1#include "cache.h"
2#include "lockfile.h"
3#include "tree.h"
4#include "tree-walk.h"
5#include "cache-tree.h"
6
7#ifndef DEBUG
8#define DEBUG 0
9#endif
10
11struct cache_tree *cache_tree(void)
12{
13 struct cache_tree *it = xcalloc(1, sizeof(struct cache_tree));
14 it->entry_count = -1;
15 return it;
16}
17
18void cache_tree_free(struct cache_tree **it_p)
19{
20 int i;
21 struct cache_tree *it = *it_p;
22
23 if (!it)
24 return;
25 for (i = 0; i < it->subtree_nr; i++)
26 if (it->down[i]) {
27 cache_tree_free(&it->down[i]->cache_tree);
28 free(it->down[i]);
29 }
30 free(it->down);
31 free(it);
32 *it_p = NULL;
33}
34
35static int subtree_name_cmp(const char *one, int onelen,
36 const char *two, int twolen)
37{
38 if (onelen < twolen)
39 return -1;
40 if (twolen < onelen)
41 return 1;
42 return memcmp(one, two, onelen);
43}
44
45static int subtree_pos(struct cache_tree *it, const char *path, int pathlen)
46{
47 struct cache_tree_sub **down = it->down;
48 int lo, hi;
49 lo = 0;
50 hi = it->subtree_nr;
51 while (lo < hi) {
52 int mi = (lo + hi) / 2;
53 struct cache_tree_sub *mdl = down[mi];
54 int cmp = subtree_name_cmp(path, pathlen,
55 mdl->name, mdl->namelen);
56 if (!cmp)
57 return mi;
58 if (cmp < 0)
59 hi = mi;
60 else
61 lo = mi + 1;
62 }
63 return -lo-1;
64}
65
66static struct cache_tree_sub *find_subtree(struct cache_tree *it,
67 const char *path,
68 int pathlen,
69 int create)
70{
71 struct cache_tree_sub *down;
72 int pos = subtree_pos(it, path, pathlen);
73 if (0 <= pos)
74 return it->down[pos];
75 if (!create)
76 return NULL;
77
78 pos = -pos-1;
79 ALLOC_GROW(it->down, it->subtree_nr + 1, it->subtree_alloc);
80 it->subtree_nr++;
81
82 FLEX_ALLOC_MEM(down, name, path, pathlen);
83 down->cache_tree = NULL;
84 down->namelen = pathlen;
85
86 if (pos < it->subtree_nr)
87 memmove(it->down + pos + 1,
88 it->down + pos,
89 sizeof(down) * (it->subtree_nr - pos - 1));
90 it->down[pos] = down;
91 return down;
92}
93
94struct cache_tree_sub *cache_tree_sub(struct cache_tree *it, const char *path)
95{
96 int pathlen = strlen(path);
97 return find_subtree(it, path, pathlen, 1);
98}
99
100static int do_invalidate_path(struct cache_tree *it, const char *path)
101{
102 /* a/b/c
103 * ==> invalidate self
104 * ==> find "a", have it invalidate "b/c"
105 * a
106 * ==> invalidate self
107 * ==> if "a" exists as a subtree, remove it.
108 */
109 const char *slash;
110 int namelen;
111 struct cache_tree_sub *down;
112
113#if DEBUG
114 fprintf(stderr, "cache-tree invalidate <%s>\n", path);
115#endif
116
117 if (!it)
118 return 0;
119 slash = strchrnul(path, '/');
120 namelen = slash - path;
121 it->entry_count = -1;
122 if (!*slash) {
123 int pos;
124 pos = subtree_pos(it, path, namelen);
125 if (0 <= pos) {
126 cache_tree_free(&it->down[pos]->cache_tree);
127 free(it->down[pos]);
128 /* 0 1 2 3 4 5
129 * ^ ^subtree_nr = 6
130 * pos
131 * move 4 and 5 up one place (2 entries)
132 * 2 = 6 - 3 - 1 = subtree_nr - pos - 1
133 */
134 memmove(it->down+pos, it->down+pos+1,
135 sizeof(struct cache_tree_sub *) *
136 (it->subtree_nr - pos - 1));
137 it->subtree_nr--;
138 }
139 return 1;
140 }
141 down = find_subtree(it, path, namelen, 0);
142 if (down)
143 do_invalidate_path(down->cache_tree, slash + 1);
144 return 1;
145}
146
147void cache_tree_invalidate_path(struct index_state *istate, const char *path)
148{
149 if (do_invalidate_path(istate->cache_tree, path))
150 istate->cache_changed |= CACHE_TREE_CHANGED;
151}
152
153static int verify_cache(struct cache_entry **cache,
154 int entries, int flags)
155{
156 int i, funny;
157 int silent = flags & WRITE_TREE_SILENT;
158
159 /* Verify that the tree is merged */
160 funny = 0;
161 for (i = 0; i < entries; i++) {
162 const struct cache_entry *ce = cache[i];
163 if (ce_stage(ce)) {
164 if (silent)
165 return -1;
166 if (10 < ++funny) {
167 fprintf(stderr, "...\n");
168 break;
169 }
170 fprintf(stderr, "%s: unmerged (%s)\n",
171 ce->name, oid_to_hex(&ce->oid));
172 }
173 }
174 if (funny)
175 return -1;
176
177 /* Also verify that the cache does not have path and path/file
178 * at the same time. At this point we know the cache has only
179 * stage 0 entries.
180 */
181 funny = 0;
182 for (i = 0; i < entries - 1; i++) {
183 /* path/file always comes after path because of the way
184 * the cache is sorted. Also path can appear only once,
185 * which means conflicting one would immediately follow.
186 */
187 const char *this_name = cache[i]->name;
188 const char *next_name = cache[i+1]->name;
189 int this_len = strlen(this_name);
190 if (this_len < strlen(next_name) &&
191 strncmp(this_name, next_name, this_len) == 0 &&
192 next_name[this_len] == '/') {
193 if (10 < ++funny) {
194 fprintf(stderr, "...\n");
195 break;
196 }
197 fprintf(stderr, "You have both %s and %s\n",
198 this_name, next_name);
199 }
200 }
201 if (funny)
202 return -1;
203 return 0;
204}
205
206static void discard_unused_subtrees(struct cache_tree *it)
207{
208 struct cache_tree_sub **down = it->down;
209 int nr = it->subtree_nr;
210 int dst, src;
211 for (dst = src = 0; src < nr; src++) {
212 struct cache_tree_sub *s = down[src];
213 if (s->used)
214 down[dst++] = s;
215 else {
216 cache_tree_free(&s->cache_tree);
217 free(s);
218 it->subtree_nr--;
219 }
220 }
221}
222
223int cache_tree_fully_valid(struct cache_tree *it)
224{
225 int i;
226 if (!it)
227 return 0;
228 if (it->entry_count < 0 || !has_sha1_file(it->oid.hash))
229 return 0;
230 for (i = 0; i < it->subtree_nr; i++) {
231 if (!cache_tree_fully_valid(it->down[i]->cache_tree))
232 return 0;
233 }
234 return 1;
235}
236
237static int update_one(struct cache_tree *it,
238 struct cache_entry **cache,
239 int entries,
240 const char *base,
241 int baselen,
242 int *skip_count,
243 int flags)
244{
245 struct strbuf buffer;
246 int missing_ok = flags & WRITE_TREE_MISSING_OK;
247 int dryrun = flags & WRITE_TREE_DRY_RUN;
248 int repair = flags & WRITE_TREE_REPAIR;
249 int to_invalidate = 0;
250 int i;
251
252 assert(!(dryrun && repair));
253
254 *skip_count = 0;
255
256 if (0 <= it->entry_count && has_sha1_file(it->oid.hash))
257 return it->entry_count;
258
259 /*
260 * We first scan for subtrees and update them; we start by
261 * marking existing subtrees -- the ones that are unmarked
262 * should not be in the result.
263 */
264 for (i = 0; i < it->subtree_nr; i++)
265 it->down[i]->used = 0;
266
267 /*
268 * Find the subtrees and update them.
269 */
270 i = 0;
271 while (i < entries) {
272 const struct cache_entry *ce = cache[i];
273 struct cache_tree_sub *sub;
274 const char *path, *slash;
275 int pathlen, sublen, subcnt, subskip;
276
277 path = ce->name;
278 pathlen = ce_namelen(ce);
279 if (pathlen <= baselen || memcmp(base, path, baselen))
280 break; /* at the end of this level */
281
282 slash = strchr(path + baselen, '/');
283 if (!slash) {
284 i++;
285 continue;
286 }
287 /*
288 * a/bbb/c (base = a/, slash = /c)
289 * ==>
290 * path+baselen = bbb/c, sublen = 3
291 */
292 sublen = slash - (path + baselen);
293 sub = find_subtree(it, path + baselen, sublen, 1);
294 if (!sub->cache_tree)
295 sub->cache_tree = cache_tree();
296 subcnt = update_one(sub->cache_tree,
297 cache + i, entries - i,
298 path,
299 baselen + sublen + 1,
300 &subskip,
301 flags);
302 if (subcnt < 0)
303 return subcnt;
304 if (!subcnt)
305 die("index cache-tree records empty sub-tree");
306 i += subcnt;
307 sub->count = subcnt; /* to be used in the next loop */
308 *skip_count += subskip;
309 sub->used = 1;
310 }
311
312 discard_unused_subtrees(it);
313
314 /*
315 * Then write out the tree object for this level.
316 */
317 strbuf_init(&buffer, 8192);
318
319 i = 0;
320 while (i < entries) {
321 const struct cache_entry *ce = cache[i];
322 struct cache_tree_sub *sub = NULL;
323 const char *path, *slash;
324 int pathlen, entlen;
325 const unsigned char *sha1;
326 unsigned mode;
327 int expected_missing = 0;
328 int contains_ita = 0;
329
330 path = ce->name;
331 pathlen = ce_namelen(ce);
332 if (pathlen <= baselen || memcmp(base, path, baselen))
333 break; /* at the end of this level */
334
335 slash = strchr(path + baselen, '/');
336 if (slash) {
337 entlen = slash - (path + baselen);
338 sub = find_subtree(it, path + baselen, entlen, 0);
339 if (!sub)
340 die("cache-tree.c: '%.*s' in '%s' not found",
341 entlen, path + baselen, path);
342 i += sub->count;
343 sha1 = sub->cache_tree->oid.hash;
344 mode = S_IFDIR;
345 contains_ita = sub->cache_tree->entry_count < 0;
346 if (contains_ita) {
347 to_invalidate = 1;
348 expected_missing = 1;
349 }
350 }
351 else {
352 sha1 = ce->oid.hash;
353 mode = ce->ce_mode;
354 entlen = pathlen - baselen;
355 i++;
356 }
357
358 if (is_null_sha1(sha1) ||
359 (mode != S_IFGITLINK && !missing_ok && !has_sha1_file(sha1))) {
360 strbuf_release(&buffer);
361 if (expected_missing)
362 return -1;
363 return error("invalid object %06o %s for '%.*s'",
364 mode, sha1_to_hex(sha1), entlen+baselen, path);
365 }
366
367 /*
368 * CE_REMOVE entries are removed before the index is
369 * written to disk. Skip them to remain consistent
370 * with the future on-disk index.
371 */
372 if (ce->ce_flags & CE_REMOVE) {
373 *skip_count = *skip_count + 1;
374 continue;
375 }
376
377 /*
378 * CE_INTENT_TO_ADD entries exist on on-disk index but
379 * they are not part of generated trees. Invalidate up
380 * to root to force cache-tree users to read elsewhere.
381 */
382 if (!sub && ce_intent_to_add(ce)) {
383 to_invalidate = 1;
384 continue;
385 }
386
387 /*
388 * "sub" can be an empty tree if all subentries are i-t-a.
389 */
390 if (contains_ita && !hashcmp(sha1, EMPTY_TREE_SHA1_BIN))
391 continue;
392
393 strbuf_grow(&buffer, entlen + 100);
394 strbuf_addf(&buffer, "%o %.*s%c", mode, entlen, path + baselen, '\0');
395 strbuf_add(&buffer, sha1, 20);
396
397#if DEBUG
398 fprintf(stderr, "cache-tree update-one %o %.*s\n",
399 mode, entlen, path + baselen);
400#endif
401 }
402
403 if (repair) {
404 unsigned char sha1[20];
405 hash_sha1_file(buffer.buf, buffer.len, tree_type, sha1);
406 if (has_sha1_file(sha1))
407 hashcpy(it->oid.hash, sha1);
408 else
409 to_invalidate = 1;
410 } else if (dryrun)
411 hash_sha1_file(buffer.buf, buffer.len, tree_type,
412 it->oid.hash);
413 else if (write_sha1_file(buffer.buf, buffer.len, tree_type, it->oid.hash)) {
414 strbuf_release(&buffer);
415 return -1;
416 }
417
418 strbuf_release(&buffer);
419 it->entry_count = to_invalidate ? -1 : i - *skip_count;
420#if DEBUG
421 fprintf(stderr, "cache-tree update-one (%d ent, %d subtree) %s\n",
422 it->entry_count, it->subtree_nr,
423 oid_to_hex(&it->oid));
424#endif
425 return i;
426}
427
428int cache_tree_update(struct index_state *istate, int flags)
429{
430 struct cache_tree *it = istate->cache_tree;
431 struct cache_entry **cache = istate->cache;
432 int entries = istate->cache_nr;
433 int skip, i = verify_cache(cache, entries, flags);
434
435 if (i)
436 return i;
437 i = update_one(it, cache, entries, "", 0, &skip, flags);
438 if (i < 0)
439 return i;
440 istate->cache_changed |= CACHE_TREE_CHANGED;
441 return 0;
442}
443
444static void write_one(struct strbuf *buffer, struct cache_tree *it,
445 const char *path, int pathlen)
446{
447 int i;
448
449 /* One "cache-tree" entry consists of the following:
450 * path (NUL terminated)
451 * entry_count, subtree_nr ("%d %d\n")
452 * tree-sha1 (missing if invalid)
453 * subtree_nr "cache-tree" entries for subtrees.
454 */
455 strbuf_grow(buffer, pathlen + 100);
456 strbuf_add(buffer, path, pathlen);
457 strbuf_addf(buffer, "%c%d %d\n", 0, it->entry_count, it->subtree_nr);
458
459#if DEBUG
460 if (0 <= it->entry_count)
461 fprintf(stderr, "cache-tree <%.*s> (%d ent, %d subtree) %s\n",
462 pathlen, path, it->entry_count, it->subtree_nr,
463 oid_to_hex(&it->oid));
464 else
465 fprintf(stderr, "cache-tree <%.*s> (%d subtree) invalid\n",
466 pathlen, path, it->subtree_nr);
467#endif
468
469 if (0 <= it->entry_count) {
470 strbuf_add(buffer, it->oid.hash, 20);
471 }
472 for (i = 0; i < it->subtree_nr; i++) {
473 struct cache_tree_sub *down = it->down[i];
474 if (i) {
475 struct cache_tree_sub *prev = it->down[i-1];
476 if (subtree_name_cmp(down->name, down->namelen,
477 prev->name, prev->namelen) <= 0)
478 die("fatal - unsorted cache subtree");
479 }
480 write_one(buffer, down->cache_tree, down->name, down->namelen);
481 }
482}
483
484void cache_tree_write(struct strbuf *sb, struct cache_tree *root)
485{
486 write_one(sb, root, "", 0);
487}
488
489static struct cache_tree *read_one(const char **buffer, unsigned long *size_p)
490{
491 const char *buf = *buffer;
492 unsigned long size = *size_p;
493 const char *cp;
494 char *ep;
495 struct cache_tree *it;
496 int i, subtree_nr;
497
498 it = NULL;
499 /* skip name, but make sure name exists */
500 while (size && *buf) {
501 size--;
502 buf++;
503 }
504 if (!size)
505 goto free_return;
506 buf++; size--;
507 it = cache_tree();
508
509 cp = buf;
510 it->entry_count = strtol(cp, &ep, 10);
511 if (cp == ep)
512 goto free_return;
513 cp = ep;
514 subtree_nr = strtol(cp, &ep, 10);
515 if (cp == ep)
516 goto free_return;
517 while (size && *buf && *buf != '\n') {
518 size--;
519 buf++;
520 }
521 if (!size)
522 goto free_return;
523 buf++; size--;
524 if (0 <= it->entry_count) {
525 if (size < 20)
526 goto free_return;
527 hashcpy(it->oid.hash, (const unsigned char*)buf);
528 buf += 20;
529 size -= 20;
530 }
531
532#if DEBUG
533 if (0 <= it->entry_count)
534 fprintf(stderr, "cache-tree <%s> (%d ent, %d subtree) %s\n",
535 *buffer, it->entry_count, subtree_nr,
536 oid_to_hex(&it->oid));
537 else
538 fprintf(stderr, "cache-tree <%s> (%d subtrees) invalid\n",
539 *buffer, subtree_nr);
540#endif
541
542 /*
543 * Just a heuristic -- we do not add directories that often but
544 * we do not want to have to extend it immediately when we do,
545 * hence +2.
546 */
547 it->subtree_alloc = subtree_nr + 2;
548 it->down = xcalloc(it->subtree_alloc, sizeof(struct cache_tree_sub *));
549 for (i = 0; i < subtree_nr; i++) {
550 /* read each subtree */
551 struct cache_tree *sub;
552 struct cache_tree_sub *subtree;
553 const char *name = buf;
554
555 sub = read_one(&buf, &size);
556 if (!sub)
557 goto free_return;
558 subtree = cache_tree_sub(it, name);
559 subtree->cache_tree = sub;
560 }
561 if (subtree_nr != it->subtree_nr)
562 die("cache-tree: internal error");
563 *buffer = buf;
564 *size_p = size;
565 return it;
566
567 free_return:
568 cache_tree_free(&it);
569 return NULL;
570}
571
572struct cache_tree *cache_tree_read(const char *buffer, unsigned long size)
573{
574 if (buffer[0])
575 return NULL; /* not the whole tree */
576 return read_one(&buffer, &size);
577}
578
579static struct cache_tree *cache_tree_find(struct cache_tree *it, const char *path)
580{
581 if (!it)
582 return NULL;
583 while (*path) {
584 const char *slash;
585 struct cache_tree_sub *sub;
586
587 slash = strchrnul(path, '/');
588 /*
589 * Between path and slash is the name of the subtree
590 * to look for.
591 */
592 sub = find_subtree(it, path, slash - path, 0);
593 if (!sub)
594 return NULL;
595 it = sub->cache_tree;
596
597 path = slash;
598 while (*path == '/')
599 path++;
600 }
601 return it;
602}
603
604int write_index_as_tree(unsigned char *sha1, struct index_state *index_state, const char *index_path, int flags, const char *prefix)
605{
606 int entries, was_valid, newfd;
607 struct lock_file *lock_file;
608
609 /*
610 * We can't free this memory, it becomes part of a linked list
611 * parsed atexit()
612 */
613 lock_file = xcalloc(1, sizeof(struct lock_file));
614
615 newfd = hold_lock_file_for_update(lock_file, index_path, LOCK_DIE_ON_ERROR);
616
617 entries = read_index_from(index_state, index_path);
618 if (entries < 0)
619 return WRITE_TREE_UNREADABLE_INDEX;
620 if (flags & WRITE_TREE_IGNORE_CACHE_TREE)
621 cache_tree_free(&index_state->cache_tree);
622
623 if (!index_state->cache_tree)
624 index_state->cache_tree = cache_tree();
625
626 was_valid = cache_tree_fully_valid(index_state->cache_tree);
627 if (!was_valid) {
628 if (cache_tree_update(index_state, flags) < 0)
629 return WRITE_TREE_UNMERGED_INDEX;
630 if (0 <= newfd) {
631 if (!write_locked_index(index_state, lock_file, COMMIT_LOCK))
632 newfd = -1;
633 }
634 /* Not being able to write is fine -- we are only interested
635 * in updating the cache-tree part, and if the next caller
636 * ends up using the old index with unupdated cache-tree part
637 * it misses the work we did here, but that is just a
638 * performance penalty and not a big deal.
639 */
640 }
641
642 if (prefix) {
643 struct cache_tree *subtree;
644 subtree = cache_tree_find(index_state->cache_tree, prefix);
645 if (!subtree)
646 return WRITE_TREE_PREFIX_ERROR;
647 hashcpy(sha1, subtree->oid.hash);
648 }
649 else
650 hashcpy(sha1, index_state->cache_tree->oid.hash);
651
652 if (0 <= newfd)
653 rollback_lock_file(lock_file);
654
655 return 0;
656}
657
658int write_cache_as_tree(unsigned char *sha1, int flags, const char *prefix)
659{
660 return write_index_as_tree(sha1, &the_index, get_index_file(), flags, prefix);
661}
662
663static void prime_cache_tree_rec(struct cache_tree *it, struct tree *tree)
664{
665 struct tree_desc desc;
666 struct name_entry entry;
667 int cnt;
668
669 oidcpy(&it->oid, &tree->object.oid);
670 init_tree_desc(&desc, tree->buffer, tree->size);
671 cnt = 0;
672 while (tree_entry(&desc, &entry)) {
673 if (!S_ISDIR(entry.mode))
674 cnt++;
675 else {
676 struct cache_tree_sub *sub;
677 struct tree *subtree = lookup_tree(entry.oid);
678 if (!subtree->object.parsed)
679 parse_tree(subtree);
680 sub = cache_tree_sub(it, entry.path);
681 sub->cache_tree = cache_tree();
682 prime_cache_tree_rec(sub->cache_tree, subtree);
683 cnt += sub->cache_tree->entry_count;
684 }
685 }
686 it->entry_count = cnt;
687}
688
689void prime_cache_tree(struct index_state *istate, struct tree *tree)
690{
691 cache_tree_free(&istate->cache_tree);
692 istate->cache_tree = cache_tree();
693 prime_cache_tree_rec(istate->cache_tree, tree);
694 istate->cache_changed |= CACHE_TREE_CHANGED;
695}
696
697/*
698 * find the cache_tree that corresponds to the current level without
699 * exploding the full path into textual form. The root of the
700 * cache tree is given as "root", and our current level is "info".
701 * (1) When at root level, info->prev is NULL, so it is "root" itself.
702 * (2) Otherwise, find the cache_tree that corresponds to one level
703 * above us, and find ourselves in there.
704 */
705static struct cache_tree *find_cache_tree_from_traversal(struct cache_tree *root,
706 struct traverse_info *info)
707{
708 struct cache_tree *our_parent;
709
710 if (!info->prev)
711 return root;
712 our_parent = find_cache_tree_from_traversal(root, info->prev);
713 return cache_tree_find(our_parent, info->name.path);
714}
715
716int cache_tree_matches_traversal(struct cache_tree *root,
717 struct name_entry *ent,
718 struct traverse_info *info)
719{
720 struct cache_tree *it;
721
722 it = find_cache_tree_from_traversal(root, info);
723 it = cache_tree_find(it, ent->path);
724 if (it && it->entry_count > 0 && !oidcmp(ent->oid, &it->oid))
725 return it->entry_count;
726 return 0;
727}
728
729int update_main_cache_tree(int flags)
730{
731 if (!the_index.cache_tree)
732 the_index.cache_tree = cache_tree();
733 return cache_tree_update(&the_index, flags);
734}