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