]> pd.if.org Git - zpackage/blob - lzma/common/index.c
allow partial package ids in packagehash
[zpackage] / lzma / common / index.c
1 ///////////////////////////////////////////////////////////////////////////////
2 //
3 /// \file       index.c
4 /// \brief      Handling of .xz Indexes and some other Stream information
5 //
6 //  Author:     Lasse Collin
7 //
8 //  This file has been put into the public domain.
9 //  You can do whatever you want with this file.
10 //
11 ///////////////////////////////////////////////////////////////////////////////
12
13 #include "index.h"
14 #include "stream_flags_common.h"
15
16
17 /// \brief      How many Records to allocate at once
18 ///
19 /// This should be big enough to avoid making lots of tiny allocations
20 /// but small enough to avoid too much unused memory at once.
21 #define INDEX_GROUP_SIZE 512
22
23
24 /// \brief      How many Records can be allocated at once at maximum
25 #define PREALLOC_MAX ((SIZE_MAX - sizeof(index_group)) / sizeof(index_record))
26
27
28 /// \brief      Base structure for index_stream and index_group structures
29 typedef struct index_tree_node_s index_tree_node;
30 struct index_tree_node_s {
31         /// Uncompressed start offset of this Stream (relative to the
32         /// beginning of the file) or Block (relative to the beginning
33         /// of the Stream)
34         lzma_vli uncompressed_base;
35
36         /// Compressed start offset of this Stream or Block
37         lzma_vli compressed_base;
38
39         index_tree_node *parent;
40         index_tree_node *left;
41         index_tree_node *right;
42 };
43
44
45 /// \brief      AVL tree to hold index_stream or index_group structures
46 typedef struct {
47         /// Root node
48         index_tree_node *root;
49
50         /// Leftmost node. Since the tree will be filled sequentially,
51         /// this won't change after the first node has been added to
52         /// the tree.
53         index_tree_node *leftmost;
54
55         /// The rightmost node in the tree. Since the tree is filled
56         /// sequentially, this is always the node where to add the new data.
57         index_tree_node *rightmost;
58
59         /// Number of nodes in the tree
60         uint32_t count;
61
62 } index_tree;
63
64
65 typedef struct {
66         lzma_vli uncompressed_sum;
67         lzma_vli unpadded_sum;
68 } index_record;
69
70
71 typedef struct {
72         /// Every Record group is part of index_stream.groups tree.
73         index_tree_node node;
74
75         /// Number of Blocks in this Stream before this group.
76         lzma_vli number_base;
77
78         /// Number of Records that can be put in records[].
79         size_t allocated;
80
81         /// Index of the last Record in use.
82         size_t last;
83
84         /// The sizes in this array are stored as cumulative sums relative
85         /// to the beginning of the Stream. This makes it possible to
86         /// use binary search in lzma_index_locate().
87         ///
88         /// Note that the cumulative summing is done specially for
89         /// unpadded_sum: The previous value is rounded up to the next
90         /// multiple of four before adding the Unpadded Size of the new
91         /// Block. The total encoded size of the Blocks in the Stream
92         /// is records[last].unpadded_sum in the last Record group of
93         /// the Stream.
94         ///
95         /// For example, if the Unpadded Sizes are 39, 57, and 81, the
96         /// stored values are 39, 97 (40 + 57), and 181 (100 + 181).
97         /// The total encoded size of these Blocks is 184.
98         ///
99         /// This is a flexible array, because it makes easy to optimize
100         /// memory usage in case someone concatenates many Streams that
101         /// have only one or few Blocks.
102         index_record records[];
103
104 } index_group;
105
106
107 typedef struct {
108         /// Every index_stream is a node in the tree of Sreams.
109         index_tree_node node;
110
111         /// Number of this Stream (first one is 1)
112         uint32_t number;
113
114         /// Total number of Blocks before this Stream
115         lzma_vli block_number_base;
116
117         /// Record groups of this Stream are stored in a tree.
118         /// It's a T-tree with AVL-tree balancing. There are
119         /// INDEX_GROUP_SIZE Records per node by default.
120         /// This keeps the number of memory allocations reasonable
121         /// and finding a Record is fast.
122         index_tree groups;
123
124         /// Number of Records in this Stream
125         lzma_vli record_count;
126
127         /// Size of the List of Records field in this Stream. This is used
128         /// together with record_count to calculate the size of the Index
129         /// field and thus the total size of the Stream.
130         lzma_vli index_list_size;
131
132         /// Stream Flags of this Stream. This is meaningful only if
133         /// the Stream Flags have been told us with lzma_index_stream_flags().
134         /// Initially stream_flags.version is set to UINT32_MAX to indicate
135         /// that the Stream Flags are unknown.
136         lzma_stream_flags stream_flags;
137
138         /// Amount of Stream Padding after this Stream. This defaults to
139         /// zero and can be set with lzma_index_stream_padding().
140         lzma_vli stream_padding;
141
142 } index_stream;
143
144
145 struct lzma_index_s {
146         /// AVL-tree containing the Stream(s). Often there is just one
147         /// Stream, but using a tree keeps lookups fast even when there
148         /// are many concatenated Streams.
149         index_tree streams;
150
151         /// Uncompressed size of all the Blocks in the Stream(s)
152         lzma_vli uncompressed_size;
153
154         /// Total size of all the Blocks in the Stream(s)
155         lzma_vli total_size;
156
157         /// Total number of Records in all Streams in this lzma_index
158         lzma_vli record_count;
159
160         /// Size of the List of Records field if all the Streams in this
161         /// lzma_index were packed into a single Stream (makes it simpler to
162         /// take many .xz files and combine them into a single Stream).
163         ///
164         /// This value together with record_count is needed to calculate
165         /// Backward Size that is stored into Stream Footer.
166         lzma_vli index_list_size;
167
168         /// How many Records to allocate at once in lzma_index_append().
169         /// This defaults to INDEX_GROUP_SIZE but can be overriden with
170         /// lzma_index_prealloc().
171         size_t prealloc;
172
173         /// Bitmask indicating what integrity check types have been used
174         /// as set by lzma_index_stream_flags(). The bit of the last Stream
175         /// is not included here, since it is possible to change it by
176         /// calling lzma_index_stream_flags() again.
177         uint32_t checks;
178 };
179
180
181 static void
182 index_tree_init(index_tree *tree)
183 {
184         tree->root = NULL;
185         tree->leftmost = NULL;
186         tree->rightmost = NULL;
187         tree->count = 0;
188         return;
189 }
190
191
192 /// Helper for index_tree_end()
193 static void
194 index_tree_node_end(index_tree_node *node, const lzma_allocator *allocator,
195                 void (*free_func)(void *node, const lzma_allocator *allocator))
196 {
197         // The tree won't ever be very huge, so recursion should be fine.
198         // 20 levels in the tree is likely quite a lot already in practice.
199         if (node->left != NULL)
200                 index_tree_node_end(node->left, allocator, free_func);
201
202         if (node->right != NULL)
203                 index_tree_node_end(node->right, allocator, free_func);
204
205         free_func(node, allocator);
206         return;
207 }
208
209
210 /// Free the memory allocated for a tree. Each node is freed using the
211 /// given free_func which is either &lzma_free or &index_stream_end.
212 /// The latter is used to free the Record groups from each index_stream
213 /// before freeing the index_stream itself.
214 static void
215 index_tree_end(index_tree *tree, const lzma_allocator *allocator,
216                 void (*free_func)(void *node, const lzma_allocator *allocator))
217 {
218         assert(free_func != NULL);
219
220         if (tree->root != NULL)
221                 index_tree_node_end(tree->root, allocator, free_func);
222
223         return;
224 }
225
226
227 /// Add a new node to the tree. node->uncompressed_base and
228 /// node->compressed_base must have been set by the caller already.
229 static void
230 index_tree_append(index_tree *tree, index_tree_node *node)
231 {
232         node->parent = tree->rightmost;
233         node->left = NULL;
234         node->right = NULL;
235
236         ++tree->count;
237
238         // Handle the special case of adding the first node.
239         if (tree->root == NULL) {
240                 tree->root = node;
241                 tree->leftmost = node;
242                 tree->rightmost = node;
243                 return;
244         }
245
246         // The tree is always filled sequentially.
247         assert(tree->rightmost->uncompressed_base <= node->uncompressed_base);
248         assert(tree->rightmost->compressed_base < node->compressed_base);
249
250         // Add the new node after the rightmost node. It's the correct
251         // place due to the reason above.
252         tree->rightmost->right = node;
253         tree->rightmost = node;
254
255         // Balance the AVL-tree if needed. We don't need to keep the balance
256         // factors in nodes, because we always fill the tree sequentially,
257         // and thus know the state of the tree just by looking at the node
258         // count. From the node count we can calculate how many steps to go
259         // up in the tree to find the rotation root.
260         uint32_t up = tree->count ^ (UINT32_C(1) << bsr32(tree->count));
261         if (up != 0) {
262                 // Locate the root node for the rotation.
263                 up = ctz32(tree->count) + 2;
264                 do {
265                         node = node->parent;
266                 } while (--up > 0);
267
268                 // Rotate left using node as the rotation root.
269                 index_tree_node *pivot = node->right;
270
271                 if (node->parent == NULL) {
272                         tree->root = pivot;
273                 } else {
274                         assert(node->parent->right == node);
275                         node->parent->right = pivot;
276                 }
277
278                 pivot->parent = node->parent;
279
280                 node->right = pivot->left;
281                 if (node->right != NULL)
282                         node->right->parent = node;
283
284                 pivot->left = node;
285                 node->parent = pivot;
286         }
287
288         return;
289 }
290
291
292 /// Get the next node in the tree. Return NULL if there are no more nodes.
293 static void *
294 index_tree_next(const index_tree_node *node)
295 {
296         if (node->right != NULL) {
297                 node = node->right;
298                 while (node->left != NULL)
299                         node = node->left;
300
301                 return (void *)(node);
302         }
303
304         while (node->parent != NULL && node->parent->right == node)
305                 node = node->parent;
306
307         return (void *)(node->parent);
308 }
309
310
311 /// Locate a node that contains the given uncompressed offset. It is
312 /// caller's job to check that target is not bigger than the uncompressed
313 /// size of the tree (the last node would be returned in that case still).
314 static void *
315 index_tree_locate(const index_tree *tree, lzma_vli target)
316 {
317         const index_tree_node *result = NULL;
318         const index_tree_node *node = tree->root;
319
320         assert(tree->leftmost == NULL
321                         || tree->leftmost->uncompressed_base == 0);
322
323         // Consecutive nodes may have the same uncompressed_base.
324         // We must pick the rightmost one.
325         while (node != NULL) {
326                 if (node->uncompressed_base > target) {
327                         node = node->left;
328                 } else {
329                         result = node;
330                         node = node->right;
331                 }
332         }
333
334         return (void *)(result);
335 }
336
337
338 /// Allocate and initialize a new Stream using the given base offsets.
339 static index_stream *
340 index_stream_init(lzma_vli compressed_base, lzma_vli uncompressed_base,
341                 uint32_t stream_number, lzma_vli block_number_base,
342                 const lzma_allocator *allocator)
343 {
344         index_stream *s = lzma_alloc(sizeof(index_stream), allocator);
345         if (s == NULL)
346                 return NULL;
347
348         s->node.uncompressed_base = uncompressed_base;
349         s->node.compressed_base = compressed_base;
350         s->node.parent = NULL;
351         s->node.left = NULL;
352         s->node.right = NULL;
353
354         s->number = stream_number;
355         s->block_number_base = block_number_base;
356
357         index_tree_init(&s->groups);
358
359         s->record_count = 0;
360         s->index_list_size = 0;
361         s->stream_flags.version = UINT32_MAX;
362         s->stream_padding = 0;
363
364         return s;
365 }
366
367
368 /// Free the memory allocated for a Stream and its Record groups.
369 static void
370 index_stream_end(void *node, const lzma_allocator *allocator)
371 {
372         index_stream *s = node;
373         index_tree_end(&s->groups, allocator, &lzma_free);
374         lzma_free(s, allocator);
375         return;
376 }
377
378
379 static lzma_index *
380 index_init_plain(const lzma_allocator *allocator)
381 {
382         lzma_index *i = lzma_alloc(sizeof(lzma_index), allocator);
383         if (i != NULL) {
384                 index_tree_init(&i->streams);
385                 i->uncompressed_size = 0;
386                 i->total_size = 0;
387                 i->record_count = 0;
388                 i->index_list_size = 0;
389                 i->prealloc = INDEX_GROUP_SIZE;
390                 i->checks = 0;
391         }
392
393         return i;
394 }
395
396
397 extern LZMA_API(lzma_index *)
398 lzma_index_init(const lzma_allocator *allocator)
399 {
400         lzma_index *i = index_init_plain(allocator);
401         if (i == NULL)
402                 return NULL;
403
404         index_stream *s = index_stream_init(0, 0, 1, 0, allocator);
405         if (s == NULL) {
406                 lzma_free(i, allocator);
407                 return NULL;
408         }
409
410         index_tree_append(&i->streams, &s->node);
411
412         return i;
413 }
414
415
416 extern LZMA_API(void)
417 lzma_index_end(lzma_index *i, const lzma_allocator *allocator)
418 {
419         // NOTE: If you modify this function, check also the bottom
420         // of lzma_index_cat().
421         if (i != NULL) {
422                 index_tree_end(&i->streams, allocator, &index_stream_end);
423                 lzma_free(i, allocator);
424         }
425
426         return;
427 }
428
429
430 extern void
431 lzma_index_prealloc(lzma_index *i, lzma_vli records)
432 {
433         if (records > PREALLOC_MAX)
434                 records = PREALLOC_MAX;
435
436         i->prealloc = (size_t)(records);
437         return;
438 }
439
440
441 extern LZMA_API(uint64_t)
442 lzma_index_memusage(lzma_vli streams, lzma_vli blocks)
443 {
444         // This calculates an upper bound that is only a little bit
445         // bigger than the exact maximum memory usage with the given
446         // parameters.
447
448         // Typical malloc() overhead is 2 * sizeof(void *) but we take
449         // a little bit extra just in case. Using LZMA_MEMUSAGE_BASE
450         // instead would give too inaccurate estimate.
451         const size_t alloc_overhead = 4 * sizeof(void *);
452
453         // Amount of memory needed for each Stream base structures.
454         // We assume that every Stream has at least one Block and
455         // thus at least one group.
456         const size_t stream_base = sizeof(index_stream)
457                         + sizeof(index_group) + 2 * alloc_overhead;
458
459         // Amount of memory needed per group.
460         const size_t group_base = sizeof(index_group)
461                         + INDEX_GROUP_SIZE * sizeof(index_record)
462                         + alloc_overhead;
463
464         // Number of groups. There may actually be more, but that overhead
465         // has been taken into account in stream_base already.
466         const lzma_vli groups
467                         = (blocks + INDEX_GROUP_SIZE - 1) / INDEX_GROUP_SIZE;
468
469         // Memory used by index_stream and index_group structures.
470         const uint64_t streams_mem = streams * stream_base;
471         const uint64_t groups_mem = groups * group_base;
472
473         // Memory used by the base structure.
474         const uint64_t index_base = sizeof(lzma_index) + alloc_overhead;
475
476         // Validate the arguments and catch integer overflows.
477         // Maximum number of Streams is "only" UINT32_MAX, because
478         // that limit is used by the tree containing the Streams.
479         const uint64_t limit = UINT64_MAX - index_base;
480         if (streams == 0 || streams > UINT32_MAX || blocks > LZMA_VLI_MAX
481                         || streams > limit / stream_base
482                         || groups > limit / group_base
483                         || limit - streams_mem < groups_mem)
484                 return UINT64_MAX;
485
486         return index_base + streams_mem + groups_mem;
487 }
488
489
490 extern LZMA_API(uint64_t)
491 lzma_index_memused(const lzma_index *i)
492 {
493         return lzma_index_memusage(i->streams.count, i->record_count);
494 }
495
496
497 extern LZMA_API(lzma_vli)
498 lzma_index_block_count(const lzma_index *i)
499 {
500         return i->record_count;
501 }
502
503
504 extern LZMA_API(lzma_vli)
505 lzma_index_stream_count(const lzma_index *i)
506 {
507         return i->streams.count;
508 }
509
510
511 extern LZMA_API(lzma_vli)
512 lzma_index_size(const lzma_index *i)
513 {
514         return index_size(i->record_count, i->index_list_size);
515 }
516
517
518 extern LZMA_API(lzma_vli)
519 lzma_index_total_size(const lzma_index *i)
520 {
521         return i->total_size;
522 }
523
524
525 extern LZMA_API(lzma_vli)
526 lzma_index_stream_size(const lzma_index *i)
527 {
528         // Stream Header + Blocks + Index + Stream Footer
529         return LZMA_STREAM_HEADER_SIZE + i->total_size
530                         + index_size(i->record_count, i->index_list_size)
531                         + LZMA_STREAM_HEADER_SIZE;
532 }
533
534
535 static lzma_vli
536 index_file_size(lzma_vli compressed_base, lzma_vli unpadded_sum,
537                 lzma_vli record_count, lzma_vli index_list_size,
538                 lzma_vli stream_padding)
539 {
540         // Earlier Streams and Stream Paddings + Stream Header
541         // + Blocks + Index + Stream Footer + Stream Padding
542         //
543         // This might go over LZMA_VLI_MAX due to too big unpadded_sum
544         // when this function is used in lzma_index_append().
545         lzma_vli file_size = compressed_base + 2 * LZMA_STREAM_HEADER_SIZE
546                         + stream_padding + vli_ceil4(unpadded_sum);
547         if (file_size > LZMA_VLI_MAX)
548                 return LZMA_VLI_UNKNOWN;
549
550         // The same applies here.
551         file_size += index_size(record_count, index_list_size);
552         if (file_size > LZMA_VLI_MAX)
553                 return LZMA_VLI_UNKNOWN;
554
555         return file_size;
556 }
557
558
559 extern LZMA_API(lzma_vli)
560 lzma_index_file_size(const lzma_index *i)
561 {
562         const index_stream *s = (const index_stream *)(i->streams.rightmost);
563         const index_group *g = (const index_group *)(s->groups.rightmost);
564         return index_file_size(s->node.compressed_base,
565                         g == NULL ? 0 : g->records[g->last].unpadded_sum,
566                         s->record_count, s->index_list_size,
567                         s->stream_padding);
568 }
569
570
571 extern LZMA_API(lzma_vli)
572 lzma_index_uncompressed_size(const lzma_index *i)
573 {
574         return i->uncompressed_size;
575 }
576
577
578 extern LZMA_API(uint32_t)
579 lzma_index_checks(const lzma_index *i)
580 {
581         uint32_t checks = i->checks;
582
583         // Get the type of the Check of the last Stream too.
584         const index_stream *s = (const index_stream *)(i->streams.rightmost);
585         if (s->stream_flags.version != UINT32_MAX)
586                 checks |= UINT32_C(1) << s->stream_flags.check;
587
588         return checks;
589 }
590
591
592 extern uint32_t
593 lzma_index_padding_size(const lzma_index *i)
594 {
595         return (LZMA_VLI_C(4) - index_size_unpadded(
596                         i->record_count, i->index_list_size)) & 3;
597 }
598
599
600 extern LZMA_API(lzma_ret)
601 lzma_index_stream_flags(lzma_index *i, const lzma_stream_flags *stream_flags)
602 {
603         if (i == NULL || stream_flags == NULL)
604                 return LZMA_PROG_ERROR;
605
606         // Validate the Stream Flags.
607         return_if_error(lzma_stream_flags_compare(
608                         stream_flags, stream_flags));
609
610         index_stream *s = (index_stream *)(i->streams.rightmost);
611         s->stream_flags = *stream_flags;
612
613         return LZMA_OK;
614 }
615
616
617 extern LZMA_API(lzma_ret)
618 lzma_index_stream_padding(lzma_index *i, lzma_vli stream_padding)
619 {
620         if (i == NULL || stream_padding > LZMA_VLI_MAX
621                         || (stream_padding & 3) != 0)
622                 return LZMA_PROG_ERROR;
623
624         index_stream *s = (index_stream *)(i->streams.rightmost);
625
626         // Check that the new value won't make the file grow too big.
627         const lzma_vli old_stream_padding = s->stream_padding;
628         s->stream_padding = 0;
629         if (lzma_index_file_size(i) + stream_padding > LZMA_VLI_MAX) {
630                 s->stream_padding = old_stream_padding;
631                 return LZMA_DATA_ERROR;
632         }
633
634         s->stream_padding = stream_padding;
635         return LZMA_OK;
636 }
637
638
639 extern LZMA_API(lzma_ret)
640 lzma_index_append(lzma_index *i, const lzma_allocator *allocator,
641                 lzma_vli unpadded_size, lzma_vli uncompressed_size)
642 {
643         // Validate.
644         if (i == NULL || unpadded_size < UNPADDED_SIZE_MIN
645                         || unpadded_size > UNPADDED_SIZE_MAX
646                         || uncompressed_size > LZMA_VLI_MAX)
647                 return LZMA_PROG_ERROR;
648
649         index_stream *s = (index_stream *)(i->streams.rightmost);
650         index_group *g = (index_group *)(s->groups.rightmost);
651
652         const lzma_vli compressed_base = g == NULL ? 0
653                         : vli_ceil4(g->records[g->last].unpadded_sum);
654         const lzma_vli uncompressed_base = g == NULL ? 0
655                         : g->records[g->last].uncompressed_sum;
656         const uint32_t index_list_size_add = lzma_vli_size(unpadded_size)
657                         + lzma_vli_size(uncompressed_size);
658
659         // Check that the file size will stay within limits.
660         if (index_file_size(s->node.compressed_base,
661                         compressed_base + unpadded_size, s->record_count + 1,
662                         s->index_list_size + index_list_size_add,
663                         s->stream_padding) == LZMA_VLI_UNKNOWN)
664                 return LZMA_DATA_ERROR;
665
666         // The size of the Index field must not exceed the maximum value
667         // that can be stored in the Backward Size field.
668         if (index_size(i->record_count + 1,
669                         i->index_list_size + index_list_size_add)
670                         > LZMA_BACKWARD_SIZE_MAX)
671                 return LZMA_DATA_ERROR;
672
673         if (g != NULL && g->last + 1 < g->allocated) {
674                 // There is space in the last group at least for one Record.
675                 ++g->last;
676         } else {
677                 // We need to allocate a new group.
678                 g = lzma_alloc(sizeof(index_group)
679                                 + i->prealloc * sizeof(index_record),
680                                 allocator);
681                 if (g == NULL)
682                         return LZMA_MEM_ERROR;
683
684                 g->last = 0;
685                 g->allocated = i->prealloc;
686
687                 // Reset prealloc so that if the application happens to
688                 // add new Records, the allocation size will be sane.
689                 i->prealloc = INDEX_GROUP_SIZE;
690
691                 // Set the start offsets of this group.
692                 g->node.uncompressed_base = uncompressed_base;
693                 g->node.compressed_base = compressed_base;
694                 g->number_base = s->record_count + 1;
695
696                 // Add the new group to the Stream.
697                 index_tree_append(&s->groups, &g->node);
698         }
699
700         // Add the new Record to the group.
701         g->records[g->last].uncompressed_sum
702                         = uncompressed_base + uncompressed_size;
703         g->records[g->last].unpadded_sum
704                         = compressed_base + unpadded_size;
705
706         // Update the totals.
707         ++s->record_count;
708         s->index_list_size += index_list_size_add;
709
710         i->total_size += vli_ceil4(unpadded_size);
711         i->uncompressed_size += uncompressed_size;
712         ++i->record_count;
713         i->index_list_size += index_list_size_add;
714
715         return LZMA_OK;
716 }
717
718
719 /// Structure to pass info to index_cat_helper()
720 typedef struct {
721         /// Uncompressed size of the destination
722         lzma_vli uncompressed_size;
723
724         /// Compressed file size of the destination
725         lzma_vli file_size;
726
727         /// Same as above but for Block numbers
728         lzma_vli block_number_add;
729
730         /// Number of Streams that were in the destination index before we
731         /// started appending new Streams from the source index. This is
732         /// used to fix the Stream numbering.
733         uint32_t stream_number_add;
734
735         /// Destination index' Stream tree
736         index_tree *streams;
737
738 } index_cat_info;
739
740
741 /// Add the Stream nodes from the source index to dest using recursion.
742 /// Simplest iterative traversal of the source tree wouldn't work, because
743 /// we update the pointers in nodes when moving them to the destination tree.
744 static void
745 index_cat_helper(const index_cat_info *info, index_stream *this)
746 {
747         index_stream *left = (index_stream *)(this->node.left);
748         index_stream *right = (index_stream *)(this->node.right);
749
750         if (left != NULL)
751                 index_cat_helper(info, left);
752
753         this->node.uncompressed_base += info->uncompressed_size;
754         this->node.compressed_base += info->file_size;
755         this->number += info->stream_number_add;
756         this->block_number_base += info->block_number_add;
757         index_tree_append(info->streams, &this->node);
758
759         if (right != NULL)
760                 index_cat_helper(info, right);
761
762         return;
763 }
764
765
766 extern LZMA_API(lzma_ret)
767 lzma_index_cat(lzma_index *restrict dest, lzma_index *restrict src,
768                 const lzma_allocator *allocator)
769 {
770         const lzma_vli dest_file_size = lzma_index_file_size(dest);
771
772         // Check that we don't exceed the file size limits.
773         if (dest_file_size + lzma_index_file_size(src) > LZMA_VLI_MAX
774                         || dest->uncompressed_size + src->uncompressed_size
775                                 > LZMA_VLI_MAX)
776                 return LZMA_DATA_ERROR;
777
778         // Check that the encoded size of the combined lzma_indexes stays
779         // within limits. In theory, this should be done only if we know
780         // that the user plans to actually combine the Streams and thus
781         // construct a single Index (probably rare). However, exceeding
782         // this limit is quite theoretical, so we do this check always
783         // to simplify things elsewhere.
784         {
785                 const lzma_vli dest_size = index_size_unpadded(
786                                 dest->record_count, dest->index_list_size);
787                 const lzma_vli src_size = index_size_unpadded(
788                                 src->record_count, src->index_list_size);
789                 if (vli_ceil4(dest_size + src_size) > LZMA_BACKWARD_SIZE_MAX)
790                         return LZMA_DATA_ERROR;
791         }
792
793         // Optimize the last group to minimize memory usage. Allocation has
794         // to be done before modifying dest or src.
795         {
796                 index_stream *s = (index_stream *)(dest->streams.rightmost);
797                 index_group *g = (index_group *)(s->groups.rightmost);
798                 if (g != NULL && g->last + 1 < g->allocated) {
799                         assert(g->node.left == NULL);
800                         assert(g->node.right == NULL);
801
802                         index_group *newg = lzma_alloc(sizeof(index_group)
803                                         + (g->last + 1)
804                                         * sizeof(index_record),
805                                         allocator);
806                         if (newg == NULL)
807                                 return LZMA_MEM_ERROR;
808
809                         newg->node = g->node;
810                         newg->allocated = g->last + 1;
811                         newg->last = g->last;
812                         newg->number_base = g->number_base;
813
814                         memcpy(newg->records, g->records, newg->allocated
815                                         * sizeof(index_record));
816
817                         if (g->node.parent != NULL) {
818                                 assert(g->node.parent->right == &g->node);
819                                 g->node.parent->right = &newg->node;
820                         }
821
822                         if (s->groups.leftmost == &g->node) {
823                                 assert(s->groups.root == &g->node);
824                                 s->groups.leftmost = &newg->node;
825                                 s->groups.root = &newg->node;
826                         }
827
828                         if (s->groups.rightmost == &g->node)
829                                 s->groups.rightmost = &newg->node;
830
831                         lzma_free(g, allocator);
832
833                         // NOTE: newg isn't leaked here because
834                         // newg == (void *)&newg->node.
835                 }
836         }
837
838         // Add all the Streams from src to dest. Update the base offsets
839         // of each Stream from src.
840         const index_cat_info info = {
841                 .uncompressed_size = dest->uncompressed_size,
842                 .file_size = dest_file_size,
843                 .stream_number_add = dest->streams.count,
844                 .block_number_add = dest->record_count,
845                 .streams = &dest->streams,
846         };
847         index_cat_helper(&info, (index_stream *)(src->streams.root));
848
849         // Update info about all the combined Streams.
850         dest->uncompressed_size += src->uncompressed_size;
851         dest->total_size += src->total_size;
852         dest->record_count += src->record_count;
853         dest->index_list_size += src->index_list_size;
854         dest->checks = lzma_index_checks(dest) | src->checks;
855
856         // There's nothing else left in src than the base structure.
857         lzma_free(src, allocator);
858
859         return LZMA_OK;
860 }
861
862
863 /// Duplicate an index_stream.
864 static index_stream *
865 index_dup_stream(const index_stream *src, const lzma_allocator *allocator)
866 {
867         // Catch a somewhat theoretical integer overflow.
868         if (src->record_count > PREALLOC_MAX)
869                 return NULL;
870
871         // Allocate and initialize a new Stream.
872         index_stream *dest = index_stream_init(src->node.compressed_base,
873                         src->node.uncompressed_base, src->number,
874                         src->block_number_base, allocator);
875         if (dest == NULL)
876                 return NULL;
877
878         // Copy the overall information.
879         dest->record_count = src->record_count;
880         dest->index_list_size = src->index_list_size;
881         dest->stream_flags = src->stream_flags;
882         dest->stream_padding = src->stream_padding;
883
884         // Return if there are no groups to duplicate.
885         if (src->groups.leftmost == NULL)
886                 return dest;
887
888         // Allocate memory for the Records. We put all the Records into
889         // a single group. It's simplest and also tends to make
890         // lzma_index_locate() a little bit faster with very big Indexes.
891         index_group *destg = lzma_alloc(sizeof(index_group)
892                         + src->record_count * sizeof(index_record),
893                         allocator);
894         if (destg == NULL) {
895                 index_stream_end(dest, allocator);
896                 return NULL;
897         }
898
899         // Initialize destg.
900         destg->node.uncompressed_base = 0;
901         destg->node.compressed_base = 0;
902         destg->number_base = 1;
903         destg->allocated = src->record_count;
904         destg->last = src->record_count - 1;
905
906         // Go through all the groups in src and copy the Records into destg.
907         const index_group *srcg = (const index_group *)(src->groups.leftmost);
908         size_t i = 0;
909         do {
910                 memcpy(destg->records + i, srcg->records,
911                                 (srcg->last + 1) * sizeof(index_record));
912                 i += srcg->last + 1;
913                 srcg = index_tree_next(&srcg->node);
914         } while (srcg != NULL);
915
916         assert(i == destg->allocated);
917
918         // Add the group to the new Stream.
919         index_tree_append(&dest->groups, &destg->node);
920
921         return dest;
922 }
923
924
925 extern LZMA_API(lzma_index *)
926 lzma_index_dup(const lzma_index *src, const lzma_allocator *allocator)
927 {
928         // Allocate the base structure (no initial Stream).
929         lzma_index *dest = index_init_plain(allocator);
930         if (dest == NULL)
931                 return NULL;
932
933         // Copy the totals.
934         dest->uncompressed_size = src->uncompressed_size;
935         dest->total_size = src->total_size;
936         dest->record_count = src->record_count;
937         dest->index_list_size = src->index_list_size;
938
939         // Copy the Streams and the groups in them.
940         const index_stream *srcstream
941                         = (const index_stream *)(src->streams.leftmost);
942         do {
943                 index_stream *deststream = index_dup_stream(
944                                 srcstream, allocator);
945                 if (deststream == NULL) {
946                         lzma_index_end(dest, allocator);
947                         return NULL;
948                 }
949
950                 index_tree_append(&dest->streams, &deststream->node);
951
952                 srcstream = index_tree_next(&srcstream->node);
953         } while (srcstream != NULL);
954
955         return dest;
956 }
957
958
959 /// Indexing for lzma_index_iter.internal[]
960 enum {
961         ITER_INDEX,
962         ITER_STREAM,
963         ITER_GROUP,
964         ITER_RECORD,
965         ITER_METHOD,
966 };
967
968
969 /// Values for lzma_index_iter.internal[ITER_METHOD].s
970 enum {
971         ITER_METHOD_NORMAL,
972         ITER_METHOD_NEXT,
973         ITER_METHOD_LEFTMOST,
974 };
975
976
977 static void
978 iter_set_info(lzma_index_iter *iter)
979 {
980         const lzma_index *i = iter->internal[ITER_INDEX].p;
981         const index_stream *stream = iter->internal[ITER_STREAM].p;
982         const index_group *group = iter->internal[ITER_GROUP].p;
983         const size_t record = iter->internal[ITER_RECORD].s;
984
985         // lzma_index_iter.internal must not contain a pointer to the last
986         // group in the index, because that may be reallocated by
987         // lzma_index_cat().
988         if (group == NULL) {
989                 // There are no groups.
990                 assert(stream->groups.root == NULL);
991                 iter->internal[ITER_METHOD].s = ITER_METHOD_LEFTMOST;
992
993         } else if (i->streams.rightmost != &stream->node
994                         || stream->groups.rightmost != &group->node) {
995                 // The group is not not the last group in the index.
996                 iter->internal[ITER_METHOD].s = ITER_METHOD_NORMAL;
997
998         } else if (stream->groups.leftmost != &group->node) {
999                 // The group isn't the only group in the Stream, thus we
1000                 // know that it must have a parent group i.e. it's not
1001                 // the root node.
1002                 assert(stream->groups.root != &group->node);
1003                 assert(group->node.parent->right == &group->node);
1004                 iter->internal[ITER_METHOD].s = ITER_METHOD_NEXT;
1005                 iter->internal[ITER_GROUP].p = group->node.parent;
1006
1007         } else {
1008                 // The Stream has only one group.
1009                 assert(stream->groups.root == &group->node);
1010                 assert(group->node.parent == NULL);
1011                 iter->internal[ITER_METHOD].s = ITER_METHOD_LEFTMOST;
1012                 iter->internal[ITER_GROUP].p = NULL;
1013         }
1014
1015         // NOTE: lzma_index_iter.stream.number is lzma_vli but we use uint32_t
1016         // internally.
1017         iter->stream.number = stream->number;
1018         iter->stream.block_count = stream->record_count;
1019         iter->stream.compressed_offset = stream->node.compressed_base;
1020         iter->stream.uncompressed_offset = stream->node.uncompressed_base;
1021
1022         // iter->stream.flags will be NULL if the Stream Flags haven't been
1023         // set with lzma_index_stream_flags().
1024         iter->stream.flags = stream->stream_flags.version == UINT32_MAX
1025                         ? NULL : &stream->stream_flags;
1026         iter->stream.padding = stream->stream_padding;
1027
1028         if (stream->groups.rightmost == NULL) {
1029                 // Stream has no Blocks.
1030                 iter->stream.compressed_size = index_size(0, 0)
1031                                 + 2 * LZMA_STREAM_HEADER_SIZE;
1032                 iter->stream.uncompressed_size = 0;
1033         } else {
1034                 const index_group *g = (const index_group *)(
1035                                 stream->groups.rightmost);
1036
1037                 // Stream Header + Stream Footer + Index + Blocks
1038                 iter->stream.compressed_size = 2 * LZMA_STREAM_HEADER_SIZE
1039                                 + index_size(stream->record_count,
1040                                         stream->index_list_size)
1041                                 + vli_ceil4(g->records[g->last].unpadded_sum);
1042                 iter->stream.uncompressed_size
1043                                 = g->records[g->last].uncompressed_sum;
1044         }
1045
1046         if (group != NULL) {
1047                 iter->block.number_in_stream = group->number_base + record;
1048                 iter->block.number_in_file = iter->block.number_in_stream
1049                                 + stream->block_number_base;
1050
1051                 iter->block.compressed_stream_offset
1052                                 = record == 0 ? group->node.compressed_base
1053                                 : vli_ceil4(group->records[
1054                                         record - 1].unpadded_sum);
1055                 iter->block.uncompressed_stream_offset
1056                                 = record == 0 ? group->node.uncompressed_base
1057                                 : group->records[record - 1].uncompressed_sum;
1058
1059                 iter->block.uncompressed_size
1060                                 = group->records[record].uncompressed_sum
1061                                 - iter->block.uncompressed_stream_offset;
1062                 iter->block.unpadded_size
1063                                 = group->records[record].unpadded_sum
1064                                 - iter->block.compressed_stream_offset;
1065                 iter->block.total_size = vli_ceil4(iter->block.unpadded_size);
1066
1067                 iter->block.compressed_stream_offset
1068                                 += LZMA_STREAM_HEADER_SIZE;
1069
1070                 iter->block.compressed_file_offset
1071                                 = iter->block.compressed_stream_offset
1072                                 + iter->stream.compressed_offset;
1073                 iter->block.uncompressed_file_offset
1074                                 = iter->block.uncompressed_stream_offset
1075                                 + iter->stream.uncompressed_offset;
1076         }
1077
1078         return;
1079 }
1080
1081
1082 extern LZMA_API(void)
1083 lzma_index_iter_init(lzma_index_iter *iter, const lzma_index *i)
1084 {
1085         iter->internal[ITER_INDEX].p = i;
1086         lzma_index_iter_rewind(iter);
1087         return;
1088 }
1089
1090
1091 extern LZMA_API(void)
1092 lzma_index_iter_rewind(lzma_index_iter *iter)
1093 {
1094         iter->internal[ITER_STREAM].p = NULL;
1095         iter->internal[ITER_GROUP].p = NULL;
1096         iter->internal[ITER_RECORD].s = 0;
1097         iter->internal[ITER_METHOD].s = ITER_METHOD_NORMAL;
1098         return;
1099 }
1100
1101
1102 extern LZMA_API(lzma_bool)
1103 lzma_index_iter_next(lzma_index_iter *iter, lzma_index_iter_mode mode)
1104 {
1105         // Catch unsupported mode values.
1106         if ((unsigned int)(mode) > LZMA_INDEX_ITER_NONEMPTY_BLOCK)
1107                 return true;
1108
1109         const lzma_index *i = iter->internal[ITER_INDEX].p;
1110         const index_stream *stream = iter->internal[ITER_STREAM].p;
1111         const index_group *group = NULL;
1112         size_t record = iter->internal[ITER_RECORD].s;
1113
1114         // If we are being asked for the next Stream, leave group to NULL
1115         // so that the rest of the this function thinks that this Stream
1116         // has no groups and will thus go to the next Stream.
1117         if (mode != LZMA_INDEX_ITER_STREAM) {
1118                 // Get the pointer to the current group. See iter_set_inf()
1119                 // for explanation.
1120                 switch (iter->internal[ITER_METHOD].s) {
1121                 case ITER_METHOD_NORMAL:
1122                         group = iter->internal[ITER_GROUP].p;
1123                         break;
1124
1125                 case ITER_METHOD_NEXT:
1126                         group = index_tree_next(iter->internal[ITER_GROUP].p);
1127                         break;
1128
1129                 case ITER_METHOD_LEFTMOST:
1130                         group = (const index_group *)(
1131                                         stream->groups.leftmost);
1132                         break;
1133                 }
1134         }
1135
1136 again:
1137         if (stream == NULL) {
1138                 // We at the beginning of the lzma_index.
1139                 // Locate the first Stream.
1140                 stream = (const index_stream *)(i->streams.leftmost);
1141                 if (mode >= LZMA_INDEX_ITER_BLOCK) {
1142                         // Since we are being asked to return information
1143                         // about the first a Block, skip Streams that have
1144                         // no Blocks.
1145                         while (stream->groups.leftmost == NULL) {
1146                                 stream = index_tree_next(&stream->node);
1147                                 if (stream == NULL)
1148                                         return true;
1149                         }
1150                 }
1151
1152                 // Start from the first Record in the Stream.
1153                 group = (const index_group *)(stream->groups.leftmost);
1154                 record = 0;
1155
1156         } else if (group != NULL && record < group->last) {
1157                 // The next Record is in the same group.
1158                 ++record;
1159
1160         } else {
1161                 // This group has no more Records or this Stream has
1162                 // no Blocks at all.
1163                 record = 0;
1164
1165                 // If group is not NULL, this Stream has at least one Block
1166                 // and thus at least one group. Find the next group.
1167                 if (group != NULL)
1168                         group = index_tree_next(&group->node);
1169
1170                 if (group == NULL) {
1171                         // This Stream has no more Records. Find the next
1172                         // Stream. If we are being asked to return information
1173                         // about a Block, we skip empty Streams.
1174                         do {
1175                                 stream = index_tree_next(&stream->node);
1176                                 if (stream == NULL)
1177                                         return true;
1178                         } while (mode >= LZMA_INDEX_ITER_BLOCK
1179                                         && stream->groups.leftmost == NULL);
1180
1181                         group = (const index_group *)(
1182                                         stream->groups.leftmost);
1183                 }
1184         }
1185
1186         if (mode == LZMA_INDEX_ITER_NONEMPTY_BLOCK) {
1187                 // We need to look for the next Block again if this Block
1188                 // is empty.
1189                 if (record == 0) {
1190                         if (group->node.uncompressed_base
1191                                         == group->records[0].uncompressed_sum)
1192                                 goto again;
1193                 } else if (group->records[record - 1].uncompressed_sum
1194                                 == group->records[record].uncompressed_sum) {
1195                         goto again;
1196                 }
1197         }
1198
1199         iter->internal[ITER_STREAM].p = stream;
1200         iter->internal[ITER_GROUP].p = group;
1201         iter->internal[ITER_RECORD].s = record;
1202
1203         iter_set_info(iter);
1204
1205         return false;
1206 }
1207
1208
1209 extern LZMA_API(lzma_bool)
1210 lzma_index_iter_locate(lzma_index_iter *iter, lzma_vli target)
1211 {
1212         const lzma_index *i = iter->internal[ITER_INDEX].p;
1213
1214         // If the target is past the end of the file, return immediately.
1215         if (i->uncompressed_size <= target)
1216                 return true;
1217
1218         // Locate the Stream containing the target offset.
1219         const index_stream *stream = index_tree_locate(&i->streams, target);
1220         assert(stream != NULL);
1221         target -= stream->node.uncompressed_base;
1222
1223         // Locate the group containing the target offset.
1224         const index_group *group = index_tree_locate(&stream->groups, target);
1225         assert(group != NULL);
1226
1227         // Use binary search to locate the exact Record. It is the first
1228         // Record whose uncompressed_sum is greater than target.
1229         // This is because we want the rightmost Record that fullfills the
1230         // search criterion. It is possible that there are empty Blocks;
1231         // we don't want to return them.
1232         size_t left = 0;
1233         size_t right = group->last;
1234
1235         while (left < right) {
1236                 const size_t pos = left + (right - left) / 2;
1237                 if (group->records[pos].uncompressed_sum <= target)
1238                         left = pos + 1;
1239                 else
1240                         right = pos;
1241         }
1242
1243         iter->internal[ITER_STREAM].p = stream;
1244         iter->internal[ITER_GROUP].p = group;
1245         iter->internal[ITER_RECORD].s = left;
1246
1247         iter_set_info(iter);
1248
1249         return false;
1250 }