2 * Copyright (c) 2018 Grzegorz Antoniak (http://antoniak.org)
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR(S) ``AS IS'' AND ANY EXPRESS OR
15 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
16 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
17 * IN NO EVENT SHALL THE AUTHOR(S) BE LIABLE FOR ANY DIRECT, INDIRECT,
18 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
19 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
20 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
21 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
22 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
23 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
26 #include "archive_platform.h"
33 #include <zlib.h> /* crc32 */
38 #include "archive_crc32.h"
41 #include "archive_entry.h"
42 #include "archive_entry_locale.h"
43 #include "archive_ppmd7_private.h"
44 #include "archive_entry_private.h"
49 #include "archive_blake2.h"
52 /*#define CHECK_CRC_ON_SOLID_SKIP*/
53 /*#define DONT_FAIL_ON_CRC_ERROR*/
56 #define rar5_min(a, b) (((a) > (b)) ? (b) : (a))
57 #define rar5_max(a, b) (((a) > (b)) ? (a) : (b))
58 #define rar5_countof(X) ((const ssize_t) (sizeof(X) / sizeof(*X)))
61 #define DEBUG_CODE if(1)
63 #define DEBUG_CODE if(0)
66 /* Real RAR5 magic number is:
68 * 0x52, 0x61, 0x72, 0x21, 0x1a, 0x07, 0x01, 0x00
71 * It's stored in `rar5_signature` after XOR'ing it with 0xA1, because I don't
72 * want to put this magic sequence in each binary that uses libarchive, so
73 * applications that scan through the file for this marker won't trigger on
76 * The array itself is decrypted in `rar5_init` function. */
78 static unsigned char rar5_signature[] = { 243, 192, 211, 128, 187, 166, 160, 161 };
79 static const ssize_t rar5_signature_size = sizeof(rar5_signature);
80 /* static const size_t g_unpack_buf_chunk_size = 1024; */
81 static const size_t g_unpack_window_size = 0x20000;
84 ssize_t bytes_remaining;
85 ssize_t unpacked_size;
86 int64_t last_offset; /* Used in sanity checks. */
87 int64_t last_size; /* Used in sanity checks. */
89 uint8_t solid : 1; /* Is this a solid stream? */
90 uint8_t service : 1; /* Is this file a service data? */
92 /* Optional time fields. */
98 /* Optional hash fields. */
99 uint32_t stored_crc32;
100 uint32_t calculated_crc32;
101 uint8_t blake2sp[32];
102 blake2sp_state b2state;
107 FILTER_DELTA = 0, /* Generic pattern. */
108 FILTER_E8 = 1, /* Intel x86 code. */
109 FILTER_E8E9 = 2, /* Intel x86 code. */
110 FILTER_ARM = 3, /* ARM code. */
111 FILTER_AUDIO = 4, /* Audio filter, not used in RARv5. */
112 FILTER_RGB = 5, /* Color palette, not used in RARv5. */
113 FILTER_ITANIUM = 6, /* Intel's Itanium, not used in RARv5. */
114 FILTER_PPM = 7, /* Predictive pattern matching, not used in RARv5. */
124 ssize_t block_length;
143 struct decode_table {
145 int32_t decode_len[16];
146 uint32_t decode_pos[16];
148 uint8_t quick_len[1 << 10];
149 uint16_t quick_num[1 << 10];
150 uint16_t decode_num[306];
154 /* Flag used to specify if unpacker needs to reinitialize the uncompression
156 uint8_t initialized : 1;
158 /* Flag used when applying filters. */
159 uint8_t all_filters_applied : 1;
161 /* Flag used to skip file context reinitialization, used when unpacker is
162 * skipping through different multivolume archives. */
163 uint8_t switch_multivolume : 1;
165 /* Flag used to specify if unpacker has processed the whole data block or
166 * just a part of it. */
167 uint8_t block_parsing_finished : 1;
171 int flags; /* Uncompression flags. */
172 int method; /* Uncompression algorithm method. */
173 int version; /* Uncompression algorithm version. */
174 ssize_t window_size; /* Size of window_buf. */
175 uint8_t* window_buf; /* Circular buffer used during
177 uint8_t* filtered_buf; /* Buffer used when applying filters. */
178 const uint8_t* block_buf; /* Buffer used when merging blocks. */
179 size_t window_mask; /* Convinience field; window_size - 1. */
180 int64_t write_ptr; /* This amount of data has been unpacked in
181 the window buffer. */
182 int64_t last_write_ptr; /* This amount of data has been stored in
184 int64_t last_unstore_ptr; /* Counter of bytes extracted during
185 unstoring. This is separate from
186 last_write_ptr because of how SERVICE
187 base blocks are handled during skipping
188 in solid multiarchive archives. */
189 int64_t solid_offset; /* Additional offset inside the window
190 buffer, used in unpacking solid
192 ssize_t cur_block_size; /* Size of current data block. */
193 int last_len; /* Flag used in lzss decompression. */
195 /* Decode tables used during lzss uncompression. */
198 struct decode_table bd; /* huffman bit lengths */
200 struct decode_table ld; /* literals */
202 struct decode_table dd; /* distances */
204 struct decode_table ldd; /* lower bits of distances */
206 struct decode_table rd; /* repeating distances */
207 #define HUFF_TABLE_SIZE (HUFF_NC + HUFF_DC + HUFF_RC + HUFF_LDC)
209 /* Circular deque for storing filters. */
210 struct cdeque filters;
211 int64_t last_block_start; /* Used for sanity checking. */
212 ssize_t last_block_length; /* Used for sanity checking. */
214 /* Distance cache used during lzss uncompression. */
217 /* Data buffer stack. */
218 struct data_ready dready[2];
221 /* Bit reader state. */
223 int8_t bit_addr; /* Current bit pointer inside current byte. */
224 int in_addr; /* Current byte pointer. */
227 /* RARv5 block header structure. */
228 struct compressed_block_header {
231 uint8_t bit_size : 3;
232 uint8_t byte_count : 3;
233 uint8_t is_last_block : 1;
234 uint8_t is_table_present : 1;
236 uint8_t block_flags_u8;
242 /* RARv5 main header structure. */
244 /* Does the archive contain solid streams? */
247 /* If this a multi-file archive? */
255 struct generic_header {
256 uint8_t split_after : 1;
257 uint8_t split_before : 1;
268 /* Main context structure. */
270 int header_initialized;
272 /* Set to 1 if current file is positioned AFTER the magic value
273 * of the archive file. This is used in header reading functions. */
276 /* Set to not zero if we're in skip mode (either by calling rar5_data_skip
277 * function or when skipping over solid streams). Set to 0 when in
278 * extraction mode. This is used during checksum calculation functions. */
281 /* An offset to QuickOpen list. This is not supported by this unpacker,
282 * becuase we're focusing on streaming interface. QuickOpen is designed
283 * to make things quicker for non-stream interfaces, so it's not our
285 uint64_t qlist_offset;
287 /* An offset to additional Recovery data. This is not supported by this
288 * unpacker. Recovery data are additional Reed-Solomon codes that could
289 * be used to calculate bytes that are missing in archive or are
293 /* Various context variables grouped to different structures. */
294 struct generic_header generic;
295 struct main_header main;
296 struct comp_state cstate;
297 struct file_header file;
298 struct bit_reader bits;
299 struct multivolume vol;
301 /* The header of currently processed RARv5 block. Used in main
302 * decompression logic loop. */
303 struct compressed_block_header last_block_hdr;
306 /* Forward function declarations. */
308 static int verify_global_checksums(struct archive_read* a);
309 static int rar5_read_data_skip(struct archive_read *a);
310 static int push_data_ready(struct archive_read* a, struct rar5* rar,
311 const uint8_t* buf, size_t size, int64_t offset);
313 /* CDE_xxx = Circular Double Ended (Queue) return values. */
314 enum CDE_RETURN_VALUES {
315 CDE_OK, CDE_ALLOC, CDE_PARAM, CDE_OUT_OF_BOUNDS,
318 /* Clears the contents of this circular deque. */
319 static void cdeque_clear(struct cdeque* d) {
325 /* Creates a new circular deque object. Capacity must be power of 2: 8, 16, 32,
326 * 64, 256, etc. When the user will add another item above current capacity,
327 * the circular deque will overwrite the oldest entry. */
328 static int cdeque_init(struct cdeque* d, int max_capacity_power_of_2) {
329 if(d == NULL || max_capacity_power_of_2 == 0)
332 d->cap_mask = max_capacity_power_of_2 - 1;
335 if((max_capacity_power_of_2 & d->cap_mask) > 0)
339 d->arr = malloc(sizeof(void*) * max_capacity_power_of_2);
341 return d->arr ? CDE_OK : CDE_ALLOC;
344 /* Return the current size (not capacity) of circular deque `d`. */
345 static size_t cdeque_size(struct cdeque* d) {
349 /* Returns the first element of current circular deque. Note that this function
350 * doesn't perform any bounds checking. If you need bounds checking, use
351 * `cdeque_front()` function instead. */
352 static void cdeque_front_fast(struct cdeque* d, void** value) {
353 *value = (void*) d->arr[d->beg_pos];
356 /* Returns the first element of current circular deque. This function
357 * performs bounds checking. */
358 static int cdeque_front(struct cdeque* d, void** value) {
360 cdeque_front_fast(d, value);
363 return CDE_OUT_OF_BOUNDS;
366 /* Pushes a new element into the end of this circular deque object. If current
367 * size will exceed capacity, the oldest element will be overwritten. */
368 static int cdeque_push_back(struct cdeque* d, void* item) {
372 if(d->size == d->cap_mask + 1)
373 return CDE_OUT_OF_BOUNDS;
375 d->arr[d->end_pos] = (size_t) item;
376 d->end_pos = (d->end_pos + 1) & d->cap_mask;
382 /* Pops a front element of this circular deque object and returns its value.
383 * This function doesn't perform any bounds checking. */
384 static void cdeque_pop_front_fast(struct cdeque* d, void** value) {
385 *value = (void*) d->arr[d->beg_pos];
386 d->beg_pos = (d->beg_pos + 1) & d->cap_mask;
390 /* Pops a front element of this cicrular deque object and returns its value.
391 * This function performs bounds checking. */
392 static int cdeque_pop_front(struct cdeque* d, void** value) {
397 return CDE_OUT_OF_BOUNDS;
399 cdeque_pop_front_fast(d, value);
403 /* Convinience function to cast filter_info** to void **. */
404 static void** cdeque_filter_p(struct filter_info** f) {
405 return (void**) (size_t) f;
408 /* Convinience function to cast filter_info* to void *. */
409 static void* cdeque_filter(struct filter_info* f) {
410 return (void**) (size_t) f;
413 /* Destroys this circular deque object. Dellocates the memory of the collection
414 * buffer, but doesn't deallocate the memory of any pointer passed to this
415 * deque as a value. */
416 static void cdeque_free(struct cdeque* d) {
431 static inline struct rar5* get_context(struct archive_read* a) {
432 return (struct rar5*) a->format->data;
435 // TODO: make sure these functions return a little endian number
437 /* Convinience functions used by filter implementations. */
439 static uint32_t read_filter_data(struct rar5* rar, uint32_t offset) {
440 uint32_t* dptr = (uint32_t*) &rar->cstate.window_buf[offset];
441 // TODO: bswap if big endian
445 static void write_filter_data(struct rar5* rar, uint32_t offset,
448 uint32_t* dptr = (uint32_t*) &rar->cstate.filtered_buf[offset];
449 // TODO: bswap if big endian
453 static void circular_memcpy(uint8_t* dst, uint8_t* window, const int mask,
454 int64_t start, int64_t end)
456 if((start & mask) > (end & mask)) {
457 ssize_t len1 = mask + 1 - (start & mask);
458 ssize_t len2 = end & mask;
460 memcpy(dst, &window[start & mask], len1);
461 memcpy(dst + len1, window, len2);
463 memcpy(dst, &window[start & mask], (size_t) (end - start));
467 /* Allocates a new filter descriptor and adds it to the filter array. */
468 static struct filter_info* add_new_filter(struct rar5* rar) {
469 struct filter_info* f =
470 (struct filter_info*) calloc(1, sizeof(struct filter_info));
476 cdeque_push_back(&rar->cstate.filters, cdeque_filter(f));
480 static int run_delta_filter(struct rar5* rar, struct filter_info* flt) {
482 ssize_t dest_pos, src_pos = 0;
484 for(i = 0; i < flt->channels; i++) {
485 uint8_t prev_byte = 0;
487 dest_pos < flt->block_length;
488 dest_pos += flt->channels)
492 byte = rar->cstate.window_buf[(rar->cstate.solid_offset +
493 flt->block_start + src_pos) & rar->cstate.window_mask];
496 rar->cstate.filtered_buf[dest_pos] = prev_byte;
504 static int run_e8e9_filter(struct rar5* rar, struct filter_info* flt,
507 const uint32_t file_size = 0x1000000;
510 circular_memcpy(rar->cstate.filtered_buf,
511 rar->cstate.window_buf,
512 rar->cstate.window_mask,
513 rar->cstate.solid_offset + flt->block_start,
514 rar->cstate.solid_offset + flt->block_start + flt->block_length);
516 for(i = 0; i < flt->block_length - 4;) {
517 uint8_t b = rar->cstate.window_buf[(rar->cstate.solid_offset +
518 flt->block_start + i++) & rar->cstate.window_mask];
520 /* 0xE8 = x86's call <relative_addr_uint32> (function call)
521 * 0xE9 = x86's jmp <relative_addr_uint32> (unconditional jump) */
522 if(b == 0xE8 || (extended && b == 0xE9)) {
525 uint32_t offset = (i + flt->block_start) % file_size;
527 addr = read_filter_data(rar, (rar->cstate.solid_offset +
528 flt->block_start + i) & rar->cstate.window_mask);
530 if(addr & 0x80000000) {
531 if(((addr + offset) & 0x80000000) == 0) {
532 write_filter_data(rar, i, addr + file_size);
535 if((addr - file_size) & 0x80000000) {
536 uint32_t naddr = addr - offset;
537 write_filter_data(rar, i, naddr);
548 static int run_arm_filter(struct rar5* rar, struct filter_info* flt) {
551 const int mask = rar->cstate.window_mask;
553 circular_memcpy(rar->cstate.filtered_buf,
554 rar->cstate.window_buf,
555 rar->cstate.window_mask,
556 rar->cstate.solid_offset + flt->block_start,
557 rar->cstate.solid_offset + flt->block_start + flt->block_length);
559 for(i = 0; i < flt->block_length - 3; i += 4) {
560 uint8_t* b = &rar->cstate.window_buf[(rar->cstate.solid_offset +
561 flt->block_start + i) & mask];
564 /* 0xEB = ARM's BL (branch + link) instruction. */
565 offset = read_filter_data(rar, (rar->cstate.solid_offset +
566 flt->block_start + i) & mask) & 0x00ffffff;
568 offset -= (uint32_t) ((i + flt->block_start) / 4);
569 offset = (offset & 0x00ffffff) | 0xeb000000;
570 write_filter_data(rar, i, offset);
577 static int run_filter(struct archive_read* a, struct filter_info* flt) {
579 struct rar5* rar = get_context(a);
581 free(rar->cstate.filtered_buf);
583 rar->cstate.filtered_buf = malloc(flt->block_length);
584 if(!rar->cstate.filtered_buf) {
585 archive_set_error(&a->archive, ENOMEM, "Can't allocate memory for "
587 return ARCHIVE_FATAL;
592 ret = run_delta_filter(rar, flt);
598 ret = run_e8e9_filter(rar, flt, flt->type == FILTER_E8E9);
602 ret = run_arm_filter(rar, flt);
606 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
607 "Unsupported filter type: 0x%02x", flt->type);
608 return ARCHIVE_FATAL;
611 if(ret != ARCHIVE_OK) {
612 /* Filter has failed. */
616 if(ARCHIVE_OK != push_data_ready(a, rar, rar->cstate.filtered_buf,
617 flt->block_length, rar->cstate.last_write_ptr))
619 archive_set_error(&a->archive, ARCHIVE_ERRNO_PROGRAMMER,
620 "Stack overflow when submitting unpacked data");
622 return ARCHIVE_FATAL;
625 rar->cstate.last_write_ptr += flt->block_length;
629 /* The `push_data` function submits the selected data range to the user.
630 * Next call of `use_data` will use the pointer, size and offset arguments
631 * that are specified here. These arguments are pushed to the FIFO stack here,
632 * and popped from the stack by the `use_data` function. */
633 static void push_data(struct archive_read* a, struct rar5* rar,
634 const uint8_t* buf, int64_t idx_begin, int64_t idx_end)
636 const int wmask = rar->cstate.window_mask;
637 const ssize_t solid_write_ptr = (rar->cstate.solid_offset +
638 rar->cstate.last_write_ptr) & wmask;
640 idx_begin += rar->cstate.solid_offset;
641 idx_end += rar->cstate.solid_offset;
643 /* Check if our unpacked data is wrapped inside the window circular buffer.
644 * If it's not wrapped, it can be copied out by using a single memcpy,
645 * but when it's wrapped, we need to copy the first part with one
646 * memcpy, and the second part with another memcpy. */
648 if((idx_begin & wmask) > (idx_end & wmask)) {
649 /* The data is wrapped (begin offset sis bigger than end offset). */
650 const ssize_t frag1_size = rar->cstate.window_size - (idx_begin & wmask);
651 const ssize_t frag2_size = idx_end & wmask;
653 /* Copy the first part of the buffer first. */
654 push_data_ready(a, rar, buf + solid_write_ptr, frag1_size,
655 rar->cstate.last_write_ptr);
657 /* Copy the second part of the buffer. */
658 push_data_ready(a, rar, buf, frag2_size,
659 rar->cstate.last_write_ptr + frag1_size);
661 rar->cstate.last_write_ptr += frag1_size + frag2_size;
663 /* Data is not wrapped, so we can just use one call to copy the
665 push_data_ready(a, rar,
666 buf + solid_write_ptr,
667 (idx_end - idx_begin) & wmask,
668 rar->cstate.last_write_ptr);
670 rar->cstate.last_write_ptr += idx_end - idx_begin;
674 /* Convinience function that submits the data to the user. It uses the
675 * unpack window buffer as a source location. */
676 static void push_window_data(struct archive_read* a, struct rar5* rar,
677 int64_t idx_begin, int64_t idx_end)
679 push_data(a, rar, rar->cstate.window_buf, idx_begin, idx_end);
682 static int apply_filters(struct archive_read* a) {
683 struct filter_info* flt;
684 struct rar5* rar = get_context(a);
687 rar->cstate.all_filters_applied = 0;
689 /* Get the first filter that can be applied to our data. The data needs to
690 * be fully unpacked before the filter can be run. */
692 cdeque_front(&rar->cstate.filters, cdeque_filter_p(&flt)))
694 /* Check if our unpacked data fully covers this filter's range. */
695 if(rar->cstate.write_ptr > flt->block_start &&
696 rar->cstate.write_ptr >= flt->block_start + flt->block_length)
698 /* Check if we have some data pending to be written right before
699 * the filter's start offset. */
700 if(rar->cstate.last_write_ptr == flt->block_start) {
701 /* Run the filter specified by descriptor `flt`. */
702 ret = run_filter(a, flt);
703 if(ret != ARCHIVE_OK) {
704 /* Filter failure, return error. */
708 /* Filter descriptor won't be needed anymore after it's used,
709 * so remove it from the filter list and free its memory. */
710 (void) cdeque_pop_front(&rar->cstate.filters,
711 cdeque_filter_p(&flt));
715 /* We can't run filters yet, dump the memory right before the
717 push_window_data(a, rar, rar->cstate.last_write_ptr,
721 /* Return 'filter applied or not needed' state to the caller. */
722 return ARCHIVE_RETRY;
726 rar->cstate.all_filters_applied = 1;
730 static void dist_cache_push(struct rar5* rar, int value) {
731 int* q = rar->cstate.dist_cache;
739 static int dist_cache_touch(struct rar5* rar, int index) {
740 int* q = rar->cstate.dist_cache;
741 int i, dist = q[index];
743 for(i = index; i > 0; i--)
750 static void free_filters(struct rar5* rar) {
751 struct cdeque* d = &rar->cstate.filters;
753 /* Free any remaining filters. All filters should be naturally consumed by
754 * the unpacking function, so remaining filters after unpacking normally
755 * mean that unpacking wasn't successfull. But still of course we shouldn't
756 * leak memory in such case. */
758 /* cdeque_size() is a fast operation, so we can use it as a loop
760 while(cdeque_size(d) > 0) {
761 struct filter_info* f = NULL;
763 /* Pop_front will also decrease the collection's size. */
764 if (CDE_OK == cdeque_pop_front(d, cdeque_filter_p(&f)))
770 /* Also clear out the variables needed for sanity checking. */
771 rar->cstate.last_block_start = 0;
772 rar->cstate.last_block_length = 0;
775 static void reset_file_context(struct rar5* rar) {
776 memset(&rar->file, 0, sizeof(rar->file));
777 blake2sp_init(&rar->file.b2state, 32);
779 if(rar->main.solid) {
780 rar->cstate.solid_offset += rar->cstate.write_ptr;
782 rar->cstate.solid_offset = 0;
785 rar->cstate.write_ptr = 0;
786 rar->cstate.last_write_ptr = 0;
787 rar->cstate.last_unstore_ptr = 0;
792 static inline int get_archive_read(struct archive* a,
793 struct archive_read** ar)
795 *ar = (struct archive_read*) a;
796 archive_check_magic(a, ARCHIVE_READ_MAGIC, ARCHIVE_STATE_NEW,
797 "archive_read_support_format_rar5");
802 static int read_ahead(struct archive_read* a, size_t how_many,
809 *ptr = __archive_read_ahead(a, how_many, &avail);
818 static int consume(struct archive_read* a, int64_t how_many) {
822 how_many == __archive_read_consume(a, how_many)
830 * Read a RAR5 variable sized numeric value. This value will be stored in
831 * `pvalue`. The `pvalue_len` argument points to a variable that will receive
832 * the byte count that was consumed in order to decode the `pvalue` value, plus
835 * pvalue_len is optional and can be NULL.
837 * NOTE: if `pvalue_len` is NOT NULL, the caller needs to manually consume
838 * the number of bytes that `pvalue_len` value contains. If the `pvalue_len`
839 * is NULL, this consuming operation is done automatically.
841 * Returns 1 if *pvalue was successfully read.
842 * Returns 0 if there was an error. In this case, *pvalue contains an
846 static int read_var(struct archive_read* a, uint64_t* pvalue,
847 uint64_t* pvalue_len)
854 /* We will read maximum of 8 bytes. We don't have to handle the situation
855 * to read the RAR5 variable-sized value stored at the end of the file,
856 * because such situation will never happen. */
857 if(!read_ahead(a, 8, &p))
860 for(shift = 0, i = 0; i < 8; i++, shift += 7) {
863 /* Strip the MSB from the input byte and add the resulting number
864 * to the `result`. */
865 result += (b & (uint64_t)0x7F) << shift;
867 /* MSB set to 1 means we need to continue decoding process. MSB set
868 * to 0 means we're done.
870 * This conditional checks for the second case. */
871 if((b & 0x80) == 0) {
876 /* If the caller has passed the `pvalue_len` pointer, store the
877 * number of consumed bytes in it and do NOT consume those bytes,
878 * since the caller has all the information it needs to perform
879 * the consuming process itself. */
883 /* If the caller did not provide the `pvalue_len` pointer,
884 * it will not have the possibility to advance the file
885 * pointer, because it will not know how many bytes it needs
886 * to consume. This is why we handle such situation here
888 if(ARCHIVE_OK != consume(a, 1 + i)) {
893 /* End of decoding process, return success. */
898 /* The decoded value takes the maximum number of 8 bytes. It's a maximum
899 * number of bytes, so end decoding process here even if the first bit
900 * of last byte is 1. */
908 if(ARCHIVE_OK != consume(a, 9)) {
916 static int read_var_sized(struct archive_read* a, size_t* pvalue,
922 const int ret = pvalue_len
923 ? read_var(a, &v, &v_size)
924 : read_var(a, &v, NULL);
926 if(ret == 1 && pvalue) {
927 *pvalue = (size_t) v;
931 /* Possible data truncation should be safe. */
932 *pvalue_len = (size_t) v_size;
938 static int read_bits_32(struct rar5* rar, const uint8_t* p, uint32_t* value) {
939 uint32_t bits = p[rar->bits.in_addr] << 24;
940 bits |= p[rar->bits.in_addr + 1] << 16;
941 bits |= p[rar->bits.in_addr + 2] << 8;
942 bits |= p[rar->bits.in_addr + 3];
943 bits <<= rar->bits.bit_addr;
944 bits |= p[rar->bits.in_addr + 4] >> (8 - rar->bits.bit_addr);
949 static int read_bits_16(struct rar5* rar, const uint8_t* p, uint16_t* value) {
950 int bits = (int) p[rar->bits.in_addr] << 16;
951 bits |= (int) p[rar->bits.in_addr + 1] << 8;
952 bits |= (int) p[rar->bits.in_addr + 2];
953 bits >>= (8 - rar->bits.bit_addr);
954 *value = bits & 0xffff;
958 static void skip_bits(struct rar5* rar, int bits) {
959 const int new_bits = rar->bits.bit_addr + bits;
960 rar->bits.in_addr += new_bits >> 3;
961 rar->bits.bit_addr = new_bits & 7;
965 static int read_consume_bits(struct rar5* rar, const uint8_t* p, int n,
971 if(n == 0 || n > 16) {
972 /* This is a programmer error and should never happen in runtime. */
973 return ARCHIVE_FATAL;
976 ret = read_bits_16(rar, p, &v);
977 if(ret != ARCHIVE_OK)
991 static int read_u32(struct archive_read* a, uint32_t* pvalue) {
993 if(!read_ahead(a, 4, &p))
996 *pvalue = *(const uint32_t*)p;
998 return ARCHIVE_OK == consume(a, 4) ? 1 : 0;
1001 static int read_u64(struct archive_read* a, uint64_t* pvalue) {
1003 if(!read_ahead(a, 8, &p))
1006 *pvalue = *(const uint64_t*)p;
1008 return ARCHIVE_OK == consume(a, 8) ? 1 : 0;
1011 static int bid_standard(struct archive_read* a) {
1014 if(!read_ahead(a, rar5_signature_size, &p))
1017 if(!memcmp(rar5_signature, p, rar5_signature_size))
1023 static int rar5_bid(struct archive_read* a, int best_bid) {
1029 my_bid = bid_standard(a);
1037 static int rar5_options(struct archive_read *a, const char *key, const char *val) {
1042 /* No options supported in this version. Return the ARCHIVE_WARN code to
1043 * signal the options supervisor that the unpacker didn't handle setting
1046 return ARCHIVE_WARN;
1049 static void init_header(struct archive_read* a) {
1050 a->archive.archive_format = ARCHIVE_FORMAT_RAR_V5;
1051 a->archive.archive_format_name = "RAR5";
1055 HFL_EXTRA_DATA = 0x0001, HFL_DATA = 0x0002, HFL_SKIP_IF_UNKNOWN = 0x0004,
1056 HFL_SPLIT_BEFORE = 0x0008, HFL_SPLIT_AFTER = 0x0010, HFL_CHILD = 0x0020,
1057 HFL_INHERITED = 0x0040
1060 static int process_main_locator_extra_block(struct archive_read* a,
1063 uint64_t locator_flags;
1065 if(!read_var(a, &locator_flags, NULL)) {
1069 enum LOCATOR_FLAGS {
1070 QLIST = 0x01, RECOVERY = 0x02,
1073 if(locator_flags & QLIST) {
1074 if(!read_var(a, &rar->qlist_offset, NULL)) {
1078 /* qlist is not used */
1081 if(locator_flags & RECOVERY) {
1082 if(!read_var(a, &rar->rr_offset, NULL)) {
1086 /* rr is not used */
1092 static int parse_file_extra_hash(struct archive_read* a, struct rar5* rar,
1093 ssize_t* extra_data_size)
1098 if(!read_var_sized(a, &hash_type, &value_len))
1101 *extra_data_size -= value_len;
1102 if(ARCHIVE_OK != consume(a, value_len)) {
1110 /* The file uses BLAKE2sp checksum algorithm instead of plain old
1112 if(hash_type == BLAKE2sp) {
1114 const int hash_size = sizeof(rar->file.blake2sp);
1116 if(!read_ahead(a, hash_size, &p))
1119 rar->file.has_blake2 = 1;
1120 memcpy(&rar->file.blake2sp, p, hash_size);
1122 if(ARCHIVE_OK != consume(a, hash_size)) {
1126 *extra_data_size -= hash_size;
1128 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
1129 "Unsupported hash type (0x%02x)", (int) hash_type);
1130 return ARCHIVE_FATAL;
1136 static uint64_t time_win_to_unix(uint64_t win_time) {
1137 const size_t ns_in_sec = 10000000;
1138 const uint64_t sec_to_unix = 11644473600LL;
1139 return win_time / ns_in_sec - sec_to_unix;
1142 static int parse_htime_item(struct archive_read* a, char unix_time,
1143 uint64_t* where, ssize_t* extra_data_size)
1147 if(!read_u32(a, &time_val))
1150 *extra_data_size -= 4;
1151 *where = (uint64_t) time_val;
1153 uint64_t windows_time;
1154 if(!read_u64(a, &windows_time))
1157 *where = time_win_to_unix(windows_time);
1158 *extra_data_size -= 8;
1164 static int parse_file_extra_htime(struct archive_read* a,
1165 struct archive_entry* e, struct rar5* rar,
1166 ssize_t* extra_data_size)
1180 if(!read_var_sized(a, &flags, &value_len))
1183 *extra_data_size -= value_len;
1184 if(ARCHIVE_OK != consume(a, value_len)) {
1188 unix_time = flags & IS_UNIX;
1190 if(flags & HAS_MTIME) {
1191 parse_htime_item(a, unix_time, &rar->file.e_mtime, extra_data_size);
1192 archive_entry_set_mtime(e, rar->file.e_mtime, 0);
1195 if(flags & HAS_CTIME) {
1196 parse_htime_item(a, unix_time, &rar->file.e_ctime, extra_data_size);
1197 archive_entry_set_ctime(e, rar->file.e_ctime, 0);
1200 if(flags & HAS_ATIME) {
1201 parse_htime_item(a, unix_time, &rar->file.e_atime, extra_data_size);
1202 archive_entry_set_atime(e, rar->file.e_atime, 0);
1205 if(flags & HAS_UNIX_NS) {
1206 if(!read_u32(a, &rar->file.e_unix_ns))
1209 *extra_data_size -= 4;
1215 static int process_head_file_extra(struct archive_read* a,
1216 struct archive_entry* e, struct rar5* rar,
1217 ssize_t extra_data_size)
1219 size_t extra_field_size;
1220 size_t extra_field_id;
1221 int ret = ARCHIVE_FATAL;
1225 CRYPT = 0x01, HASH = 0x02, HTIME = 0x03, VERSION_ = 0x04,
1226 REDIR = 0x05, UOWNER = 0x06, SUBDATA = 0x07
1229 while(extra_data_size > 0) {
1230 if(!read_var_sized(a, &extra_field_size, &var_size))
1233 extra_data_size -= var_size;
1234 if(ARCHIVE_OK != consume(a, var_size)) {
1238 if(!read_var_sized(a, &extra_field_id, &var_size))
1241 extra_data_size -= var_size;
1242 if(ARCHIVE_OK != consume(a, var_size)) {
1246 switch(extra_field_id) {
1248 ret = parse_file_extra_hash(a, rar, &extra_data_size);
1251 ret = parse_file_extra_htime(a, e, rar, &extra_data_size);
1264 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
1265 "Unknown extra field in file/service block: 0x%02x",
1266 (int) extra_field_id);
1267 return ARCHIVE_FATAL;
1271 if(ret != ARCHIVE_OK) {
1272 /* Attribute not implemented. */
1279 static int process_head_file(struct archive_read* a, struct rar5* rar,
1280 struct archive_entry* entry, size_t block_flags)
1282 ssize_t extra_data_size = 0;
1283 size_t data_size, file_flags, file_attr, compression_info, host_os,
1285 uint64_t unpacked_size;
1286 uint32_t mtime = 0, crc;
1287 int c_method = 0, c_version = 0, is_dir;
1288 char name_utf8_buf[2048 * 4];
1291 archive_entry_clear(entry);
1293 /* Do not reset file context if we're switching archives. */
1294 if(!rar->cstate.switch_multivolume) {
1295 reset_file_context(rar);
1298 if(block_flags & HFL_EXTRA_DATA) {
1300 if(!read_var_sized(a, &edata_size, NULL))
1303 /* Intentional type cast from unsigned to signed. */
1304 extra_data_size = (ssize_t) edata_size;
1307 if(block_flags & HFL_DATA) {
1308 if(!read_var_sized(a, &data_size, NULL))
1311 rar->file.bytes_remaining = data_size;
1313 rar->file.bytes_remaining = 0;
1315 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
1316 "no data found in file/service block");
1317 return ARCHIVE_FATAL;
1321 DIRECTORY = 0x0001, UTIME = 0x0002, CRC32 = 0x0004,
1322 UNKNOWN_UNPACKED_SIZE = 0x0008,
1325 enum COMP_INFO_FLAGS {
1329 if(!read_var_sized(a, &file_flags, NULL))
1332 if(!read_var(a, &unpacked_size, NULL))
1335 if(file_flags & UNKNOWN_UNPACKED_SIZE) {
1336 archive_set_error(&a->archive, ARCHIVE_ERRNO_PROGRAMMER,
1337 "Files with unknown unpacked size are not supported");
1338 return ARCHIVE_FATAL;
1341 is_dir = (int) (file_flags & DIRECTORY);
1343 if(!read_var_sized(a, &file_attr, NULL))
1346 if(file_flags & UTIME) {
1347 if(!read_u32(a, &mtime))
1351 if(file_flags & CRC32) {
1352 if(!read_u32(a, &crc))
1356 if(!read_var_sized(a, &compression_info, NULL))
1359 c_method = (int) (compression_info >> 7) & 0x7;
1360 c_version = (int) (compression_info & 0x3f);
1362 rar->cstate.window_size = is_dir ?
1364 g_unpack_window_size << ((compression_info >> 10) & 15);
1365 rar->cstate.method = c_method;
1366 rar->cstate.version = c_version + 50;
1368 rar->file.solid = (compression_info & SOLID) > 0;
1369 rar->file.service = 0;
1371 if(!read_var_sized(a, &host_os, NULL))
1379 if(host_os == HOST_WINDOWS) {
1380 /* Host OS is Windows */
1382 unsigned short mode = 0660;
1389 archive_entry_set_mode(entry, mode);
1390 } else if(host_os == HOST_UNIX) {
1391 /* Host OS is Unix */
1392 archive_entry_set_mode(entry, (unsigned short) file_attr);
1394 /* Unknown host OS */
1395 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
1396 "Unsupported Host OS: 0x%02x", (int) host_os);
1398 return ARCHIVE_FATAL;
1401 if(!read_var_sized(a, &name_size, NULL))
1404 if(!read_ahead(a, name_size, &p))
1407 if(name_size > 2047) {
1408 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
1409 "Filename is too long");
1411 return ARCHIVE_FATAL;
1414 if(name_size == 0) {
1415 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
1416 "No filename specified");
1418 return ARCHIVE_FATAL;
1421 memcpy(name_utf8_buf, p, name_size);
1422 name_utf8_buf[name_size] = 0;
1423 if(ARCHIVE_OK != consume(a, name_size)) {
1427 if(extra_data_size > 0) {
1428 int ret = process_head_file_extra(a, entry, rar, extra_data_size);
1431 if(extra_data_size < 0) {
1432 archive_set_error(&a->archive, ARCHIVE_ERRNO_PROGRAMMER,
1433 "File extra data size is not zero");
1434 return ARCHIVE_FATAL;
1437 if(ret != ARCHIVE_OK)
1441 if((file_flags & UNKNOWN_UNPACKED_SIZE) == 0) {
1442 rar->file.unpacked_size = (ssize_t) unpacked_size;
1443 archive_entry_set_size(entry, unpacked_size);
1446 if(file_flags & UTIME) {
1447 archive_entry_set_mtime(entry, (time_t) mtime, 0);
1450 if(file_flags & CRC32) {
1451 rar->file.stored_crc32 = crc;
1454 archive_entry_update_pathname_utf8(entry, name_utf8_buf);
1456 if(!rar->cstate.switch_multivolume) {
1457 /* Do not reinitialize unpacking state if we're switching archives. */
1458 rar->cstate.block_parsing_finished = 1;
1459 rar->cstate.all_filters_applied = 1;
1460 rar->cstate.initialized = 0;
1463 if(rar->generic.split_before > 0) {
1464 /* If now we're standing on a header that has a 'split before' mark,
1465 * it means we're standing on a 'continuation' file header. Signal
1466 * the caller that if it wants to move to another file, it must call
1467 * rar5_read_header() function again. */
1469 return ARCHIVE_RETRY;
1475 static int process_head_service(struct archive_read* a, struct rar5* rar,
1476 struct archive_entry* entry, size_t block_flags)
1478 /* Process this SERVICE block the same way as FILE blocks. */
1479 int ret = process_head_file(a, rar, entry, block_flags);
1480 if(ret != ARCHIVE_OK)
1483 rar->file.service = 1;
1485 /* But skip the data part automatically. It's no use for the user anyway.
1486 * It contains only service data, not even needed to properly unpack the
1488 ret = rar5_read_data_skip(a);
1489 if(ret != ARCHIVE_OK)
1492 /* After skipping, try parsing another block automatically. */
1493 return ARCHIVE_RETRY;
1496 static int process_head_main(struct archive_read* a, struct rar5* rar,
1497 struct archive_entry* entry, size_t block_flags)
1502 size_t extra_data_size,
1507 if(block_flags & HFL_EXTRA_DATA) {
1508 if(!read_var_sized(a, &extra_data_size, NULL))
1511 extra_data_size = 0;
1514 if(!read_var_sized(a, &archive_flags, NULL)) {
1519 VOLUME = 0x0001, /* multi-volume archive */
1520 VOLUME_NUMBER = 0x0002, /* volume number, first vol doesnt have it */
1521 SOLID = 0x0004, /* solid archive */
1522 PROTECT = 0x0008, /* contains Recovery info */
1523 LOCK = 0x0010, /* readonly flag, not used */
1526 rar->main.volume = (archive_flags & VOLUME) > 0;
1527 rar->main.solid = (archive_flags & SOLID) > 0;
1529 if(archive_flags & VOLUME_NUMBER) {
1531 if(!read_var_sized(a, &v, NULL)) {
1535 rar->main.vol_no = (int) v;
1537 rar->main.vol_no = 0;
1540 if(rar->vol.expected_vol_no > 0 &&
1541 rar->main.vol_no != rar->vol.expected_vol_no)
1543 /* Returning EOF instead of FATAL because of strange libarchive
1544 * behavior. When opening multiple files via
1545 * archive_read_open_filenames(), after reading up the whole last file,
1546 * the __archive_read_ahead function wraps up to the first archive
1547 * instead of returning EOF. */
1551 if(extra_data_size == 0) {
1556 if(!read_var_sized(a, &extra_field_size, NULL)) {
1560 if(!read_var_sized(a, &extra_field_id, NULL)) {
1564 if(extra_field_size == 0) {
1565 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
1566 "Invalid extra field size");
1567 return ARCHIVE_FATAL;
1571 // Just one attribute here.
1575 switch(extra_field_id) {
1577 ret = process_main_locator_extra_block(a, rar);
1578 if(ret != ARCHIVE_OK) {
1579 /* Error while parsing main locator extra block. */
1585 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
1586 "Unsupported extra type (0x%02x)", (int) extra_field_id);
1587 return ARCHIVE_FATAL;
1593 static int scan_for_signature(struct archive_read* a);
1595 /* Base block processing function. A 'base block' is a RARv5 header block
1596 * that tells the reader what kind of data is stored inside the block.
1598 * From the birds-eye view a RAR file looks file this:
1600 * <magic><base_block_1><base_block_2>...<base_block_n>
1602 * There are a few types of base blocks. Those types are specified inside
1603 * the 'switch' statement in this function. For example purposes, I'll write
1604 * how a standard RARv5 file could look like here:
1606 * <magic><MAIN><FILE><FILE><FILE><SERVICE><ENDARC>
1608 * The structure above could describe an archive file with 3 files in it,
1609 * one service "QuickOpen" block (that is ignored by this parser), and an
1610 * end of file base block marker.
1612 * If the file is stored in multiple archive files ("multiarchive"), it might
1615 * .part01.rar: <magic><MAIN><FILE><ENDARC>
1616 * .part02.rar: <magic><MAIN><FILE><ENDARC>
1617 * .part03.rar: <magic><MAIN><FILE><ENDARC>
1619 * This example could describe 3 RAR files that contain ONE archived file.
1620 * Or it could describe 3 RAR files that contain 3 different files. Or 3
1621 * RAR files than contain 2 files. It all depends what metadata is stored in
1622 * the headers of <FILE> blocks.
1624 * Each <FILE> block contains info about its size, the name of the file it's
1625 * storing inside, and whether this FILE block is a continuation block of
1626 * previous archive ('split before'), and is this FILE block should be
1627 * continued in another archive ('split after'). By parsing the 'split before'
1628 * and 'split after' flags, we're able to tell if multiple <FILE> base blocks
1629 * are describing one file, or multiple files (with the same filename, for
1632 * One thing to note is that if we're parsing the first <FILE> block, and
1633 * we see 'split after' flag, then we need to jump over to another <FILE>
1634 * block to be able to decompress rest of the data. To do this, we need
1635 * to skip the <ENDARC> block, then switch to another file, then skip the
1636 * <magic> block, <MAIN> block, and then we're standing on the proper
1640 static int process_base_block(struct archive_read* a,
1641 struct archive_entry* entry)
1643 struct rar5* rar = get_context(a);
1644 uint32_t hdr_crc, computed_crc;
1645 size_t raw_hdr_size, hdr_size_len, hdr_size;
1646 size_t header_id, header_flags;
1650 /* Skip any unprocessed data for this file. */
1651 if(rar->file.bytes_remaining) {
1652 ret = rar5_read_data_skip(a);
1653 if(ret != ARCHIVE_OK) {
1658 /* Read the expected CRC32 checksum. */
1659 if(!read_u32(a, &hdr_crc)) {
1663 /* Read header size. */
1664 if(!read_var_sized(a, &raw_hdr_size, &hdr_size_len)) {
1668 /* Sanity check, maximum header size for RAR5 is 2MB. */
1669 if(raw_hdr_size > (2 * 1024 * 1024)) {
1670 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
1671 "Base block header is too large");
1673 return ARCHIVE_FATAL;
1676 hdr_size = raw_hdr_size + hdr_size_len;
1678 /* Read the whole header data into memory, maximum memory use here is
1680 if(!read_ahead(a, hdr_size, &p)) {
1684 /* Verify the CRC32 of the header data. */
1685 computed_crc = (uint32_t) crc32(0, p, (int) hdr_size);
1686 if(computed_crc != hdr_crc) {
1687 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
1688 "Header CRC error");
1690 return ARCHIVE_FATAL;
1693 /* If the checksum is OK, we proceed with parsing. */
1694 if(ARCHIVE_OK != consume(a, hdr_size_len)) {
1698 if(!read_var_sized(a, &header_id, NULL))
1701 if(!read_var_sized(a, &header_flags, NULL))
1704 rar->generic.split_after = (header_flags & HFL_SPLIT_AFTER) > 0;
1705 rar->generic.split_before = (header_flags & HFL_SPLIT_BEFORE) > 0;
1706 rar->generic.size = hdr_size;
1707 rar->generic.last_header_id = header_id;
1708 rar->main.endarc = 0;
1710 /* Those are possible header ids in RARv5. */
1712 HEAD_MARK = 0x00, HEAD_MAIN = 0x01, HEAD_FILE = 0x02,
1713 HEAD_SERVICE = 0x03, HEAD_CRYPT = 0x04, HEAD_ENDARC = 0x05,
1714 HEAD_UNKNOWN = 0xff,
1719 ret = process_head_main(a, rar, entry, header_flags);
1721 /* Main header doesn't have any files in it, so it's pointless
1722 * to return to the caller. Retry to next header, which should be
1723 * HEAD_FILE/HEAD_SERVICE. */
1724 if(ret == ARCHIVE_OK)
1725 return ARCHIVE_RETRY;
1729 ret = process_head_service(a, rar, entry, header_flags);
1732 ret = process_head_file(a, rar, entry, header_flags);
1735 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
1736 "Encryption is not supported");
1737 return ARCHIVE_FATAL;
1739 rar->main.endarc = 1;
1741 /* After encountering an end of file marker, we need to take
1742 * into consideration if this archive is continued in another
1743 * file (i.e. is it part01.rar: is there a part02.rar?) */
1744 if(rar->main.volume) {
1745 /* In case there is part02.rar, position the read pointer
1746 * in a proper place, so we can resume parsing. */
1748 ret = scan_for_signature(a);
1749 if(ret == ARCHIVE_FATAL) {
1752 rar->vol.expected_vol_no = rar->main.vol_no + 1;
1761 if((header_flags & HFL_SKIP_IF_UNKNOWN) == 0) {
1762 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
1763 "Header type error");
1764 return ARCHIVE_FATAL;
1766 /* If the block is marked as 'skip if unknown', do as the flag
1767 * says: skip the block instead on failing on it. */
1768 return ARCHIVE_RETRY;
1774 archive_set_error(&a->archive, ARCHIVE_ERRNO_PROGRAMMER,
1775 "Internal unpacker error");
1776 return ARCHIVE_FATAL;
1780 static int skip_base_block(struct archive_read* a) {
1782 struct rar5* rar = get_context(a);
1784 /* Create a new local archive_entry structure that will be operated on
1785 * by header reader; operations on this archive_entry will be discarded.
1787 struct archive_entry* entry = archive_entry_new();
1788 ret = process_base_block(a, entry);
1790 /* Discard operations on this archive_entry structure. */
1791 archive_entry_free(entry);
1793 if(rar->generic.last_header_id == 2 && rar->generic.split_before > 0)
1796 if(ret == ARCHIVE_OK)
1797 return ARCHIVE_RETRY;
1802 static int rar5_read_header(struct archive_read *a,
1803 struct archive_entry *entry)
1805 struct rar5* rar = get_context(a);
1808 if(rar->header_initialized == 0) {
1810 rar->header_initialized = 1;
1813 if(rar->skipped_magic == 0) {
1814 if(ARCHIVE_OK != consume(a, rar5_signature_size)) {
1818 rar->skipped_magic = 1;
1822 ret = process_base_block(a, entry);
1823 } while(ret == ARCHIVE_RETRY ||
1824 (rar->main.endarc > 0 && ret == ARCHIVE_OK));
1829 static void init_unpack(struct rar5* rar) {
1830 rar->file.calculated_crc32 = 0;
1831 if (rar->cstate.window_size)
1832 rar->cstate.window_mask = rar->cstate.window_size - 1;
1834 rar->cstate.window_mask = 0;
1836 free(rar->cstate.window_buf);
1838 free(rar->cstate.filtered_buf);
1840 rar->cstate.window_buf = calloc(1, rar->cstate.window_size);
1841 rar->cstate.filtered_buf = calloc(1, rar->cstate.window_size);
1843 rar->cstate.write_ptr = 0;
1844 rar->cstate.last_write_ptr = 0;
1846 memset(&rar->cstate.bd, 0, sizeof(rar->cstate.bd));
1847 memset(&rar->cstate.ld, 0, sizeof(rar->cstate.ld));
1848 memset(&rar->cstate.dd, 0, sizeof(rar->cstate.dd));
1849 memset(&rar->cstate.ldd, 0, sizeof(rar->cstate.ldd));
1850 memset(&rar->cstate.rd, 0, sizeof(rar->cstate.rd));
1853 static void update_crc(struct rar5* rar, const uint8_t* p, size_t to_read) {
1856 if(rar->skip_mode) {
1857 #if defined CHECK_CRC_ON_SOLID_SKIP
1866 /* Don't update CRC32 if the file doesn't have the `stored_crc32` info
1868 if(rar->file.stored_crc32 > 0) {
1869 rar->file.calculated_crc32 =
1870 crc32(rar->file.calculated_crc32, p, to_read);
1873 /* Check if the file uses an optional BLAKE2sp checksum algorithm. */
1874 if(rar->file.has_blake2 > 0) {
1875 /* Return value of the `update` function is always 0, so we can
1876 * explicitly ignore it here. */
1877 (void) blake2sp_update(&rar->file.b2state, p, to_read);
1882 static int create_decode_tables(uint8_t* bit_length,
1883 struct decode_table* table,
1886 int code, upper_limit = 0, i, lc[16];
1887 uint32_t decode_pos_clone[rar5_countof(table->decode_pos)];
1888 ssize_t cur_len, quick_data_size;
1890 memset(&lc, 0, sizeof(lc));
1891 memset(table->decode_num, 0, sizeof(table->decode_num));
1893 table->quick_bits = size == HUFF_NC ? 10 : 7;
1895 for(i = 0; i < size; i++) {
1896 lc[bit_length[i] & 15]++;
1900 table->decode_pos[0] = 0;
1901 table->decode_len[0] = 0;
1903 for(i = 1; i < 16; i++) {
1904 upper_limit += lc[i];
1906 table->decode_len[i] = upper_limit << (16 - i);
1907 table->decode_pos[i] = table->decode_pos[i - 1] + lc[i - 1];
1912 memcpy(decode_pos_clone, table->decode_pos, sizeof(decode_pos_clone));
1914 for(i = 0; i < size; i++) {
1915 uint8_t clen = bit_length[i] & 15;
1917 int last_pos = decode_pos_clone[clen];
1918 table->decode_num[last_pos] = i;
1919 decode_pos_clone[clen]++;
1923 quick_data_size = 1 << table->quick_bits;
1925 for(code = 0; code < quick_data_size; code++) {
1926 int bit_field = code << (16 - table->quick_bits);
1929 while(cur_len < rar5_countof(table->decode_len) &&
1930 bit_field >= table->decode_len[cur_len]) {
1934 table->quick_len[code] = (uint8_t) cur_len;
1936 dist = bit_field - table->decode_len[cur_len - 1];
1937 dist >>= (16 - cur_len);
1939 pos = table->decode_pos[cur_len] + dist;
1940 if(cur_len < rar5_countof(table->decode_pos) && pos < size) {
1941 table->quick_num[code] = table->decode_num[pos];
1943 table->quick_num[code] = 0;
1950 static int decode_number(struct archive_read* a, struct decode_table* table,
1951 const uint8_t* p, uint16_t* num)
1956 struct rar5* rar = get_context(a);
1958 if(ARCHIVE_OK != read_bits_16(rar, p, &bitfield)) {
1964 if(bitfield < table->decode_len[table->quick_bits]) {
1965 int code = bitfield >> (16 - table->quick_bits);
1966 skip_bits(rar, table->quick_len[code]);
1967 *num = table->quick_num[code];
1973 for(i = table->quick_bits + 1; i < 15; i++) {
1974 if(bitfield < table->decode_len[i]) {
1980 skip_bits(rar, bits);
1982 dist = bitfield - table->decode_len[bits - 1];
1983 dist >>= (16 - bits);
1984 pos = table->decode_pos[bits] + dist;
1986 if(pos >= table->size)
1989 *num = table->decode_num[pos];
1993 /* Reads and parses Huffman tables from the beginning of the block. */
1994 static int parse_tables(struct archive_read* a, struct rar5* rar,
1997 int ret, value, i, w, idx = 0;
1998 uint8_t bit_length[HUFF_BC],
1999 table[HUFF_TABLE_SIZE],
2003 enum { ESCAPE = 15 };
2005 /* The data for table generation is compressed using a simple RLE-like
2006 * algorithm when storing zeroes, so we need to unpack it first. */
2007 for(w = 0, i = 0; w < HUFF_BC;) {
2008 value = (p[i] & nibble_mask) >> nibble_shift;
2010 if(nibble_mask == 0x0F)
2013 nibble_mask ^= 0xFF;
2016 /* Values smaller than 15 is data, so we write it directly. Value 15
2017 * is a flag telling us that we need to unpack more bytes. */
2018 if(value == ESCAPE) {
2019 value = (p[i] & nibble_mask) >> nibble_shift;
2020 if(nibble_mask == 0x0F)
2022 nibble_mask ^= 0xFF;
2026 /* We sometimes need to write the actual value of 15, so this
2027 * case handles that. */
2028 bit_length[w++] = ESCAPE;
2033 for(k = 0; k < value + 2; k++) {
2034 bit_length[w++] = 0;
2038 bit_length[w++] = value;
2042 rar->bits.in_addr = i;
2043 rar->bits.bit_addr = nibble_shift ^ 4;
2045 ret = create_decode_tables(bit_length, &rar->cstate.bd, HUFF_BC);
2046 if(ret != ARCHIVE_OK) {
2047 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
2048 "Decoding huffman tables failed");
2049 return ARCHIVE_FATAL;
2052 for(i = 0; i < HUFF_TABLE_SIZE;) {
2055 ret = decode_number(a, &rar->cstate.bd, p, &num);
2056 if(ret != ARCHIVE_OK) {
2057 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
2058 "Decoding huffman tables failed");
2059 return ARCHIVE_FATAL;
2063 /* 0..15: store directly */
2064 table[i] = (uint8_t) num;
2070 /* 16..17: repeat previous code */
2072 if(ARCHIVE_OK != read_bits_16(rar, p, &n))
2086 while(n-- > 0 && i < HUFF_TABLE_SIZE) {
2087 table[i] = table[i - 1];
2091 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
2092 "Unexpected error when decoding huffman tables");
2093 return ARCHIVE_FATAL;
2099 /* other codes: fill with zeroes `n` times */
2101 if(ARCHIVE_OK != read_bits_16(rar, p, &n))
2114 while(n-- > 0 && i < HUFF_TABLE_SIZE)
2118 ret = create_decode_tables(&table[idx], &rar->cstate.ld, HUFF_NC);
2119 if(ret != ARCHIVE_OK) {
2120 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
2121 "Failed to create literal table");
2122 return ARCHIVE_FATAL;
2127 ret = create_decode_tables(&table[idx], &rar->cstate.dd, HUFF_DC);
2128 if(ret != ARCHIVE_OK) {
2129 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
2130 "Failed to create distance table");
2131 return ARCHIVE_FATAL;
2136 ret = create_decode_tables(&table[idx], &rar->cstate.ldd, HUFF_LDC);
2137 if(ret != ARCHIVE_OK) {
2138 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
2139 "Failed to create lower bits of distances table");
2140 return ARCHIVE_FATAL;
2145 ret = create_decode_tables(&table[idx], &rar->cstate.rd, HUFF_RC);
2146 if(ret != ARCHIVE_OK) {
2147 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
2148 "Failed to create repeating distances table");
2149 return ARCHIVE_FATAL;
2155 /* Parses the block header, verifies its CRC byte, and saves the header
2156 * fields inside the `hdr` pointer. */
2157 static int parse_block_header(struct archive_read* a, const uint8_t* p,
2158 ssize_t* block_size, struct compressed_block_header* hdr)
2160 memcpy(hdr, p, sizeof(struct compressed_block_header));
2162 if(hdr->block_flags.byte_count > 2) {
2163 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
2164 "Unsupported block header size (was %d, max is 2)",
2165 hdr->block_flags.byte_count);
2166 return ARCHIVE_FATAL;
2169 /* This should probably use bit reader interface in order to be more
2172 switch(hdr->block_flags.byte_count) {
2173 /* 1-byte block size */
2175 *block_size = *(const uint8_t*) &p[2];
2178 /* 2-byte block size */
2180 *block_size = *(const uint16_t*) &p[2];
2183 /* 3-byte block size */
2185 *block_size = *(const uint32_t*) &p[2];
2186 *block_size &= 0x00FFFFFF;
2189 /* Other block sizes are not supported. This case is not reached,
2190 * because we have an 'if' guard before the switch that makes sure
2193 return ARCHIVE_FATAL;
2196 /* Verify the block header checksum. 0x5A is a magic value and is always
2198 uint8_t calculated_cksum = 0x5A
2199 ^ (uint8_t) hdr->block_flags_u8
2200 ^ (uint8_t) *block_size
2201 ^ (uint8_t) (*block_size >> 8)
2202 ^ (uint8_t) (*block_size >> 16);
2204 if(calculated_cksum != hdr->block_cksum) {
2205 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
2206 "Block checksum error: got 0x%02x, expected 0x%02x",
2207 hdr->block_cksum, calculated_cksum);
2209 return ARCHIVE_FATAL;
2215 /* Convinience function used during filter processing. */
2216 static int parse_filter_data(struct rar5* rar, const uint8_t* p,
2217 uint32_t* filter_data)
2222 if(ARCHIVE_OK != read_consume_bits(rar, p, 2, &bytes))
2227 for(i = 0; i < bytes; i++) {
2230 if(ARCHIVE_OK != read_bits_16(rar, p, &byte)) {
2234 data += (byte >> 8) << (i * 8);
2238 *filter_data = data;
2242 /* Function is used during sanity checking. */
2243 static int is_valid_filter_block_start(struct rar5* rar,
2246 const int64_t block_start = (ssize_t) start + rar->cstate.write_ptr;
2247 const int64_t last_bs = rar->cstate.last_block_start;
2248 const ssize_t last_bl = rar->cstate.last_block_length;
2250 if(last_bs == 0 || last_bl == 0) {
2251 /* We didn't have any filters yet, so accept this offset. */
2255 if(block_start >= last_bs + last_bl) {
2256 /* Current offset is bigger than last block's end offset, so
2257 * accept current offset. */
2261 /* Any other case is not a normal situation and we should fail. */
2265 /* The function will create a new filter, read its parameters from the input
2266 * stream and add it to the filter collection. */
2267 static int parse_filter(struct archive_read* ar, const uint8_t* p) {
2268 uint32_t block_start, block_length;
2269 uint16_t filter_type;
2270 struct rar5* rar = get_context(ar);
2272 /* Read the parameters from the input stream. */
2273 if(ARCHIVE_OK != parse_filter_data(rar, p, &block_start))
2276 if(ARCHIVE_OK != parse_filter_data(rar, p, &block_length))
2279 if(ARCHIVE_OK != read_bits_16(rar, p, &filter_type))
2285 /* Perform some sanity checks on this filter parameters. Note that we
2286 * allow only DELTA, E8/E9 and ARM filters here, because rest of filters
2287 * are not used in RARv5. */
2289 if(block_length < 4 ||
2290 block_length > 0x400000 ||
2291 filter_type > FILTER_ARM ||
2292 !is_valid_filter_block_start(rar, block_start))
2294 archive_set_error(&ar->archive, ARCHIVE_ERRNO_FILE_FORMAT, "Invalid "
2295 "filter encountered");
2296 return ARCHIVE_FATAL;
2299 /* Allocate a new filter. */
2300 struct filter_info* filt = add_new_filter(rar);
2302 archive_set_error(&ar->archive, ENOMEM, "Can't allocate memory for a "
2303 "filter descriptor.");
2304 return ARCHIVE_FATAL;
2307 filt->type = filter_type;
2308 filt->block_start = rar->cstate.write_ptr + block_start;
2309 filt->block_length = block_length;
2311 rar->cstate.last_block_start = filt->block_start;
2312 rar->cstate.last_block_length = filt->block_length;
2314 /* Read some more data in case this is a DELTA filter. Other filter types
2315 * don't require any additional data over what was already read. */
2316 if(filter_type == FILTER_DELTA) {
2319 if(ARCHIVE_OK != read_consume_bits(rar, p, 5, &channels))
2322 filt->channels = channels + 1;
2328 static int decode_code_length(struct rar5* rar, const uint8_t* p,
2331 int lbits, length = 2;
2336 lbits = code / 4 - 1;
2337 length += (4 | (code & 3)) << lbits;
2343 if(ARCHIVE_OK != read_consume_bits(rar, p, lbits, &add))
2352 static int copy_string(struct archive_read* a, int len, int dist) {
2353 struct rar5* rar = get_context(a);
2354 const int cmask = rar->cstate.window_mask;
2355 const int64_t write_ptr = rar->cstate.write_ptr + rar->cstate.solid_offset;
2358 /* The unpacker spends most of the time in this function. It would be
2359 * a good idea to introduce some optimizations here.
2361 * Just remember that this loop treats buffers that overlap differently
2362 * than buffers that do not overlap. This is why a simple memcpy(3) call
2363 * will not be enough. */
2365 for(i = 0; i < len; i++) {
2366 const ssize_t write_idx = (write_ptr + i) & cmask;
2367 const ssize_t read_idx = (write_ptr + i - dist) & cmask;
2368 rar->cstate.window_buf[write_idx] = rar->cstate.window_buf[read_idx];
2371 rar->cstate.write_ptr += len;
2375 static int do_uncompress_block(struct archive_read* a, const uint8_t* p) {
2376 struct rar5* rar = get_context(a);
2380 const int cmask = rar->cstate.window_mask;
2381 const struct compressed_block_header* hdr = &rar->last_block_hdr;
2382 const uint8_t bit_size = 1 + hdr->block_flags.bit_size;
2385 if(rar->cstate.write_ptr - rar->cstate.last_write_ptr >
2386 (rar->cstate.window_size >> 1)) {
2388 /* Don't allow growing data by more than half of the window size
2389 * at a time. In such case, break the loop; next call to this
2390 * function will continue processing from this moment. */
2395 if(rar->bits.in_addr > rar->cstate.cur_block_size - 1 ||
2396 (rar->bits.in_addr == rar->cstate.cur_block_size - 1 &&
2397 rar->bits.bit_addr >= bit_size))
2399 /* If the program counter is here, it means the function has
2400 * finished processing the block. */
2401 rar->cstate.block_parsing_finished = 1;
2405 /* Decode the next literal. */
2406 if(ARCHIVE_OK != decode_number(a, &rar->cstate.ld, p, &num)) {
2410 /* Num holds a decompression literal, or 'command code'.
2412 * - Values lower than 256 are just bytes. Those codes can be stored
2413 * in the output buffer directly.
2415 * - Code 256 defines a new filter, which is later used to transform
2416 * the data block accordingly to the filter type. The data block
2417 * needs to be fully uncompressed first.
2419 * - Code bigger than 257 and smaller than 262 define a repetition
2420 * pattern that should be copied from an already uncompressed chunk
2425 /* Directly store the byte. */
2427 int64_t write_idx = rar->cstate.solid_offset +
2428 rar->cstate.write_ptr++;
2430 rar->cstate.window_buf[write_idx & cmask] = (uint8_t) num;
2432 } else if(num >= 262) {
2434 int len = decode_code_length(rar, p, num - 262),
2439 archive_set_error(&a->archive, ARCHIVE_ERRNO_PROGRAMMER,
2440 "Failed to decode the code length");
2442 return ARCHIVE_FATAL;
2445 if(ARCHIVE_OK != decode_number(a, &rar->cstate.dd, p, &dist_slot))
2447 archive_set_error(&a->archive, ARCHIVE_ERRNO_PROGRAMMER,
2448 "Failed to decode the distance slot");
2450 return ARCHIVE_FATAL;
2457 dbits = dist_slot / 2 - 1;
2458 dist += (2 | (dist_slot & 1)) << dbits;
2467 if(ARCHIVE_OK != read_bits_32(rar, p, &add)) {
2468 /* Return EOF if we can't read more data. */
2472 skip_bits(rar, dbits - 4);
2473 add = (add >> (36 - dbits)) << 4;
2477 if(ARCHIVE_OK != decode_number(a, &rar->cstate.ldd, p,
2480 archive_set_error(&a->archive,
2481 ARCHIVE_ERRNO_PROGRAMMER,
2482 "Failed to decode the distance slot");
2484 return ARCHIVE_FATAL;
2489 /* dbits is one of [0,1,2,3] */
2492 if(ARCHIVE_OK != read_consume_bits(rar, p, dbits, &add)) {
2493 /* Return EOF if we can't read more data. */
2507 if(dist > 0x40000) {
2513 dist_cache_push(rar, dist);
2514 rar->cstate.last_len = len;
2516 if(ARCHIVE_OK != copy_string(a, len, dist))
2517 return ARCHIVE_FATAL;
2520 } else if(num == 256) {
2521 /* Create a filter. */
2522 ret = parse_filter(a, p);
2523 if(ret != ARCHIVE_OK)
2527 } else if(num == 257) {
2528 if(rar->cstate.last_len != 0) {
2529 if(ARCHIVE_OK != copy_string(a, rar->cstate.last_len,
2530 rar->cstate.dist_cache[0]))
2532 return ARCHIVE_FATAL;
2537 } else if(num < 262) {
2538 const int index = num - 258;
2539 const int dist = dist_cache_touch(rar, index);
2544 if(ARCHIVE_OK != decode_number(a, &rar->cstate.rd, p, &len_slot)) {
2545 return ARCHIVE_FATAL;
2548 len = decode_code_length(rar, p, len_slot);
2549 rar->cstate.last_len = len;
2551 if(ARCHIVE_OK != copy_string(a, len, dist))
2552 return ARCHIVE_FATAL;
2557 /* The program counter shouldn't reach here. */
2558 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
2559 "Unsupported block code: 0x%02x", num);
2561 return ARCHIVE_FATAL;
2567 /* Binary search for the RARv5 signature. */
2568 static int scan_for_signature(struct archive_read* a) {
2570 const int chunk_size = 512;
2573 /* If we're here, it means we're on an 'unknown territory' data.
2574 * There's no indication what kind of data we're reading here. It could be
2575 * some text comment, any kind of binary data, digital sign, dragons, etc.
2577 * We want to find a valid RARv5 magic header inside this unknown data. */
2579 /* Is it possible in libarchive to just skip everything until the
2580 * end of the file? If so, it would be a better approach than the
2581 * current implementation of this function. */
2584 if(!read_ahead(a, chunk_size, &p))
2587 for(i = 0; i < chunk_size - rar5_signature_size; i++) {
2588 if(memcmp(&p[i], rar5_signature, rar5_signature_size) == 0) {
2589 /* Consume the number of bytes we've used to search for the
2590 * signature, as well as the number of bytes used by the
2591 * signature itself. After this we should be standing on a
2592 * valid base block header. */
2593 (void) consume(a, i + rar5_signature_size);
2598 consume(a, chunk_size);
2601 return ARCHIVE_FATAL;
2604 /* This function will switch the multivolume archive file to another file,
2605 * i.e. from part03 to part 04. */
2606 static int advance_multivolume(struct archive_read* a) {
2608 struct rar5* rar = get_context(a);
2610 /* A small state machine that will skip unnecessary data, needed to
2611 * switch from one multivolume to another. Such skipping is needed if
2612 * we want to be an stream-oriented (instead of file-oriented)
2615 * The state machine starts with `rar->main.endarc` == 0. It also
2616 * assumes that current stream pointer points to some base block header.
2618 * The `endarc` field is being set when the base block parsing function
2619 * encounters the 'end of archive' marker.
2623 if(rar->main.endarc == 1) {
2624 rar->main.endarc = 0;
2625 while(ARCHIVE_RETRY == skip_base_block(a));
2628 /* Skip current base block. In order to properly skip it,
2629 * we really need to simply parse it and discard the results. */
2631 lret = skip_base_block(a);
2633 /* The `skip_base_block` function tells us if we should continue
2634 * with skipping, or we should stop skipping. We're trying to skip
2635 * everything up to a base FILE block. */
2637 if(lret != ARCHIVE_RETRY) {
2638 /* If there was an error during skipping, or we have just
2639 * skipped a FILE base block... */
2641 if(rar->main.endarc == 0) {
2653 /* Merges the partial block from the first multivolume archive file, and
2654 * partial block from the second multivolume archive file. The result is
2655 * a chunk of memory containing the whole block, and the stream pointer
2656 * is advanced to the next block in the second multivolume archive file. */
2657 static int merge_block(struct archive_read* a, ssize_t block_size,
2660 struct rar5* rar = get_context(a);
2661 ssize_t cur_block_size, partial_offset = 0;
2665 /* Set a flag that we're in the switching mode. */
2666 rar->cstate.switch_multivolume = 1;
2668 /* Reallocate the memory which will hold the whole block. */
2669 if(rar->vol.push_buf)
2670 free((void*) rar->vol.push_buf);
2672 /* Increasing the allocation block by 8 is due to bit reading functions,
2673 * which are using additional 2 or 4 bytes. Allocating the block size
2674 * by exact value would make bit reader perform reads from invalid memory
2675 * block when reading the last byte from the buffer. */
2676 rar->vol.push_buf = malloc(block_size + 8);
2677 if(!rar->vol.push_buf) {
2678 archive_set_error(&a->archive, ENOMEM, "Can't allocate memory for a "
2679 "merge block buffer.");
2680 return ARCHIVE_FATAL;
2683 /* Valgrind complains if the extension block for bit reader is not
2684 * initialized, so initialize it. */
2685 memset(&rar->vol.push_buf[block_size], 0, 8);
2687 /* A single block can span across multiple multivolume archive files,
2688 * so we use a loop here. This loop will consume enough multivolume
2689 * archive files until the whole block is read. */
2692 /* Get the size of current block chunk in this multivolume archive
2693 * file and read it. */
2695 rar5_min(rar->file.bytes_remaining, block_size - partial_offset);
2697 if(!read_ahead(a, cur_block_size, &lp))
2700 /* Sanity check; there should never be a situation where this function
2701 * reads more data than the block's size. */
2702 if(partial_offset + cur_block_size > block_size) {
2703 archive_set_error(&a->archive, ARCHIVE_ERRNO_PROGRAMMER,
2704 "Consumed too much data when merging blocks.");
2705 return ARCHIVE_FATAL;
2708 /* Merge previous block chunk with current block chunk, or create
2709 * first block chunk if this is our first iteration. */
2710 memcpy(&rar->vol.push_buf[partial_offset], lp, cur_block_size);
2712 /* Advance the stream read pointer by this block chunk size. */
2713 if(ARCHIVE_OK != consume(a, cur_block_size))
2716 /* Update the pointers. `partial_offset` contains information about
2717 * the sum of merged block chunks. */
2718 partial_offset += cur_block_size;
2719 rar->file.bytes_remaining -= cur_block_size;
2721 /* If `partial_offset` is the same as `block_size`, this means we've
2722 * merged all block chunks and we have a valid full block. */
2723 if(partial_offset == block_size) {
2727 /* If we don't have any bytes to read, this means we should switch
2728 * to another multivolume archive file. */
2729 if(rar->file.bytes_remaining == 0) {
2730 ret = advance_multivolume(a);
2731 if(ret != ARCHIVE_OK)
2736 *p = rar->vol.push_buf;
2738 /* If we're here, we can resume unpacking by processing the block pointed
2739 * to by the `*p` memory pointer. */
2744 static int process_block(struct archive_read* a) {
2746 struct rar5* rar = get_context(a);
2749 /* If we don't have any data to be processed, this most probably means
2750 * we need to switch to the next volume. */
2751 if(rar->main.volume && rar->file.bytes_remaining == 0) {
2752 ret = advance_multivolume(a);
2753 if(ret != ARCHIVE_OK)
2757 if(rar->cstate.block_parsing_finished) {
2760 rar->cstate.block_parsing_finished = 0;
2762 /* The header size won't be bigger than 6 bytes. */
2763 if(!read_ahead(a, 6, &p)) {
2764 /* Failed to prefetch data block header. */
2769 * Read block_size by parsing block header. Validate the header by
2770 * calculating CRC byte stored inside the header. Size of the header is
2771 * not constant (block size can be stored either in 1 or 2 bytes),
2772 * that's why block size is left out from the `compressed_block_header`
2773 * structure and returned by `parse_block_header` as the second
2776 ret = parse_block_header(a, p, &block_size, &rar->last_block_hdr);
2777 if(ret != ARCHIVE_OK)
2780 /* Skip block header. Next data is huffman tables, if present. */
2781 ssize_t to_skip = sizeof(struct compressed_block_header) +
2782 rar->last_block_hdr.block_flags.byte_count + 1;
2784 if(ARCHIVE_OK != consume(a, to_skip))
2787 rar->file.bytes_remaining -= to_skip;
2789 /* The block size gives information about the whole block size, but
2790 * the block could be stored in split form when using multi-volume
2791 * archives. In this case, the block size will be bigger than the
2792 * actual data stored in this file. Remaining part of the data will
2793 * be in another file. */
2795 ssize_t cur_block_size =
2796 rar5_min(rar->file.bytes_remaining, block_size);
2798 if(block_size > rar->file.bytes_remaining) {
2799 /* If current blocks' size is bigger than our data size, this
2800 * means we have a multivolume archive. In this case, skip
2801 * all base headers until the end of the file, proceed to next
2802 * "partXXX.rar" volume, find its signature, skip all headers up
2803 * to the first FILE base header, and continue from there.
2805 * Note that `merge_block` will update the `rar` context structure
2806 * quite extensively. */
2808 ret = merge_block(a, block_size, &p);
2809 if(ret != ARCHIVE_OK) {
2813 cur_block_size = block_size;
2815 /* Current stream pointer should be now directly *after* the
2816 * block that spanned through multiple archive files. `p` pointer
2817 * should have the data of the *whole* block (merged from
2818 * partial blocks stored in multiple archives files). */
2820 rar->cstate.switch_multivolume = 0;
2822 /* Read the whole block size into memory. This can take up to
2823 * 8 megabytes of memory in theoretical cases. Might be worth to
2824 * optimize this and use a standard chunk of 4kb's. */
2826 if(!read_ahead(a, 4 + cur_block_size, &p)) {
2827 /* Failed to prefetch block data. */
2832 rar->cstate.block_buf = p;
2833 rar->cstate.cur_block_size = cur_block_size;
2835 rar->bits.in_addr = 0;
2836 rar->bits.bit_addr = 0;
2838 if(rar->last_block_hdr.block_flags.is_table_present) {
2839 /* Load Huffman tables. */
2840 ret = parse_tables(a, rar, p);
2841 if(ret != ARCHIVE_OK) {
2842 /* Error during decompression of Huffman tables. */
2847 p = rar->cstate.block_buf;
2850 /* Uncompress the block, or a part of it, depending on how many bytes
2851 * will be generated by uncompressing the block.
2853 * In case too many bytes will be generated, calling this function again
2854 * will resume the uncompression operation. */
2855 ret = do_uncompress_block(a, p);
2856 if(ret != ARCHIVE_OK) {
2860 if(rar->cstate.block_parsing_finished &&
2861 rar->cstate.switch_multivolume == 0 &&
2862 rar->cstate.cur_block_size > 0)
2864 /* If we're processing a normal block, consume the whole block. We
2865 * can do this because we've already read the whole block to memory.
2867 if(ARCHIVE_OK != consume(a, rar->cstate.cur_block_size))
2868 return ARCHIVE_FATAL;
2870 rar->file.bytes_remaining -= rar->cstate.cur_block_size;
2871 } else if(rar->cstate.switch_multivolume) {
2872 /* Don't consume the block if we're doing multivolume processing.
2873 * The volume switching function will consume the proper count of
2876 rar->cstate.switch_multivolume = 0;
2882 /* Pops the `buf`, `size` and `offset` from the "data ready" stack.
2884 * Returns ARCHIVE_OK when those arguments can be used, ARCHIVE_RETRY
2885 * when there is no data on the stack. */
2886 static int use_data(struct rar5* rar, const void** buf, size_t* size,
2891 for(i = 0; i < rar5_countof(rar->cstate.dready); i++) {
2892 struct data_ready *d = &rar->cstate.dready[i];
2895 if(buf) *buf = d->buf;
2896 if(size) *size = d->size;
2897 if(offset) *offset = d->offset;
2904 return ARCHIVE_RETRY;
2907 /* Pushes the `buf`, `size` and `offset` arguments to the rar->cstate.dready
2908 * FIFO stack. Those values will be popped from this stack by the `use_data`
2910 static int push_data_ready(struct archive_read* a, struct rar5* rar,
2911 const uint8_t* buf, size_t size, int64_t offset)
2915 /* Don't push if we're in skip mode. This is needed because solid
2916 * streams need full processing even if we're skipping data. After fully
2917 * processing the stream, we need to discard the generated bytes, because
2918 * we're interested only in the side effect: building up the internal
2919 * window circular buffer. This window buffer will be used later during
2920 * unpacking of requested data. */
2925 if(offset != rar->file.last_offset + rar->file.last_size) {
2926 archive_set_error(&a->archive, ARCHIVE_ERRNO_PROGRAMMER, "Sanity "
2927 "check error: output stream is not continuous");
2928 return ARCHIVE_FATAL;
2931 for(i = 0; i < rar5_countof(rar->cstate.dready); i++) {
2932 struct data_ready* d = &rar->cstate.dready[i];
2939 /* These fields are used only in sanity checking. */
2940 rar->file.last_offset = offset;
2941 rar->file.last_size = size;
2943 /* Calculate the checksum of this new block before submitting
2944 * data to libarchive's engine. */
2945 update_crc(rar, d->buf, d->size);
2951 /* Program counter will reach this code if the `rar->cstate.data_ready`
2952 * stack will be filled up so that no new entries will be allowed. The
2953 * code shouldn't allow such situation to occur. So we treat this case
2954 * as an internal error. */
2956 archive_set_error(&a->archive, ARCHIVE_ERRNO_PROGRAMMER, "Error: "
2957 "premature end of data_ready stack");
2958 return ARCHIVE_FATAL;
2961 /* This function uncompresses the data that is stored in the <FILE> base
2964 * The FILE base block looks like this:
2966 * <header><huffman tables><block_1><block_2>...<block_n>
2968 * The <header> is a block header, that is parsed in parse_block_header().
2969 * It's a "compressed_block_header" structure, containing metadata needed
2970 * to know when we should stop looking for more <block_n> blocks.
2972 * <huffman tables> contain data needed to set up the huffman tables, needed
2973 * for the actual decompression.
2975 * Each <block_n> consists of series of literals:
2977 * <literal><literal><literal>...<literal>
2979 * Those literals generate the uncompression data. They operate on a circular
2980 * buffer, sometimes writing raw data into it, sometimes referencing
2981 * some previous data inside this buffer, and sometimes declaring a filter
2982 * that will need to be executed on the data stored in the circular buffer.
2983 * It all depends on the literal that is used.
2985 * Sometimes blocks produce output data, sometimes they don't. For example, for
2986 * some huge files that use lots of filters, sometimes a block is filled with
2987 * only filter declaration literals. Such blocks won't produce any data in the
2990 * Sometimes blocks will produce 4 bytes of data, and sometimes 1 megabyte,
2991 * because a literal can reference previously decompressed data. For example,
2992 * there can be a literal that says: 'append a byte 0xFE here', and after
2993 * it another literal can say 'append 1 megabyte of data from circular buffer
2994 * offset 0x12345'. This is how RAR format handles compressing repeated
2997 * The RAR compressor creates those literals and the actual efficiency of
2998 * compression depends on what those literals are. The literals can also
2999 * be seen as a kind of a non-turing-complete virtual machine that simply
3000 * tells the decompressor what it should do.
3003 static int do_uncompress_file(struct archive_read* a) {
3004 struct rar5* rar = get_context(a);
3006 int64_t max_end_pos;
3008 if(!rar->cstate.initialized) {
3009 /* Don't perform full context reinitialization if we're processing
3010 * a solid archive. */
3011 if(!rar->main.solid || !rar->cstate.window_buf) {
3015 rar->cstate.initialized = 1;
3018 if(rar->cstate.all_filters_applied == 1) {
3019 /* We use while(1) here, but standard case allows for just 1 iteration.
3020 * The loop will iterate if process_block() didn't generate any data at
3021 * all. This can happen if the block contains only filter definitions
3022 * (this is common in big files). */
3025 ret = process_block(a);
3026 if(ret == ARCHIVE_EOF || ret == ARCHIVE_FATAL)
3029 if(rar->cstate.last_write_ptr == rar->cstate.write_ptr) {
3030 /* The block didn't generate any new data, so just process
3035 /* The block has generated some new data, so break the loop. */
3040 /* Try to run filters. If filters won't be applied, it means that
3041 * insufficient data was generated. */
3042 ret = apply_filters(a);
3043 if(ret == ARCHIVE_RETRY) {
3045 } else if(ret == ARCHIVE_FATAL) {
3046 return ARCHIVE_FATAL;
3049 /* If apply_filters() will return ARCHIVE_OK, we can continue here. */
3051 if(cdeque_size(&rar->cstate.filters) > 0) {
3052 /* Check if we can write something before hitting first filter. */
3053 struct filter_info* flt;
3055 /* Get the block_start offset from the first filter. */
3056 if(CDE_OK != cdeque_front(&rar->cstate.filters, cdeque_filter_p(&flt)))
3058 archive_set_error(&a->archive, ARCHIVE_ERRNO_PROGRAMMER,
3059 "Can't read first filter");
3060 return ARCHIVE_FATAL;
3063 max_end_pos = rar5_min(flt->block_start, rar->cstate.write_ptr);
3065 /* There are no filters defined, or all filters were applied. This
3066 * means we can just store the data without any postprocessing. */
3067 max_end_pos = rar->cstate.write_ptr;
3070 if(max_end_pos == rar->cstate.last_write_ptr) {
3071 /* We can't write anything yet. The block uncompression function did
3072 * not generate enough data, and no filter can be applied. At the same
3073 * time we don't have any data that can be stored without filter
3074 * postprocessing. This means we need to wait for more data to be
3075 * generated, so we can apply the filters.
3077 * Signal the caller that we need more data to be able to do anything.
3079 return ARCHIVE_RETRY;
3081 /* We can write the data before hitting the first filter. So let's
3082 * do it. The push_window_data() function will effectively return
3083 * the selected data block to the user application. */
3084 push_window_data(a, rar, rar->cstate.last_write_ptr, max_end_pos);
3085 rar->cstate.last_write_ptr = max_end_pos;
3091 static int uncompress_file(struct archive_read* a) {
3095 /* Sometimes the uncompression function will return a 'retry' signal.
3096 * If this will happen, we have to retry the function. */
3097 ret = do_uncompress_file(a);
3098 if(ret != ARCHIVE_RETRY)
3104 static int do_unstore_file(struct archive_read* a,
3112 if(rar->file.bytes_remaining == 0 && rar->main.volume > 0 &&
3113 rar->generic.split_after > 0)
3117 rar->cstate.switch_multivolume = 1;
3118 ret = advance_multivolume(a);
3119 rar->cstate.switch_multivolume = 0;
3121 if(ret != ARCHIVE_OK) {
3122 /* Failed to advance to next multivolume archive file. */
3127 size_t to_read = rar5_min(rar->file.bytes_remaining, 64 * 1024);
3129 if(!read_ahead(a, to_read, &p)) {
3130 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT, "I/O error "
3131 "when unstoring file");
3132 return ARCHIVE_FATAL;
3135 if(ARCHIVE_OK != consume(a, to_read)) {
3140 if(size) *size = to_read;
3141 if(offset) *offset = rar->cstate.last_unstore_ptr;
3143 rar->file.bytes_remaining -= to_read;
3144 rar->cstate.last_unstore_ptr += to_read;
3146 update_crc(rar, p, to_read);
3150 static int do_unpack(struct archive_read* a, struct rar5* rar,
3151 const void** buf, size_t* size, int64_t* offset)
3153 enum COMPRESSION_METHOD {
3154 STORE = 0, FASTEST = 1, FAST = 2, NORMAL = 3, GOOD = 4, BEST = 5
3157 if(rar->file.service > 0) {
3158 return do_unstore_file(a, rar, buf, size, offset);
3160 switch(rar->cstate.method) {
3162 return do_unstore_file(a, rar, buf, size, offset);
3172 return uncompress_file(a);
3174 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
3175 "Compression method not supported: 0x%08x",
3176 rar->cstate.method);
3178 return ARCHIVE_FATAL;
3188 static int verify_checksums(struct archive_read* a) {
3190 struct rar5* rar = get_context(a);
3192 /* Check checksums only when actually unpacking the data. There's no need
3193 * to calculate checksum when we're skipping data in solid archives
3194 * (skipping in solid archives is the same thing as unpacking compressed
3195 * data and discarding the result). */
3197 if(!rar->skip_mode) {
3198 /* Always check checkums if we're not in skip mode */
3201 /* We can override the logic above with a compile-time option
3202 * NO_CRC_ON_SOLID_SKIP. This option is used during debugging, and it
3203 * will check checksums of unpacked data even when we're skipping it.
3206 #if defined CHECK_CRC_ON_SOLID_SKIP
3216 /* During unpacking, on each unpacked block we're calling the
3217 * update_crc() function. Since we are here, the unpacking process is
3218 * already over and we can check if calculated checksum (CRC32 or
3219 * BLAKE2sp) is the same as what is stored in the archive.
3221 if(rar->file.stored_crc32 > 0) {
3222 /* Check CRC32 only when the file contains a CRC32 value for this
3225 if(rar->file.calculated_crc32 != rar->file.stored_crc32) {
3226 /* Checksums do not match; the unpacked file is corrupted. */
3229 printf("Checksum error: CRC32 (was: %08x, expected: %08x)\n",
3230 rar->file.calculated_crc32, rar->file.stored_crc32);
3233 #ifndef DONT_FAIL_ON_CRC_ERROR
3234 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
3235 "Checksum error: CRC32");
3236 return ARCHIVE_FATAL;
3240 printf("Checksum OK: CRC32 (%08x/%08x)\n",
3241 rar->file.stored_crc32,
3242 rar->file.calculated_crc32);
3247 if(rar->file.has_blake2 > 0) {
3248 /* BLAKE2sp is an optional checksum algorithm that is added to
3249 * RARv5 archives when using the `-htb` switch during creation of
3252 * We now finalize the hash calculation by calling the `final`
3253 * function. This will generate the final hash value we can use to
3254 * compare it with the BLAKE2sp checksum that is stored in the
3257 * The return value of this `final` function is not very helpful,
3258 * as it guards only against improper use. This is why we're
3259 * explicitly ignoring it. */
3262 (void) blake2sp_final(&rar->file.b2state, b2_buf, 32);
3264 if(memcmp(&rar->file.blake2sp, b2_buf, 32) != 0) {
3265 #ifndef DONT_FAIL_ON_CRC_ERROR
3266 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
3267 "Checksum error: BLAKE2");
3269 return ARCHIVE_FATAL;
3275 /* Finalization for this file has been successfully completed. */
3279 static int verify_global_checksums(struct archive_read* a) {
3280 return verify_checksums(a);
3283 static int rar5_read_data(struct archive_read *a, const void **buff,
3284 size_t *size, int64_t *offset) {
3286 struct rar5* rar = get_context(a);
3288 if(!rar->skip_mode && (rar->cstate.last_write_ptr > rar->file.unpacked_size)) {
3289 archive_set_error(&a->archive, ARCHIVE_ERRNO_PROGRAMMER,
3290 "Unpacker has written too many bytes");
3291 return ARCHIVE_FATAL;
3294 ret = use_data(rar, buff, size, offset);
3295 if(ret == ARCHIVE_OK)
3298 ret = do_unpack(a, rar, buff, size, offset);
3299 if(ret != ARCHIVE_OK) {
3303 if(rar->file.bytes_remaining == 0 &&
3304 rar->cstate.last_write_ptr == rar->file.unpacked_size)
3306 /* If all bytes of current file were processed, run finalization.
3308 * Finalization will check checksum against proper values. If
3309 * some of the checksums will not match, we'll return an error
3310 * value in the last `archive_read_data` call to signal an error
3313 return verify_global_checksums(a);
3319 static int rar5_read_data_skip(struct archive_read *a) {
3320 struct rar5* rar = get_context(a);
3322 if(rar->main.solid) {
3323 /* In solid archives, instead of skipping the data, we need to extract
3324 * it, and dispose the result. The side effect of this operation will
3325 * be setting up the initial window buffer state needed to be able to
3326 * extract the selected file. */
3330 /* Make sure to process all blocks in the compressed stream. */
3331 while(rar->file.bytes_remaining > 0) {
3332 /* Setting the "skip mode" will allow us to skip checksum checks
3333 * during data skipping. Checking the checksum of skipped data
3334 * isn't really necessary and it's only slowing things down.
3336 * This is incremented instead of setting to 1 because this data
3337 * skipping function can be called recursively. */
3340 /* We're disposing 1 block of data, so we use triple NULLs in
3343 ret = rar5_read_data(a, NULL, NULL, NULL);
3345 /* Turn off "skip mode". */
3349 /* Propagate any potential error conditions to the caller. */
3354 /* In standard archives, we can just jump over the compressed stream.
3355 * Each file in non-solid archives starts from an empty window buffer.
3358 if(ARCHIVE_OK != consume(a, rar->file.bytes_remaining)) {
3359 return ARCHIVE_FATAL;
3362 rar->file.bytes_remaining = 0;
3368 static int64_t rar5_seek_data(struct archive_read *a, int64_t offset,
3375 /* We're a streaming unpacker, and we don't support seeking. */
3377 return ARCHIVE_FATAL;
3380 static int rar5_cleanup(struct archive_read *a) {
3381 struct rar5* rar = get_context(a);
3383 free(rar->cstate.window_buf);
3385 free(rar->cstate.filtered_buf);
3387 free(rar->vol.push_buf);
3390 cdeque_free(&rar->cstate.filters);
3393 a->format->data = NULL;
3398 static int rar5_capabilities(struct archive_read * a) {
3403 static int rar5_has_encrypted_entries(struct archive_read *_a) {
3406 /* Unsupported for now. */
3407 return ARCHIVE_READ_FORMAT_ENCRYPTION_UNSUPPORTED;
3410 static int rar5_init(struct rar5* rar) {
3413 memset(rar, 0, sizeof(struct rar5));
3415 /* Decrypt the magic signature pattern. Check the comment near the
3416 * `rar5_signature` symbol to read the rationale behind this. */
3418 if(rar5_signature[0] == 243) {
3419 for(i = 0; i < rar5_signature_size; i++) {
3420 rar5_signature[i] ^= 0xA1;
3424 if(CDE_OK != cdeque_init(&rar->cstate.filters, 8192))
3425 return ARCHIVE_FATAL;
3430 int archive_read_support_format_rar5(struct archive *_a) {
3431 struct archive_read* ar;
3435 if(ARCHIVE_OK != (ret = get_archive_read(_a, &ar)))
3438 rar = malloc(sizeof(*rar));
3440 archive_set_error(&ar->archive, ENOMEM, "Can't allocate rar5 data");
3441 return ARCHIVE_FATAL;
3444 if(ARCHIVE_OK != rar5_init(rar)) {
3445 archive_set_error(&ar->archive, ENOMEM, "Can't allocate rar5 filter "
3447 return ARCHIVE_FATAL;
3450 ret = __archive_read_register_format(ar,
3457 rar5_read_data_skip,
3461 rar5_has_encrypted_entries);
3463 if(ret != ARCHIVE_OK) {
3464 (void) rar5_cleanup(ar);