Merge branch 'es/doc-worktree-guessremote-config'
[git.git] / read-cache.c
blobbfff271a3db92cfaaa93d23c5bf52297dda0419e
1 /*
2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
5 */
6 #define NO_THE_INDEX_COMPATIBILITY_MACROS
7 #include "cache.h"
8 #include "config.h"
9 #include "diff.h"
10 #include "diffcore.h"
11 #include "tempfile.h"
12 #include "lockfile.h"
13 #include "cache-tree.h"
14 #include "refs.h"
15 #include "dir.h"
16 #include "object-store.h"
17 #include "tree.h"
18 #include "commit.h"
19 #include "blob.h"
20 #include "resolve-undo.h"
21 #include "strbuf.h"
22 #include "varint.h"
23 #include "split-index.h"
24 #include "utf8.h"
25 #include "fsmonitor.h"
26 #include "thread-utils.h"
27 #include "progress.h"
29 /* Mask for the name length in ce_flags in the on-disk index */
31 #define CE_NAMEMASK (0x0fff)
33 /* Index extensions.
35 * The first letter should be 'A'..'Z' for extensions that are not
36 * necessary for a correct operation (i.e. optimization data).
37 * When new extensions are added that _needs_ to be understood in
38 * order to correctly interpret the index file, pick character that
39 * is outside the range, to cause the reader to abort.
42 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
43 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
44 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
45 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
46 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
47 #define CACHE_EXT_FSMONITOR 0x46534D4E /* "FSMN" */
48 #define CACHE_EXT_ENDOFINDEXENTRIES 0x454F4945 /* "EOIE" */
49 #define CACHE_EXT_INDEXENTRYOFFSETTABLE 0x49454F54 /* "IEOT" */
51 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
52 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
53 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
54 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED | FSMONITOR_CHANGED)
58 * This is an estimate of the pathname length in the index. We use
59 * this for V4 index files to guess the un-deltafied size of the index
60 * in memory because of pathname deltafication. This is not required
61 * for V2/V3 index formats because their pathnames are not compressed.
62 * If the initial amount of memory set aside is not sufficient, the
63 * mem pool will allocate extra memory.
65 #define CACHE_ENTRY_PATH_LENGTH 80
67 static inline struct cache_entry *mem_pool__ce_alloc(struct mem_pool *mem_pool, size_t len)
69 struct cache_entry *ce;
70 ce = mem_pool_alloc(mem_pool, cache_entry_size(len));
71 ce->mem_pool_allocated = 1;
72 return ce;
75 static inline struct cache_entry *mem_pool__ce_calloc(struct mem_pool *mem_pool, size_t len)
77 struct cache_entry * ce;
78 ce = mem_pool_calloc(mem_pool, 1, cache_entry_size(len));
79 ce->mem_pool_allocated = 1;
80 return ce;
83 static struct mem_pool *find_mem_pool(struct index_state *istate)
85 struct mem_pool **pool_ptr;
87 if (istate->split_index && istate->split_index->base)
88 pool_ptr = &istate->split_index->base->ce_mem_pool;
89 else
90 pool_ptr = &istate->ce_mem_pool;
92 if (!*pool_ptr)
93 mem_pool_init(pool_ptr, 0);
95 return *pool_ptr;
98 struct index_state the_index;
99 static const char *alternate_index_output;
101 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
103 istate->cache[nr] = ce;
104 add_name_hash(istate, ce);
107 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
109 struct cache_entry *old = istate->cache[nr];
111 replace_index_entry_in_base(istate, old, ce);
112 remove_name_hash(istate, old);
113 discard_cache_entry(old);
114 ce->ce_flags &= ~CE_HASHED;
115 set_index_entry(istate, nr, ce);
116 ce->ce_flags |= CE_UPDATE_IN_BASE;
117 mark_fsmonitor_invalid(istate, ce);
118 istate->cache_changed |= CE_ENTRY_CHANGED;
121 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
123 struct cache_entry *old_entry = istate->cache[nr], *new_entry;
124 int namelen = strlen(new_name);
126 new_entry = make_empty_cache_entry(istate, namelen);
127 copy_cache_entry(new_entry, old_entry);
128 new_entry->ce_flags &= ~CE_HASHED;
129 new_entry->ce_namelen = namelen;
130 new_entry->index = 0;
131 memcpy(new_entry->name, new_name, namelen + 1);
133 cache_tree_invalidate_path(istate, old_entry->name);
134 untracked_cache_remove_from_index(istate, old_entry->name);
135 remove_index_entry_at(istate, nr);
136 add_index_entry(istate, new_entry, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
139 void fill_stat_data(struct stat_data *sd, struct stat *st)
141 sd->sd_ctime.sec = (unsigned int)st->st_ctime;
142 sd->sd_mtime.sec = (unsigned int)st->st_mtime;
143 sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
144 sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
145 sd->sd_dev = st->st_dev;
146 sd->sd_ino = st->st_ino;
147 sd->sd_uid = st->st_uid;
148 sd->sd_gid = st->st_gid;
149 sd->sd_size = st->st_size;
152 int match_stat_data(const struct stat_data *sd, struct stat *st)
154 int changed = 0;
156 if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
157 changed |= MTIME_CHANGED;
158 if (trust_ctime && check_stat &&
159 sd->sd_ctime.sec != (unsigned int)st->st_ctime)
160 changed |= CTIME_CHANGED;
162 #ifdef USE_NSEC
163 if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
164 changed |= MTIME_CHANGED;
165 if (trust_ctime && check_stat &&
166 sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
167 changed |= CTIME_CHANGED;
168 #endif
170 if (check_stat) {
171 if (sd->sd_uid != (unsigned int) st->st_uid ||
172 sd->sd_gid != (unsigned int) st->st_gid)
173 changed |= OWNER_CHANGED;
174 if (sd->sd_ino != (unsigned int) st->st_ino)
175 changed |= INODE_CHANGED;
178 #ifdef USE_STDEV
180 * st_dev breaks on network filesystems where different
181 * clients will have different views of what "device"
182 * the filesystem is on
184 if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
185 changed |= INODE_CHANGED;
186 #endif
188 if (sd->sd_size != (unsigned int) st->st_size)
189 changed |= DATA_CHANGED;
191 return changed;
195 * This only updates the "non-critical" parts of the directory
196 * cache, ie the parts that aren't tracked by GIT, and only used
197 * to validate the cache.
199 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
201 fill_stat_data(&ce->ce_stat_data, st);
203 if (assume_unchanged)
204 ce->ce_flags |= CE_VALID;
206 if (S_ISREG(st->st_mode)) {
207 ce_mark_uptodate(ce);
208 mark_fsmonitor_valid(ce);
212 static int ce_compare_data(struct index_state *istate,
213 const struct cache_entry *ce,
214 struct stat *st)
216 int match = -1;
217 int fd = git_open_cloexec(ce->name, O_RDONLY);
219 if (fd >= 0) {
220 struct object_id oid;
221 if (!index_fd(istate, &oid, fd, st, OBJ_BLOB, ce->name, 0))
222 match = !oideq(&oid, &ce->oid);
223 /* index_fd() closed the file descriptor already */
225 return match;
228 static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
230 int match = -1;
231 void *buffer;
232 unsigned long size;
233 enum object_type type;
234 struct strbuf sb = STRBUF_INIT;
236 if (strbuf_readlink(&sb, ce->name, expected_size))
237 return -1;
239 buffer = read_object_file(&ce->oid, &type, &size);
240 if (buffer) {
241 if (size == sb.len)
242 match = memcmp(buffer, sb.buf, size);
243 free(buffer);
245 strbuf_release(&sb);
246 return match;
249 static int ce_compare_gitlink(const struct cache_entry *ce)
251 struct object_id oid;
254 * We don't actually require that the .git directory
255 * under GITLINK directory be a valid git directory. It
256 * might even be missing (in case nobody populated that
257 * sub-project).
259 * If so, we consider it always to match.
261 if (resolve_gitlink_ref(ce->name, "HEAD", &oid) < 0)
262 return 0;
263 return !oideq(&oid, &ce->oid);
266 static int ce_modified_check_fs(struct index_state *istate,
267 const struct cache_entry *ce,
268 struct stat *st)
270 switch (st->st_mode & S_IFMT) {
271 case S_IFREG:
272 if (ce_compare_data(istate, ce, st))
273 return DATA_CHANGED;
274 break;
275 case S_IFLNK:
276 if (ce_compare_link(ce, xsize_t(st->st_size)))
277 return DATA_CHANGED;
278 break;
279 case S_IFDIR:
280 if (S_ISGITLINK(ce->ce_mode))
281 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
282 /* else fallthrough */
283 default:
284 return TYPE_CHANGED;
286 return 0;
289 static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
291 unsigned int changed = 0;
293 if (ce->ce_flags & CE_REMOVE)
294 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
296 switch (ce->ce_mode & S_IFMT) {
297 case S_IFREG:
298 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
299 /* We consider only the owner x bit to be relevant for
300 * "mode changes"
302 if (trust_executable_bit &&
303 (0100 & (ce->ce_mode ^ st->st_mode)))
304 changed |= MODE_CHANGED;
305 break;
306 case S_IFLNK:
307 if (!S_ISLNK(st->st_mode) &&
308 (has_symlinks || !S_ISREG(st->st_mode)))
309 changed |= TYPE_CHANGED;
310 break;
311 case S_IFGITLINK:
312 /* We ignore most of the st_xxx fields for gitlinks */
313 if (!S_ISDIR(st->st_mode))
314 changed |= TYPE_CHANGED;
315 else if (ce_compare_gitlink(ce))
316 changed |= DATA_CHANGED;
317 return changed;
318 default:
319 BUG("unsupported ce_mode: %o", ce->ce_mode);
322 changed |= match_stat_data(&ce->ce_stat_data, st);
324 /* Racily smudged entry? */
325 if (!ce->ce_stat_data.sd_size) {
326 if (!is_empty_blob_sha1(ce->oid.hash))
327 changed |= DATA_CHANGED;
330 return changed;
333 static int is_racy_stat(const struct index_state *istate,
334 const struct stat_data *sd)
336 return (istate->timestamp.sec &&
337 #ifdef USE_NSEC
338 /* nanosecond timestamped files can also be racy! */
339 (istate->timestamp.sec < sd->sd_mtime.sec ||
340 (istate->timestamp.sec == sd->sd_mtime.sec &&
341 istate->timestamp.nsec <= sd->sd_mtime.nsec))
342 #else
343 istate->timestamp.sec <= sd->sd_mtime.sec
344 #endif
348 int is_racy_timestamp(const struct index_state *istate,
349 const struct cache_entry *ce)
351 return (!S_ISGITLINK(ce->ce_mode) &&
352 is_racy_stat(istate, &ce->ce_stat_data));
355 int match_stat_data_racy(const struct index_state *istate,
356 const struct stat_data *sd, struct stat *st)
358 if (is_racy_stat(istate, sd))
359 return MTIME_CHANGED;
360 return match_stat_data(sd, st);
363 int ie_match_stat(struct index_state *istate,
364 const struct cache_entry *ce, struct stat *st,
365 unsigned int options)
367 unsigned int changed;
368 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
369 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
370 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
371 int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
373 if (!ignore_fsmonitor)
374 refresh_fsmonitor(istate);
376 * If it's marked as always valid in the index, it's
377 * valid whatever the checked-out copy says.
379 * skip-worktree has the same effect with higher precedence
381 if (!ignore_skip_worktree && ce_skip_worktree(ce))
382 return 0;
383 if (!ignore_valid && (ce->ce_flags & CE_VALID))
384 return 0;
385 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID))
386 return 0;
389 * Intent-to-add entries have not been added, so the index entry
390 * by definition never matches what is in the work tree until it
391 * actually gets added.
393 if (ce_intent_to_add(ce))
394 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
396 changed = ce_match_stat_basic(ce, st);
399 * Within 1 second of this sequence:
400 * echo xyzzy >file && git-update-index --add file
401 * running this command:
402 * echo frotz >file
403 * would give a falsely clean cache entry. The mtime and
404 * length match the cache, and other stat fields do not change.
406 * We could detect this at update-index time (the cache entry
407 * being registered/updated records the same time as "now")
408 * and delay the return from git-update-index, but that would
409 * effectively mean we can make at most one commit per second,
410 * which is not acceptable. Instead, we check cache entries
411 * whose mtime are the same as the index file timestamp more
412 * carefully than others.
414 if (!changed && is_racy_timestamp(istate, ce)) {
415 if (assume_racy_is_modified)
416 changed |= DATA_CHANGED;
417 else
418 changed |= ce_modified_check_fs(istate, ce, st);
421 return changed;
424 int ie_modified(struct index_state *istate,
425 const struct cache_entry *ce,
426 struct stat *st, unsigned int options)
428 int changed, changed_fs;
430 changed = ie_match_stat(istate, ce, st, options);
431 if (!changed)
432 return 0;
434 * If the mode or type has changed, there's no point in trying
435 * to refresh the entry - it's not going to match
437 if (changed & (MODE_CHANGED | TYPE_CHANGED))
438 return changed;
441 * Immediately after read-tree or update-index --cacheinfo,
442 * the length field is zero, as we have never even read the
443 * lstat(2) information once, and we cannot trust DATA_CHANGED
444 * returned by ie_match_stat() which in turn was returned by
445 * ce_match_stat_basic() to signal that the filesize of the
446 * blob changed. We have to actually go to the filesystem to
447 * see if the contents match, and if so, should answer "unchanged".
449 * The logic does not apply to gitlinks, as ce_match_stat_basic()
450 * already has checked the actual HEAD from the filesystem in the
451 * subproject. If ie_match_stat() already said it is different,
452 * then we know it is.
454 if ((changed & DATA_CHANGED) &&
455 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
456 return changed;
458 changed_fs = ce_modified_check_fs(istate, ce, st);
459 if (changed_fs)
460 return changed | changed_fs;
461 return 0;
464 int base_name_compare(const char *name1, int len1, int mode1,
465 const char *name2, int len2, int mode2)
467 unsigned char c1, c2;
468 int len = len1 < len2 ? len1 : len2;
469 int cmp;
471 cmp = memcmp(name1, name2, len);
472 if (cmp)
473 return cmp;
474 c1 = name1[len];
475 c2 = name2[len];
476 if (!c1 && S_ISDIR(mode1))
477 c1 = '/';
478 if (!c2 && S_ISDIR(mode2))
479 c2 = '/';
480 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
484 * df_name_compare() is identical to base_name_compare(), except it
485 * compares conflicting directory/file entries as equal. Note that
486 * while a directory name compares as equal to a regular file, they
487 * then individually compare _differently_ to a filename that has
488 * a dot after the basename (because '\0' < '.' < '/').
490 * This is used by routines that want to traverse the git namespace
491 * but then handle conflicting entries together when possible.
493 int df_name_compare(const char *name1, int len1, int mode1,
494 const char *name2, int len2, int mode2)
496 int len = len1 < len2 ? len1 : len2, cmp;
497 unsigned char c1, c2;
499 cmp = memcmp(name1, name2, len);
500 if (cmp)
501 return cmp;
502 /* Directories and files compare equal (same length, same name) */
503 if (len1 == len2)
504 return 0;
505 c1 = name1[len];
506 if (!c1 && S_ISDIR(mode1))
507 c1 = '/';
508 c2 = name2[len];
509 if (!c2 && S_ISDIR(mode2))
510 c2 = '/';
511 if (c1 == '/' && !c2)
512 return 0;
513 if (c2 == '/' && !c1)
514 return 0;
515 return c1 - c2;
518 int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
520 size_t min_len = (len1 < len2) ? len1 : len2;
521 int cmp = memcmp(name1, name2, min_len);
522 if (cmp)
523 return cmp;
524 if (len1 < len2)
525 return -1;
526 if (len1 > len2)
527 return 1;
528 return 0;
531 int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
533 int cmp;
535 cmp = name_compare(name1, len1, name2, len2);
536 if (cmp)
537 return cmp;
539 if (stage1 < stage2)
540 return -1;
541 if (stage1 > stage2)
542 return 1;
543 return 0;
546 static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
548 int first, last;
550 first = 0;
551 last = istate->cache_nr;
552 while (last > first) {
553 int next = (last + first) >> 1;
554 struct cache_entry *ce = istate->cache[next];
555 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
556 if (!cmp)
557 return next;
558 if (cmp < 0) {
559 last = next;
560 continue;
562 first = next+1;
564 return -first-1;
567 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
569 return index_name_stage_pos(istate, name, namelen, 0);
572 int remove_index_entry_at(struct index_state *istate, int pos)
574 struct cache_entry *ce = istate->cache[pos];
576 record_resolve_undo(istate, ce);
577 remove_name_hash(istate, ce);
578 save_or_free_index_entry(istate, ce);
579 istate->cache_changed |= CE_ENTRY_REMOVED;
580 istate->cache_nr--;
581 if (pos >= istate->cache_nr)
582 return 0;
583 MOVE_ARRAY(istate->cache + pos, istate->cache + pos + 1,
584 istate->cache_nr - pos);
585 return 1;
589 * Remove all cache entries marked for removal, that is where
590 * CE_REMOVE is set in ce_flags. This is much more effective than
591 * calling remove_index_entry_at() for each entry to be removed.
593 void remove_marked_cache_entries(struct index_state *istate)
595 struct cache_entry **ce_array = istate->cache;
596 unsigned int i, j;
598 for (i = j = 0; i < istate->cache_nr; i++) {
599 if (ce_array[i]->ce_flags & CE_REMOVE) {
600 remove_name_hash(istate, ce_array[i]);
601 save_or_free_index_entry(istate, ce_array[i]);
603 else
604 ce_array[j++] = ce_array[i];
606 if (j == istate->cache_nr)
607 return;
608 istate->cache_changed |= CE_ENTRY_REMOVED;
609 istate->cache_nr = j;
612 int remove_file_from_index(struct index_state *istate, const char *path)
614 int pos = index_name_pos(istate, path, strlen(path));
615 if (pos < 0)
616 pos = -pos-1;
617 cache_tree_invalidate_path(istate, path);
618 untracked_cache_remove_from_index(istate, path);
619 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
620 remove_index_entry_at(istate, pos);
621 return 0;
624 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
626 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
629 static int index_name_pos_also_unmerged(struct index_state *istate,
630 const char *path, int namelen)
632 int pos = index_name_pos(istate, path, namelen);
633 struct cache_entry *ce;
635 if (pos >= 0)
636 return pos;
638 /* maybe unmerged? */
639 pos = -1 - pos;
640 if (pos >= istate->cache_nr ||
641 compare_name((ce = istate->cache[pos]), path, namelen))
642 return -1;
644 /* order of preference: stage 2, 1, 3 */
645 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
646 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
647 !compare_name(ce, path, namelen))
648 pos++;
649 return pos;
652 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
654 int len = ce_namelen(ce);
655 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
659 * If we add a filename that aliases in the cache, we will use the
660 * name that we already have - but we don't want to update the same
661 * alias twice, because that implies that there were actually two
662 * different files with aliasing names!
664 * So we use the CE_ADDED flag to verify that the alias was an old
665 * one before we accept it as
667 static struct cache_entry *create_alias_ce(struct index_state *istate,
668 struct cache_entry *ce,
669 struct cache_entry *alias)
671 int len;
672 struct cache_entry *new_entry;
674 if (alias->ce_flags & CE_ADDED)
675 die(_("will not add file alias '%s' ('%s' already exists in index)"),
676 ce->name, alias->name);
678 /* Ok, create the new entry using the name of the existing alias */
679 len = ce_namelen(alias);
680 new_entry = make_empty_cache_entry(istate, len);
681 memcpy(new_entry->name, alias->name, len);
682 copy_cache_entry(new_entry, ce);
683 save_or_free_index_entry(istate, ce);
684 return new_entry;
687 void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
689 struct object_id oid;
690 if (write_object_file("", 0, blob_type, &oid))
691 die(_("cannot create an empty blob in the object database"));
692 oidcpy(&ce->oid, &oid);
695 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
697 int namelen, was_same;
698 mode_t st_mode = st->st_mode;
699 struct cache_entry *ce, *alias = NULL;
700 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
701 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
702 int pretend = flags & ADD_CACHE_PRETEND;
703 int intent_only = flags & ADD_CACHE_INTENT;
704 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
705 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
706 int newflags = HASH_WRITE_OBJECT;
708 if (flags & HASH_RENORMALIZE)
709 newflags |= HASH_RENORMALIZE;
711 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
712 return error(_("%s: can only add regular files, symbolic links or git-directories"), path);
714 namelen = strlen(path);
715 if (S_ISDIR(st_mode)) {
716 while (namelen && path[namelen-1] == '/')
717 namelen--;
719 ce = make_empty_cache_entry(istate, namelen);
720 memcpy(ce->name, path, namelen);
721 ce->ce_namelen = namelen;
722 if (!intent_only)
723 fill_stat_cache_info(ce, st);
724 else
725 ce->ce_flags |= CE_INTENT_TO_ADD;
728 if (trust_executable_bit && has_symlinks) {
729 ce->ce_mode = create_ce_mode(st_mode);
730 } else {
731 /* If there is an existing entry, pick the mode bits and type
732 * from it, otherwise assume unexecutable regular file.
734 struct cache_entry *ent;
735 int pos = index_name_pos_also_unmerged(istate, path, namelen);
737 ent = (0 <= pos) ? istate->cache[pos] : NULL;
738 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
741 /* When core.ignorecase=true, determine if a directory of the same name but differing
742 * case already exists within the Git repository. If it does, ensure the directory
743 * case of the file being added to the repository matches (is folded into) the existing
744 * entry's directory case.
746 if (ignore_case) {
747 adjust_dirname_case(istate, ce->name);
749 if (!(flags & HASH_RENORMALIZE)) {
750 alias = index_file_exists(istate, ce->name,
751 ce_namelen(ce), ignore_case);
752 if (alias &&
753 !ce_stage(alias) &&
754 !ie_match_stat(istate, alias, st, ce_option)) {
755 /* Nothing changed, really */
756 if (!S_ISGITLINK(alias->ce_mode))
757 ce_mark_uptodate(alias);
758 alias->ce_flags |= CE_ADDED;
760 discard_cache_entry(ce);
761 return 0;
764 if (!intent_only) {
765 if (index_path(istate, &ce->oid, path, st, newflags)) {
766 discard_cache_entry(ce);
767 return error(_("unable to index file '%s'"), path);
769 } else
770 set_object_name_for_intent_to_add_entry(ce);
772 if (ignore_case && alias && different_name(ce, alias))
773 ce = create_alias_ce(istate, ce, alias);
774 ce->ce_flags |= CE_ADDED;
776 /* It was suspected to be racily clean, but it turns out to be Ok */
777 was_same = (alias &&
778 !ce_stage(alias) &&
779 oideq(&alias->oid, &ce->oid) &&
780 ce->ce_mode == alias->ce_mode);
782 if (pretend)
783 discard_cache_entry(ce);
784 else if (add_index_entry(istate, ce, add_option)) {
785 discard_cache_entry(ce);
786 return error(_("unable to add '%s' to index"), path);
788 if (verbose && !was_same)
789 printf("add '%s'\n", path);
790 return 0;
793 int add_file_to_index(struct index_state *istate, const char *path, int flags)
795 struct stat st;
796 if (lstat(path, &st))
797 die_errno(_("unable to stat '%s'"), path);
798 return add_to_index(istate, path, &st, flags);
801 struct cache_entry *make_empty_cache_entry(struct index_state *istate, size_t len)
803 return mem_pool__ce_calloc(find_mem_pool(istate), len);
806 struct cache_entry *make_empty_transient_cache_entry(size_t len)
808 return xcalloc(1, cache_entry_size(len));
811 struct cache_entry *make_cache_entry(struct index_state *istate,
812 unsigned int mode,
813 const struct object_id *oid,
814 const char *path,
815 int stage,
816 unsigned int refresh_options)
818 struct cache_entry *ce, *ret;
819 int len;
821 if (!verify_path(path, mode)) {
822 error(_("invalid path '%s'"), path);
823 return NULL;
826 len = strlen(path);
827 ce = make_empty_cache_entry(istate, len);
829 oidcpy(&ce->oid, oid);
830 memcpy(ce->name, path, len);
831 ce->ce_flags = create_ce_flags(stage);
832 ce->ce_namelen = len;
833 ce->ce_mode = create_ce_mode(mode);
835 ret = refresh_cache_entry(istate, ce, refresh_options);
836 if (ret != ce)
837 discard_cache_entry(ce);
838 return ret;
841 struct cache_entry *make_transient_cache_entry(unsigned int mode, const struct object_id *oid,
842 const char *path, int stage)
844 struct cache_entry *ce;
845 int len;
847 if (!verify_path(path, mode)) {
848 error(_("invalid path '%s'"), path);
849 return NULL;
852 len = strlen(path);
853 ce = make_empty_transient_cache_entry(len);
855 oidcpy(&ce->oid, oid);
856 memcpy(ce->name, path, len);
857 ce->ce_flags = create_ce_flags(stage);
858 ce->ce_namelen = len;
859 ce->ce_mode = create_ce_mode(mode);
861 return ce;
865 * Chmod an index entry with either +x or -x.
867 * Returns -1 if the chmod for the particular cache entry failed (if it's
868 * not a regular file), -2 if an invalid flip argument is passed in, 0
869 * otherwise.
871 int chmod_index_entry(struct index_state *istate, struct cache_entry *ce,
872 char flip)
874 if (!S_ISREG(ce->ce_mode))
875 return -1;
876 switch (flip) {
877 case '+':
878 ce->ce_mode |= 0111;
879 break;
880 case '-':
881 ce->ce_mode &= ~0111;
882 break;
883 default:
884 return -2;
886 cache_tree_invalidate_path(istate, ce->name);
887 ce->ce_flags |= CE_UPDATE_IN_BASE;
888 mark_fsmonitor_invalid(istate, ce);
889 istate->cache_changed |= CE_ENTRY_CHANGED;
891 return 0;
894 int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
896 int len = ce_namelen(a);
897 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
901 * We fundamentally don't like some paths: we don't want
902 * dot or dot-dot anywhere, and for obvious reasons don't
903 * want to recurse into ".git" either.
905 * Also, we don't want double slashes or slashes at the
906 * end that can make pathnames ambiguous.
908 static int verify_dotfile(const char *rest, unsigned mode)
911 * The first character was '.', but that
912 * has already been discarded, we now test
913 * the rest.
916 /* "." is not allowed */
917 if (*rest == '\0' || is_dir_sep(*rest))
918 return 0;
920 switch (*rest) {
922 * ".git" followed by NUL or slash is bad. Note that we match
923 * case-insensitively here, even if ignore_case is not set.
924 * This outlaws ".GIT" everywhere out of an abundance of caution,
925 * since there's really no good reason to allow it.
927 * Once we've seen ".git", we can also find ".gitmodules", etc (also
928 * case-insensitively).
930 case 'g':
931 case 'G':
932 if (rest[1] != 'i' && rest[1] != 'I')
933 break;
934 if (rest[2] != 't' && rest[2] != 'T')
935 break;
936 if (rest[3] == '\0' || is_dir_sep(rest[3]))
937 return 0;
938 if (S_ISLNK(mode)) {
939 rest += 3;
940 if (skip_iprefix(rest, "modules", &rest) &&
941 (*rest == '\0' || is_dir_sep(*rest)))
942 return 0;
944 break;
945 case '.':
946 if (rest[1] == '\0' || is_dir_sep(rest[1]))
947 return 0;
949 return 1;
952 int verify_path(const char *path, unsigned mode)
954 char c;
956 if (has_dos_drive_prefix(path))
957 return 0;
959 goto inside;
960 for (;;) {
961 if (!c)
962 return 1;
963 if (is_dir_sep(c)) {
964 inside:
965 if (protect_hfs) {
966 if (is_hfs_dotgit(path))
967 return 0;
968 if (S_ISLNK(mode)) {
969 if (is_hfs_dotgitmodules(path))
970 return 0;
973 if (protect_ntfs) {
974 if (is_ntfs_dotgit(path))
975 return 0;
976 if (S_ISLNK(mode)) {
977 if (is_ntfs_dotgitmodules(path))
978 return 0;
982 c = *path++;
983 if ((c == '.' && !verify_dotfile(path, mode)) ||
984 is_dir_sep(c) || c == '\0')
985 return 0;
987 c = *path++;
992 * Do we have another file that has the beginning components being a
993 * proper superset of the name we're trying to add?
995 static int has_file_name(struct index_state *istate,
996 const struct cache_entry *ce, int pos, int ok_to_replace)
998 int retval = 0;
999 int len = ce_namelen(ce);
1000 int stage = ce_stage(ce);
1001 const char *name = ce->name;
1003 while (pos < istate->cache_nr) {
1004 struct cache_entry *p = istate->cache[pos++];
1006 if (len >= ce_namelen(p))
1007 break;
1008 if (memcmp(name, p->name, len))
1009 break;
1010 if (ce_stage(p) != stage)
1011 continue;
1012 if (p->name[len] != '/')
1013 continue;
1014 if (p->ce_flags & CE_REMOVE)
1015 continue;
1016 retval = -1;
1017 if (!ok_to_replace)
1018 break;
1019 remove_index_entry_at(istate, --pos);
1021 return retval;
1026 * Like strcmp(), but also return the offset of the first change.
1027 * If strings are equal, return the length.
1029 int strcmp_offset(const char *s1, const char *s2, size_t *first_change)
1031 size_t k;
1033 if (!first_change)
1034 return strcmp(s1, s2);
1036 for (k = 0; s1[k] == s2[k]; k++)
1037 if (s1[k] == '\0')
1038 break;
1040 *first_change = k;
1041 return (unsigned char)s1[k] - (unsigned char)s2[k];
1045 * Do we have another file with a pathname that is a proper
1046 * subset of the name we're trying to add?
1048 * That is, is there another file in the index with a path
1049 * that matches a sub-directory in the given entry?
1051 static int has_dir_name(struct index_state *istate,
1052 const struct cache_entry *ce, int pos, int ok_to_replace)
1054 int retval = 0;
1055 int stage = ce_stage(ce);
1056 const char *name = ce->name;
1057 const char *slash = name + ce_namelen(ce);
1058 size_t len_eq_last;
1059 int cmp_last = 0;
1062 * We are frequently called during an iteration on a sorted
1063 * list of pathnames and while building a new index. Therefore,
1064 * there is a high probability that this entry will eventually
1065 * be appended to the index, rather than inserted in the middle.
1066 * If we can confirm that, we can avoid binary searches on the
1067 * components of the pathname.
1069 * Compare the entry's full path with the last path in the index.
1071 if (istate->cache_nr > 0) {
1072 cmp_last = strcmp_offset(name,
1073 istate->cache[istate->cache_nr - 1]->name,
1074 &len_eq_last);
1075 if (cmp_last > 0) {
1076 if (len_eq_last == 0) {
1078 * The entry sorts AFTER the last one in the
1079 * index and their paths have no common prefix,
1080 * so there cannot be a F/D conflict.
1082 return retval;
1083 } else {
1085 * The entry sorts AFTER the last one in the
1086 * index, but has a common prefix. Fall through
1087 * to the loop below to disect the entry's path
1088 * and see where the difference is.
1091 } else if (cmp_last == 0) {
1093 * The entry exactly matches the last one in the
1094 * index, but because of multiple stage and CE_REMOVE
1095 * items, we fall through and let the regular search
1096 * code handle it.
1101 for (;;) {
1102 size_t len;
1104 for (;;) {
1105 if (*--slash == '/')
1106 break;
1107 if (slash <= ce->name)
1108 return retval;
1110 len = slash - name;
1112 if (cmp_last > 0) {
1114 * (len + 1) is a directory boundary (including
1115 * the trailing slash). And since the loop is
1116 * decrementing "slash", the first iteration is
1117 * the longest directory prefix; subsequent
1118 * iterations consider parent directories.
1121 if (len + 1 <= len_eq_last) {
1123 * The directory prefix (including the trailing
1124 * slash) also appears as a prefix in the last
1125 * entry, so the remainder cannot collide (because
1126 * strcmp said the whole path was greater).
1128 * EQ: last: xxx/A
1129 * this: xxx/B
1131 * LT: last: xxx/file_A
1132 * this: xxx/file_B
1134 return retval;
1137 if (len > len_eq_last) {
1139 * This part of the directory prefix (excluding
1140 * the trailing slash) is longer than the known
1141 * equal portions, so this sub-directory cannot
1142 * collide with a file.
1144 * GT: last: xxxA
1145 * this: xxxB/file
1147 return retval;
1150 if (istate->cache_nr > 0 &&
1151 ce_namelen(istate->cache[istate->cache_nr - 1]) > len) {
1153 * The directory prefix lines up with part of
1154 * a longer file or directory name, but sorts
1155 * after it, so this sub-directory cannot
1156 * collide with a file.
1158 * last: xxx/yy-file (because '-' sorts before '/')
1159 * this: xxx/yy/abc
1161 return retval;
1165 * This is a possible collision. Fall through and
1166 * let the regular search code handle it.
1168 * last: xxx
1169 * this: xxx/file
1173 pos = index_name_stage_pos(istate, name, len, stage);
1174 if (pos >= 0) {
1176 * Found one, but not so fast. This could
1177 * be a marker that says "I was here, but
1178 * I am being removed". Such an entry is
1179 * not a part of the resulting tree, and
1180 * it is Ok to have a directory at the same
1181 * path.
1183 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
1184 retval = -1;
1185 if (!ok_to_replace)
1186 break;
1187 remove_index_entry_at(istate, pos);
1188 continue;
1191 else
1192 pos = -pos-1;
1195 * Trivial optimization: if we find an entry that
1196 * already matches the sub-directory, then we know
1197 * we're ok, and we can exit.
1199 while (pos < istate->cache_nr) {
1200 struct cache_entry *p = istate->cache[pos];
1201 if ((ce_namelen(p) <= len) ||
1202 (p->name[len] != '/') ||
1203 memcmp(p->name, name, len))
1204 break; /* not our subdirectory */
1205 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
1207 * p is at the same stage as our entry, and
1208 * is a subdirectory of what we are looking
1209 * at, so we cannot have conflicts at our
1210 * level or anything shorter.
1212 return retval;
1213 pos++;
1216 return retval;
1219 /* We may be in a situation where we already have path/file and path
1220 * is being added, or we already have path and path/file is being
1221 * added. Either one would result in a nonsense tree that has path
1222 * twice when git-write-tree tries to write it out. Prevent it.
1224 * If ok-to-replace is specified, we remove the conflicting entries
1225 * from the cache so the caller should recompute the insert position.
1226 * When this happens, we return non-zero.
1228 static int check_file_directory_conflict(struct index_state *istate,
1229 const struct cache_entry *ce,
1230 int pos, int ok_to_replace)
1232 int retval;
1235 * When ce is an "I am going away" entry, we allow it to be added
1237 if (ce->ce_flags & CE_REMOVE)
1238 return 0;
1241 * We check if the path is a sub-path of a subsequent pathname
1242 * first, since removing those will not change the position
1243 * in the array.
1245 retval = has_file_name(istate, ce, pos, ok_to_replace);
1248 * Then check if the path might have a clashing sub-directory
1249 * before it.
1251 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
1254 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
1256 int pos;
1257 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
1258 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
1259 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
1260 int new_only = option & ADD_CACHE_NEW_ONLY;
1262 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1263 cache_tree_invalidate_path(istate, ce->name);
1266 * If this entry's path sorts after the last entry in the index,
1267 * we can avoid searching for it.
1269 if (istate->cache_nr > 0 &&
1270 strcmp(ce->name, istate->cache[istate->cache_nr - 1]->name) > 0)
1271 pos = -istate->cache_nr - 1;
1272 else
1273 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1275 /* existing match? Just replace it. */
1276 if (pos >= 0) {
1277 if (!new_only)
1278 replace_index_entry(istate, pos, ce);
1279 return 0;
1281 pos = -pos-1;
1283 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1284 untracked_cache_add_to_index(istate, ce->name);
1287 * Inserting a merged entry ("stage 0") into the index
1288 * will always replace all non-merged entries..
1290 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1291 while (ce_same_name(istate->cache[pos], ce)) {
1292 ok_to_add = 1;
1293 if (!remove_index_entry_at(istate, pos))
1294 break;
1298 if (!ok_to_add)
1299 return -1;
1300 if (!verify_path(ce->name, ce->ce_mode))
1301 return error(_("invalid path '%s'"), ce->name);
1303 if (!skip_df_check &&
1304 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
1305 if (!ok_to_replace)
1306 return error(_("'%s' appears as both a file and as a directory"),
1307 ce->name);
1308 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1309 pos = -pos-1;
1311 return pos + 1;
1314 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1316 int pos;
1318 if (option & ADD_CACHE_JUST_APPEND)
1319 pos = istate->cache_nr;
1320 else {
1321 int ret;
1322 ret = add_index_entry_with_check(istate, ce, option);
1323 if (ret <= 0)
1324 return ret;
1325 pos = ret - 1;
1328 /* Make sure the array is big enough .. */
1329 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1331 /* Add it in.. */
1332 istate->cache_nr++;
1333 if (istate->cache_nr > pos + 1)
1334 MOVE_ARRAY(istate->cache + pos + 1, istate->cache + pos,
1335 istate->cache_nr - pos - 1);
1336 set_index_entry(istate, pos, ce);
1337 istate->cache_changed |= CE_ENTRY_ADDED;
1338 return 0;
1342 * "refresh" does not calculate a new sha1 file or bring the
1343 * cache up-to-date for mode/content changes. But what it
1344 * _does_ do is to "re-match" the stat information of a file
1345 * with the cache, so that you can refresh the cache for a
1346 * file that hasn't been changed but where the stat entry is
1347 * out of date.
1349 * For example, you'd want to do this after doing a "git-read-tree",
1350 * to link up the stat cache details with the proper files.
1352 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1353 struct cache_entry *ce,
1354 unsigned int options, int *err,
1355 int *changed_ret)
1357 struct stat st;
1358 struct cache_entry *updated;
1359 int changed;
1360 int refresh = options & CE_MATCH_REFRESH;
1361 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1362 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1363 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1364 int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
1366 if (!refresh || ce_uptodate(ce))
1367 return ce;
1369 if (!ignore_fsmonitor)
1370 refresh_fsmonitor(istate);
1372 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1373 * that the change to the work tree does not matter and told
1374 * us not to worry.
1376 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1377 ce_mark_uptodate(ce);
1378 return ce;
1380 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1381 ce_mark_uptodate(ce);
1382 return ce;
1384 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID)) {
1385 ce_mark_uptodate(ce);
1386 return ce;
1389 if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1390 if (ignore_missing)
1391 return ce;
1392 if (err)
1393 *err = ENOENT;
1394 return NULL;
1397 if (lstat(ce->name, &st) < 0) {
1398 if (ignore_missing && errno == ENOENT)
1399 return ce;
1400 if (err)
1401 *err = errno;
1402 return NULL;
1405 changed = ie_match_stat(istate, ce, &st, options);
1406 if (changed_ret)
1407 *changed_ret = changed;
1408 if (!changed) {
1410 * The path is unchanged. If we were told to ignore
1411 * valid bit, then we did the actual stat check and
1412 * found that the entry is unmodified. If the entry
1413 * is not marked VALID, this is the place to mark it
1414 * valid again, under "assume unchanged" mode.
1416 if (ignore_valid && assume_unchanged &&
1417 !(ce->ce_flags & CE_VALID))
1418 ; /* mark this one VALID again */
1419 else {
1421 * We do not mark the index itself "modified"
1422 * because CE_UPTODATE flag is in-core only;
1423 * we are not going to write this change out.
1425 if (!S_ISGITLINK(ce->ce_mode)) {
1426 ce_mark_uptodate(ce);
1427 mark_fsmonitor_valid(ce);
1429 return ce;
1433 if (ie_modified(istate, ce, &st, options)) {
1434 if (err)
1435 *err = EINVAL;
1436 return NULL;
1439 updated = make_empty_cache_entry(istate, ce_namelen(ce));
1440 copy_cache_entry(updated, ce);
1441 memcpy(updated->name, ce->name, ce->ce_namelen + 1);
1442 fill_stat_cache_info(updated, &st);
1444 * If ignore_valid is not set, we should leave CE_VALID bit
1445 * alone. Otherwise, paths marked with --no-assume-unchanged
1446 * (i.e. things to be edited) will reacquire CE_VALID bit
1447 * automatically, which is not really what we want.
1449 if (!ignore_valid && assume_unchanged &&
1450 !(ce->ce_flags & CE_VALID))
1451 updated->ce_flags &= ~CE_VALID;
1453 /* istate->cache_changed is updated in the caller */
1454 return updated;
1457 static void show_file(const char * fmt, const char * name, int in_porcelain,
1458 int * first, const char *header_msg)
1460 if (in_porcelain && *first && header_msg) {
1461 printf("%s\n", header_msg);
1462 *first = 0;
1464 printf(fmt, name);
1467 int refresh_index(struct index_state *istate, unsigned int flags,
1468 const struct pathspec *pathspec,
1469 char *seen, const char *header_msg)
1471 int i;
1472 int has_errors = 0;
1473 int really = (flags & REFRESH_REALLY) != 0;
1474 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1475 int quiet = (flags & REFRESH_QUIET) != 0;
1476 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1477 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1478 int first = 1;
1479 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1480 unsigned int options = (CE_MATCH_REFRESH |
1481 (really ? CE_MATCH_IGNORE_VALID : 0) |
1482 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1483 const char *modified_fmt;
1484 const char *deleted_fmt;
1485 const char *typechange_fmt;
1486 const char *added_fmt;
1487 const char *unmerged_fmt;
1488 struct progress *progress = NULL;
1490 if (flags & REFRESH_PROGRESS && isatty(2))
1491 progress = start_delayed_progress(_("Refresh index"),
1492 istate->cache_nr);
1494 trace_performance_enter();
1495 modified_fmt = in_porcelain ? "M\t%s\n" : "%s: needs update\n";
1496 deleted_fmt = in_porcelain ? "D\t%s\n" : "%s: needs update\n";
1497 typechange_fmt = in_porcelain ? "T\t%s\n" : "%s: needs update\n";
1498 added_fmt = in_porcelain ? "A\t%s\n" : "%s: needs update\n";
1499 unmerged_fmt = in_porcelain ? "U\t%s\n" : "%s: needs merge\n";
1501 * Use the multi-threaded preload_index() to refresh most of the
1502 * cache entries quickly then in the single threaded loop below,
1503 * we only have to do the special cases that are left.
1505 preload_index(istate, pathspec, 0);
1506 for (i = 0; i < istate->cache_nr; i++) {
1507 struct cache_entry *ce, *new_entry;
1508 int cache_errno = 0;
1509 int changed = 0;
1510 int filtered = 0;
1512 ce = istate->cache[i];
1513 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1514 continue;
1516 if (pathspec && !ce_path_match(istate, ce, pathspec, seen))
1517 filtered = 1;
1519 if (ce_stage(ce)) {
1520 while ((i < istate->cache_nr) &&
1521 ! strcmp(istate->cache[i]->name, ce->name))
1522 i++;
1523 i--;
1524 if (allow_unmerged)
1525 continue;
1526 if (!filtered)
1527 show_file(unmerged_fmt, ce->name, in_porcelain,
1528 &first, header_msg);
1529 has_errors = 1;
1530 continue;
1533 if (filtered)
1534 continue;
1536 new_entry = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1537 if (new_entry == ce)
1538 continue;
1539 if (progress)
1540 display_progress(progress, i);
1541 if (!new_entry) {
1542 const char *fmt;
1544 if (really && cache_errno == EINVAL) {
1545 /* If we are doing --really-refresh that
1546 * means the index is not valid anymore.
1548 ce->ce_flags &= ~CE_VALID;
1549 ce->ce_flags |= CE_UPDATE_IN_BASE;
1550 mark_fsmonitor_invalid(istate, ce);
1551 istate->cache_changed |= CE_ENTRY_CHANGED;
1553 if (quiet)
1554 continue;
1556 if (cache_errno == ENOENT)
1557 fmt = deleted_fmt;
1558 else if (ce_intent_to_add(ce))
1559 fmt = added_fmt; /* must be before other checks */
1560 else if (changed & TYPE_CHANGED)
1561 fmt = typechange_fmt;
1562 else
1563 fmt = modified_fmt;
1564 show_file(fmt,
1565 ce->name, in_porcelain, &first, header_msg);
1566 has_errors = 1;
1567 continue;
1570 replace_index_entry(istate, i, new_entry);
1572 if (progress) {
1573 display_progress(progress, istate->cache_nr);
1574 stop_progress(&progress);
1576 trace_performance_leave("refresh index");
1577 return has_errors;
1580 struct cache_entry *refresh_cache_entry(struct index_state *istate,
1581 struct cache_entry *ce,
1582 unsigned int options)
1584 return refresh_cache_ent(istate, ce, options, NULL, NULL);
1588 /*****************************************************************
1589 * Index File I/O
1590 *****************************************************************/
1592 #define INDEX_FORMAT_DEFAULT 3
1594 static unsigned int get_index_format_default(void)
1596 char *envversion = getenv("GIT_INDEX_VERSION");
1597 char *endp;
1598 int value;
1599 unsigned int version = INDEX_FORMAT_DEFAULT;
1601 if (!envversion) {
1602 if (!git_config_get_int("index.version", &value))
1603 version = value;
1604 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1605 warning(_("index.version set, but the value is invalid.\n"
1606 "Using version %i"), INDEX_FORMAT_DEFAULT);
1607 return INDEX_FORMAT_DEFAULT;
1609 return version;
1612 version = strtoul(envversion, &endp, 10);
1613 if (*endp ||
1614 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1615 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1616 "Using version %i"), INDEX_FORMAT_DEFAULT);
1617 version = INDEX_FORMAT_DEFAULT;
1619 return version;
1623 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1624 * Again - this is just a (very strong in practice) heuristic that
1625 * the inode hasn't changed.
1627 * We save the fields in big-endian order to allow using the
1628 * index file over NFS transparently.
1630 struct ondisk_cache_entry {
1631 struct cache_time ctime;
1632 struct cache_time mtime;
1633 uint32_t dev;
1634 uint32_t ino;
1635 uint32_t mode;
1636 uint32_t uid;
1637 uint32_t gid;
1638 uint32_t size;
1639 unsigned char sha1[20];
1640 uint16_t flags;
1641 char name[FLEX_ARRAY]; /* more */
1645 * This struct is used when CE_EXTENDED bit is 1
1646 * The struct must match ondisk_cache_entry exactly from
1647 * ctime till flags
1649 struct ondisk_cache_entry_extended {
1650 struct cache_time ctime;
1651 struct cache_time mtime;
1652 uint32_t dev;
1653 uint32_t ino;
1654 uint32_t mode;
1655 uint32_t uid;
1656 uint32_t gid;
1657 uint32_t size;
1658 unsigned char sha1[20];
1659 uint16_t flags;
1660 uint16_t flags2;
1661 char name[FLEX_ARRAY]; /* more */
1664 /* These are only used for v3 or lower */
1665 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1666 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1667 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1668 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1669 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1670 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1671 ondisk_cache_entry_size(ce_namelen(ce)))
1673 /* Allow fsck to force verification of the index checksum. */
1674 int verify_index_checksum;
1676 /* Allow fsck to force verification of the cache entry order. */
1677 int verify_ce_order;
1679 static int verify_hdr(const struct cache_header *hdr, unsigned long size)
1681 git_hash_ctx c;
1682 unsigned char hash[GIT_MAX_RAWSZ];
1683 int hdr_version;
1685 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1686 return error(_("bad signature 0x%08x"), hdr->hdr_signature);
1687 hdr_version = ntohl(hdr->hdr_version);
1688 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1689 return error(_("bad index version %d"), hdr_version);
1691 if (!verify_index_checksum)
1692 return 0;
1694 the_hash_algo->init_fn(&c);
1695 the_hash_algo->update_fn(&c, hdr, size - the_hash_algo->rawsz);
1696 the_hash_algo->final_fn(hash, &c);
1697 if (!hasheq(hash, (unsigned char *)hdr + size - the_hash_algo->rawsz))
1698 return error(_("bad index file sha1 signature"));
1699 return 0;
1702 static int read_index_extension(struct index_state *istate,
1703 const char *ext, const char *data, unsigned long sz)
1705 switch (CACHE_EXT(ext)) {
1706 case CACHE_EXT_TREE:
1707 istate->cache_tree = cache_tree_read(data, sz);
1708 break;
1709 case CACHE_EXT_RESOLVE_UNDO:
1710 istate->resolve_undo = resolve_undo_read(data, sz);
1711 break;
1712 case CACHE_EXT_LINK:
1713 if (read_link_extension(istate, data, sz))
1714 return -1;
1715 break;
1716 case CACHE_EXT_UNTRACKED:
1717 istate->untracked = read_untracked_extension(data, sz);
1718 break;
1719 case CACHE_EXT_FSMONITOR:
1720 read_fsmonitor_extension(istate, data, sz);
1721 break;
1722 case CACHE_EXT_ENDOFINDEXENTRIES:
1723 case CACHE_EXT_INDEXENTRYOFFSETTABLE:
1724 /* already handled in do_read_index() */
1725 break;
1726 default:
1727 if (*ext < 'A' || 'Z' < *ext)
1728 return error(_("index uses %.4s extension, which we do not understand"),
1729 ext);
1730 fprintf_ln(stderr, _("ignoring %.4s extension"), ext);
1731 break;
1733 return 0;
1736 int hold_locked_index(struct lock_file *lk, int lock_flags)
1738 return hold_lock_file_for_update(lk, get_index_file(), lock_flags);
1741 int read_index(struct index_state *istate)
1743 return read_index_from(istate, get_index_file(), get_git_dir());
1746 static struct cache_entry *create_from_disk(struct mem_pool *ce_mem_pool,
1747 unsigned int version,
1748 struct ondisk_cache_entry *ondisk,
1749 unsigned long *ent_size,
1750 const struct cache_entry *previous_ce)
1752 struct cache_entry *ce;
1753 size_t len;
1754 const char *name;
1755 unsigned int flags;
1756 size_t copy_len = 0;
1758 * Adjacent cache entries tend to share the leading paths, so it makes
1759 * sense to only store the differences in later entries. In the v4
1760 * on-disk format of the index, each on-disk cache entry stores the
1761 * number of bytes to be stripped from the end of the previous name,
1762 * and the bytes to append to the result, to come up with its name.
1764 int expand_name_field = version == 4;
1766 /* On-disk flags are just 16 bits */
1767 flags = get_be16(&ondisk->flags);
1768 len = flags & CE_NAMEMASK;
1770 if (flags & CE_EXTENDED) {
1771 struct ondisk_cache_entry_extended *ondisk2;
1772 int extended_flags;
1773 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1774 extended_flags = get_be16(&ondisk2->flags2) << 16;
1775 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1776 if (extended_flags & ~CE_EXTENDED_FLAGS)
1777 die(_("unknown index entry format 0x%08x"), extended_flags);
1778 flags |= extended_flags;
1779 name = ondisk2->name;
1781 else
1782 name = ondisk->name;
1784 if (expand_name_field) {
1785 const unsigned char *cp = (const unsigned char *)name;
1786 size_t strip_len, previous_len;
1788 /* If we're at the begining of a block, ignore the previous name */
1789 strip_len = decode_varint(&cp);
1790 if (previous_ce) {
1791 previous_len = previous_ce->ce_namelen;
1792 if (previous_len < strip_len)
1793 die(_("malformed name field in the index, near path '%s'"),
1794 previous_ce->name);
1795 copy_len = previous_len - strip_len;
1797 name = (const char *)cp;
1800 if (len == CE_NAMEMASK) {
1801 len = strlen(name);
1802 if (expand_name_field)
1803 len += copy_len;
1806 ce = mem_pool__ce_alloc(ce_mem_pool, len);
1808 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1809 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1810 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1811 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1812 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1813 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1814 ce->ce_mode = get_be32(&ondisk->mode);
1815 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1816 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1817 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
1818 ce->ce_flags = flags & ~CE_NAMEMASK;
1819 ce->ce_namelen = len;
1820 ce->index = 0;
1821 hashcpy(ce->oid.hash, ondisk->sha1);
1823 if (expand_name_field) {
1824 if (copy_len)
1825 memcpy(ce->name, previous_ce->name, copy_len);
1826 memcpy(ce->name + copy_len, name, len + 1 - copy_len);
1827 *ent_size = (name - ((char *)ondisk)) + len + 1 - copy_len;
1828 } else {
1829 memcpy(ce->name, name, len + 1);
1830 *ent_size = ondisk_ce_size(ce);
1832 return ce;
1835 static void check_ce_order(struct index_state *istate)
1837 unsigned int i;
1839 if (!verify_ce_order)
1840 return;
1842 for (i = 1; i < istate->cache_nr; i++) {
1843 struct cache_entry *ce = istate->cache[i - 1];
1844 struct cache_entry *next_ce = istate->cache[i];
1845 int name_compare = strcmp(ce->name, next_ce->name);
1847 if (0 < name_compare)
1848 die(_("unordered stage entries in index"));
1849 if (!name_compare) {
1850 if (!ce_stage(ce))
1851 die(_("multiple stage entries for merged file '%s'"),
1852 ce->name);
1853 if (ce_stage(ce) > ce_stage(next_ce))
1854 die(_("unordered stage entries for '%s'"),
1855 ce->name);
1860 static void tweak_untracked_cache(struct index_state *istate)
1862 switch (git_config_get_untracked_cache()) {
1863 case -1: /* keep: do nothing */
1864 break;
1865 case 0: /* false */
1866 remove_untracked_cache(istate);
1867 break;
1868 case 1: /* true */
1869 add_untracked_cache(istate);
1870 break;
1871 default: /* unknown value: do nothing */
1872 break;
1876 static void tweak_split_index(struct index_state *istate)
1878 switch (git_config_get_split_index()) {
1879 case -1: /* unset: do nothing */
1880 break;
1881 case 0: /* false */
1882 remove_split_index(istate);
1883 break;
1884 case 1: /* true */
1885 add_split_index(istate);
1886 break;
1887 default: /* unknown value: do nothing */
1888 break;
1892 static void post_read_index_from(struct index_state *istate)
1894 check_ce_order(istate);
1895 tweak_untracked_cache(istate);
1896 tweak_split_index(istate);
1897 tweak_fsmonitor(istate);
1900 static size_t estimate_cache_size_from_compressed(unsigned int entries)
1902 return entries * (sizeof(struct cache_entry) + CACHE_ENTRY_PATH_LENGTH);
1905 static size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1907 long per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1910 * Account for potential alignment differences.
1912 per_entry += align_padding_size(sizeof(struct cache_entry), -sizeof(struct ondisk_cache_entry));
1913 return ondisk_size + entries * per_entry;
1916 struct index_entry_offset
1918 /* starting byte offset into index file, count of index entries in this block */
1919 int offset, nr;
1922 struct index_entry_offset_table
1924 int nr;
1925 struct index_entry_offset entries[FLEX_ARRAY];
1928 static struct index_entry_offset_table *read_ieot_extension(const char *mmap, size_t mmap_size, size_t offset);
1929 static void write_ieot_extension(struct strbuf *sb, struct index_entry_offset_table *ieot);
1931 static size_t read_eoie_extension(const char *mmap, size_t mmap_size);
1932 static void write_eoie_extension(struct strbuf *sb, git_hash_ctx *eoie_context, size_t offset);
1934 struct load_index_extensions
1936 pthread_t pthread;
1937 struct index_state *istate;
1938 const char *mmap;
1939 size_t mmap_size;
1940 unsigned long src_offset;
1943 static void *load_index_extensions(void *_data)
1945 struct load_index_extensions *p = _data;
1946 unsigned long src_offset = p->src_offset;
1948 while (src_offset <= p->mmap_size - the_hash_algo->rawsz - 8) {
1949 /* After an array of active_nr index entries,
1950 * there can be arbitrary number of extended
1951 * sections, each of which is prefixed with
1952 * extension name (4-byte) and section length
1953 * in 4-byte network byte order.
1955 uint32_t extsize = get_be32(p->mmap + src_offset + 4);
1956 if (read_index_extension(p->istate,
1957 p->mmap + src_offset,
1958 p->mmap + src_offset + 8,
1959 extsize) < 0) {
1960 munmap((void *)p->mmap, p->mmap_size);
1961 die(_("index file corrupt"));
1963 src_offset += 8;
1964 src_offset += extsize;
1967 return NULL;
1971 * A helper function that will load the specified range of cache entries
1972 * from the memory mapped file and add them to the given index.
1974 static unsigned long load_cache_entry_block(struct index_state *istate,
1975 struct mem_pool *ce_mem_pool, int offset, int nr, const char *mmap,
1976 unsigned long start_offset, const struct cache_entry *previous_ce)
1978 int i;
1979 unsigned long src_offset = start_offset;
1981 for (i = offset; i < offset + nr; i++) {
1982 struct ondisk_cache_entry *disk_ce;
1983 struct cache_entry *ce;
1984 unsigned long consumed;
1986 disk_ce = (struct ondisk_cache_entry *)(mmap + src_offset);
1987 ce = create_from_disk(ce_mem_pool, istate->version, disk_ce, &consumed, previous_ce);
1988 set_index_entry(istate, i, ce);
1990 src_offset += consumed;
1991 previous_ce = ce;
1993 return src_offset - start_offset;
1996 static unsigned long load_all_cache_entries(struct index_state *istate,
1997 const char *mmap, size_t mmap_size, unsigned long src_offset)
1999 unsigned long consumed;
2001 if (istate->version == 4) {
2002 mem_pool_init(&istate->ce_mem_pool,
2003 estimate_cache_size_from_compressed(istate->cache_nr));
2004 } else {
2005 mem_pool_init(&istate->ce_mem_pool,
2006 estimate_cache_size(mmap_size, istate->cache_nr));
2009 consumed = load_cache_entry_block(istate, istate->ce_mem_pool,
2010 0, istate->cache_nr, mmap, src_offset, NULL);
2011 return consumed;
2015 * Mostly randomly chosen maximum thread counts: we
2016 * cap the parallelism to online_cpus() threads, and we want
2017 * to have at least 10000 cache entries per thread for it to
2018 * be worth starting a thread.
2021 #define THREAD_COST (10000)
2023 struct load_cache_entries_thread_data
2025 pthread_t pthread;
2026 struct index_state *istate;
2027 struct mem_pool *ce_mem_pool;
2028 int offset;
2029 const char *mmap;
2030 struct index_entry_offset_table *ieot;
2031 int ieot_start; /* starting index into the ieot array */
2032 int ieot_blocks; /* count of ieot entries to process */
2033 unsigned long consumed; /* return # of bytes in index file processed */
2037 * A thread proc to run the load_cache_entries() computation
2038 * across multiple background threads.
2040 static void *load_cache_entries_thread(void *_data)
2042 struct load_cache_entries_thread_data *p = _data;
2043 int i;
2045 /* iterate across all ieot blocks assigned to this thread */
2046 for (i = p->ieot_start; i < p->ieot_start + p->ieot_blocks; i++) {
2047 p->consumed += load_cache_entry_block(p->istate, p->ce_mem_pool,
2048 p->offset, p->ieot->entries[i].nr, p->mmap, p->ieot->entries[i].offset, NULL);
2049 p->offset += p->ieot->entries[i].nr;
2051 return NULL;
2054 static unsigned long load_cache_entries_threaded(struct index_state *istate, const char *mmap, size_t mmap_size,
2055 unsigned long src_offset, int nr_threads, struct index_entry_offset_table *ieot)
2057 int i, offset, ieot_blocks, ieot_start, err;
2058 struct load_cache_entries_thread_data *data;
2059 unsigned long consumed = 0;
2061 /* a little sanity checking */
2062 if (istate->name_hash_initialized)
2063 BUG("the name hash isn't thread safe");
2065 mem_pool_init(&istate->ce_mem_pool, 0);
2067 /* ensure we have no more threads than we have blocks to process */
2068 if (nr_threads > ieot->nr)
2069 nr_threads = ieot->nr;
2070 data = xcalloc(nr_threads, sizeof(*data));
2072 offset = ieot_start = 0;
2073 ieot_blocks = DIV_ROUND_UP(ieot->nr, nr_threads);
2074 for (i = 0; i < nr_threads; i++) {
2075 struct load_cache_entries_thread_data *p = &data[i];
2076 int nr, j;
2078 if (ieot_start + ieot_blocks > ieot->nr)
2079 ieot_blocks = ieot->nr - ieot_start;
2081 p->istate = istate;
2082 p->offset = offset;
2083 p->mmap = mmap;
2084 p->ieot = ieot;
2085 p->ieot_start = ieot_start;
2086 p->ieot_blocks = ieot_blocks;
2088 /* create a mem_pool for each thread */
2089 nr = 0;
2090 for (j = p->ieot_start; j < p->ieot_start + p->ieot_blocks; j++)
2091 nr += p->ieot->entries[j].nr;
2092 if (istate->version == 4) {
2093 mem_pool_init(&p->ce_mem_pool,
2094 estimate_cache_size_from_compressed(nr));
2095 } else {
2096 mem_pool_init(&p->ce_mem_pool,
2097 estimate_cache_size(mmap_size, nr));
2100 err = pthread_create(&p->pthread, NULL, load_cache_entries_thread, p);
2101 if (err)
2102 die(_("unable to create load_cache_entries thread: %s"), strerror(err));
2104 /* increment by the number of cache entries in the ieot block being processed */
2105 for (j = 0; j < ieot_blocks; j++)
2106 offset += ieot->entries[ieot_start + j].nr;
2107 ieot_start += ieot_blocks;
2110 for (i = 0; i < nr_threads; i++) {
2111 struct load_cache_entries_thread_data *p = &data[i];
2113 err = pthread_join(p->pthread, NULL);
2114 if (err)
2115 die(_("unable to join load_cache_entries thread: %s"), strerror(err));
2116 mem_pool_combine(istate->ce_mem_pool, p->ce_mem_pool);
2117 consumed += p->consumed;
2120 free(data);
2122 return consumed;
2125 /* remember to discard_cache() before reading a different cache! */
2126 int do_read_index(struct index_state *istate, const char *path, int must_exist)
2128 int fd;
2129 struct stat st;
2130 unsigned long src_offset;
2131 const struct cache_header *hdr;
2132 const char *mmap;
2133 size_t mmap_size;
2134 struct load_index_extensions p;
2135 size_t extension_offset = 0;
2136 int nr_threads, cpus;
2137 struct index_entry_offset_table *ieot = NULL;
2139 if (istate->initialized)
2140 return istate->cache_nr;
2142 istate->timestamp.sec = 0;
2143 istate->timestamp.nsec = 0;
2144 fd = open(path, O_RDONLY);
2145 if (fd < 0) {
2146 if (!must_exist && errno == ENOENT)
2147 return 0;
2148 die_errno(_("%s: index file open failed"), path);
2151 if (fstat(fd, &st))
2152 die_errno(_("%s: cannot stat the open index"), path);
2154 mmap_size = xsize_t(st.st_size);
2155 if (mmap_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
2156 die(_("%s: index file smaller than expected"), path);
2158 mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
2159 if (mmap == MAP_FAILED)
2160 die_errno(_("%s: unable to map index file"), path);
2161 close(fd);
2163 hdr = (const struct cache_header *)mmap;
2164 if (verify_hdr(hdr, mmap_size) < 0)
2165 goto unmap;
2167 hashcpy(istate->oid.hash, (const unsigned char *)hdr + mmap_size - the_hash_algo->rawsz);
2168 istate->version = ntohl(hdr->hdr_version);
2169 istate->cache_nr = ntohl(hdr->hdr_entries);
2170 istate->cache_alloc = alloc_nr(istate->cache_nr);
2171 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
2172 istate->initialized = 1;
2174 p.istate = istate;
2175 p.mmap = mmap;
2176 p.mmap_size = mmap_size;
2178 src_offset = sizeof(*hdr);
2180 if (git_config_get_index_threads(&nr_threads))
2181 nr_threads = 1;
2183 /* TODO: does creating more threads than cores help? */
2184 if (!nr_threads) {
2185 nr_threads = istate->cache_nr / THREAD_COST;
2186 cpus = online_cpus();
2187 if (nr_threads > cpus)
2188 nr_threads = cpus;
2191 if (!HAVE_THREADS)
2192 nr_threads = 1;
2194 if (nr_threads > 1) {
2195 extension_offset = read_eoie_extension(mmap, mmap_size);
2196 if (extension_offset) {
2197 int err;
2199 p.src_offset = extension_offset;
2200 err = pthread_create(&p.pthread, NULL, load_index_extensions, &p);
2201 if (err)
2202 die(_("unable to create load_index_extensions thread: %s"), strerror(err));
2204 nr_threads--;
2209 * Locate and read the index entry offset table so that we can use it
2210 * to multi-thread the reading of the cache entries.
2212 if (extension_offset && nr_threads > 1)
2213 ieot = read_ieot_extension(mmap, mmap_size, extension_offset);
2215 if (ieot) {
2216 src_offset += load_cache_entries_threaded(istate, mmap, mmap_size, src_offset, nr_threads, ieot);
2217 free(ieot);
2218 } else {
2219 src_offset += load_all_cache_entries(istate, mmap, mmap_size, src_offset);
2222 istate->timestamp.sec = st.st_mtime;
2223 istate->timestamp.nsec = ST_MTIME_NSEC(st);
2225 /* if we created a thread, join it otherwise load the extensions on the primary thread */
2226 if (extension_offset) {
2227 int ret = pthread_join(p.pthread, NULL);
2228 if (ret)
2229 die(_("unable to join load_index_extensions thread: %s"), strerror(ret));
2230 } else {
2231 p.src_offset = src_offset;
2232 load_index_extensions(&p);
2234 munmap((void *)mmap, mmap_size);
2235 return istate->cache_nr;
2237 unmap:
2238 munmap((void *)mmap, mmap_size);
2239 die(_("index file corrupt"));
2243 * Signal that the shared index is used by updating its mtime.
2245 * This way, shared index can be removed if they have not been used
2246 * for some time.
2248 static void freshen_shared_index(const char *shared_index, int warn)
2250 if (!check_and_freshen_file(shared_index, 1) && warn)
2251 warning(_("could not freshen shared index '%s'"), shared_index);
2254 int read_index_from(struct index_state *istate, const char *path,
2255 const char *gitdir)
2257 struct split_index *split_index;
2258 int ret;
2259 char *base_oid_hex;
2260 char *base_path;
2262 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2263 if (istate->initialized)
2264 return istate->cache_nr;
2266 trace_performance_enter();
2267 ret = do_read_index(istate, path, 0);
2268 trace_performance_leave("read cache %s", path);
2270 split_index = istate->split_index;
2271 if (!split_index || is_null_oid(&split_index->base_oid)) {
2272 post_read_index_from(istate);
2273 return ret;
2276 trace_performance_enter();
2277 if (split_index->base)
2278 discard_index(split_index->base);
2279 else
2280 split_index->base = xcalloc(1, sizeof(*split_index->base));
2282 base_oid_hex = oid_to_hex(&split_index->base_oid);
2283 base_path = xstrfmt("%s/sharedindex.%s", gitdir, base_oid_hex);
2284 ret = do_read_index(split_index->base, base_path, 1);
2285 if (!oideq(&split_index->base_oid, &split_index->base->oid))
2286 die(_("broken index, expect %s in %s, got %s"),
2287 base_oid_hex, base_path,
2288 oid_to_hex(&split_index->base->oid));
2290 freshen_shared_index(base_path, 0);
2291 merge_base_index(istate);
2292 post_read_index_from(istate);
2293 trace_performance_leave("read cache %s", base_path);
2294 free(base_path);
2295 return ret;
2298 int is_index_unborn(struct index_state *istate)
2300 return (!istate->cache_nr && !istate->timestamp.sec);
2303 int discard_index(struct index_state *istate)
2306 * Cache entries in istate->cache[] should have been allocated
2307 * from the memory pool associated with this index, or from an
2308 * associated split_index. There is no need to free individual
2309 * cache entries. validate_cache_entries can detect when this
2310 * assertion does not hold.
2312 validate_cache_entries(istate);
2314 resolve_undo_clear_index(istate);
2315 istate->cache_nr = 0;
2316 istate->cache_changed = 0;
2317 istate->timestamp.sec = 0;
2318 istate->timestamp.nsec = 0;
2319 free_name_hash(istate);
2320 cache_tree_free(&(istate->cache_tree));
2321 istate->initialized = 0;
2322 FREE_AND_NULL(istate->cache);
2323 istate->cache_alloc = 0;
2324 discard_split_index(istate);
2325 free_untracked_cache(istate->untracked);
2326 istate->untracked = NULL;
2328 if (istate->ce_mem_pool) {
2329 mem_pool_discard(istate->ce_mem_pool, should_validate_cache_entries());
2330 istate->ce_mem_pool = NULL;
2333 return 0;
2337 * Validate the cache entries of this index.
2338 * All cache entries associated with this index
2339 * should have been allocated by the memory pool
2340 * associated with this index, or by a referenced
2341 * split index.
2343 void validate_cache_entries(const struct index_state *istate)
2345 int i;
2347 if (!should_validate_cache_entries() ||!istate || !istate->initialized)
2348 return;
2350 for (i = 0; i < istate->cache_nr; i++) {
2351 if (!istate) {
2352 BUG("cache entry is not allocated from expected memory pool");
2353 } else if (!istate->ce_mem_pool ||
2354 !mem_pool_contains(istate->ce_mem_pool, istate->cache[i])) {
2355 if (!istate->split_index ||
2356 !istate->split_index->base ||
2357 !istate->split_index->base->ce_mem_pool ||
2358 !mem_pool_contains(istate->split_index->base->ce_mem_pool, istate->cache[i])) {
2359 BUG("cache entry is not allocated from expected memory pool");
2364 if (istate->split_index)
2365 validate_cache_entries(istate->split_index->base);
2368 int unmerged_index(const struct index_state *istate)
2370 int i;
2371 for (i = 0; i < istate->cache_nr; i++) {
2372 if (ce_stage(istate->cache[i]))
2373 return 1;
2375 return 0;
2378 int index_has_changes(struct index_state *istate,
2379 struct tree *tree,
2380 struct strbuf *sb)
2382 struct object_id cmp;
2383 int i;
2385 if (istate != &the_index) {
2386 BUG("index_has_changes cannot yet accept istate != &the_index; do_diff_cache needs updating first.");
2388 if (tree)
2389 cmp = tree->object.oid;
2390 if (tree || !get_oid_tree("HEAD", &cmp)) {
2391 struct diff_options opt;
2393 repo_diff_setup(the_repository, &opt);
2394 opt.flags.exit_with_status = 1;
2395 if (!sb)
2396 opt.flags.quick = 1;
2397 do_diff_cache(&cmp, &opt);
2398 diffcore_std(&opt);
2399 for (i = 0; sb && i < diff_queued_diff.nr; i++) {
2400 if (i)
2401 strbuf_addch(sb, ' ');
2402 strbuf_addstr(sb, diff_queued_diff.queue[i]->two->path);
2404 diff_flush(&opt);
2405 return opt.flags.has_changes != 0;
2406 } else {
2407 for (i = 0; sb && i < istate->cache_nr; i++) {
2408 if (i)
2409 strbuf_addch(sb, ' ');
2410 strbuf_addstr(sb, istate->cache[i]->name);
2412 return !!istate->cache_nr;
2416 #define WRITE_BUFFER_SIZE 8192
2417 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
2418 static unsigned long write_buffer_len;
2420 static int ce_write_flush(git_hash_ctx *context, int fd)
2422 unsigned int buffered = write_buffer_len;
2423 if (buffered) {
2424 the_hash_algo->update_fn(context, write_buffer, buffered);
2425 if (write_in_full(fd, write_buffer, buffered) < 0)
2426 return -1;
2427 write_buffer_len = 0;
2429 return 0;
2432 static int ce_write(git_hash_ctx *context, int fd, void *data, unsigned int len)
2434 while (len) {
2435 unsigned int buffered = write_buffer_len;
2436 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
2437 if (partial > len)
2438 partial = len;
2439 memcpy(write_buffer + buffered, data, partial);
2440 buffered += partial;
2441 if (buffered == WRITE_BUFFER_SIZE) {
2442 write_buffer_len = buffered;
2443 if (ce_write_flush(context, fd))
2444 return -1;
2445 buffered = 0;
2447 write_buffer_len = buffered;
2448 len -= partial;
2449 data = (char *) data + partial;
2451 return 0;
2454 static int write_index_ext_header(git_hash_ctx *context, git_hash_ctx *eoie_context,
2455 int fd, unsigned int ext, unsigned int sz)
2457 ext = htonl(ext);
2458 sz = htonl(sz);
2459 if (eoie_context) {
2460 the_hash_algo->update_fn(eoie_context, &ext, 4);
2461 the_hash_algo->update_fn(eoie_context, &sz, 4);
2463 return ((ce_write(context, fd, &ext, 4) < 0) ||
2464 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
2467 static int ce_flush(git_hash_ctx *context, int fd, unsigned char *hash)
2469 unsigned int left = write_buffer_len;
2471 if (left) {
2472 write_buffer_len = 0;
2473 the_hash_algo->update_fn(context, write_buffer, left);
2476 /* Flush first if not enough space for hash signature */
2477 if (left + the_hash_algo->rawsz > WRITE_BUFFER_SIZE) {
2478 if (write_in_full(fd, write_buffer, left) < 0)
2479 return -1;
2480 left = 0;
2483 /* Append the hash signature at the end */
2484 the_hash_algo->final_fn(write_buffer + left, context);
2485 hashcpy(hash, write_buffer + left);
2486 left += the_hash_algo->rawsz;
2487 return (write_in_full(fd, write_buffer, left) < 0) ? -1 : 0;
2490 static void ce_smudge_racily_clean_entry(struct index_state *istate,
2491 struct cache_entry *ce)
2494 * The only thing we care about in this function is to smudge the
2495 * falsely clean entry due to touch-update-touch race, so we leave
2496 * everything else as they are. We are called for entries whose
2497 * ce_stat_data.sd_mtime match the index file mtime.
2499 * Note that this actually does not do much for gitlinks, for
2500 * which ce_match_stat_basic() always goes to the actual
2501 * contents. The caller checks with is_racy_timestamp() which
2502 * always says "no" for gitlinks, so we are not called for them ;-)
2504 struct stat st;
2506 if (lstat(ce->name, &st) < 0)
2507 return;
2508 if (ce_match_stat_basic(ce, &st))
2509 return;
2510 if (ce_modified_check_fs(istate, ce, &st)) {
2511 /* This is "racily clean"; smudge it. Note that this
2512 * is a tricky code. At first glance, it may appear
2513 * that it can break with this sequence:
2515 * $ echo xyzzy >frotz
2516 * $ git-update-index --add frotz
2517 * $ : >frotz
2518 * $ sleep 3
2519 * $ echo filfre >nitfol
2520 * $ git-update-index --add nitfol
2522 * but it does not. When the second update-index runs,
2523 * it notices that the entry "frotz" has the same timestamp
2524 * as index, and if we were to smudge it by resetting its
2525 * size to zero here, then the object name recorded
2526 * in index is the 6-byte file but the cached stat information
2527 * becomes zero --- which would then match what we would
2528 * obtain from the filesystem next time we stat("frotz").
2530 * However, the second update-index, before calling
2531 * this function, notices that the cached size is 6
2532 * bytes and what is on the filesystem is an empty
2533 * file, and never calls us, so the cached size information
2534 * for "frotz" stays 6 which does not match the filesystem.
2536 ce->ce_stat_data.sd_size = 0;
2540 /* Copy miscellaneous fields but not the name */
2541 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
2542 struct cache_entry *ce)
2544 short flags;
2546 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
2547 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
2548 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
2549 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
2550 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
2551 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
2552 ondisk->mode = htonl(ce->ce_mode);
2553 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
2554 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
2555 ondisk->size = htonl(ce->ce_stat_data.sd_size);
2556 hashcpy(ondisk->sha1, ce->oid.hash);
2558 flags = ce->ce_flags & ~CE_NAMEMASK;
2559 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
2560 ondisk->flags = htons(flags);
2561 if (ce->ce_flags & CE_EXTENDED) {
2562 struct ondisk_cache_entry_extended *ondisk2;
2563 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
2564 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
2568 static int ce_write_entry(git_hash_ctx *c, int fd, struct cache_entry *ce,
2569 struct strbuf *previous_name, struct ondisk_cache_entry *ondisk)
2571 int size;
2572 int result;
2573 unsigned int saved_namelen;
2574 int stripped_name = 0;
2575 static unsigned char padding[8] = { 0x00 };
2577 if (ce->ce_flags & CE_STRIP_NAME) {
2578 saved_namelen = ce_namelen(ce);
2579 ce->ce_namelen = 0;
2580 stripped_name = 1;
2583 if (ce->ce_flags & CE_EXTENDED)
2584 size = offsetof(struct ondisk_cache_entry_extended, name);
2585 else
2586 size = offsetof(struct ondisk_cache_entry, name);
2588 if (!previous_name) {
2589 int len = ce_namelen(ce);
2590 copy_cache_entry_to_ondisk(ondisk, ce);
2591 result = ce_write(c, fd, ondisk, size);
2592 if (!result)
2593 result = ce_write(c, fd, ce->name, len);
2594 if (!result)
2595 result = ce_write(c, fd, padding, align_padding_size(size, len));
2596 } else {
2597 int common, to_remove, prefix_size;
2598 unsigned char to_remove_vi[16];
2599 for (common = 0;
2600 (ce->name[common] &&
2601 common < previous_name->len &&
2602 ce->name[common] == previous_name->buf[common]);
2603 common++)
2604 ; /* still matching */
2605 to_remove = previous_name->len - common;
2606 prefix_size = encode_varint(to_remove, to_remove_vi);
2608 copy_cache_entry_to_ondisk(ondisk, ce);
2609 result = ce_write(c, fd, ondisk, size);
2610 if (!result)
2611 result = ce_write(c, fd, to_remove_vi, prefix_size);
2612 if (!result)
2613 result = ce_write(c, fd, ce->name + common, ce_namelen(ce) - common);
2614 if (!result)
2615 result = ce_write(c, fd, padding, 1);
2617 strbuf_splice(previous_name, common, to_remove,
2618 ce->name + common, ce_namelen(ce) - common);
2620 if (stripped_name) {
2621 ce->ce_namelen = saved_namelen;
2622 ce->ce_flags &= ~CE_STRIP_NAME;
2625 return result;
2629 * This function verifies if index_state has the correct sha1 of the
2630 * index file. Don't die if we have any other failure, just return 0.
2632 static int verify_index_from(const struct index_state *istate, const char *path)
2634 int fd;
2635 ssize_t n;
2636 struct stat st;
2637 unsigned char hash[GIT_MAX_RAWSZ];
2639 if (!istate->initialized)
2640 return 0;
2642 fd = open(path, O_RDONLY);
2643 if (fd < 0)
2644 return 0;
2646 if (fstat(fd, &st))
2647 goto out;
2649 if (st.st_size < sizeof(struct cache_header) + the_hash_algo->rawsz)
2650 goto out;
2652 n = pread_in_full(fd, hash, the_hash_algo->rawsz, st.st_size - the_hash_algo->rawsz);
2653 if (n != the_hash_algo->rawsz)
2654 goto out;
2656 if (!hasheq(istate->oid.hash, hash))
2657 goto out;
2659 close(fd);
2660 return 1;
2662 out:
2663 close(fd);
2664 return 0;
2667 static int verify_index(const struct index_state *istate)
2669 return verify_index_from(istate, get_index_file());
2672 static int has_racy_timestamp(struct index_state *istate)
2674 int entries = istate->cache_nr;
2675 int i;
2677 for (i = 0; i < entries; i++) {
2678 struct cache_entry *ce = istate->cache[i];
2679 if (is_racy_timestamp(istate, ce))
2680 return 1;
2682 return 0;
2685 void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
2687 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
2688 verify_index(istate))
2689 write_locked_index(istate, lockfile, COMMIT_LOCK);
2690 else
2691 rollback_lock_file(lockfile);
2694 static int record_eoie(void)
2696 int val;
2698 if (!git_config_get_bool("index.recordendofindexentries", &val))
2699 return val;
2702 * As a convenience, the end of index entries extension
2703 * used for threading is written by default if the user
2704 * explicitly requested threaded index reads.
2706 return !git_config_get_index_threads(&val) && val != 1;
2709 static int record_ieot(void)
2711 int val;
2713 if (!git_config_get_bool("index.recordoffsettable", &val))
2714 return val;
2717 * As a convenience, the offset table used for threading is
2718 * written by default if the user explicitly requested
2719 * threaded index reads.
2721 return !git_config_get_index_threads(&val) && val != 1;
2725 * On success, `tempfile` is closed. If it is the temporary file
2726 * of a `struct lock_file`, we will therefore effectively perform
2727 * a 'close_lock_file_gently()`. Since that is an implementation
2728 * detail of lockfiles, callers of `do_write_index()` should not
2729 * rely on it.
2731 static int do_write_index(struct index_state *istate, struct tempfile *tempfile,
2732 int strip_extensions)
2734 uint64_t start = getnanotime();
2735 int newfd = tempfile->fd;
2736 git_hash_ctx c, eoie_c;
2737 struct cache_header hdr;
2738 int i, err = 0, removed, extended, hdr_version;
2739 struct cache_entry **cache = istate->cache;
2740 int entries = istate->cache_nr;
2741 struct stat st;
2742 struct ondisk_cache_entry_extended ondisk;
2743 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
2744 int drop_cache_tree = istate->drop_cache_tree;
2745 off_t offset;
2746 int ieot_entries = 1;
2747 struct index_entry_offset_table *ieot = NULL;
2748 int nr, nr_threads;
2750 for (i = removed = extended = 0; i < entries; i++) {
2751 if (cache[i]->ce_flags & CE_REMOVE)
2752 removed++;
2754 /* reduce extended entries if possible */
2755 cache[i]->ce_flags &= ~CE_EXTENDED;
2756 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2757 extended++;
2758 cache[i]->ce_flags |= CE_EXTENDED;
2762 if (!istate->version) {
2763 istate->version = get_index_format_default();
2764 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0))
2765 init_split_index(istate);
2768 /* demote version 3 to version 2 when the latter suffices */
2769 if (istate->version == 3 || istate->version == 2)
2770 istate->version = extended ? 3 : 2;
2772 hdr_version = istate->version;
2774 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
2775 hdr.hdr_version = htonl(hdr_version);
2776 hdr.hdr_entries = htonl(entries - removed);
2778 the_hash_algo->init_fn(&c);
2779 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
2780 return -1;
2782 if (!HAVE_THREADS || git_config_get_index_threads(&nr_threads))
2783 nr_threads = 1;
2785 if (nr_threads != 1 && record_ieot()) {
2786 int ieot_blocks, cpus;
2789 * ensure default number of ieot blocks maps evenly to the
2790 * default number of threads that will process them leaving
2791 * room for the thread to load the index extensions.
2793 if (!nr_threads) {
2794 ieot_blocks = istate->cache_nr / THREAD_COST;
2795 cpus = online_cpus();
2796 if (ieot_blocks > cpus - 1)
2797 ieot_blocks = cpus - 1;
2798 } else {
2799 ieot_blocks = nr_threads;
2800 if (ieot_blocks > istate->cache_nr)
2801 ieot_blocks = istate->cache_nr;
2805 * no reason to write out the IEOT extension if we don't
2806 * have enough blocks to utilize multi-threading
2808 if (ieot_blocks > 1) {
2809 ieot = xcalloc(1, sizeof(struct index_entry_offset_table)
2810 + (ieot_blocks * sizeof(struct index_entry_offset)));
2811 ieot_entries = DIV_ROUND_UP(entries, ieot_blocks);
2815 offset = lseek(newfd, 0, SEEK_CUR);
2816 if (offset < 0) {
2817 free(ieot);
2818 return -1;
2820 offset += write_buffer_len;
2821 nr = 0;
2822 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2824 for (i = 0; i < entries; i++) {
2825 struct cache_entry *ce = cache[i];
2826 if (ce->ce_flags & CE_REMOVE)
2827 continue;
2828 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2829 ce_smudge_racily_clean_entry(istate, ce);
2830 if (is_null_oid(&ce->oid)) {
2831 static const char msg[] = "cache entry has null sha1: %s";
2832 static int allow = -1;
2834 if (allow < 0)
2835 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2836 if (allow)
2837 warning(msg, ce->name);
2838 else
2839 err = error(msg, ce->name);
2841 drop_cache_tree = 1;
2843 if (ieot && i && (i % ieot_entries == 0)) {
2844 ieot->entries[ieot->nr].nr = nr;
2845 ieot->entries[ieot->nr].offset = offset;
2846 ieot->nr++;
2848 * If we have a V4 index, set the first byte to an invalid
2849 * character to ensure there is nothing common with the previous
2850 * entry
2852 if (previous_name)
2853 previous_name->buf[0] = 0;
2854 nr = 0;
2855 offset = lseek(newfd, 0, SEEK_CUR);
2856 if (offset < 0) {
2857 free(ieot);
2858 return -1;
2860 offset += write_buffer_len;
2862 if (ce_write_entry(&c, newfd, ce, previous_name, (struct ondisk_cache_entry *)&ondisk) < 0)
2863 err = -1;
2865 if (err)
2866 break;
2867 nr++;
2869 if (ieot && nr) {
2870 ieot->entries[ieot->nr].nr = nr;
2871 ieot->entries[ieot->nr].offset = offset;
2872 ieot->nr++;
2874 strbuf_release(&previous_name_buf);
2876 if (err) {
2877 free(ieot);
2878 return err;
2881 /* Write extension data here */
2882 offset = lseek(newfd, 0, SEEK_CUR);
2883 if (offset < 0) {
2884 free(ieot);
2885 return -1;
2887 offset += write_buffer_len;
2888 the_hash_algo->init_fn(&eoie_c);
2891 * Lets write out CACHE_EXT_INDEXENTRYOFFSETTABLE first so that we
2892 * can minimize the number of extensions we have to scan through to
2893 * find it during load. Write it out regardless of the
2894 * strip_extensions parameter as we need it when loading the shared
2895 * index.
2897 if (ieot) {
2898 struct strbuf sb = STRBUF_INIT;
2900 write_ieot_extension(&sb, ieot);
2901 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_INDEXENTRYOFFSETTABLE, sb.len) < 0
2902 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2903 strbuf_release(&sb);
2904 free(ieot);
2905 if (err)
2906 return -1;
2909 if (!strip_extensions && istate->split_index) {
2910 struct strbuf sb = STRBUF_INIT;
2912 err = write_link_extension(&sb, istate) < 0 ||
2913 write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_LINK,
2914 sb.len) < 0 ||
2915 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2916 strbuf_release(&sb);
2917 if (err)
2918 return -1;
2920 if (!strip_extensions && !drop_cache_tree && istate->cache_tree) {
2921 struct strbuf sb = STRBUF_INIT;
2923 cache_tree_write(&sb, istate->cache_tree);
2924 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_TREE, sb.len) < 0
2925 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2926 strbuf_release(&sb);
2927 if (err)
2928 return -1;
2930 if (!strip_extensions && istate->resolve_undo) {
2931 struct strbuf sb = STRBUF_INIT;
2933 resolve_undo_write(&sb, istate->resolve_undo);
2934 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_RESOLVE_UNDO,
2935 sb.len) < 0
2936 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2937 strbuf_release(&sb);
2938 if (err)
2939 return -1;
2941 if (!strip_extensions && istate->untracked) {
2942 struct strbuf sb = STRBUF_INIT;
2944 write_untracked_extension(&sb, istate->untracked);
2945 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_UNTRACKED,
2946 sb.len) < 0 ||
2947 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2948 strbuf_release(&sb);
2949 if (err)
2950 return -1;
2952 if (!strip_extensions && istate->fsmonitor_last_update) {
2953 struct strbuf sb = STRBUF_INIT;
2955 write_fsmonitor_extension(&sb, istate);
2956 err = write_index_ext_header(&c, &eoie_c, newfd, CACHE_EXT_FSMONITOR, sb.len) < 0
2957 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2958 strbuf_release(&sb);
2959 if (err)
2960 return -1;
2964 * CACHE_EXT_ENDOFINDEXENTRIES must be written as the last entry before the SHA1
2965 * so that it can be found and processed before all the index entries are
2966 * read. Write it out regardless of the strip_extensions parameter as we need it
2967 * when loading the shared index.
2969 if (offset && record_eoie()) {
2970 struct strbuf sb = STRBUF_INIT;
2972 write_eoie_extension(&sb, &eoie_c, offset);
2973 err = write_index_ext_header(&c, NULL, newfd, CACHE_EXT_ENDOFINDEXENTRIES, sb.len) < 0
2974 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2975 strbuf_release(&sb);
2976 if (err)
2977 return -1;
2980 if (ce_flush(&c, newfd, istate->oid.hash))
2981 return -1;
2982 if (close_tempfile_gently(tempfile)) {
2983 error(_("could not close '%s'"), tempfile->filename.buf);
2984 return -1;
2986 if (stat(tempfile->filename.buf, &st))
2987 return -1;
2988 istate->timestamp.sec = (unsigned int)st.st_mtime;
2989 istate->timestamp.nsec = ST_MTIME_NSEC(st);
2990 trace_performance_since(start, "write index, changed mask = %x", istate->cache_changed);
2991 return 0;
2994 void set_alternate_index_output(const char *name)
2996 alternate_index_output = name;
2999 static int commit_locked_index(struct lock_file *lk)
3001 if (alternate_index_output)
3002 return commit_lock_file_to(lk, alternate_index_output);
3003 else
3004 return commit_lock_file(lk);
3007 static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
3008 unsigned flags)
3010 int ret = do_write_index(istate, lock->tempfile, 0);
3011 if (ret)
3012 return ret;
3013 if (flags & COMMIT_LOCK)
3014 return commit_locked_index(lock);
3015 return close_lock_file_gently(lock);
3018 static int write_split_index(struct index_state *istate,
3019 struct lock_file *lock,
3020 unsigned flags)
3022 int ret;
3023 prepare_to_write_split_index(istate);
3024 ret = do_write_locked_index(istate, lock, flags);
3025 finish_writing_split_index(istate);
3026 return ret;
3029 static const char *shared_index_expire = "2.weeks.ago";
3031 static unsigned long get_shared_index_expire_date(void)
3033 static unsigned long shared_index_expire_date;
3034 static int shared_index_expire_date_prepared;
3036 if (!shared_index_expire_date_prepared) {
3037 git_config_get_expiry("splitindex.sharedindexexpire",
3038 &shared_index_expire);
3039 shared_index_expire_date = approxidate(shared_index_expire);
3040 shared_index_expire_date_prepared = 1;
3043 return shared_index_expire_date;
3046 static int should_delete_shared_index(const char *shared_index_path)
3048 struct stat st;
3049 unsigned long expiration;
3051 /* Check timestamp */
3052 expiration = get_shared_index_expire_date();
3053 if (!expiration)
3054 return 0;
3055 if (stat(shared_index_path, &st))
3056 return error_errno(_("could not stat '%s'"), shared_index_path);
3057 if (st.st_mtime > expiration)
3058 return 0;
3060 return 1;
3063 static int clean_shared_index_files(const char *current_hex)
3065 struct dirent *de;
3066 DIR *dir = opendir(get_git_dir());
3068 if (!dir)
3069 return error_errno(_("unable to open git dir: %s"), get_git_dir());
3071 while ((de = readdir(dir)) != NULL) {
3072 const char *sha1_hex;
3073 const char *shared_index_path;
3074 if (!skip_prefix(de->d_name, "sharedindex.", &sha1_hex))
3075 continue;
3076 if (!strcmp(sha1_hex, current_hex))
3077 continue;
3078 shared_index_path = git_path("%s", de->d_name);
3079 if (should_delete_shared_index(shared_index_path) > 0 &&
3080 unlink(shared_index_path))
3081 warning_errno(_("unable to unlink: %s"), shared_index_path);
3083 closedir(dir);
3085 return 0;
3088 static int write_shared_index(struct index_state *istate,
3089 struct tempfile **temp)
3091 struct split_index *si = istate->split_index;
3092 int ret;
3094 move_cache_to_base_index(istate);
3095 ret = do_write_index(si->base, *temp, 1);
3096 if (ret)
3097 return ret;
3098 ret = adjust_shared_perm(get_tempfile_path(*temp));
3099 if (ret) {
3100 error(_("cannot fix permission bits on '%s'"), get_tempfile_path(*temp));
3101 return ret;
3103 ret = rename_tempfile(temp,
3104 git_path("sharedindex.%s", oid_to_hex(&si->base->oid)));
3105 if (!ret) {
3106 oidcpy(&si->base_oid, &si->base->oid);
3107 clean_shared_index_files(oid_to_hex(&si->base->oid));
3110 return ret;
3113 static const int default_max_percent_split_change = 20;
3115 static int too_many_not_shared_entries(struct index_state *istate)
3117 int i, not_shared = 0;
3118 int max_split = git_config_get_max_percent_split_change();
3120 switch (max_split) {
3121 case -1:
3122 /* not or badly configured: use the default value */
3123 max_split = default_max_percent_split_change;
3124 break;
3125 case 0:
3126 return 1; /* 0% means always write a new shared index */
3127 case 100:
3128 return 0; /* 100% means never write a new shared index */
3129 default:
3130 break; /* just use the configured value */
3133 /* Count not shared entries */
3134 for (i = 0; i < istate->cache_nr; i++) {
3135 struct cache_entry *ce = istate->cache[i];
3136 if (!ce->index)
3137 not_shared++;
3140 return (int64_t)istate->cache_nr * max_split < (int64_t)not_shared * 100;
3143 int write_locked_index(struct index_state *istate, struct lock_file *lock,
3144 unsigned flags)
3146 int new_shared_index, ret;
3147 struct split_index *si = istate->split_index;
3149 if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0))
3150 cache_tree_verify(the_repository, istate);
3152 if ((flags & SKIP_IF_UNCHANGED) && !istate->cache_changed) {
3153 if (flags & COMMIT_LOCK)
3154 rollback_lock_file(lock);
3155 return 0;
3158 if (istate->fsmonitor_last_update)
3159 fill_fsmonitor_bitmap(istate);
3161 if (!si || alternate_index_output ||
3162 (istate->cache_changed & ~EXTMASK)) {
3163 if (si)
3164 oidclr(&si->base_oid);
3165 ret = do_write_locked_index(istate, lock, flags);
3166 goto out;
3169 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0)) {
3170 int v = si->base_oid.hash[0];
3171 if ((v & 15) < 6)
3172 istate->cache_changed |= SPLIT_INDEX_ORDERED;
3174 if (too_many_not_shared_entries(istate))
3175 istate->cache_changed |= SPLIT_INDEX_ORDERED;
3177 new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED;
3179 if (new_shared_index) {
3180 struct tempfile *temp;
3181 int saved_errno;
3183 /* Same initial permissions as the main .git/index file */
3184 temp = mks_tempfile_sm(git_path("sharedindex_XXXXXX"), 0, 0666);
3185 if (!temp) {
3186 oidclr(&si->base_oid);
3187 ret = do_write_locked_index(istate, lock, flags);
3188 goto out;
3190 ret = write_shared_index(istate, &temp);
3192 saved_errno = errno;
3193 if (is_tempfile_active(temp))
3194 delete_tempfile(&temp);
3195 errno = saved_errno;
3197 if (ret)
3198 goto out;
3201 ret = write_split_index(istate, lock, flags);
3203 /* Freshen the shared index only if the split-index was written */
3204 if (!ret && !new_shared_index) {
3205 const char *shared_index = git_path("sharedindex.%s",
3206 oid_to_hex(&si->base_oid));
3207 freshen_shared_index(shared_index, 1);
3210 out:
3211 if (flags & COMMIT_LOCK)
3212 rollback_lock_file(lock);
3213 return ret;
3217 * Read the index file that is potentially unmerged into given
3218 * index_state, dropping any unmerged entries to stage #0 (potentially
3219 * resulting in a path appearing as both a file and a directory in the
3220 * index; the caller is responsible to clear out the extra entries
3221 * before writing the index to a tree). Returns true if the index is
3222 * unmerged. Callers who want to refuse to work from an unmerged
3223 * state can call this and check its return value, instead of calling
3224 * read_cache().
3226 int read_index_unmerged(struct index_state *istate)
3228 int i;
3229 int unmerged = 0;
3231 read_index(istate);
3232 for (i = 0; i < istate->cache_nr; i++) {
3233 struct cache_entry *ce = istate->cache[i];
3234 struct cache_entry *new_ce;
3235 int len;
3237 if (!ce_stage(ce))
3238 continue;
3239 unmerged = 1;
3240 len = ce_namelen(ce);
3241 new_ce = make_empty_cache_entry(istate, len);
3242 memcpy(new_ce->name, ce->name, len);
3243 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
3244 new_ce->ce_namelen = len;
3245 new_ce->ce_mode = ce->ce_mode;
3246 if (add_index_entry(istate, new_ce, ADD_CACHE_SKIP_DFCHECK))
3247 return error(_("%s: cannot drop to stage #0"),
3248 new_ce->name);
3250 return unmerged;
3254 * Returns 1 if the path is an "other" path with respect to
3255 * the index; that is, the path is not mentioned in the index at all,
3256 * either as a file, a directory with some files in the index,
3257 * or as an unmerged entry.
3259 * We helpfully remove a trailing "/" from directories so that
3260 * the output of read_directory can be used as-is.
3262 int index_name_is_other(const struct index_state *istate, const char *name,
3263 int namelen)
3265 int pos;
3266 if (namelen && name[namelen - 1] == '/')
3267 namelen--;
3268 pos = index_name_pos(istate, name, namelen);
3269 if (0 <= pos)
3270 return 0; /* exact match */
3271 pos = -pos - 1;
3272 if (pos < istate->cache_nr) {
3273 struct cache_entry *ce = istate->cache[pos];
3274 if (ce_namelen(ce) == namelen &&
3275 !memcmp(ce->name, name, namelen))
3276 return 0; /* Yup, this one exists unmerged */
3278 return 1;
3281 void *read_blob_data_from_index(const struct index_state *istate,
3282 const char *path, unsigned long *size)
3284 int pos, len;
3285 unsigned long sz;
3286 enum object_type type;
3287 void *data;
3289 len = strlen(path);
3290 pos = index_name_pos(istate, path, len);
3291 if (pos < 0) {
3293 * We might be in the middle of a merge, in which
3294 * case we would read stage #2 (ours).
3296 int i;
3297 for (i = -pos - 1;
3298 (pos < 0 && i < istate->cache_nr &&
3299 !strcmp(istate->cache[i]->name, path));
3300 i++)
3301 if (ce_stage(istate->cache[i]) == 2)
3302 pos = i;
3304 if (pos < 0)
3305 return NULL;
3306 data = read_object_file(&istate->cache[pos]->oid, &type, &sz);
3307 if (!data || type != OBJ_BLOB) {
3308 free(data);
3309 return NULL;
3311 if (size)
3312 *size = sz;
3313 return data;
3316 void stat_validity_clear(struct stat_validity *sv)
3318 FREE_AND_NULL(sv->sd);
3321 int stat_validity_check(struct stat_validity *sv, const char *path)
3323 struct stat st;
3325 if (stat(path, &st) < 0)
3326 return sv->sd == NULL;
3327 if (!sv->sd)
3328 return 0;
3329 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
3332 void stat_validity_update(struct stat_validity *sv, int fd)
3334 struct stat st;
3336 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
3337 stat_validity_clear(sv);
3338 else {
3339 if (!sv->sd)
3340 sv->sd = xcalloc(1, sizeof(struct stat_data));
3341 fill_stat_data(sv->sd, &st);
3345 void move_index_extensions(struct index_state *dst, struct index_state *src)
3347 dst->untracked = src->untracked;
3348 src->untracked = NULL;
3349 dst->cache_tree = src->cache_tree;
3350 src->cache_tree = NULL;
3353 struct cache_entry *dup_cache_entry(const struct cache_entry *ce,
3354 struct index_state *istate)
3356 unsigned int size = ce_size(ce);
3357 int mem_pool_allocated;
3358 struct cache_entry *new_entry = make_empty_cache_entry(istate, ce_namelen(ce));
3359 mem_pool_allocated = new_entry->mem_pool_allocated;
3361 memcpy(new_entry, ce, size);
3362 new_entry->mem_pool_allocated = mem_pool_allocated;
3363 return new_entry;
3366 void discard_cache_entry(struct cache_entry *ce)
3368 if (ce && should_validate_cache_entries())
3369 memset(ce, 0xCD, cache_entry_size(ce->ce_namelen));
3371 if (ce && ce->mem_pool_allocated)
3372 return;
3374 free(ce);
3377 int should_validate_cache_entries(void)
3379 static int validate_index_cache_entries = -1;
3381 if (validate_index_cache_entries < 0) {
3382 if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))
3383 validate_index_cache_entries = 1;
3384 else
3385 validate_index_cache_entries = 0;
3388 return validate_index_cache_entries;
3391 #define EOIE_SIZE (4 + GIT_SHA1_RAWSZ) /* <4-byte offset> + <20-byte hash> */
3392 #define EOIE_SIZE_WITH_HEADER (4 + 4 + EOIE_SIZE) /* <4-byte signature> + <4-byte length> + EOIE_SIZE */
3394 static size_t read_eoie_extension(const char *mmap, size_t mmap_size)
3397 * The end of index entries (EOIE) extension is guaranteed to be last
3398 * so that it can be found by scanning backwards from the EOF.
3400 * "EOIE"
3401 * <4-byte length>
3402 * <4-byte offset>
3403 * <20-byte hash>
3405 const char *index, *eoie;
3406 uint32_t extsize;
3407 size_t offset, src_offset;
3408 unsigned char hash[GIT_MAX_RAWSZ];
3409 git_hash_ctx c;
3411 /* ensure we have an index big enough to contain an EOIE extension */
3412 if (mmap_size < sizeof(struct cache_header) + EOIE_SIZE_WITH_HEADER + the_hash_algo->rawsz)
3413 return 0;
3415 /* validate the extension signature */
3416 index = eoie = mmap + mmap_size - EOIE_SIZE_WITH_HEADER - the_hash_algo->rawsz;
3417 if (CACHE_EXT(index) != CACHE_EXT_ENDOFINDEXENTRIES)
3418 return 0;
3419 index += sizeof(uint32_t);
3421 /* validate the extension size */
3422 extsize = get_be32(index);
3423 if (extsize != EOIE_SIZE)
3424 return 0;
3425 index += sizeof(uint32_t);
3428 * Validate the offset we're going to look for the first extension
3429 * signature is after the index header and before the eoie extension.
3431 offset = get_be32(index);
3432 if (mmap + offset < mmap + sizeof(struct cache_header))
3433 return 0;
3434 if (mmap + offset >= eoie)
3435 return 0;
3436 index += sizeof(uint32_t);
3439 * The hash is computed over extension types and their sizes (but not
3440 * their contents). E.g. if we have "TREE" extension that is N-bytes
3441 * long, "REUC" extension that is M-bytes long, followed by "EOIE",
3442 * then the hash would be:
3444 * SHA-1("TREE" + <binary representation of N> +
3445 * "REUC" + <binary representation of M>)
3447 src_offset = offset;
3448 the_hash_algo->init_fn(&c);
3449 while (src_offset < mmap_size - the_hash_algo->rawsz - EOIE_SIZE_WITH_HEADER) {
3450 /* After an array of active_nr index entries,
3451 * there can be arbitrary number of extended
3452 * sections, each of which is prefixed with
3453 * extension name (4-byte) and section length
3454 * in 4-byte network byte order.
3456 uint32_t extsize;
3457 memcpy(&extsize, mmap + src_offset + 4, 4);
3458 extsize = ntohl(extsize);
3460 /* verify the extension size isn't so large it will wrap around */
3461 if (src_offset + 8 + extsize < src_offset)
3462 return 0;
3464 the_hash_algo->update_fn(&c, mmap + src_offset, 8);
3466 src_offset += 8;
3467 src_offset += extsize;
3469 the_hash_algo->final_fn(hash, &c);
3470 if (!hasheq(hash, (const unsigned char *)index))
3471 return 0;
3473 /* Validate that the extension offsets returned us back to the eoie extension. */
3474 if (src_offset != mmap_size - the_hash_algo->rawsz - EOIE_SIZE_WITH_HEADER)
3475 return 0;
3477 return offset;
3480 static void write_eoie_extension(struct strbuf *sb, git_hash_ctx *eoie_context, size_t offset)
3482 uint32_t buffer;
3483 unsigned char hash[GIT_MAX_RAWSZ];
3485 /* offset */
3486 put_be32(&buffer, offset);
3487 strbuf_add(sb, &buffer, sizeof(uint32_t));
3489 /* hash */
3490 the_hash_algo->final_fn(hash, eoie_context);
3491 strbuf_add(sb, hash, the_hash_algo->rawsz);
3494 #define IEOT_VERSION (1)
3496 static struct index_entry_offset_table *read_ieot_extension(const char *mmap, size_t mmap_size, size_t offset)
3498 const char *index = NULL;
3499 uint32_t extsize, ext_version;
3500 struct index_entry_offset_table *ieot;
3501 int i, nr;
3503 /* find the IEOT extension */
3504 if (!offset)
3505 return NULL;
3506 while (offset <= mmap_size - the_hash_algo->rawsz - 8) {
3507 extsize = get_be32(mmap + offset + 4);
3508 if (CACHE_EXT((mmap + offset)) == CACHE_EXT_INDEXENTRYOFFSETTABLE) {
3509 index = mmap + offset + 4 + 4;
3510 break;
3512 offset += 8;
3513 offset += extsize;
3515 if (!index)
3516 return NULL;
3518 /* validate the version is IEOT_VERSION */
3519 ext_version = get_be32(index);
3520 if (ext_version != IEOT_VERSION) {
3521 error("invalid IEOT version %d", ext_version);
3522 return NULL;
3524 index += sizeof(uint32_t);
3526 /* extension size - version bytes / bytes per entry */
3527 nr = (extsize - sizeof(uint32_t)) / (sizeof(uint32_t) + sizeof(uint32_t));
3528 if (!nr) {
3529 error("invalid number of IEOT entries %d", nr);
3530 return NULL;
3532 ieot = xmalloc(sizeof(struct index_entry_offset_table)
3533 + (nr * sizeof(struct index_entry_offset)));
3534 ieot->nr = nr;
3535 for (i = 0; i < nr; i++) {
3536 ieot->entries[i].offset = get_be32(index);
3537 index += sizeof(uint32_t);
3538 ieot->entries[i].nr = get_be32(index);
3539 index += sizeof(uint32_t);
3542 return ieot;
3545 static void write_ieot_extension(struct strbuf *sb, struct index_entry_offset_table *ieot)
3547 uint32_t buffer;
3548 int i;
3550 /* version */
3551 put_be32(&buffer, IEOT_VERSION);
3552 strbuf_add(sb, &buffer, sizeof(uint32_t));
3554 /* ieot */
3555 for (i = 0; i < ieot->nr; i++) {
3557 /* offset */
3558 put_be32(&buffer, ieot->entries[i].offset);
3559 strbuf_add(sb, &buffer, sizeof(uint32_t));
3561 /* count */
3562 put_be32(&buffer, ieot->entries[i].nr);
3563 strbuf_add(sb, &buffer, sizeof(uint32_t));