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? */
91 uint8_t eof : 1; /* Did we finish unpacking the file? */
93 /* Optional time fields. */
99 /* Optional hash fields. */
100 uint32_t stored_crc32;
101 uint32_t calculated_crc32;
102 uint8_t blake2sp[32];
103 blake2sp_state b2state;
108 FILTER_DELTA = 0, /* Generic pattern. */
109 FILTER_E8 = 1, /* Intel x86 code. */
110 FILTER_E8E9 = 2, /* Intel x86 code. */
111 FILTER_ARM = 3, /* ARM code. */
112 FILTER_AUDIO = 4, /* Audio filter, not used in RARv5. */
113 FILTER_RGB = 5, /* Color palette, not used in RARv5. */
114 FILTER_ITANIUM = 6, /* Intel's Itanium, not used in RARv5. */
115 FILTER_PPM = 7, /* Predictive pattern matching, not used in RARv5. */
125 ssize_t block_length;
144 struct decode_table {
146 int32_t decode_len[16];
147 uint32_t decode_pos[16];
149 uint8_t quick_len[1 << 10];
150 uint16_t quick_num[1 << 10];
151 uint16_t decode_num[306];
155 /* Flag used to specify if unpacker needs to reinitialize the uncompression
157 uint8_t initialized : 1;
159 /* Flag used when applying filters. */
160 uint8_t all_filters_applied : 1;
162 /* Flag used to skip file context reinitialization, used when unpacker is
163 * skipping through different multivolume archives. */
164 uint8_t switch_multivolume : 1;
166 /* Flag used to specify if unpacker has processed the whole data block or
167 * just a part of it. */
168 uint8_t block_parsing_finished : 1;
172 int flags; /* Uncompression flags. */
173 int method; /* Uncompression algorithm method. */
174 int version; /* Uncompression algorithm version. */
175 ssize_t window_size; /* Size of window_buf. */
176 uint8_t* window_buf; /* Circular buffer used during
178 uint8_t* filtered_buf; /* Buffer used when applying filters. */
179 const uint8_t* block_buf; /* Buffer used when merging blocks. */
180 size_t window_mask; /* Convenience field; window_size - 1. */
181 int64_t write_ptr; /* This amount of data has been unpacked in
182 the window buffer. */
183 int64_t last_write_ptr; /* This amount of data has been stored in
185 int64_t last_unstore_ptr; /* Counter of bytes extracted during
186 unstoring. This is separate from
187 last_write_ptr because of how SERVICE
188 base blocks are handled during skipping
189 in solid multiarchive archives. */
190 int64_t solid_offset; /* Additional offset inside the window
191 buffer, used in unpacking solid
193 ssize_t cur_block_size; /* Size of current data block. */
194 int last_len; /* Flag used in lzss decompression. */
196 /* Decode tables used during lzss uncompression. */
199 struct decode_table bd; /* huffman bit lengths */
201 struct decode_table ld; /* literals */
203 struct decode_table dd; /* distances */
205 struct decode_table ldd; /* lower bits of distances */
207 struct decode_table rd; /* repeating distances */
208 #define HUFF_TABLE_SIZE (HUFF_NC + HUFF_DC + HUFF_RC + HUFF_LDC)
210 /* Circular deque for storing filters. */
211 struct cdeque filters;
212 int64_t last_block_start; /* Used for sanity checking. */
213 ssize_t last_block_length; /* Used for sanity checking. */
215 /* Distance cache used during lzss uncompression. */
218 /* Data buffer stack. */
219 struct data_ready dready[2];
222 /* Bit reader state. */
224 int8_t bit_addr; /* Current bit pointer inside current byte. */
225 int in_addr; /* Current byte pointer. */
228 /* RARv5 block header structure. */
229 struct compressed_block_header {
232 uint8_t bit_size : 3;
233 uint8_t byte_count : 3;
234 uint8_t is_last_block : 1;
235 uint8_t is_table_present : 1;
237 uint8_t block_flags_u8;
243 /* RARv5 main header structure. */
245 /* Does the archive contain solid streams? */
248 /* If this a multi-file archive? */
256 struct generic_header {
257 uint8_t split_after : 1;
258 uint8_t split_before : 1;
269 /* Main context structure. */
271 int header_initialized;
273 /* Set to 1 if current file is positioned AFTER the magic value
274 * of the archive file. This is used in header reading functions. */
277 /* Set to not zero if we're in skip mode (either by calling rar5_data_skip
278 * function or when skipping over solid streams). Set to 0 when in
279 * extraction mode. This is used during checksum calculation functions. */
282 /* An offset to QuickOpen list. This is not supported by this unpacker,
283 * because we're focusing on streaming interface. QuickOpen is designed
284 * to make things quicker for non-stream interfaces, so it's not our
286 uint64_t qlist_offset;
288 /* An offset to additional Recovery data. This is not supported by this
289 * unpacker. Recovery data are additional Reed-Solomon codes that could
290 * be used to calculate bytes that are missing in archive or are
294 /* Various context variables grouped to different structures. */
295 struct generic_header generic;
296 struct main_header main;
297 struct comp_state cstate;
298 struct file_header file;
299 struct bit_reader bits;
300 struct multivolume vol;
302 /* The header of currently processed RARv5 block. Used in main
303 * decompression logic loop. */
304 struct compressed_block_header last_block_hdr;
307 /* Forward function declarations. */
309 static int verify_global_checksums(struct archive_read* a);
310 static int rar5_read_data_skip(struct archive_read *a);
311 static int push_data_ready(struct archive_read* a, struct rar5* rar,
312 const uint8_t* buf, size_t size, int64_t offset);
314 /* CDE_xxx = Circular Double Ended (Queue) return values. */
315 enum CDE_RETURN_VALUES {
316 CDE_OK, CDE_ALLOC, CDE_PARAM, CDE_OUT_OF_BOUNDS,
319 /* Clears the contents of this circular deque. */
320 static void cdeque_clear(struct cdeque* d) {
326 /* Creates a new circular deque object. Capacity must be power of 2: 8, 16, 32,
327 * 64, 256, etc. When the user will add another item above current capacity,
328 * the circular deque will overwrite the oldest entry. */
329 static int cdeque_init(struct cdeque* d, int max_capacity_power_of_2) {
330 if(d == NULL || max_capacity_power_of_2 == 0)
333 d->cap_mask = max_capacity_power_of_2 - 1;
336 if((max_capacity_power_of_2 & d->cap_mask) > 0)
340 d->arr = malloc(sizeof(void*) * max_capacity_power_of_2);
342 return d->arr ? CDE_OK : CDE_ALLOC;
345 /* Return the current size (not capacity) of circular deque `d`. */
346 static size_t cdeque_size(struct cdeque* d) {
350 /* Returns the first element of current circular deque. Note that this function
351 * doesn't perform any bounds checking. If you need bounds checking, use
352 * `cdeque_front()` function instead. */
353 static void cdeque_front_fast(struct cdeque* d, void** value) {
354 *value = (void*) d->arr[d->beg_pos];
357 /* Returns the first element of current circular deque. This function
358 * performs bounds checking. */
359 static int cdeque_front(struct cdeque* d, void** value) {
361 cdeque_front_fast(d, value);
364 return CDE_OUT_OF_BOUNDS;
367 /* Pushes a new element into the end of this circular deque object. If current
368 * size will exceed capacity, the oldest element will be overwritten. */
369 static int cdeque_push_back(struct cdeque* d, void* item) {
373 if(d->size == d->cap_mask + 1)
374 return CDE_OUT_OF_BOUNDS;
376 d->arr[d->end_pos] = (size_t) item;
377 d->end_pos = (d->end_pos + 1) & d->cap_mask;
383 /* Pops a front element of this circular deque object and returns its value.
384 * This function doesn't perform any bounds checking. */
385 static void cdeque_pop_front_fast(struct cdeque* d, void** value) {
386 *value = (void*) d->arr[d->beg_pos];
387 d->beg_pos = (d->beg_pos + 1) & d->cap_mask;
391 /* Pops a front element of this circular deque object and returns its value.
392 * This function performs bounds checking. */
393 static int cdeque_pop_front(struct cdeque* d, void** value) {
398 return CDE_OUT_OF_BOUNDS;
400 cdeque_pop_front_fast(d, value);
404 /* Convenience function to cast filter_info** to void **. */
405 static void** cdeque_filter_p(struct filter_info** f) {
406 return (void**) (size_t) f;
409 /* Convenience function to cast filter_info* to void *. */
410 static void* cdeque_filter(struct filter_info* f) {
411 return (void**) (size_t) f;
414 /* Destroys this circular deque object. Deallocates the memory of the collection
415 * buffer, but doesn't deallocate the memory of any pointer passed to this
416 * deque as a value. */
417 static void cdeque_free(struct cdeque* d) {
432 static inline struct rar5* get_context(struct archive_read* a) {
433 return (struct rar5*) a->format->data;
436 // TODO: make sure these functions return a little endian number
438 /* Convenience functions used by filter implementations. */
440 static uint32_t read_filter_data(struct rar5* rar, uint32_t offset) {
441 uint32_t* dptr = (uint32_t*) &rar->cstate.window_buf[offset];
442 // TODO: bswap if big endian
446 static void write_filter_data(struct rar5* rar, uint32_t offset,
449 uint32_t* dptr = (uint32_t*) &rar->cstate.filtered_buf[offset];
450 // TODO: bswap if big endian
454 static void circular_memcpy(uint8_t* dst, uint8_t* window, const int mask,
455 int64_t start, int64_t end)
457 if((start & mask) > (end & mask)) {
458 ssize_t len1 = mask + 1 - (start & mask);
459 ssize_t len2 = end & mask;
461 memcpy(dst, &window[start & mask], len1);
462 memcpy(dst + len1, window, len2);
464 memcpy(dst, &window[start & mask], (size_t) (end - start));
468 /* Allocates a new filter descriptor and adds it to the filter array. */
469 static struct filter_info* add_new_filter(struct rar5* rar) {
470 struct filter_info* f =
471 (struct filter_info*) calloc(1, sizeof(struct filter_info));
477 cdeque_push_back(&rar->cstate.filters, cdeque_filter(f));
481 static int run_delta_filter(struct rar5* rar, struct filter_info* flt) {
483 ssize_t dest_pos, src_pos = 0;
485 for(i = 0; i < flt->channels; i++) {
486 uint8_t prev_byte = 0;
488 dest_pos < flt->block_length;
489 dest_pos += flt->channels)
493 byte = rar->cstate.window_buf[(rar->cstate.solid_offset +
494 flt->block_start + src_pos) & rar->cstate.window_mask];
497 rar->cstate.filtered_buf[dest_pos] = prev_byte;
505 static int run_e8e9_filter(struct rar5* rar, struct filter_info* flt,
508 const uint32_t file_size = 0x1000000;
511 circular_memcpy(rar->cstate.filtered_buf,
512 rar->cstate.window_buf,
513 rar->cstate.window_mask,
514 rar->cstate.solid_offset + flt->block_start,
515 rar->cstate.solid_offset + flt->block_start + flt->block_length);
517 for(i = 0; i < flt->block_length - 4;) {
518 uint8_t b = rar->cstate.window_buf[(rar->cstate.solid_offset +
519 flt->block_start + i++) & rar->cstate.window_mask];
521 /* 0xE8 = x86's call <relative_addr_uint32> (function call)
522 * 0xE9 = x86's jmp <relative_addr_uint32> (unconditional jump) */
523 if(b == 0xE8 || (extended && b == 0xE9)) {
526 uint32_t offset = (i + flt->block_start) % file_size;
528 addr = read_filter_data(rar, (rar->cstate.solid_offset +
529 flt->block_start + i) & rar->cstate.window_mask);
531 if(addr & 0x80000000) {
532 if(((addr + offset) & 0x80000000) == 0) {
533 write_filter_data(rar, i, addr + file_size);
536 if((addr - file_size) & 0x80000000) {
537 uint32_t naddr = addr - offset;
538 write_filter_data(rar, i, naddr);
549 static int run_arm_filter(struct rar5* rar, struct filter_info* flt) {
552 const int mask = rar->cstate.window_mask;
554 circular_memcpy(rar->cstate.filtered_buf,
555 rar->cstate.window_buf,
556 rar->cstate.window_mask,
557 rar->cstate.solid_offset + flt->block_start,
558 rar->cstate.solid_offset + flt->block_start + flt->block_length);
560 for(i = 0; i < flt->block_length - 3; i += 4) {
561 uint8_t* b = &rar->cstate.window_buf[(rar->cstate.solid_offset +
562 flt->block_start + i) & mask];
565 /* 0xEB = ARM's BL (branch + link) instruction. */
566 offset = read_filter_data(rar, (rar->cstate.solid_offset +
567 flt->block_start + i) & mask) & 0x00ffffff;
569 offset -= (uint32_t) ((i + flt->block_start) / 4);
570 offset = (offset & 0x00ffffff) | 0xeb000000;
571 write_filter_data(rar, i, offset);
578 static int run_filter(struct archive_read* a, struct filter_info* flt) {
580 struct rar5* rar = get_context(a);
582 if(rar->cstate.filtered_buf)
583 free(rar->cstate.filtered_buf);
585 rar->cstate.filtered_buf = malloc(flt->block_length);
586 if(!rar->cstate.filtered_buf) {
587 archive_set_error(&a->archive, ENOMEM, "Can't allocate memory for "
589 return ARCHIVE_FATAL;
594 ret = run_delta_filter(rar, flt);
600 ret = run_e8e9_filter(rar, flt, flt->type == FILTER_E8E9);
604 ret = run_arm_filter(rar, flt);
608 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
609 "Unsupported filter type: 0x%02x", flt->type);
610 return ARCHIVE_FATAL;
613 if(ret != ARCHIVE_OK) {
614 /* Filter has failed. */
618 if(ARCHIVE_OK != push_data_ready(a, rar, rar->cstate.filtered_buf,
619 flt->block_length, rar->cstate.last_write_ptr))
621 archive_set_error(&a->archive, ARCHIVE_ERRNO_PROGRAMMER,
622 "Stack overflow when submitting unpacked data");
624 return ARCHIVE_FATAL;
627 rar->cstate.last_write_ptr += flt->block_length;
631 /* The `push_data` function submits the selected data range to the user.
632 * Next call of `use_data` will use the pointer, size and offset arguments
633 * that are specified here. These arguments are pushed to the FIFO stack here,
634 * and popped from the stack by the `use_data` function. */
635 static void push_data(struct archive_read* a, struct rar5* rar,
636 const uint8_t* buf, int64_t idx_begin, int64_t idx_end)
638 const int wmask = rar->cstate.window_mask;
639 const ssize_t solid_write_ptr = (rar->cstate.solid_offset +
640 rar->cstate.last_write_ptr) & wmask;
642 idx_begin += rar->cstate.solid_offset;
643 idx_end += rar->cstate.solid_offset;
645 /* Check if our unpacked data is wrapped inside the window circular buffer.
646 * If it's not wrapped, it can be copied out by using a single memcpy,
647 * but when it's wrapped, we need to copy the first part with one
648 * memcpy, and the second part with another memcpy. */
650 if((idx_begin & wmask) > (idx_end & wmask)) {
651 /* The data is wrapped (begin offset sis bigger than end offset). */
652 const ssize_t frag1_size = rar->cstate.window_size - (idx_begin & wmask);
653 const ssize_t frag2_size = idx_end & wmask;
655 /* Copy the first part of the buffer first. */
656 push_data_ready(a, rar, buf + solid_write_ptr, frag1_size,
657 rar->cstate.last_write_ptr);
659 /* Copy the second part of the buffer. */
660 push_data_ready(a, rar, buf, frag2_size,
661 rar->cstate.last_write_ptr + frag1_size);
663 rar->cstate.last_write_ptr += frag1_size + frag2_size;
665 /* Data is not wrapped, so we can just use one call to copy the
667 push_data_ready(a, rar,
668 buf + solid_write_ptr,
669 (idx_end - idx_begin) & wmask,
670 rar->cstate.last_write_ptr);
672 rar->cstate.last_write_ptr += idx_end - idx_begin;
676 /* Convenience function that submits the data to the user. It uses the
677 * unpack window buffer as a source location. */
678 static void push_window_data(struct archive_read* a, struct rar5* rar,
679 int64_t idx_begin, int64_t idx_end)
681 push_data(a, rar, rar->cstate.window_buf, idx_begin, idx_end);
684 static int apply_filters(struct archive_read* a) {
685 struct filter_info* flt;
686 struct rar5* rar = get_context(a);
689 rar->cstate.all_filters_applied = 0;
691 /* Get the first filter that can be applied to our data. The data needs to
692 * be fully unpacked before the filter can be run. */
694 cdeque_front(&rar->cstate.filters, cdeque_filter_p(&flt)))
696 /* Check if our unpacked data fully covers this filter's range. */
697 if(rar->cstate.write_ptr > flt->block_start &&
698 rar->cstate.write_ptr >= flt->block_start + flt->block_length)
700 /* Check if we have some data pending to be written right before
701 * the filter's start offset. */
702 if(rar->cstate.last_write_ptr == flt->block_start) {
703 /* Run the filter specified by descriptor `flt`. */
704 ret = run_filter(a, flt);
705 if(ret != ARCHIVE_OK) {
706 /* Filter failure, return error. */
710 /* Filter descriptor won't be needed anymore after it's used,
711 * so remove it from the filter list and free its memory. */
712 (void) cdeque_pop_front(&rar->cstate.filters,
713 cdeque_filter_p(&flt));
717 /* We can't run filters yet, dump the memory right before the
719 push_window_data(a, rar, rar->cstate.last_write_ptr,
723 /* Return 'filter applied or not needed' state to the caller. */
724 return ARCHIVE_RETRY;
728 rar->cstate.all_filters_applied = 1;
732 static void dist_cache_push(struct rar5* rar, int value) {
733 int* q = rar->cstate.dist_cache;
741 static int dist_cache_touch(struct rar5* rar, int idx) {
742 int* q = rar->cstate.dist_cache;
743 int i, dist = q[idx];
745 for(i = idx; i > 0; i--)
752 static void free_filters(struct rar5* rar) {
753 struct cdeque* d = &rar->cstate.filters;
755 /* Free any remaining filters. All filters should be naturally consumed by
756 * the unpacking function, so remaining filters after unpacking normally
757 * mean that unpacking wasn't successful. But still of course we shouldn't
758 * leak memory in such case. */
760 /* cdeque_size() is a fast operation, so we can use it as a loop
762 while(cdeque_size(d) > 0) {
763 struct filter_info* f = NULL;
765 /* Pop_front will also decrease the collection's size. */
766 if(CDE_OK == cdeque_pop_front(d, cdeque_filter_p(&f)) && f != NULL)
772 /* Also clear out the variables needed for sanity checking. */
773 rar->cstate.last_block_start = 0;
774 rar->cstate.last_block_length = 0;
777 static void reset_file_context(struct rar5* rar) {
778 memset(&rar->file, 0, sizeof(rar->file));
779 blake2sp_init(&rar->file.b2state, 32);
781 if(rar->main.solid) {
782 rar->cstate.solid_offset += rar->cstate.write_ptr;
784 rar->cstate.solid_offset = 0;
787 rar->cstate.write_ptr = 0;
788 rar->cstate.last_write_ptr = 0;
789 rar->cstate.last_unstore_ptr = 0;
794 static inline int get_archive_read(struct archive* a,
795 struct archive_read** ar)
797 *ar = (struct archive_read*) a;
798 archive_check_magic(a, ARCHIVE_READ_MAGIC, ARCHIVE_STATE_NEW,
799 "archive_read_support_format_rar5");
804 static int read_ahead(struct archive_read* a, size_t how_many,
811 *ptr = __archive_read_ahead(a, how_many, &avail);
820 static int consume(struct archive_read* a, int64_t how_many) {
824 how_many == __archive_read_consume(a, how_many)
832 * Read a RAR5 variable sized numeric value. This value will be stored in
833 * `pvalue`. The `pvalue_len` argument points to a variable that will receive
834 * the byte count that was consumed in order to decode the `pvalue` value, plus
837 * pvalue_len is optional and can be NULL.
839 * NOTE: if `pvalue_len` is NOT NULL, the caller needs to manually consume
840 * the number of bytes that `pvalue_len` value contains. If the `pvalue_len`
841 * is NULL, this consuming operation is done automatically.
843 * Returns 1 if *pvalue was successfully read.
844 * Returns 0 if there was an error. In this case, *pvalue contains an
848 static int read_var(struct archive_read* a, uint64_t* pvalue,
849 uint64_t* pvalue_len)
856 /* We will read maximum of 8 bytes. We don't have to handle the situation
857 * to read the RAR5 variable-sized value stored at the end of the file,
858 * because such situation will never happen. */
859 if(!read_ahead(a, 8, &p))
862 for(shift = 0, i = 0; i < 8; i++, shift += 7) {
865 /* Strip the MSB from the input byte and add the resulting number
866 * to the `result`. */
867 result += (b & 0x7F) << shift;
869 /* MSB set to 1 means we need to continue decoding process. MSB set
870 * to 0 means we're done.
872 * This conditional checks for the second case. */
873 if((b & 0x80) == 0) {
878 /* If the caller has passed the `pvalue_len` pointer, store the
879 * number of consumed bytes in it and do NOT consume those bytes,
880 * since the caller has all the information it needs to perform
881 * the consuming process itself. */
885 /* If the caller did not provide the `pvalue_len` pointer,
886 * it will not have the possibility to advance the file
887 * pointer, because it will not know how many bytes it needs
888 * to consume. This is why we handle such situation here
890 if(ARCHIVE_OK != consume(a, 1 + i)) {
895 /* End of decoding process, return success. */
900 /* The decoded value takes the maximum number of 8 bytes. It's a maximum
901 * number of bytes, so end decoding process here even if the first bit
902 * of last byte is 1. */
910 if(ARCHIVE_OK != consume(a, 9)) {
918 static int read_var_sized(struct archive_read* a, size_t* pvalue,
924 const int ret = pvalue_len
925 ? read_var(a, &v, &v_size)
926 : read_var(a, &v, NULL);
928 if(ret == 1 && pvalue) {
929 *pvalue = (size_t) v;
933 /* Possible data truncation should be safe. */
934 *pvalue_len = (size_t) v_size;
940 static int read_bits_32(struct rar5* rar, const uint8_t* p, uint32_t* value) {
941 uint32_t bits = p[rar->bits.in_addr] << 24;
942 bits |= p[rar->bits.in_addr + 1] << 16;
943 bits |= p[rar->bits.in_addr + 2] << 8;
944 bits |= p[rar->bits.in_addr + 3];
945 bits <<= rar->bits.bit_addr;
946 bits |= p[rar->bits.in_addr + 4] >> (8 - rar->bits.bit_addr);
951 static int read_bits_16(struct rar5* rar, const uint8_t* p, uint16_t* value) {
952 int bits = (int) p[rar->bits.in_addr] << 16;
953 bits |= (int) p[rar->bits.in_addr + 1] << 8;
954 bits |= (int) p[rar->bits.in_addr + 2];
955 bits >>= (8 - rar->bits.bit_addr);
956 *value = bits & 0xffff;
960 static void skip_bits(struct rar5* rar, int bits) {
961 const int new_bits = rar->bits.bit_addr + bits;
962 rar->bits.in_addr += new_bits >> 3;
963 rar->bits.bit_addr = new_bits & 7;
967 static int read_consume_bits(struct rar5* rar, const uint8_t* p, int n,
973 if(n == 0 || n > 16) {
974 /* This is a programmer error and should never happen in runtime. */
975 return ARCHIVE_FATAL;
978 ret = read_bits_16(rar, p, &v);
979 if(ret != ARCHIVE_OK)
993 static int read_u32(struct archive_read* a, uint32_t* pvalue) {
995 if(!read_ahead(a, 4, &p))
998 *pvalue = *(const uint32_t*)p;
1000 return ARCHIVE_OK == consume(a, 4) ? 1 : 0;
1003 static int read_u64(struct archive_read* a, uint64_t* pvalue) {
1005 if(!read_ahead(a, 8, &p))
1008 *pvalue = *(const uint64_t*)p;
1010 return ARCHIVE_OK == consume(a, 8) ? 1 : 0;
1013 static int bid_standard(struct archive_read* a) {
1016 if(!read_ahead(a, rar5_signature_size, &p))
1019 if(!memcmp(rar5_signature, p, rar5_signature_size))
1025 static int rar5_bid(struct archive_read* a, int best_bid) {
1031 my_bid = bid_standard(a);
1039 static int rar5_options(struct archive_read *a, const char *key, const char *val) {
1044 /* No options supported in this version. Return the ARCHIVE_WARN code to
1045 * signal the options supervisor that the unpacker didn't handle setting
1048 return ARCHIVE_WARN;
1051 static void init_header(struct archive_read* a) {
1052 a->archive.archive_format = ARCHIVE_FORMAT_RAR_V5;
1053 a->archive.archive_format_name = "RAR5";
1057 HFL_EXTRA_DATA = 0x0001, HFL_DATA = 0x0002, HFL_SKIP_IF_UNKNOWN = 0x0004,
1058 HFL_SPLIT_BEFORE = 0x0008, HFL_SPLIT_AFTER = 0x0010, HFL_CHILD = 0x0020,
1059 HFL_INHERITED = 0x0040
1062 static int process_main_locator_extra_block(struct archive_read* a,
1065 uint64_t locator_flags;
1067 if(!read_var(a, &locator_flags, NULL)) {
1071 enum LOCATOR_FLAGS {
1072 QLIST = 0x01, RECOVERY = 0x02,
1075 if(locator_flags & QLIST) {
1076 if(!read_var(a, &rar->qlist_offset, NULL)) {
1080 /* qlist is not used */
1083 if(locator_flags & RECOVERY) {
1084 if(!read_var(a, &rar->rr_offset, NULL)) {
1088 /* rr is not used */
1094 static int parse_file_extra_hash(struct archive_read* a, struct rar5* rar,
1095 ssize_t* extra_data_size)
1100 if(!read_var_sized(a, &hash_type, &value_len))
1103 *extra_data_size -= value_len;
1104 if(ARCHIVE_OK != consume(a, value_len)) {
1112 /* The file uses BLAKE2sp checksum algorithm instead of plain old
1114 if(hash_type == BLAKE2sp) {
1116 const int hash_size = sizeof(rar->file.blake2sp);
1118 if(!read_ahead(a, hash_size, &p))
1121 rar->file.has_blake2 = 1;
1122 memcpy(&rar->file.blake2sp, p, hash_size);
1124 if(ARCHIVE_OK != consume(a, hash_size)) {
1128 *extra_data_size -= hash_size;
1130 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
1131 "Unsupported hash type (0x%02x)", (int) hash_type);
1132 return ARCHIVE_FATAL;
1138 static uint64_t time_win_to_unix(uint64_t win_time) {
1139 const size_t ns_in_sec = 10000000;
1140 const uint64_t sec_to_unix = 11644473600LL;
1141 return win_time / ns_in_sec - sec_to_unix;
1144 static int parse_htime_item(struct archive_read* a, char unix_time,
1145 uint64_t* where, ssize_t* extra_data_size)
1149 if(!read_u32(a, &time_val))
1152 *extra_data_size -= 4;
1153 *where = (uint64_t) time_val;
1155 uint64_t windows_time;
1156 if(!read_u64(a, &windows_time))
1159 *where = time_win_to_unix(windows_time);
1160 *extra_data_size -= 8;
1166 static int parse_file_extra_htime(struct archive_read* a,
1167 struct archive_entry* e, struct rar5* rar,
1168 ssize_t* extra_data_size)
1182 if(!read_var_sized(a, &flags, &value_len))
1185 *extra_data_size -= value_len;
1186 if(ARCHIVE_OK != consume(a, value_len)) {
1190 unix_time = flags & IS_UNIX;
1192 if(flags & HAS_MTIME) {
1193 parse_htime_item(a, unix_time, &rar->file.e_mtime, extra_data_size);
1194 archive_entry_set_mtime(e, rar->file.e_mtime, 0);
1197 if(flags & HAS_CTIME) {
1198 parse_htime_item(a, unix_time, &rar->file.e_ctime, extra_data_size);
1199 archive_entry_set_ctime(e, rar->file.e_ctime, 0);
1202 if(flags & HAS_ATIME) {
1203 parse_htime_item(a, unix_time, &rar->file.e_atime, extra_data_size);
1204 archive_entry_set_atime(e, rar->file.e_atime, 0);
1207 if(flags & HAS_UNIX_NS) {
1208 if(!read_u32(a, &rar->file.e_unix_ns))
1211 *extra_data_size -= 4;
1217 static int process_head_file_extra(struct archive_read* a,
1218 struct archive_entry* e, struct rar5* rar,
1219 ssize_t extra_data_size)
1221 size_t extra_field_size;
1222 size_t extra_field_id = 0;
1223 int ret = ARCHIVE_FATAL;
1227 CRYPT = 0x01, HASH = 0x02, HTIME = 0x03, VERSION_ = 0x04,
1228 REDIR = 0x05, UOWNER = 0x06, SUBDATA = 0x07
1231 while(extra_data_size > 0) {
1232 if(!read_var_sized(a, &extra_field_size, &var_size))
1235 extra_data_size -= var_size;
1236 if(ARCHIVE_OK != consume(a, var_size)) {
1240 if(!read_var_sized(a, &extra_field_id, &var_size))
1243 extra_data_size -= var_size;
1244 if(ARCHIVE_OK != consume(a, var_size)) {
1248 switch(extra_field_id) {
1250 ret = parse_file_extra_hash(a, rar, &extra_data_size);
1253 ret = parse_file_extra_htime(a, e, rar, &extra_data_size);
1266 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
1267 "Unknown extra field in file/service block: 0x%02x",
1268 (int) extra_field_id);
1269 return ARCHIVE_FATAL;
1273 if(ret != ARCHIVE_OK) {
1274 /* Attribute not implemented. */
1281 static int process_head_file(struct archive_read* a, struct rar5* rar,
1282 struct archive_entry* entry, size_t block_flags)
1284 ssize_t extra_data_size = 0;
1285 size_t data_size = 0;
1286 size_t file_flags = 0;
1287 size_t file_attr = 0;
1288 size_t compression_info = 0;
1290 size_t name_size = 0;
1291 uint64_t unpacked_size;
1292 uint32_t mtime = 0, crc = 0;
1293 int c_method = 0, c_version = 0, is_dir;
1294 char name_utf8_buf[2048 * 4];
1297 memset(entry, 0, sizeof(struct archive_entry));
1299 /* Do not reset file context if we're switching archives. */
1300 if(!rar->cstate.switch_multivolume) {
1301 reset_file_context(rar);
1304 if(block_flags & HFL_EXTRA_DATA) {
1305 size_t edata_size = 0;
1306 if(!read_var_sized(a, &edata_size, NULL))
1309 /* Intentional type cast from unsigned to signed. */
1310 extra_data_size = (ssize_t) edata_size;
1313 if(block_flags & HFL_DATA) {
1314 if(!read_var_sized(a, &data_size, NULL))
1317 rar->file.bytes_remaining = data_size;
1319 rar->file.bytes_remaining = 0;
1321 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
1322 "no data found in file/service block");
1323 return ARCHIVE_FATAL;
1327 DIRECTORY = 0x0001, UTIME = 0x0002, CRC32 = 0x0004,
1328 UNKNOWN_UNPACKED_SIZE = 0x0008,
1331 enum COMP_INFO_FLAGS {
1335 if(!read_var_sized(a, &file_flags, NULL))
1338 if(!read_var(a, &unpacked_size, NULL))
1341 if(file_flags & UNKNOWN_UNPACKED_SIZE) {
1342 archive_set_error(&a->archive, ARCHIVE_ERRNO_PROGRAMMER,
1343 "Files with unknown unpacked size are not supported");
1344 return ARCHIVE_FATAL;
1347 is_dir = (int) (file_flags & DIRECTORY);
1349 if(!read_var_sized(a, &file_attr, NULL))
1352 if(file_flags & UTIME) {
1353 if(!read_u32(a, &mtime))
1357 if(file_flags & CRC32) {
1358 if(!read_u32(a, &crc))
1362 if(!read_var_sized(a, &compression_info, NULL))
1365 c_method = (int) (compression_info >> 7) & 0x7;
1366 c_version = (int) (compression_info & 0x3f);
1368 rar->cstate.window_size = is_dir ?
1370 g_unpack_window_size << ((compression_info >> 10) & 15);
1371 rar->cstate.method = c_method;
1372 rar->cstate.version = c_version + 50;
1374 rar->file.solid = (compression_info & SOLID) > 0;
1375 rar->file.service = 0;
1377 if(!read_var_sized(a, &host_os, NULL))
1385 if(host_os == HOST_WINDOWS) {
1386 /* Host OS is Windows */
1388 unsigned short mode = 0660;
1395 archive_entry_set_mode(entry, mode);
1396 } else if(host_os == HOST_UNIX) {
1397 /* Host OS is Unix */
1398 archive_entry_set_mode(entry, (unsigned short) file_attr);
1400 /* Unknown host OS */
1401 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
1402 "Unsupported Host OS: 0x%02x", (int) host_os);
1404 return ARCHIVE_FATAL;
1407 if(!read_var_sized(a, &name_size, NULL))
1410 if(!read_ahead(a, name_size, &p))
1413 if(name_size > 2047) {
1414 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
1415 "Filename is too long");
1417 return ARCHIVE_FATAL;
1420 if(name_size == 0) {
1421 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
1422 "No filename specified");
1424 return ARCHIVE_FATAL;
1427 memcpy(name_utf8_buf, p, name_size);
1428 name_utf8_buf[name_size] = 0;
1429 if(ARCHIVE_OK != consume(a, name_size)) {
1433 if(extra_data_size > 0) {
1434 int ret = process_head_file_extra(a, entry, rar, extra_data_size);
1437 if(extra_data_size < 0) {
1438 archive_set_error(&a->archive, ARCHIVE_ERRNO_PROGRAMMER,
1439 "File extra data size is not zero");
1440 return ARCHIVE_FATAL;
1443 if(ret != ARCHIVE_OK)
1447 if((file_flags & UNKNOWN_UNPACKED_SIZE) == 0) {
1448 rar->file.unpacked_size = (ssize_t) unpacked_size;
1449 archive_entry_set_size(entry, unpacked_size);
1452 if(file_flags & UTIME) {
1453 archive_entry_set_mtime(entry, (time_t) mtime, 0);
1456 if(file_flags & CRC32) {
1457 rar->file.stored_crc32 = crc;
1460 archive_entry_update_pathname_utf8(entry, name_utf8_buf);
1462 if(!rar->cstate.switch_multivolume) {
1463 /* Do not reinitialize unpacking state if we're switching archives. */
1464 rar->cstate.block_parsing_finished = 1;
1465 rar->cstate.all_filters_applied = 1;
1466 rar->cstate.initialized = 0;
1469 if(rar->generic.split_before > 0) {
1470 /* If now we're standing on a header that has a 'split before' mark,
1471 * it means we're standing on a 'continuation' file header. Signal
1472 * the caller that if it wants to move to another file, it must call
1473 * rar5_read_header() function again. */
1475 return ARCHIVE_RETRY;
1481 static int process_head_service(struct archive_read* a, struct rar5* rar,
1482 struct archive_entry* entry, size_t block_flags)
1484 /* Process this SERVICE block the same way as FILE blocks. */
1485 int ret = process_head_file(a, rar, entry, block_flags);
1486 if(ret != ARCHIVE_OK)
1489 rar->file.service = 1;
1491 /* But skip the data part automatically. It's no use for the user anyway.
1492 * It contains only service data, not even needed to properly unpack the
1494 ret = rar5_read_data_skip(a);
1495 if(ret != ARCHIVE_OK)
1498 /* After skipping, try parsing another block automatically. */
1499 return ARCHIVE_RETRY;
1502 static int process_head_main(struct archive_read* a, struct rar5* rar,
1503 struct archive_entry* entry, size_t block_flags)
1508 size_t extra_data_size = 0;
1509 size_t extra_field_size = 0;
1510 size_t extra_field_id = 0;
1511 size_t archive_flags = 0;
1513 if(block_flags & HFL_EXTRA_DATA) {
1514 if(!read_var_sized(a, &extra_data_size, NULL))
1517 extra_data_size = 0;
1520 if(!read_var_sized(a, &archive_flags, NULL)) {
1525 VOLUME = 0x0001, /* multi-volume archive */
1526 VOLUME_NUMBER = 0x0002, /* volume number, first vol doesn't have it */
1527 SOLID = 0x0004, /* solid archive */
1528 PROTECT = 0x0008, /* contains Recovery info */
1529 LOCK = 0x0010, /* readonly flag, not used */
1532 rar->main.volume = (archive_flags & VOLUME) > 0;
1533 rar->main.solid = (archive_flags & SOLID) > 0;
1535 if(archive_flags & VOLUME_NUMBER) {
1537 if(!read_var_sized(a, &v, NULL)) {
1541 rar->main.vol_no = (int) v;
1543 rar->main.vol_no = 0;
1546 if(rar->vol.expected_vol_no > 0 &&
1547 rar->main.vol_no != rar->vol.expected_vol_no)
1549 /* Returning EOF instead of FATAL because of strange libarchive
1550 * behavior. When opening multiple files via
1551 * archive_read_open_filenames(), after reading up the whole last file,
1552 * the __archive_read_ahead function wraps up to the first archive
1553 * instead of returning EOF. */
1557 if(extra_data_size == 0) {
1562 if(!read_var_sized(a, &extra_field_size, NULL)) {
1566 if(!read_var_sized(a, &extra_field_id, NULL)) {
1570 if(extra_field_size == 0) {
1571 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
1572 "Invalid extra field size");
1573 return ARCHIVE_FATAL;
1577 // Just one attribute here.
1581 switch(extra_field_id) {
1583 ret = process_main_locator_extra_block(a, rar);
1584 if(ret != ARCHIVE_OK) {
1585 /* Error while parsing main locator extra block. */
1591 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
1592 "Unsupported extra type (0x%02x)", (int) extra_field_id);
1593 return ARCHIVE_FATAL;
1599 static int scan_for_signature(struct archive_read* a);
1601 /* Base block processing function. A 'base block' is a RARv5 header block
1602 * that tells the reader what kind of data is stored inside the block.
1604 * From the birds-eye view a RAR file looks file this:
1606 * <magic><base_block_1><base_block_2>...<base_block_n>
1608 * There are a few types of base blocks. Those types are specified inside
1609 * the 'switch' statement in this function. For example purposes, I'll write
1610 * how a standard RARv5 file could look like here:
1612 * <magic><MAIN><FILE><FILE><FILE><SERVICE><ENDARC>
1614 * The structure above could describe an archive file with 3 files in it,
1615 * one service "QuickOpen" block (that is ignored by this parser), and an
1616 * end of file base block marker.
1618 * If the file is stored in multiple archive files ("multiarchive"), it might
1621 * .part01.rar: <magic><MAIN><FILE><ENDARC>
1622 * .part02.rar: <magic><MAIN><FILE><ENDARC>
1623 * .part03.rar: <magic><MAIN><FILE><ENDARC>
1625 * This example could describe 3 RAR files that contain ONE archived file.
1626 * Or it could describe 3 RAR files that contain 3 different files. Or 3
1627 * RAR files than contain 2 files. It all depends what metadata is stored in
1628 * the headers of <FILE> blocks.
1630 * Each <FILE> block contains info about its size, the name of the file it's
1631 * storing inside, and whether this FILE block is a continuation block of
1632 * previous archive ('split before'), and is this FILE block should be
1633 * continued in another archive ('split after'). By parsing the 'split before'
1634 * and 'split after' flags, we're able to tell if multiple <FILE> base blocks
1635 * are describing one file, or multiple files (with the same filename, for
1638 * One thing to note is that if we're parsing the first <FILE> block, and
1639 * we see 'split after' flag, then we need to jump over to another <FILE>
1640 * block to be able to decompress rest of the data. To do this, we need
1641 * to skip the <ENDARC> block, then switch to another file, then skip the
1642 * <magic> block, <MAIN> block, and then we're standing on the proper
1646 static int process_base_block(struct archive_read* a,
1647 struct archive_entry* entry)
1649 struct rar5* rar = get_context(a);
1650 uint32_t hdr_crc, computed_crc;
1651 size_t raw_hdr_size = 0, hdr_size_len, hdr_size;
1652 size_t header_id = 0;
1653 size_t header_flags = 0;
1657 /* Skip any unprocessed data for this file. */
1658 if(rar->file.bytes_remaining) {
1659 ret = rar5_read_data_skip(a);
1660 if(ret != ARCHIVE_OK) {
1665 /* Read the expected CRC32 checksum. */
1666 if(!read_u32(a, &hdr_crc)) {
1670 /* Read header size. */
1671 if(!read_var_sized(a, &raw_hdr_size, &hdr_size_len)) {
1675 /* Sanity check, maximum header size for RAR5 is 2MB. */
1676 if(raw_hdr_size > (2 * 1024 * 1024)) {
1677 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
1678 "Base block header is too large");
1680 return ARCHIVE_FATAL;
1683 hdr_size = raw_hdr_size + hdr_size_len;
1685 /* Read the whole header data into memory, maximum memory use here is
1687 if(!read_ahead(a, hdr_size, &p)) {
1691 /* Verify the CRC32 of the header data. */
1692 computed_crc = (uint32_t) crc32(0, p, (int) hdr_size);
1693 if(computed_crc != hdr_crc) {
1694 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
1695 "Header CRC error");
1697 return ARCHIVE_FATAL;
1700 /* If the checksum is OK, we proceed with parsing. */
1701 if(ARCHIVE_OK != consume(a, hdr_size_len)) {
1705 if(!read_var_sized(a, &header_id, NULL))
1708 if(!read_var_sized(a, &header_flags, NULL))
1711 rar->generic.split_after = (header_flags & HFL_SPLIT_AFTER) > 0;
1712 rar->generic.split_before = (header_flags & HFL_SPLIT_BEFORE) > 0;
1713 rar->generic.size = hdr_size;
1714 rar->generic.last_header_id = header_id;
1715 rar->main.endarc = 0;
1717 /* Those are possible header ids in RARv5. */
1719 HEAD_MARK = 0x00, HEAD_MAIN = 0x01, HEAD_FILE = 0x02,
1720 HEAD_SERVICE = 0x03, HEAD_CRYPT = 0x04, HEAD_ENDARC = 0x05,
1721 HEAD_UNKNOWN = 0xff,
1726 ret = process_head_main(a, rar, entry, header_flags);
1728 /* Main header doesn't have any files in it, so it's pointless
1729 * to return to the caller. Retry to next header, which should be
1730 * HEAD_FILE/HEAD_SERVICE. */
1731 if(ret == ARCHIVE_OK)
1732 return ARCHIVE_RETRY;
1736 ret = process_head_service(a, rar, entry, header_flags);
1739 ret = process_head_file(a, rar, entry, header_flags);
1742 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
1743 "Encryption is not supported");
1744 return ARCHIVE_FATAL;
1746 rar->main.endarc = 1;
1748 /* After encountering an end of file marker, we need to take
1749 * into consideration if this archive is continued in another
1750 * file (i.e. is it part01.rar: is there a part02.rar?) */
1751 if(rar->main.volume) {
1752 /* In case there is part02.rar, position the read pointer
1753 * in a proper place, so we can resume parsing. */
1755 ret = scan_for_signature(a);
1756 if(ret == ARCHIVE_FATAL) {
1759 rar->vol.expected_vol_no = rar->main.vol_no + 1;
1768 if((header_flags & HFL_SKIP_IF_UNKNOWN) == 0) {
1769 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
1770 "Header type error");
1771 return ARCHIVE_FATAL;
1773 /* If the block is marked as 'skip if unknown', do as the flag
1774 * says: skip the block instead on failing on it. */
1775 return ARCHIVE_RETRY;
1781 archive_set_error(&a->archive, ARCHIVE_ERRNO_PROGRAMMER,
1782 "Internal unpacker error");
1783 return ARCHIVE_FATAL;
1787 static int skip_base_block(struct archive_read* a) {
1789 struct rar5* rar = get_context(a);
1791 struct archive_entry entry;
1792 ret = process_base_block(a, &entry);
1794 if(rar->generic.last_header_id == 2 && rar->generic.split_before > 0)
1797 if(ret == ARCHIVE_OK)
1798 return ARCHIVE_RETRY;
1803 static int rar5_read_header(struct archive_read *a,
1804 struct archive_entry *entry)
1806 struct rar5* rar = get_context(a);
1809 if(rar->header_initialized == 0) {
1811 rar->header_initialized = 1;
1814 if(rar->skipped_magic == 0) {
1815 if(ARCHIVE_OK != consume(a, rar5_signature_size)) {
1819 rar->skipped_magic = 1;
1823 ret = process_base_block(a, entry);
1824 } while(ret == ARCHIVE_RETRY ||
1825 (rar->main.endarc > 0 && ret == ARCHIVE_OK));
1830 static void init_unpack(struct rar5* rar) {
1831 rar->file.calculated_crc32 = 0;
1832 rar->cstate.window_mask = rar->cstate.window_size - 1;
1834 if(rar->cstate.window_buf)
1835 free(rar->cstate.window_buf);
1837 if(rar->cstate.filtered_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 /* Convenience 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 idx = num - 258;
2539 const int dist = dist_cache_touch(rar, idx);
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 rar->vol.push_buf = malloc(block_size);
2673 if(!rar->vol.push_buf) {
2674 archive_set_error(&a->archive, ENOMEM, "Can't allocate memory for a "
2675 "merge block buffer.");
2676 return ARCHIVE_FATAL;
2679 /* A single block can span across multiple multivolume archive files,
2680 * so we use a loop here. This loop will consume enough multivolume
2681 * archive files until the whole block is read. */
2684 /* Get the size of current block chunk in this multivolume archive
2685 * file and read it. */
2687 rar5_min(rar->file.bytes_remaining, block_size - partial_offset);
2689 if(cur_block_size == 0) {
2690 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
2691 "Encountered block size == 0 during block merge");
2692 return ARCHIVE_FATAL;
2695 if(!read_ahead(a, cur_block_size, &lp))
2698 /* Sanity check; there should never be a situation where this function
2699 * reads more data than the block's size. */
2700 if(partial_offset + cur_block_size > block_size) {
2701 archive_set_error(&a->archive, ARCHIVE_ERRNO_PROGRAMMER,
2702 "Consumed too much data when merging blocks.");
2703 return ARCHIVE_FATAL;
2706 /* Merge previous block chunk with current block chunk, or create
2707 * first block chunk if this is our first iteration. */
2708 memcpy(&rar->vol.push_buf[partial_offset], lp, cur_block_size);
2710 /* Advance the stream read pointer by this block chunk size. */
2711 if(ARCHIVE_OK != consume(a, cur_block_size))
2714 /* Update the pointers. `partial_offset` contains information about
2715 * the sum of merged block chunks. */
2716 partial_offset += cur_block_size;
2717 rar->file.bytes_remaining -= cur_block_size;
2719 /* If `partial_offset` is the same as `block_size`, this means we've
2720 * merged all block chunks and we have a valid full block. */
2721 if(partial_offset == block_size) {
2725 /* If we don't have any bytes to read, this means we should switch
2726 * to another multivolume archive file. */
2727 if(rar->file.bytes_remaining == 0) {
2728 ret = advance_multivolume(a);
2729 if(ret != ARCHIVE_OK)
2734 *p = rar->vol.push_buf;
2736 /* If we're here, we can resume unpacking by processing the block pointed
2737 * to by the `*p` memory pointer. */
2742 static int process_block(struct archive_read* a) {
2744 struct rar5* rar = get_context(a);
2747 /* If we don't have any data to be processed, this most probably means
2748 * we need to switch to the next volume. */
2749 if(rar->main.volume && rar->file.bytes_remaining == 0) {
2750 ret = advance_multivolume(a);
2751 if(ret != ARCHIVE_OK)
2755 if(rar->cstate.block_parsing_finished) {
2758 rar->cstate.block_parsing_finished = 0;
2760 /* The header size won't be bigger than 6 bytes. */
2761 if(!read_ahead(a, 6, &p)) {
2762 /* Failed to prefetch data block header. */
2767 * Read block_size by parsing block header. Validate the header by
2768 * calculating CRC byte stored inside the header. Size of the header is
2769 * not constant (block size can be stored either in 1 or 2 bytes),
2770 * that's why block size is left out from the `compressed_block_header`
2771 * structure and returned by `parse_block_header` as the second
2774 ret = parse_block_header(a, p, &block_size, &rar->last_block_hdr);
2775 if(ret != ARCHIVE_OK)
2778 /* Skip block header. Next data is huffman tables, if present. */
2779 ssize_t to_skip = sizeof(struct compressed_block_header) +
2780 rar->last_block_hdr.block_flags.byte_count + 1;
2782 if(ARCHIVE_OK != consume(a, to_skip))
2785 rar->file.bytes_remaining -= to_skip;
2787 /* The block size gives information about the whole block size, but
2788 * the block could be stored in split form when using multi-volume
2789 * archives. In this case, the block size will be bigger than the
2790 * actual data stored in this file. Remaining part of the data will
2791 * be in another file. */
2793 ssize_t cur_block_size =
2794 rar5_min(rar->file.bytes_remaining, block_size);
2796 if(block_size > rar->file.bytes_remaining) {
2797 /* If current blocks' size is bigger than our data size, this
2798 * means we have a multivolume archive. In this case, skip
2799 * all base headers until the end of the file, proceed to next
2800 * "partXXX.rar" volume, find its signature, skip all headers up
2801 * to the first FILE base header, and continue from there.
2803 * Note that `merge_block` will update the `rar` context structure
2804 * quite extensively. */
2806 ret = merge_block(a, block_size, &p);
2807 if(ret != ARCHIVE_OK) {
2811 cur_block_size = block_size;
2813 /* Current stream pointer should be now directly *after* the
2814 * block that spanned through multiple archive files. `p` pointer
2815 * should have the data of the *whole* block (merged from
2816 * partial blocks stored in multiple archives files). */
2818 rar->cstate.switch_multivolume = 0;
2820 /* Read the whole block size into memory. This can take up to
2821 * 8 megabytes of memory in theoretical cases. Might be worth to
2822 * optimize this and use a standard chunk of 4kb's. */
2824 if(!read_ahead(a, 4 + cur_block_size, &p)) {
2825 /* Failed to prefetch block data. */
2830 rar->cstate.block_buf = p;
2831 rar->cstate.cur_block_size = cur_block_size;
2833 rar->bits.in_addr = 0;
2834 rar->bits.bit_addr = 0;
2836 if(rar->last_block_hdr.block_flags.is_table_present) {
2837 /* Load Huffman tables. */
2838 ret = parse_tables(a, rar, p);
2839 if(ret != ARCHIVE_OK) {
2840 /* Error during decompression of Huffman tables. */
2845 p = rar->cstate.block_buf;
2848 /* Uncompress the block, or a part of it, depending on how many bytes
2849 * will be generated by uncompressing the block.
2851 * In case too many bytes will be generated, calling this function again
2852 * will resume the uncompression operation. */
2853 ret = do_uncompress_block(a, p);
2854 if(ret != ARCHIVE_OK) {
2858 if(rar->cstate.block_parsing_finished &&
2859 rar->cstate.switch_multivolume == 0 &&
2860 rar->cstate.cur_block_size > 0)
2862 /* If we're processing a normal block, consume the whole block. We
2863 * can do this because we've already read the whole block to memory.
2865 if(ARCHIVE_OK != consume(a, rar->cstate.cur_block_size))
2866 return ARCHIVE_FATAL;
2868 rar->file.bytes_remaining -= rar->cstate.cur_block_size;
2869 } else if(rar->cstate.switch_multivolume) {
2870 /* Don't consume the block if we're doing multivolume processing.
2871 * The volume switching function will consume the proper count of
2874 rar->cstate.switch_multivolume = 0;
2880 /* Pops the `buf`, `size` and `offset` from the "data ready" stack.
2882 * Returns ARCHIVE_OK when those arguments can be used, ARCHIVE_RETRY
2883 * when there is no data on the stack. */
2884 static int use_data(struct rar5* rar, const void** buf, size_t* size,
2889 for(i = 0; i < rar5_countof(rar->cstate.dready); i++) {
2890 struct data_ready *d = &rar->cstate.dready[i];
2893 if(buf) *buf = d->buf;
2894 if(size) *size = d->size;
2895 if(offset) *offset = d->offset;
2902 return ARCHIVE_RETRY;
2905 /* Pushes the `buf`, `size` and `offset` arguments to the rar->cstate.dready
2906 * FIFO stack. Those values will be popped from this stack by the `use_data`
2908 static int push_data_ready(struct archive_read* a, struct rar5* rar,
2909 const uint8_t* buf, size_t size, int64_t offset)
2913 /* Don't push if we're in skip mode. This is needed because solid
2914 * streams need full processing even if we're skipping data. After fully
2915 * processing the stream, we need to discard the generated bytes, because
2916 * we're interested only in the side effect: building up the internal
2917 * window circular buffer. This window buffer will be used later during
2918 * unpacking of requested data. */
2923 if(offset != rar->file.last_offset + rar->file.last_size) {
2924 archive_set_error(&a->archive, ARCHIVE_ERRNO_PROGRAMMER, "Sanity "
2925 "check error: output stream is not continuous");
2926 return ARCHIVE_FATAL;
2929 for(i = 0; i < rar5_countof(rar->cstate.dready); i++) {
2930 struct data_ready* d = &rar->cstate.dready[i];
2937 /* These fields are used only in sanity checking. */
2938 rar->file.last_offset = offset;
2939 rar->file.last_size = size;
2941 /* Calculate the checksum of this new block before submitting
2942 * data to libarchive's engine. */
2943 update_crc(rar, d->buf, d->size);
2949 /* Program counter will reach this code if the `rar->cstate.data_ready`
2950 * stack will be filled up so that no new entries will be allowed. The
2951 * code shouldn't allow such situation to occur. So we treat this case
2952 * as an internal error. */
2954 archive_set_error(&a->archive, ARCHIVE_ERRNO_PROGRAMMER, "Error: "
2955 "premature end of data_ready stack");
2956 return ARCHIVE_FATAL;
2959 /* This function uncompresses the data that is stored in the <FILE> base
2962 * The FILE base block looks like this:
2964 * <header><huffman tables><block_1><block_2>...<block_n>
2966 * The <header> is a block header, that is parsed in parse_block_header().
2967 * It's a "compressed_block_header" structure, containing metadata needed
2968 * to know when we should stop looking for more <block_n> blocks.
2970 * <huffman tables> contain data needed to set up the huffman tables, needed
2971 * for the actual decompression.
2973 * Each <block_n> consists of series of literals:
2975 * <literal><literal><literal>...<literal>
2977 * Those literals generate the uncompression data. They operate on a circular
2978 * buffer, sometimes writing raw data into it, sometimes referencing
2979 * some previous data inside this buffer, and sometimes declaring a filter
2980 * that will need to be executed on the data stored in the circular buffer.
2981 * It all depends on the literal that is used.
2983 * Sometimes blocks produce output data, sometimes they don't. For example, for
2984 * some huge files that use lots of filters, sometimes a block is filled with
2985 * only filter declaration literals. Such blocks won't produce any data in the
2988 * Sometimes blocks will produce 4 bytes of data, and sometimes 1 megabyte,
2989 * because a literal can reference previously decompressed data. For example,
2990 * there can be a literal that says: 'append a byte 0xFE here', and after
2991 * it another literal can say 'append 1 megabyte of data from circular buffer
2992 * offset 0x12345'. This is how RAR format handles compressing repeated
2995 * The RAR compressor creates those literals and the actual efficiency of
2996 * compression depends on what those literals are. The literals can also
2997 * be seen as a kind of a non-turing-complete virtual machine that simply
2998 * tells the decompressor what it should do.
3001 static int do_uncompress_file(struct archive_read* a) {
3002 struct rar5* rar = get_context(a);
3004 int64_t max_end_pos;
3006 if(!rar->cstate.initialized) {
3007 /* Don't perform full context reinitialization if we're processing
3008 * a solid archive. */
3009 if(!rar->main.solid || !rar->cstate.window_buf) {
3013 rar->cstate.initialized = 1;
3016 if(rar->cstate.all_filters_applied == 1) {
3017 /* We use while(1) here, but standard case allows for just 1 iteration.
3018 * The loop will iterate if process_block() didn't generate any data at
3019 * all. This can happen if the block contains only filter definitions
3020 * (this is common in big files). */
3023 ret = process_block(a);
3024 if(ret == ARCHIVE_EOF || ret == ARCHIVE_FATAL)
3027 if(rar->cstate.last_write_ptr == rar->cstate.write_ptr) {
3028 /* The block didn't generate any new data, so just process
3033 /* The block has generated some new data, so break the loop. */
3038 /* Try to run filters. If filters won't be applied, it means that
3039 * insufficient data was generated. */
3040 ret = apply_filters(a);
3041 if(ret == ARCHIVE_RETRY) {
3043 } else if(ret == ARCHIVE_FATAL) {
3044 return ARCHIVE_FATAL;
3047 /* If apply_filters() will return ARCHIVE_OK, we can continue here. */
3049 if(cdeque_size(&rar->cstate.filters) > 0) {
3050 /* Check if we can write something before hitting first filter. */
3051 struct filter_info* flt;
3053 /* Get the block_start offset from the first filter. */
3054 if(CDE_OK != cdeque_front(&rar->cstate.filters, cdeque_filter_p(&flt)))
3056 archive_set_error(&a->archive, ARCHIVE_ERRNO_PROGRAMMER,
3057 "Can't read first filter");
3058 return ARCHIVE_FATAL;
3061 max_end_pos = rar5_min(flt->block_start, rar->cstate.write_ptr);
3063 /* There are no filters defined, or all filters were applied. This
3064 * means we can just store the data without any postprocessing. */
3065 max_end_pos = rar->cstate.write_ptr;
3068 if(max_end_pos == rar->cstate.last_write_ptr) {
3069 /* We can't write anything yet. The block uncompression function did
3070 * not generate enough data, and no filter can be applied. At the same
3071 * time we don't have any data that can be stored without filter
3072 * postprocessing. This means we need to wait for more data to be
3073 * generated, so we can apply the filters.
3075 * Signal the caller that we need more data to be able to do anything.
3077 return ARCHIVE_RETRY;
3079 /* We can write the data before hitting the first filter. So let's
3080 * do it. The push_window_data() function will effectively return
3081 * the selected data block to the user application. */
3082 push_window_data(a, rar, rar->cstate.last_write_ptr, max_end_pos);
3083 rar->cstate.last_write_ptr = max_end_pos;
3089 static int uncompress_file(struct archive_read* a) {
3093 /* Sometimes the uncompression function will return a 'retry' signal.
3094 * If this will happen, we have to retry the function. */
3095 ret = do_uncompress_file(a);
3096 if(ret != ARCHIVE_RETRY)
3102 static int do_unstore_file(struct archive_read* a,
3110 if(rar->file.bytes_remaining == 0 && rar->main.volume > 0 &&
3111 rar->generic.split_after > 0)
3115 rar->cstate.switch_multivolume = 1;
3116 ret = advance_multivolume(a);
3117 rar->cstate.switch_multivolume = 0;
3119 if(ret != ARCHIVE_OK) {
3120 /* Failed to advance to next multivolume archive file. */
3125 size_t to_read = rar5_min(rar->file.bytes_remaining, 64 * 1024);
3130 if(!read_ahead(a, to_read, &p)) {
3131 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT, "I/O error "
3132 "when unstoring file");
3133 return ARCHIVE_FATAL;
3136 if(ARCHIVE_OK != consume(a, to_read)) {
3141 if(size) *size = to_read;
3142 if(offset) *offset = rar->cstate.last_unstore_ptr;
3144 rar->file.bytes_remaining -= to_read;
3145 rar->cstate.last_unstore_ptr += to_read;
3147 update_crc(rar, p, to_read);
3151 static int do_unpack(struct archive_read* a, struct rar5* rar,
3152 const void** buf, size_t* size, int64_t* offset)
3154 enum COMPRESSION_METHOD {
3155 STORE = 0, FASTEST = 1, FAST = 2, NORMAL = 3, GOOD = 4, BEST = 5
3158 if(rar->file.service > 0) {
3159 return do_unstore_file(a, rar, buf, size, offset);
3161 switch(rar->cstate.method) {
3163 return do_unstore_file(a, rar, buf, size, offset);
3173 return uncompress_file(a);
3175 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
3176 "Compression method not supported: 0x%08x",
3177 rar->cstate.method);
3179 return ARCHIVE_FATAL;
3189 static int verify_checksums(struct archive_read* a) {
3191 struct rar5* rar = get_context(a);
3193 /* Check checksums only when actually unpacking the data. There's no need
3194 * to calculate checksum when we're skipping data in solid archives
3195 * (skipping in solid archives is the same thing as unpacking compressed
3196 * data and discarding the result). */
3198 if(!rar->skip_mode) {
3199 /* Always check checksums if we're not in skip mode */
3202 /* We can override the logic above with a compile-time option
3203 * NO_CRC_ON_SOLID_SKIP. This option is used during debugging, and it
3204 * will check checksums of unpacked data even when we're skipping it.
3207 #if defined CHECK_CRC_ON_SOLID_SKIP
3217 /* During unpacking, on each unpacked block we're calling the
3218 * update_crc() function. Since we are here, the unpacking process is
3219 * already over and we can check if calculated checksum (CRC32 or
3220 * BLAKE2sp) is the same as what is stored in the archive.
3222 if(rar->file.stored_crc32 > 0) {
3223 /* Check CRC32 only when the file contains a CRC32 value for this
3226 if(rar->file.calculated_crc32 != rar->file.stored_crc32) {
3227 /* Checksums do not match; the unpacked file is corrupted. */
3230 printf("Checksum error: CRC32 (was: %08x, expected: %08x)\n",
3231 rar->file.calculated_crc32, rar->file.stored_crc32);
3234 #ifndef DONT_FAIL_ON_CRC_ERROR
3235 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
3236 "Checksum error: CRC32");
3237 return ARCHIVE_FATAL;
3241 printf("Checksum OK: CRC32 (%08x/%08x)\n",
3242 rar->file.stored_crc32,
3243 rar->file.calculated_crc32);
3248 if(rar->file.has_blake2 > 0) {
3249 /* BLAKE2sp is an optional checksum algorithm that is added to
3250 * RARv5 archives when using the `-htb` switch during creation of
3253 * We now finalize the hash calculation by calling the `final`
3254 * function. This will generate the final hash value we can use to
3255 * compare it with the BLAKE2sp checksum that is stored in the
3258 * The return value of this `final` function is not very helpful,
3259 * as it guards only against improper use. This is why we're
3260 * explicitly ignoring it. */
3263 (void) blake2sp_final(&rar->file.b2state, b2_buf, 32);
3265 if(memcmp(&rar->file.blake2sp, b2_buf, 32) != 0) {
3266 #ifndef DONT_FAIL_ON_CRC_ERROR
3267 archive_set_error(&a->archive, ARCHIVE_ERRNO_FILE_FORMAT,
3268 "Checksum error: BLAKE2");
3270 return ARCHIVE_FATAL;
3276 /* Finalization for this file has been successfully completed. */
3280 static int verify_global_checksums(struct archive_read* a) {
3281 return verify_checksums(a);
3284 static int rar5_read_data(struct archive_read *a, const void **buff,
3285 size_t *size, int64_t *offset) {
3287 struct rar5* rar = get_context(a);
3289 if(!rar->skip_mode && (rar->cstate.last_write_ptr > rar->file.unpacked_size)) {
3290 archive_set_error(&a->archive, ARCHIVE_ERRNO_PROGRAMMER,
3291 "Unpacker has written too many bytes");
3292 return ARCHIVE_FATAL;
3295 ret = use_data(rar, buff, size, offset);
3296 if(ret == ARCHIVE_OK) {
3300 if(rar->file.eof == 1) {
3304 ret = do_unpack(a, rar, buff, size, offset);
3305 if(ret != ARCHIVE_OK) {
3309 if(rar->file.bytes_remaining == 0 &&
3310 rar->cstate.last_write_ptr == rar->file.unpacked_size)
3312 /* If all bytes of current file were processed, run finalization.
3314 * Finalization will check checksum against proper values. If
3315 * some of the checksums will not match, we'll return an error
3316 * value in the last `archive_read_data` call to signal an error
3320 return verify_global_checksums(a);
3326 static int rar5_read_data_skip(struct archive_read *a) {
3327 struct rar5* rar = get_context(a);
3329 if(rar->main.solid) {
3330 /* In solid archives, instead of skipping the data, we need to extract
3331 * it, and dispose the result. The side effect of this operation will
3332 * be setting up the initial window buffer state needed to be able to
3333 * extract the selected file. */
3337 /* Make sure to process all blocks in the compressed stream. */
3338 while(rar->file.bytes_remaining > 0) {
3339 /* Setting the "skip mode" will allow us to skip checksum checks
3340 * during data skipping. Checking the checksum of skipped data
3341 * isn't really necessary and it's only slowing things down.
3343 * This is incremented instead of setting to 1 because this data
3344 * skipping function can be called recursively. */
3347 /* We're disposing 1 block of data, so we use triple NULLs in
3350 ret = rar5_read_data(a, NULL, NULL, NULL);
3352 /* Turn off "skip mode". */
3356 /* Propagate any potential error conditions to the caller. */
3361 /* In standard archives, we can just jump over the compressed stream.
3362 * Each file in non-solid archives starts from an empty window buffer.
3365 if(ARCHIVE_OK != consume(a, rar->file.bytes_remaining)) {
3366 return ARCHIVE_FATAL;
3369 rar->file.bytes_remaining = 0;
3375 static int64_t rar5_seek_data(struct archive_read *a, int64_t offset,
3382 /* We're a streaming unpacker, and we don't support seeking. */
3384 return ARCHIVE_FATAL;
3387 static int rar5_cleanup(struct archive_read *a) {
3388 struct rar5* rar = get_context(a);
3390 if(rar->cstate.window_buf)
3391 free(rar->cstate.window_buf);
3393 if(rar->cstate.filtered_buf)
3394 free(rar->cstate.filtered_buf);
3396 if(rar->vol.push_buf)
3397 free(rar->vol.push_buf);
3400 cdeque_free(&rar->cstate.filters);
3403 a->format->data = NULL;
3408 static int rar5_capabilities(struct archive_read * a) {
3413 static int rar5_has_encrypted_entries(struct archive_read *_a) {
3416 /* Unsupported for now. */
3417 return ARCHIVE_READ_FORMAT_ENCRYPTION_UNSUPPORTED;
3420 static int rar5_init(struct rar5* rar) {
3423 memset(rar, 0, sizeof(struct rar5));
3425 /* Decrypt the magic signature pattern. Check the comment near the
3426 * `rar5_signature` symbol to read the rationale behind this. */
3428 if(rar5_signature[0] == 243) {
3429 for(i = 0; i < rar5_signature_size; i++) {
3430 rar5_signature[i] ^= 0xA1;
3434 if(CDE_OK != cdeque_init(&rar->cstate.filters, 8192))
3435 return ARCHIVE_FATAL;
3440 int archive_read_support_format_rar5(struct archive *_a) {
3441 struct archive_read* ar;
3445 if(ARCHIVE_OK != (ret = get_archive_read(_a, &ar)))
3448 rar = malloc(sizeof(*rar));
3450 archive_set_error(&ar->archive, ENOMEM, "Can't allocate rar5 data");
3451 return ARCHIVE_FATAL;
3454 if(ARCHIVE_OK != rar5_init(rar)) {
3455 archive_set_error(&ar->archive, ENOMEM, "Can't allocate rar5 filter "
3457 return ARCHIVE_FATAL;
3460 ret = __archive_read_register_format(ar,
3467 rar5_read_data_skip,
3471 rar5_has_encrypted_entries);
3473 if(ret != ARCHIVE_OK) {
3474 (void) rar5_cleanup(ar);