read-cache.c · leroysheep/LeeRoySheep

test

bisect jch maint master next seen test todo

c 4,056 lines 112 KB

Raw

1	/*
2	* GIT - The information manager from hell
3	*
4	* Copyright (C) Linus Torvalds, 2005
5	*/
6
7	#define USE_THE_REPOSITORY_VARIABLE
8	#define DISABLE_SIGN_COMPARE_WARNINGS
9
10	#include "git-compat-util.h"
11	#include "config.h"
12	#include "date.h"
13	#include "diff.h"
14	#include "diffcore.h"
15	#include "hex.h"
16	#include "tempfile.h"
17	#include "lockfile.h"
18	#include "cache-tree.h"
19	#include "refs.h"
20	#include "dir.h"
21	#include "object-file.h"
22	#include "odb.h"
23	#include "odb/transaction.h"
24	#include "oid-array.h"
25	#include "tree.h"
26	#include "commit.h"
27	#include "environment.h"
28	#include "gettext.h"
29	#include "mem-pool.h"
30	#include "name-hash.h"
31	#include "object-name.h"
32	#include "path.h"
33	#include "preload-index.h"
34	#include "read-cache.h"
35	#include "repository.h"
36	#include "resolve-undo.h"
37	#include "revision.h"
38	#include "strbuf.h"
39	#include "trace2.h"
40	#include "varint.h"
41	#include "split-index.h"
42	#include "symlinks.h"
43	#include "utf8.h"
44	#include "fsmonitor.h"
45	#include "thread-utils.h"
46	#include "progress.h"
47	#include "sparse-index.h"
48	#include "csum-file.h"
49	#include "promisor-remote.h"
50	#include "hook.h"
51	#include "submodule.h"
52	#include "submodule-config.h"
53	#include "advice.h"
54
55	/* Mask for the name length in ce_flags in the on-disk index */
56
57	#define CE_NAMEMASK (0x0fff)
58
59	/* Index extensions.
60	*
61	* The first letter should be 'A'..'Z' for extensions that are not
62	* necessary for a correct operation (i.e. optimization data).
63	* When new extensions are added that _needs_ to be understood in
64	* order to correctly interpret the index file, pick character that
65	* is outside the range, to cause the reader to abort.
66	*/
67
68	#define CACHE_EXT(s) ( (s[0]<<24)\|(s[1]<<16)\|(s[2]<<8)\|(s[3]) )
69	#define CACHE_EXT_TREE 0x54524545 /* "TREE" */
70	#define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
71	#define CACHE_EXT_LINK 0x6c696e6b /* "link" */
72	#define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
73	#define CACHE_EXT_FSMONITOR 0x46534D4E /* "FSMN" */
74	#define CACHE_EXT_ENDOFINDEXENTRIES 0x454F4945 /* "EOIE" */
75	#define CACHE_EXT_INDEXENTRYOFFSETTABLE 0x49454F54 /* "IEOT" */
76	#define CACHE_EXT_SPARSE_DIRECTORIES 0x73646972 /* "sdir" */
77
78	/* changes that can be kept in $GIT_DIR/index (basically all extensions) */
79	#define EXTMASK (RESOLVE_UNDO_CHANGED \| CACHE_TREE_CHANGED \| \
80	CE_ENTRY_ADDED \| CE_ENTRY_REMOVED \| CE_ENTRY_CHANGED \| \
81	SPLIT_INDEX_ORDERED \| UNTRACKED_CHANGED \| FSMONITOR_CHANGED)
82
83
84	/*
85	* This is an estimate of the pathname length in the index. We use
86	* this for V4 index files to guess the un-deltafied size of the index
87	* in memory because of pathname deltafication. This is not required
88	* for V2/V3 index formats because their pathnames are not compressed.
89	* If the initial amount of memory set aside is not sufficient, the
90	* mem pool will allocate extra memory.
91	*/
92	#define CACHE_ENTRY_PATH_LENGTH 80
93
94	enum index_search_mode {
95	NO_EXPAND_SPARSE = 0,
96	EXPAND_SPARSE = 1
97	};
98
99	static inline struct cache_entry mem_pool__ce_alloc(struct mem_pool mem_pool, size_t len)
100	{
101	struct cache_entry *ce;
102	ce = mem_pool_alloc(mem_pool, cache_entry_size(len));
103	ce->mem_pool_allocated = 1;
104	return ce;
105	}
106
107	static inline struct cache_entry mem_pool__ce_calloc(struct mem_pool mem_pool, size_t len)
108	{
109	struct cache_entry * ce;
110	ce = mem_pool_calloc(mem_pool, 1, cache_entry_size(len));
111	ce->mem_pool_allocated = 1;
112	return ce;
113	}
114
115	static struct mem_pool find_mem_pool(struct index_state istate)
116	{
117	struct mem_pool **pool_ptr;
118
119	if (istate->split_index && istate->split_index->base)
120	pool_ptr = &istate->split_index->base->ce_mem_pool;
121	else
122	pool_ptr = &istate->ce_mem_pool;
123
124	if (!*pool_ptr) {
125	pool_ptr = xmalloc(sizeof(*pool_ptr));
126	mem_pool_init(*pool_ptr, 0);
127	}
128
129	return *pool_ptr;
130	}
131
132	static const char *alternate_index_output;
133
134	static void set_index_entry(struct index_state istate, int nr, struct cache_entry ce)
135	{
136	if (S_ISSPARSEDIR(ce->ce_mode))
137	istate->sparse_index = INDEX_COLLAPSED;
138
139	istate->cache[nr] = ce;
140	add_name_hash(istate, ce);
141	}
142
143	static void replace_index_entry(struct index_state istate, int nr, struct cache_entry ce)
144	{
145	struct cache_entry *old = istate->cache[nr];
146
147	replace_index_entry_in_base(istate, old, ce);
148	remove_name_hash(istate, old);
149	discard_cache_entry(old);
150	ce->ce_flags &= ~CE_HASHED;
151	set_index_entry(istate, nr, ce);
152	ce->ce_flags \|= CE_UPDATE_IN_BASE;
153	mark_fsmonitor_invalid(istate, ce);
154	istate->cache_changed \|= CE_ENTRY_CHANGED;
155	}
156
157	void rename_index_entry_at(struct index_state istate, int nr, const char new_name)
158	{
159	struct cache_entry old_entry = istate->cache[nr], new_entry, *refreshed;
160	int namelen = strlen(new_name);
161
162	new_entry = make_empty_cache_entry(istate, namelen);
163	copy_cache_entry(new_entry, old_entry);
164	new_entry->ce_flags &= ~CE_HASHED;
165	new_entry->ce_namelen = namelen;
166	new_entry->index = 0;
167	memcpy(new_entry->name, new_name, namelen + 1);
168
169	cache_tree_invalidate_path(istate, old_entry->name);
170	untracked_cache_remove_from_index(istate, old_entry->name);
171	remove_index_entry_at(istate, nr);
172
173	/*
174	* Refresh the new index entry. Using 'refresh_cache_entry' ensures
175	* we only update stat info if the entry is otherwise up-to-date (i.e.,
176	* the contents/mode haven't changed). This ensures that we reflect the
177	* 'ctime' of the rename in the index without (incorrectly) updating
178	* the cached stat info to reflect unstaged changes on disk.
179	*/
180	refreshed = refresh_cache_entry(istate, new_entry, CE_MATCH_REFRESH);
181	if (refreshed && refreshed != new_entry) {
182	add_index_entry(istate, refreshed, ADD_CACHE_OK_TO_ADD\|ADD_CACHE_OK_TO_REPLACE);
183	discard_cache_entry(new_entry);
184	} else
185	add_index_entry(istate, new_entry, ADD_CACHE_OK_TO_ADD\|ADD_CACHE_OK_TO_REPLACE);
186	}
187
188	/*
189	* This only updates the "non-critical" parts of the directory
190	* cache, ie the parts that aren't tracked by GIT, and only used
191	* to validate the cache.
192	*/
193	void fill_stat_cache_info(struct index_state istate, struct cache_entry ce, struct stat *st)
194	{
195	fill_stat_data(&ce->ce_stat_data, st);
196
197	if (assume_unchanged)
198	ce->ce_flags \|= CE_VALID;
199
200	if (S_ISREG(st->st_mode)) {
201	ce_mark_uptodate(ce);
202	mark_fsmonitor_valid(istate, ce);
203	}
204	}
205
206	static unsigned int st_mode_from_ce(const struct cache_entry *ce)
207	{
208	switch (ce->ce_mode & S_IFMT) {
209	case S_IFLNK:
210	return repo_has_symlinks(the_repository) ? S_IFLNK : (S_IFREG \| 0644);
211	case S_IFREG:
212	return (ce->ce_mode & (repo_trust_executable_bit(the_repository) ? 0755 : 0644)) \| S_IFREG;
213	case S_IFGITLINK:
214	return S_IFDIR \| 0755;
215	case S_IFDIR:
216	return ce->ce_mode;
217	default:
218	BUG("unsupported ce_mode: %o", ce->ce_mode);
219	}
220	}
221
222	int fake_lstat(const struct cache_entry ce, struct stat st)
223	{
224	fake_lstat_data(&ce->ce_stat_data, st);
225	st->st_mode = st_mode_from_ce(ce);
226
227	/* always succeed as lstat() replacement */
228	return 0;
229	}
230
231	static int ce_compare_data(struct index_state *istate,
232	const struct cache_entry *ce,
233	struct stat *st)
234	{
235	int match = -1;
236	int fd = git_open_cloexec(ce->name, O_RDONLY);
237
238	if (fd >= 0) {
239	struct object_id oid;
240	if (!index_fd(istate, &oid, fd, st, OBJ_BLOB, ce->name, 0))
241	match = !oideq(&oid, &ce->oid);
242	/* index_fd() closed the file descriptor already */
243	}
244	return match;
245	}
246
247	static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
248	{
249	int match = -1;
250	void *buffer;
251	size_t size;
252	enum object_type type;
253	struct strbuf sb = STRBUF_INIT;
254
255	if (strbuf_readlink(&sb, ce->name, expected_size))
256	return -1;
257
258	buffer = odb_read_object(the_repository->objects, &ce->oid, &type, &size);
259	if (buffer) {
260	if (size == sb.len)
261	match = memcmp(buffer, sb.buf, size);
262	free(buffer);
263	}
264	strbuf_release(&sb);
265	return match;
266	}
267
268	static int ce_compare_gitlink(const struct cache_entry *ce)
269	{
270	struct object_id oid;
271
272	/*
273	* We don't actually require that the .git directory
274	* under GITLINK directory be a valid git directory. It
275	* might even be missing (in case nobody populated that
276	* sub-project).
277	*
278	* If so, we consider it always to match.
279	*/
280	if (repo_resolve_gitlink_ref(the_repository, ce->name,
281	"HEAD", &oid) < 0)
282	return 0;
283	return !oideq(&oid, &ce->oid);
284	}
285
286	static int ce_modified_check_fs(struct index_state *istate,
287	const struct cache_entry *ce,
288	struct stat *st)
289	{
290	switch (st->st_mode & S_IFMT) {
291	case S_IFREG:
292	if (ce_compare_data(istate, ce, st))
293	return DATA_CHANGED;
294	break;
295	case S_IFLNK:
296	if (ce_compare_link(ce, xsize_t(st->st_size)))
297	return DATA_CHANGED;
298	break;
299	case S_IFDIR:
300	if (S_ISGITLINK(ce->ce_mode))
301	return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
302	/* else fallthrough */
303	default:
304	return TYPE_CHANGED;
305	}
306	return 0;
307	}
308
309	static int ce_match_stat_basic(const struct cache_entry ce, struct stat st)
310	{
311	unsigned int changed = 0;
312
313	if (ce->ce_flags & CE_REMOVE)
314	return MODE_CHANGED \| DATA_CHANGED \| TYPE_CHANGED;
315
316	switch (ce->ce_mode & S_IFMT) {
317	case S_IFREG:
318	changed \|= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
319	/* We consider only the owner x bit to be relevant for
320	* "mode changes"
321	*/
322	if (repo_trust_executable_bit(the_repository) &&
323	(0100 & (ce->ce_mode ^ st->st_mode)))
324	changed \|= MODE_CHANGED;
325	break;
326	case S_IFLNK:
327	if (!S_ISLNK(st->st_mode) &&
328	(repo_has_symlinks(the_repository) \|\| !S_ISREG(st->st_mode)))
329	changed \|= TYPE_CHANGED;
330	break;
331	case S_IFGITLINK:
332	/* We ignore most of the st_xxx fields for gitlinks */
333	if (!S_ISDIR(st->st_mode))
334	changed \|= TYPE_CHANGED;
335	else if (ce_compare_gitlink(ce))
336	changed \|= DATA_CHANGED;
337	return changed;
338	default:
339	BUG("unsupported ce_mode: %o", ce->ce_mode);
340	}
341
342	changed \|= match_stat_data(&ce->ce_stat_data, st);
343
344	/* Racily smudged entry? */
345	if (!ce->ce_stat_data.sd_size) {
346	if (!is_empty_blob_oid(&ce->oid, the_repository->hash_algo))
347	changed \|= DATA_CHANGED;
348	}
349
350	return changed;
351	}
352
353	static int is_racy_stat(const struct index_state *istate,
354	const struct stat_data *sd)
355	{
356	return (istate->timestamp.sec &&
357	#ifdef USE_NSEC
358	/* nanosecond timestamped files can also be racy! */
359	(istate->timestamp.sec < sd->sd_mtime.sec \|\|
360	(istate->timestamp.sec == sd->sd_mtime.sec &&
361	istate->timestamp.nsec <= sd->sd_mtime.nsec))
362	#else
363	istate->timestamp.sec <= sd->sd_mtime.sec
364	#endif
365	);
366	}
367
368	int is_racy_timestamp(const struct index_state *istate,
369	const struct cache_entry *ce)
370	{
371	return (!S_ISGITLINK(ce->ce_mode) &&
372	is_racy_stat(istate, &ce->ce_stat_data));
373	}
374
375	int match_stat_data_racy(const struct index_state *istate,
376	const struct stat_data sd, struct stat st)
377	{
378	if (is_racy_stat(istate, sd))
379	return MTIME_CHANGED;
380	return match_stat_data(sd, st);
381	}
382
383	int ie_match_stat(struct index_state *istate,
384	const struct cache_entry ce, struct stat st,
385	unsigned int options)
386	{
387	unsigned int changed;
388	int ignore_valid = options & CE_MATCH_IGNORE_VALID;
389	int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
390	int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
391	int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
392
393	if (!ignore_fsmonitor)
394	refresh_fsmonitor(istate);
395	/*
396	* If it's marked as always valid in the index, it's
397	* valid whatever the checked-out copy says.
398	*
399	* skip-worktree has the same effect with higher precedence
400	*/
401	if (!ignore_skip_worktree && ce_skip_worktree(ce))
402	return 0;
403	if (!ignore_valid && (ce->ce_flags & CE_VALID))
404	return 0;
405	if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID))
406	return 0;
407
408	/*
409	* Intent-to-add entries have not been added, so the index entry
410	* by definition never matches what is in the work tree until it
411	* actually gets added.
412	*/
413	if (ce_intent_to_add(ce))
414	return DATA_CHANGED \| TYPE_CHANGED \| MODE_CHANGED;
415
416	changed = ce_match_stat_basic(ce, st);
417
418	/*
419	* Within 1 second of this sequence:
420	* echo xyzzy >file && git-update-index --add file
421	* running this command:
422	* echo frotz >file
423	* would give a falsely clean cache entry. The mtime and
424	* length match the cache, and other stat fields do not change.
425	*
426	* We could detect this at update-index time (the cache entry
427	* being registered/updated records the same time as "now")
428	* and delay the return from git-update-index, but that would
429	* effectively mean we can make at most one commit per second,
430	* which is not acceptable. Instead, we check cache entries
431	* whose mtime are the same as the index file timestamp more
432	* carefully than others.
433	*/
434	if (!changed && is_racy_timestamp(istate, ce)) {
435	if (assume_racy_is_modified)
436	changed \|= DATA_CHANGED;
437	else
438	changed \|= ce_modified_check_fs(istate, ce, st);
439	}
440
441	return changed;
442	}
443
444	int ie_modified(struct index_state *istate,
445	const struct cache_entry *ce,
446	struct stat *st, unsigned int options)
447	{
448	int changed, changed_fs;
449
450	changed = ie_match_stat(istate, ce, st, options);
451	if (!changed)
452	return 0;
453	/*
454	* If the mode or type has changed, there's no point in trying
455	* to refresh the entry - it's not going to match
456	*/
457	if (changed & (MODE_CHANGED \| TYPE_CHANGED))
458	return changed;
459
460	/*
461	* Immediately after read-tree or update-index --cacheinfo,
462	* the length field is zero, as we have never even read the
463	* lstat(2) information once, and we cannot trust DATA_CHANGED
464	* returned by ie_match_stat() which in turn was returned by
465	* ce_match_stat_basic() to signal that the filesize of the
466	* blob changed. We have to actually go to the filesystem to
467	* see if the contents match, and if so, should answer "unchanged".
468	*
469	* The logic does not apply to gitlinks, as ce_match_stat_basic()
470	* already has checked the actual HEAD from the filesystem in the
471	* subproject. If ie_match_stat() already said it is different,
472	* then we know it is.
473	*/
474	if ((changed & DATA_CHANGED) &&
475	#ifdef GIT_WINDOWS_NATIVE
476	/*
477	* Work around Git for Windows v2.27.0 fixing a bug where symlinks'
478	* target path lengths were not read at all, and instead recorded
479	* as 4096: now, all symlinks would appear as modified.
480	*
481	* So let's just special-case symlinks with a target path length
482	* (i.e. `sd_size`) of 4096 and force them to be re-checked.
483	*/
484	(!S_ISLNK(st->st_mode) \|\| ce->ce_stat_data.sd_size != MAX_PATH) &&
485	#endif
486	(S_ISGITLINK(ce->ce_mode) \|\| ce->ce_stat_data.sd_size != 0))
487	return changed;
488
489	changed_fs = ce_modified_check_fs(istate, ce, st);
490	if (changed_fs)
491	return changed \| changed_fs;
492	return 0;
493	}
494
495	static int cache_name_stage_compare(const char *name1, int len1, int stage1,
496	const char *name2, int len2, int stage2)
497	{
498	int cmp;
499
500	cmp = name_compare(name1, len1, name2, len2);
501	if (cmp)
502	return cmp;
503
504	if (stage1 < stage2)
505	return -1;
506	if (stage1 > stage2)
507	return 1;
508	return 0;
509	}
510
511	int cmp_cache_name_compare(const void a_, const void b_)
512	{
513	const struct cache_entry ce1, ce2;
514
515	ce1 = ((const struct cache_entry *)a_);
516	ce2 = ((const struct cache_entry *)b_);
517	return cache_name_stage_compare(ce1->name, ce1->ce_namelen, ce_stage(ce1),
518	ce2->name, ce2->ce_namelen, ce_stage(ce2));
519	}
520
521	static int index_name_stage_pos(struct index_state *istate,
522	const char *name, int namelen,
523	int stage,
524	enum index_search_mode search_mode)
525	{
526	int first, last;
527
528	first = 0;
529	last = istate->cache_nr;
530	while (last > first) {
531	int next = first + ((last - first) >> 1);
532	struct cache_entry *ce = istate->cache[next];
533	int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
534	if (!cmp)
535	return next;
536	if (cmp < 0) {
537	last = next;
538	continue;
539	}
540	first = next+1;
541	}
542
543	if (search_mode == EXPAND_SPARSE && istate->sparse_index &&
544	first > 0) {
545	/* Note: first <= istate->cache_nr */
546	struct cache_entry *ce = istate->cache[first - 1];
547
548	/*
549	* If we are in a sparse-index _and_ the entry before the
550	* insertion position is a sparse-directory entry that is
551	* an ancestor of 'name', then we need to expand the index
552	* and search again. This will only trigger once, because
553	* thereafter the index is fully expanded.
554	*/
555	if (S_ISSPARSEDIR(ce->ce_mode) &&
556	ce_namelen(ce) < namelen &&
557	!strncmp(name, ce->name, ce_namelen(ce))) {
558	ensure_full_index(istate);
559	return index_name_stage_pos(istate, name, namelen, stage, search_mode);
560	}
561	}
562
563	return -first-1;
564	}
565
566	int index_name_pos(struct index_state istate, const char name, int namelen)
567	{
568	return index_name_stage_pos(istate, name, namelen, 0, EXPAND_SPARSE);
569	}
570
571	int index_name_pos_sparse(struct index_state istate, const char name, int namelen)
572	{
573	return index_name_stage_pos(istate, name, namelen, 0, NO_EXPAND_SPARSE);
574	}
575
576	int index_entry_exists(struct index_state istate, const char name, int namelen)
577	{
578	return index_name_stage_pos(istate, name, namelen, 0, NO_EXPAND_SPARSE) >= 0;
579	}
580
581	int remove_index_entry_at(struct index_state *istate, int pos)
582	{
583	struct cache_entry *ce = istate->cache[pos];
584
585	record_resolve_undo(istate, ce);
586	remove_name_hash(istate, ce);
587	save_or_free_index_entry(istate, ce);
588	istate->cache_changed \|= CE_ENTRY_REMOVED;
589	istate->cache_nr--;
590	if (pos >= istate->cache_nr)
591	return 0;
592	MOVE_ARRAY(istate->cache + pos, istate->cache + pos + 1,
593	istate->cache_nr - pos);
594	return 1;
595	}
596
597	/*
598	* Remove all cache entries marked for removal, that is where
599	* CE_REMOVE is set in ce_flags. This is much more effective than
600	* calling remove_index_entry_at() for each entry to be removed.
601	*/
602	void remove_marked_cache_entries(struct index_state *istate, int invalidate)
603	{
604	struct cache_entry **ce_array = istate->cache;
605	unsigned int i, j;
606
607	for (i = j = 0; i < istate->cache_nr; i++) {
608	if (ce_array[i]->ce_flags & CE_REMOVE) {
609	if (invalidate) {
610	cache_tree_invalidate_path(istate,
611	ce_array[i]->name);
612	untracked_cache_remove_from_index(istate,
613	ce_array[i]->name);
614	}
615	remove_name_hash(istate, ce_array[i]);
616	save_or_free_index_entry(istate, ce_array[i]);
617	}
618	else
619	ce_array[j++] = ce_array[i];
620	}
621	if (j == istate->cache_nr)
622	return;
623	istate->cache_changed \|= CE_ENTRY_REMOVED;
624	istate->cache_nr = j;
625	}
626
627	int remove_file_from_index(struct index_state istate, const char path)
628	{
629	int pos = index_name_pos(istate, path, strlen(path));
630	if (pos < 0)
631	pos = -pos-1;
632	cache_tree_invalidate_path(istate, path);
633	untracked_cache_remove_from_index(istate, path);
634	while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
635	remove_index_entry_at(istate, pos);
636	return 0;
637	}
638
639	static int compare_name(struct cache_entry ce, const char path, int namelen)
640	{
641	return namelen != ce_namelen(ce) \|\| memcmp(path, ce->name, namelen);
642	}
643
644	static int index_name_pos_also_unmerged(struct index_state *istate,
645	const char *path, int namelen)
646	{
647	int pos = index_name_pos(istate, path, namelen);
648	struct cache_entry *ce;
649
650	if (pos >= 0)
651	return pos;
652
653	/* maybe unmerged? */
654	pos = -1 - pos;
655	if (pos >= istate->cache_nr \|\|
656	compare_name((ce = istate->cache[pos]), path, namelen))
657	return -1;
658
659	/* order of preference: stage 2, 1, 3 */
660	if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
661	ce_stage((ce = istate->cache[pos + 1])) == 2 &&
662	!compare_name(ce, path, namelen))
663	pos++;
664	return pos;
665	}
666
667	static int different_name(struct cache_entry ce, struct cache_entry alias)
668	{
669	int len = ce_namelen(ce);
670	return ce_namelen(alias) != len \|\| memcmp(ce->name, alias->name, len);
671	}
672
673	/*
674	* If we add a filename that aliases in the cache, we will use the
675	* name that we already have - but we don't want to update the same
676	* alias twice, because that implies that there were actually two
677	* different files with aliasing names!
678	*
679	* So we use the CE_ADDED flag to verify that the alias was an old
680	* one before we accept it as
681	*/
682	static struct cache_entry create_alias_ce(struct index_state istate,
683	struct cache_entry *ce,
684	struct cache_entry *alias)
685	{
686	int len;
687	struct cache_entry *new_entry;
688
689	if (alias->ce_flags & CE_ADDED)
690	die(_("will not add file alias '%s' ('%s' already exists in index)"),
691	ce->name, alias->name);
692
693	/* Ok, create the new entry using the name of the existing alias */
694	len = ce_namelen(alias);
695	new_entry = make_empty_cache_entry(istate, len);
696	memcpy(new_entry->name, alias->name, len);
697	copy_cache_entry(new_entry, ce);
698	save_or_free_index_entry(istate, ce);
699	return new_entry;
700	}
701
702	void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
703	{
704	struct object_id oid;
705	if (odb_write_object(the_repository->objects, "", 0, OBJ_BLOB, &oid))
706	die(_("cannot create an empty blob in the object database"));
707	oidcpy(&ce->oid, &oid);
708	}
709
710	int add_to_index(struct index_state istate, const char path, struct stat *st, int flags)
711	{
712	int namelen, was_same;
713	mode_t st_mode = st->st_mode;
714	struct cache_entry ce, alias = NULL;
715	unsigned ce_option = CE_MATCH_IGNORE_VALID\|CE_MATCH_IGNORE_SKIP_WORKTREE\|CE_MATCH_RACY_IS_DIRTY;
716	int verbose = flags & (ADD_CACHE_VERBOSE \| ADD_CACHE_PRETEND);
717	int pretend = flags & ADD_CACHE_PRETEND;
718	int intent_only = flags & ADD_CACHE_INTENT;
719	int add_option = (ADD_CACHE_OK_TO_ADD\|ADD_CACHE_OK_TO_REPLACE\|
720	(intent_only ? ADD_CACHE_NEW_ONLY : 0));
721	unsigned hash_flags = pretend ? 0 : INDEX_WRITE_OBJECT;
722
723	if (flags & ADD_CACHE_RENORMALIZE)
724	hash_flags \|= INDEX_RENORMALIZE;
725
726	if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
727	return error(_("%s: can only add regular files, symbolic links or git-directories"), path);
728
729	namelen = strlen(path);
730	if (S_ISDIR(st_mode)) {
731	while (namelen && path[namelen-1] == '/')
732	namelen--;
733	}
734	ce = make_empty_cache_entry(istate, namelen);
735	memcpy(ce->name, path, namelen);
736	ce->ce_namelen = namelen;
737	if (!intent_only)
738	fill_stat_cache_info(istate, ce, st);
739	else
740	ce->ce_flags \|= CE_INTENT_TO_ADD;
741
742
743	if (repo_trust_executable_bit(istate->repo) &&
744	repo_has_symlinks(istate->repo)) {
745	ce->ce_mode = create_ce_mode(st_mode);
746	} else {
747	/* If there is an existing entry, pick the mode bits and type
748	* from it, otherwise assume unexecutable regular file.
749	*/
750	struct cache_entry *ent;
751	int pos = index_name_pos_also_unmerged(istate, path, namelen);
752
753	ent = (0 <= pos) ? istate->cache[pos] : NULL;
754	ce->ce_mode = ce_mode_from_stat(istate->repo, ent, st_mode);
755	}
756
757	/* When core.ignorecase=true, determine if a directory of the same name but differing
758	* case already exists within the Git repository. If it does, ensure the directory
759	* case of the file being added to the repository matches (is folded into) the existing
760	* entry's directory case.
761	*/
762	if (repo_ignore_case(the_repository)) {
763	adjust_dirname_case(istate, ce->name);
764	}
765	if (!(flags & ADD_CACHE_RENORMALIZE)) {
766	alias = index_file_exists(istate, ce->name,
767	ce_namelen(ce), repo_ignore_case(the_repository));
768	if (alias &&
769	!ce_stage(alias) &&
770	!ie_match_stat(istate, alias, st, ce_option)) {
771	/* Nothing changed, really */
772	if (!S_ISGITLINK(alias->ce_mode))
773	ce_mark_uptodate(alias);
774	alias->ce_flags \|= CE_ADDED;
775
776	discard_cache_entry(ce);
777	return 0;
778	}
779	}
780	if (!intent_only) {
781	if (index_path(istate, &ce->oid, path, st, hash_flags)) {
782	discard_cache_entry(ce);
783	return error(_("unable to index file '%s'"), path);
784	}
785	} else
786	set_object_name_for_intent_to_add_entry(ce);
787
788	if (repo_ignore_case(the_repository) && alias && different_name(ce, alias))
789	ce = create_alias_ce(istate, ce, alias);
790	ce->ce_flags \|= CE_ADDED;
791
792	/* It was suspected to be racily clean, but it turns out to be Ok */
793	was_same = (alias &&
794	!ce_stage(alias) &&
795	oideq(&alias->oid, &ce->oid) &&
796	ce->ce_mode == alias->ce_mode);
797
798	if (pretend)
799	discard_cache_entry(ce);
800	else if (add_index_entry(istate, ce, add_option)) {
801	discard_cache_entry(ce);
802	return error(_("unable to add '%s' to index"), path);
803	}
804	if (verbose && !was_same)
805	printf("add '%s'\n", path);
806	return 0;
807	}
808
809	int add_file_to_index(struct index_state istate, const char path, int flags)
810	{
811	struct stat st;
812	if (lstat(path, &st))
813	die_errno(_("unable to stat '%s'"), path);
814	return add_to_index(istate, path, &st, flags);
815	}
816
817	struct cache_entry make_empty_cache_entry(struct index_state istate, size_t len)
818	{
819	return mem_pool__ce_calloc(find_mem_pool(istate), len);
820	}
821
822	struct cache_entry *make_empty_transient_cache_entry(size_t len,
823	struct mem_pool *ce_mem_pool)
824	{
825	if (ce_mem_pool)
826	return mem_pool__ce_calloc(ce_mem_pool, len);
827	return xcalloc(1, cache_entry_size(len));
828	}
829
830	enum verify_path_result {
831	PATH_OK,
832	PATH_INVALID,
833	PATH_DIR_WITH_SEP,
834	};
835
836	static enum verify_path_result verify_path_internal(const char *, unsigned);
837
838	int verify_path(const char *path, unsigned mode)
839	{
840	return verify_path_internal(path, mode) == PATH_OK;
841	}
842
843	struct cache_entry make_cache_entry(struct index_state istate,
844	unsigned int mode,
845	const struct object_id *oid,
846	const char *path,
847	int stage,
848	unsigned int refresh_options)
849	{
850	struct cache_entry ce, ret;
851	int len;
852
853	if (verify_path_internal(path, mode) == PATH_INVALID) {
854	error(_("invalid path '%s'"), path);
855	return NULL;
856	}
857
858	len = strlen(path);
859	ce = make_empty_cache_entry(istate, len);
860
861	oidcpy(&ce->oid, oid);
862	memcpy(ce->name, path, len);
863	ce->ce_flags = create_ce_flags(stage);
864	ce->ce_namelen = len;
865	ce->ce_mode = create_ce_mode(mode);
866
867	ret = refresh_cache_entry(istate, ce, refresh_options);
868	if (ret != ce)
869	discard_cache_entry(ce);
870	return ret;
871	}
872
873	struct cache_entry *make_transient_cache_entry(unsigned int mode,
874	const struct object_id *oid,
875	const char *path,
876	int stage,
877	struct mem_pool *ce_mem_pool)
878	{
879	struct cache_entry *ce;
880	int len;
881
882	if (!verify_path(path, mode)) {
883	error(_("invalid path '%s'"), path);
884	return NULL;
885	}
886
887	len = strlen(path);
888	ce = make_empty_transient_cache_entry(len, ce_mem_pool);
889
890	oidcpy(&ce->oid, oid);
891	memcpy(ce->name, path, len);
892	ce->ce_flags = create_ce_flags(stage);
893	ce->ce_namelen = len;
894	ce->ce_mode = create_ce_mode(mode);
895
896	return ce;
897	}
898
899	/*
900	* Chmod an index entry with either +x or -x.
901	*
902	* Returns -1 if the chmod for the particular cache entry failed (if it's
903	* not a regular file), -2 if an invalid flip argument is passed in, 0
904	* otherwise.
905	*/
906	int chmod_index_entry(struct index_state istate, struct cache_entry ce,
907	char flip)
908	{
909	if (!S_ISREG(ce->ce_mode))
910	return -1;
911	switch (flip) {
912	case '+':
913	ce->ce_mode \|= 0111;
914	break;
915	case '-':
916	ce->ce_mode &= ~0111;
917	break;
918	default:
919	return -2;
920	}
921	cache_tree_invalidate_path(istate, ce->name);
922	ce->ce_flags \|= CE_UPDATE_IN_BASE;
923	mark_fsmonitor_invalid(istate, ce);
924	istate->cache_changed \|= CE_ENTRY_CHANGED;
925
926	return 0;
927	}
928
929	int ce_same_name(const struct cache_entry a, const struct cache_entry b)
930	{
931	int len = ce_namelen(a);
932	return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
933	}
934
935	/*
936	* We fundamentally don't like some paths: we don't want
937	* dot or dot-dot anywhere, and for obvious reasons don't
938	* want to recurse into ".git" either.
939	*
940	* Also, we don't want double slashes or slashes at the
941	* end that can make pathnames ambiguous.
942	*/
943	static int verify_dotfile(const char *rest, unsigned mode)
944	{
945	/*
946	* The first character was '.', but that
947	* has already been discarded, we now test
948	* the rest.
949	*/
950
951	/* "." is not allowed */
952	if (rest == '\0' \|\| is_dir_sep(rest))
953	return 0;
954
955	switch (*rest) {
956	/*
957	* ".git" followed by NUL or slash is bad. Note that we match
958	* case-insensitively here, even if ignore_case is not set.
959	* This outlaws ".GIT" everywhere out of an abundance of caution,
960	* since there's really no good reason to allow it.
961	*
962	* Once we've seen ".git", we can also find ".gitmodules", etc (also
963	* case-insensitively).
964	*/
965	case 'g':
966	case 'G':
967	if (rest[1] != 'i' && rest[1] != 'I')
968	break;
969	if (rest[2] != 't' && rest[2] != 'T')
970	break;
971	if (rest[3] == '\0' \|\| is_dir_sep(rest[3]))
972	return 0;
973	if (S_ISLNK(mode)) {
974	rest += 3;
975	if (skip_iprefix(rest, "modules", &rest) &&
976	(rest == '\0' \|\| is_dir_sep(rest)))
977	return 0;
978	}
979	break;
980	case '.':
981	if (rest[1] == '\0' \|\| is_dir_sep(rest[1]))
982	return 0;
983	}
984	return 1;
985	}
986
987	static enum verify_path_result verify_path_internal(const char *path,
988	unsigned mode)
989	{
990	char c = 0;
991
992	if (has_dos_drive_prefix(path))
993	return PATH_INVALID;
994
995	if (!is_valid_path(path))
996	return PATH_INVALID;
997
998	goto inside;
999	for (;;) {
1000	if (!c)
1001	return PATH_OK;
1002	if (is_dir_sep(c)) {
1003	inside:
1004	if (repo_protect_hfs(the_repository)) {
1005
1006	if (is_hfs_dotgit(path))
1007	return PATH_INVALID;
1008	if (S_ISLNK(mode)) {
1009	if (is_hfs_dotgitmodules(path))
1010	return PATH_INVALID;
1011	}
1012	}
1013	if (repo_protect_ntfs(the_repository)) {
1014	#if defined GIT_WINDOWS_NATIVE \|\| defined __CYGWIN__
1015	if (c == '\\')
1016	return PATH_INVALID;
1017	#endif
1018	if (is_ntfs_dotgit(path))
1019	return PATH_INVALID;
1020	if (S_ISLNK(mode)) {
1021	if (is_ntfs_dotgitmodules(path))
1022	return PATH_INVALID;
1023	}
1024	}
1025
1026	c = *path++;
1027	if ((c == '.' && !verify_dotfile(path, mode)) \|\|
1028	is_dir_sep(c))
1029	return PATH_INVALID;
1030	/*
1031	* allow terminating directory separators for
1032	* sparse directory entries.
1033	*/
1034	if (c == '\0')
1035	return S_ISDIR(mode) ? PATH_DIR_WITH_SEP :
1036	PATH_INVALID;
1037	} else if (c == '\\' &&
1038	repo_protect_ntfs(the_repository)) {
1039	if (is_ntfs_dotgit(path))
1040	return PATH_INVALID;
1041	if (S_ISLNK(mode)) {
1042	if (is_ntfs_dotgitmodules(path))
1043	return PATH_INVALID;
1044	}
1045	}
1046
1047	c = *path++;
1048	}
1049	}
1050
1051	/*
1052	* Do we have another file that has the beginning components being a
1053	* proper superset of the name we're trying to add?
1054	*/
1055	static int has_file_name(struct index_state *istate,
1056	const struct cache_entry *ce, int pos, int ok_to_replace)
1057	{
1058	int retval = 0;
1059	int len = ce_namelen(ce);
1060	int stage = ce_stage(ce);
1061	const char *name = ce->name;
1062
1063	while (pos < istate->cache_nr) {
1064	struct cache_entry *p = istate->cache[pos++];
1065
1066	if (len >= ce_namelen(p))
1067	break;
1068	if (memcmp(name, p->name, len))
1069	break;
1070	if (ce_stage(p) != stage)
1071	continue;
1072	if (p->name[len] != '/')
1073	continue;
1074	if (p->ce_flags & CE_REMOVE)
1075	continue;
1076	retval = -1;
1077	if (!ok_to_replace)
1078	break;
1079	remove_index_entry_at(istate, --pos);
1080	}
1081	return retval;
1082	}
1083
1084
1085	/*
1086	* Like strcmp(), but also return the offset of the first change.
1087	* If strings are equal, return the length.
1088	*/
1089	int strcmp_offset(const char s1, const char s2, size_t *first_change)
1090	{
1091	size_t k;
1092
1093	if (!first_change)
1094	return strcmp(s1, s2);
1095
1096	for (k = 0; s1[k] == s2[k]; k++)
1097	if (s1[k] == '\0')
1098	break;
1099
1100	*first_change = k;
1101	return (unsigned char)s1[k] - (unsigned char)s2[k];
1102	}
1103
1104	/*
1105	* Do we have another file with a pathname that is a proper
1106	* subset of the name we're trying to add?
1107	*
1108	* That is, is there another file in the index with a path
1109	* that matches a sub-directory in the given entry?
1110	*/
1111	static int has_dir_name(struct index_state *istate,
1112	const struct cache_entry *ce, int pos, int ok_to_replace)
1113	{
1114	int retval = 0;
1115	int stage = ce_stage(ce);
1116	const char *name = ce->name;
1117	const char *slash = name + ce_namelen(ce);
1118	size_t len_eq_last;
1119	int cmp_last = 0;
1120
1121	/*
1122	* We are frequently called during an iteration on a sorted
1123	* list of pathnames and while building a new index. Therefore,
1124	* there is a high probability that this entry will eventually
1125	* be appended to the index, rather than inserted in the middle.
1126	* If we can confirm that, we can avoid binary searches on the
1127	* components of the pathname.
1128	*
1129	* Compare the entry's full path with the last path in the index.
1130	*/
1131	if (!istate->cache_nr)
1132	return 0;
1133
1134	cmp_last = strcmp_offset(name,
1135	istate->cache[istate->cache_nr - 1]->name,
1136	&len_eq_last);
1137	if (cmp_last > 0 && name[len_eq_last] != '/')
1138	/*
1139	* The entry sorts AFTER the last one in the
1140	* index and their paths have no common prefix,
1141	* so there cannot be a F/D conflict.
1142	*/
1143	return 0;
1144
1145	for (;;) {
1146	size_t len;
1147
1148	for (;;) {
1149	if (*--slash == '/')
1150	break;
1151	if (slash <= ce->name)
1152	return retval;
1153	}
1154	len = slash - name;
1155
1156	pos = index_name_stage_pos(istate, name, len, stage, EXPAND_SPARSE);
1157	if (pos >= 0) {
1158	/*
1159	* Found one, but not so fast. This could
1160	* be a marker that says "I was here, but
1161	* I am being removed". Such an entry is
1162	* not a part of the resulting tree, and
1163	* it is Ok to have a directory at the same
1164	* path.
1165	*/
1166	if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
1167	retval = -1;
1168	if (!ok_to_replace)
1169	break;
1170	remove_index_entry_at(istate, pos);
1171	continue;
1172	}
1173	}
1174	else
1175	pos = -pos-1;
1176
1177	/*
1178	* Trivial optimization: if we find an entry that
1179	* already matches the sub-directory, then we know
1180	* we're ok, and we can exit.
1181	*/
1182	while (pos < istate->cache_nr) {
1183	struct cache_entry *p = istate->cache[pos];
1184	if ((ce_namelen(p) <= len) \|\|
1185	(p->name[len] != '/') \|\|
1186	memcmp(p->name, name, len))
1187	break; /* not our subdirectory */
1188	if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
1189	/*
1190	* p is at the same stage as our entry, and
1191	* is a subdirectory of what we are looking
1192	* at, so we cannot have conflicts at our
1193	* level or anything shorter.
1194	*/
1195	return retval;
1196	pos++;
1197	}
1198	}
1199	return retval;
1200	}
1201
1202	/* We may be in a situation where we already have path/file and path
1203	* is being added, or we already have path and path/file is being
1204	* added. Either one would result in a nonsense tree that has path
1205	* twice when git-write-tree tries to write it out. Prevent it.
1206	*
1207	* If ok-to-replace is specified, we remove the conflicting entries
1208	* from the cache so the caller should recompute the insert position.
1209	* When this happens, we return non-zero.
1210	*/
1211	static int check_file_directory_conflict(struct index_state *istate,
1212	const struct cache_entry *ce,
1213	int pos, int ok_to_replace)
1214	{
1215	int retval;
1216
1217	/*
1218	* When ce is an "I am going away" entry, we allow it to be added
1219	*/
1220	if (ce->ce_flags & CE_REMOVE)
1221	return 0;
1222
1223	/*
1224	* We check if the path is a sub-path of a subsequent pathname
1225	* first, since removing those will not change the position
1226	* in the array.
1227	*/
1228	retval = has_file_name(istate, ce, pos, ok_to_replace);
1229
1230	/*
1231	* Then check if the path might have a clashing sub-directory
1232	* before it.
1233	*/
1234	return retval + has_dir_name(istate, ce, pos, ok_to_replace);
1235	}
1236
1237	static int add_index_entry_with_check(struct index_state istate, struct cache_entry ce, int option)
1238	{
1239	int pos;
1240	int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
1241	int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
1242	int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
1243	int new_only = option & ADD_CACHE_NEW_ONLY;
1244
1245	/*
1246	* If this entry's path sorts after the last entry in the index,
1247	* we can avoid searching for it.
1248	*/
1249	if (istate->cache_nr > 0 &&
1250	strcmp(ce->name, istate->cache[istate->cache_nr - 1]->name) > 0)
1251	pos = index_pos_to_insert_pos(istate->cache_nr);
1252	else
1253	pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce), EXPAND_SPARSE);
1254
1255	/*
1256	* Cache tree path should be invalidated only after index_name_stage_pos,
1257	* in case it expands a sparse index.
1258	*/
1259	if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1260	cache_tree_invalidate_path(istate, ce->name);
1261
1262	/* existing match? Just replace it. */
1263	if (pos >= 0) {
1264	if (!new_only)
1265	replace_index_entry(istate, pos, ce);
1266	return 0;
1267	}
1268	pos = -pos-1;
1269
1270	if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1271	untracked_cache_add_to_index(istate, ce->name);
1272
1273	/*
1274	* Inserting a merged entry ("stage 0") into the index
1275	* will always replace all non-merged entries..
1276	*/
1277	if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1278	while (ce_same_name(istate->cache[pos], ce)) {
1279	ok_to_add = 1;
1280	if (!remove_index_entry_at(istate, pos))
1281	break;
1282	}
1283	}
1284
1285	if (!ok_to_add)
1286	return -1;
1287	if (verify_path_internal(ce->name, ce->ce_mode) == PATH_INVALID)
1288	return error(_("invalid path '%s'"), ce->name);
1289
1290	if (!skip_df_check &&
1291	check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
1292	if (!ok_to_replace)
1293	return error(_("'%s' appears as both a file and as a directory"),
1294	ce->name);
1295	pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce), EXPAND_SPARSE);
1296	pos = -pos-1;
1297	}
1298	return pos + 1;
1299	}
1300
1301	int add_index_entry(struct index_state istate, struct cache_entry ce, int option)
1302	{
1303	int pos;
1304
1305	if (option & ADD_CACHE_JUST_APPEND)
1306	pos = istate->cache_nr;
1307	else {
1308	int ret;
1309	ret = add_index_entry_with_check(istate, ce, option);
1310	if (ret <= 0)
1311	return ret;
1312	pos = ret - 1;
1313	}
1314
1315	/* Make sure the array is big enough .. */
1316	ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1317
1318	/* Add it in.. */
1319	istate->cache_nr++;
1320	if (istate->cache_nr > pos + 1)
1321	MOVE_ARRAY(istate->cache + pos + 1, istate->cache + pos,
1322	istate->cache_nr - pos - 1);
1323	set_index_entry(istate, pos, ce);
1324	istate->cache_changed \|= CE_ENTRY_ADDED;
1325	return 0;
1326	}
1327
1328	/*
1329	* "refresh" does not calculate a new sha1 file or bring the
1330	* cache up-to-date for mode/content changes. But what it
1331	* _does_ do is to "re-match" the stat information of a file
1332	* with the cache, so that you can refresh the cache for a
1333	* file that hasn't been changed but where the stat entry is
1334	* out of date.
1335	*
1336	* For example, you'd want to do this after doing a "git-read-tree",
1337	* to link up the stat cache details with the proper files.
1338	*/
1339	static struct cache_entry refresh_cache_ent(struct index_state istate,
1340	struct cache_entry *ce,
1341	unsigned int options, int *err,
1342	int *changed_ret,
1343	int *t2_did_lstat,
1344	int *t2_did_scan)
1345	{
1346	struct stat st;
1347	struct cache_entry *updated;
1348	int changed;
1349	int refresh = options & CE_MATCH_REFRESH;
1350	int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1351	int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1352	int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1353	int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
1354
1355	if (!refresh \|\| ce_uptodate(ce))
1356	return ce;
1357
1358	if (!ignore_fsmonitor)
1359	refresh_fsmonitor(istate);
1360	/*
1361	* CE_VALID or CE_SKIP_WORKTREE means the user promised us
1362	* that the change to the work tree does not matter and told
1363	* us not to worry.
1364	*/
1365	if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1366	ce_mark_uptodate(ce);
1367	return ce;
1368	}
1369	if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1370	ce_mark_uptodate(ce);
1371	return ce;
1372	}
1373	if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID)) {
1374	ce_mark_uptodate(ce);
1375	return ce;
1376	}
1377
1378	if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1379	if (ignore_missing)
1380	return ce;
1381	if (err)
1382	*err = ENOENT;
1383	return NULL;
1384	}
1385
1386	if (t2_did_lstat)
1387	*t2_did_lstat = 1;
1388	if (lstat(ce->name, &st) < 0) {
1389	if (ignore_missing && errno == ENOENT)
1390	return ce;
1391	if (err)
1392	*err = errno;
1393	return NULL;
1394	}
1395
1396	changed = ie_match_stat(istate, ce, &st, options);
1397	if (changed_ret)
1398	*changed_ret = changed;
1399	if (!changed) {
1400	/*
1401	* The path is unchanged. If we were told to ignore
1402	* valid bit, then we did the actual stat check and
1403	* found that the entry is unmodified. If the entry
1404	* is not marked VALID, this is the place to mark it
1405	* valid again, under "assume unchanged" mode.
1406	*/
1407	if (ignore_valid && assume_unchanged &&
1408	!(ce->ce_flags & CE_VALID))
1409	; /* mark this one VALID again */
1410	else {
1411	/*
1412	* We do not mark the index itself "modified"
1413	* because CE_UPTODATE flag is in-core only;
1414	* we are not going to write this change out.
1415	*/
1416	if (!S_ISGITLINK(ce->ce_mode)) {
1417	ce_mark_uptodate(ce);
1418	mark_fsmonitor_valid(istate, ce);
1419	}
1420	return ce;
1421	}
1422	}
1423
1424	if (t2_did_scan)
1425	*t2_did_scan = 1;
1426	if (ie_modified(istate, ce, &st, options)) {
1427	if (err)
1428	*err = EINVAL;
1429	return NULL;
1430	}
1431
1432	updated = make_empty_cache_entry(istate, ce_namelen(ce));
1433	copy_cache_entry(updated, ce);
1434	memcpy(updated->name, ce->name, ce->ce_namelen + 1);
1435	fill_stat_cache_info(istate, updated, &st);
1436	/*
1437	* If ignore_valid is not set, we should leave CE_VALID bit
1438	* alone. Otherwise, paths marked with --no-assume-unchanged
1439	* (i.e. things to be edited) will reacquire CE_VALID bit
1440	* automatically, which is not really what we want.
1441	*/
1442	if (!ignore_valid && assume_unchanged &&
1443	!(ce->ce_flags & CE_VALID))
1444	updated->ce_flags &= ~CE_VALID;
1445
1446	/* istate->cache_changed is updated in the caller */
1447	return updated;
1448	}
1449
1450	static void show_file(const char * fmt, const char * name, int in_porcelain,
1451	int * first, const char *header_msg)
1452	{
1453	if (in_porcelain && *first && header_msg) {
1454	printf("%s\n", header_msg);
1455	*first = 0;
1456	}
1457	printf(fmt, name);
1458	}
1459
1460	int repo_refresh_and_write_index(struct repository *repo,
1461	unsigned int refresh_flags,
1462	unsigned int write_flags,
1463	int gentle,
1464	const struct pathspec *pathspec,
1465	char seen, const char header_msg)
1466	{
1467	struct lock_file lock_file = LOCK_INIT;
1468	int fd, ret = 0;
1469
1470	fd = repo_hold_locked_index(repo, &lock_file,
1471	gentle ? 0 : LOCK_REPORT_ON_ERROR);
1472	if (!gentle && fd < 0)
1473	return -1;
1474	if (refresh_index(repo->index, refresh_flags, pathspec, seen, header_msg))
1475	ret = 1;
1476	if (0 <= fd && write_locked_index(repo->index, &lock_file, COMMIT_LOCK \| write_flags))
1477	ret = -1;
1478	return ret;
1479	}
1480
1481
1482	int refresh_index(struct index_state *istate, unsigned int flags,
1483	const struct pathspec *pathspec,
1484	char seen, const char header_msg)
1485	{
1486	int i;
1487	int has_errors = 0;
1488	int really = (flags & REFRESH_REALLY) != 0;
1489	int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1490	int quiet = (flags & REFRESH_QUIET) != 0;
1491	int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1492	int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1493	int ignore_skip_worktree = (flags & REFRESH_IGNORE_SKIP_WORKTREE) != 0;
1494	int first = 1;
1495	int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1496	unsigned int options = (CE_MATCH_REFRESH \|
1497	(really ? CE_MATCH_IGNORE_VALID : 0) \|
1498	(not_new ? CE_MATCH_IGNORE_MISSING : 0));
1499	const char *modified_fmt;
1500	const char *deleted_fmt;
1501	const char *typechange_fmt;
1502	const char *added_fmt;
1503	const char *unmerged_fmt;
1504	struct progress *progress = NULL;
1505	int t2_sum_lstat = 0;
1506	int t2_sum_scan = 0;
1507
1508	if (flags & REFRESH_PROGRESS && isatty(2))
1509	progress = start_delayed_progress(the_repository,
1510	_("Refresh index"),
1511	istate->cache_nr);
1512
1513	trace_performance_enter();
1514	modified_fmt = in_porcelain ? "M\t%s\n" : "%s: needs update\n";
1515	deleted_fmt = in_porcelain ? "D\t%s\n" : "%s: needs update\n";
1516	typechange_fmt = in_porcelain ? "T\t%s\n" : "%s: needs update\n";
1517	added_fmt = in_porcelain ? "A\t%s\n" : "%s: needs update\n";
1518	unmerged_fmt = in_porcelain ? "U\t%s\n" : "%s: needs merge\n";
1519	/*
1520	* Use the multi-threaded preload_index() to refresh most of the
1521	* cache entries quickly then in the single threaded loop below,
1522	* we only have to do the special cases that are left.
1523	*/
1524	preload_index(istate, pathspec, 0);
1525	trace2_region_enter("index", "refresh", NULL);
1526
1527	for (i = 0; i < istate->cache_nr; i++) {
1528	struct cache_entry ce, new_entry;
1529	int cache_errno = 0;
1530	int changed = 0;
1531	int filtered = 0;
1532	int t2_did_lstat = 0;
1533	int t2_did_scan = 0;
1534
1535	ce = istate->cache[i];
1536	if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1537	continue;
1538	if (ignore_skip_worktree && ce_skip_worktree(ce))
1539	continue;
1540
1541	/*
1542	* If this entry is a sparse directory, then there isn't
1543	* any stat() information to update. Ignore the entry.
1544	*/
1545	if (S_ISSPARSEDIR(ce->ce_mode))
1546	continue;
1547
1548	if (pathspec && !ce_path_match(istate, ce, pathspec, seen))
1549	filtered = 1;
1550
1551	if (ce_stage(ce)) {
1552	while ((i < istate->cache_nr) &&
1553	! strcmp(istate->cache[i]->name, ce->name))
1554	i++;
1555	i--;
1556	if (allow_unmerged)
1557	continue;
1558	if (!filtered)
1559	show_file(unmerged_fmt, ce->name, in_porcelain,
1560	&first, header_msg);
1561	has_errors = 1;
1562	continue;
1563	}
1564
1565	if (filtered)
1566	continue;
1567
1568	new_entry = refresh_cache_ent(istate, ce, options,
1569	&cache_errno, &changed,
1570	&t2_did_lstat, &t2_did_scan);
1571	t2_sum_lstat += t2_did_lstat;
1572	t2_sum_scan += t2_did_scan;
1573	if (new_entry == ce)
1574	continue;
1575	display_progress(progress, i);
1576	if (!new_entry) {
1577	const char *fmt;
1578
1579	if (really && cache_errno == EINVAL) {
1580	/* If we are doing --really-refresh that
1581	* means the index is not valid anymore.
1582	*/
1583	ce->ce_flags &= ~CE_VALID;
1584	ce->ce_flags \|= CE_UPDATE_IN_BASE;
1585	mark_fsmonitor_invalid(istate, ce);
1586	istate->cache_changed \|= CE_ENTRY_CHANGED;
1587	}
1588	if (quiet)
1589	continue;
1590
1591	if (cache_errno == ENOENT)
1592	fmt = deleted_fmt;
1593	else if (ce_intent_to_add(ce))
1594	fmt = added_fmt; /* must be before other checks */
1595	else if (changed & TYPE_CHANGED)
1596	fmt = typechange_fmt;
1597	else
1598	fmt = modified_fmt;
1599	show_file(fmt,
1600	ce->name, in_porcelain, &first, header_msg);
1601	has_errors = 1;
1602	continue;
1603	}
1604
1605	replace_index_entry(istate, i, new_entry);
1606	}
1607	trace2_data_intmax("index", NULL, "refresh/sum_lstat", t2_sum_lstat);
1608	trace2_data_intmax("index", NULL, "refresh/sum_scan", t2_sum_scan);
1609	trace2_region_leave("index", "refresh", NULL);
1610	display_progress(progress, istate->cache_nr);
1611	stop_progress(&progress);
1612	trace_performance_leave("refresh index");
1613	return has_errors;
1614	}
1615
1616	struct cache_entry refresh_cache_entry(struct index_state istate,
1617	struct cache_entry *ce,
1618	unsigned int options)
1619	{
1620	return refresh_cache_ent(istate, ce, options, NULL, NULL, NULL, NULL);
1621	}
1622
1623
1624	/*****************************************************************
1625	* Index File I/O
1626	*****************************************************************/
1627
1628	#define INDEX_FORMAT_DEFAULT 3
1629
1630	static unsigned int get_index_format_default(struct repository *r)
1631	{
1632	char *envversion = getenv("GIT_INDEX_VERSION");
1633	char *endp;
1634	unsigned int version = INDEX_FORMAT_DEFAULT;
1635
1636	if (!envversion) {
1637	prepare_repo_settings(r);
1638
1639	if (r->settings.index_version >= 0)
1640	version = r->settings.index_version;
1641	if (version < INDEX_FORMAT_LB \|\| INDEX_FORMAT_UB < version) {
1642	warning(_("index.version set, but the value is invalid.\n"
1643	"Using version %i"), INDEX_FORMAT_DEFAULT);
1644	return INDEX_FORMAT_DEFAULT;
1645	}
1646	return version;
1647	}
1648
1649	version = strtoul(envversion, &endp, 10);
1650	if (*endp \|\|
1651	version < INDEX_FORMAT_LB \|\| INDEX_FORMAT_UB < version) {
1652	warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1653	"Using version %i"), INDEX_FORMAT_DEFAULT);
1654	version = INDEX_FORMAT_DEFAULT;
1655	}
1656	return version;
1657	}
1658
1659	/*
1660	* dev/ino/uid/gid/size are also just tracked to the low 32 bits
1661	* Again - this is just a (very strong in practice) heuristic that
1662	* the inode hasn't changed.
1663	*
1664	* We save the fields in big-endian order to allow using the
1665	* index file over NFS transparently.
1666	*/
1667	struct ondisk_cache_entry {
1668	struct cache_time ctime;
1669	struct cache_time mtime;
1670	uint32_t dev;
1671	uint32_t ino;
1672	uint32_t mode;
1673	uint32_t uid;
1674	uint32_t gid;
1675	uint32_t size;
1676	/*
1677	* unsigned char hash[hashsz];
1678	* uint16_t flags;
1679	* if (flags & CE_EXTENDED)
1680	* uint16_t flags2;
1681	*/
1682	unsigned char data[GIT_MAX_RAWSZ + 2 * sizeof(uint16_t)];
1683	char name[FLEX_ARRAY];
1684	};
1685
1686	/* These are only used for v3 or lower */
1687	#define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1688	#define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,data) + (len) + 8) & ~7)
1689	#define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1690	#define ondisk_data_size(flags, len) (the_hash_algo->rawsz + \
1691	((flags & CE_EXTENDED) ? 2 : 1) * sizeof(uint16_t) + len)
1692	#define ondisk_data_size_max(len) (ondisk_data_size(CE_EXTENDED, len))
1693	#define ondisk_ce_size(ce) (ondisk_cache_entry_size(ondisk_data_size((ce)->ce_flags, ce_namelen(ce))))
1694
1695	/* Allow fsck to force verification of the index checksum. */
1696	int verify_index_checksum;
1697
1698	/* Allow fsck to force verification of the cache entry order. */
1699	int verify_ce_order;
1700
1701	static int verify_hdr(const struct cache_header *hdr, unsigned long size)
1702	{
1703	struct git_hash_ctx c;
1704	unsigned char hash[GIT_MAX_RAWSZ];
1705	int hdr_version;
1706	unsigned char start, end;
1707	struct object_id oid;
1708
1709	if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1710	return error(_("bad signature 0x%08x"), hdr->hdr_signature);
1711	hdr_version = ntohl(hdr->hdr_version);
1712	if (hdr_version < INDEX_FORMAT_LB \|\| INDEX_FORMAT_UB < hdr_version)
1713	return error(_("bad index version %d"), hdr_version);
1714
1715	if (!verify_index_checksum)
1716	return 0;
1717
1718	end = (unsigned char *)hdr + size;
1719	start = end - the_hash_algo->rawsz;
1720	oidread(&oid, start, the_repository->hash_algo);
1721	if (oideq(&oid, null_oid(the_hash_algo)))
1722	return 0;
1723
1724	git_hash_init(&c, the_hash_algo);
1725	git_hash_update(&c, hdr, size - the_hash_algo->rawsz);
1726	git_hash_final(hash, &c);
1727	if (!hasheq(hash, start, the_repository->hash_algo))
1728	return error(_("bad index file sha1 signature"));
1729	return 0;
1730	}
1731
1732	static int read_index_extension(struct index_state *istate,
1733	const char ext, const char data, unsigned long sz)
1734	{
1735	switch (CACHE_EXT(ext)) {
1736	case CACHE_EXT_TREE:
1737	istate->cache_tree = cache_tree_read(data, sz);
1738	break;
1739	case CACHE_EXT_RESOLVE_UNDO:
1740	istate->resolve_undo = resolve_undo_read(data, sz, the_hash_algo);
1741	break;
1742	case CACHE_EXT_LINK:
1743	if (read_link_extension(istate, data, sz))
1744	return -1;
1745	break;
1746	case CACHE_EXT_UNTRACKED:
1747	istate->untracked = read_untracked_extension(data, sz);
1748	break;
1749	case CACHE_EXT_FSMONITOR:
1750	read_fsmonitor_extension(istate, data, sz);
1751	break;
1752	case CACHE_EXT_ENDOFINDEXENTRIES:
1753	case CACHE_EXT_INDEXENTRYOFFSETTABLE:
1754	/* already handled in do_read_index() */
1755	break;
1756	case CACHE_EXT_SPARSE_DIRECTORIES:
1757	/* no content, only an indicator */
1758	istate->sparse_index = INDEX_COLLAPSED;
1759	break;
1760	default:
1761	if (ext < 'A' \|\| 'Z' < ext)
1762	return error(_("index uses %.4s extension, which we do not understand"),
1763	ext);
1764	fprintf_ln(stderr, _("ignoring %.4s extension"), ext);
1765	break;
1766	}
1767	return 0;
1768	}
1769
1770	/*
1771	* Parses the contents of the cache entry contained within the 'ondisk' buffer
1772	* into a new incore 'cache_entry'.
1773	*
1774	* Note that 'char *ondisk' may not be aligned to a 4-byte address interval in
1775	* index v4, so we cannot cast it to 'struct ondisk_cache_entry *' and access
1776	* its members. Instead, we use the byte offsets of members within the struct to
1777	* identify where 'get_be16()', 'get_be32()', and 'oidread()' (which can all
1778	* read from an unaligned memory buffer) should read from the 'ondisk' buffer
1779	* into the corresponding incore 'cache_entry' members.
1780	*/
1781	static struct cache_entry create_from_disk(struct mem_pool ce_mem_pool,
1782	unsigned int version,
1783	const char *ondisk,
1784	unsigned long *ent_size,
1785	const struct cache_entry *previous_ce)
1786	{
1787	struct cache_entry *ce;
1788	size_t len;
1789	const char *name;
1790	const unsigned hashsz = the_hash_algo->rawsz;
1791	const char *flagsp = ondisk + offsetof(struct ondisk_cache_entry, data) + hashsz;
1792	unsigned int flags;
1793	size_t copy_len = 0;
1794	/*
1795	* Adjacent cache entries tend to share the leading paths, so it makes
1796	* sense to only store the differences in later entries. In the v4
1797	* on-disk format of the index, each on-disk cache entry stores the
1798	* number of bytes to be stripped from the end of the previous name,
1799	* and the bytes to append to the result, to come up with its name.
1800	*/
1801	int expand_name_field = version == 4;
1802
1803	/* On-disk flags are just 16 bits */
1804	flags = get_be16(flagsp);
1805	len = flags & CE_NAMEMASK;
1806
1807	if (flags & CE_EXTENDED) {
1808	int extended_flags;
1809	extended_flags = get_be16(flagsp + sizeof(uint16_t)) << 16;
1810	/* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1811	if (extended_flags & ~CE_EXTENDED_FLAGS)
1812	die(_("unknown index entry format 0x%08x"), extended_flags);
1813	flags \|= extended_flags;
1814	name = (const char )(flagsp + 2 sizeof(uint16_t));
1815	}
1816	else
1817	name = (const char *)(flagsp + sizeof(uint16_t));
1818
1819	if (expand_name_field) {
1820	const unsigned char cp = (const unsigned char )name;
1821	uint64_t strip_len, previous_len;
1822
1823	/* If we're at the beginning of a block, ignore the previous name */
1824	strip_len = decode_varint(&cp);
1825	if (previous_ce) {
1826	previous_len = previous_ce->ce_namelen;
1827	if (previous_len < strip_len)
1828	die(_("malformed name field in the index, near path '%s'"),
1829	previous_ce->name);
1830	copy_len = previous_len - strip_len;
1831	}
1832	name = (const char *)cp;
1833	}
1834
1835	if (len == CE_NAMEMASK) {
1836	len = strlen(name);
1837	if (expand_name_field)
1838	len += copy_len;
1839	}
1840
1841	ce = mem_pool__ce_alloc(ce_mem_pool, len);
1842
1843	/*
1844	* NEEDSWORK: using 'offsetof()' is cumbersome and should be replaced
1845	* with something more akin to 'load_bitmap_entries_v1()'s use of
1846	* 'read_be16'/'read_be32'. For consistency with the corresponding
1847	* ondisk entry write function ('copy_cache_entry_to_ondisk()'), this
1848	* should be done at the same time as removing references to
1849	* 'ondisk_cache_entry' there.
1850	*/
1851	ce->ce_stat_data.sd_ctime.sec = get_be32(ondisk + offsetof(struct ondisk_cache_entry, ctime)
1852	+ offsetof(struct cache_time, sec));
1853	ce->ce_stat_data.sd_mtime.sec = get_be32(ondisk + offsetof(struct ondisk_cache_entry, mtime)
1854	+ offsetof(struct cache_time, sec));
1855	ce->ce_stat_data.sd_ctime.nsec = get_be32(ondisk + offsetof(struct ondisk_cache_entry, ctime)
1856	+ offsetof(struct cache_time, nsec));
1857	ce->ce_stat_data.sd_mtime.nsec = get_be32(ondisk + offsetof(struct ondisk_cache_entry, mtime)
1858	+ offsetof(struct cache_time, nsec));
1859	ce->ce_stat_data.sd_dev = get_be32(ondisk + offsetof(struct ondisk_cache_entry, dev));
1860	ce->ce_stat_data.sd_ino = get_be32(ondisk + offsetof(struct ondisk_cache_entry, ino));
1861	ce->ce_mode = get_be32(ondisk + offsetof(struct ondisk_cache_entry, mode));
1862	ce->ce_stat_data.sd_uid = get_be32(ondisk + offsetof(struct ondisk_cache_entry, uid));
1863	ce->ce_stat_data.sd_gid = get_be32(ondisk + offsetof(struct ondisk_cache_entry, gid));
1864	ce->ce_stat_data.sd_size = get_be32(ondisk + offsetof(struct ondisk_cache_entry, size));
1865	ce->ce_flags = flags & ~CE_NAMEMASK;
1866	ce->ce_namelen = len;
1867	ce->index = 0;
1868	oidread(&ce->oid, (const unsigned char *)ondisk + offsetof(struct ondisk_cache_entry, data),
1869	the_repository->hash_algo);
1870
1871	if (expand_name_field) {
1872	if (copy_len)
1873	memcpy(ce->name, previous_ce->name, copy_len);
1874	memcpy(ce->name + copy_len, name, len + 1 - copy_len);
1875	ent_size = (name - ((char )ondisk)) + len + 1 - copy_len;
1876	} else {
1877	memcpy(ce->name, name, len + 1);
1878	*ent_size = ondisk_ce_size(ce);
1879	}
1880	return ce;
1881	}
1882
1883	static void check_ce_order(struct index_state *istate)
1884	{
1885	unsigned int i;
1886
1887	if (!verify_ce_order)
1888	return;
1889
1890	for (i = 1; i < istate->cache_nr; i++) {
1891	struct cache_entry *ce = istate->cache[i - 1];
1892	struct cache_entry *next_ce = istate->cache[i];
1893	int name_compare = strcmp(ce->name, next_ce->name);
1894
1895	if (0 < name_compare)
1896	die(_("unordered stage entries in index"));
1897	if (!name_compare) {
1898	if (!ce_stage(ce))
1899	die(_("multiple stage entries for merged file '%s'"),
1900	ce->name);
1901	if (ce_stage(ce) > ce_stage(next_ce))
1902	die(_("unordered stage entries for '%s'"),
1903	ce->name);
1904	}
1905	}
1906	}
1907
1908	static void tweak_untracked_cache(struct index_state *istate)
1909	{
1910	struct repository *r = the_repository;
1911
1912	prepare_repo_settings(r);
1913
1914	switch (r->settings.core_untracked_cache) {
1915	case UNTRACKED_CACHE_REMOVE:
1916	remove_untracked_cache(istate);
1917	break;
1918	case UNTRACKED_CACHE_WRITE:
1919	add_untracked_cache(istate);
1920	break;
1921	case UNTRACKED_CACHE_KEEP:
1922	/*
1923	* Either an explicit "core.untrackedCache=keep", the
1924	* default if "core.untrackedCache" isn't configured,
1925	* or a fallback on an unknown "core.untrackedCache"
1926	* value.
1927	*/
1928	break;
1929	}
1930	}
1931
1932	static void tweak_split_index(struct index_state *istate)
1933	{
1934	switch (repo_config_get_split_index(the_repository)) {
1935	case -1: /* unset: do nothing */
1936	break;
1937	case 0: /* false */
1938	remove_split_index(istate);
1939	break;
1940	case 1: /* true */
1941	add_split_index(istate);
1942	break;
1943	default: /* unknown value: do nothing */
1944	break;
1945	}
1946	}
1947
1948	static void post_read_index_from(struct index_state *istate)
1949	{
1950	check_ce_order(istate);
1951	tweak_untracked_cache(istate);
1952	tweak_split_index(istate);
1953	tweak_fsmonitor(istate);
1954	}
1955
1956	static size_t estimate_cache_size_from_compressed(unsigned int entries)
1957	{
1958	return entries * (sizeof(struct cache_entry) + CACHE_ENTRY_PATH_LENGTH);
1959	}
1960
1961	static size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1962	{
1963	long per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1964
1965	/*
1966	* Account for potential alignment differences.
1967	*/
1968	per_entry += align_padding_size(per_entry, 0);
1969	return ondisk_size + entries * per_entry;
1970	}
1971
1972	struct index_entry_offset
1973	{
1974	/* starting byte offset into index file, count of index entries in this block */
1975	int offset, nr;
1976	};
1977
1978	struct index_entry_offset_table
1979	{
1980	int nr;
1981	struct index_entry_offset entries[FLEX_ARRAY];
1982	};
1983
1984	static struct index_entry_offset_table read_ieot_extension(const char mmap, size_t mmap_size, size_t offset);
1985	static void write_ieot_extension(struct strbuf sb, struct index_entry_offset_table ieot);
1986
1987	static size_t read_eoie_extension(const char *mmap, size_t mmap_size);
1988	static void write_eoie_extension(struct strbuf sb, struct git_hash_ctx eoie_context, size_t offset);
1989
1990	struct load_index_extensions
1991	{
1992	pthread_t pthread;
1993	struct index_state *istate;
1994	const char *mmap;
1995	size_t mmap_size;
1996	unsigned long src_offset;
1997	};
1998
1999	static void load_index_extensions(void _data)
2000	{
2001	struct load_index_extensions *p = _data;
2002	unsigned long src_offset = p->src_offset;
2003
2004	while (src_offset <= p->mmap_size - the_hash_algo->rawsz - 8) {
2005	/* After an array of active_nr index entries,
2006	* there can be arbitrary number of extended
2007	* sections, each of which is prefixed with
2008	* extension name (4-byte) and section length
2009	* in 4-byte network byte order.
2010	*/
2011	uint32_t extsize = get_be32(p->mmap + src_offset + 4);
2012	if (read_index_extension(p->istate,
2013	p->mmap + src_offset,
2014	p->mmap + src_offset + 8,
2015	extsize) < 0) {
2016	munmap((void *)p->mmap, p->mmap_size);
2017	die(_("index file corrupt"));
2018	}
2019	src_offset += 8;
2020	src_offset += extsize;
2021	}
2022
2023	return NULL;
2024	}
2025
2026	/*
2027	* A helper function that will load the specified range of cache entries
2028	* from the memory mapped file and add them to the given index.
2029	*/
2030	static unsigned long load_cache_entry_block(struct index_state *istate,
2031	struct mem_pool ce_mem_pool, int offset, int nr, const char mmap,
2032	unsigned long start_offset, const struct cache_entry *previous_ce)
2033	{
2034	int i;
2035	unsigned long src_offset = start_offset;
2036
2037	for (i = offset; i < offset + nr; i++) {
2038	struct cache_entry *ce;
2039	unsigned long consumed;
2040
2041	ce = create_from_disk(ce_mem_pool, istate->version,
2042	mmap + src_offset,
2043	&consumed, previous_ce);
2044	set_index_entry(istate, i, ce);
2045
2046	src_offset += consumed;
2047	previous_ce = ce;
2048	}
2049	return src_offset - start_offset;
2050	}
2051
2052	static unsigned long load_all_cache_entries(struct index_state *istate,
2053	const char *mmap, size_t mmap_size, unsigned long src_offset)
2054	{
2055	unsigned long consumed;
2056
2057	istate->ce_mem_pool = xmalloc(sizeof(*istate->ce_mem_pool));
2058	if (istate->version == 4) {
2059	mem_pool_init(istate->ce_mem_pool,
2060	estimate_cache_size_from_compressed(istate->cache_nr));
2061	} else {
2062	mem_pool_init(istate->ce_mem_pool,
2063	estimate_cache_size(mmap_size, istate->cache_nr));
2064	}
2065
2066	consumed = load_cache_entry_block(istate, istate->ce_mem_pool,
2067	0, istate->cache_nr, mmap, src_offset, NULL);
2068	return consumed;
2069	}
2070
2071	/*
2072	* Mostly randomly chosen maximum thread counts: we
2073	* cap the parallelism to online_cpus() threads, and we want
2074	* to have at least 10000 cache entries per thread for it to
2075	* be worth starting a thread.
2076	*/
2077
2078	#define THREAD_COST (10000)
2079
2080	struct load_cache_entries_thread_data
2081	{
2082	pthread_t pthread;
2083	struct index_state *istate;
2084	struct mem_pool *ce_mem_pool;
2085	int offset;
2086	const char *mmap;
2087	struct index_entry_offset_table *ieot;
2088	int ieot_start; /* starting index into the ieot array */
2089	int ieot_blocks; /* count of ieot entries to process */
2090	unsigned long consumed; /* return # of bytes in index file processed */
2091	};
2092
2093	/*
2094	* A thread proc to run the load_cache_entries() computation
2095	* across multiple background threads.
2096	*/
2097	static void load_cache_entries_thread(void _data)
2098	{
2099	struct load_cache_entries_thread_data *p = _data;
2100	int i;
2101
2102	/* iterate across all ieot blocks assigned to this thread */
2103	for (i = p->ieot_start; i < p->ieot_start + p->ieot_blocks; i++) {
2104	p->consumed += load_cache_entry_block(p->istate, p->ce_mem_pool,
2105	p->offset, p->ieot->entries[i].nr, p->mmap, p->ieot->entries[i].offset, NULL);
2106	p->offset += p->ieot->entries[i].nr;
2107	}
2108	return NULL;
2109	}
2110
2111	static unsigned long load_cache_entries_threaded(struct index_state istate, const char mmap, size_t mmap_size,
2112	int nr_threads, struct index_entry_offset_table *ieot)
2113	{
2114	int i, offset, ieot_blocks, ieot_start, err;
2115	struct load_cache_entries_thread_data *data;
2116	unsigned long consumed = 0;
2117
2118	/* a little sanity checking */
2119	if (istate->name_hash_initialized)
2120	BUG("the name hash isn't thread safe");
2121
2122	istate->ce_mem_pool = xmalloc(sizeof(*istate->ce_mem_pool));
2123	mem_pool_init(istate->ce_mem_pool, 0);
2124
2125	/* ensure we have no more threads than we have blocks to process */
2126	if (nr_threads > ieot->nr)
2127	nr_threads = ieot->nr;
2128	CALLOC_ARRAY(data, nr_threads);
2129
2130	offset = ieot_start = 0;
2131	ieot_blocks = DIV_ROUND_UP(ieot->nr, nr_threads);
2132	for (i = 0; i < nr_threads; i++) {
2133	struct load_cache_entries_thread_data *p = &data[i];
2134	int nr, j;
2135
2136	if (ieot_start + ieot_blocks > ieot->nr)
2137	ieot_blocks = ieot->nr - ieot_start;
2138
2139	p->istate = istate;
2140	p->offset = offset;
2141	p->mmap = mmap;
2142	p->ieot = ieot;
2143	p->ieot_start = ieot_start;
2144	p->ieot_blocks = ieot_blocks;
2145
2146	/* create a mem_pool for each thread */
2147	nr = 0;
2148	for (j = p->ieot_start; j < p->ieot_start + p->ieot_blocks; j++)
2149	nr += p->ieot->entries[j].nr;
2150	p->ce_mem_pool = xmalloc(sizeof(*istate->ce_mem_pool));
2151	if (istate->version == 4) {
2152	mem_pool_init(p->ce_mem_pool,
2153	estimate_cache_size_from_compressed(nr));
2154	} else {
2155	mem_pool_init(p->ce_mem_pool,
2156	estimate_cache_size(mmap_size, nr));
2157	}
2158
2159	err = pthread_create(&p->pthread, NULL, load_cache_entries_thread, p);
2160	if (err)
2161	die(_("unable to create load_cache_entries thread: %s"), strerror(err));
2162
2163	/* increment by the number of cache entries in the ieot block being processed */
2164	for (j = 0; j < ieot_blocks; j++)
2165	offset += ieot->entries[ieot_start + j].nr;
2166	ieot_start += ieot_blocks;
2167	}
2168
2169	for (i = 0; i < nr_threads; i++) {
2170	struct load_cache_entries_thread_data *p = &data[i];
2171
2172	err = pthread_join(p->pthread, NULL);
2173	if (err)
2174	die(_("unable to join load_cache_entries thread: %s"), strerror(err));
2175	mem_pool_combine(istate->ce_mem_pool, p->ce_mem_pool);
2176	free(p->ce_mem_pool);
2177	consumed += p->consumed;
2178	}
2179
2180	free(data);
2181
2182	return consumed;
2183	}
2184
2185	static void set_new_index_sparsity(struct index_state *istate)
2186	{
2187	/*
2188	* If the index's repo exists, mark it sparse according to
2189	* repo settings.
2190	*/
2191	prepare_repo_settings(istate->repo);
2192	if (!istate->repo->settings.command_requires_full_index &&
2193	is_sparse_index_allowed(istate, 0))
2194	istate->sparse_index = 1;
2195	}
2196
2197	/* remember to discard_cache() before reading a different cache! */
2198	int do_read_index(struct index_state istate, const char path, int must_exist)
2199	{
2200	int fd;
2201	struct stat st;
2202	unsigned long src_offset;
2203	const struct cache_header *hdr;
2204	const char *mmap;
2205	size_t mmap_size;
2206	struct load_index_extensions p;
2207	size_t extension_offset = 0;
2208	int nr_threads, cpus;
2209	struct index_entry_offset_table *ieot = NULL;
2210
2211	if (istate->initialized)
2212	return istate->cache_nr;
2213
2214	istate->timestamp.sec = 0;
2215	istate->timestamp.nsec = 0;
2216	fd = open(path, O_RDONLY);
2217	if (fd < 0) {
2218	if (!must_exist && errno == ENOENT) {
2219	set_new_index_sparsity(istate);
2220	istate->initialized = 1;
2221	return 0;
2222	}
2223	die_errno(_("%s: index file open failed"), path);
2224	}
2225
2226	if (fstat(fd, &st))
2227	die_errno(_("%s: cannot stat the open index"), path);
2228
2229	mmap_size = xsize_t(st.st_size);
2230	if (mmap_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
2231	die(_("%s: index file smaller than expected"), path);
2232
2233	mmap = xmmap_gently(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
2234	if (mmap == MAP_FAILED)
2235	die_errno(_("%s: unable to map index file%s"), path,
2236	mmap_os_err());
2237	close(fd);
2238
2239	hdr = (const struct cache_header *)mmap;
2240	if (verify_hdr(hdr, mmap_size) < 0)
2241	goto unmap;
2242
2243	oidread(&istate->oid, (const unsigned char *)hdr + mmap_size - the_hash_algo->rawsz,
2244	the_repository->hash_algo);
2245	istate->version = ntohl(hdr->hdr_version);
2246	istate->cache_nr = ntohl(hdr->hdr_entries);
2247	istate->cache_alloc = alloc_nr(istate->cache_nr);
2248	CALLOC_ARRAY(istate->cache, istate->cache_alloc);
2249	istate->initialized = 1;
2250
2251	p.istate = istate;
2252	p.mmap = mmap;
2253	p.mmap_size = mmap_size;
2254
2255	src_offset = sizeof(*hdr);
2256
2257	if (repo_config_get_index_threads(the_repository, &nr_threads))
2258	nr_threads = 1;
2259
2260	/* TODO: does creating more threads than cores help? */
2261	if (!nr_threads) {
2262	nr_threads = istate->cache_nr / THREAD_COST;
2263	cpus = online_cpus();
2264	if (nr_threads > cpus)
2265	nr_threads = cpus;
2266	}
2267
2268	if (!HAVE_THREADS)
2269	nr_threads = 1;
2270
2271	if (nr_threads > 1) {
2272	extension_offset = read_eoie_extension(mmap, mmap_size);
2273	if (extension_offset) {
2274	int err;
2275
2276	p.src_offset = extension_offset;
2277	err = pthread_create(&p.pthread, NULL, load_index_extensions, &p);
2278	if (err)
2279	die(_("unable to create load_index_extensions thread: %s"), strerror(err));
2280
2281	nr_threads--;
2282	}
2283	}
2284
2285	/*
2286	* Locate and read the index entry offset table so that we can use it
2287	* to multi-thread the reading of the cache entries.
2288	*/
2289	if (extension_offset && nr_threads > 1)
2290	ieot = read_ieot_extension(mmap, mmap_size, extension_offset);
2291
2292	if (ieot) {
2293	src_offset += load_cache_entries_threaded(istate, mmap, mmap_size, nr_threads, ieot);
2294	free(ieot);
2295	} else {
2296	src_offset += load_all_cache_entries(istate, mmap, mmap_size, src_offset);
2297	}
2298
2299	istate->timestamp.sec = st.st_mtime;
2300	istate->timestamp.nsec = ST_MTIME_NSEC(st);
2301
2302	/* if we created a thread, join it otherwise load the extensions on the primary thread */
2303	if (extension_offset) {
2304	int ret = pthread_join(p.pthread, NULL);
2305	if (ret)
2306	die(_("unable to join load_index_extensions thread: %s"), strerror(ret));
2307	} else {
2308	p.src_offset = src_offset;
2309	load_index_extensions(&p);
2310	}
2311	munmap((void *)mmap, mmap_size);
2312
2313	trace2_data_intmax("index", istate->repo, "read/version",
2314	istate->version);
2315	trace2_data_intmax("index", istate->repo, "read/cache_nr",
2316	istate->cache_nr);
2317
2318	/*
2319	* If the command explicitly requires a full index, force it
2320	* to be full. Otherwise, correct the sparsity based on repository
2321	* settings and other properties of the index (if necessary).
2322	*/
2323	prepare_repo_settings(istate->repo);
2324	if (istate->repo->settings.command_requires_full_index)
2325	ensure_full_index(istate);
2326	else
2327	ensure_correct_sparsity(istate);
2328
2329	return istate->cache_nr;
2330
2331	unmap:
2332	munmap((void *)mmap, mmap_size);
2333	die(_("index file corrupt"));
2334	}
2335
2336	/*
2337	* Signal that the shared index is used by updating its mtime.
2338	*
2339	* This way, shared index can be removed if they have not been used
2340	* for some time.
2341	*/
2342	static void freshen_shared_index(const char *shared_index, int warn)
2343	{
2344	if (!check_and_freshen_file(shared_index, 1, NULL) && warn)
2345	warning(_("could not freshen shared index '%s'"), shared_index);
2346	}
2347
2348	int read_index_from(struct index_state istate, const char path,
2349	const char *gitdir)
2350	{
2351	struct split_index *split_index;
2352	int ret;
2353	char *base_oid_hex;
2354	char *base_path;
2355
2356	/* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2357	if (istate->initialized)
2358	return istate->cache_nr;
2359
2360	trace2_region_enter_printf("index", "do_read_index", istate->repo,
2361	"%s", path);
2362	trace_performance_enter();
2363	ret = do_read_index(istate, path, 0);
2364	trace_performance_leave("read cache %s", path);
2365	trace2_region_leave_printf("index", "do_read_index", istate->repo,
2366	"%s", path);
2367
2368	split_index = istate->split_index;
2369	if (!split_index \|\| is_null_oid(&split_index->base_oid)) {
2370	post_read_index_from(istate);
2371	return ret;
2372	}
2373
2374	trace_performance_enter();
2375	if (split_index->base)
2376	release_index(split_index->base);
2377	else
2378	ALLOC_ARRAY(split_index->base, 1);
2379	index_state_init(split_index->base, istate->repo);
2380
2381	base_oid_hex = oid_to_hex(&split_index->base_oid);
2382	base_path = xstrfmt("%s/sharedindex.%s", gitdir, base_oid_hex);
2383	if (file_exists(base_path)) {
2384	trace2_region_enter_printf("index", "shared/do_read_index",
2385	the_repository, "%s", base_path);
2386
2387	ret = do_read_index(split_index->base, base_path, 0);
2388	trace2_region_leave_printf("index", "shared/do_read_index",
2389	the_repository, "%s", base_path);
2390	} else {
2391	char *path_copy = xstrdup(path);
2392	char *base_path2 = xstrfmt("%s/sharedindex.%s",
2393	dirname(path_copy), base_oid_hex);
2394	free(path_copy);
2395	trace2_region_enter_printf("index", "shared/do_read_index",
2396	the_repository, "%s", base_path2);
2397	ret = do_read_index(split_index->base, base_path2, 1);
2398	trace2_region_leave_printf("index", "shared/do_read_index",
2399	the_repository, "%s", base_path2);
2400	free(base_path2);
2401	}
2402	if (!oideq(&split_index->base_oid, &split_index->base->oid))
2403	die(_("broken index, expect %s in %s, got %s"),
2404	base_oid_hex, base_path,
2405	oid_to_hex(&split_index->base->oid));
2406
2407	freshen_shared_index(base_path, 0);
2408	merge_base_index(istate);
2409	post_read_index_from(istate);
2410	trace_performance_leave("read cache %s", base_path);
2411	free(base_path);
2412	return ret;
2413	}
2414
2415	int is_index_unborn(struct index_state *istate)
2416	{
2417	return (!istate->cache_nr && !istate->timestamp.sec);
2418	}
2419
2420	void index_state_init(struct index_state istate, struct repository r)
2421	{
2422	struct index_state blank = INDEX_STATE_INIT(r);
2423	memcpy(istate, &blank, sizeof(*istate));
2424	}
2425
2426	void release_index(struct index_state *istate)
2427	{
2428	/*
2429	* Cache entries in istate->cache[] should have been allocated
2430	* from the memory pool associated with this index, or from an
2431	* associated split_index. There is no need to free individual
2432	* cache entries. validate_cache_entries can detect when this
2433	* assertion does not hold.
2434	*/
2435	validate_cache_entries(istate);
2436
2437	resolve_undo_clear_index(istate);
2438	free_name_hash(istate);
2439	cache_tree_free(&(istate->cache_tree));
2440	free(istate->fsmonitor_last_update);
2441	free(istate->cache);
2442	discard_split_index(istate);
2443	free_untracked_cache(istate->untracked);
2444
2445	if (istate->sparse_checkout_patterns) {
2446	clear_pattern_list(istate->sparse_checkout_patterns);
2447	FREE_AND_NULL(istate->sparse_checkout_patterns);
2448	}
2449
2450	if (istate->ce_mem_pool) {
2451	mem_pool_discard(istate->ce_mem_pool, should_validate_cache_entries());
2452	FREE_AND_NULL(istate->ce_mem_pool);
2453	}
2454	}
2455
2456	void discard_index(struct index_state *istate)
2457	{
2458	release_index(istate);
2459	index_state_init(istate, istate->repo);
2460	}
2461
2462	/*
2463	* Validate the cache entries of this index.
2464	* All cache entries associated with this index
2465	* should have been allocated by the memory pool
2466	* associated with this index, or by a referenced
2467	* split index.
2468	*/
2469	void validate_cache_entries(const struct index_state *istate)
2470	{
2471	int i;
2472
2473	if (!should_validate_cache_entries() \|\|!istate \|\| !istate->initialized)
2474	return;
2475
2476	for (i = 0; i < istate->cache_nr; i++) {
2477	if (!istate) {
2478	BUG("cache entry is not allocated from expected memory pool");
2479	} else if (!istate->ce_mem_pool \|\|
2480	!mem_pool_contains(istate->ce_mem_pool, istate->cache[i])) {
2481	if (!istate->split_index \|\|
2482	!istate->split_index->base \|\|
2483	!istate->split_index->base->ce_mem_pool \|\|
2484	!mem_pool_contains(istate->split_index->base->ce_mem_pool, istate->cache[i])) {
2485	BUG("cache entry is not allocated from expected memory pool");
2486	}
2487	}
2488	}
2489
2490	if (istate->split_index)
2491	validate_cache_entries(istate->split_index->base);
2492	}
2493
2494	int unmerged_index(const struct index_state *istate)
2495	{
2496	int i;
2497	for (i = 0; i < istate->cache_nr; i++) {
2498	if (ce_stage(istate->cache[i]))
2499	return 1;
2500	}
2501	return 0;
2502	}
2503
2504	int repo_index_has_changes(struct repository *repo,
2505	struct tree *tree,
2506	struct strbuf *sb)
2507	{
2508	struct index_state *istate = repo->index;
2509	struct object_id cmp;
2510	int i;
2511
2512	if (tree)
2513	cmp = tree->object.oid;
2514	if (tree \|\| !repo_get_oid_tree(repo, "HEAD", &cmp)) {
2515	struct diff_options opt;
2516
2517	repo_diff_setup(repo, &opt);
2518	opt.flags.exit_with_status = 1;
2519	if (!sb)
2520	opt.flags.quick = 1;
2521	diff_setup_done(&opt);
2522	do_diff_cache(&cmp, &opt);
2523	diffcore_std(&opt);
2524	for (i = 0; sb && i < diff_queued_diff.nr; i++) {
2525	if (i)
2526	strbuf_addch(sb, ' ');
2527	strbuf_addstr(sb, diff_queued_diff.queue[i]->two->path);
2528	}
2529	diff_flush(&opt);
2530	return opt.flags.has_changes != 0;
2531	} else {
2532	/* TODO: audit for interaction with sparse-index. */
2533	ensure_full_index(istate);
2534	for (i = 0; sb && i < istate->cache_nr; i++) {
2535	if (i)
2536	strbuf_addch(sb, ' ');
2537	strbuf_addstr(sb, istate->cache[i]->name);
2538	}
2539	return !!istate->cache_nr;
2540	}
2541	}
2542
2543	static int write_index_ext_header(struct hashfile *f,
2544	struct git_hash_ctx *eoie_f,
2545	unsigned int ext,
2546	unsigned int sz)
2547	{
2548	hashwrite_be32(f, ext);
2549	hashwrite_be32(f, sz);
2550
2551	if (eoie_f) {
2552	ext = htonl(ext);
2553	sz = htonl(sz);
2554	git_hash_update(eoie_f, &ext, sizeof(ext));
2555	git_hash_update(eoie_f, &sz, sizeof(sz));
2556	}
2557	return 0;
2558	}
2559
2560	static void ce_smudge_racily_clean_entry(struct index_state *istate,
2561	struct cache_entry *ce)
2562	{
2563	/*
2564	* The only thing we care about in this function is to smudge the
2565	* falsely clean entry due to touch-update-touch race, so we leave
2566	* everything else as they are. We are called for entries whose
2567	* ce_stat_data.sd_mtime match the index file mtime.
2568	*
2569	* Note that this actually does not do much for gitlinks, for
2570	* which ce_match_stat_basic() always goes to the actual
2571	* contents. The caller checks with is_racy_timestamp() which
2572	* always says "no" for gitlinks, so we are not called for them ;-)
2573	*/
2574	struct stat st;
2575
2576	if (lstat(ce->name, &st) < 0)
2577	return;
2578	if (ce_match_stat_basic(ce, &st))
2579	return;
2580	if (ce_modified_check_fs(istate, ce, &st)) {
2581	/* This is "racily clean"; smudge it. Note that this
2582	* is a tricky code. At first glance, it may appear
2583	* that it can break with this sequence:
2584	*
2585	* $ echo xyzzy >frotz
2586	* $ git-update-index --add frotz
2587	* $ : >frotz
2588	* $ sleep 3
2589	* $ echo filfre >nitfol
2590	* $ git-update-index --add nitfol
2591	*
2592	* but it does not. When the second update-index runs,
2593	* it notices that the entry "frotz" has the same timestamp
2594	* as index, and if we were to smudge it by resetting its
2595	* size to zero here, then the object name recorded
2596	* in index is the 6-byte file but the cached stat information
2597	* becomes zero --- which would then match what we would
2598	* obtain from the filesystem next time we stat("frotz").
2599	*
2600	* However, the second update-index, before calling
2601	* this function, notices that the cached size is 6
2602	* bytes and what is on the filesystem is an empty
2603	* file, and never calls us, so the cached size information
2604	* for "frotz" stays 6 which does not match the filesystem.
2605	*/
2606	ce->ce_stat_data.sd_size = 0;
2607	}
2608	}
2609
2610	/* Copy miscellaneous fields but not the name */
2611	static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
2612	struct cache_entry *ce)
2613	{
2614	short flags;
2615	const unsigned hashsz = the_hash_algo->rawsz;
2616	uint16_t flagsp = (uint16_t )(ondisk->data + hashsz);
2617
2618	ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
2619	ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
2620	ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
2621	ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
2622	ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
2623	ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
2624	ondisk->mode = htonl(ce->ce_mode);
2625	ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
2626	ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
2627	ondisk->size = htonl(ce->ce_stat_data.sd_size);
2628	hashcpy(ondisk->data, ce->oid.hash, the_repository->hash_algo);
2629
2630	flags = ce->ce_flags & ~CE_NAMEMASK;
2631	flags \|= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
2632	flagsp[0] = htons(flags);
2633	if (ce->ce_flags & CE_EXTENDED) {
2634	flagsp[1] = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
2635	}
2636	}
2637
2638	static int ce_write_entry(struct hashfile f, struct cache_entry ce,
2639	struct strbuf previous_name, struct ondisk_cache_entry ondisk)
2640	{
2641	int size;
2642	unsigned int saved_namelen;
2643	int stripped_name = 0;
2644	static unsigned char padding[8] = { 0x00 };
2645
2646	if (ce->ce_flags & CE_STRIP_NAME) {
2647	saved_namelen = ce_namelen(ce);
2648	ce->ce_namelen = 0;
2649	stripped_name = 1;
2650	}
2651
2652	size = offsetof(struct ondisk_cache_entry,data) + ondisk_data_size(ce->ce_flags, 0);
2653
2654	if (!previous_name) {
2655	int len = ce_namelen(ce);
2656	copy_cache_entry_to_ondisk(ondisk, ce);
2657	hashwrite(f, ondisk, size);
2658	hashwrite(f, ce->name, len);
2659	hashwrite(f, padding, align_padding_size(size, len));
2660	} else {
2661	int common, to_remove;
2662	uint8_t prefix_size;
2663	unsigned char to_remove_vi[16];
2664
2665	for (common = 0;
2666	(common < previous_name->len &&
2667	ce->name[common] &&
2668	ce->name[common] == previous_name->buf[common]);
2669	common++)
2670	; /* still matching */
2671	to_remove = previous_name->len - common;
2672	prefix_size = encode_varint(to_remove, to_remove_vi);
2673
2674	copy_cache_entry_to_ondisk(ondisk, ce);
2675	hashwrite(f, ondisk, size);
2676	hashwrite(f, to_remove_vi, prefix_size);
2677	hashwrite(f, ce->name + common, ce_namelen(ce) - common);
2678	hashwrite(f, padding, 1);
2679
2680	strbuf_splice(previous_name, common, to_remove,
2681	ce->name + common, ce_namelen(ce) - common);
2682	}
2683	if (stripped_name) {
2684	ce->ce_namelen = saved_namelen;
2685	ce->ce_flags &= ~CE_STRIP_NAME;
2686	}
2687
2688	return 0;
2689	}
2690
2691	/*
2692	* This function verifies if index_state has the correct sha1 of the
2693	* index file. Don't die if we have any other failure, just return 0.
2694	*/
2695	static int verify_index_from(const struct index_state istate, const char path)
2696	{
2697	int fd;
2698	ssize_t n;
2699	struct stat st;
2700	unsigned char hash[GIT_MAX_RAWSZ];
2701
2702	if (!istate->initialized)
2703	return 0;
2704
2705	fd = open(path, O_RDONLY);
2706	if (fd < 0)
2707	return 0;
2708
2709	if (fstat(fd, &st))
2710	goto out;
2711
2712	if (st.st_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
2713	goto out;
2714
2715	n = pread_in_full(fd, hash, the_hash_algo->rawsz, st.st_size - the_hash_algo->rawsz);
2716	if (n != the_hash_algo->rawsz)
2717	goto out;
2718
2719	if (!hasheq(istate->oid.hash, hash, the_repository->hash_algo))
2720	goto out;
2721
2722	close(fd);
2723	return 1;
2724
2725	out:
2726	close(fd);
2727	return 0;
2728	}
2729
2730	static int repo_verify_index(struct repository *repo)
2731	{
2732	return verify_index_from(repo->index, repo->index_file);
2733	}
2734
2735	int has_racy_timestamp(struct index_state *istate)
2736	{
2737	int entries = istate->cache_nr;
2738	int i;
2739
2740	for (i = 0; i < entries; i++) {
2741	struct cache_entry *ce = istate->cache[i];
2742	if (is_racy_timestamp(istate, ce))
2743	return 1;
2744	}
2745	return 0;
2746	}
2747
2748	void repo_update_index_if_able(struct repository *repo,
2749	struct lock_file *lockfile)
2750	{
2751	if ((repo->index->cache_changed \|\|
2752	has_racy_timestamp(repo->index)) &&
2753	repo_verify_index(repo))
2754	write_locked_index(repo->index, lockfile, COMMIT_LOCK);
2755	else
2756	rollback_lock_file(lockfile);
2757	}
2758
2759	static int record_eoie(void)
2760	{
2761	int val;
2762
2763	if (!repo_config_get_bool(the_repository, "index.recordendofindexentries", &val))
2764	return val;
2765
2766	/*
2767	* As a convenience, the end of index entries extension
2768	* used for threading is written by default if the user
2769	* explicitly requested threaded index reads.
2770	*/
2771	return !repo_config_get_index_threads(the_repository, &val) && val != 1;
2772	}
2773
2774	static int record_ieot(void)
2775	{
2776	int val;
2777
2778	if (!repo_config_get_bool(the_repository, "index.recordoffsettable", &val))
2779	return val;
2780
2781	/*
2782	* As a convenience, the offset table used for threading is
2783	* written by default if the user explicitly requested
2784	* threaded index reads.
2785	*/
2786	return !repo_config_get_index_threads(the_repository, &val) && val != 1;
2787	}
2788
2789	enum write_extensions {
2790	WRITE_NO_EXTENSION = 0,
2791	WRITE_SPLIT_INDEX_EXTENSION = 1<<0,
2792	WRITE_CACHE_TREE_EXTENSION = 1<<1,
2793	WRITE_RESOLVE_UNDO_EXTENSION = 1<<2,
2794	WRITE_UNTRACKED_CACHE_EXTENSION = 1<<3,
2795	WRITE_FSMONITOR_EXTENSION = 1<<4,
2796	};
2797	#define WRITE_ALL_EXTENSIONS ((enum write_extensions)-1)
2798
2799	/*
2800	* On success, `tempfile` is closed. If it is the temporary file
2801	* of a `struct lock_file`, we will therefore effectively perform
2802	* a 'close_lock_file_gently()`. Since that is an implementation
2803	* detail of lockfiles, callers of `do_write_index()` should not
2804	* rely on it.
2805	*/
2806	static int do_write_index(struct index_state istate, struct tempfile tempfile,
2807	enum write_extensions write_extensions, unsigned flags)
2808	{
2809	uint64_t start = getnanotime();
2810	struct hashfile *f;
2811	struct git_hash_ctx *eoie_c = NULL;
2812	struct cache_header hdr;
2813	int i, err = 0, removed, extended, hdr_version;
2814	struct cache_entry **cache = istate->cache;
2815	int entries = istate->cache_nr;
2816	struct stat st;
2817	struct ondisk_cache_entry ondisk;
2818	struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
2819	int drop_cache_tree = istate->drop_cache_tree;
2820	off_t offset;
2821	int csum_fsync_flag;
2822	int ieot_entries = 1;
2823	struct index_entry_offset_table *ieot = NULL;
2824	struct repository *r = istate->repo;
2825	struct strbuf sb = STRBUF_INIT;
2826	int nr, nr_threads, ret;
2827
2828	f = hashfd(the_repository->hash_algo, tempfile->fd, tempfile->filename.buf);
2829
2830	prepare_repo_settings(r);
2831	f->skip_hash = r->settings.index_skip_hash;
2832
2833	for (i = removed = extended = 0; i < entries; i++) {
2834	if (cache[i]->ce_flags & CE_REMOVE)
2835	removed++;
2836
2837	/* reduce extended entries if possible */
2838	cache[i]->ce_flags &= ~CE_EXTENDED;
2839	if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2840	extended++;
2841	cache[i]->ce_flags \|= CE_EXTENDED;
2842	}
2843	}
2844
2845	if (!istate->version)
2846	istate->version = get_index_format_default(r);
2847
2848	/* demote version 3 to version 2 when the latter suffices */
2849	if (istate->version == 3 \|\| istate->version == 2)
2850	istate->version = extended ? 3 : 2;
2851
2852	hdr_version = istate->version;
2853
2854	hdr.hdr_signature = htonl(CACHE_SIGNATURE);
2855	hdr.hdr_version = htonl(hdr_version);
2856	hdr.hdr_entries = htonl(entries - removed);
2857
2858	hashwrite(f, &hdr, sizeof(hdr));
2859
2860	if (!HAVE_THREADS \|\| repo_config_get_index_threads(the_repository, &nr_threads))
2861	nr_threads = 1;
2862
2863	if (nr_threads != 1 && record_ieot()) {
2864	int ieot_blocks, cpus;
2865
2866	/*
2867	* ensure default number of ieot blocks maps evenly to the
2868	* default number of threads that will process them leaving
2869	* room for the thread to load the index extensions.
2870	*/
2871	if (!nr_threads) {
2872	ieot_blocks = istate->cache_nr / THREAD_COST;
2873	cpus = online_cpus();
2874	if (ieot_blocks > cpus - 1)
2875	ieot_blocks = cpus - 1;
2876	} else {
2877	ieot_blocks = nr_threads;
2878	if (ieot_blocks > istate->cache_nr)
2879	ieot_blocks = istate->cache_nr;
2880	}
2881
2882	/*
2883	* no reason to write out the IEOT extension if we don't
2884	* have enough blocks to utilize multi-threading
2885	*/
2886	if (ieot_blocks > 1) {
2887	ieot = xcalloc(1, sizeof(struct index_entry_offset_table)
2888	+ (ieot_blocks * sizeof(struct index_entry_offset)));
2889	ieot_entries = DIV_ROUND_UP(entries, ieot_blocks);
2890	}
2891	}
2892
2893	offset = hashfile_total(f);
2894
2895	nr = 0;
2896	previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2897
2898	for (i = 0; i < entries; i++) {
2899	struct cache_entry *ce = cache[i];
2900	if (ce->ce_flags & CE_REMOVE)
2901	continue;
2902	if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2903	ce_smudge_racily_clean_entry(istate, ce);
2904	if (is_null_oid(&ce->oid)) {
2905	static const char msg[] = "cache entry has null sha1: %s";
2906	static int allow = -1;
2907
2908	if (allow < 0)
2909	allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2910	if (allow)
2911	warning(msg, ce->name);
2912	else
2913	err = error(msg, ce->name);
2914
2915	drop_cache_tree = 1;
2916	}
2917	if (ieot && i && (i % ieot_entries == 0)) {
2918	ieot->entries[ieot->nr].nr = nr;
2919	ieot->entries[ieot->nr].offset = offset;
2920	ieot->nr++;
2921	/*
2922	* If we have a V4 index, set the first byte to an invalid
2923	* character to ensure there is nothing common with the previous
2924	* entry
2925	*/
2926	if (previous_name)
2927	previous_name->buf[0] = 0;
2928	nr = 0;
2929
2930	offset = hashfile_total(f);
2931	}
2932	if (ce_write_entry(f, ce, previous_name, (struct ondisk_cache_entry *)&ondisk) < 0)
2933	err = -1;
2934
2935	if (err)
2936	break;
2937	nr++;
2938	}
2939	if (ieot && nr) {
2940	ieot->entries[ieot->nr].nr = nr;
2941	ieot->entries[ieot->nr].offset = offset;
2942	ieot->nr++;
2943	}
2944	strbuf_release(&previous_name_buf);
2945
2946	if (err) {
2947	ret = err;
2948	goto out;
2949	}
2950
2951	offset = hashfile_total(f);
2952
2953	/*
2954	* The extension headers must be hashed on their own for the
2955	* EOIE extension. Create a hashfile here to compute that hash.
2956	*/
2957	if (offset && record_eoie()) {
2958	CALLOC_ARRAY(eoie_c, 1);
2959	git_hash_init(eoie_c, the_hash_algo);
2960	}
2961
2962	/*
2963	* Lets write out CACHE_EXT_INDEXENTRYOFFSETTABLE first so that we
2964	* can minimize the number of extensions we have to scan through to
2965	* find it during load. Write it out regardless of the
2966	* strip_extensions parameter as we need it when loading the shared
2967	* index.
2968	*/
2969	if (ieot) {
2970	strbuf_reset(&sb);
2971
2972	write_ieot_extension(&sb, ieot);
2973	err = write_index_ext_header(f, eoie_c, CACHE_EXT_INDEXENTRYOFFSETTABLE, sb.len) < 0;
2974	hashwrite(f, sb.buf, sb.len);
2975	if (err) {
2976	ret = -1;
2977	goto out;
2978	}
2979	}
2980
2981	if (write_extensions & WRITE_SPLIT_INDEX_EXTENSION &&
2982	istate->split_index) {
2983	strbuf_reset(&sb);
2984
2985	if (istate->sparse_index)
2986	die(_("cannot write split index for a sparse index"));
2987
2988	err = write_link_extension(&sb, istate) < 0 \|\|
2989	write_index_ext_header(f, eoie_c, CACHE_EXT_LINK,
2990	sb.len) < 0;
2991	hashwrite(f, sb.buf, sb.len);
2992	if (err) {
2993	ret = -1;
2994	goto out;
2995	}
2996	}
2997	if (write_extensions & WRITE_CACHE_TREE_EXTENSION &&
2998	!drop_cache_tree && istate->cache_tree) {
2999	strbuf_reset(&sb);
3000
3001	cache_tree_write(&sb, istate->cache_tree);
3002	err = write_index_ext_header(f, eoie_c, CACHE_EXT_TREE, sb.len) < 0;
3003	hashwrite(f, sb.buf, sb.len);
3004	if (err) {
3005	ret = -1;
3006	goto out;
3007	}
3008	}
3009	if (write_extensions & WRITE_RESOLVE_UNDO_EXTENSION &&
3010	istate->resolve_undo) {
3011	strbuf_reset(&sb);
3012
3013	resolve_undo_write(&sb, istate->resolve_undo, the_hash_algo);
3014	err = write_index_ext_header(f, eoie_c, CACHE_EXT_RESOLVE_UNDO,
3015	sb.len) < 0;
3016	hashwrite(f, sb.buf, sb.len);
3017	if (err) {
3018	ret = -1;
3019	goto out;
3020	}
3021	}
3022	if (write_extensions & WRITE_UNTRACKED_CACHE_EXTENSION &&
3023	istate->untracked) {
3024	strbuf_reset(&sb);
3025
3026	write_untracked_extension(&sb, istate->untracked);
3027	err = write_index_ext_header(f, eoie_c, CACHE_EXT_UNTRACKED,
3028	sb.len) < 0;
3029	hashwrite(f, sb.buf, sb.len);
3030	if (err) {
3031	ret = -1;
3032	goto out;
3033	}
3034	}
3035	if (write_extensions & WRITE_FSMONITOR_EXTENSION &&
3036	istate->fsmonitor_last_update) {
3037	strbuf_reset(&sb);
3038
3039	write_fsmonitor_extension(&sb, istate);
3040	err = write_index_ext_header(f, eoie_c, CACHE_EXT_FSMONITOR, sb.len) < 0;
3041	hashwrite(f, sb.buf, sb.len);
3042	if (err) {
3043	ret = -1;
3044	goto out;
3045	}
3046	}
3047	if (istate->sparse_index) {
3048	if (write_index_ext_header(f, eoie_c, CACHE_EXT_SPARSE_DIRECTORIES, 0) < 0) {
3049	ret = -1;
3050	goto out;
3051	}
3052	}
3053
3054	/*
3055	* CACHE_EXT_ENDOFINDEXENTRIES must be written as the last entry before the SHA1
3056	* so that it can be found and processed before all the index entries are
3057	* read. Write it out regardless of the strip_extensions parameter as we need it
3058	* when loading the shared index.
3059	*/
3060	if (eoie_c) {
3061	strbuf_reset(&sb);
3062
3063	write_eoie_extension(&sb, eoie_c, offset);
3064	err = write_index_ext_header(f, NULL, CACHE_EXT_ENDOFINDEXENTRIES, sb.len) < 0;
3065	hashwrite(f, sb.buf, sb.len);
3066	if (err) {
3067	ret = -1;
3068	goto out;
3069	}
3070	}
3071
3072	csum_fsync_flag = 0;
3073	if (!alternate_index_output && (flags & COMMIT_LOCK))
3074	csum_fsync_flag = CSUM_FSYNC;
3075
3076	finalize_hashfile(f, istate->oid.hash, FSYNC_COMPONENT_INDEX,
3077	CSUM_HASH_IN_STREAM \| csum_fsync_flag);
3078	f = NULL;
3079
3080	if (close_tempfile_gently(tempfile)) {
3081	ret = error(_("could not close '%s'"), get_tempfile_path(tempfile));
3082	goto out;
3083	}
3084	if (stat(get_tempfile_path(tempfile), &st)) {
3085	ret = -1;
3086	goto out;
3087	}
3088	istate->timestamp.sec = (unsigned int)st.st_mtime;
3089	istate->timestamp.nsec = ST_MTIME_NSEC(st);
3090	trace_performance_since(start, "write index, changed mask = %x", istate->cache_changed);
3091
3092	trace2_data_intmax("index", istate->repo, "write/version",
3093	istate->version);
3094	trace2_data_intmax("index", istate->repo, "write/cache_nr",
3095	istate->cache_nr);
3096
3097	ret = 0;
3098
3099	out:
3100	if (f)
3101	free_hashfile(f);
3102	strbuf_release(&sb);
3103	free(eoie_c);
3104	free(ieot);
3105	return ret;
3106	}
3107
3108	void set_alternate_index_output(const char *name)
3109	{
3110	alternate_index_output = name;
3111	}
3112
3113	static int commit_locked_index(struct lock_file *lk)
3114	{
3115	if (alternate_index_output)
3116	return commit_lock_file_to(lk, alternate_index_output);
3117	else
3118	return commit_lock_file(lk);
3119	}
3120
3121	static int do_write_locked_index(struct index_state *istate,
3122	struct lock_file *lock,
3123	unsigned flags,
3124	enum write_extensions write_extensions)
3125	{
3126	int ret;
3127	int was_full = istate->sparse_index == INDEX_EXPANDED;
3128
3129	ret = convert_to_sparse(istate, 0);
3130
3131	if (ret) {
3132	warning(_("failed to convert to a sparse-index"));
3133	return ret;
3134	}
3135
3136	trace2_region_enter_printf("index", "do_write_index", istate->repo,
3137	"%s", get_lock_file_path(lock));
3138	ret = do_write_index(istate, lock->tempfile, write_extensions, flags);
3139	trace2_region_leave_printf("index", "do_write_index", istate->repo,
3140	"%s", get_lock_file_path(lock));
3141
3142	if (was_full)
3143	ensure_full_index(istate);
3144
3145	if (ret)
3146	return ret;
3147	if (flags & COMMIT_LOCK)
3148	ret = commit_locked_index(lock);
3149	else
3150	ret = close_lock_file_gently(lock);
3151
3152	run_hooks_l(the_repository, "post-index-change",
3153	istate->updated_workdir ? "1" : "0",
3154	istate->updated_skipworktree ? "1" : "0", NULL);
3155	istate->updated_workdir = 0;
3156	istate->updated_skipworktree = 0;
3157
3158	return ret;
3159	}
3160
3161	static int write_split_index(struct index_state *istate,
3162	struct lock_file *lock,
3163	unsigned flags)
3164	{
3165	int ret;
3166	prepare_to_write_split_index(istate);
3167	ret = do_write_locked_index(istate, lock, flags, WRITE_ALL_EXTENSIONS);
3168	finish_writing_split_index(istate);
3169	return ret;
3170	}
3171
3172	static unsigned long get_shared_index_expire_date(void)
3173	{
3174	static unsigned long shared_index_expire_date;
3175	static int shared_index_expire_date_prepared;
3176
3177	if (!shared_index_expire_date_prepared) {
3178	const char *shared_index_expire = "2.weeks.ago";
3179	char *value = NULL;
3180
3181	repo_config_get_expiry(the_repository, "splitindex.sharedindexexpire",
3182	&value);
3183	if (value)
3184	shared_index_expire = value;
3185
3186	shared_index_expire_date = approxidate(shared_index_expire);
3187	shared_index_expire_date_prepared = 1;
3188
3189	free(value);
3190	}
3191
3192	return shared_index_expire_date;
3193	}
3194
3195	static int should_delete_shared_index(const char *shared_index_path)
3196	{
3197	struct stat st;
3198	unsigned long expiration;
3199
3200	/* Check timestamp */
3201	expiration = get_shared_index_expire_date();
3202	if (!expiration)
3203	return 0;
3204	if (stat(shared_index_path, &st))
3205	return error_errno(_("could not stat '%s'"), shared_index_path);
3206	if (st.st_mtime > expiration)
3207	return 0;
3208
3209	return 1;
3210	}
3211
3212	static int clean_shared_index_files(const char *current_hex)
3213	{
3214	struct dirent *de;
3215	DIR *dir = opendir(repo_get_git_dir(the_repository));
3216
3217	if (!dir)
3218	return error_errno(_("unable to open git dir: %s"),
3219	repo_get_git_dir(the_repository));
3220
3221	while ((de = readdir(dir)) != NULL) {
3222	const char *sha1_hex;
3223	char *shared_index_path;
3224	if (!skip_prefix(de->d_name, "sharedindex.", &sha1_hex))
3225	continue;
3226	if (!strcmp(sha1_hex, current_hex))
3227	continue;
3228
3229	shared_index_path = repo_git_path(the_repository, "%s", de->d_name);
3230	if (should_delete_shared_index(shared_index_path) > 0 &&
3231	unlink(shared_index_path))
3232	warning_errno(_("unable to unlink: %s"), shared_index_path);
3233
3234	free(shared_index_path);
3235	}
3236	closedir(dir);
3237
3238	return 0;
3239	}
3240
3241	static int write_shared_index(struct index_state *istate,
3242	struct tempfile **temp, unsigned flags)
3243	{
3244	struct split_index *si = istate->split_index;
3245	int ret, was_full = !istate->sparse_index;
3246	char *path;
3247
3248	move_cache_to_base_index(istate);
3249	convert_to_sparse(istate, 0);
3250
3251	trace2_region_enter_printf("index", "shared/do_write_index",
3252	the_repository, "%s", get_tempfile_path(*temp));
3253	ret = do_write_index(si->base, *temp, WRITE_NO_EXTENSION, flags);
3254	trace2_region_leave_printf("index", "shared/do_write_index",
3255	the_repository, "%s", get_tempfile_path(*temp));
3256
3257	if (was_full)
3258	ensure_full_index(istate);
3259
3260	if (ret)
3261	return ret;
3262	ret = adjust_shared_perm(the_repository, get_tempfile_path(*temp));
3263	if (ret) {
3264	error(_("cannot fix permission bits on '%s'"), get_tempfile_path(*temp));
3265	return ret;
3266	}
3267
3268	path = repo_git_path(the_repository, "sharedindex.%s", oid_to_hex(&si->base->oid));
3269	ret = rename_tempfile(temp, path);
3270	if (!ret) {
3271	oidcpy(&si->base_oid, &si->base->oid);
3272	clean_shared_index_files(oid_to_hex(&si->base->oid));
3273	}
3274
3275	free(path);
3276	return ret;
3277	}
3278
3279	static const int default_max_percent_split_change = 20;
3280
3281	static int too_many_not_shared_entries(struct index_state *istate)
3282	{
3283	int i, not_shared = 0;
3284	int max_split = repo_config_get_max_percent_split_change(the_repository);
3285
3286	switch (max_split) {
3287	case -1:
3288	/* not or badly configured: use the default value */
3289	max_split = default_max_percent_split_change;
3290	break;
3291	case 0:
3292	return 1; /* 0% means always write a new shared index */
3293	case 100:
3294	return 0; /* 100% means never write a new shared index */
3295	default:
3296	break; /* just use the configured value */
3297	}
3298
3299	/* Count not shared entries */
3300	for (i = 0; i < istate->cache_nr; i++) {
3301	struct cache_entry *ce = istate->cache[i];
3302	if (!ce->index)
3303	not_shared++;
3304	}
3305
3306	return (int64_t)istate->cache_nr * max_split < (int64_t)not_shared * 100;
3307	}
3308
3309	int write_locked_index(struct index_state istate, struct lock_file lock,
3310	unsigned flags)
3311	{
3312	int new_shared_index, ret, test_split_index_env;
3313	struct split_index *si = istate->split_index;
3314
3315	if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0) &&
3316	cache_tree_verify(the_repository, istate) < 0)
3317	return -1;
3318
3319	if ((flags & SKIP_IF_UNCHANGED) && !istate->cache_changed) {
3320	if (flags & COMMIT_LOCK)
3321	rollback_lock_file(lock);
3322	return 0;
3323	}
3324
3325	if (istate->fsmonitor_last_update)
3326	fill_fsmonitor_bitmap(istate);
3327
3328	test_split_index_env = git_env_bool("GIT_TEST_SPLIT_INDEX", 0);
3329
3330	if ((!si && !test_split_index_env) \|\|
3331	alternate_index_output \|\|
3332	(istate->cache_changed & ~EXTMASK)) {
3333	ret = do_write_locked_index(istate, lock, flags,
3334	~WRITE_SPLIT_INDEX_EXTENSION);
3335	goto out;
3336	}
3337
3338	if (test_split_index_env) {
3339	if (!si) {
3340	si = init_split_index(istate);
3341	istate->cache_changed \|= SPLIT_INDEX_ORDERED;
3342	} else {
3343	int v = si->base_oid.hash[0];
3344	if ((v & 15) < 6)
3345	istate->cache_changed \|= SPLIT_INDEX_ORDERED;
3346	}
3347	}
3348	if (too_many_not_shared_entries(istate))
3349	istate->cache_changed \|= SPLIT_INDEX_ORDERED;
3350
3351	new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED;
3352
3353	if (new_shared_index) {
3354	struct tempfile *temp;
3355	int saved_errno;
3356	char *path;
3357
3358	/* Same initial permissions as the main .git/index file */
3359	path = repo_git_path(the_repository, "sharedindex_XXXXXX");
3360	temp = mks_tempfile_sm(path, 0, 0666);
3361	free(path);
3362	if (!temp) {
3363	ret = do_write_locked_index(istate, lock, flags,
3364	~WRITE_SPLIT_INDEX_EXTENSION);
3365	goto out;
3366	}
3367	ret = write_shared_index(istate, &temp, flags);
3368
3369	saved_errno = errno;
3370	if (is_tempfile_active(temp))
3371	delete_tempfile(&temp);
3372	errno = saved_errno;
3373
3374	if (ret)
3375	goto out;
3376	}
3377
3378	ret = write_split_index(istate, lock, flags);
3379
3380	/* Freshen the shared index only if the split-index was written */
3381	if (!ret && !new_shared_index && !is_null_oid(&si->base_oid)) {
3382	char *shared_index = repo_git_path(the_repository, "sharedindex.%s",
3383	oid_to_hex(&si->base_oid));
3384	freshen_shared_index(shared_index, 1);
3385	free(shared_index);
3386	}
3387
3388	out:
3389	if (flags & COMMIT_LOCK)
3390	rollback_lock_file(lock);
3391	return ret;
3392	}
3393
3394	/*
3395	* Read the index file that is potentially unmerged into given
3396	* index_state, dropping any unmerged entries to stage #0 (potentially
3397	* resulting in a path appearing as both a file and a directory in the
3398	* index; the caller is responsible to clear out the extra entries
3399	* before writing the index to a tree). Returns true if the index is
3400	* unmerged. Callers who want to refuse to work from an unmerged
3401	* state can call this and check its return value, instead of calling
3402	* read_cache().
3403	*/
3404	int repo_read_index_unmerged(struct repository *repo)
3405	{
3406	repo_read_index(repo);
3407	return index_state_unmerged_to_stage0(repo->index);
3408	}
3409
3410	int index_state_unmerged_to_stage0(struct index_state *istate)
3411	{
3412	int unmerged = 0;
3413
3414	for (unsigned int i = 0; i < istate->cache_nr; i++) {
3415	struct cache_entry *ce = istate->cache[i];
3416	struct cache_entry *new_ce;
3417	int len;
3418
3419	if (!ce_stage(ce))
3420	continue;
3421	unmerged = 1;
3422	len = ce_namelen(ce);
3423	new_ce = make_empty_cache_entry(istate, len);
3424	memcpy(new_ce->name, ce->name, len);
3425	new_ce->ce_flags = create_ce_flags(0) \| CE_CONFLICTED;
3426	new_ce->ce_namelen = len;
3427	new_ce->ce_mode = ce->ce_mode;
3428	if (add_index_entry(istate, new_ce, ADD_CACHE_SKIP_DFCHECK))
3429	return error(_("%s: cannot drop to stage #0"),
3430	new_ce->name);
3431	}
3432	return unmerged;
3433	}
3434
3435	/*
3436	* Returns 1 if the path is an "other" path with respect to
3437	* the index; that is, the path is not mentioned in the index at all,
3438	* either as a file, a directory with some files in the index,
3439	* or as an unmerged entry.
3440	*
3441	* We helpfully remove a trailing "/" from directories so that
3442	* the output of read_directory can be used as-is.
3443	*/
3444	int index_name_is_other(struct index_state istate, const char name,
3445	int namelen)
3446	{
3447	int pos;
3448	if (namelen && name[namelen - 1] == '/')
3449	namelen--;
3450	pos = index_name_pos(istate, name, namelen);
3451	if (0 <= pos)
3452	return 0; /* exact match */
3453	pos = -pos - 1;
3454	if (pos < istate->cache_nr) {
3455	struct cache_entry *ce = istate->cache[pos];
3456	if (ce_namelen(ce) == namelen &&
3457	!memcmp(ce->name, name, namelen))
3458	return 0; /* Yup, this one exists unmerged */
3459	}
3460	return 1;
3461	}
3462
3463	void read_blob_data_from_index(struct index_state istate,
3464	const char path, unsigned long size)
3465	{
3466	int pos, len;
3467	size_t sz;
3468	enum object_type type;
3469	void *data;
3470
3471	len = strlen(path);
3472	pos = index_name_pos(istate, path, len);
3473	if (pos < 0) {
3474	/*
3475	* We might be in the middle of a merge, in which
3476	* case we would read stage #2 (ours).
3477	*/
3478	int i;
3479	for (i = -pos - 1;
3480	(pos < 0 && i < istate->cache_nr &&
3481	!strcmp(istate->cache[i]->name, path));
3482	i++)
3483	if (ce_stage(istate->cache[i]) == 2)
3484	pos = i;
3485	}
3486	if (pos < 0)
3487	return NULL;
3488	data = odb_read_object(the_repository->objects, &istate->cache[pos]->oid,
3489	&type, &sz);
3490	if (!data \|\| type != OBJ_BLOB) {
3491	free(data);
3492	return NULL;
3493	}
3494	if (size)
3495	*size = cast_size_t_to_ulong(sz);
3496	return data;
3497	}
3498
3499	void move_index_extensions(struct index_state dst, struct index_state src)
3500	{
3501	dst->untracked = src->untracked;
3502	src->untracked = NULL;
3503	dst->cache_tree = src->cache_tree;
3504	src->cache_tree = NULL;
3505	}
3506
3507	struct cache_entry dup_cache_entry(const struct cache_entry ce,
3508	struct index_state *istate)
3509	{
3510	unsigned int size = ce_size(ce);
3511	int mem_pool_allocated;
3512	struct cache_entry *new_entry = make_empty_cache_entry(istate, ce_namelen(ce));
3513	mem_pool_allocated = new_entry->mem_pool_allocated;
3514
3515	memcpy(new_entry, ce, size);
3516	new_entry->mem_pool_allocated = mem_pool_allocated;
3517	return new_entry;
3518	}
3519
3520	void discard_cache_entry(struct cache_entry *ce)
3521	{
3522	if (ce && should_validate_cache_entries())
3523	memset(ce, 0xCD, cache_entry_size(ce->ce_namelen));
3524
3525	if (ce && ce->mem_pool_allocated)
3526	return;
3527
3528	free(ce);
3529	}
3530
3531	int should_validate_cache_entries(void)
3532	{
3533	static int validate_index_cache_entries = -1;
3534
3535	if (validate_index_cache_entries < 0) {
3536	if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))
3537	validate_index_cache_entries = 1;
3538	else
3539	validate_index_cache_entries = 0;
3540	}
3541
3542	return validate_index_cache_entries;
3543	}
3544
3545	#define EOIE_SIZE (4 + GIT_SHA1_RAWSZ) /* <4-byte offset> + <20-byte hash> */
3546	#define EOIE_SIZE_WITH_HEADER (4 + 4 + EOIE_SIZE) /* <4-byte signature> + <4-byte length> + EOIE_SIZE */
3547
3548	static size_t read_eoie_extension(const char *mmap, size_t mmap_size)
3549	{
3550	/*
3551	* The end of index entries (EOIE) extension is guaranteed to be last
3552	* so that it can be found by scanning backwards from the EOF.
3553	*
3554	* "EOIE"
3555	* <4-byte length>
3556	* <4-byte offset>
3557	* <20-byte hash>
3558	*/
3559	const char index, eoie;
3560	uint32_t extsize;
3561	size_t offset, src_offset;
3562	unsigned char hash[GIT_MAX_RAWSZ];
3563	struct git_hash_ctx c;
3564
3565	/* ensure we have an index big enough to contain an EOIE extension */
3566	if (mmap_size < sizeof(struct cache_header) + EOIE_SIZE_WITH_HEADER + the_hash_algo->rawsz)
3567	return 0;
3568
3569	/* validate the extension signature */
3570	index = eoie = mmap + mmap_size - EOIE_SIZE_WITH_HEADER - the_hash_algo->rawsz;
3571	if (CACHE_EXT(index) != CACHE_EXT_ENDOFINDEXENTRIES)
3572	return 0;
3573	index += sizeof(uint32_t);
3574
3575	/* validate the extension size */
3576	extsize = get_be32(index);
3577	if (extsize != EOIE_SIZE)
3578	return 0;
3579	index += sizeof(uint32_t);
3580
3581	/*
3582	* Validate the offset we're going to look for the first extension
3583	* signature is after the index header and before the eoie extension.
3584	*/
3585	offset = get_be32(index);
3586	if (mmap + offset < mmap + sizeof(struct cache_header))
3587	return 0;
3588	if (mmap + offset >= eoie)
3589	return 0;
3590	index += sizeof(uint32_t);
3591
3592	/*
3593	* The hash is computed over extension types and their sizes (but not
3594	* their contents). E.g. if we have "TREE" extension that is N-bytes
3595	* long, "REUC" extension that is M-bytes long, followed by "EOIE",
3596	* then the hash would be:
3597	*
3598	* SHA-1("TREE" + <binary representation of N> +
3599	* "REUC" + <binary representation of M>)
3600	*/
3601	src_offset = offset;
3602	git_hash_init(&c, the_hash_algo);
3603	while (src_offset < mmap_size - the_hash_algo->rawsz - EOIE_SIZE_WITH_HEADER) {
3604	/* After an array of active_nr index entries,
3605	* there can be arbitrary number of extended
3606	* sections, each of which is prefixed with
3607	* extension name (4-byte) and section length
3608	* in 4-byte network byte order.
3609	*/
3610	uint32_t extsize;
3611	memcpy(&extsize, mmap + src_offset + 4, 4);
3612	extsize = ntohl(extsize);
3613
3614	/* verify the extension size isn't so large it will wrap around */
3615	if (src_offset + 8 + extsize < src_offset)
3616	return 0;
3617
3618	git_hash_update(&c, mmap + src_offset, 8);
3619
3620	src_offset += 8;
3621	src_offset += extsize;
3622	}
3623	git_hash_final(hash, &c);
3624	if (!hasheq(hash, (const unsigned char *)index, the_repository->hash_algo))
3625	return 0;
3626
3627	/* Validate that the extension offsets returned us back to the eoie extension. */
3628	if (src_offset != mmap_size - the_hash_algo->rawsz - EOIE_SIZE_WITH_HEADER)
3629	return 0;
3630
3631	return offset;
3632	}
3633
3634	static void write_eoie_extension(struct strbuf sb, struct git_hash_ctx eoie_context, size_t offset)
3635	{
3636	uint32_t buffer;
3637	unsigned char hash[GIT_MAX_RAWSZ];
3638
3639	/* offset */
3640	put_be32(&buffer, offset);
3641	strbuf_add(sb, &buffer, sizeof(uint32_t));
3642
3643	/* hash */
3644	git_hash_final(hash, eoie_context);
3645	strbuf_add(sb, hash, the_hash_algo->rawsz);
3646	}
3647
3648	#define IEOT_VERSION (1)
3649
3650	static struct index_entry_offset_table read_ieot_extension(const char mmap, size_t mmap_size, size_t offset)
3651	{
3652	const char *index = NULL;
3653	uint32_t extsize, ext_version;
3654	struct index_entry_offset_table *ieot;
3655	int i, nr;
3656
3657	/* find the IEOT extension */
3658	if (!offset)
3659	return NULL;
3660	while (offset <= mmap_size - the_hash_algo->rawsz - 8) {
3661	extsize = get_be32(mmap + offset + 4);
3662	if (CACHE_EXT((mmap + offset)) == CACHE_EXT_INDEXENTRYOFFSETTABLE) {
3663	index = mmap + offset + 4 + 4;
3664	break;
3665	}
3666	offset += 8;
3667	offset += extsize;
3668	}
3669	if (!index)
3670	return NULL;
3671
3672	/* validate the version is IEOT_VERSION */
3673	ext_version = get_be32(index);
3674	if (ext_version != IEOT_VERSION) {
3675	error("invalid IEOT version %d", ext_version);
3676	return NULL;
3677	}
3678	index += sizeof(uint32_t);
3679
3680	/* extension size - version bytes / bytes per entry */
3681	nr = (extsize - sizeof(uint32_t)) / (sizeof(uint32_t) + sizeof(uint32_t));
3682	if (!nr) {
3683	error("invalid number of IEOT entries %d", nr);
3684	return NULL;
3685	}
3686	ieot = xmalloc(sizeof(struct index_entry_offset_table)
3687	+ (nr * sizeof(struct index_entry_offset)));
3688	ieot->nr = nr;
3689	for (i = 0; i < nr; i++) {
3690	ieot->entries[i].offset = get_be32(index);
3691	index += sizeof(uint32_t);
3692	ieot->entries[i].nr = get_be32(index);
3693	index += sizeof(uint32_t);
3694	}
3695
3696	return ieot;
3697	}
3698
3699	static void write_ieot_extension(struct strbuf sb, struct index_entry_offset_table ieot)
3700	{
3701	uint32_t buffer;
3702	int i;
3703
3704	/* version */
3705	put_be32(&buffer, IEOT_VERSION);
3706	strbuf_add(sb, &buffer, sizeof(uint32_t));
3707
3708	/* ieot */
3709	for (i = 0; i < ieot->nr; i++) {
3710
3711	/* offset */
3712	put_be32(&buffer, ieot->entries[i].offset);
3713	strbuf_add(sb, &buffer, sizeof(uint32_t));
3714
3715	/* count */
3716	put_be32(&buffer, ieot->entries[i].nr);
3717	strbuf_add(sb, &buffer, sizeof(uint32_t));
3718	}
3719	}
3720
3721	void prefetch_cache_entries(const struct index_state *istate,
3722	must_prefetch_predicate must_prefetch)
3723	{
3724	int i;
3725	struct oid_array to_fetch = OID_ARRAY_INIT;
3726
3727	for (i = 0; i < istate->cache_nr; i++) {
3728	struct cache_entry *ce = istate->cache[i];
3729
3730	if (S_ISGITLINK(ce->ce_mode) \|\| !must_prefetch(ce))
3731	continue;
3732	if (!odb_read_object_info_extended(the_repository->objects,
3733	&ce->oid, NULL,
3734	OBJECT_INFO_FOR_PREFETCH))
3735	continue;
3736	oid_array_append(&to_fetch, &ce->oid);
3737	}
3738	promisor_remote_get_direct(the_repository,
3739	to_fetch.oid, to_fetch.nr);
3740	oid_array_clear(&to_fetch);
3741	}
3742
3743	static int read_one_entry_opt(struct index_state *istate,
3744	const struct object_id *oid,
3745	struct strbuf *base,
3746	const char *pathname,
3747	unsigned mode, int opt)
3748	{
3749	int len;
3750	struct cache_entry *ce;
3751
3752	if (S_ISDIR(mode))
3753	return READ_TREE_RECURSIVE;
3754
3755	len = strlen(pathname);
3756	ce = make_empty_cache_entry(istate, base->len + len);
3757
3758	ce->ce_mode = create_ce_mode(mode);
3759	ce->ce_flags = create_ce_flags(1);
3760	ce->ce_namelen = base->len + len;
3761	memcpy(ce->name, base->buf, base->len);
3762	memcpy(ce->name + base->len, pathname, len+1);
3763	oidcpy(&ce->oid, oid);
3764	return add_index_entry(istate, ce, opt);
3765	}
3766
3767	static int read_one_entry(const struct object_id oid, struct strbuf base,
3768	const char *pathname, unsigned mode,
3769	void *context)
3770	{
3771	struct index_state *istate = context;
3772	return read_one_entry_opt(istate, oid, base, pathname,
3773	mode,
3774	ADD_CACHE_OK_TO_ADD\|ADD_CACHE_SKIP_DFCHECK);
3775	}
3776
3777	/*
3778	* This is used when the caller knows there is no existing entries at
3779	* the stage that will conflict with the entry being added.
3780	*/
3781	static int read_one_entry_quick(const struct object_id oid, struct strbuf base,
3782	const char *pathname, unsigned mode,
3783	void *context)
3784	{
3785	struct index_state *istate = context;
3786	return read_one_entry_opt(istate, oid, base, pathname,
3787	mode, ADD_CACHE_JUST_APPEND);
3788	}
3789
3790	/*
3791	* Read the tree specified with --with-tree option
3792	* (typically, HEAD) into stage #1 and then
3793	* squash them down to stage #0. This is used for
3794	* --error-unmatch to list and check the path patterns
3795	* that were given from the command line. We are not
3796	* going to write this index out.
3797	*/
3798	void overlay_tree_on_index(struct index_state *istate,
3799	const char tree_name, const char prefix)
3800	{
3801	struct tree *tree;
3802	struct object_id oid;
3803	struct pathspec pathspec;
3804	struct cache_entry *last_stage0 = NULL;
3805	int i;
3806	read_tree_fn_t fn = NULL;
3807	int err;
3808
3809	if (repo_get_oid(the_repository, tree_name, &oid))
3810	die("tree-ish %s not found.", tree_name);
3811	tree = repo_parse_tree_indirect(the_repository, &oid);
3812	if (!tree)
3813	die("bad tree-ish %s", tree_name);
3814
3815	/* Hoist the unmerged entries up to stage #3 to make room */
3816	/* TODO: audit for interaction with sparse-index. */
3817	ensure_full_index(istate);
3818	for (i = 0; i < istate->cache_nr; i++) {
3819	struct cache_entry *ce = istate->cache[i];
3820	if (!ce_stage(ce))
3821	continue;
3822	ce->ce_flags \|= CE_STAGEMASK;
3823	}
3824
3825	if (prefix) {
3826	static const char *(matchbuf[1]);
3827	matchbuf[0] = NULL;
3828	parse_pathspec(&pathspec, PATHSPEC_ALL_MAGIC,
3829	PATHSPEC_PREFER_CWD, prefix, matchbuf);
3830	} else
3831	memset(&pathspec, 0, sizeof(pathspec));
3832
3833	/*
3834	* See if we have cache entry at the stage. If so,
3835	* do it the original slow way, otherwise, append and then
3836	* sort at the end.
3837	*/
3838	for (i = 0; !fn && i < istate->cache_nr; i++) {
3839	const struct cache_entry *ce = istate->cache[i];
3840	if (ce_stage(ce) == 1)
3841	fn = read_one_entry;
3842	}
3843
3844	if (!fn)
3845	fn = read_one_entry_quick;
3846	err = read_tree(the_repository, tree, &pathspec, fn, istate);
3847	clear_pathspec(&pathspec);
3848	if (err)
3849	die("unable to read tree entries %s", tree_name);
3850
3851	/*
3852	* Sort the cache entry -- we need to nuke the cache tree, though.
3853	*/
3854	if (fn == read_one_entry_quick) {
3855	cache_tree_free(&istate->cache_tree);
3856	QSORT(istate->cache, istate->cache_nr, cmp_cache_name_compare);
3857	}
3858
3859	for (i = 0; i < istate->cache_nr; i++) {
3860	struct cache_entry *ce = istate->cache[i];
3861	switch (ce_stage(ce)) {
3862	case 0:
3863	last_stage0 = ce;
3864	/* fallthru */
3865	default:
3866	continue;
3867	case 1:
3868	/*
3869	* If there is stage #0 entry for this, we do not
3870	* need to show it. We use CE_UPDATE bit to mark
3871	* such an entry.
3872	*/
3873	if (last_stage0 &&
3874	!strcmp(last_stage0->name, ce->name))
3875	ce->ce_flags \|= CE_UPDATE;
3876	}
3877	}
3878	}
3879
3880	struct update_callback_data {
3881	struct index_state *index;
3882	struct repository *repo;
3883	struct pathspec *pathspec;
3884	int include_sparse;
3885	int flags;
3886	int add_errors;
3887	int ignored_too;
3888	};
3889
3890	static int fix_unmerged_status(struct diff_filepair *p,
3891	struct update_callback_data *data)
3892	{
3893	if (p->status != DIFF_STATUS_UNMERGED)
3894	return p->status;
3895	if (!(data->flags & ADD_CACHE_IGNORE_REMOVAL) && !p->two->mode)
3896	/*
3897	* This is not an explicit add request, and the
3898	* path is missing from the working tree (deleted)
3899	*/
3900	return DIFF_STATUS_DELETED;
3901	else
3902	/*
3903	* Either an explicit add request, or path exists
3904	* in the working tree. An attempt to explicitly
3905	* add a path that does not exist in the working tree
3906	* will be caught as an error by the caller immediately.
3907	*/
3908	return DIFF_STATUS_MODIFIED;
3909	}
3910
3911	static int skip_submodule(const char *path,
3912	struct repository *repo,
3913	struct pathspec *pathspec,
3914	int ignored_too)
3915	{
3916	struct stat st;
3917	const struct submodule *sub;
3918	int pathspec_matches = 0;
3919	int ps_i;
3920	char *norm_pathspec = NULL;
3921
3922	/* Only consider if path is a directory */
3923	if (lstat(path, &st) \|\| !S_ISDIR(st.st_mode))
3924	return 0;
3925
3926	/* Check if it's a submodule with ignore=all */
3927	sub = submodule_from_path(repo, null_oid(the_hash_algo), path);
3928	if (!sub \|\| !sub->name \|\| !sub->ignore \|\| strcmp(sub->ignore, "all"))
3929	return 0;
3930
3931	trace_printf("ignore=all: %s\n", path);
3932	trace_printf("pathspec %s\n", (pathspec && pathspec->nr)
3933	? "has pathspec"
3934	: "no pathspec");
3935
3936	/* Check if submodule path is explicitly mentioned in pathspec */
3937	if (pathspec) {
3938	for (ps_i = 0; ps_i < pathspec->nr; ps_i++) {
3939	const char *m = pathspec->items[ps_i].match;
3940	if (!m)
3941	continue;
3942	norm_pathspec = xstrdup(m);
3943	strip_dir_trailing_slashes(norm_pathspec);
3944	if (!strcmp(path, norm_pathspec)) {
3945	pathspec_matches = 1;
3946	FREE_AND_NULL(norm_pathspec);
3947	break;
3948	}
3949	FREE_AND_NULL(norm_pathspec);
3950	}
3951	}
3952
3953	/* If explicitly matched and forced, allow adding */
3954	if (pathspec_matches) {
3955	if (ignored_too && ignored_too > 0) {
3956	trace_printf("Add submodule due to --force: %s\n", path);
3957	return 0;
3958	} else {
3959	advise_if_enabled(ADVICE_ADD_IGNORED_FILE,
3960	_("Skipping submodule due to ignore=all: %s\n"
3961	"Use --force if you really want to add the submodule."), path);
3962	return 1;
3963	}
3964	}
3965
3966	/* No explicit pathspec match -> skip silently */
3967	trace_printf("Pathspec to submodule does not match explicitly: %s\n", path);
3968	return 1;
3969	}
3970
3971	static void update_callback(struct diff_queue_struct *q,
3972	struct diff_options opt UNUSED, void cbdata)
3973	{
3974	int i;
3975	struct update_callback_data *data = cbdata;
3976
3977	for (i = 0; i < q->nr; i++) {
3978	struct diff_filepair *p = q->queue[i];
3979	const char *path = p->one->path;
3980
3981	if (!data->include_sparse &&
3982	!path_in_sparse_checkout(path, data->index))
3983	continue;
3984
3985	switch (fix_unmerged_status(p, data)) {
3986	default:
3987	die(_("unexpected diff status %c"), p->status);
3988	case DIFF_STATUS_MODIFIED:
3989	case DIFF_STATUS_TYPE_CHANGED:
3990	if (skip_submodule(path, data->repo,
3991	data->pathspec,
3992	data->ignored_too))
3993	continue;
3994
3995	if (add_file_to_index(data->index, path, data->flags)) {
3996	if (!(data->flags & ADD_CACHE_IGNORE_ERRORS))
3997	die(_("updating files failed"));
3998	data->add_errors++;
3999	}
4000	break;
4001	case DIFF_STATUS_DELETED:
4002	if (data->flags & ADD_CACHE_IGNORE_REMOVAL)
4003	break;
4004	if (!(data->flags & ADD_CACHE_PRETEND))
4005	remove_file_from_index(data->index, path);
4006	if (data->flags & (ADD_CACHE_PRETEND\|ADD_CACHE_VERBOSE))
4007	printf(_("remove '%s'\n"), path);
4008	break;
4009	}
4010	}
4011	}
4012
4013	int add_files_to_cache(struct repository repo, const char prefix,
4014	const struct pathspec pathspec, char ps_matched,
4015	int include_sparse, int flags, int ignored_too )
4016	{
4017	int inflight = !!repo->objects->transaction;
4018	struct odb_transaction *transaction;
4019	struct update_callback_data data;
4020	struct rev_info rev;
4021
4022	memset(&data, 0, sizeof(data));
4023	data.index = repo->index;
4024	data.include_sparse = include_sparse;
4025	data.flags = flags;
4026	data.repo = repo;
4027	data.ignored_too = ignored_too;
4028	data.pathspec = (struct pathspec *)pathspec;
4029
4030	repo_init_revisions(repo, &rev, prefix);
4031	setup_revisions(0, NULL, &rev, NULL);
4032	if (pathspec) {
4033	copy_pathspec(&rev.prune_data, pathspec);
4034	rev.ps_matched = ps_matched;
4035	}
4036	rev.diffopt.output_format = DIFF_FORMAT_CALLBACK;
4037	rev.diffopt.format_callback = update_callback;
4038	rev.diffopt.format_callback_data = &data;
4039	rev.diffopt.flags.override_submodule_config = 1;
4040	rev.diffopt.detect_rename = 0; /* staging worktree changes does not need renames */
4041	rev.max_count = 0; /* do not compare unmerged paths with stage #2 */
4042
4043	/*
4044	* Use an ODB transaction to optimize adding multiple objects.
4045	* This function is invoked from commands other than 'add', which
4046	* may not have their own transaction active.
4047	*/
4048	if (!inflight)
4049	odb_transaction_begin_or_die(repo->objects, &transaction, 0);
4050	run_diff_files(&rev, DIFF_RACY_IS_MODIFIED);
4051	if (!inflight)
4052	odb_transaction_commit(transaction);
4053
4054	release_revisions(&rev);
4055	return !!data.add_errors;
4056	}