2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 #define NO_THE_INDEX_COMPATIBILITY_MACROS
11 #include "cache-tree.h"
17 #include "resolve-undo.h"
20 #include "split-index.h"
23 /* Mask for the name length in ce_flags in the on-disk index */
25 #define CE_NAMEMASK (0x0fff)
29 * The first letter should be 'A'..'Z' for extensions that are not
30 * necessary for a correct operation (i.e. optimization data).
31 * When new extensions are added that _needs_ to be understood in
32 * order to correctly interpret the index file, pick character that
33 * is outside the range, to cause the reader to abort.
36 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
37 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
38 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
39 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
40 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
42 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
43 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
44 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
45 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED)
47 struct index_state the_index
;
48 static const char *alternate_index_output
;
50 static void set_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
52 istate
->cache
[nr
] = ce
;
53 add_name_hash(istate
, ce
);
56 static void replace_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
58 struct cache_entry
*old
= istate
->cache
[nr
];
60 replace_index_entry_in_base(istate
, old
, ce
);
61 remove_name_hash(istate
, old
);
63 set_index_entry(istate
, nr
, ce
);
64 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
65 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
68 void rename_index_entry_at(struct index_state
*istate
, int nr
, const char *new_name
)
70 struct cache_entry
*old
= istate
->cache
[nr
], *new;
71 int namelen
= strlen(new_name
);
73 new = xmalloc(cache_entry_size(namelen
));
74 copy_cache_entry(new, old
);
75 new->ce_flags
&= ~CE_HASHED
;
76 new->ce_namelen
= namelen
;
78 memcpy(new->name
, new_name
, namelen
+ 1);
80 cache_tree_invalidate_path(istate
, old
->name
);
81 untracked_cache_remove_from_index(istate
, old
->name
);
82 remove_index_entry_at(istate
, nr
);
83 add_index_entry(istate
, new, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
86 void fill_stat_data(struct stat_data
*sd
, struct stat
*st
)
88 sd
->sd_ctime
.sec
= (unsigned int)st
->st_ctime
;
89 sd
->sd_mtime
.sec
= (unsigned int)st
->st_mtime
;
90 sd
->sd_ctime
.nsec
= ST_CTIME_NSEC(*st
);
91 sd
->sd_mtime
.nsec
= ST_MTIME_NSEC(*st
);
92 sd
->sd_dev
= st
->st_dev
;
93 sd
->sd_ino
= st
->st_ino
;
94 sd
->sd_uid
= st
->st_uid
;
95 sd
->sd_gid
= st
->st_gid
;
96 sd
->sd_size
= st
->st_size
;
99 int match_stat_data(const struct stat_data
*sd
, struct stat
*st
)
103 if (sd
->sd_mtime
.sec
!= (unsigned int)st
->st_mtime
)
104 changed
|= MTIME_CHANGED
;
105 if (trust_ctime
&& check_stat
&&
106 sd
->sd_ctime
.sec
!= (unsigned int)st
->st_ctime
)
107 changed
|= CTIME_CHANGED
;
110 if (check_stat
&& sd
->sd_mtime
.nsec
!= ST_MTIME_NSEC(*st
))
111 changed
|= MTIME_CHANGED
;
112 if (trust_ctime
&& check_stat
&&
113 sd
->sd_ctime
.nsec
!= ST_CTIME_NSEC(*st
))
114 changed
|= CTIME_CHANGED
;
118 if (sd
->sd_uid
!= (unsigned int) st
->st_uid
||
119 sd
->sd_gid
!= (unsigned int) st
->st_gid
)
120 changed
|= OWNER_CHANGED
;
121 if (sd
->sd_ino
!= (unsigned int) st
->st_ino
)
122 changed
|= INODE_CHANGED
;
127 * st_dev breaks on network filesystems where different
128 * clients will have different views of what "device"
129 * the filesystem is on
131 if (check_stat
&& sd
->sd_dev
!= (unsigned int) st
->st_dev
)
132 changed
|= INODE_CHANGED
;
135 if (sd
->sd_size
!= (unsigned int) st
->st_size
)
136 changed
|= DATA_CHANGED
;
142 * This only updates the "non-critical" parts of the directory
143 * cache, ie the parts that aren't tracked by GIT, and only used
144 * to validate the cache.
146 void fill_stat_cache_info(struct cache_entry
*ce
, struct stat
*st
)
148 fill_stat_data(&ce
->ce_stat_data
, st
);
150 if (assume_unchanged
)
151 ce
->ce_flags
|= CE_VALID
;
153 if (S_ISREG(st
->st_mode
))
154 ce_mark_uptodate(ce
);
157 static int ce_compare_data(const struct cache_entry
*ce
, struct stat
*st
)
160 int fd
= git_open_cloexec(ce
->name
, O_RDONLY
);
163 struct object_id oid
;
164 if (!index_fd(&oid
, fd
, st
, OBJ_BLOB
, ce
->name
, 0))
165 match
= oidcmp(&oid
, &ce
->oid
);
166 /* index_fd() closed the file descriptor already */
171 static int ce_compare_link(const struct cache_entry
*ce
, size_t expected_size
)
176 enum object_type type
;
177 struct strbuf sb
= STRBUF_INIT
;
179 if (strbuf_readlink(&sb
, ce
->name
, expected_size
))
182 buffer
= read_sha1_file(ce
->oid
.hash
, &type
, &size
);
185 match
= memcmp(buffer
, sb
.buf
, size
);
192 static int ce_compare_gitlink(const struct cache_entry
*ce
)
194 unsigned char sha1
[20];
197 * We don't actually require that the .git directory
198 * under GITLINK directory be a valid git directory. It
199 * might even be missing (in case nobody populated that
202 * If so, we consider it always to match.
204 if (resolve_gitlink_ref(ce
->name
, "HEAD", sha1
) < 0)
206 return hashcmp(sha1
, ce
->oid
.hash
);
209 static int ce_modified_check_fs(const struct cache_entry
*ce
, struct stat
*st
)
211 switch (st
->st_mode
& S_IFMT
) {
213 if (ce_compare_data(ce
, st
))
217 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
221 if (S_ISGITLINK(ce
->ce_mode
))
222 return ce_compare_gitlink(ce
) ? DATA_CHANGED
: 0;
223 /* else fallthrough */
230 static int ce_match_stat_basic(const struct cache_entry
*ce
, struct stat
*st
)
232 unsigned int changed
= 0;
234 if (ce
->ce_flags
& CE_REMOVE
)
235 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
237 switch (ce
->ce_mode
& S_IFMT
) {
239 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
240 /* We consider only the owner x bit to be relevant for
243 if (trust_executable_bit
&&
244 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
245 changed
|= MODE_CHANGED
;
248 if (!S_ISLNK(st
->st_mode
) &&
249 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
250 changed
|= TYPE_CHANGED
;
253 /* We ignore most of the st_xxx fields for gitlinks */
254 if (!S_ISDIR(st
->st_mode
))
255 changed
|= TYPE_CHANGED
;
256 else if (ce_compare_gitlink(ce
))
257 changed
|= DATA_CHANGED
;
260 die("internal error: ce_mode is %o", ce
->ce_mode
);
263 changed
|= match_stat_data(&ce
->ce_stat_data
, st
);
265 /* Racily smudged entry? */
266 if (!ce
->ce_stat_data
.sd_size
) {
267 if (!is_empty_blob_sha1(ce
->oid
.hash
))
268 changed
|= DATA_CHANGED
;
274 static int is_racy_stat(const struct index_state
*istate
,
275 const struct stat_data
*sd
)
277 return (istate
->timestamp
.sec
&&
279 /* nanosecond timestamped files can also be racy! */
280 (istate
->timestamp
.sec
< sd
->sd_mtime
.sec
||
281 (istate
->timestamp
.sec
== sd
->sd_mtime
.sec
&&
282 istate
->timestamp
.nsec
<= sd
->sd_mtime
.nsec
))
284 istate
->timestamp
.sec
<= sd
->sd_mtime
.sec
289 static int is_racy_timestamp(const struct index_state
*istate
,
290 const struct cache_entry
*ce
)
292 return (!S_ISGITLINK(ce
->ce_mode
) &&
293 is_racy_stat(istate
, &ce
->ce_stat_data
));
296 int match_stat_data_racy(const struct index_state
*istate
,
297 const struct stat_data
*sd
, struct stat
*st
)
299 if (is_racy_stat(istate
, sd
))
300 return MTIME_CHANGED
;
301 return match_stat_data(sd
, st
);
304 int ie_match_stat(const struct index_state
*istate
,
305 const struct cache_entry
*ce
, struct stat
*st
,
306 unsigned int options
)
308 unsigned int changed
;
309 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
310 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
311 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
314 * If it's marked as always valid in the index, it's
315 * valid whatever the checked-out copy says.
317 * skip-worktree has the same effect with higher precedence
319 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
))
321 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
325 * Intent-to-add entries have not been added, so the index entry
326 * by definition never matches what is in the work tree until it
327 * actually gets added.
329 if (ce_intent_to_add(ce
))
330 return DATA_CHANGED
| TYPE_CHANGED
| MODE_CHANGED
;
332 changed
= ce_match_stat_basic(ce
, st
);
335 * Within 1 second of this sequence:
336 * echo xyzzy >file && git-update-index --add file
337 * running this command:
339 * would give a falsely clean cache entry. The mtime and
340 * length match the cache, and other stat fields do not change.
342 * We could detect this at update-index time (the cache entry
343 * being registered/updated records the same time as "now")
344 * and delay the return from git-update-index, but that would
345 * effectively mean we can make at most one commit per second,
346 * which is not acceptable. Instead, we check cache entries
347 * whose mtime are the same as the index file timestamp more
348 * carefully than others.
350 if (!changed
&& is_racy_timestamp(istate
, ce
)) {
351 if (assume_racy_is_modified
)
352 changed
|= DATA_CHANGED
;
354 changed
|= ce_modified_check_fs(ce
, st
);
360 int ie_modified(const struct index_state
*istate
,
361 const struct cache_entry
*ce
,
362 struct stat
*st
, unsigned int options
)
364 int changed
, changed_fs
;
366 changed
= ie_match_stat(istate
, ce
, st
, options
);
370 * If the mode or type has changed, there's no point in trying
371 * to refresh the entry - it's not going to match
373 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
377 * Immediately after read-tree or update-index --cacheinfo,
378 * the length field is zero, as we have never even read the
379 * lstat(2) information once, and we cannot trust DATA_CHANGED
380 * returned by ie_match_stat() which in turn was returned by
381 * ce_match_stat_basic() to signal that the filesize of the
382 * blob changed. We have to actually go to the filesystem to
383 * see if the contents match, and if so, should answer "unchanged".
385 * The logic does not apply to gitlinks, as ce_match_stat_basic()
386 * already has checked the actual HEAD from the filesystem in the
387 * subproject. If ie_match_stat() already said it is different,
388 * then we know it is.
390 if ((changed
& DATA_CHANGED
) &&
391 (S_ISGITLINK(ce
->ce_mode
) || ce
->ce_stat_data
.sd_size
!= 0))
394 changed_fs
= ce_modified_check_fs(ce
, st
);
396 return changed
| changed_fs
;
400 int base_name_compare(const char *name1
, int len1
, int mode1
,
401 const char *name2
, int len2
, int mode2
)
403 unsigned char c1
, c2
;
404 int len
= len1
< len2
? len1
: len2
;
407 cmp
= memcmp(name1
, name2
, len
);
412 if (!c1
&& S_ISDIR(mode1
))
414 if (!c2
&& S_ISDIR(mode2
))
416 return (c1
< c2
) ? -1 : (c1
> c2
) ? 1 : 0;
420 * df_name_compare() is identical to base_name_compare(), except it
421 * compares conflicting directory/file entries as equal. Note that
422 * while a directory name compares as equal to a regular file, they
423 * then individually compare _differently_ to a filename that has
424 * a dot after the basename (because '\0' < '.' < '/').
426 * This is used by routines that want to traverse the git namespace
427 * but then handle conflicting entries together when possible.
429 int df_name_compare(const char *name1
, int len1
, int mode1
,
430 const char *name2
, int len2
, int mode2
)
432 int len
= len1
< len2
? len1
: len2
, cmp
;
433 unsigned char c1
, c2
;
435 cmp
= memcmp(name1
, name2
, len
);
438 /* Directories and files compare equal (same length, same name) */
442 if (!c1
&& S_ISDIR(mode1
))
445 if (!c2
&& S_ISDIR(mode2
))
447 if (c1
== '/' && !c2
)
449 if (c2
== '/' && !c1
)
454 int name_compare(const char *name1
, size_t len1
, const char *name2
, size_t len2
)
456 size_t min_len
= (len1
< len2
) ? len1
: len2
;
457 int cmp
= memcmp(name1
, name2
, min_len
);
467 int cache_name_stage_compare(const char *name1
, int len1
, int stage1
, const char *name2
, int len2
, int stage2
)
471 cmp
= name_compare(name1
, len1
, name2
, len2
);
482 static int index_name_stage_pos(const struct index_state
*istate
, const char *name
, int namelen
, int stage
)
487 last
= istate
->cache_nr
;
488 while (last
> first
) {
489 int next
= (last
+ first
) >> 1;
490 struct cache_entry
*ce
= istate
->cache
[next
];
491 int cmp
= cache_name_stage_compare(name
, namelen
, stage
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
503 int index_name_pos(const struct index_state
*istate
, const char *name
, int namelen
)
505 return index_name_stage_pos(istate
, name
, namelen
, 0);
508 int remove_index_entry_at(struct index_state
*istate
, int pos
)
510 struct cache_entry
*ce
= istate
->cache
[pos
];
512 record_resolve_undo(istate
, ce
);
513 remove_name_hash(istate
, ce
);
514 save_or_free_index_entry(istate
, ce
);
515 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
517 if (pos
>= istate
->cache_nr
)
519 MOVE_ARRAY(istate
->cache
+ pos
, istate
->cache
+ pos
+ 1,
520 istate
->cache_nr
- pos
);
525 * Remove all cache entries marked for removal, that is where
526 * CE_REMOVE is set in ce_flags. This is much more effective than
527 * calling remove_index_entry_at() for each entry to be removed.
529 void remove_marked_cache_entries(struct index_state
*istate
)
531 struct cache_entry
**ce_array
= istate
->cache
;
534 for (i
= j
= 0; i
< istate
->cache_nr
; i
++) {
535 if (ce_array
[i
]->ce_flags
& CE_REMOVE
) {
536 remove_name_hash(istate
, ce_array
[i
]);
537 save_or_free_index_entry(istate
, ce_array
[i
]);
540 ce_array
[j
++] = ce_array
[i
];
542 if (j
== istate
->cache_nr
)
544 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
545 istate
->cache_nr
= j
;
548 int remove_file_from_index(struct index_state
*istate
, const char *path
)
550 int pos
= index_name_pos(istate
, path
, strlen(path
));
553 cache_tree_invalidate_path(istate
, path
);
554 untracked_cache_remove_from_index(istate
, path
);
555 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
556 remove_index_entry_at(istate
, pos
);
560 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
562 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
565 static int index_name_pos_also_unmerged(struct index_state
*istate
,
566 const char *path
, int namelen
)
568 int pos
= index_name_pos(istate
, path
, namelen
);
569 struct cache_entry
*ce
;
574 /* maybe unmerged? */
576 if (pos
>= istate
->cache_nr
||
577 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
580 /* order of preference: stage 2, 1, 3 */
581 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
582 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
583 !compare_name(ce
, path
, namelen
))
588 static int different_name(struct cache_entry
*ce
, struct cache_entry
*alias
)
590 int len
= ce_namelen(ce
);
591 return ce_namelen(alias
) != len
|| memcmp(ce
->name
, alias
->name
, len
);
595 * If we add a filename that aliases in the cache, we will use the
596 * name that we already have - but we don't want to update the same
597 * alias twice, because that implies that there were actually two
598 * different files with aliasing names!
600 * So we use the CE_ADDED flag to verify that the alias was an old
601 * one before we accept it as
603 static struct cache_entry
*create_alias_ce(struct index_state
*istate
,
604 struct cache_entry
*ce
,
605 struct cache_entry
*alias
)
608 struct cache_entry
*new;
610 if (alias
->ce_flags
& CE_ADDED
)
611 die("Will not add file alias '%s' ('%s' already exists in index)", ce
->name
, alias
->name
);
613 /* Ok, create the new entry using the name of the existing alias */
614 len
= ce_namelen(alias
);
615 new = xcalloc(1, cache_entry_size(len
));
616 memcpy(new->name
, alias
->name
, len
);
617 copy_cache_entry(new, ce
);
618 save_or_free_index_entry(istate
, ce
);
622 void set_object_name_for_intent_to_add_entry(struct cache_entry
*ce
)
624 unsigned char sha1
[20];
625 if (write_sha1_file("", 0, blob_type
, sha1
))
626 die("cannot create an empty blob in the object database");
627 hashcpy(ce
->oid
.hash
, sha1
);
630 int add_to_index(struct index_state
*istate
, const char *path
, struct stat
*st
, int flags
)
632 int size
, namelen
, was_same
;
633 mode_t st_mode
= st
->st_mode
;
634 struct cache_entry
*ce
, *alias
;
635 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_IGNORE_SKIP_WORKTREE
|CE_MATCH_RACY_IS_DIRTY
;
636 int verbose
= flags
& (ADD_CACHE_VERBOSE
| ADD_CACHE_PRETEND
);
637 int pretend
= flags
& ADD_CACHE_PRETEND
;
638 int intent_only
= flags
& ADD_CACHE_INTENT
;
639 int add_option
= (ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
|
640 (intent_only
? ADD_CACHE_NEW_ONLY
: 0));
642 if (!S_ISREG(st_mode
) && !S_ISLNK(st_mode
) && !S_ISDIR(st_mode
))
643 return error("%s: can only add regular files, symbolic links or git-directories", path
);
645 namelen
= strlen(path
);
646 if (S_ISDIR(st_mode
)) {
647 while (namelen
&& path
[namelen
-1] == '/')
650 size
= cache_entry_size(namelen
);
651 ce
= xcalloc(1, size
);
652 memcpy(ce
->name
, path
, namelen
);
653 ce
->ce_namelen
= namelen
;
655 fill_stat_cache_info(ce
, st
);
657 ce
->ce_flags
|= CE_INTENT_TO_ADD
;
660 if (trust_executable_bit
&& has_symlinks
) {
661 ce
->ce_mode
= create_ce_mode(st_mode
);
663 /* If there is an existing entry, pick the mode bits and type
664 * from it, otherwise assume unexecutable regular file.
666 struct cache_entry
*ent
;
667 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
669 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
670 ce
->ce_mode
= ce_mode_from_stat(ent
, st_mode
);
673 /* When core.ignorecase=true, determine if a directory of the same name but differing
674 * case already exists within the Git repository. If it does, ensure the directory
675 * case of the file being added to the repository matches (is folded into) the existing
676 * entry's directory case.
679 adjust_dirname_case(istate
, ce
->name
);
682 alias
= index_file_exists(istate
, ce
->name
, ce_namelen(ce
), ignore_case
);
683 if (alias
&& !ce_stage(alias
) && !ie_match_stat(istate
, alias
, st
, ce_option
)) {
684 /* Nothing changed, really */
685 if (!S_ISGITLINK(alias
->ce_mode
))
686 ce_mark_uptodate(alias
);
687 alias
->ce_flags
|= CE_ADDED
;
693 if (index_path(&ce
->oid
, path
, st
, HASH_WRITE_OBJECT
)) {
695 return error("unable to index file %s", path
);
698 set_object_name_for_intent_to_add_entry(ce
);
700 if (ignore_case
&& alias
&& different_name(ce
, alias
))
701 ce
= create_alias_ce(istate
, ce
, alias
);
702 ce
->ce_flags
|= CE_ADDED
;
704 /* It was suspected to be racily clean, but it turns out to be Ok */
707 !oidcmp(&alias
->oid
, &ce
->oid
) &&
708 ce
->ce_mode
== alias
->ce_mode
);
712 else if (add_index_entry(istate
, ce
, add_option
)) {
714 return error("unable to add %s to index", path
);
716 if (verbose
&& !was_same
)
717 printf("add '%s'\n", path
);
721 int add_file_to_index(struct index_state
*istate
, const char *path
, int flags
)
724 if (lstat(path
, &st
))
725 die_errno("unable to stat '%s'", path
);
726 return add_to_index(istate
, path
, &st
, flags
);
729 struct cache_entry
*make_cache_entry(unsigned int mode
,
730 const unsigned char *sha1
, const char *path
, int stage
,
731 unsigned int refresh_options
)
734 struct cache_entry
*ce
, *ret
;
736 if (!verify_path(path
, mode
)) {
737 error("Invalid path '%s'", path
);
742 size
= cache_entry_size(len
);
743 ce
= xcalloc(1, size
);
745 hashcpy(ce
->oid
.hash
, sha1
);
746 memcpy(ce
->name
, path
, len
);
747 ce
->ce_flags
= create_ce_flags(stage
);
748 ce
->ce_namelen
= len
;
749 ce
->ce_mode
= create_ce_mode(mode
);
751 ret
= refresh_cache_entry(ce
, refresh_options
);
758 * Chmod an index entry with either +x or -x.
760 * Returns -1 if the chmod for the particular cache entry failed (if it's
761 * not a regular file), -2 if an invalid flip argument is passed in, 0
764 int chmod_index_entry(struct index_state
*istate
, struct cache_entry
*ce
,
767 if (!S_ISREG(ce
->ce_mode
))
774 ce
->ce_mode
&= ~0111;
779 cache_tree_invalidate_path(istate
, ce
->name
);
780 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
781 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
786 int ce_same_name(const struct cache_entry
*a
, const struct cache_entry
*b
)
788 int len
= ce_namelen(a
);
789 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
793 * We fundamentally don't like some paths: we don't want
794 * dot or dot-dot anywhere, and for obvious reasons don't
795 * want to recurse into ".git" either.
797 * Also, we don't want double slashes or slashes at the
798 * end that can make pathnames ambiguous.
800 static int verify_dotfile(const char *rest
, unsigned mode
)
803 * The first character was '.', but that
804 * has already been discarded, we now test
808 /* "." is not allowed */
809 if (*rest
== '\0' || is_dir_sep(*rest
))
814 * ".git" followed by NUL or slash is bad. Note that we match
815 * case-insensitively here, even if ignore_case is not set.
816 * This outlaws ".GIT" everywhere out of an abundance of caution,
817 * since there's really no good reason to allow it.
819 * Once we've seen ".git", we can also find ".gitmodules", etc (also
820 * case-insensitively).
824 if (rest
[1] != 'i' && rest
[1] != 'I')
826 if (rest
[2] != 't' && rest
[2] != 'T')
828 if (rest
[3] == '\0' || is_dir_sep(rest
[3]))
832 if (skip_iprefix(rest
, "modules", &rest
) &&
833 (*rest
== '\0' || is_dir_sep(*rest
)))
838 if (rest
[1] == '\0' || is_dir_sep(rest
[1]))
844 int verify_path(const char *path
, unsigned mode
)
848 if (has_dos_drive_prefix(path
))
851 if (!is_valid_path(path
))
861 if (is_hfs_dotgit(path
))
864 if (is_hfs_dotgitmodules(path
))
869 if (is_ntfs_dotgit(path
))
872 if (is_ntfs_dotgitmodules(path
))
878 if ((c
== '.' && !verify_dotfile(path
, mode
)) ||
879 is_dir_sep(c
) || c
== '\0')
881 } else if (c
== '\\' && protect_ntfs
) {
882 if (is_ntfs_dotgit(path
))
885 if (is_ntfs_dotgitmodules(path
))
895 * Do we have another file that has the beginning components being a
896 * proper superset of the name we're trying to add?
898 static int has_file_name(struct index_state
*istate
,
899 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
902 int len
= ce_namelen(ce
);
903 int stage
= ce_stage(ce
);
904 const char *name
= ce
->name
;
906 while (pos
< istate
->cache_nr
) {
907 struct cache_entry
*p
= istate
->cache
[pos
++];
909 if (len
>= ce_namelen(p
))
911 if (memcmp(name
, p
->name
, len
))
913 if (ce_stage(p
) != stage
)
915 if (p
->name
[len
] != '/')
917 if (p
->ce_flags
& CE_REMOVE
)
922 remove_index_entry_at(istate
, --pos
);
929 * Like strcmp(), but also return the offset of the first change.
930 * If strings are equal, return the length.
932 int strcmp_offset(const char *s1
, const char *s2
, size_t *first_change
)
937 return strcmp(s1
, s2
);
939 for (k
= 0; s1
[k
] == s2
[k
]; k
++)
944 return (unsigned char)s1
[k
] - (unsigned char)s2
[k
];
948 * Do we have another file with a pathname that is a proper
949 * subset of the name we're trying to add?
951 * That is, is there another file in the index with a path
952 * that matches a sub-directory in the given entry?
954 static int has_dir_name(struct index_state
*istate
,
955 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
958 int stage
= ce_stage(ce
);
959 const char *name
= ce
->name
;
960 const char *slash
= name
+ ce_namelen(ce
);
965 * We are frequently called during an iteration on a sorted
966 * list of pathnames and while building a new index. Therefore,
967 * there is a high probability that this entry will eventually
968 * be appended to the index, rather than inserted in the middle.
969 * If we can confirm that, we can avoid binary searches on the
970 * components of the pathname.
972 * Compare the entry's full path with the last path in the index.
974 if (istate
->cache_nr
> 0) {
975 cmp_last
= strcmp_offset(name
,
976 istate
->cache
[istate
->cache_nr
- 1]->name
,
979 if (len_eq_last
== 0) {
981 * The entry sorts AFTER the last one in the
982 * index and their paths have no common prefix,
983 * so there cannot be a F/D conflict.
988 * The entry sorts AFTER the last one in the
989 * index, but has a common prefix. Fall through
990 * to the loop below to disect the entry's path
991 * and see where the difference is.
994 } else if (cmp_last
== 0) {
996 * The entry exactly matches the last one in the
997 * index, but because of multiple stage and CE_REMOVE
998 * items, we fall through and let the regular search
1008 if (*--slash
== '/')
1010 if (slash
<= ce
->name
)
1017 * (len + 1) is a directory boundary (including
1018 * the trailing slash). And since the loop is
1019 * decrementing "slash", the first iteration is
1020 * the longest directory prefix; subsequent
1021 * iterations consider parent directories.
1024 if (len
+ 1 <= len_eq_last
) {
1026 * The directory prefix (including the trailing
1027 * slash) also appears as a prefix in the last
1028 * entry, so the remainder cannot collide (because
1029 * strcmp said the whole path was greater).
1034 * LT: last: xxx/file_A
1040 if (len
> len_eq_last
) {
1042 * This part of the directory prefix (excluding
1043 * the trailing slash) is longer than the known
1044 * equal portions, so this sub-directory cannot
1045 * collide with a file.
1053 if (istate
->cache_nr
> 0 &&
1054 ce_namelen(istate
->cache
[istate
->cache_nr
- 1]) > len
) {
1056 * The directory prefix lines up with part of
1057 * a longer file or directory name, but sorts
1058 * after it, so this sub-directory cannot
1059 * collide with a file.
1061 * last: xxx/yy-file (because '-' sorts before '/')
1068 * This is a possible collision. Fall through and
1069 * let the regular search code handle it.
1076 pos
= index_name_stage_pos(istate
, name
, len
, stage
);
1079 * Found one, but not so fast. This could
1080 * be a marker that says "I was here, but
1081 * I am being removed". Such an entry is
1082 * not a part of the resulting tree, and
1083 * it is Ok to have a directory at the same
1086 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
1090 remove_index_entry_at(istate
, pos
);
1098 * Trivial optimization: if we find an entry that
1099 * already matches the sub-directory, then we know
1100 * we're ok, and we can exit.
1102 while (pos
< istate
->cache_nr
) {
1103 struct cache_entry
*p
= istate
->cache
[pos
];
1104 if ((ce_namelen(p
) <= len
) ||
1105 (p
->name
[len
] != '/') ||
1106 memcmp(p
->name
, name
, len
))
1107 break; /* not our subdirectory */
1108 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
1110 * p is at the same stage as our entry, and
1111 * is a subdirectory of what we are looking
1112 * at, so we cannot have conflicts at our
1113 * level or anything shorter.
1122 /* We may be in a situation where we already have path/file and path
1123 * is being added, or we already have path and path/file is being
1124 * added. Either one would result in a nonsense tree that has path
1125 * twice when git-write-tree tries to write it out. Prevent it.
1127 * If ok-to-replace is specified, we remove the conflicting entries
1128 * from the cache so the caller should recompute the insert position.
1129 * When this happens, we return non-zero.
1131 static int check_file_directory_conflict(struct index_state
*istate
,
1132 const struct cache_entry
*ce
,
1133 int pos
, int ok_to_replace
)
1138 * When ce is an "I am going away" entry, we allow it to be added
1140 if (ce
->ce_flags
& CE_REMOVE
)
1144 * We check if the path is a sub-path of a subsequent pathname
1145 * first, since removing those will not change the position
1148 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
1151 * Then check if the path might have a clashing sub-directory
1154 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
1157 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1160 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
1161 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
1162 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
1163 int new_only
= option
& ADD_CACHE_NEW_ONLY
;
1165 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1166 cache_tree_invalidate_path(istate
, ce
->name
);
1169 * If this entry's path sorts after the last entry in the index,
1170 * we can avoid searching for it.
1172 if (istate
->cache_nr
> 0 &&
1173 strcmp(ce
->name
, istate
->cache
[istate
->cache_nr
- 1]->name
) > 0)
1174 pos
= -istate
->cache_nr
- 1;
1176 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1178 /* existing match? Just replace it. */
1181 replace_index_entry(istate
, pos
, ce
);
1186 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1187 untracked_cache_add_to_index(istate
, ce
->name
);
1190 * Inserting a merged entry ("stage 0") into the index
1191 * will always replace all non-merged entries..
1193 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
1194 while (ce_same_name(istate
->cache
[pos
], ce
)) {
1196 if (!remove_index_entry_at(istate
, pos
))
1203 if (!verify_path(ce
->name
, ce
->ce_mode
))
1204 return error("Invalid path '%s'", ce
->name
);
1206 if (!skip_df_check
&&
1207 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
1209 return error("'%s' appears as both a file and as a directory",
1211 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1217 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1221 if (option
& ADD_CACHE_JUST_APPEND
)
1222 pos
= istate
->cache_nr
;
1225 ret
= add_index_entry_with_check(istate
, ce
, option
);
1231 /* Make sure the array is big enough .. */
1232 ALLOC_GROW(istate
->cache
, istate
->cache_nr
+ 1, istate
->cache_alloc
);
1236 if (istate
->cache_nr
> pos
+ 1)
1237 memmove(istate
->cache
+ pos
+ 1,
1238 istate
->cache
+ pos
,
1239 (istate
->cache_nr
- pos
- 1) * sizeof(ce
));
1240 set_index_entry(istate
, pos
, ce
);
1241 istate
->cache_changed
|= CE_ENTRY_ADDED
;
1246 * "refresh" does not calculate a new sha1 file or bring the
1247 * cache up-to-date for mode/content changes. But what it
1248 * _does_ do is to "re-match" the stat information of a file
1249 * with the cache, so that you can refresh the cache for a
1250 * file that hasn't been changed but where the stat entry is
1253 * For example, you'd want to do this after doing a "git-read-tree",
1254 * to link up the stat cache details with the proper files.
1256 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
1257 struct cache_entry
*ce
,
1258 unsigned int options
, int *err
,
1262 struct cache_entry
*updated
;
1264 int refresh
= options
& CE_MATCH_REFRESH
;
1265 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
1266 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
1267 int ignore_missing
= options
& CE_MATCH_IGNORE_MISSING
;
1269 if (!refresh
|| ce_uptodate(ce
))
1273 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1274 * that the change to the work tree does not matter and told
1277 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
)) {
1278 ce_mark_uptodate(ce
);
1281 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
)) {
1282 ce_mark_uptodate(ce
);
1286 if (has_symlink_leading_path(ce
->name
, ce_namelen(ce
))) {
1294 if (lstat(ce
->name
, &st
) < 0) {
1295 if (ignore_missing
&& errno
== ENOENT
)
1302 changed
= ie_match_stat(istate
, ce
, &st
, options
);
1304 *changed_ret
= changed
;
1307 * The path is unchanged. If we were told to ignore
1308 * valid bit, then we did the actual stat check and
1309 * found that the entry is unmodified. If the entry
1310 * is not marked VALID, this is the place to mark it
1311 * valid again, under "assume unchanged" mode.
1313 if (ignore_valid
&& assume_unchanged
&&
1314 !(ce
->ce_flags
& CE_VALID
))
1315 ; /* mark this one VALID again */
1318 * We do not mark the index itself "modified"
1319 * because CE_UPTODATE flag is in-core only;
1320 * we are not going to write this change out.
1322 if (!S_ISGITLINK(ce
->ce_mode
))
1323 ce_mark_uptodate(ce
);
1328 if (ie_modified(istate
, ce
, &st
, options
)) {
1335 updated
= xmalloc(size
);
1336 memcpy(updated
, ce
, size
);
1337 fill_stat_cache_info(updated
, &st
);
1339 * If ignore_valid is not set, we should leave CE_VALID bit
1340 * alone. Otherwise, paths marked with --no-assume-unchanged
1341 * (i.e. things to be edited) will reacquire CE_VALID bit
1342 * automatically, which is not really what we want.
1344 if (!ignore_valid
&& assume_unchanged
&&
1345 !(ce
->ce_flags
& CE_VALID
))
1346 updated
->ce_flags
&= ~CE_VALID
;
1348 /* istate->cache_changed is updated in the caller */
1352 static void show_file(const char * fmt
, const char * name
, int in_porcelain
,
1353 int * first
, const char *header_msg
)
1355 if (in_porcelain
&& *first
&& header_msg
) {
1356 printf("%s\n", header_msg
);
1362 int refresh_index(struct index_state
*istate
, unsigned int flags
,
1363 const struct pathspec
*pathspec
,
1364 char *seen
, const char *header_msg
)
1368 int really
= (flags
& REFRESH_REALLY
) != 0;
1369 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
1370 int quiet
= (flags
& REFRESH_QUIET
) != 0;
1371 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
1372 int ignore_submodules
= (flags
& REFRESH_IGNORE_SUBMODULES
) != 0;
1374 int in_porcelain
= (flags
& REFRESH_IN_PORCELAIN
);
1375 unsigned int options
= (CE_MATCH_REFRESH
|
1376 (really
? CE_MATCH_IGNORE_VALID
: 0) |
1377 (not_new
? CE_MATCH_IGNORE_MISSING
: 0));
1378 const char *modified_fmt
;
1379 const char *deleted_fmt
;
1380 const char *typechange_fmt
;
1381 const char *added_fmt
;
1382 const char *unmerged_fmt
;
1384 modified_fmt
= (in_porcelain
? "M\t%s\n" : "%s: needs update\n");
1385 deleted_fmt
= (in_porcelain
? "D\t%s\n" : "%s: needs update\n");
1386 typechange_fmt
= (in_porcelain
? "T\t%s\n" : "%s needs update\n");
1387 added_fmt
= (in_porcelain
? "A\t%s\n" : "%s needs update\n");
1388 unmerged_fmt
= (in_porcelain
? "U\t%s\n" : "%s: needs merge\n");
1389 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1390 struct cache_entry
*ce
, *new;
1391 int cache_errno
= 0;
1395 ce
= istate
->cache
[i
];
1396 if (ignore_submodules
&& S_ISGITLINK(ce
->ce_mode
))
1399 if (pathspec
&& !ce_path_match(ce
, pathspec
, seen
))
1403 while ((i
< istate
->cache_nr
) &&
1404 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
1410 show_file(unmerged_fmt
, ce
->name
, in_porcelain
,
1411 &first
, header_msg
);
1419 new = refresh_cache_ent(istate
, ce
, options
, &cache_errno
, &changed
);
1425 if (really
&& cache_errno
== EINVAL
) {
1426 /* If we are doing --really-refresh that
1427 * means the index is not valid anymore.
1429 ce
->ce_flags
&= ~CE_VALID
;
1430 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
1431 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
1436 if (cache_errno
== ENOENT
)
1438 else if (ce_intent_to_add(ce
))
1439 fmt
= added_fmt
; /* must be before other checks */
1440 else if (changed
& TYPE_CHANGED
)
1441 fmt
= typechange_fmt
;
1445 ce
->name
, in_porcelain
, &first
, header_msg
);
1450 replace_index_entry(istate
, i
, new);
1455 struct cache_entry
*refresh_cache_entry(struct cache_entry
*ce
,
1456 unsigned int options
)
1458 return refresh_cache_ent(&the_index
, ce
, options
, NULL
, NULL
);
1462 /*****************************************************************
1464 *****************************************************************/
1466 #define INDEX_FORMAT_DEFAULT 3
1468 static unsigned int get_index_format_default(void)
1470 char *envversion
= getenv("GIT_INDEX_VERSION");
1473 unsigned int version
= INDEX_FORMAT_DEFAULT
;
1476 if (!git_config_get_int("index.version", &value
))
1478 if (version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1479 warning(_("index.version set, but the value is invalid.\n"
1480 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1481 return INDEX_FORMAT_DEFAULT
;
1486 version
= strtoul(envversion
, &endp
, 10);
1488 version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1489 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1490 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1491 version
= INDEX_FORMAT_DEFAULT
;
1497 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1498 * Again - this is just a (very strong in practice) heuristic that
1499 * the inode hasn't changed.
1501 * We save the fields in big-endian order to allow using the
1502 * index file over NFS transparently.
1504 struct ondisk_cache_entry
{
1505 struct cache_time ctime
;
1506 struct cache_time mtime
;
1513 unsigned char sha1
[20];
1515 char name
[FLEX_ARRAY
]; /* more */
1519 * This struct is used when CE_EXTENDED bit is 1
1520 * The struct must match ondisk_cache_entry exactly from
1523 struct ondisk_cache_entry_extended
{
1524 struct cache_time ctime
;
1525 struct cache_time mtime
;
1532 unsigned char sha1
[20];
1535 char name
[FLEX_ARRAY
]; /* more */
1538 /* These are only used for v3 or lower */
1539 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1540 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1541 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1542 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1543 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1544 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1545 ondisk_cache_entry_size(ce_namelen(ce)))
1547 /* Allow fsck to force verification of the index checksum. */
1548 int verify_index_checksum
;
1550 static int verify_hdr(struct cache_header
*hdr
, unsigned long size
)
1553 unsigned char sha1
[20];
1556 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
1557 return error("bad signature");
1558 hdr_version
= ntohl(hdr
->hdr_version
);
1559 if (hdr_version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< hdr_version
)
1560 return error("bad index version %d", hdr_version
);
1562 if (!verify_index_checksum
)
1566 git_SHA1_Update(&c
, hdr
, size
- 20);
1567 git_SHA1_Final(sha1
, &c
);
1568 if (hashcmp(sha1
, (unsigned char *)hdr
+ size
- 20))
1569 return error("bad index file sha1 signature");
1573 static int read_index_extension(struct index_state
*istate
,
1574 const char *ext
, void *data
, unsigned long sz
)
1576 switch (CACHE_EXT(ext
)) {
1577 case CACHE_EXT_TREE
:
1578 istate
->cache_tree
= cache_tree_read(data
, sz
);
1580 case CACHE_EXT_RESOLVE_UNDO
:
1581 istate
->resolve_undo
= resolve_undo_read(data
, sz
);
1583 case CACHE_EXT_LINK
:
1584 if (read_link_extension(istate
, data
, sz
))
1587 case CACHE_EXT_UNTRACKED
:
1588 istate
->untracked
= read_untracked_extension(data
, sz
);
1591 if (*ext
< 'A' || 'Z' < *ext
)
1592 return error("index uses %.4s extension, which we do not understand",
1594 fprintf(stderr
, "ignoring %.4s extension\n", ext
);
1600 int hold_locked_index(struct lock_file
*lk
, int lock_flags
)
1602 return hold_lock_file_for_update(lk
, get_index_file(), lock_flags
);
1605 int read_index(struct index_state
*istate
)
1607 return read_index_from(istate
, get_index_file());
1610 static struct cache_entry
*cache_entry_from_ondisk(struct ondisk_cache_entry
*ondisk
,
1615 struct cache_entry
*ce
= xmalloc(cache_entry_size(len
));
1617 ce
->ce_stat_data
.sd_ctime
.sec
= get_be32(&ondisk
->ctime
.sec
);
1618 ce
->ce_stat_data
.sd_mtime
.sec
= get_be32(&ondisk
->mtime
.sec
);
1619 ce
->ce_stat_data
.sd_ctime
.nsec
= get_be32(&ondisk
->ctime
.nsec
);
1620 ce
->ce_stat_data
.sd_mtime
.nsec
= get_be32(&ondisk
->mtime
.nsec
);
1621 ce
->ce_stat_data
.sd_dev
= get_be32(&ondisk
->dev
);
1622 ce
->ce_stat_data
.sd_ino
= get_be32(&ondisk
->ino
);
1623 ce
->ce_mode
= get_be32(&ondisk
->mode
);
1624 ce
->ce_stat_data
.sd_uid
= get_be32(&ondisk
->uid
);
1625 ce
->ce_stat_data
.sd_gid
= get_be32(&ondisk
->gid
);
1626 ce
->ce_stat_data
.sd_size
= get_be32(&ondisk
->size
);
1627 ce
->ce_flags
= flags
& ~CE_NAMEMASK
;
1628 ce
->ce_namelen
= len
;
1630 hashcpy(ce
->oid
.hash
, ondisk
->sha1
);
1631 memcpy(ce
->name
, name
, len
);
1632 ce
->name
[len
] = '\0';
1637 * Adjacent cache entries tend to share the leading paths, so it makes
1638 * sense to only store the differences in later entries. In the v4
1639 * on-disk format of the index, each on-disk cache entry stores the
1640 * number of bytes to be stripped from the end of the previous name,
1641 * and the bytes to append to the result, to come up with its name.
1643 static unsigned long expand_name_field(struct strbuf
*name
, const char *cp_
)
1645 const unsigned char *ep
, *cp
= (const unsigned char *)cp_
;
1646 size_t len
= decode_varint(&cp
);
1648 if (name
->len
< len
)
1649 die("malformed name field in the index");
1650 strbuf_remove(name
, name
->len
- len
, len
);
1651 for (ep
= cp
; *ep
; ep
++)
1652 ; /* find the end */
1653 strbuf_add(name
, cp
, ep
- cp
);
1654 return (const char *)ep
+ 1 - cp_
;
1657 static struct cache_entry
*create_from_disk(struct ondisk_cache_entry
*ondisk
,
1658 unsigned long *ent_size
,
1659 struct strbuf
*previous_name
)
1661 struct cache_entry
*ce
;
1666 /* On-disk flags are just 16 bits */
1667 flags
= get_be16(&ondisk
->flags
);
1668 len
= flags
& CE_NAMEMASK
;
1670 if (flags
& CE_EXTENDED
) {
1671 struct ondisk_cache_entry_extended
*ondisk2
;
1673 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
1674 extended_flags
= get_be16(&ondisk2
->flags2
) << 16;
1675 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1676 if (extended_flags
& ~CE_EXTENDED_FLAGS
)
1677 die("Unknown index entry format %08x", extended_flags
);
1678 flags
|= extended_flags
;
1679 name
= ondisk2
->name
;
1682 name
= ondisk
->name
;
1684 if (!previous_name
) {
1685 /* v3 and earlier */
1686 if (len
== CE_NAMEMASK
)
1688 ce
= cache_entry_from_ondisk(ondisk
, flags
, name
, len
);
1690 *ent_size
= ondisk_ce_size(ce
);
1692 unsigned long consumed
;
1693 consumed
= expand_name_field(previous_name
, name
);
1694 ce
= cache_entry_from_ondisk(ondisk
, flags
,
1696 previous_name
->len
);
1698 *ent_size
= (name
- ((char *)ondisk
)) + consumed
;
1703 static void check_ce_order(struct index_state
*istate
)
1707 for (i
= 1; i
< istate
->cache_nr
; i
++) {
1708 struct cache_entry
*ce
= istate
->cache
[i
- 1];
1709 struct cache_entry
*next_ce
= istate
->cache
[i
];
1710 int name_compare
= strcmp(ce
->name
, next_ce
->name
);
1712 if (0 < name_compare
)
1713 die("unordered stage entries in index");
1714 if (!name_compare
) {
1716 die("multiple stage entries for merged file '%s'",
1718 if (ce_stage(ce
) > ce_stage(next_ce
))
1719 die("unordered stage entries for '%s'",
1725 static void tweak_untracked_cache(struct index_state
*istate
)
1727 switch (git_config_get_untracked_cache()) {
1728 case -1: /* keep: do nothing */
1731 remove_untracked_cache(istate
);
1734 add_untracked_cache(istate
);
1736 default: /* unknown value: do nothing */
1741 static void tweak_split_index(struct index_state
*istate
)
1743 switch (git_config_get_split_index()) {
1744 case -1: /* unset: do nothing */
1747 remove_split_index(istate
);
1750 add_split_index(istate
);
1752 default: /* unknown value: do nothing */
1757 static void post_read_index_from(struct index_state
*istate
)
1759 check_ce_order(istate
);
1760 tweak_untracked_cache(istate
);
1761 tweak_split_index(istate
);
1764 /* remember to discard_cache() before reading a different cache! */
1765 int do_read_index(struct index_state
*istate
, const char *path
, int must_exist
)
1769 unsigned long src_offset
;
1770 struct cache_header
*hdr
;
1773 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
1775 if (istate
->initialized
)
1776 return istate
->cache_nr
;
1778 istate
->timestamp
.sec
= 0;
1779 istate
->timestamp
.nsec
= 0;
1780 fd
= open(path
, O_RDONLY
);
1782 if (!must_exist
&& errno
== ENOENT
)
1784 die_errno("%s: index file open failed", path
);
1788 die_errno("cannot stat the open index");
1790 mmap_size
= xsize_t(st
.st_size
);
1791 if (mmap_size
< sizeof(struct cache_header
) + 20)
1792 die("index file smaller than expected");
1794 mmap
= xmmap(NULL
, mmap_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1795 if (mmap
== MAP_FAILED
)
1796 die_errno("unable to map index file");
1800 if (verify_hdr(hdr
, mmap_size
) < 0)
1803 hashcpy(istate
->sha1
, (const unsigned char *)hdr
+ mmap_size
- 20);
1804 istate
->version
= ntohl(hdr
->hdr_version
);
1805 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
1806 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
1807 istate
->cache
= xcalloc(istate
->cache_alloc
, sizeof(*istate
->cache
));
1808 istate
->initialized
= 1;
1810 if (istate
->version
== 4)
1811 previous_name
= &previous_name_buf
;
1813 previous_name
= NULL
;
1815 src_offset
= sizeof(*hdr
);
1816 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1817 struct ondisk_cache_entry
*disk_ce
;
1818 struct cache_entry
*ce
;
1819 unsigned long consumed
;
1821 disk_ce
= (struct ondisk_cache_entry
*)((char *)mmap
+ src_offset
);
1822 ce
= create_from_disk(disk_ce
, &consumed
, previous_name
);
1823 set_index_entry(istate
, i
, ce
);
1825 src_offset
+= consumed
;
1827 strbuf_release(&previous_name_buf
);
1828 istate
->timestamp
.sec
= st
.st_mtime
;
1829 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
1831 while (src_offset
<= mmap_size
- 20 - 8) {
1832 /* After an array of active_nr index entries,
1833 * there can be arbitrary number of extended
1834 * sections, each of which is prefixed with
1835 * extension name (4-byte) and section length
1836 * in 4-byte network byte order.
1839 memcpy(&extsize
, (char *)mmap
+ src_offset
+ 4, 4);
1840 extsize
= ntohl(extsize
);
1841 if (read_index_extension(istate
,
1842 (const char *) mmap
+ src_offset
,
1843 (char *) mmap
+ src_offset
+ 8,
1847 src_offset
+= extsize
;
1849 munmap(mmap
, mmap_size
);
1850 return istate
->cache_nr
;
1853 munmap(mmap
, mmap_size
);
1854 die("index file corrupt");
1858 * Signal that the shared index is used by updating its mtime.
1860 * This way, shared index can be removed if they have not been used
1863 static void freshen_shared_index(char *base_sha1_hex
, int warn
)
1865 char *shared_index
= git_pathdup("sharedindex.%s", base_sha1_hex
);
1866 if (!check_and_freshen_file(shared_index
, 1) && warn
)
1867 warning("could not freshen shared index '%s'", shared_index
);
1871 int read_index_from(struct index_state
*istate
, const char *path
)
1873 struct split_index
*split_index
;
1875 char *base_sha1_hex
;
1876 const char *base_path
;
1878 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1879 if (istate
->initialized
)
1880 return istate
->cache_nr
;
1882 ret
= do_read_index(istate
, path
, 0);
1884 split_index
= istate
->split_index
;
1885 if (!split_index
|| is_null_sha1(split_index
->base_sha1
)) {
1886 post_read_index_from(istate
);
1890 if (split_index
->base
)
1891 discard_index(split_index
->base
);
1893 split_index
->base
= xcalloc(1, sizeof(*split_index
->base
));
1895 base_sha1_hex
= sha1_to_hex(split_index
->base_sha1
);
1896 base_path
= git_path("sharedindex.%s", base_sha1_hex
);
1897 ret
= do_read_index(split_index
->base
, base_path
, 1);
1898 if (hashcmp(split_index
->base_sha1
, split_index
->base
->sha1
))
1899 die("broken index, expect %s in %s, got %s",
1900 base_sha1_hex
, base_path
,
1901 sha1_to_hex(split_index
->base
->sha1
));
1903 freshen_shared_index(base_sha1_hex
, 0);
1904 merge_base_index(istate
);
1905 post_read_index_from(istate
);
1909 int is_index_unborn(struct index_state
*istate
)
1911 return (!istate
->cache_nr
&& !istate
->timestamp
.sec
);
1914 int discard_index(struct index_state
*istate
)
1918 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1919 if (istate
->cache
[i
]->index
&&
1920 istate
->split_index
&&
1921 istate
->split_index
->base
&&
1922 istate
->cache
[i
]->index
<= istate
->split_index
->base
->cache_nr
&&
1923 istate
->cache
[i
] == istate
->split_index
->base
->cache
[istate
->cache
[i
]->index
- 1])
1925 free(istate
->cache
[i
]);
1927 resolve_undo_clear_index(istate
);
1928 istate
->cache_nr
= 0;
1929 istate
->cache_changed
= 0;
1930 istate
->timestamp
.sec
= 0;
1931 istate
->timestamp
.nsec
= 0;
1932 free_name_hash(istate
);
1933 cache_tree_free(&(istate
->cache_tree
));
1934 istate
->initialized
= 0;
1935 FREE_AND_NULL(istate
->cache
);
1936 istate
->cache_alloc
= 0;
1937 discard_split_index(istate
);
1938 free_untracked_cache(istate
->untracked
);
1939 istate
->untracked
= NULL
;
1943 int unmerged_index(const struct index_state
*istate
)
1946 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1947 if (ce_stage(istate
->cache
[i
]))
1953 #define WRITE_BUFFER_SIZE 8192
1954 static unsigned char write_buffer
[WRITE_BUFFER_SIZE
];
1955 static unsigned long write_buffer_len
;
1957 static int ce_write_flush(git_SHA_CTX
*context
, int fd
)
1959 unsigned int buffered
= write_buffer_len
;
1961 git_SHA1_Update(context
, write_buffer
, buffered
);
1962 if (write_in_full(fd
, write_buffer
, buffered
) < 0)
1964 write_buffer_len
= 0;
1969 static int ce_write(git_SHA_CTX
*context
, int fd
, void *data
, unsigned int len
)
1972 unsigned int buffered
= write_buffer_len
;
1973 unsigned int partial
= WRITE_BUFFER_SIZE
- buffered
;
1976 memcpy(write_buffer
+ buffered
, data
, partial
);
1977 buffered
+= partial
;
1978 if (buffered
== WRITE_BUFFER_SIZE
) {
1979 write_buffer_len
= buffered
;
1980 if (ce_write_flush(context
, fd
))
1984 write_buffer_len
= buffered
;
1986 data
= (char *) data
+ partial
;
1991 static int write_index_ext_header(git_SHA_CTX
*context
, int fd
,
1992 unsigned int ext
, unsigned int sz
)
1996 return ((ce_write(context
, fd
, &ext
, 4) < 0) ||
1997 (ce_write(context
, fd
, &sz
, 4) < 0)) ? -1 : 0;
2000 static int ce_flush(git_SHA_CTX
*context
, int fd
, unsigned char *sha1
)
2002 unsigned int left
= write_buffer_len
;
2005 write_buffer_len
= 0;
2006 git_SHA1_Update(context
, write_buffer
, left
);
2009 /* Flush first if not enough space for SHA1 signature */
2010 if (left
+ 20 > WRITE_BUFFER_SIZE
) {
2011 if (write_in_full(fd
, write_buffer
, left
) < 0)
2016 /* Append the SHA1 signature at the end */
2017 git_SHA1_Final(write_buffer
+ left
, context
);
2018 hashcpy(sha1
, write_buffer
+ left
);
2020 return (write_in_full(fd
, write_buffer
, left
) < 0) ? -1 : 0;
2023 static void ce_smudge_racily_clean_entry(struct cache_entry
*ce
)
2026 * The only thing we care about in this function is to smudge the
2027 * falsely clean entry due to touch-update-touch race, so we leave
2028 * everything else as they are. We are called for entries whose
2029 * ce_stat_data.sd_mtime match the index file mtime.
2031 * Note that this actually does not do much for gitlinks, for
2032 * which ce_match_stat_basic() always goes to the actual
2033 * contents. The caller checks with is_racy_timestamp() which
2034 * always says "no" for gitlinks, so we are not called for them ;-)
2038 if (lstat(ce
->name
, &st
) < 0)
2040 if (ce_match_stat_basic(ce
, &st
))
2042 if (ce_modified_check_fs(ce
, &st
)) {
2043 /* This is "racily clean"; smudge it. Note that this
2044 * is a tricky code. At first glance, it may appear
2045 * that it can break with this sequence:
2047 * $ echo xyzzy >frotz
2048 * $ git-update-index --add frotz
2051 * $ echo filfre >nitfol
2052 * $ git-update-index --add nitfol
2054 * but it does not. When the second update-index runs,
2055 * it notices that the entry "frotz" has the same timestamp
2056 * as index, and if we were to smudge it by resetting its
2057 * size to zero here, then the object name recorded
2058 * in index is the 6-byte file but the cached stat information
2059 * becomes zero --- which would then match what we would
2060 * obtain from the filesystem next time we stat("frotz").
2062 * However, the second update-index, before calling
2063 * this function, notices that the cached size is 6
2064 * bytes and what is on the filesystem is an empty
2065 * file, and never calls us, so the cached size information
2066 * for "frotz" stays 6 which does not match the filesystem.
2068 ce
->ce_stat_data
.sd_size
= 0;
2072 /* Copy miscellaneous fields but not the name */
2073 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry
*ondisk
,
2074 struct cache_entry
*ce
)
2078 ondisk
->ctime
.sec
= htonl(ce
->ce_stat_data
.sd_ctime
.sec
);
2079 ondisk
->mtime
.sec
= htonl(ce
->ce_stat_data
.sd_mtime
.sec
);
2080 ondisk
->ctime
.nsec
= htonl(ce
->ce_stat_data
.sd_ctime
.nsec
);
2081 ondisk
->mtime
.nsec
= htonl(ce
->ce_stat_data
.sd_mtime
.nsec
);
2082 ondisk
->dev
= htonl(ce
->ce_stat_data
.sd_dev
);
2083 ondisk
->ino
= htonl(ce
->ce_stat_data
.sd_ino
);
2084 ondisk
->mode
= htonl(ce
->ce_mode
);
2085 ondisk
->uid
= htonl(ce
->ce_stat_data
.sd_uid
);
2086 ondisk
->gid
= htonl(ce
->ce_stat_data
.sd_gid
);
2087 ondisk
->size
= htonl(ce
->ce_stat_data
.sd_size
);
2088 hashcpy(ondisk
->sha1
, ce
->oid
.hash
);
2090 flags
= ce
->ce_flags
& ~CE_NAMEMASK
;
2091 flags
|= (ce_namelen(ce
) >= CE_NAMEMASK
? CE_NAMEMASK
: ce_namelen(ce
));
2092 ondisk
->flags
= htons(flags
);
2093 if (ce
->ce_flags
& CE_EXTENDED
) {
2094 struct ondisk_cache_entry_extended
*ondisk2
;
2095 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
2096 ondisk2
->flags2
= htons((ce
->ce_flags
& CE_EXTENDED_FLAGS
) >> 16);
2100 static int ce_write_entry(git_SHA_CTX
*c
, int fd
, struct cache_entry
*ce
,
2101 struct strbuf
*previous_name
, struct ondisk_cache_entry
*ondisk
)
2104 int saved_namelen
= saved_namelen
; /* compiler workaround */
2106 static unsigned char padding
[8] = { 0x00 };
2108 if (ce
->ce_flags
& CE_STRIP_NAME
) {
2109 saved_namelen
= ce_namelen(ce
);
2113 if (ce
->ce_flags
& CE_EXTENDED
)
2114 size
= offsetof(struct ondisk_cache_entry_extended
, name
);
2116 size
= offsetof(struct ondisk_cache_entry
, name
);
2118 if (!previous_name
) {
2119 int len
= ce_namelen(ce
);
2120 copy_cache_entry_to_ondisk(ondisk
, ce
);
2121 result
= ce_write(c
, fd
, ondisk
, size
);
2123 result
= ce_write(c
, fd
, ce
->name
, len
);
2125 result
= ce_write(c
, fd
, padding
, align_padding_size(size
, len
));
2127 int common
, to_remove
, prefix_size
;
2128 unsigned char to_remove_vi
[16];
2130 (ce
->name
[common
] &&
2131 common
< previous_name
->len
&&
2132 ce
->name
[common
] == previous_name
->buf
[common
]);
2134 ; /* still matching */
2135 to_remove
= previous_name
->len
- common
;
2136 prefix_size
= encode_varint(to_remove
, to_remove_vi
);
2138 copy_cache_entry_to_ondisk(ondisk
, ce
);
2139 result
= ce_write(c
, fd
, ondisk
, size
);
2141 result
= ce_write(c
, fd
, to_remove_vi
, prefix_size
);
2143 result
= ce_write(c
, fd
, ce
->name
+ common
, ce_namelen(ce
) - common
);
2145 result
= ce_write(c
, fd
, padding
, 1);
2147 strbuf_splice(previous_name
, common
, to_remove
,
2148 ce
->name
+ common
, ce_namelen(ce
) - common
);
2150 if (ce
->ce_flags
& CE_STRIP_NAME
) {
2151 ce
->ce_namelen
= saved_namelen
;
2152 ce
->ce_flags
&= ~CE_STRIP_NAME
;
2159 * This function verifies if index_state has the correct sha1 of the
2160 * index file. Don't die if we have any other failure, just return 0.
2162 static int verify_index_from(const struct index_state
*istate
, const char *path
)
2167 unsigned char sha1
[20];
2169 if (!istate
->initialized
)
2172 fd
= open(path
, O_RDONLY
);
2179 if (st
.st_size
< sizeof(struct cache_header
) + 20)
2182 n
= pread_in_full(fd
, sha1
, 20, st
.st_size
- 20);
2186 if (hashcmp(istate
->sha1
, sha1
))
2197 static int verify_index(const struct index_state
*istate
)
2199 return verify_index_from(istate
, get_index_file());
2202 static int has_racy_timestamp(struct index_state
*istate
)
2204 int entries
= istate
->cache_nr
;
2207 for (i
= 0; i
< entries
; i
++) {
2208 struct cache_entry
*ce
= istate
->cache
[i
];
2209 if (is_racy_timestamp(istate
, ce
))
2216 * Opportunistically update the index but do not complain if we can't
2218 void update_index_if_able(struct index_state
*istate
, struct lock_file
*lockfile
)
2220 if ((istate
->cache_changed
|| has_racy_timestamp(istate
)) &&
2221 verify_index(istate
) &&
2222 write_locked_index(istate
, lockfile
, COMMIT_LOCK
))
2223 rollback_lock_file(lockfile
);
2226 static int do_write_index(struct index_state
*istate
, struct tempfile
*tempfile
,
2227 int strip_extensions
)
2229 int newfd
= tempfile
->fd
;
2231 struct cache_header hdr
;
2232 int i
, err
= 0, removed
, extended
, hdr_version
;
2233 struct cache_entry
**cache
= istate
->cache
;
2234 int entries
= istate
->cache_nr
;
2236 struct ondisk_cache_entry_extended ondisk
;
2237 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
2238 int drop_cache_tree
= 0;
2240 for (i
= removed
= extended
= 0; i
< entries
; i
++) {
2241 if (cache
[i
]->ce_flags
& CE_REMOVE
)
2244 /* reduce extended entries if possible */
2245 cache
[i
]->ce_flags
&= ~CE_EXTENDED
;
2246 if (cache
[i
]->ce_flags
& CE_EXTENDED_FLAGS
) {
2248 cache
[i
]->ce_flags
|= CE_EXTENDED
;
2252 if (!istate
->version
) {
2253 istate
->version
= get_index_format_default();
2254 if (getenv("GIT_TEST_SPLIT_INDEX"))
2255 init_split_index(istate
);
2258 /* demote version 3 to version 2 when the latter suffices */
2259 if (istate
->version
== 3 || istate
->version
== 2)
2260 istate
->version
= extended
? 3 : 2;
2262 hdr_version
= istate
->version
;
2264 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
2265 hdr
.hdr_version
= htonl(hdr_version
);
2266 hdr
.hdr_entries
= htonl(entries
- removed
);
2269 if (ce_write(&c
, newfd
, &hdr
, sizeof(hdr
)) < 0)
2272 previous_name
= (hdr_version
== 4) ? &previous_name_buf
: NULL
;
2274 for (i
= 0; i
< entries
; i
++) {
2275 struct cache_entry
*ce
= cache
[i
];
2276 if (ce
->ce_flags
& CE_REMOVE
)
2278 if (!ce_uptodate(ce
) && is_racy_timestamp(istate
, ce
))
2279 ce_smudge_racily_clean_entry(ce
);
2280 if (is_null_oid(&ce
->oid
)) {
2281 static const char msg
[] = "cache entry has null sha1: %s";
2282 static int allow
= -1;
2285 allow
= git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2287 warning(msg
, ce
->name
);
2289 err
= error(msg
, ce
->name
);
2291 drop_cache_tree
= 1;
2293 if (ce_write_entry(&c
, newfd
, ce
, previous_name
, (struct ondisk_cache_entry
*)&ondisk
) < 0)
2299 strbuf_release(&previous_name_buf
);
2304 /* Write extension data here */
2305 if (!strip_extensions
&& istate
->split_index
) {
2306 struct strbuf sb
= STRBUF_INIT
;
2308 err
= write_link_extension(&sb
, istate
) < 0 ||
2309 write_index_ext_header(&c
, newfd
, CACHE_EXT_LINK
,
2311 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2312 strbuf_release(&sb
);
2316 if (!strip_extensions
&& !drop_cache_tree
&& istate
->cache_tree
) {
2317 struct strbuf sb
= STRBUF_INIT
;
2319 cache_tree_write(&sb
, istate
->cache_tree
);
2320 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_TREE
, sb
.len
) < 0
2321 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2322 strbuf_release(&sb
);
2326 if (!strip_extensions
&& istate
->resolve_undo
) {
2327 struct strbuf sb
= STRBUF_INIT
;
2329 resolve_undo_write(&sb
, istate
->resolve_undo
);
2330 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_RESOLVE_UNDO
,
2332 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2333 strbuf_release(&sb
);
2337 if (!strip_extensions
&& istate
->untracked
) {
2338 struct strbuf sb
= STRBUF_INIT
;
2340 write_untracked_extension(&sb
, istate
->untracked
);
2341 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_UNTRACKED
,
2343 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2344 strbuf_release(&sb
);
2349 if (ce_flush(&c
, newfd
, istate
->sha1
))
2351 if (close_tempfile_gently(tempfile
)) {
2352 error(_("could not close '%s'"), tempfile
->filename
.buf
);
2353 delete_tempfile(&tempfile
);
2356 if (stat(tempfile
->filename
.buf
, &st
))
2358 istate
->timestamp
.sec
= (unsigned int)st
.st_mtime
;
2359 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
2363 void set_alternate_index_output(const char *name
)
2365 alternate_index_output
= name
;
2368 static int commit_locked_index(struct lock_file
*lk
)
2370 if (alternate_index_output
)
2371 return commit_lock_file_to(lk
, alternate_index_output
);
2373 return commit_lock_file(lk
);
2376 static int do_write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
2379 int ret
= do_write_index(istate
, lock
->tempfile
, 0);
2382 assert((flags
& (COMMIT_LOCK
| CLOSE_LOCK
)) !=
2383 (COMMIT_LOCK
| CLOSE_LOCK
));
2384 if (flags
& COMMIT_LOCK
)
2385 return commit_locked_index(lock
);
2386 else if (flags
& CLOSE_LOCK
)
2387 return close_lock_file_gently(lock
);
2392 static int write_split_index(struct index_state
*istate
,
2393 struct lock_file
*lock
,
2397 prepare_to_write_split_index(istate
);
2398 ret
= do_write_locked_index(istate
, lock
, flags
);
2399 finish_writing_split_index(istate
);
2403 static const char *shared_index_expire
= "2.weeks.ago";
2405 static unsigned long get_shared_index_expire_date(void)
2407 static unsigned long shared_index_expire_date
;
2408 static int shared_index_expire_date_prepared
;
2410 if (!shared_index_expire_date_prepared
) {
2411 git_config_get_expiry("splitindex.sharedindexexpire",
2412 &shared_index_expire
);
2413 shared_index_expire_date
= approxidate(shared_index_expire
);
2414 shared_index_expire_date_prepared
= 1;
2417 return shared_index_expire_date
;
2420 static int should_delete_shared_index(const char *shared_index_path
)
2423 unsigned long expiration
;
2425 /* Check timestamp */
2426 expiration
= get_shared_index_expire_date();
2429 if (stat(shared_index_path
, &st
))
2430 return error_errno(_("could not stat '%s'"), shared_index_path
);
2431 if (st
.st_mtime
> expiration
)
2437 static int clean_shared_index_files(const char *current_hex
)
2440 DIR *dir
= opendir(get_git_dir());
2443 return error_errno(_("unable to open git dir: %s"), get_git_dir());
2445 while ((de
= readdir(dir
)) != NULL
) {
2446 const char *sha1_hex
;
2447 const char *shared_index_path
;
2448 if (!skip_prefix(de
->d_name
, "sharedindex.", &sha1_hex
))
2450 if (!strcmp(sha1_hex
, current_hex
))
2452 shared_index_path
= git_path("%s", de
->d_name
);
2453 if (should_delete_shared_index(shared_index_path
) > 0 &&
2454 unlink(shared_index_path
))
2455 warning_errno(_("unable to unlink: %s"), shared_index_path
);
2462 static int write_shared_index(struct index_state
*istate
,
2463 struct lock_file
*lock
, unsigned flags
)
2465 struct tempfile
*temp
;
2466 struct split_index
*si
= istate
->split_index
;
2469 temp
= mks_tempfile(git_path("sharedindex_XXXXXX"));
2471 hashclr(si
->base_sha1
);
2472 return do_write_locked_index(istate
, lock
, flags
);
2474 move_cache_to_base_index(istate
);
2475 ret
= do_write_index(si
->base
, temp
, 1);
2477 delete_tempfile(&temp
);
2480 ret
= adjust_shared_perm(get_tempfile_path(temp
));
2482 int save_errno
= errno
;
2483 error("cannot fix permission bits on %s", get_tempfile_path(temp
));
2484 delete_tempfile(&temp
);
2488 ret
= rename_tempfile(&temp
,
2489 git_path("sharedindex.%s", sha1_to_hex(si
->base
->sha1
)));
2491 hashcpy(si
->base_sha1
, si
->base
->sha1
);
2492 clean_shared_index_files(sha1_to_hex(si
->base
->sha1
));
2498 static const int default_max_percent_split_change
= 20;
2500 static int too_many_not_shared_entries(struct index_state
*istate
)
2502 int i
, not_shared
= 0;
2503 int max_split
= git_config_get_max_percent_split_change();
2505 switch (max_split
) {
2507 /* not or badly configured: use the default value */
2508 max_split
= default_max_percent_split_change
;
2511 return 1; /* 0% means always write a new shared index */
2513 return 0; /* 100% means never write a new shared index */
2515 break; /* just use the configured value */
2518 /* Count not shared entries */
2519 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2520 struct cache_entry
*ce
= istate
->cache
[i
];
2525 return (int64_t)istate
->cache_nr
* max_split
< (int64_t)not_shared
* 100;
2528 int write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
2531 int new_shared_index
, ret
;
2532 struct split_index
*si
= istate
->split_index
;
2534 if (!si
|| alternate_index_output
||
2535 (istate
->cache_changed
& ~EXTMASK
)) {
2537 hashclr(si
->base_sha1
);
2538 return do_write_locked_index(istate
, lock
, flags
);
2541 if (getenv("GIT_TEST_SPLIT_INDEX")) {
2542 int v
= si
->base_sha1
[0];
2544 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
2546 if (too_many_not_shared_entries(istate
))
2547 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
2549 new_shared_index
= istate
->cache_changed
& SPLIT_INDEX_ORDERED
;
2551 if (new_shared_index
) {
2552 ret
= write_shared_index(istate
, lock
, flags
);
2557 ret
= write_split_index(istate
, lock
, flags
);
2559 /* Freshen the shared index only if the split-index was written */
2560 if (!ret
&& !new_shared_index
)
2561 freshen_shared_index(sha1_to_hex(si
->base_sha1
), 1);
2567 * Read the index file that is potentially unmerged into given
2568 * index_state, dropping any unmerged entries. Returns true if
2569 * the index is unmerged. Callers who want to refuse to work
2570 * from an unmerged state can call this and check its return value,
2571 * instead of calling read_cache().
2573 int read_index_unmerged(struct index_state
*istate
)
2579 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2580 struct cache_entry
*ce
= istate
->cache
[i
];
2581 struct cache_entry
*new_ce
;
2587 len
= ce_namelen(ce
);
2588 size
= cache_entry_size(len
);
2589 new_ce
= xcalloc(1, size
);
2590 memcpy(new_ce
->name
, ce
->name
, len
);
2591 new_ce
->ce_flags
= create_ce_flags(0) | CE_CONFLICTED
;
2592 new_ce
->ce_namelen
= len
;
2593 new_ce
->ce_mode
= ce
->ce_mode
;
2594 if (add_index_entry(istate
, new_ce
, 0))
2595 return error("%s: cannot drop to stage #0",
2602 * Returns 1 if the path is an "other" path with respect to
2603 * the index; that is, the path is not mentioned in the index at all,
2604 * either as a file, a directory with some files in the index,
2605 * or as an unmerged entry.
2607 * We helpfully remove a trailing "/" from directories so that
2608 * the output of read_directory can be used as-is.
2610 int index_name_is_other(const struct index_state
*istate
, const char *name
,
2614 if (namelen
&& name
[namelen
- 1] == '/')
2616 pos
= index_name_pos(istate
, name
, namelen
);
2618 return 0; /* exact match */
2620 if (pos
< istate
->cache_nr
) {
2621 struct cache_entry
*ce
= istate
->cache
[pos
];
2622 if (ce_namelen(ce
) == namelen
&&
2623 !memcmp(ce
->name
, name
, namelen
))
2624 return 0; /* Yup, this one exists unmerged */
2629 void *read_blob_data_from_index(const struct index_state
*istate
,
2630 const char *path
, unsigned long *size
)
2634 enum object_type type
;
2638 pos
= index_name_pos(istate
, path
, len
);
2641 * We might be in the middle of a merge, in which
2642 * case we would read stage #2 (ours).
2646 (pos
< 0 && i
< istate
->cache_nr
&&
2647 !strcmp(istate
->cache
[i
]->name
, path
));
2649 if (ce_stage(istate
->cache
[i
]) == 2)
2654 data
= read_sha1_file(istate
->cache
[pos
]->oid
.hash
, &type
, &sz
);
2655 if (!data
|| type
!= OBJ_BLOB
) {
2664 void stat_validity_clear(struct stat_validity
*sv
)
2666 FREE_AND_NULL(sv
->sd
);
2669 int stat_validity_check(struct stat_validity
*sv
, const char *path
)
2673 if (stat(path
, &st
) < 0)
2674 return sv
->sd
== NULL
;
2677 return S_ISREG(st
.st_mode
) && !match_stat_data(sv
->sd
, &st
);
2680 void stat_validity_update(struct stat_validity
*sv
, int fd
)
2684 if (fstat(fd
, &st
) < 0 || !S_ISREG(st
.st_mode
))
2685 stat_validity_clear(sv
);
2688 sv
->sd
= xcalloc(1, sizeof(struct stat_data
));
2689 fill_stat_data(sv
->sd
, &st
);
2693 void move_index_extensions(struct index_state
*dst
, struct index_state
*src
)
2695 dst
->untracked
= src
->untracked
;
2696 src
->untracked
= NULL
;