calculate checksum when we're skipping data in solid archives
	 * (skipping in solid archives is the same thing as unpacking compressed
	 * data and discarding the result). */

	if(!rar5->skip_mode) {
		/* Always check checksums if we're not in skip mode */
		verify_crc = 1;
	} else {
		/* We can override the logic above with a compile-time option
		 * NO_CRC_ON_SOLID_SKIP. This option is used during debugging,
		 * and it will check checksums of unpacked data even when
		 * we're skipping it. */

#if defined CHECK_CRC_ON_SOLID_SKIP
		/* Debug case */
		verify_crc = 1;
#else
		/* Normal case */
		verify_crc = 0;
#endif
	}

	if(verify_crc) {
		/* During unpacking, on each unpacked block we're calling the
		 * update_crc() function. Since we are here, the unpacking
		 * process is already over and we can check if calculated
		 * checksum (CRC32 or BLAKE2sp) is the same as what is stored
		 * in the archive. */
		if(rar5->file.stored_crc32 > 0) {
			/* Check CRC32 only when the file contains a CRC32
			 * value for this file. */

			if(rar5->file.calculated_crc32 !=
			    rar5->file.stored_crc32) {
				/* Checksums do not match; the unpacked file
				 * is corrupted. */

				DEBUG_CODE {
					printf("Checksum error: CRC32 "
					    "(was: %08" PRIx32 ", expected: %08" PRIx32 ")\n",
					    rar5->file.calculated_crc32,
					    rar5->file.stored_crc32);
				}

#ifndef DONT_FAIL_ON_CRC_ERROR
				archive_set_error(&a->archive,
				    ARCHIVE_ERRNO_FILE_FORMAT,
				    "Checksum error: CRC32");
				return ARCHIVE_FAILED;
#endif
			} else {
				DEBUG_CODE {
					printf("Checksum OK: CRC32 "
					    "(%08" PRIx32 "/%08" PRIx32 ")\n",
					    rar5->file.stored_crc32,
					    rar5->file.calculated_crc32);
				}
			}
		}

		if(rar5->file.has_blake2 > 0) {
			/* BLAKE2sp is an optional checksum algorithm that is
			 * added to RARv5 archives when using the `-htb` switch
			 *  during creation of archive.
			 *
			 * We now finalize the hash calculation by calling the
			 * `final` function. This will generate the final hash
			 * value we can use to compare it with the BLAKE2sp
			 * checksum that is stored in the archive.
			 *
			 * The return value of this `final` function is not
			 * very helpful, as it guards only against improper use.
 			 * This is why we're explicitly ignoring it. */

			uint8_t b2_buf[32];
			(void) blake2sp_final(&rar5->file.b2state, b2_buf, 32);

			if(memcmp(&rar5->file.blake2sp, b2_buf, 32) != 0) {
#ifndef DONT_FAIL_ON_CRC_ERROR
				archive_set_error(&a->archive,
				    ARCHIVE_ERRNO_FILE_FORMAT,
				    "Checksum error: BLAKE2");

				return ARCHIVE_FAILED;
#endif
			}
		}
	}

	/* Finalization for this file has been successfully completed. */
	return ARCHIVE_OK;
}

static int verify_global_checksums(struct archive_read* a) {
	return verify_checksums(a);
}

/*
 * Decryption function for the magic signature pattern. Check the comment near
 * the `rar5_signature_xor` symbol to read the rationale behind this.
 */
static void rar5_signature(char *buf) {
		size_t i;

		for(i = 0; i < sizeof(rar5_signature_xor); i++) {
			buf[i] = rar5_signature_xor[i] ^ 0xA1;
		}
}

static int rar5_read_data(struct archive_read *a, const void **buff,
    size_t *size, int64_t *offset) {
	struct rar5 *rar5 = a->format->data;
	int ret;

	if (size)
		*size = 0;

	if (rar5->has_encrypted_entries == ARCHIVE_READ_FORMAT_ENCRYPTION_DONT_KNOW) {
		rar5->has_encrypted_entries = 0;
	}

	if (rar5->headers_are_encrypted || rar5->cstate.data_encrypted) {
		archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
		    "Reading encrypted data is not currently supported");
		return ARCHIVE_FAILED;
	}

	if(rar5->file.dir > 0) {
		/* Don't process any data if this file entry was declared
		 * as a directory. This is needed, because entries marked as
		 * directory doesn't have any dictionary buffer allocated, so
		 * it's impossible to perform any decompression. */
		archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
		    "Can't decompress an entry marked as a directory");
		return ARCHIVE_FATAL;
	}

	if(!rar5->skip_mode && (rar5->cstate.last_write_ptr > rar5->file.unpacked_size)) {
		archive_set_error(&a->archive, ARCHIVE_ERRNO_PROGRAMMER,
		    "Unpacker has written too many bytes");
		return ARCHIVE_FATAL;
	}

	ret = use_data(rar5, buff, size, offset);
	if(ret == ARCHIVE_OK) {
		return ret;
	}

	if(rar5->file.eof == 1) {
		return ARCHIVE_EOF;
	}

	ret = do_unpack(a, rar5, buff, size, offset);
	if(ret != ARCHIVE_OK) {
		return ret;
	}

	if(rar5->file.bytes_remaining == 0 &&
			rar5->cstate.last_write_ptr == rar5->file.unpacked_size)
	{
		/* If all bytes of current file were processed, run
		 * finalization.
		 *
		 * Finalization will check checksum against proper values. If
		 * some of the checksums will not match, we'll return an error
		 * value in the last `archive_read_data` call to signal an error
		 * to the user. */

		rar5->file.eof = 1;
		return verify_global_checksums(a);
	}

	return ARCHIVE_OK;
}

static int rar5_read_data_skip(struct archive_read *a) {
	struct rar5 *rar5 = a->format->data;

	if(rar5->main.solid && (rar5->cstate.data_encrypted == 0)) {
		/* In solid archives, instead of skipping the data, we need to
		 * extract it, and dispose the result. The side effect of this
		 * operation will be setting up the initial window buffer state
		 * needed to be able to extract the selected file. Note that
		 * this is only possible when data within this solid block is
		 * not encrypted, in which case we'll skip and fail if the user
		 * tries to read data. */

		int ret;

		/* Make sure to process all blocks in the compressed stream. */
		while(rar5->file.bytes_remaining > 0) {
			/* Setting the "skip mode" will allow us to skip
			 * checksum checks during data skipping. Checking the
			 * checksum of skipped data isn't really necessary and
			 * it's only slowing things down.
			 *
			 * This is incremented instead of setting to 1 because
			 * this data skipping function can be called
			 * recursively. */
			rar5->skip_mode++;

			/* We're disposing 1 block of data, so we use triple
			 * NULLs in arguments. */
			ret = rar5_read_data(a, NULL, NULL, NULL);

			/* Turn off "skip mode". */
			rar5->skip_mode--;

			if(ret < 0 || ret == ARCHIVE_EOF) {
				/* Propagate any potential error conditions
				 * to the caller. */
				return ret;
			}
		}
	} else {
		/* In standard archives, we can just jump over the compressed
		 * stream. Each file in non-solid archives starts from an empty
		 * window buffer. */

		if(ARCHIVE_OK != consume(a, rar5->file.bytes_remaining)) {
			return ARCHIVE_FATAL;
		}

		rar5->file.bytes_remaining = 0;
	}

	return ARCHIVE_OK;
}

static int64_t rar5_seek_data(struct archive_read *a, int64_t offset,
    int whence)
{
	(void) offset;
	(void) whence;

	/* We're a streaming unpacker, and we don't support seeking.
	 * That's a capability gap, not a fatal error. */
	archive_set_error(&a->archive, ARCHIVE_ERRNO_MISC,
	    "Seeking of RAR5 files is unsupported");

	return (ARCHIVE_FAILED);
}

static int rar5_cleanup(struct archive_read *a) {
	struct rar5 *rar5 = a->format->data;

	free(rar5->cstate.window_buf);
	free(rar5->cstate.filtered_buf);
	clear_data_ready_stack(rar5);

	free(rar5->vol.push_buf);

	free_filters(rar5);
	rar5_deinit(rar5);

	free(rar5);
	a->format->data = NULL;

	return ARCHIVE_OK;
}

static int rar5_capabilities(struct archive_read * a) {
	(void) a;
	return (ARCHIVE_READ_FORMAT_CAPS_ENCRYPT_DATA
			| ARCHIVE_READ_FORMAT_CAPS_ENCRYPT_METADATA);
}

static int rar5_has_encrypted_entries(struct archive_read *_a) {
	if (_a && _a->format) {
		struct rar5 *rar5 = _a->format->data;
		if (rar5) {
			return rar5->has_encrypted_entries;
		}
	}

	return ARCHIVE_READ_FORMAT_ENCRYPTION_DONT_KNOW;
}

/* Must match deallocations in rar5_deinit */
static int rar5_init(struct rar5 *rar5) {
	memset(rar5, 0, sizeof(struct rar5));

	if(CDE_OK != cdeque_init(&rar5->cstate.filters, 8192))
		return ARCHIVE_FATAL;

	/*
	 * Until enough data has been read, we cannot tell about
	 * any encrypted entries yet.
	 */
	rar5->has_encrypted_entries = ARCHIVE_READ_FORMAT_ENCRYPTION_DONT_KNOW;

	return ARCHIVE_OK;
}

/* Must match allocations in rar5_init */
static void rar5_deinit(struct rar5 *rar5) {
	cdeque_free(&rar5->cstate.filters);
}

int archive_read_support_format_rar5(struct archive *_a) {
	struct archive_read* ar;
	int ret;
	struct rar5 *rar5;

	if(ARCHIVE_OK != (ret = get_archive_read(_a, &ar)))
		return ret;

	rar5 = malloc(sizeof(*rar5));
	if(rar5 == NULL) {
		archive_set_error(&ar->archive, ENOMEM,
		    "Can't allocate rar5 data");
		return ARCHIVE_FATAL;
	}

	if(ARCHIVE_OK != rar5_init(rar5)) {
		archive_set_error(&ar->archive, ENOMEM,
		    "Can't allocate rar5 filter buffer");
		free(rar5);
		return ARCHIVE_FATAL;
	}

	ret = __archive_read_register_format(ar,
	    rar5,
	    "rar5",
	    rar5_bid,
	    rar5_options,
	    rar5_read_header,
	    rar5_read_data,
	    rar5_read_data_skip,
	    rar5_seek_data,
	    rar5_cleanup,
	    rar5_capabilities,
	    rar5_has_encrypted_entries);

	if(ret != ARCHIVE_OK) {
		rar5_deinit(rar5);
		free(rar5);
	}

	return ARCHIVE_OK;
}