read-cache.c · leroysheep/rag-system

seen

bisect jch maint master next seen test todo

c 4,069 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	int remove_file_from_index_with_flags(struct index_state *istate,
640	const char *path,
641	int flags)
642	{
643	int verbose = flags & (ADD_CACHE_VERBOSE \| ADD_CACHE_PRETEND);
644	int pretend = flags & ADD_CACHE_PRETEND;
645
646	if (verbose)
647	printf(_("remove '%s'\n"), path);
648	if (pretend)
649	return 0;
650	return remove_file_from_index(istate, path);
651	}
652
653	static int compare_name(struct cache_entry ce, const char path, int namelen)
654	{
655	return namelen != ce_namelen(ce) \|\| memcmp(path, ce->name, namelen);
656	}
657
658	static int index_name_pos_also_unmerged(struct index_state *istate,
659	const char *path, int namelen)
660	{
661	int pos = index_name_pos(istate, path, namelen);
662	struct cache_entry *ce;
663
664	if (pos >= 0)
665	return pos;
666
667	/* maybe unmerged? */
668	pos = -1 - pos;
669	if (pos >= istate->cache_nr \|\|
670	compare_name((ce = istate->cache[pos]), path, namelen))
671	return -1;
672
673	/* order of preference: stage 2, 1, 3 */
674	if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
675	ce_stage((ce = istate->cache[pos + 1])) == 2 &&
676	!compare_name(ce, path, namelen))
677	pos++;
678	return pos;
679	}
680
681	static int different_name(struct cache_entry ce, struct cache_entry alias)
682	{
683	int len = ce_namelen(ce);
684	return ce_namelen(alias) != len \|\| memcmp(ce->name, alias->name, len);
685	}
686
687	/*
688	* If we add a filename that aliases in the cache, we will use the
689	* name that we already have - but we don't want to update the same
690	* alias twice, because that implies that there were actually two
691	* different files with aliasing names!
692	*
693	* So we use the CE_ADDED flag to verify that the alias was an old
694	* one before we accept it as
695	*/
696	static struct cache_entry create_alias_ce(struct index_state istate,
697	struct cache_entry *ce,
698	struct cache_entry *alias)
699	{
700	int len;
701	struct cache_entry *new_entry;
702
703	if (alias->ce_flags & CE_ADDED)
704	die(_("will not add file alias '%s' ('%s' already exists in index)"),
705	ce->name, alias->name);
706
707	/* Ok, create the new entry using the name of the existing alias */
708	len = ce_namelen(alias);
709	new_entry = make_empty_cache_entry(istate, len);
710	memcpy(new_entry->name, alias->name, len);
711	copy_cache_entry(new_entry, ce);
712	save_or_free_index_entry(istate, ce);
713	return new_entry;
714	}
715
716	void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
717	{
718	struct object_id oid;
719	if (odb_write_object(the_repository->objects, "", 0, OBJ_BLOB, &oid))
720	die(_("cannot create an empty blob in the object database"));
721	oidcpy(&ce->oid, &oid);
722	}
723
724	int add_to_index(struct index_state istate, const char path, struct stat *st, int flags)
725	{
726	int namelen, was_same;
727	mode_t st_mode = st->st_mode;
728	struct cache_entry ce, alias = NULL;
729	unsigned ce_option = CE_MATCH_IGNORE_VALID\|CE_MATCH_IGNORE_SKIP_WORKTREE\|CE_MATCH_RACY_IS_DIRTY;
730	int verbose = flags & (ADD_CACHE_VERBOSE \| ADD_CACHE_PRETEND);
731	int pretend = flags & ADD_CACHE_PRETEND;
732	int intent_only = flags & ADD_CACHE_INTENT;
733	int add_option = (ADD_CACHE_OK_TO_ADD\|ADD_CACHE_OK_TO_REPLACE\|
734	(intent_only ? ADD_CACHE_NEW_ONLY : 0));
735	unsigned hash_flags = pretend ? 0 : INDEX_WRITE_OBJECT;
736
737	if (flags & ADD_CACHE_RENORMALIZE)
738	hash_flags \|= INDEX_RENORMALIZE;
739
740	if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
741	return error(_("%s: can only add regular files, symbolic links or git-directories"), path);
742
743	namelen = strlen(path);
744	if (S_ISDIR(st_mode)) {
745	while (namelen && path[namelen-1] == '/')
746	namelen--;
747	}
748	ce = make_empty_cache_entry(istate, namelen);
749	memcpy(ce->name, path, namelen);
750	ce->ce_namelen = namelen;
751	if (!intent_only)
752	fill_stat_cache_info(istate, ce, st);
753	else
754	ce->ce_flags \|= CE_INTENT_TO_ADD;
755
756
757	if (repo_trust_executable_bit(istate->repo) &&
758	repo_has_symlinks(istate->repo)) {
759	ce->ce_mode = create_ce_mode(st_mode);
760	} else {
761	/* If there is an existing entry, pick the mode bits and type
762	* from it, otherwise assume unexecutable regular file.
763	*/
764	struct cache_entry *ent;
765	int pos = index_name_pos_also_unmerged(istate, path, namelen);
766
767	ent = (0 <= pos) ? istate->cache[pos] : NULL;
768	ce->ce_mode = ce_mode_from_stat(istate->repo, ent, st_mode);
769	}
770
771	/* When core.ignorecase=true, determine if a directory of the same name but differing
772	* case already exists within the Git repository. If it does, ensure the directory
773	* case of the file being added to the repository matches (is folded into) the existing
774	* entry's directory case.
775	*/
776	if (repo_ignore_case(the_repository)) {
777	adjust_dirname_case(istate, ce->name);
778	}
779	if (!(flags & ADD_CACHE_RENORMALIZE)) {
780	alias = index_file_exists(istate, ce->name,
781	ce_namelen(ce), repo_ignore_case(the_repository));
782	if (alias &&
783	!ce_stage(alias) &&
784	!ie_match_stat(istate, alias, st, ce_option)) {
785	/* Nothing changed, really */
786	if (!S_ISGITLINK(alias->ce_mode))
787	ce_mark_uptodate(alias);
788	alias->ce_flags \|= CE_ADDED;
789
790	discard_cache_entry(ce);
791	return 0;
792	}
793	}
794	if (!intent_only) {
795	if (index_path(istate, &ce->oid, path, st, hash_flags)) {
796	discard_cache_entry(ce);
797	return error(_("unable to index file '%s'"), path);
798	}
799	} else
800	set_object_name_for_intent_to_add_entry(ce);
801
802	if (repo_ignore_case(the_repository) && alias && different_name(ce, alias))
803	ce = create_alias_ce(istate, ce, alias);
804	ce->ce_flags \|= CE_ADDED;
805
806	/* It was suspected to be racily clean, but it turns out to be Ok */
807	was_same = (alias &&
808	!ce_stage(alias) &&
809	oideq(&alias->oid, &ce->oid) &&
810	ce->ce_mode == alias->ce_mode);
811
812	if (pretend)
813	discard_cache_entry(ce);
814	else if (add_index_entry(istate, ce, add_option)) {
815	discard_cache_entry(ce);
816	return error(_("unable to add '%s' to index"), path);
817	}
818	if (verbose && !was_same)
819	printf("add '%s'\n", path);
820	return 0;
821	}
822
823	int add_file_to_index(struct index_state istate, const char path, int flags)
824	{
825	struct stat st;
826	if (lstat(path, &st))
827	die_errno(_("unable to stat '%s'"), path);
828	return add_to_index(istate, path, &st, flags);
829	}
830
831	struct cache_entry make_empty_cache_entry(struct index_state istate, size_t len)
832	{
833	return mem_pool__ce_calloc(find_mem_pool(istate), len);
834	}
835
836	struct cache_entry *make_empty_transient_cache_entry(size_t len,
837	struct mem_pool *ce_mem_pool)
838	{
839	if (ce_mem_pool)
840	return mem_pool__ce_calloc(ce_mem_pool, len);
841	return xcalloc(1, cache_entry_size(len));
842	}
843
844	enum verify_path_result {
845	PATH_OK,
846	PATH_INVALID,
847	PATH_DIR_WITH_SEP,
848	};
849
850	static enum verify_path_result verify_path_internal(const char *, unsigned);
851
852	int verify_path(const char *path, unsigned mode)
853	{
854	return verify_path_internal(path, mode) == PATH_OK;
855	}
856
857	struct cache_entry make_cache_entry(struct index_state istate,
858	unsigned int mode,
859	const struct object_id *oid,
860	const char *path,
861	int stage,
862	unsigned int refresh_options)
863	{
864	struct cache_entry ce, ret;
865	int len;
866
867	if (verify_path_internal(path, mode) == PATH_INVALID) {
868	error(_("invalid path '%s'"), path);
869	return NULL;
870	}
871
872	len = strlen(path);
873	ce = make_empty_cache_entry(istate, len);
874
875	oidcpy(&ce->oid, oid);
876	memcpy(ce->name, path, len);
877	ce->ce_flags = create_ce_flags(stage);
878	ce->ce_namelen = len;
879	ce->ce_mode = create_ce_mode(mode);
880
881	ret = refresh_cache_entry(istate, ce, refresh_options);
882	if (ret != ce)
883	discard_cache_entry(ce);
884	return ret;
885	}
886
887	struct cache_entry *make_transient_cache_entry(unsigned int mode,
888	const struct object_id *oid,
889	const char *path,
890	int stage,
891	struct mem_pool *ce_mem_pool)
892	{
893	struct cache_entry *ce;
894	int len;
895
896	if (!verify_path(path, mode)) {
897	error(_("invalid path '%s'"), path);
898	return NULL;
899	}
900
901	len = strlen(path);
902	ce = make_empty_transient_cache_entry(len, ce_mem_pool);
903
904	oidcpy(&ce->oid, oid);
905	memcpy(ce->name, path, len);
906	ce->ce_flags = create_ce_flags(stage);
907	ce->ce_namelen = len;
908	ce->ce_mode = create_ce_mode(mode);
909
910	return ce;
911	}
912
913	/*
914	* Chmod an index entry with either +x or -x.
915	*
916	* Returns -1 if the chmod for the particular cache entry failed (if it's
917	* not a regular file), -2 if an invalid flip argument is passed in, 0
918	* otherwise.
919	*/
920	int chmod_index_entry(struct index_state istate, struct cache_entry ce,
921	char flip)
922	{
923	if (!S_ISREG(ce->ce_mode))
924	return -1;
925	switch (flip) {
926	case '+':
927	ce->ce_mode \|= 0111;
928	break;
929	case '-':
930	ce->ce_mode &= ~0111;
931	break;
932	default:
933	return -2;
934	}
935	cache_tree_invalidate_path(istate, ce->name);
936	ce->ce_flags \|= CE_UPDATE_IN_BASE;
937	mark_fsmonitor_invalid(istate, ce);
938	istate->cache_changed \|= CE_ENTRY_CHANGED;
939
940	return 0;
941	}
942
943	int ce_same_name(const struct cache_entry a, const struct cache_entry b)
944	{
945	int len = ce_namelen(a);
946	return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
947	}
948
949	/*
950	* We fundamentally don't like some paths: we don't want
951	* dot or dot-dot anywhere, and for obvious reasons don't
952	* want to recurse into ".git" either.
953	*
954	* Also, we don't want double slashes or slashes at the
955	* end that can make pathnames ambiguous.
956	*/
957	static int verify_dotfile(const char *rest, unsigned mode)
958	{
959	/*
960	* The first character was '.', but that
961	* has already been discarded, we now test
962	* the rest.
963	*/
964
965	/* "." is not allowed */
966	if (rest == '\0' \|\| is_dir_sep(rest))
967	return 0;
968
969	switch (*rest) {
970	/*
971	* ".git" followed by NUL or slash is bad. Note that we match
972	* case-insensitively here, even if ignore_case is not set.
973	* This outlaws ".GIT" everywhere out of an abundance of caution,
974	* since there's really no good reason to allow it.
975	*
976	* Once we've seen ".git", we can also find ".gitmodules", etc (also
977	* case-insensitively).
978	*/
979	case 'g':
980	case 'G':
981	if (rest[1] != 'i' && rest[1] != 'I')
982	break;
983	if (rest[2] != 't' && rest[2] != 'T')
984	break;
985	if (rest[3] == '\0' \|\| is_dir_sep(rest[3]))
986	return 0;
987	if (S_ISLNK(mode)) {
988	rest += 3;
989	if (skip_iprefix(rest, "modules", &rest) &&
990	(rest == '\0' \|\| is_dir_sep(rest)))
991	return 0;
992	}
993	break;
994	case '.':
995	if (rest[1] == '\0' \|\| is_dir_sep(rest[1]))
996	return 0;
997	}
998	return 1;
999	}
1000
1001	static enum verify_path_result verify_path_internal(const char *path,
1002	unsigned mode)
1003	{
1004	char c = 0;
1005
1006	if (has_dos_drive_prefix(path))
1007	return PATH_INVALID;
1008
1009	if (!is_valid_path(path))
1010	return PATH_INVALID;
1011
1012	goto inside;
1013	for (;;) {
1014	if (!c)
1015	return PATH_OK;
1016	if (is_dir_sep(c)) {
1017	inside:
1018	if (repo_protect_hfs(the_repository)) {
1019
1020	if (is_hfs_dotgit(path))
1021	return PATH_INVALID;
1022	if (S_ISLNK(mode)) {
1023	if (is_hfs_dotgitmodules(path))
1024	return PATH_INVALID;
1025	}
1026	}
1027	if (repo_protect_ntfs(the_repository)) {
1028	#if defined GIT_WINDOWS_NATIVE \|\| defined __CYGWIN__
1029	if (c == '\\')
1030	return PATH_INVALID;
1031	#endif
1032	if (is_ntfs_dotgit(path))
1033	return PATH_INVALID;
1034	if (S_ISLNK(mode)) {
1035	if (is_ntfs_dotgitmodules(path))
1036	return PATH_INVALID;
1037	}
1038	}
1039
1040	c = *path++;
1041	if ((c == '.' && !verify_dotfile(path, mode)) \|\|
1042	is_dir_sep(c))
1043	return PATH_INVALID;
1044	/*
1045	* allow terminating directory separators for
1046	* sparse directory entries.
1047	*/
1048	if (c == '\0')
1049	return S_ISDIR(mode) ? PATH_DIR_WITH_SEP :
1050	PATH_INVALID;
1051	} else if (c == '\\' &&
1052	repo_protect_ntfs(the_repository)) {
1053	if (is_ntfs_dotgit(path))
1054	return PATH_INVALID;
1055	if (S_ISLNK(mode)) {
1056	if (is_ntfs_dotgitmodules(path))
1057	return PATH_INVALID;
1058	}
1059	}
1060
1061	c = *path++;
1062	}
1063	}
1064
1065	/*
1066	* Do we have another file that has the beginning components being a
1067	* proper superset of the name we're trying to add?
1068	*/
1069	static int has_file_name(struct index_state *istate,
1070	const struct cache_entry *ce, int pos, int ok_to_replace)
1071	{
1072	int retval = 0;
1073	int len = ce_namelen(ce);
1074	int stage = ce_stage(ce);
1075	const char *name = ce->name;
1076
1077	while (pos < istate->cache_nr) {
1078	struct cache_entry *p = istate->cache[pos++];
1079
1080	if (len >= ce_namelen(p))
1081	break;
1082	if (memcmp(name, p->name, len))
1083	break;
1084	if (ce_stage(p) != stage)
1085	continue;
1086	if (p->name[len] != '/')
1087	continue;
1088	if (p->ce_flags & CE_REMOVE)
1089	continue;
1090	retval = -1;
1091	if (!ok_to_replace)
1092	break;
1093	remove_index_entry_at(istate, --pos);
1094	}
1095	return retval;
1096	}
1097
1098
1099	/*
1100	* Like strcmp(), but also return the offset of the first change.
1101	* If strings are equal, return the length.
1102	*/
1103	int strcmp_offset(const char s1, const char s2, size_t *first_change)
1104	{
1105	size_t k;
1106
1107	if (!first_change)
1108	return strcmp(s1, s2);
1109
1110	for (k = 0; s1[k] == s2[k]; k++)
1111	if (s1[k] == '\0')
1112	break;
1113
1114	*first_change = k;
1115	return (unsigned char)s1[k] - (unsigned char)s2[k];
1116	}
1117
1118	/*
1119	* Do we have another file with a pathname that is a proper
1120	* subset of the name we're trying to add?
1121	*
1122	* That is, is there another file in the index with a path
1123	* that matches a sub-directory in the given entry?
1124	*/
1125	static int has_dir_name(struct index_state *istate,
1126	const struct cache_entry *ce, int pos, int ok_to_replace)
1127	{
1128	int retval = 0;
1129	int stage = ce_stage(ce);
1130	const char *name = ce->name;
1131	const char *slash = name + ce_namelen(ce);
1132	size_t len_eq_last;
1133	int cmp_last = 0;
1134
1135	/*
1136	* We are frequently called during an iteration on a sorted
1137	* list of pathnames and while building a new index. Therefore,
1138	* there is a high probability that this entry will eventually
1139	* be appended to the index, rather than inserted in the middle.
1140	* If we can confirm that, we can avoid binary searches on the
1141	* components of the pathname.
1142	*
1143	* Compare the entry's full path with the last path in the index.
1144	*/
1145	if (!istate->cache_nr)
1146	return 0;
1147
1148	cmp_last = strcmp_offset(name,
1149	istate->cache[istate->cache_nr - 1]->name,
1150	&len_eq_last);
1151	if (cmp_last > 0 && name[len_eq_last] != '/')
1152	/*
1153	* The entry sorts AFTER the last one in the
1154	* index and their paths have no common prefix,
1155	* so there cannot be a F/D conflict.
1156	*/
1157	return 0;
1158
1159	for (;;) {
1160	size_t len;
1161
1162	for (;;) {
1163	if (*--slash == '/')
1164	break;
1165	if (slash <= ce->name)
1166	return retval;
1167	}
1168	len = slash - name;
1169
1170	pos = index_name_stage_pos(istate, name, len, stage, EXPAND_SPARSE);
1171	if (pos >= 0) {
1172	/*
1173	* Found one, but not so fast. This could
1174	* be a marker that says "I was here, but
1175	* I am being removed". Such an entry is
1176	* not a part of the resulting tree, and
1177	* it is Ok to have a directory at the same
1178	* path.
1179	*/
1180	if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
1181	retval = -1;
1182	if (!ok_to_replace)
1183	break;
1184	remove_index_entry_at(istate, pos);
1185	continue;
1186	}
1187	}
1188	else
1189	pos = -pos-1;
1190
1191	/*
1192	* Trivial optimization: if we find an entry that
1193	* already matches the sub-directory, then we know
1194	* we're ok, and we can exit.
1195	*/
1196	while (pos < istate->cache_nr) {
1197	struct cache_entry *p = istate->cache[pos];
1198	if ((ce_namelen(p) <= len) \|\|
1199	(p->name[len] != '/') \|\|
1200	memcmp(p->name, name, len))
1201	break; /* not our subdirectory */
1202	if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
1203	/*
1204	* p is at the same stage as our entry, and
1205	* is a subdirectory of what we are looking
1206	* at, so we cannot have conflicts at our
1207	* level or anything shorter.
1208	*/
1209	return retval;
1210	pos++;
1211	}
1212	}
1213	return retval;
1214	}
1215
1216	/* We may be in a situation where we already have path/file and path
1217	* is being added, or we already have path and path/file is being
1218	* added. Either one would result in a nonsense tree that has path
1219	* twice when git-write-tree tries to write it out. Prevent it.
1220	*
1221	* If ok-to-replace is specified, we remove the conflicting entries
1222	* from the cache so the caller should recompute the insert position.
1223	* When this happens, we return non-zero.
1224	*/
1225	static int check_file_directory_conflict(struct index_state *istate,
1226	const struct cache_entry *ce,
1227	int pos, int ok_to_replace)
1228	{
1229	int retval;
1230
1231	/*
1232	* When ce is an "I am going away" entry, we allow it to be added
1233	*/
1234	if (ce->ce_flags & CE_REMOVE)
1235	return 0;
1236
1237	/*
1238	* We check if the path is a sub-path of a subsequent pathname
1239	* first, since removing those will not change the position
1240	* in the array.
1241	*/
1242	retval = has_file_name(istate, ce, pos, ok_to_replace);
1243
1244	/*
1245	* Then check if the path might have a clashing sub-directory
1246	* before it.
1247	*/
1248	return retval + has_dir_name(istate, ce, pos, ok_to_replace);
1249	}
1250
1251	static int add_index_entry_with_check(struct index_state istate, struct cache_entry ce, int option)
1252	{
1253	int pos;
1254	int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
1255	int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
1256	int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
1257	int new_only = option & ADD_CACHE_NEW_ONLY;
1258
1259	/*
1260	* If this entry's path sorts after the last entry in the index,
1261	* we can avoid searching for it.
1262	*/
1263	if (istate->cache_nr > 0 &&
1264	strcmp(ce->name, istate->cache[istate->cache_nr - 1]->name) > 0)
1265	pos = index_pos_to_insert_pos(istate->cache_nr);
1266	else
1267	pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce), EXPAND_SPARSE);
1268
1269	/*
1270	* Cache tree path should be invalidated only after index_name_stage_pos,
1271	* in case it expands a sparse index.
1272	*/
1273	if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1274	cache_tree_invalidate_path(istate, ce->name);
1275
1276	/* existing match? Just replace it. */
1277	if (pos >= 0) {
1278	if (!new_only)
1279	replace_index_entry(istate, pos, ce);
1280	return 0;
1281	}
1282	pos = -pos-1;
1283
1284	if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1285	untracked_cache_add_to_index(istate, ce->name);
1286
1287	/*
1288	* Inserting a merged entry ("stage 0") into the index
1289	* will always replace all non-merged entries..
1290	*/
1291	if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1292	while (ce_same_name(istate->cache[pos], ce)) {
1293	ok_to_add = 1;
1294	if (!remove_index_entry_at(istate, pos))
1295	break;
1296	}
1297	}
1298
1299	if (!ok_to_add)
1300	return -1;
1301	if (verify_path_internal(ce->name, ce->ce_mode) == PATH_INVALID)
1302	return error(_("invalid path '%s'"), ce->name);
1303
1304	if (!skip_df_check &&
1305	check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
1306	if (!ok_to_replace)
1307	return error(_("'%s' appears as both a file and as a directory"),
1308	ce->name);
1309	pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce), EXPAND_SPARSE);
1310	pos = -pos-1;
1311	}
1312	return pos + 1;
1313	}
1314
1315	int add_index_entry(struct index_state istate, struct cache_entry ce, int option)
1316	{
1317	int pos;
1318
1319	if (option & ADD_CACHE_JUST_APPEND)
1320	pos = istate->cache_nr;
1321	else {
1322	int ret;
1323	ret = add_index_entry_with_check(istate, ce, option);
1324	if (ret <= 0)
1325	return ret;
1326	pos = ret - 1;
1327	}
1328
1329	/* Make sure the array is big enough .. */
1330	ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1331
1332	/* Add it in.. */
1333	istate->cache_nr++;
1334	if (istate->cache_nr > pos + 1)
1335	MOVE_ARRAY(istate->cache + pos + 1, istate->cache + pos,
1336	istate->cache_nr - pos - 1);
1337	set_index_entry(istate, pos, ce);
1338	istate->cache_changed \|= CE_ENTRY_ADDED;
1339	return 0;
1340	}
1341
1342	/*
1343	* "refresh" does not calculate a new sha1 file or bring the
1344	* cache up-to-date for mode/content changes. But what it
1345	* _does_ do is to "re-match" the stat information of a file
1346	* with the cache, so that you can refresh the cache for a
1347	* file that hasn't been changed but where the stat entry is
1348	* out of date.
1349	*
1350	* For example, you'd want to do this after doing a "git-read-tree",
1351	* to link up the stat cache details with the proper files.
1352	*/
1353	static struct cache_entry refresh_cache_ent(struct index_state istate,
1354	struct cache_entry *ce,
1355	unsigned int options, int *err,
1356	int *changed_ret,
1357	int *t2_did_lstat,
1358	int *t2_did_scan)
1359	{
1360	struct stat st;
1361	struct cache_entry *updated;
1362	int changed;
1363	int refresh = options & CE_MATCH_REFRESH;
1364	int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1365	int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1366	int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1367	int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
1368
1369	if (!refresh \|\| ce_uptodate(ce))
1370	return ce;
1371
1372	if (!ignore_fsmonitor)
1373	refresh_fsmonitor(istate);
1374	/*
1375	* CE_VALID or CE_SKIP_WORKTREE means the user promised us
1376	* that the change to the work tree does not matter and told
1377	* us not to worry.
1378	*/
1379	if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1380	ce_mark_uptodate(ce);
1381	return ce;
1382	}
1383	if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1384	ce_mark_uptodate(ce);
1385	return ce;
1386	}
1387	if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID)) {
1388	ce_mark_uptodate(ce);
1389	return ce;
1390	}
1391
1392	if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1393	if (ignore_missing)
1394	return ce;
1395	if (err)
1396	*err = ENOENT;
1397	return NULL;
1398	}
1399
1400	if (t2_did_lstat)
1401	*t2_did_lstat = 1;
1402	if (lstat(ce->name, &st) < 0) {
1403	if (ignore_missing && errno == ENOENT)
1404	return ce;
1405	if (err)
1406	*err = errno;
1407	return NULL;
1408	}
1409
1410	changed = ie_match_stat(istate, ce, &st, options);
1411	if (changed_ret)
1412	*changed_ret = changed;
1413	if (!changed) {
1414	/*
1415	* The path is unchanged. If we were told to ignore
1416	* valid bit, then we did the actual stat check and
1417	* found that the entry is unmodified. If the entry
1418	* is not marked VALID, this is the place to mark it
1419	* valid again, under "assume unchanged" mode.
1420	*/
1421	if (ignore_valid && assume_unchanged &&
1422	!(ce->ce_flags & CE_VALID))
1423	; /* mark this one VALID again */
1424	else {
1425	/*
1426	* We do not mark the index itself "modified"
1427	* because CE_UPTODATE flag is in-core only;
1428	* we are not going to write this change out.
1429	*/
1430	if (!S_ISGITLINK(ce->ce_mode)) {
1431	ce_mark_uptodate(ce);
1432	mark_fsmonitor_valid(istate, ce);
1433	}
1434	return ce;
1435	}
1436	}
1437
1438	if (t2_did_scan)
1439	*t2_did_scan = 1;
1440	if (ie_modified(istate, ce, &st, options)) {
1441	if (err)
1442	*err = EINVAL;
1443	return NULL;
1444	}
1445
1446	updated = make_empty_cache_entry(istate, ce_namelen(ce));
1447	copy_cache_entry(updated, ce);
1448	memcpy(updated->name, ce->name, ce->ce_namelen + 1);
1449	fill_stat_cache_info(istate, updated, &st);
1450	/*
1451	* If ignore_valid is not set, we should leave CE_VALID bit
1452	* alone. Otherwise, paths marked with --no-assume-unchanged
1453	* (i.e. things to be edited) will reacquire CE_VALID bit
1454	* automatically, which is not really what we want.
1455	*/
1456	if (!ignore_valid && assume_unchanged &&
1457	!(ce->ce_flags & CE_VALID))
1458	updated->ce_flags &= ~CE_VALID;
1459
1460	/* istate->cache_changed is updated in the caller */
1461	return updated;
1462	}
1463
1464	static void show_file(const char * fmt, const char * name, int in_porcelain,
1465	int * first, const char *header_msg)
1466	{
1467	if (in_porcelain && *first && header_msg) {
1468	printf("%s\n", header_msg);
1469	*first = 0;
1470	}
1471	printf(fmt, name);
1472	}
1473
1474	int repo_refresh_and_write_index(struct repository *repo,
1475	unsigned int refresh_flags,
1476	unsigned int write_flags,
1477	int gentle,
1478	const struct pathspec *pathspec,
1479	char seen, const char header_msg)
1480	{
1481	struct lock_file lock_file = LOCK_INIT;
1482	int fd, ret = 0;
1483
1484	fd = repo_hold_locked_index(repo, &lock_file,
1485	gentle ? 0 : LOCK_REPORT_ON_ERROR);
1486	if (!gentle && fd < 0)
1487	return -1;
1488	if (refresh_index(repo->index, refresh_flags, pathspec, seen, header_msg))
1489	ret = 1;
1490	if (0 <= fd && write_locked_index(repo->index, &lock_file, COMMIT_LOCK \| write_flags))
1491	ret = -1;
1492	return ret;
1493	}
1494
1495
1496	int refresh_index(struct index_state *istate, unsigned int flags,
1497	const struct pathspec *pathspec,
1498	char seen, const char header_msg)
1499	{
1500	int i;
1501	int has_errors = 0;
1502	int really = (flags & REFRESH_REALLY) != 0;
1503	int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1504	int quiet = (flags & REFRESH_QUIET) != 0;
1505	int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1506	int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1507	int ignore_skip_worktree = (flags & REFRESH_IGNORE_SKIP_WORKTREE) != 0;
1508	int first = 1;
1509	int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1510	unsigned int options = (CE_MATCH_REFRESH \|
1511	(really ? CE_MATCH_IGNORE_VALID : 0) \|
1512	(not_new ? CE_MATCH_IGNORE_MISSING : 0));
1513	const char *modified_fmt;
1514	const char *deleted_fmt;
1515	const char *typechange_fmt;
1516	const char *added_fmt;
1517	const char *unmerged_fmt;
1518	struct progress *progress = NULL;
1519	int t2_sum_lstat = 0;
1520	int t2_sum_scan = 0;
1521
1522	if (flags & REFRESH_PROGRESS && isatty(2))
1523	progress = start_delayed_progress(the_repository,
1524	_("Refresh index"),
1525	istate->cache_nr);
1526
1527	trace_performance_enter();
1528	modified_fmt = in_porcelain ? "M\t%s\n" : "%s: needs update\n";
1529	deleted_fmt = in_porcelain ? "D\t%s\n" : "%s: needs update\n";
1530	typechange_fmt = in_porcelain ? "T\t%s\n" : "%s: needs update\n";
1531	added_fmt = in_porcelain ? "A\t%s\n" : "%s: needs update\n";
1532	unmerged_fmt = in_porcelain ? "U\t%s\n" : "%s: needs merge\n";
1533	/*
1534	* Use the multi-threaded preload_index() to refresh most of the
1535	* cache entries quickly then in the single threaded loop below,
1536	* we only have to do the special cases that are left.
1537	*/
1538	preload_index(istate, pathspec, 0);
1539	trace2_region_enter("index", "refresh", NULL);
1540
1541	for (i = 0; i < istate->cache_nr; i++) {
1542	struct cache_entry ce, new_entry;
1543	int cache_errno = 0;
1544	int changed = 0;
1545	int filtered = 0;
1546	int t2_did_lstat = 0;
1547	int t2_did_scan = 0;
1548
1549	ce = istate->cache[i];
1550	if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1551	continue;
1552	if (ignore_skip_worktree && ce_skip_worktree(ce))
1553	continue;
1554
1555	/*
1556	* If this entry is a sparse directory, then there isn't
1557	* any stat() information to update. Ignore the entry.
1558	*/
1559	if (S_ISSPARSEDIR(ce->ce_mode))
1560	continue;
1561
1562	if (pathspec && !ce_path_match(istate, ce, pathspec, seen))
1563	filtered = 1;
1564
1565	if (ce_stage(ce)) {
1566	while ((i < istate->cache_nr) &&
1567	! strcmp(istate->cache[i]->name, ce->name))
1568	i++;
1569	i--;
1570	if (allow_unmerged)
1571	continue;
1572	if (!filtered)
1573	show_file(unmerged_fmt, ce->name, in_porcelain,
1574	&first, header_msg);
1575	has_errors = 1;
1576	continue;
1577	}
1578
1579	if (filtered)
1580	continue;
1581
1582	new_entry = refresh_cache_ent(istate, ce, options,
1583	&cache_errno, &changed,
1584	&t2_did_lstat, &t2_did_scan);
1585	t2_sum_lstat += t2_did_lstat;
1586	t2_sum_scan += t2_did_scan;
1587	if (new_entry == ce)
1588	continue;
1589	display_progress(progress, i);
1590	if (!new_entry) {
1591	const char *fmt;
1592
1593	if (really && cache_errno == EINVAL) {
1594	/* If we are doing --really-refresh that
1595	* means the index is not valid anymore.
1596	*/
1597	ce->ce_flags &= ~CE_VALID;
1598	ce->ce_flags \|= CE_UPDATE_IN_BASE;
1599	mark_fsmonitor_invalid(istate, ce);
1600	istate->cache_changed \|= CE_ENTRY_CHANGED;
1601	}
1602	if (quiet)
1603	continue;
1604
1605	if (cache_errno == ENOENT)
1606	fmt = deleted_fmt;
1607	else if (ce_intent_to_add(ce))
1608	fmt = added_fmt; /* must be before other checks */
1609	else if (changed & TYPE_CHANGED)
1610	fmt = typechange_fmt;
1611	else
1612	fmt = modified_fmt;
1613	show_file(fmt,
1614	ce->name, in_porcelain, &first, header_msg);
1615	has_errors = 1;
1616	continue;
1617	}
1618
1619	replace_index_entry(istate, i, new_entry);
1620	}
1621	trace2_data_intmax("index", NULL, "refresh/sum_lstat", t2_sum_lstat);
1622	trace2_data_intmax("index", NULL, "refresh/sum_scan", t2_sum_scan);
1623	trace2_region_leave("index", "refresh", NULL);
1624	display_progress(progress, istate->cache_nr);
1625	stop_progress(&progress);
1626	trace_performance_leave("refresh index");
1627	return has_errors;
1628	}
1629
1630	struct cache_entry refresh_cache_entry(struct index_state istate,
1631	struct cache_entry *ce,
1632	unsigned int options)
1633	{
1634	return refresh_cache_ent(istate, ce, options, NULL, NULL, NULL, NULL);
1635	}
1636
1637
1638	/*****************************************************************
1639	* Index File I/O
1640	*****************************************************************/
1641
1642	#define INDEX_FORMAT_DEFAULT 3
1643
1644	static unsigned int get_index_format_default(struct repository *r)
1645	{
1646	char *envversion = getenv("GIT_INDEX_VERSION");
1647	char *endp;
1648	unsigned int version = INDEX_FORMAT_DEFAULT;
1649
1650	if (!envversion) {
1651	prepare_repo_settings(r);
1652
1653	if (r->settings.index_version >= 0)
1654	version = r->settings.index_version;
1655	if (version < INDEX_FORMAT_LB \|\| INDEX_FORMAT_UB < version) {
1656	warning(_("index.version set, but the value is invalid.\n"
1657	"Using version %i"), INDEX_FORMAT_DEFAULT);
1658	return INDEX_FORMAT_DEFAULT;
1659	}
1660	return version;
1661	}
1662
1663	version = strtoul(envversion, &endp, 10);
1664	if (*endp \|\|
1665	version < INDEX_FORMAT_LB \|\| INDEX_FORMAT_UB < version) {
1666	warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1667	"Using version %i"), INDEX_FORMAT_DEFAULT);
1668	version = INDEX_FORMAT_DEFAULT;
1669	}
1670	return version;
1671	}
1672
1673	/*
1674	* dev/ino/uid/gid/size are also just tracked to the low 32 bits
1675	* Again - this is just a (very strong in practice) heuristic that
1676	* the inode hasn't changed.
1677	*
1678	* We save the fields in big-endian order to allow using the
1679	* index file over NFS transparently.
1680	*/
1681	struct ondisk_cache_entry {
1682	struct cache_time ctime;
1683	struct cache_time mtime;
1684	uint32_t dev;
1685	uint32_t ino;
1686	uint32_t mode;
1687	uint32_t uid;
1688	uint32_t gid;
1689	uint32_t size;
1690	/*
1691	* unsigned char hash[hashsz];
1692	* uint16_t flags;
1693	* if (flags & CE_EXTENDED)
1694	* uint16_t flags2;
1695	*/
1696	unsigned char data[GIT_MAX_RAWSZ + 2 * sizeof(uint16_t)];
1697	char name[FLEX_ARRAY];
1698	};
1699
1700	/* These are only used for v3 or lower */
1701	#define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1702	#define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,data) + (len) + 8) & ~7)
1703	#define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1704	#define ondisk_data_size(flags, len) (the_hash_algo->rawsz + \
1705	((flags & CE_EXTENDED) ? 2 : 1) * sizeof(uint16_t) + len)
1706	#define ondisk_data_size_max(len) (ondisk_data_size(CE_EXTENDED, len))
1707	#define ondisk_ce_size(ce) (ondisk_cache_entry_size(ondisk_data_size((ce)->ce_flags, ce_namelen(ce))))
1708
1709	/* Allow fsck to force verification of the index checksum. */
1710	int verify_index_checksum;
1711
1712	/* Allow fsck to force verification of the cache entry order. */
1713	int verify_ce_order;
1714
1715	static int verify_hdr(const struct cache_header *hdr, unsigned long size)
1716	{
1717	struct git_hash_ctx c;
1718	unsigned char hash[GIT_MAX_RAWSZ];
1719	int hdr_version;
1720	unsigned char start, end;
1721	struct object_id oid;
1722
1723	if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1724	return error(_("bad signature 0x%08x"), hdr->hdr_signature);
1725	hdr_version = ntohl(hdr->hdr_version);
1726	if (hdr_version < INDEX_FORMAT_LB \|\| INDEX_FORMAT_UB < hdr_version)
1727	return error(_("bad index version %d"), hdr_version);
1728
1729	if (!verify_index_checksum)
1730	return 0;
1731
1732	end = (unsigned char *)hdr + size;
1733	start = end - the_hash_algo->rawsz;
1734	oidread(&oid, start, the_repository->hash_algo);
1735	if (oideq(&oid, null_oid(the_hash_algo)))
1736	return 0;
1737
1738	git_hash_init(&c, the_hash_algo);
1739	git_hash_update(&c, hdr, size - the_hash_algo->rawsz);
1740	git_hash_final(hash, &c);
1741	if (!hasheq(hash, start, the_repository->hash_algo))
1742	return error(_("bad index file sha1 signature"));
1743	return 0;
1744	}
1745
1746	static int read_index_extension(struct index_state *istate,
1747	const char ext, const char data, unsigned long sz)
1748	{
1749	switch (CACHE_EXT(ext)) {
1750	case CACHE_EXT_TREE:
1751	istate->cache_tree = cache_tree_read(data, sz);
1752	break;
1753	case CACHE_EXT_RESOLVE_UNDO:
1754	istate->resolve_undo = resolve_undo_read(data, sz, the_hash_algo);
1755	break;
1756	case CACHE_EXT_LINK:
1757	if (read_link_extension(istate, data, sz))
1758	return -1;
1759	break;
1760	case CACHE_EXT_UNTRACKED:
1761	istate->untracked = read_untracked_extension(data, sz);
1762	break;
1763	case CACHE_EXT_FSMONITOR:
1764	read_fsmonitor_extension(istate, data, sz);
1765	break;
1766	case CACHE_EXT_ENDOFINDEXENTRIES:
1767	case CACHE_EXT_INDEXENTRYOFFSETTABLE:
1768	/* already handled in do_read_index() */
1769	break;
1770	case CACHE_EXT_SPARSE_DIRECTORIES:
1771	/* no content, only an indicator */
1772	istate->sparse_index = INDEX_COLLAPSED;
1773	break;
1774	default:
1775	if (ext < 'A' \|\| 'Z' < ext)
1776	return error(_("index uses %.4s extension, which we do not understand"),
1777	ext);
1778	fprintf_ln(stderr, _("ignoring %.4s extension"), ext);
1779	break;
1780	}
1781	return 0;
1782	}
1783
1784	/*
1785	* Parses the contents of the cache entry contained within the 'ondisk' buffer
1786	* into a new incore 'cache_entry'.
1787	*
1788	* Note that 'char *ondisk' may not be aligned to a 4-byte address interval in
1789	* index v4, so we cannot cast it to 'struct ondisk_cache_entry *' and access
1790	* its members. Instead, we use the byte offsets of members within the struct to
1791	* identify where 'get_be16()', 'get_be32()', and 'oidread()' (which can all
1792	* read from an unaligned memory buffer) should read from the 'ondisk' buffer
1793	* into the corresponding incore 'cache_entry' members.
1794	*/
1795	static struct cache_entry create_from_disk(struct mem_pool ce_mem_pool,
1796	unsigned int version,
1797	const char *ondisk,
1798	unsigned long *ent_size,
1799	const struct cache_entry *previous_ce)
1800	{
1801	struct cache_entry *ce;
1802	size_t len;
1803	const char *name;
1804	const unsigned hashsz = the_hash_algo->rawsz;
1805	const char *flagsp = ondisk + offsetof(struct ondisk_cache_entry, data) + hashsz;
1806	unsigned int flags;
1807	size_t copy_len = 0;
1808	/*
1809	* Adjacent cache entries tend to share the leading paths, so it makes
1810	* sense to only store the differences in later entries. In the v4
1811	* on-disk format of the index, each on-disk cache entry stores the
1812	* number of bytes to be stripped from the end of the previous name,
1813	* and the bytes to append to the result, to come up with its name.
1814	*/
1815	int expand_name_field = version == 4;
1816
1817	/* On-disk flags are just 16 bits */
1818	flags = get_be16(flagsp);
1819	len = flags & CE_NAMEMASK;
1820
1821	if (flags & CE_EXTENDED) {
1822	int extended_flags;
1823	extended_flags = get_be16(flagsp + sizeof(uint16_t)) << 16;
1824	/* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1825	if (extended_flags & ~CE_EXTENDED_FLAGS)
1826	die(_("unknown index entry format 0x%08x"), extended_flags);
1827	flags \|= extended_flags;
1828	name = (const char )(flagsp + 2 sizeof(uint16_t));
1829	}
1830	else
1831	name = (const char *)(flagsp + sizeof(uint16_t));
1832
1833	if (expand_name_field) {
1834	const unsigned char cp = (const unsigned char )name;
1835	uint64_t strip_len, previous_len;
1836
1837	/* If we're at the beginning of a block, ignore the previous name */
1838	strip_len = decode_varint(&cp);
1839	if (previous_ce) {
1840	previous_len = previous_ce->ce_namelen;
1841	if (previous_len < strip_len)
1842	die(_("malformed name field in the index, near path '%s'"),
1843	previous_ce->name);
1844	copy_len = previous_len - strip_len;
1845	}
1846	name = (const char *)cp;
1847	}
1848
1849	if (len == CE_NAMEMASK) {
1850	len = strlen(name);
1851	if (expand_name_field)
1852	len += copy_len;
1853	}
1854
1855	ce = mem_pool__ce_alloc(ce_mem_pool, len);
1856
1857	/*
1858	* NEEDSWORK: using 'offsetof()' is cumbersome and should be replaced
1859	* with something more akin to 'load_bitmap_entries_v1()'s use of
1860	* 'read_be16'/'read_be32'. For consistency with the corresponding
1861	* ondisk entry write function ('copy_cache_entry_to_ondisk()'), this
1862	* should be done at the same time as removing references to
1863	* 'ondisk_cache_entry' there.
1864	*/
1865	ce->ce_stat_data.sd_ctime.sec = get_be32(ondisk + offsetof(struct ondisk_cache_entry, ctime)
1866	+ offsetof(struct cache_time, sec));
1867	ce->ce_stat_data.sd_mtime.sec = get_be32(ondisk + offsetof(struct ondisk_cache_entry, mtime)
1868	+ offsetof(struct cache_time, sec));
1869	ce->ce_stat_data.sd_ctime.nsec = get_be32(ondisk + offsetof(struct ondisk_cache_entry, ctime)
1870	+ offsetof(struct cache_time, nsec));
1871	ce->ce_stat_data.sd_mtime.nsec = get_be32(ondisk + offsetof(struct ondisk_cache_entry, mtime)
1872	+ offsetof(struct cache_time, nsec));
1873	ce->ce_stat_data.sd_dev = get_be32(ondisk + offsetof(struct ondisk_cache_entry, dev));
1874	ce->ce_stat_data.sd_ino = get_be32(ondisk + offsetof(struct ondisk_cache_entry, ino));
1875	ce->ce_mode = get_be32(ondisk + offsetof(struct ondisk_cache_entry, mode));
1876	ce->ce_stat_data.sd_uid = get_be32(ondisk + offsetof(struct ondisk_cache_entry, uid));
1877	ce->ce_stat_data.sd_gid = get_be32(ondisk + offsetof(struct ondisk_cache_entry, gid));
1878	ce->ce_stat_data.sd_size = get_be32(ondisk + offsetof(struct ondisk_cache_entry, size));
1879	ce->ce_flags = flags & ~CE_NAMEMASK;
1880	ce->ce_namelen = len;
1881	ce->index = 0;
1882	oidread(&ce->oid, (const unsigned char *)ondisk + offsetof(struct ondisk_cache_entry, data),
1883	the_repository->hash_algo);
1884
1885	if (expand_name_field) {
1886	if (copy_len)
1887	memcpy(ce->name, previous_ce->name, copy_len);
1888	memcpy(ce->name + copy_len, name, len + 1 - copy_len);
1889	ent_size = (name - ((char )ondisk)) + len + 1 - copy_len;
1890	} else {
1891	memcpy(ce->name, name, len + 1);
1892	*ent_size = ondisk_ce_size(ce);
1893	}
1894	return ce;
1895	}
1896
1897	static void check_ce_order(struct index_state *istate)
1898	{
1899	unsigned int i;
1900
1901	if (!verify_ce_order)
1902	return;
1903
1904	for (i = 1; i < istate->cache_nr; i++) {
1905	struct cache_entry *ce = istate->cache[i - 1];
1906	struct cache_entry *next_ce = istate->cache[i];
1907	int name_compare = strcmp(ce->name, next_ce->name);
1908
1909	if (0 < name_compare)
1910	die(_("unordered stage entries in index"));
1911	if (!name_compare) {
1912	if (!ce_stage(ce))
1913	die(_("multiple stage entries for merged file '%s'"),
1914	ce->name);
1915	if (ce_stage(ce) > ce_stage(next_ce))
1916	die(_("unordered stage entries for '%s'"),
1917	ce->name);
1918	}
1919	}
1920	}
1921
1922	static void tweak_untracked_cache(struct index_state *istate)
1923	{
1924	struct repository *r = the_repository;
1925
1926	prepare_repo_settings(r);
1927
1928	switch (r->settings.core_untracked_cache) {
1929	case UNTRACKED_CACHE_REMOVE:
1930	remove_untracked_cache(istate);
1931	break;
1932	case UNTRACKED_CACHE_WRITE:
1933	add_untracked_cache(istate);
1934	break;
1935	case UNTRACKED_CACHE_KEEP:
1936	/*
1937	* Either an explicit "core.untrackedCache=keep", the
1938	* default if "core.untrackedCache" isn't configured,
1939	* or a fallback on an unknown "core.untrackedCache"
1940	* value.
1941	*/
1942	break;
1943	}
1944	}
1945
1946	static void tweak_split_index(struct index_state *istate)
1947	{
1948	switch (repo_config_get_split_index(the_repository)) {
1949	case -1: /* unset: do nothing */
1950	break;
1951	case 0: /* false */
1952	remove_split_index(istate);
1953	break;
1954	case 1: /* true */
1955	add_split_index(istate);
1956	break;
1957	default: /* unknown value: do nothing */
1958	break;
1959	}
1960	}
1961
1962	static void post_read_index_from(struct index_state *istate)
1963	{
1964	check_ce_order(istate);
1965	tweak_untracked_cache(istate);
1966	tweak_split_index(istate);
1967	tweak_fsmonitor(istate);
1968	}
1969
1970	static size_t estimate_cache_size_from_compressed(unsigned int entries)
1971	{
1972	return entries * (sizeof(struct cache_entry) + CACHE_ENTRY_PATH_LENGTH);
1973	}
1974
1975	static size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1976	{
1977	long per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1978
1979	/*
1980	* Account for potential alignment differences.
1981	*/
1982	per_entry += align_padding_size(per_entry, 0);
1983	return ondisk_size + entries * per_entry;
1984	}
1985
1986	struct index_entry_offset
1987	{
1988	/* starting byte offset into index file, count of index entries in this block */
1989	int offset, nr;
1990	};
1991
1992	struct index_entry_offset_table
1993	{
1994	int nr;
1995	struct index_entry_offset entries[FLEX_ARRAY];
1996	};
1997
1998	static struct index_entry_offset_table read_ieot_extension(const char mmap, size_t mmap_size, size_t offset);
1999	static void write_ieot_extension(struct strbuf sb, struct index_entry_offset_table ieot);
2000
2001	static size_t read_eoie_extension(const char *mmap, size_t mmap_size);
2002	static void write_eoie_extension(struct strbuf sb, struct git_hash_ctx eoie_context, size_t offset);
2003
2004	struct load_index_extensions
2005	{
2006	pthread_t pthread;
2007	struct index_state *istate;
2008	const char *mmap;
2009	size_t mmap_size;
2010	unsigned long src_offset;
2011	};
2012
2013	static void load_index_extensions(void _data)
2014	{
2015	struct load_index_extensions *p = _data;
2016	unsigned long src_offset = p->src_offset;
2017
2018	while (src_offset <= p->mmap_size - the_hash_algo->rawsz - 8) {
2019	/* After an array of active_nr index entries,
2020	* there can be arbitrary number of extended
2021	* sections, each of which is prefixed with
2022	* extension name (4-byte) and section length
2023	* in 4-byte network byte order.
2024	*/
2025	uint32_t extsize = get_be32(p->mmap + src_offset + 4);
2026	if (read_index_extension(p->istate,
2027	p->mmap + src_offset,
2028	p->mmap + src_offset + 8,
2029	extsize) < 0) {
2030	munmap((void *)p->mmap, p->mmap_size);
2031	die(_("index file corrupt"));
2032	}
2033	src_offset += 8;
2034	src_offset += extsize;
2035	}
2036
2037	return NULL;
2038	}
2039
2040	/*
2041	* A helper function that will load the specified range of cache entries
2042	* from the memory mapped file and add them to the given index.
2043	*/
2044	static unsigned long load_cache_entry_block(struct index_state *istate,
2045	struct mem_pool ce_mem_pool, int offset, int nr, const char mmap,
2046	unsigned long start_offset, const struct cache_entry *previous_ce)
2047	{
2048	int i;
2049	unsigned long src_offset = start_offset;
2050
2051	for (i = offset; i < offset + nr; i++) {
2052	struct cache_entry *ce;
2053	unsigned long consumed;
2054
2055	ce = create_from_disk(ce_mem_pool, istate->version,
2056	mmap + src_offset,
2057	&consumed, previous_ce);
2058	set_index_entry(istate, i, ce);
2059
2060	src_offset += consumed;
2061	previous_ce = ce;
2062	}
2063	return src_offset - start_offset;
2064	}
2065
2066	static unsigned long load_all_cache_entries(struct index_state *istate,
2067	const char *mmap, size_t mmap_size, unsigned long src_offset)
2068	{
2069	unsigned long consumed;
2070
2071	istate->ce_mem_pool = xmalloc(sizeof(*istate->ce_mem_pool));
2072	if (istate->version == 4) {
2073	mem_pool_init(istate->ce_mem_pool,
2074	estimate_cache_size_from_compressed(istate->cache_nr));
2075	} else {
2076	mem_pool_init(istate->ce_mem_pool,
2077	estimate_cache_size(mmap_size, istate->cache_nr));
2078	}
2079
2080	consumed = load_cache_entry_block(istate, istate->ce_mem_pool,
2081	0, istate->cache_nr, mmap, src_offset, NULL);
2082	return consumed;
2083	}
2084
2085	/*
2086	* Mostly randomly chosen maximum thread counts: we
2087	* cap the parallelism to online_cpus() threads, and we want
2088	* to have at least 10000 cache entries per thread for it to
2089	* be worth starting a thread.
2090	*/
2091
2092	#define THREAD_COST (10000)
2093
2094	struct load_cache_entries_thread_data
2095	{
2096	pthread_t pthread;
2097	struct index_state *istate;
2098	struct mem_pool *ce_mem_pool;
2099	int offset;
2100	const char *mmap;
2101	struct index_entry_offset_table *ieot;
2102	int ieot_start; /* starting index into the ieot array */
2103	int ieot_blocks; /* count of ieot entries to process */
2104	unsigned long consumed; /* return # of bytes in index file processed */
2105	};
2106
2107	/*
2108	* A thread proc to run the load_cache_entries() computation
2109	* across multiple background threads.
2110	*/
2111	static void load_cache_entries_thread(void _data)
2112	{
2113	struct load_cache_entries_thread_data *p = _data;
2114	int i;
2115
2116	/* iterate across all ieot blocks assigned to this thread */
2117	for (i = p->ieot_start; i < p->ieot_start + p->ieot_blocks; i++) {
2118	p->consumed += load_cache_entry_block(p->istate, p->ce_mem_pool,
2119	p->offset, p->ieot->entries[i].nr, p->mmap, p->ieot->entries[i].offset, NULL);
2120	p->offset += p->ieot->entries[i].nr;
2121	}
2122	return NULL;
2123	}
2124
2125	static unsigned long load_cache_entries_threaded(struct index_state istate, const char mmap, size_t mmap_size,
2126	int nr_threads, struct index_entry_offset_table *ieot)
2127	{
2128	int i, offset, ieot_blocks, ieot_start, err;
2129	struct load_cache_entries_thread_data *data;
2130	unsigned long consumed = 0;
2131
2132	/* a little sanity checking */
2133	if (istate->name_hash_initialized)
2134	BUG("the name hash isn't thread safe");
2135
2136	istate->ce_mem_pool = xmalloc(sizeof(*istate->ce_mem_pool));
2137	mem_pool_init(istate->ce_mem_pool, 0);
2138
2139	/* ensure we have no more threads than we have blocks to process */
2140	if (nr_threads > ieot->nr)
2141	nr_threads = ieot->nr;
2142	CALLOC_ARRAY(data, nr_threads);
2143
2144	offset = ieot_start = 0;
2145	ieot_blocks = DIV_ROUND_UP(ieot->nr, nr_threads);
2146	for (i = 0; i < nr_threads; i++) {
2147	struct load_cache_entries_thread_data *p = &data[i];
2148	int nr, j;
2149
2150	if (ieot_start + ieot_blocks > ieot->nr)
2151	ieot_blocks = ieot->nr - ieot_start;
2152
2153	p->istate = istate;
2154	p->offset = offset;
2155	p->mmap = mmap;
2156	p->ieot = ieot;
2157	p->ieot_start = ieot_start;
2158	p->ieot_blocks = ieot_blocks;
2159
2160	/* create a mem_pool for each thread */
2161	nr = 0;
2162	for (j = p->ieot_start; j < p->ieot_start + p->ieot_blocks; j++)
2163	nr += p->ieot->entries[j].nr;
2164	p->ce_mem_pool = xmalloc(sizeof(*istate->ce_mem_pool));
2165	if (istate->version == 4) {
2166	mem_pool_init(p->ce_mem_pool,
2167	estimate_cache_size_from_compressed(nr));
2168	} else {
2169	mem_pool_init(p->ce_mem_pool,
2170	estimate_cache_size(mmap_size, nr));
2171	}
2172
2173	err = pthread_create(&p->pthread, NULL, load_cache_entries_thread, p);
2174	if (err)
2175	die(_("unable to create load_cache_entries thread: %s"), strerror(err));
2176
2177	/* increment by the number of cache entries in the ieot block being processed */
2178	for (j = 0; j < ieot_blocks; j++)
2179	offset += ieot->entries[ieot_start + j].nr;
2180	ieot_start += ieot_blocks;
2181	}
2182
2183	for (i = 0; i < nr_threads; i++) {
2184	struct load_cache_entries_thread_data *p = &data[i];
2185
2186	err = pthread_join(p->pthread, NULL);
2187	if (err)
2188	die(_("unable to join load_cache_entries thread: %s"), strerror(err));
2189	mem_pool_combine(istate->ce_mem_pool, p->ce_mem_pool);
2190	free(p->ce_mem_pool);
2191	consumed += p->consumed;
2192	}
2193
2194	free(data);
2195
2196	return consumed;
2197	}
2198
2199	static void set_new_index_sparsity(struct index_state *istate)
2200	{
2201	/*
2202	* If the index's repo exists, mark it sparse according to
2203	* repo settings.
2204	*/
2205	prepare_repo_settings(istate->repo);
2206	if (!istate->repo->settings.command_requires_full_index &&
2207	is_sparse_index_allowed(istate, 0))
2208	istate->sparse_index = 1;
2209	}
2210
2211	/* remember to discard_cache() before reading a different cache! */
2212	int do_read_index(struct index_state istate, const char path, int must_exist)
2213	{
2214	int fd;
2215	struct stat st;
2216	unsigned long src_offset;
2217	const struct cache_header *hdr;
2218	const char *mmap;
2219	size_t mmap_size;
2220	struct load_index_extensions p;
2221	size_t extension_offset = 0;
2222	int nr_threads, cpus;
2223	struct index_entry_offset_table *ieot = NULL;
2224
2225	if (istate->initialized)
2226	return istate->cache_nr;
2227
2228	istate->timestamp.sec = 0;
2229	istate->timestamp.nsec = 0;
2230	fd = open(path, O_RDONLY);
2231	if (fd < 0) {
2232	if (!must_exist && errno == ENOENT) {
2233	set_new_index_sparsity(istate);
2234	istate->initialized = 1;
2235	return 0;
2236	}
2237	die_errno(_("%s: index file open failed"), path);
2238	}
2239
2240	if (fstat(fd, &st))
2241	die_errno(_("%s: cannot stat the open index"), path);
2242
2243	mmap_size = xsize_t(st.st_size);
2244	if (mmap_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
2245	die(_("%s: index file smaller than expected"), path);
2246
2247	mmap = xmmap_gently(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
2248	if (mmap == MAP_FAILED)
2249	die_errno(_("%s: unable to map index file%s"), path,
2250	mmap_os_err());
2251	close(fd);
2252
2253	hdr = (const struct cache_header *)mmap;
2254	if (verify_hdr(hdr, mmap_size) < 0)
2255	goto unmap;
2256
2257	oidread(&istate->oid, (const unsigned char *)hdr + mmap_size - the_hash_algo->rawsz,
2258	the_repository->hash_algo);
2259	istate->version = ntohl(hdr->hdr_version);
2260	istate->cache_nr = ntohl(hdr->hdr_entries);
2261	istate->cache_alloc = alloc_nr(istate->cache_nr);
2262	CALLOC_ARRAY(istate->cache, istate->cache_alloc);
2263	istate->initialized = 1;
2264
2265	p.istate = istate;
2266	p.mmap = mmap;
2267	p.mmap_size = mmap_size;
2268
2269	src_offset = sizeof(*hdr);
2270
2271	if (repo_config_get_index_threads(the_repository, &nr_threads))
2272	nr_threads = 1;
2273
2274	/* TODO: does creating more threads than cores help? */
2275	if (!nr_threads) {
2276	nr_threads = istate->cache_nr / THREAD_COST;
2277	cpus = online_cpus();
2278	if (nr_threads > cpus)
2279	nr_threads = cpus;
2280	}
2281
2282	if (!HAVE_THREADS)
2283	nr_threads = 1;
2284
2285	if (nr_threads > 1) {
2286	extension_offset = read_eoie_extension(mmap, mmap_size);
2287	if (extension_offset) {
2288	int err;
2289
2290	p.src_offset = extension_offset;
2291	err = pthread_create(&p.pthread, NULL, load_index_extensions, &p);
2292	if (err)
2293	die(_("unable to create load_index_extensions thread: %s"), strerror(err));
2294
2295	nr_threads--;
2296	}
2297	}
2298
2299	/*
2300	* Locate and read the index entry offset table so that we can use it
2301	* to multi-thread the reading of the cache entries.
2302	*/
2303	if (extension_offset && nr_threads > 1)
2304	ieot = read_ieot_extension(mmap, mmap_size, extension_offset);
2305
2306	if (ieot) {
2307	src_offset += load_cache_entries_threaded(istate, mmap, mmap_size, nr_threads, ieot);
2308	free(ieot);
2309	} else {
2310	src_offset += load_all_cache_entries(istate, mmap, mmap_size, src_offset);
2311	}
2312
2313	istate->timestamp.sec = st.st_mtime;
2314	istate->timestamp.nsec = ST_MTIME_NSEC(st);
2315
2316	/* if we created a thread, join it otherwise load the extensions on the primary thread */
2317	if (extension_offset) {
2318	int ret = pthread_join(p.pthread, NULL);
2319	if (ret)
2320	die(_("unable to join load_index_extensions thread: %s"), strerror(ret));
2321	} else {
2322	p.src_offset = src_offset;
2323	load_index_extensions(&p);
2324	}
2325	munmap((void *)mmap, mmap_size);
2326
2327	trace2_data_intmax("index", istate->repo, "read/version",
2328	istate->version);
2329	trace2_data_intmax("index", istate->repo, "read/cache_nr",
2330	istate->cache_nr);
2331
2332	/*
2333	* If the command explicitly requires a full index, force it
2334	* to be full. Otherwise, correct the sparsity based on repository
2335	* settings and other properties of the index (if necessary).
2336	*/
2337	prepare_repo_settings(istate->repo);
2338	if (istate->repo->settings.command_requires_full_index)
2339	ensure_full_index(istate);
2340	else
2341	ensure_correct_sparsity(istate);
2342
2343	return istate->cache_nr;
2344
2345	unmap:
2346	munmap((void *)mmap, mmap_size);
2347	die(_("index file corrupt"));
2348	}
2349
2350	/*
2351	* Signal that the shared index is used by updating its mtime.
2352	*
2353	* This way, shared index can be removed if they have not been used
2354	* for some time.
2355	*/
2356	static void freshen_shared_index(const char *shared_index, int warn)
2357	{
2358	if (!check_and_freshen_file(shared_index, 1, NULL) && warn)
2359	warning(_("could not freshen shared index '%s'"), shared_index);
2360	}
2361
2362	int read_index_from(struct index_state istate, const char path,
2363	const char *gitdir)
2364	{
2365	struct split_index *split_index;
2366	int ret;
2367	char *base_oid_hex;
2368	char *base_path;
2369
2370	/* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2371	if (istate->initialized)
2372	return istate->cache_nr;
2373
2374	trace2_region_enter_printf("index", "do_read_index", istate->repo,
2375	"%s", path);
2376	trace_performance_enter();
2377	ret = do_read_index(istate, path, 0);
2378	trace_performance_leave("read cache %s", path);
2379	trace2_region_leave_printf("index", "do_read_index", istate->repo,
2380	"%s", path);
2381
2382	split_index = istate->split_index;
2383	if (!split_index \|\| is_null_oid(&split_index->base_oid)) {
2384	post_read_index_from(istate);
2385	return ret;
2386	}
2387
2388	trace_performance_enter();
2389	if (split_index->base)
2390	release_index(split_index->base);
2391	else
2392	ALLOC_ARRAY(split_index->base, 1);
2393	index_state_init(split_index->base, istate->repo);
2394
2395	base_oid_hex = oid_to_hex(&split_index->base_oid);
2396	base_path = xstrfmt("%s/sharedindex.%s", gitdir, base_oid_hex);
2397	if (file_exists(base_path)) {
2398	trace2_region_enter_printf("index", "shared/do_read_index",
2399	the_repository, "%s", base_path);
2400
2401	ret = do_read_index(split_index->base, base_path, 0);
2402	trace2_region_leave_printf("index", "shared/do_read_index",
2403	the_repository, "%s", base_path);
2404	} else {
2405	char *path_copy = xstrdup(path);
2406	char *base_path2 = xstrfmt("%s/sharedindex.%s",
2407	dirname(path_copy), base_oid_hex);
2408	free(path_copy);
2409	trace2_region_enter_printf("index", "shared/do_read_index",
2410	the_repository, "%s", base_path2);
2411	ret = do_read_index(split_index->base, base_path2, 1);
2412	trace2_region_leave_printf("index", "shared/do_read_index",
2413	the_repository, "%s", base_path2);
2414	free(base_path2);
2415	}
2416	if (!oideq(&split_index->base_oid, &split_index->base->oid))
2417	die(_("broken index, expect %s in %s, got %s"),
2418	base_oid_hex, base_path,
2419	oid_to_hex(&split_index->base->oid));
2420
2421	freshen_shared_index(base_path, 0);
2422	merge_base_index(istate);
2423	post_read_index_from(istate);
2424	trace_performance_leave("read cache %s", base_path);
2425	free(base_path);
2426	return ret;
2427	}
2428
2429	int is_index_unborn(struct index_state *istate)
2430	{
2431	return (!istate->cache_nr && !istate->timestamp.sec);
2432	}
2433
2434	void index_state_init(struct index_state istate, struct repository r)
2435	{
2436	struct index_state blank = INDEX_STATE_INIT(r);
2437	memcpy(istate, &blank, sizeof(*istate));
2438	}
2439
2440	void release_index(struct index_state *istate)
2441	{
2442	/*
2443	* Cache entries in istate->cache[] should have been allocated
2444	* from the memory pool associated with this index, or from an
2445	* associated split_index. There is no need to free individual
2446	* cache entries. validate_cache_entries can detect when this
2447	* assertion does not hold.
2448	*/
2449	validate_cache_entries(istate);
2450
2451	resolve_undo_clear_index(istate);
2452	free_name_hash(istate);
2453	cache_tree_free(&(istate->cache_tree));
2454	free(istate->fsmonitor_last_update);
2455	free(istate->cache);
2456	discard_split_index(istate);
2457	free_untracked_cache(istate->untracked);
2458
2459	if (istate->sparse_checkout_patterns) {
2460	clear_pattern_list(istate->sparse_checkout_patterns);
2461	FREE_AND_NULL(istate->sparse_checkout_patterns);
2462	}
2463
2464	if (istate->ce_mem_pool) {
2465	mem_pool_discard(istate->ce_mem_pool, should_validate_cache_entries());
2466	FREE_AND_NULL(istate->ce_mem_pool);
2467	}
2468	}
2469
2470	void discard_index(struct index_state *istate)
2471	{
2472	release_index(istate);
2473	index_state_init(istate, istate->repo);
2474	}
2475
2476	/*
2477	* Validate the cache entries of this index.
2478	* All cache entries associated with this index
2479	* should have been allocated by the memory pool
2480	* associated with this index, or by a referenced
2481	* split index.
2482	*/
2483	void validate_cache_entries(const struct index_state *istate)
2484	{
2485	int i;
2486
2487	if (!should_validate_cache_entries() \|\|!istate \|\| !istate->initialized)
2488	return;
2489
2490	for (i = 0; i < istate->cache_nr; i++) {
2491	if (!istate) {
2492	BUG("cache entry is not allocated from expected memory pool");
2493	} else if (!istate->ce_mem_pool \|\|
2494	!mem_pool_contains(istate->ce_mem_pool, istate->cache[i])) {
2495	if (!istate->split_index \|\|
2496	!istate->split_index->base \|\|
2497	!istate->split_index->base->ce_mem_pool \|\|
2498	!mem_pool_contains(istate->split_index->base->ce_mem_pool, istate->cache[i])) {
2499	BUG("cache entry is not allocated from expected memory pool");
2500	}
2501	}
2502	}
2503
2504	if (istate->split_index)
2505	validate_cache_entries(istate->split_index->base);
2506	}
2507
2508	int unmerged_index(const struct index_state *istate)
2509	{
2510	int i;
2511	for (i = 0; i < istate->cache_nr; i++) {
2512	if (ce_stage(istate->cache[i]))
2513	return 1;
2514	}
2515	return 0;
2516	}
2517
2518	int repo_index_has_changes(struct repository *repo,
2519	struct tree *tree,
2520	struct strbuf *sb)
2521	{
2522	struct index_state *istate = repo->index;
2523	struct object_id cmp;
2524	int i;
2525
2526	if (tree)
2527	cmp = tree->object.oid;
2528	if (tree \|\| !repo_get_oid_tree(repo, "HEAD", &cmp)) {
2529	struct diff_options opt;
2530
2531	repo_diff_setup(repo, &opt);
2532	opt.flags.exit_with_status = 1;
2533	if (!sb)
2534	opt.flags.quick = 1;
2535	diff_setup_done(&opt);
2536	do_diff_cache(&cmp, &opt);
2537	diffcore_std(&opt);
2538	for (i = 0; sb && i < diff_queued_diff.nr; i++) {
2539	if (i)
2540	strbuf_addch(sb, ' ');
2541	strbuf_addstr(sb, diff_queued_diff.queue[i]->two->path);
2542	}
2543	diff_flush(&opt);
2544	return opt.flags.has_changes != 0;
2545	} else {
2546	/* TODO: audit for interaction with sparse-index. */
2547	ensure_full_index(istate);
2548	for (i = 0; sb && i < istate->cache_nr; i++) {
2549	if (i)
2550	strbuf_addch(sb, ' ');
2551	strbuf_addstr(sb, istate->cache[i]->name);
2552	}
2553	return !!istate->cache_nr;
2554	}
2555	}
2556
2557	static int write_index_ext_header(struct hashfile *f,
2558	struct git_hash_ctx *eoie_f,
2559	unsigned int ext,
2560	unsigned int sz)
2561	{
2562	hashwrite_be32(f, ext);
2563	hashwrite_be32(f, sz);
2564
2565	if (eoie_f) {
2566	ext = htonl(ext);
2567	sz = htonl(sz);
2568	git_hash_update(eoie_f, &ext, sizeof(ext));
2569	git_hash_update(eoie_f, &sz, sizeof(sz));
2570	}
2571	return 0;
2572	}
2573
2574	static void ce_smudge_racily_clean_entry(struct index_state *istate,
2575	struct cache_entry *ce)
2576	{
2577	/*
2578	* The only thing we care about in this function is to smudge the
2579	* falsely clean entry due to touch-update-touch race, so we leave
2580	* everything else as they are. We are called for entries whose
2581	* ce_stat_data.sd_mtime match the index file mtime.
2582	*
2583	* Note that this actually does not do much for gitlinks, for
2584	* which ce_match_stat_basic() always goes to the actual
2585	* contents. The caller checks with is_racy_timestamp() which
2586	* always says "no" for gitlinks, so we are not called for them ;-)
2587	*/
2588	struct stat st;
2589
2590	if (lstat(ce->name, &st) < 0)
2591	return;
2592	if (ce_match_stat_basic(ce, &st))
2593	return;
2594	if (ce_modified_check_fs(istate, ce, &st)) {
2595	/* This is "racily clean"; smudge it. Note that this
2596	* is a tricky code. At first glance, it may appear
2597	* that it can break with this sequence:
2598	*
2599	* $ echo xyzzy >frotz
2600	* $ git-update-index --add frotz
2601	* $ : >frotz
2602	* $ sleep 3
2603	* $ echo filfre >nitfol
2604	* $ git-update-index --add nitfol
2605	*
2606	* but it does not. When the second update-index runs,
2607	* it notices that the entry "frotz" has the same timestamp
2608	* as index, and if we were to smudge it by resetting its
2609	* size to zero here, then the object name recorded
2610	* in index is the 6-byte file but the cached stat information
2611	* becomes zero --- which would then match what we would
2612	* obtain from the filesystem next time we stat("frotz").
2613	*
2614	* However, the second update-index, before calling
2615	* this function, notices that the cached size is 6
2616	* bytes and what is on the filesystem is an empty
2617	* file, and never calls us, so the cached size information
2618	* for "frotz" stays 6 which does not match the filesystem.
2619	*/
2620	ce->ce_stat_data.sd_size = 0;
2621	}
2622	}
2623
2624	/* Copy miscellaneous fields but not the name */
2625	static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
2626	struct cache_entry *ce)
2627	{
2628	short flags;
2629	const unsigned hashsz = the_hash_algo->rawsz;
2630	uint16_t flagsp = (uint16_t )(ondisk->data + hashsz);
2631
2632	ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
2633	ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
2634	ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
2635	ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
2636	ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
2637	ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
2638	ondisk->mode = htonl(ce->ce_mode);
2639	ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
2640	ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
2641	ondisk->size = htonl(ce->ce_stat_data.sd_size);
2642	hashcpy(ondisk->data, ce->oid.hash, the_repository->hash_algo);
2643
2644	flags = ce->ce_flags & ~CE_NAMEMASK;
2645	flags \|= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
2646	flagsp[0] = htons(flags);
2647	if (ce->ce_flags & CE_EXTENDED) {
2648	flagsp[1] = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
2649	}
2650	}
2651
2652	static int ce_write_entry(struct hashfile f, struct cache_entry ce,
2653	struct strbuf previous_name, struct ondisk_cache_entry ondisk)
2654	{
2655	int size;
2656	unsigned int saved_namelen;
2657	int stripped_name = 0;
2658	static unsigned char padding[8] = { 0x00 };
2659
2660	if (ce->ce_flags & CE_STRIP_NAME) {
2661	saved_namelen = ce_namelen(ce);
2662	ce->ce_namelen = 0;
2663	stripped_name = 1;
2664	}
2665
2666	size = offsetof(struct ondisk_cache_entry,data) + ondisk_data_size(ce->ce_flags, 0);
2667
2668	if (!previous_name) {
2669	int len = ce_namelen(ce);
2670	copy_cache_entry_to_ondisk(ondisk, ce);
2671	hashwrite(f, ondisk, size);
2672	hashwrite(f, ce->name, len);
2673	hashwrite(f, padding, align_padding_size(size, len));
2674	} else {
2675	int common, to_remove;
2676	uint8_t prefix_size;
2677	unsigned char to_remove_vi[16];
2678
2679	for (common = 0;
2680	(common < previous_name->len &&
2681	ce->name[common] &&
2682	ce->name[common] == previous_name->buf[common]);
2683	common++)
2684	; /* still matching */
2685	to_remove = previous_name->len - common;
2686	prefix_size = encode_varint(to_remove, to_remove_vi);
2687
2688	copy_cache_entry_to_ondisk(ondisk, ce);
2689	hashwrite(f, ondisk, size);
2690	hashwrite(f, to_remove_vi, prefix_size);
2691	hashwrite(f, ce->name + common, ce_namelen(ce) - common);
2692	hashwrite(f, padding, 1);
2693
2694	strbuf_splice(previous_name, common, to_remove,
2695	ce->name + common, ce_namelen(ce) - common);
2696	}
2697	if (stripped_name) {
2698	ce->ce_namelen = saved_namelen;
2699	ce->ce_flags &= ~CE_STRIP_NAME;
2700	}
2701
2702	return 0;
2703	}
2704
2705	/*
2706	* This function verifies if index_state has the correct sha1 of the
2707	* index file. Don't die if we have any other failure, just return 0.
2708	*/
2709	static int verify_index_from(const struct index_state istate, const char path)
2710	{
2711	int fd;
2712	ssize_t n;
2713	struct stat st;
2714	unsigned char hash[GIT_MAX_RAWSZ];
2715
2716	if (!istate->initialized)
2717	return 0;
2718
2719	fd = open(path, O_RDONLY);
2720	if (fd < 0)
2721	return 0;
2722
2723	if (fstat(fd, &st))
2724	goto out;
2725
2726	if (st.st_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
2727	goto out;
2728
2729	n = pread_in_full(fd, hash, the_hash_algo->rawsz, st.st_size - the_hash_algo->rawsz);
2730	if (n != the_hash_algo->rawsz)
2731	goto out;
2732
2733	if (!hasheq(istate->oid.hash, hash, the_repository->hash_algo))
2734	goto out;
2735
2736	close(fd);
2737	return 1;
2738
2739	out:
2740	close(fd);
2741	return 0;
2742	}
2743
2744	static int repo_verify_index(struct repository *repo)
2745	{
2746	return verify_index_from(repo->index, repo->index_file);
2747	}
2748
2749	int has_racy_timestamp(struct index_state *istate)
2750	{
2751	int entries = istate->cache_nr;
2752	int i;
2753
2754	for (i = 0; i < entries; i++) {
2755	struct cache_entry *ce = istate->cache[i];
2756	if (is_racy_timestamp(istate, ce))
2757	return 1;
2758	}
2759	return 0;
2760	}
2761
2762	void repo_update_index_if_able(struct repository *repo,
2763	struct lock_file *lockfile)
2764	{
2765	if ((repo->index->cache_changed \|\|
2766	has_racy_timestamp(repo->index)) &&
2767	repo_verify_index(repo))
2768	write_locked_index(repo->index, lockfile, COMMIT_LOCK);
2769	else
2770	rollback_lock_file(lockfile);
2771	}
2772
2773	static int record_eoie(void)
2774	{
2775	int val;
2776
2777	if (!repo_config_get_bool(the_repository, "index.recordendofindexentries", &val))
2778	return val;
2779
2780	/*
2781	* As a convenience, the end of index entries extension
2782	* used for threading is written by default if the user
2783	* explicitly requested threaded index reads.
2784	*/
2785	return !repo_config_get_index_threads(the_repository, &val) && val != 1;
2786	}
2787
2788	static int record_ieot(void)
2789	{
2790	int val;
2791
2792	if (!repo_config_get_bool(the_repository, "index.recordoffsettable", &val))
2793	return val;
2794
2795	/*
2796	* As a convenience, the offset table used for threading is
2797	* written by default if the user explicitly requested
2798	* threaded index reads.
2799	*/
2800	return !repo_config_get_index_threads(the_repository, &val) && val != 1;
2801	}
2802
2803	enum write_extensions {
2804	WRITE_NO_EXTENSION = 0,
2805	WRITE_SPLIT_INDEX_EXTENSION = 1<<0,
2806	WRITE_CACHE_TREE_EXTENSION = 1<<1,
2807	WRITE_RESOLVE_UNDO_EXTENSION = 1<<2,
2808	WRITE_UNTRACKED_CACHE_EXTENSION = 1<<3,
2809	WRITE_FSMONITOR_EXTENSION = 1<<4,
2810	};
2811	#define WRITE_ALL_EXTENSIONS ((enum write_extensions)-1)
2812
2813	/*
2814	* On success, `tempfile` is closed. If it is the temporary file
2815	* of a `struct lock_file`, we will therefore effectively perform
2816	* a 'close_lock_file_gently()`. Since that is an implementation
2817	* detail of lockfiles, callers of `do_write_index()` should not
2818	* rely on it.
2819	*/
2820	static int do_write_index(struct index_state istate, struct tempfile tempfile,
2821	enum write_extensions write_extensions, unsigned flags)
2822	{
2823	uint64_t start = getnanotime();
2824	struct hashfile *f;
2825	struct git_hash_ctx *eoie_c = NULL;
2826	struct cache_header hdr;
2827	int i, err = 0, removed, extended, hdr_version;
2828	struct cache_entry **cache = istate->cache;
2829	int entries = istate->cache_nr;
2830	struct stat st;
2831	struct ondisk_cache_entry ondisk;
2832	struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
2833	int drop_cache_tree = istate->drop_cache_tree;
2834	off_t offset;
2835	int csum_fsync_flag;
2836	int ieot_entries = 1;
2837	struct index_entry_offset_table *ieot = NULL;
2838	struct repository *r = istate->repo;
2839	struct strbuf sb = STRBUF_INIT;
2840	int nr, nr_threads, ret;
2841
2842	f = hashfd(the_repository->hash_algo, tempfile->fd, tempfile->filename.buf);
2843
2844	prepare_repo_settings(r);
2845	f->skip_hash = r->settings.index_skip_hash;
2846
2847	for (i = removed = extended = 0; i < entries; i++) {
2848	if (cache[i]->ce_flags & CE_REMOVE)
2849	removed++;
2850
2851	/* reduce extended entries if possible */
2852	cache[i]->ce_flags &= ~CE_EXTENDED;
2853	if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2854	extended++;
2855	cache[i]->ce_flags \|= CE_EXTENDED;
2856	}
2857	}
2858
2859	if (!istate->version)
2860	istate->version = get_index_format_default(r);
2861
2862	/* demote version 3 to version 2 when the latter suffices */
2863	if (istate->version == 3 \|\| istate->version == 2)
2864	istate->version = extended ? 3 : 2;
2865
2866	hdr_version = istate->version;
2867
2868	hdr.hdr_signature = htonl(CACHE_SIGNATURE);
2869	hdr.hdr_version = htonl(hdr_version);
2870	hdr.hdr_entries = htonl(entries - removed);
2871
2872	hashwrite(f, &hdr, sizeof(hdr));
2873
2874	if (!HAVE_THREADS \|\| repo_config_get_index_threads(the_repository, &nr_threads))
2875	nr_threads = 1;
2876
2877	if (nr_threads != 1 && record_ieot()) {
2878	int ieot_blocks, cpus;
2879
2880	/*
2881	* ensure default number of ieot blocks maps evenly to the
2882	* default number of threads that will process them leaving
2883	* room for the thread to load the index extensions.
2884	*/
2885	if (!nr_threads) {
2886	ieot_blocks = istate->cache_nr / THREAD_COST;
2887	cpus = online_cpus();
2888	if (ieot_blocks > cpus - 1)
2889	ieot_blocks = cpus - 1;
2890	} else {
2891	ieot_blocks = nr_threads;
2892	if (ieot_blocks > istate->cache_nr)
2893	ieot_blocks = istate->cache_nr;
2894	}
2895
2896	/*
2897	* no reason to write out the IEOT extension if we don't
2898	* have enough blocks to utilize multi-threading
2899	*/
2900	if (ieot_blocks > 1) {
2901	ieot = xcalloc(1, sizeof(struct index_entry_offset_table)
2902	+ (ieot_blocks * sizeof(struct index_entry_offset)));
2903	ieot_entries = DIV_ROUND_UP(entries, ieot_blocks);
2904	}
2905	}
2906
2907	offset = hashfile_total(f);
2908
2909	nr = 0;
2910	previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2911
2912	for (i = 0; i < entries; i++) {
2913	struct cache_entry *ce = cache[i];
2914	if (ce->ce_flags & CE_REMOVE)
2915	continue;
2916	if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2917	ce_smudge_racily_clean_entry(istate, ce);
2918	if (is_null_oid(&ce->oid)) {
2919	static const char msg[] = "cache entry has null sha1: %s";
2920	static int allow = -1;
2921
2922	if (allow < 0)
2923	allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2924	if (allow)
2925	warning(msg, ce->name);
2926	else
2927	err = error(msg, ce->name);
2928
2929	drop_cache_tree = 1;
2930	}
2931	if (ieot && i && (i % ieot_entries == 0)) {
2932	ieot->entries[ieot->nr].nr = nr;
2933	ieot->entries[ieot->nr].offset = offset;
2934	ieot->nr++;
2935	/*
2936	* If we have a V4 index, set the first byte to an invalid
2937	* character to ensure there is nothing common with the previous
2938	* entry
2939	*/
2940	if (previous_name)
2941	previous_name->buf[0] = 0;
2942	nr = 0;
2943
2944	offset = hashfile_total(f);
2945	}
2946	if (ce_write_entry(f, ce, previous_name, (struct ondisk_cache_entry *)&ondisk) < 0)
2947	err = -1;
2948
2949	if (err)
2950	break;
2951	nr++;
2952	}
2953	if (ieot && nr) {
2954	ieot->entries[ieot->nr].nr = nr;
2955	ieot->entries[ieot->nr].offset = offset;
2956	ieot->nr++;
2957	}
2958	strbuf_release(&previous_name_buf);
2959
2960	if (err) {
2961	ret = err;
2962	goto out;
2963	}
2964
2965	offset = hashfile_total(f);
2966
2967	/*
2968	* The extension headers must be hashed on their own for the
2969	* EOIE extension. Create a hashfile here to compute that hash.
2970	*/
2971	if (offset && record_eoie()) {
2972	CALLOC_ARRAY(eoie_c, 1);
2973	git_hash_init(eoie_c, the_hash_algo);
2974	}
2975
2976	/*
2977	* Lets write out CACHE_EXT_INDEXENTRYOFFSETTABLE first so that we
2978	* can minimize the number of extensions we have to scan through to
2979	* find it during load. Write it out regardless of the
2980	* strip_extensions parameter as we need it when loading the shared
2981	* index.
2982	*/
2983	if (ieot) {
2984	strbuf_reset(&sb);
2985
2986	write_ieot_extension(&sb, ieot);
2987	err = write_index_ext_header(f, eoie_c, CACHE_EXT_INDEXENTRYOFFSETTABLE, sb.len) < 0;
2988	hashwrite(f, sb.buf, sb.len);
2989	if (err) {
2990	ret = -1;
2991	goto out;
2992	}
2993	}
2994
2995	if (write_extensions & WRITE_SPLIT_INDEX_EXTENSION &&
2996	istate->split_index) {
2997	strbuf_reset(&sb);
2998
2999	if (istate->sparse_index)
3000	die(_("cannot write split index for a sparse index"));
3001
3002	err = write_link_extension(&sb, istate) < 0 \|\|
3003	write_index_ext_header(f, eoie_c, CACHE_EXT_LINK,
3004	sb.len) < 0;
3005	hashwrite(f, sb.buf, sb.len);
3006	if (err) {
3007	ret = -1;
3008	goto out;
3009	}
3010	}
3011	if (write_extensions & WRITE_CACHE_TREE_EXTENSION &&
3012	!drop_cache_tree && istate->cache_tree) {
3013	strbuf_reset(&sb);
3014
3015	cache_tree_write(&sb, istate->cache_tree);
3016	err = write_index_ext_header(f, eoie_c, CACHE_EXT_TREE, sb.len) < 0;
3017	hashwrite(f, sb.buf, sb.len);
3018	if (err) {
3019	ret = -1;
3020	goto out;
3021	}
3022	}
3023	if (write_extensions & WRITE_RESOLVE_UNDO_EXTENSION &&
3024	istate->resolve_undo) {
3025	strbuf_reset(&sb);
3026
3027	resolve_undo_write(&sb, istate->resolve_undo, the_hash_algo);
3028	err = write_index_ext_header(f, eoie_c, CACHE_EXT_RESOLVE_UNDO,
3029	sb.len) < 0;
3030	hashwrite(f, sb.buf, sb.len);
3031	if (err) {
3032	ret = -1;
3033	goto out;
3034	}
3035	}
3036	if (write_extensions & WRITE_UNTRACKED_CACHE_EXTENSION &&
3037	istate->untracked) {
3038	strbuf_reset(&sb);
3039
3040	write_untracked_extension(&sb, istate->untracked);
3041	err = write_index_ext_header(f, eoie_c, CACHE_EXT_UNTRACKED,
3042	sb.len) < 0;
3043	hashwrite(f, sb.buf, sb.len);
3044	if (err) {
3045	ret = -1;
3046	goto out;
3047	}
3048	}
3049	if (write_extensions & WRITE_FSMONITOR_EXTENSION &&
3050	istate->fsmonitor_last_update) {
3051	strbuf_reset(&sb);
3052
3053	write_fsmonitor_extension(&sb, istate);
3054	err = write_index_ext_header(f, eoie_c, CACHE_EXT_FSMONITOR, sb.len) < 0;
3055	hashwrite(f, sb.buf, sb.len);
3056	if (err) {
3057	ret = -1;
3058	goto out;
3059	}
3060	}
3061	if (istate->sparse_index) {
3062	if (write_index_ext_header(f, eoie_c, CACHE_EXT_SPARSE_DIRECTORIES, 0) < 0) {
3063	ret = -1;
3064	goto out;
3065	}
3066	}
3067
3068	/*
3069	* CACHE_EXT_ENDOFINDEXENTRIES must be written as the last entry before the SHA1
3070	* so that it can be found and processed before all the index entries are
3071	* read. Write it out regardless of the strip_extensions parameter as we need it
3072	* when loading the shared index.
3073	*/
3074	if (eoie_c) {
3075	strbuf_reset(&sb);
3076
3077	write_eoie_extension(&sb, eoie_c, offset);
3078	err = write_index_ext_header(f, NULL, CACHE_EXT_ENDOFINDEXENTRIES, sb.len) < 0;
3079	hashwrite(f, sb.buf, sb.len);
3080	if (err) {
3081	ret = -1;
3082	goto out;
3083	}
3084	}
3085
3086	csum_fsync_flag = 0;
3087	if (!alternate_index_output && (flags & COMMIT_LOCK))
3088	csum_fsync_flag = CSUM_FSYNC;
3089
3090	finalize_hashfile(f, istate->oid.hash, FSYNC_COMPONENT_INDEX,
3091	CSUM_HASH_IN_STREAM \| csum_fsync_flag);
3092	f = NULL;
3093
3094	if (close_tempfile_gently(tempfile)) {
3095	ret = error(_("could not close '%s'"), get_tempfile_path(tempfile));
3096	goto out;
3097	}
3098	if (stat(get_tempfile_path(tempfile), &st)) {
3099	ret = -1;
3100	goto out;
3101	}
3102	istate->timestamp.sec = (unsigned int)st.st_mtime;
3103	istate->timestamp.nsec = ST_MTIME_NSEC(st);
3104	trace_performance_since(start, "write index, changed mask = %x", istate->cache_changed);
3105
3106	trace2_data_intmax("index", istate->repo, "write/version",
3107	istate->version);
3108	trace2_data_intmax("index", istate->repo, "write/cache_nr",
3109	istate->cache_nr);
3110
3111	ret = 0;
3112
3113	out:
3114	if (f)
3115	free_hashfile(f);
3116	strbuf_release(&sb);
3117	free(eoie_c);
3118	free(ieot);
3119	return ret;
3120	}
3121
3122	void set_alternate_index_output(const char *name)
3123	{
3124	alternate_index_output = name;
3125	}
3126
3127	static int commit_locked_index(struct lock_file *lk)
3128	{
3129	if (alternate_index_output)
3130	return commit_lock_file_to(lk, alternate_index_output);
3131	else
3132	return commit_lock_file(lk);
3133	}
3134
3135	static int do_write_locked_index(struct index_state *istate,
3136	struct lock_file *lock,
3137	unsigned flags,
3138	enum write_extensions write_extensions)
3139	{
3140	int ret;
3141	int was_full = istate->sparse_index == INDEX_EXPANDED;
3142
3143	ret = convert_to_sparse(istate, 0);
3144
3145	if (ret) {
3146	warning(_("failed to convert to a sparse-index"));
3147	return ret;
3148	}
3149
3150	trace2_region_enter_printf("index", "do_write_index", istate->repo,
3151	"%s", get_lock_file_path(lock));
3152	ret = do_write_index(istate, lock->tempfile, write_extensions, flags);
3153	trace2_region_leave_printf("index", "do_write_index", istate->repo,
3154	"%s", get_lock_file_path(lock));
3155
3156	if (was_full)
3157	ensure_full_index(istate);
3158
3159	if (ret)
3160	return ret;
3161	if (flags & COMMIT_LOCK)
3162	ret = commit_locked_index(lock);
3163	else
3164	ret = close_lock_file_gently(lock);
3165
3166	run_hooks_l(the_repository, "post-index-change",
3167	istate->updated_workdir ? "1" : "0",
3168	istate->updated_skipworktree ? "1" : "0", NULL);
3169	istate->updated_workdir = 0;
3170	istate->updated_skipworktree = 0;
3171
3172	return ret;
3173	}
3174
3175	static int write_split_index(struct index_state *istate,
3176	struct lock_file *lock,
3177	unsigned flags)
3178	{
3179	int ret;
3180	prepare_to_write_split_index(istate);
3181	ret = do_write_locked_index(istate, lock, flags, WRITE_ALL_EXTENSIONS);
3182	finish_writing_split_index(istate);
3183	return ret;
3184	}
3185
3186	static unsigned long get_shared_index_expire_date(void)
3187	{
3188	static unsigned long shared_index_expire_date;
3189	static int shared_index_expire_date_prepared;
3190
3191	if (!shared_index_expire_date_prepared) {
3192	const char *shared_index_expire = "2.weeks.ago";
3193	char *value = NULL;
3194
3195	repo_config_get_expiry(the_repository, "splitindex.sharedindexexpire",
3196	&value);
3197	if (value)
3198	shared_index_expire = value;
3199
3200	shared_index_expire_date = approxidate(shared_index_expire);
3201	shared_index_expire_date_prepared = 1;
3202
3203	free(value);
3204	}
3205
3206	return shared_index_expire_date;
3207	}
3208
3209	static int should_delete_shared_index(const char *shared_index_path)
3210	{
3211	struct stat st;
3212	unsigned long expiration;
3213
3214	/* Check timestamp */
3215	expiration = get_shared_index_expire_date();
3216	if (!expiration)
3217	return 0;
3218	if (stat(shared_index_path, &st))
3219	return error_errno(_("could not stat '%s'"), shared_index_path);
3220	if (st.st_mtime > expiration)
3221	return 0;
3222
3223	return 1;
3224	}
3225
3226	static int clean_shared_index_files(const char *current_hex)
3227	{
3228	struct dirent *de;
3229	DIR *dir = opendir(repo_get_git_dir(the_repository));
3230
3231	if (!dir)
3232	return error_errno(_("unable to open git dir: %s"),
3233	repo_get_git_dir(the_repository));
3234
3235	while ((de = readdir(dir)) != NULL) {
3236	const char *sha1_hex;
3237	char *shared_index_path;
3238	if (!skip_prefix(de->d_name, "sharedindex.", &sha1_hex))
3239	continue;
3240	if (!strcmp(sha1_hex, current_hex))
3241	continue;
3242
3243	shared_index_path = repo_git_path(the_repository, "%s", de->d_name);
3244	if (should_delete_shared_index(shared_index_path) > 0 &&
3245	unlink(shared_index_path))
3246	warning_errno(_("unable to unlink: %s"), shared_index_path);
3247
3248	free(shared_index_path);
3249	}
3250	closedir(dir);
3251
3252	return 0;
3253	}
3254
3255	static int write_shared_index(struct index_state *istate,
3256	struct tempfile **temp, unsigned flags)
3257	{
3258	struct split_index *si = istate->split_index;
3259	int ret, was_full = !istate->sparse_index;
3260	char *path;
3261
3262	move_cache_to_base_index(istate);
3263	convert_to_sparse(istate, 0);
3264
3265	trace2_region_enter_printf("index", "shared/do_write_index",
3266	the_repository, "%s", get_tempfile_path(*temp));
3267	ret = do_write_index(si->base, *temp, WRITE_NO_EXTENSION, flags);
3268	trace2_region_leave_printf("index", "shared/do_write_index",
3269	the_repository, "%s", get_tempfile_path(*temp));
3270
3271	if (was_full)
3272	ensure_full_index(istate);
3273
3274	if (ret)
3275	return ret;
3276	ret = adjust_shared_perm(the_repository, get_tempfile_path(*temp));
3277	if (ret) {
3278	error(_("cannot fix permission bits on '%s'"), get_tempfile_path(*temp));
3279	return ret;
3280	}
3281
3282	path = repo_git_path(the_repository, "sharedindex.%s", oid_to_hex(&si->base->oid));
3283	ret = rename_tempfile(temp, path);
3284	if (!ret) {
3285	oidcpy(&si->base_oid, &si->base->oid);
3286	clean_shared_index_files(oid_to_hex(&si->base->oid));
3287	}
3288
3289	free(path);
3290	return ret;
3291	}
3292
3293	static const int default_max_percent_split_change = 20;
3294
3295	static int too_many_not_shared_entries(struct index_state *istate)
3296	{
3297	int i, not_shared = 0;
3298	int max_split = repo_config_get_max_percent_split_change(the_repository);
3299
3300	switch (max_split) {
3301	case -1:
3302	/* not or badly configured: use the default value */
3303	max_split = default_max_percent_split_change;
3304	break;
3305	case 0:
3306	return 1; /* 0% means always write a new shared index */
3307	case 100:
3308	return 0; /* 100% means never write a new shared index */
3309	default:
3310	break; /* just use the configured value */
3311	}
3312
3313	/* Count not shared entries */
3314	for (i = 0; i < istate->cache_nr; i++) {
3315	struct cache_entry *ce = istate->cache[i];
3316	if (!ce->index)
3317	not_shared++;
3318	}
3319
3320	return (int64_t)istate->cache_nr * max_split < (int64_t)not_shared * 100;
3321	}
3322
3323	int write_locked_index(struct index_state istate, struct lock_file lock,
3324	unsigned flags)
3325	{
3326	int new_shared_index, ret, test_split_index_env;
3327	struct split_index *si = istate->split_index;
3328
3329	if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0) &&
3330	cache_tree_verify(the_repository, istate) < 0)
3331	return -1;
3332
3333	if ((flags & SKIP_IF_UNCHANGED) && !istate->cache_changed) {
3334	if (flags & COMMIT_LOCK)
3335	rollback_lock_file(lock);
3336	return 0;
3337	}
3338
3339	if (istate->fsmonitor_last_update)
3340	fill_fsmonitor_bitmap(istate);
3341
3342	test_split_index_env = git_env_bool("GIT_TEST_SPLIT_INDEX", 0);
3343
3344	if ((!si && !test_split_index_env) \|\|
3345	alternate_index_output \|\|
3346	(istate->cache_changed & ~EXTMASK)) {
3347	ret = do_write_locked_index(istate, lock, flags,
3348	~WRITE_SPLIT_INDEX_EXTENSION);
3349	goto out;
3350	}
3351
3352	if (test_split_index_env) {
3353	if (!si) {
3354	si = init_split_index(istate);
3355	istate->cache_changed \|= SPLIT_INDEX_ORDERED;
3356	} else {
3357	int v = si->base_oid.hash[0];
3358	if ((v & 15) < 6)
3359	istate->cache_changed \|= SPLIT_INDEX_ORDERED;
3360	}
3361	}
3362	if (too_many_not_shared_entries(istate))
3363	istate->cache_changed \|= SPLIT_INDEX_ORDERED;
3364
3365	new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED;
3366
3367	if (new_shared_index) {
3368	struct tempfile *temp;
3369	int saved_errno;
3370	char *path;
3371
3372	/* Same initial permissions as the main .git/index file */
3373	path = repo_git_path(the_repository, "sharedindex_XXXXXX");
3374	temp = mks_tempfile_sm(path, 0, 0666);
3375	free(path);
3376	if (!temp) {
3377	ret = do_write_locked_index(istate, lock, flags,
3378	~WRITE_SPLIT_INDEX_EXTENSION);
3379	goto out;
3380	}
3381	ret = write_shared_index(istate, &temp, flags);
3382
3383	saved_errno = errno;
3384	if (is_tempfile_active(temp))
3385	delete_tempfile(&temp);
3386	errno = saved_errno;
3387
3388	if (ret)
3389	goto out;
3390	}
3391
3392	ret = write_split_index(istate, lock, flags);
3393
3394	/* Freshen the shared index only if the split-index was written */
3395	if (!ret && !new_shared_index && !is_null_oid(&si->base_oid)) {
3396	char *shared_index = repo_git_path(the_repository, "sharedindex.%s",
3397	oid_to_hex(&si->base_oid));
3398	freshen_shared_index(shared_index, 1);
3399	free(shared_index);
3400	}
3401
3402	out:
3403	if (flags & COMMIT_LOCK)
3404	rollback_lock_file(lock);
3405	return ret;
3406	}
3407
3408	/*
3409	* Read the index file that is potentially unmerged into given
3410	* index_state, dropping any unmerged entries to stage #0 (potentially
3411	* resulting in a path appearing as both a file and a directory in the
3412	* index; the caller is responsible to clear out the extra entries
3413	* before writing the index to a tree). Returns true if the index is
3414	* unmerged. Callers who want to refuse to work from an unmerged
3415	* state can call this and check its return value, instead of calling
3416	* read_cache().
3417	*/
3418	int repo_read_index_unmerged(struct repository *repo)
3419	{
3420	repo_read_index(repo);
3421	return index_state_unmerged_to_stage0(repo->index);
3422	}
3423
3424	int index_state_unmerged_to_stage0(struct index_state *istate)
3425	{
3426	int unmerged = 0;
3427
3428	for (unsigned int i = 0; i < istate->cache_nr; i++) {
3429	struct cache_entry *ce = istate->cache[i];
3430	struct cache_entry *new_ce;
3431	int len;
3432
3433	if (!ce_stage(ce))
3434	continue;
3435	unmerged = 1;
3436	len = ce_namelen(ce);
3437	new_ce = make_empty_cache_entry(istate, len);
3438	memcpy(new_ce->name, ce->name, len);
3439	new_ce->ce_flags = create_ce_flags(0) \| CE_CONFLICTED;
3440	new_ce->ce_namelen = len;
3441	new_ce->ce_mode = ce->ce_mode;
3442	if (add_index_entry(istate, new_ce, ADD_CACHE_SKIP_DFCHECK))
3443	return error(_("%s: cannot drop to stage #0"),
3444	new_ce->name);
3445	}
3446	return unmerged;
3447	}
3448
3449	/*
3450	* Returns 1 if the path is an "other" path with respect to
3451	* the index; that is, the path is not mentioned in the index at all,
3452	* either as a file, a directory with some files in the index,
3453	* or as an unmerged entry.
3454	*
3455	* We helpfully remove a trailing "/" from directories so that
3456	* the output of read_directory can be used as-is.
3457	*/
3458	int index_name_is_other(struct index_state istate, const char name,
3459	int namelen)
3460	{
3461	int pos;
3462	if (namelen && name[namelen - 1] == '/')
3463	namelen--;
3464	pos = index_name_pos(istate, name, namelen);
3465	if (0 <= pos)
3466	return 0; /* exact match */
3467	pos = -pos - 1;
3468	if (pos < istate->cache_nr) {
3469	struct cache_entry *ce = istate->cache[pos];
3470	if (ce_namelen(ce) == namelen &&
3471	!memcmp(ce->name, name, namelen))
3472	return 0; /* Yup, this one exists unmerged */
3473	}
3474	return 1;
3475	}
3476
3477	void read_blob_data_from_index(struct index_state istate,
3478	const char path, unsigned long size)
3479	{
3480	int pos, len;
3481	size_t sz;
3482	enum object_type type;
3483	void *data;
3484
3485	len = strlen(path);
3486	pos = index_name_pos(istate, path, len);
3487	if (pos < 0) {
3488	/*
3489	* We might be in the middle of a merge, in which
3490	* case we would read stage #2 (ours).
3491	*/
3492	int i;
3493	for (i = -pos - 1;
3494	(pos < 0 && i < istate->cache_nr &&
3495	!strcmp(istate->cache[i]->name, path));
3496	i++)
3497	if (ce_stage(istate->cache[i]) == 2)
3498	pos = i;
3499	}
3500	if (pos < 0)
3501	return NULL;
3502	data = odb_read_object(the_repository->objects, &istate->cache[pos]->oid,
3503	&type, &sz);
3504	if (!data \|\| type != OBJ_BLOB) {
3505	free(data);
3506	return NULL;
3507	}
3508	if (size)
3509	*size = cast_size_t_to_ulong(sz);
3510	return data;
3511	}
3512
3513	void move_index_extensions(struct index_state dst, struct index_state src)
3514	{
3515	dst->untracked = src->untracked;
3516	src->untracked = NULL;
3517	dst->cache_tree = src->cache_tree;
3518	src->cache_tree = NULL;
3519	}
3520
3521	struct cache_entry dup_cache_entry(const struct cache_entry ce,
3522	struct index_state *istate)
3523	{
3524	unsigned int size = ce_size(ce);
3525	int mem_pool_allocated;
3526	struct cache_entry *new_entry = make_empty_cache_entry(istate, ce_namelen(ce));
3527	mem_pool_allocated = new_entry->mem_pool_allocated;
3528
3529	memcpy(new_entry, ce, size);
3530	new_entry->mem_pool_allocated = mem_pool_allocated;
3531	return new_entry;
3532	}
3533
3534	void discard_cache_entry(struct cache_entry *ce)
3535	{
3536	if (ce && should_validate_cache_entries())
3537	memset(ce, 0xCD, cache_entry_size(ce->ce_namelen));
3538
3539	if (ce && ce->mem_pool_allocated)
3540	return;
3541
3542	free(ce);
3543	}
3544
3545	int should_validate_cache_entries(void)
3546	{
3547	static int validate_index_cache_entries = -1;
3548
3549	if (validate_index_cache_entries < 0) {
3550	if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))
3551	validate_index_cache_entries = 1;
3552	else
3553	validate_index_cache_entries = 0;
3554	}
3555
3556	return validate_index_cache_entries;
3557	}
3558
3559	#define EOIE_SIZE (4 + GIT_SHA1_RAWSZ) /* <4-byte offset> + <20-byte hash> */
3560	#define EOIE_SIZE_WITH_HEADER (4 + 4 + EOIE_SIZE) /* <4-byte signature> + <4-byte length> + EOIE_SIZE */
3561
3562	static size_t read_eoie_extension(const char *mmap, size_t mmap_size)
3563	{
3564	/*
3565	* The end of index entries (EOIE) extension is guaranteed to be last
3566	* so that it can be found by scanning backwards from the EOF.
3567	*
3568	* "EOIE"
3569	* <4-byte length>
3570	* <4-byte offset>
3571	* <20-byte hash>
3572	*/
3573	const char index, eoie;
3574	uint32_t extsize;
3575	size_t offset, src_offset;
3576	unsigned char hash[GIT_MAX_RAWSZ];
3577	struct git_hash_ctx c;
3578
3579	/* ensure we have an index big enough to contain an EOIE extension */
3580	if (mmap_size < sizeof(struct cache_header) + EOIE_SIZE_WITH_HEADER + the_hash_algo->rawsz)
3581	return 0;
3582
3583	/* validate the extension signature */
3584	index = eoie = mmap + mmap_size - EOIE_SIZE_WITH_HEADER - the_hash_algo->rawsz;
3585	if (CACHE_EXT(index) != CACHE_EXT_ENDOFINDEXENTRIES)
3586	return 0;
3587	index += sizeof(uint32_t);
3588
3589	/* validate the extension size */
3590	extsize = get_be32(index);
3591	if (extsize != EOIE_SIZE)
3592	return 0;
3593	index += sizeof(uint32_t);
3594
3595	/*
3596	* Validate the offset we're going to look for the first extension
3597	* signature is after the index header and before the eoie extension.
3598	*/
3599	offset = get_be32(index);
3600	if (mmap + offset < mmap + sizeof(struct cache_header))
3601	return 0;
3602	if (mmap + offset >= eoie)
3603	return 0;
3604	index += sizeof(uint32_t);
3605
3606	/*
3607	* The hash is computed over extension types and their sizes (but not
3608	* their contents). E.g. if we have "TREE" extension that is N-bytes
3609	* long, "REUC" extension that is M-bytes long, followed by "EOIE",
3610	* then the hash would be:
3611	*
3612	* SHA-1("TREE" + <binary representation of N> +
3613	* "REUC" + <binary representation of M>)
3614	*/
3615	src_offset = offset;
3616	git_hash_init(&c, the_hash_algo);
3617	while (src_offset < mmap_size - the_hash_algo->rawsz - EOIE_SIZE_WITH_HEADER) {
3618	/* After an array of active_nr index entries,
3619	* there can be arbitrary number of extended
3620	* sections, each of which is prefixed with
3621	* extension name (4-byte) and section length
3622	* in 4-byte network byte order.
3623	*/
3624	uint32_t extsize;
3625	memcpy(&extsize, mmap + src_offset + 4, 4);
3626	extsize = ntohl(extsize);
3627
3628	/* verify the extension size isn't so large it will wrap around */
3629	if (src_offset + 8 + extsize < src_offset)
3630	return 0;
3631
3632	git_hash_update(&c, mmap + src_offset, 8);
3633
3634	src_offset += 8;
3635	src_offset += extsize;
3636	}
3637	git_hash_final(hash, &c);
3638	if (!hasheq(hash, (const unsigned char *)index, the_repository->hash_algo))
3639	return 0;
3640
3641	/* Validate that the extension offsets returned us back to the eoie extension. */
3642	if (src_offset != mmap_size - the_hash_algo->rawsz - EOIE_SIZE_WITH_HEADER)
3643	return 0;
3644
3645	return offset;
3646	}
3647
3648	static void write_eoie_extension(struct strbuf sb, struct git_hash_ctx eoie_context, size_t offset)
3649	{
3650	uint32_t buffer;
3651	unsigned char hash[GIT_MAX_RAWSZ];
3652
3653	/* offset */
3654	put_be32(&buffer, offset);
3655	strbuf_add(sb, &buffer, sizeof(uint32_t));
3656
3657	/* hash */
3658	git_hash_final(hash, eoie_context);
3659	strbuf_add(sb, hash, the_hash_algo->rawsz);
3660	}
3661
3662	#define IEOT_VERSION (1)
3663
3664	static struct index_entry_offset_table read_ieot_extension(const char mmap, size_t mmap_size, size_t offset)
3665	{
3666	const char *index = NULL;
3667	uint32_t extsize, ext_version;
3668	struct index_entry_offset_table *ieot;
3669	int i, nr;
3670
3671	/* find the IEOT extension */
3672	if (!offset)
3673	return NULL;
3674	while (offset <= mmap_size - the_hash_algo->rawsz - 8) {
3675	extsize = get_be32(mmap + offset + 4);
3676	if (CACHE_EXT((mmap + offset)) == CACHE_EXT_INDEXENTRYOFFSETTABLE) {
3677	index = mmap + offset + 4 + 4;
3678	break;
3679	}
3680	offset += 8;
3681	offset += extsize;
3682	}
3683	if (!index)
3684	return NULL;
3685
3686	/* validate the version is IEOT_VERSION */
3687	ext_version = get_be32(index);
3688	if (ext_version != IEOT_VERSION) {
3689	error("invalid IEOT version %d", ext_version);
3690	return NULL;
3691	}
3692	index += sizeof(uint32_t);
3693
3694	/* extension size - version bytes / bytes per entry */
3695	nr = (extsize - sizeof(uint32_t)) / (sizeof(uint32_t) + sizeof(uint32_t));
3696	if (!nr) {
3697	error("invalid number of IEOT entries %d", nr);
3698	return NULL;
3699	}
3700	ieot = xmalloc(sizeof(struct index_entry_offset_table)
3701	+ (nr * sizeof(struct index_entry_offset)));
3702	ieot->nr = nr;
3703	for (i = 0; i < nr; i++) {
3704	ieot->entries[i].offset = get_be32(index);
3705	index += sizeof(uint32_t);
3706	ieot->entries[i].nr = get_be32(index);
3707	index += sizeof(uint32_t);
3708	}
3709
3710	return ieot;
3711	}
3712
3713	static void write_ieot_extension(struct strbuf sb, struct index_entry_offset_table ieot)
3714	{
3715	uint32_t buffer;
3716	int i;
3717
3718	/* version */
3719	put_be32(&buffer, IEOT_VERSION);
3720	strbuf_add(sb, &buffer, sizeof(uint32_t));
3721
3722	/* ieot */
3723	for (i = 0; i < ieot->nr; i++) {
3724
3725	/* offset */
3726	put_be32(&buffer, ieot->entries[i].offset);
3727	strbuf_add(sb, &buffer, sizeof(uint32_t));
3728
3729	/* count */
3730	put_be32(&buffer, ieot->entries[i].nr);
3731	strbuf_add(sb, &buffer, sizeof(uint32_t));
3732	}
3733	}
3734
3735	void prefetch_cache_entries(const struct index_state *istate,
3736	must_prefetch_predicate must_prefetch)
3737	{
3738	int i;
3739	struct oid_array to_fetch = OID_ARRAY_INIT;
3740
3741	for (i = 0; i < istate->cache_nr; i++) {
3742	struct cache_entry *ce = istate->cache[i];
3743
3744	if (S_ISGITLINK(ce->ce_mode) \|\| !must_prefetch(ce))
3745	continue;
3746	if (!odb_read_object_info_extended(the_repository->objects,
3747	&ce->oid, NULL,
3748	OBJECT_INFO_FOR_PREFETCH))
3749	continue;
3750	oid_array_append(&to_fetch, &ce->oid);
3751	}
3752	promisor_remote_get_direct(the_repository,
3753	to_fetch.oid, to_fetch.nr);
3754	oid_array_clear(&to_fetch);
3755	}
3756
3757	static int read_one_entry_opt(struct index_state *istate,
3758	const struct object_id *oid,
3759	struct strbuf *base,
3760	const char *pathname,
3761	unsigned mode, int opt)
3762	{
3763	int len;
3764	struct cache_entry *ce;
3765
3766	if (S_ISDIR(mode))
3767	return READ_TREE_RECURSIVE;
3768
3769	len = strlen(pathname);
3770	ce = make_empty_cache_entry(istate, base->len + len);
3771
3772	ce->ce_mode = create_ce_mode(mode);
3773	ce->ce_flags = create_ce_flags(1);
3774	ce->ce_namelen = base->len + len;
3775	memcpy(ce->name, base->buf, base->len);
3776	memcpy(ce->name + base->len, pathname, len+1);
3777	oidcpy(&ce->oid, oid);
3778	return add_index_entry(istate, ce, opt);
3779	}
3780
3781	static int read_one_entry(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,
3788	ADD_CACHE_OK_TO_ADD\|ADD_CACHE_SKIP_DFCHECK);
3789	}
3790
3791	/*
3792	* This is used when the caller knows there is no existing entries at
3793	* the stage that will conflict with the entry being added.
3794	*/
3795	static int read_one_entry_quick(const struct object_id oid, struct strbuf base,
3796	const char *pathname, unsigned mode,
3797	void *context)
3798	{
3799	struct index_state *istate = context;
3800	return read_one_entry_opt(istate, oid, base, pathname,
3801	mode, ADD_CACHE_JUST_APPEND);
3802	}
3803
3804	/*
3805	* Read the tree specified with --with-tree option
3806	* (typically, HEAD) into stage #1 and then
3807	* squash them down to stage #0. This is used for
3808	* --error-unmatch to list and check the path patterns
3809	* that were given from the command line. We are not
3810	* going to write this index out.
3811	*/
3812	void overlay_tree_on_index(struct index_state *istate,
3813	const char tree_name, const char prefix)
3814	{
3815	struct tree *tree;
3816	struct object_id oid;
3817	struct pathspec pathspec;
3818	struct cache_entry *last_stage0 = NULL;
3819	int i;
3820	read_tree_fn_t fn = NULL;
3821	int err;
3822
3823	if (repo_get_oid(the_repository, tree_name, &oid))
3824	die("tree-ish %s not found.", tree_name);
3825	tree = repo_parse_tree_indirect(the_repository, &oid);
3826	if (!tree)
3827	die("bad tree-ish %s", tree_name);
3828
3829	/* Hoist the unmerged entries up to stage #3 to make room */
3830	/* TODO: audit for interaction with sparse-index. */
3831	ensure_full_index(istate);
3832	for (i = 0; i < istate->cache_nr; i++) {
3833	struct cache_entry *ce = istate->cache[i];
3834	if (!ce_stage(ce))
3835	continue;
3836	ce->ce_flags \|= CE_STAGEMASK;
3837	}
3838
3839	if (prefix) {
3840	static const char *(matchbuf[1]);
3841	matchbuf[0] = NULL;
3842	parse_pathspec(&pathspec, PATHSPEC_ALL_MAGIC,
3843	PATHSPEC_PREFER_CWD, prefix, matchbuf);
3844	} else
3845	memset(&pathspec, 0, sizeof(pathspec));
3846
3847	/*
3848	* See if we have cache entry at the stage. If so,
3849	* do it the original slow way, otherwise, append and then
3850	* sort at the end.
3851	*/
3852	for (i = 0; !fn && i < istate->cache_nr; i++) {
3853	const struct cache_entry *ce = istate->cache[i];
3854	if (ce_stage(ce) == 1)
3855	fn = read_one_entry;
3856	}
3857
3858	if (!fn)
3859	fn = read_one_entry_quick;
3860	err = read_tree(the_repository, tree, &pathspec, fn, istate);
3861	clear_pathspec(&pathspec);
3862	if (err)
3863	die("unable to read tree entries %s", tree_name);
3864
3865	/*
3866	* Sort the cache entry -- we need to nuke the cache tree, though.
3867	*/
3868	if (fn == read_one_entry_quick) {
3869	cache_tree_free(&istate->cache_tree);
3870	QSORT(istate->cache, istate->cache_nr, cmp_cache_name_compare);
3871	}
3872
3873	for (i = 0; i < istate->cache_nr; i++) {
3874	struct cache_entry *ce = istate->cache[i];
3875	switch (ce_stage(ce)) {
3876	case 0:
3877	last_stage0 = ce;
3878	/* fallthru */
3879	default:
3880	continue;
3881	case 1:
3882	/*
3883	* If there is stage #0 entry for this, we do not
3884	* need to show it. We use CE_UPDATE bit to mark
3885	* such an entry.
3886	*/
3887	if (last_stage0 &&
3888	!strcmp(last_stage0->name, ce->name))
3889	ce->ce_flags \|= CE_UPDATE;
3890	}
3891	}
3892	}
3893
3894	struct update_callback_data {
3895	struct index_state *index;
3896	struct repository *repo;
3897	struct pathspec *pathspec;
3898	int include_sparse;
3899	int flags;
3900	int add_errors;
3901	int ignored_too;
3902	};
3903
3904	static int fix_unmerged_status(struct diff_filepair *p,
3905	struct update_callback_data *data)
3906	{
3907	if (p->status != DIFF_STATUS_UNMERGED)
3908	return p->status;
3909	if (!(data->flags & ADD_CACHE_IGNORE_REMOVAL) && !p->two->mode)
3910	/*
3911	* This is not an explicit add request, and the
3912	* path is missing from the working tree (deleted)
3913	*/
3914	return DIFF_STATUS_DELETED;
3915	else
3916	/*
3917	* Either an explicit add request, or path exists
3918	* in the working tree. An attempt to explicitly
3919	* add a path that does not exist in the working tree
3920	* will be caught as an error by the caller immediately.
3921	*/
3922	return DIFF_STATUS_MODIFIED;
3923	}
3924
3925	static int skip_submodule(const char *path,
3926	struct repository *repo,
3927	struct pathspec *pathspec,
3928	int ignored_too)
3929	{
3930	struct stat st;
3931	const struct submodule *sub;
3932	int pathspec_matches = 0;
3933	int ps_i;
3934	char *norm_pathspec = NULL;
3935
3936	/* Only consider if path is a directory */
3937	if (lstat(path, &st) \|\| !S_ISDIR(st.st_mode))
3938	return 0;
3939
3940	/* Check if it's a submodule with ignore=all */
3941	sub = submodule_from_path(repo, null_oid(the_hash_algo), path);
3942	if (!sub \|\| !sub->name \|\| !sub->ignore \|\| strcmp(sub->ignore, "all"))
3943	return 0;
3944
3945	trace_printf("ignore=all: %s\n", path);
3946	trace_printf("pathspec %s\n",
3947	((pathspec && pathspec->nr)
3948	? "has pathspec"
3949	: "no pathspec"));
3950
3951	/* Check if submodule path is explicitly mentioned in pathspec */
3952	if (pathspec) {
3953	for (ps_i = 0; ps_i < pathspec->nr; ps_i++) {
3954	const char *m = pathspec->items[ps_i].match;
3955	if (!m)
3956	continue;
3957	norm_pathspec = xstrdup(m);
3958	strip_dir_trailing_slashes(norm_pathspec);
3959	if (!strcmp(path, norm_pathspec)) {
3960	pathspec_matches = 1;
3961	FREE_AND_NULL(norm_pathspec);
3962	break;
3963	}
3964	FREE_AND_NULL(norm_pathspec);
3965	}
3966	}
3967
3968	/* If explicitly matched and forced, allow adding */
3969	if (pathspec_matches) {
3970	if (ignored_too && ignored_too > 0) {
3971	trace_printf("Add submodule due to --force: %s\n", path);
3972	return 0;
3973	} else {
3974	advise_if_enabled(ADVICE_ADD_IGNORED_FILE,
3975	_("Skipping submodule due to ignore=all: %s\n"
3976	"Use --force if you really want to "
3977	"add the submodule."), path);
3978	return 1;
3979	}
3980	}
3981
3982	/* No explicit pathspec match -> skip silently */
3983	trace_printf("Pathspec to submodule does not match explicitly: %s\n", path);
3984	return 1;
3985	}
3986
3987	static void update_callback(struct diff_queue_struct *q,
3988	struct diff_options opt UNUSED, void cbdata)
3989	{
3990	int i;
3991	struct update_callback_data *data = cbdata;
3992
3993	for (i = 0; i < q->nr; i++) {
3994	struct diff_filepair *p = q->queue[i];
3995	const char *path = p->one->path;
3996
3997	if (!data->include_sparse &&
3998	!path_in_sparse_checkout(path, data->index))
3999	continue;
4000
4001	switch (fix_unmerged_status(p, data)) {
4002	default:
4003	die(_("unexpected diff status %c"), p->status);
4004	case DIFF_STATUS_MODIFIED:
4005	case DIFF_STATUS_TYPE_CHANGED:
4006	if (skip_submodule(path, data->repo,
4007	data->pathspec,
4008	data->ignored_too))
4009	continue;
4010
4011	if (add_file_to_index(data->index, path, data->flags)) {
4012	if (!(data->flags & ADD_CACHE_IGNORE_ERRORS))
4013	die(_("updating files failed"));
4014	data->add_errors++;
4015	}
4016	break;
4017	case DIFF_STATUS_DELETED:
4018	if (data->flags & ADD_CACHE_IGNORE_REMOVAL)
4019	break;
4020	remove_file_from_index_with_flags(data->index, path, data->flags);
4021	break;
4022	}
4023	}
4024	}
4025
4026	int add_files_to_cache(struct repository repo, const char prefix,
4027	const struct pathspec pathspec, char ps_matched,
4028	int include_sparse, int flags, int ignored_too )
4029	{
4030	int inflight = !!repo->objects->transaction;
4031	struct odb_transaction *transaction;
4032	struct update_callback_data data;
4033	struct rev_info rev;
4034
4035	memset(&data, 0, sizeof(data));
4036	data.index = repo->index;
4037	data.include_sparse = include_sparse;
4038	data.flags = flags;
4039	data.repo = repo;
4040	data.ignored_too = ignored_too;
4041	data.pathspec = (struct pathspec *)pathspec;
4042
4043	repo_init_revisions(repo, &rev, prefix);
4044	setup_revisions(0, NULL, &rev, NULL);
4045	if (pathspec) {
4046	copy_pathspec(&rev.prune_data, pathspec);
4047	rev.ps_matched = ps_matched;
4048	}
4049	rev.diffopt.output_format = DIFF_FORMAT_CALLBACK;
4050	rev.diffopt.format_callback = update_callback;
4051	rev.diffopt.format_callback_data = &data;
4052	rev.diffopt.flags.override_submodule_config = 1;
4053	rev.diffopt.detect_rename = 0; /* staging worktree changes does not need renames */
4054	rev.max_count = 0; /* do not compare unmerged paths with stage #2 */
4055
4056	/*
4057	* Use an ODB transaction to optimize adding multiple objects.
4058	* This function is invoked from commands other than 'add', which
4059	* may not have their own transaction active.
4060	*/
4061	if (!inflight)
4062	odb_transaction_begin_or_die(repo->objects, &transaction, 0);
4063	run_diff_files(&rev, DIFF_RACY_IS_MODIFIED);
4064	if (!inflight)
4065	odb_transaction_commit(transaction);
4066
4067	release_revisions(&rev);
4068	return !!data.add_errors;
4069	}