2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 #define NO_THE_INDEX_COMPATIBILITY_MACROS
10 #include "cache-tree.h"
16 #include "resolve-undo.h"
19 #include "split-index.h"
22 static struct cache_entry
*refresh_cache_entry(struct cache_entry
*ce
,
23 unsigned int options
);
25 /* Mask for the name length in ce_flags in the on-disk index */
27 #define CE_NAMEMASK (0x0fff)
31 * The first letter should be 'A'..'Z' for extensions that are not
32 * necessary for a correct operation (i.e. optimization data).
33 * When new extensions are added that _needs_ to be understood in
34 * order to correctly interpret the index file, pick character that
35 * is outside the range, to cause the reader to abort.
38 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
39 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
40 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
41 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
42 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
44 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
45 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
46 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
47 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED)
49 struct index_state the_index
;
50 static const char *alternate_index_output
;
52 static void set_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
54 istate
->cache
[nr
] = ce
;
55 add_name_hash(istate
, ce
);
58 static void replace_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
60 struct cache_entry
*old
= istate
->cache
[nr
];
62 replace_index_entry_in_base(istate
, old
, ce
);
63 remove_name_hash(istate
, old
);
65 set_index_entry(istate
, nr
, ce
);
66 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
67 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
70 void rename_index_entry_at(struct index_state
*istate
, int nr
, const char *new_name
)
72 struct cache_entry
*old
= istate
->cache
[nr
], *new;
73 int namelen
= strlen(new_name
);
75 new = xmalloc(cache_entry_size(namelen
));
76 copy_cache_entry(new, old
);
77 new->ce_flags
&= ~CE_HASHED
;
78 new->ce_namelen
= namelen
;
80 memcpy(new->name
, new_name
, namelen
+ 1);
82 cache_tree_invalidate_path(istate
, old
->name
);
83 untracked_cache_remove_from_index(istate
, old
->name
);
84 remove_index_entry_at(istate
, nr
);
85 add_index_entry(istate
, new, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
88 void fill_stat_data(struct stat_data
*sd
, struct stat
*st
)
90 sd
->sd_ctime
.sec
= (unsigned int)st
->st_ctime
;
91 sd
->sd_mtime
.sec
= (unsigned int)st
->st_mtime
;
92 sd
->sd_ctime
.nsec
= ST_CTIME_NSEC(*st
);
93 sd
->sd_mtime
.nsec
= ST_MTIME_NSEC(*st
);
94 sd
->sd_dev
= st
->st_dev
;
95 sd
->sd_ino
= st
->st_ino
;
96 sd
->sd_uid
= st
->st_uid
;
97 sd
->sd_gid
= st
->st_gid
;
98 sd
->sd_size
= st
->st_size
;
101 int match_stat_data(const struct stat_data
*sd
, struct stat
*st
)
105 if (sd
->sd_mtime
.sec
!= (unsigned int)st
->st_mtime
)
106 changed
|= MTIME_CHANGED
;
107 if (trust_ctime
&& check_stat
&&
108 sd
->sd_ctime
.sec
!= (unsigned int)st
->st_ctime
)
109 changed
|= CTIME_CHANGED
;
112 if (check_stat
&& sd
->sd_mtime
.nsec
!= ST_MTIME_NSEC(*st
))
113 changed
|= MTIME_CHANGED
;
114 if (trust_ctime
&& check_stat
&&
115 sd
->sd_ctime
.nsec
!= ST_CTIME_NSEC(*st
))
116 changed
|= CTIME_CHANGED
;
120 if (sd
->sd_uid
!= (unsigned int) st
->st_uid
||
121 sd
->sd_gid
!= (unsigned int) st
->st_gid
)
122 changed
|= OWNER_CHANGED
;
123 if (sd
->sd_ino
!= (unsigned int) st
->st_ino
)
124 changed
|= INODE_CHANGED
;
129 * st_dev breaks on network filesystems where different
130 * clients will have different views of what "device"
131 * the filesystem is on
133 if (check_stat
&& sd
->sd_dev
!= (unsigned int) st
->st_dev
)
134 changed
|= INODE_CHANGED
;
137 if (sd
->sd_size
!= (unsigned int) st
->st_size
)
138 changed
|= DATA_CHANGED
;
144 * This only updates the "non-critical" parts of the directory
145 * cache, ie the parts that aren't tracked by GIT, and only used
146 * to validate the cache.
148 void fill_stat_cache_info(struct cache_entry
*ce
, struct stat
*st
)
150 fill_stat_data(&ce
->ce_stat_data
, st
);
152 if (assume_unchanged
)
153 ce
->ce_flags
|= CE_VALID
;
155 if (S_ISREG(st
->st_mode
))
156 ce_mark_uptodate(ce
);
159 static int ce_compare_data(const struct cache_entry
*ce
, struct stat
*st
)
162 int fd
= open(ce
->name
, O_RDONLY
);
165 unsigned char sha1
[20];
166 if (!index_fd(sha1
, fd
, st
, OBJ_BLOB
, ce
->name
, 0))
167 match
= hashcmp(sha1
, ce
->sha1
);
168 /* index_fd() closed the file descriptor already */
173 static int ce_compare_link(const struct cache_entry
*ce
, size_t expected_size
)
178 enum object_type type
;
179 struct strbuf sb
= STRBUF_INIT
;
181 if (strbuf_readlink(&sb
, ce
->name
, expected_size
))
184 buffer
= read_sha1_file(ce
->sha1
, &type
, &size
);
187 match
= memcmp(buffer
, sb
.buf
, size
);
194 static int ce_compare_gitlink(const struct cache_entry
*ce
)
196 unsigned char sha1
[20];
199 * We don't actually require that the .git directory
200 * under GITLINK directory be a valid git directory. It
201 * might even be missing (in case nobody populated that
204 * If so, we consider it always to match.
206 if (resolve_gitlink_ref(ce
->name
, "HEAD", sha1
) < 0)
208 return hashcmp(sha1
, ce
->sha1
);
211 static int ce_modified_check_fs(const struct cache_entry
*ce
, struct stat
*st
)
213 switch (st
->st_mode
& S_IFMT
) {
215 if (ce_compare_data(ce
, st
))
219 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
223 if (S_ISGITLINK(ce
->ce_mode
))
224 return ce_compare_gitlink(ce
) ? DATA_CHANGED
: 0;
231 static int ce_match_stat_basic(const struct cache_entry
*ce
, struct stat
*st
)
233 unsigned int changed
= 0;
235 if (ce
->ce_flags
& CE_REMOVE
)
236 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
238 switch (ce
->ce_mode
& S_IFMT
) {
240 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
241 /* We consider only the owner x bit to be relevant for
244 if (trust_executable_bit
&&
245 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
246 changed
|= MODE_CHANGED
;
249 if (!S_ISLNK(st
->st_mode
) &&
250 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
251 changed
|= TYPE_CHANGED
;
254 /* We ignore most of the st_xxx fields for gitlinks */
255 if (!S_ISDIR(st
->st_mode
))
256 changed
|= TYPE_CHANGED
;
257 else if (ce_compare_gitlink(ce
))
258 changed
|= DATA_CHANGED
;
261 die("internal error: ce_mode is %o", ce
->ce_mode
);
264 changed
|= match_stat_data(&ce
->ce_stat_data
, st
);
266 /* Racily smudged entry? */
267 if (!ce
->ce_stat_data
.sd_size
) {
268 if (!is_empty_blob_sha1(ce
->sha1
))
269 changed
|= DATA_CHANGED
;
275 static int is_racy_stat(const struct index_state
*istate
,
276 const struct stat_data
*sd
)
278 return (istate
->timestamp
.sec
&&
280 /* nanosecond timestamped files can also be racy! */
281 (istate
->timestamp
.sec
< sd
->sd_mtime
.sec
||
282 (istate
->timestamp
.sec
== sd
->sd_mtime
.sec
&&
283 istate
->timestamp
.nsec
<= sd
->sd_mtime
.nsec
))
285 istate
->timestamp
.sec
<= sd
->sd_mtime
.sec
290 static int is_racy_timestamp(const struct index_state
*istate
,
291 const struct cache_entry
*ce
)
293 return (!S_ISGITLINK(ce
->ce_mode
) &&
294 is_racy_stat(istate
, &ce
->ce_stat_data
));
297 int match_stat_data_racy(const struct index_state
*istate
,
298 const struct stat_data
*sd
, struct stat
*st
)
300 if (is_racy_stat(istate
, sd
))
301 return MTIME_CHANGED
;
302 return match_stat_data(sd
, st
);
305 int ie_match_stat(const struct index_state
*istate
,
306 const struct cache_entry
*ce
, struct stat
*st
,
307 unsigned int options
)
309 unsigned int changed
;
310 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
311 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
312 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
315 * If it's marked as always valid in the index, it's
316 * valid whatever the checked-out copy says.
318 * skip-worktree has the same effect with higher precedence
320 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
))
322 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
326 * Intent-to-add entries have not been added, so the index entry
327 * by definition never matches what is in the work tree until it
328 * actually gets added.
330 if (ce_intent_to_add(ce
))
331 return DATA_CHANGED
| TYPE_CHANGED
| MODE_CHANGED
;
333 changed
= ce_match_stat_basic(ce
, st
);
336 * Within 1 second of this sequence:
337 * echo xyzzy >file && git-update-index --add file
338 * running this command:
340 * would give a falsely clean cache entry. The mtime and
341 * length match the cache, and other stat fields do not change.
343 * We could detect this at update-index time (the cache entry
344 * being registered/updated records the same time as "now")
345 * and delay the return from git-update-index, but that would
346 * effectively mean we can make at most one commit per second,
347 * which is not acceptable. Instead, we check cache entries
348 * whose mtime are the same as the index file timestamp more
349 * carefully than others.
351 if (!changed
&& is_racy_timestamp(istate
, ce
)) {
352 if (assume_racy_is_modified
)
353 changed
|= DATA_CHANGED
;
355 changed
|= ce_modified_check_fs(ce
, st
);
361 int ie_modified(const struct index_state
*istate
,
362 const struct cache_entry
*ce
,
363 struct stat
*st
, unsigned int options
)
365 int changed
, changed_fs
;
367 changed
= ie_match_stat(istate
, ce
, st
, options
);
371 * If the mode or type has changed, there's no point in trying
372 * to refresh the entry - it's not going to match
374 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
378 * Immediately after read-tree or update-index --cacheinfo,
379 * the length field is zero, as we have never even read the
380 * lstat(2) information once, and we cannot trust DATA_CHANGED
381 * returned by ie_match_stat() which in turn was returned by
382 * ce_match_stat_basic() to signal that the filesize of the
383 * blob changed. We have to actually go to the filesystem to
384 * see if the contents match, and if so, should answer "unchanged".
386 * The logic does not apply to gitlinks, as ce_match_stat_basic()
387 * already has checked the actual HEAD from the filesystem in the
388 * subproject. If ie_match_stat() already said it is different,
389 * then we know it is.
391 if ((changed
& DATA_CHANGED
) &&
392 (S_ISGITLINK(ce
->ce_mode
) || ce
->ce_stat_data
.sd_size
!= 0))
395 changed_fs
= ce_modified_check_fs(ce
, st
);
397 return changed
| changed_fs
;
401 int base_name_compare(const char *name1
, int len1
, int mode1
,
402 const char *name2
, int len2
, int mode2
)
404 unsigned char c1
, c2
;
405 int len
= len1
< len2
? len1
: len2
;
408 cmp
= memcmp(name1
, name2
, len
);
413 if (!c1
&& S_ISDIR(mode1
))
415 if (!c2
&& S_ISDIR(mode2
))
417 return (c1
< c2
) ? -1 : (c1
> c2
) ? 1 : 0;
421 * df_name_compare() is identical to base_name_compare(), except it
422 * compares conflicting directory/file entries as equal. Note that
423 * while a directory name compares as equal to a regular file, they
424 * then individually compare _differently_ to a filename that has
425 * a dot after the basename (because '\0' < '.' < '/').
427 * This is used by routines that want to traverse the git namespace
428 * but then handle conflicting entries together when possible.
430 int df_name_compare(const char *name1
, int len1
, int mode1
,
431 const char *name2
, int len2
, int mode2
)
433 int len
= len1
< len2
? len1
: len2
, cmp
;
434 unsigned char c1
, c2
;
436 cmp
= memcmp(name1
, name2
, len
);
439 /* Directories and files compare equal (same length, same name) */
443 if (!c1
&& S_ISDIR(mode1
))
446 if (!c2
&& S_ISDIR(mode2
))
448 if (c1
== '/' && !c2
)
450 if (c2
== '/' && !c1
)
455 int name_compare(const char *name1
, size_t len1
, const char *name2
, size_t len2
)
457 size_t min_len
= (len1
< len2
) ? len1
: len2
;
458 int cmp
= memcmp(name1
, name2
, min_len
);
468 int cache_name_stage_compare(const char *name1
, int len1
, int stage1
, const char *name2
, int len2
, int stage2
)
472 cmp
= name_compare(name1
, len1
, name2
, len2
);
483 static int index_name_stage_pos(const struct index_state
*istate
, const char *name
, int namelen
, int stage
)
488 last
= istate
->cache_nr
;
489 while (last
> first
) {
490 int next
= (last
+ first
) >> 1;
491 struct cache_entry
*ce
= istate
->cache
[next
];
492 int cmp
= cache_name_stage_compare(name
, namelen
, stage
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
504 int index_name_pos(const struct index_state
*istate
, const char *name
, int namelen
)
506 return index_name_stage_pos(istate
, name
, namelen
, 0);
509 /* Remove entry, return true if there are more entries to go.. */
510 int remove_index_entry_at(struct index_state
*istate
, int pos
)
512 struct cache_entry
*ce
= istate
->cache
[pos
];
514 record_resolve_undo(istate
, ce
);
515 remove_name_hash(istate
, ce
);
516 save_or_free_index_entry(istate
, ce
);
517 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
519 if (pos
>= istate
->cache_nr
)
521 memmove(istate
->cache
+ pos
,
522 istate
->cache
+ pos
+ 1,
523 (istate
->cache_nr
- pos
) * sizeof(struct cache_entry
*));
528 * Remove all cache entries marked for removal, that is where
529 * CE_REMOVE is set in ce_flags. This is much more effective than
530 * calling remove_index_entry_at() for each entry to be removed.
532 void remove_marked_cache_entries(struct index_state
*istate
)
534 struct cache_entry
**ce_array
= istate
->cache
;
537 for (i
= j
= 0; i
< istate
->cache_nr
; i
++) {
538 if (ce_array
[i
]->ce_flags
& CE_REMOVE
) {
539 remove_name_hash(istate
, ce_array
[i
]);
540 save_or_free_index_entry(istate
, ce_array
[i
]);
543 ce_array
[j
++] = ce_array
[i
];
545 if (j
== istate
->cache_nr
)
547 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
548 istate
->cache_nr
= j
;
551 int remove_file_from_index(struct index_state
*istate
, const char *path
)
553 int pos
= index_name_pos(istate
, path
, strlen(path
));
556 cache_tree_invalidate_path(istate
, path
);
557 untracked_cache_remove_from_index(istate
, path
);
558 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
559 remove_index_entry_at(istate
, pos
);
563 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
565 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
568 static int index_name_pos_also_unmerged(struct index_state
*istate
,
569 const char *path
, int namelen
)
571 int pos
= index_name_pos(istate
, path
, namelen
);
572 struct cache_entry
*ce
;
577 /* maybe unmerged? */
579 if (pos
>= istate
->cache_nr
||
580 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
583 /* order of preference: stage 2, 1, 3 */
584 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
585 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
586 !compare_name(ce
, path
, namelen
))
591 static int different_name(struct cache_entry
*ce
, struct cache_entry
*alias
)
593 int len
= ce_namelen(ce
);
594 return ce_namelen(alias
) != len
|| memcmp(ce
->name
, alias
->name
, len
);
598 * If we add a filename that aliases in the cache, we will use the
599 * name that we already have - but we don't want to update the same
600 * alias twice, because that implies that there were actually two
601 * different files with aliasing names!
603 * So we use the CE_ADDED flag to verify that the alias was an old
604 * one before we accept it as
606 static struct cache_entry
*create_alias_ce(struct index_state
*istate
,
607 struct cache_entry
*ce
,
608 struct cache_entry
*alias
)
611 struct cache_entry
*new;
613 if (alias
->ce_flags
& CE_ADDED
)
614 die("Will not add file alias '%s' ('%s' already exists in index)", ce
->name
, alias
->name
);
616 /* Ok, create the new entry using the name of the existing alias */
617 len
= ce_namelen(alias
);
618 new = xcalloc(1, cache_entry_size(len
));
619 memcpy(new->name
, alias
->name
, len
);
620 copy_cache_entry(new, ce
);
621 save_or_free_index_entry(istate
, ce
);
625 void set_object_name_for_intent_to_add_entry(struct cache_entry
*ce
)
627 unsigned char sha1
[20];
628 if (write_sha1_file("", 0, blob_type
, sha1
))
629 die("cannot create an empty blob in the object database");
630 hashcpy(ce
->sha1
, sha1
);
633 int add_to_index(struct index_state
*istate
, const char *path
, struct stat
*st
, int flags
)
635 int size
, namelen
, was_same
;
636 mode_t st_mode
= st
->st_mode
;
637 struct cache_entry
*ce
, *alias
;
638 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_IGNORE_SKIP_WORKTREE
|CE_MATCH_RACY_IS_DIRTY
;
639 int verbose
= flags
& (ADD_CACHE_VERBOSE
| ADD_CACHE_PRETEND
);
640 int pretend
= flags
& ADD_CACHE_PRETEND
;
641 int intent_only
= flags
& ADD_CACHE_INTENT
;
642 int add_option
= (ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
|
643 (intent_only
? ADD_CACHE_NEW_ONLY
: 0));
645 if (!S_ISREG(st_mode
) && !S_ISLNK(st_mode
) && !S_ISDIR(st_mode
))
646 return error("%s: can only add regular files, symbolic links or git-directories", path
);
648 namelen
= strlen(path
);
649 if (S_ISDIR(st_mode
)) {
650 while (namelen
&& path
[namelen
-1] == '/')
653 size
= cache_entry_size(namelen
);
654 ce
= xcalloc(1, size
);
655 memcpy(ce
->name
, path
, namelen
);
656 ce
->ce_namelen
= namelen
;
658 fill_stat_cache_info(ce
, st
);
660 ce
->ce_flags
|= CE_INTENT_TO_ADD
;
663 if (trust_executable_bit
&& has_symlinks
) {
664 ce
->ce_mode
= create_ce_mode(st_mode
);
666 /* If there is an existing entry, pick the mode bits and type
667 * from it, otherwise assume unexecutable regular file.
669 struct cache_entry
*ent
;
670 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
672 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
673 ce
->ce_mode
= ce_mode_from_stat(ent
, st_mode
);
676 /* When core.ignorecase=true, determine if a directory of the same name but differing
677 * case already exists within the Git repository. If it does, ensure the directory
678 * case of the file being added to the repository matches (is folded into) the existing
679 * entry's directory case.
682 adjust_dirname_case(istate
, ce
->name
);
685 alias
= index_file_exists(istate
, ce
->name
, ce_namelen(ce
), ignore_case
);
686 if (alias
&& !ce_stage(alias
) && !ie_match_stat(istate
, alias
, st
, ce_option
)) {
687 /* Nothing changed, really */
688 if (!S_ISGITLINK(alias
->ce_mode
))
689 ce_mark_uptodate(alias
);
690 alias
->ce_flags
|= CE_ADDED
;
696 if (index_path(ce
->sha1
, path
, st
, HASH_WRITE_OBJECT
)) {
698 return error("unable to index file %s", path
);
701 set_object_name_for_intent_to_add_entry(ce
);
703 if (ignore_case
&& alias
&& different_name(ce
, alias
))
704 ce
= create_alias_ce(istate
, ce
, alias
);
705 ce
->ce_flags
|= CE_ADDED
;
707 /* It was suspected to be racily clean, but it turns out to be Ok */
710 !hashcmp(alias
->sha1
, ce
->sha1
) &&
711 ce
->ce_mode
== alias
->ce_mode
);
715 else if (add_index_entry(istate
, ce
, add_option
)) {
717 return error("unable to add %s to index", path
);
719 if (verbose
&& !was_same
)
720 printf("add '%s'\n", path
);
724 int add_file_to_index(struct index_state
*istate
, const char *path
, int flags
)
727 if (lstat(path
, &st
))
728 die_errno("unable to stat '%s'", path
);
729 return add_to_index(istate
, path
, &st
, flags
);
732 struct cache_entry
*make_cache_entry(unsigned int mode
,
733 const unsigned char *sha1
, const char *path
, int stage
,
734 unsigned int refresh_options
)
737 struct cache_entry
*ce
, *ret
;
739 if (!verify_path(path
)) {
740 error("Invalid path '%s'", path
);
745 size
= cache_entry_size(len
);
746 ce
= xcalloc(1, size
);
748 hashcpy(ce
->sha1
, sha1
);
749 memcpy(ce
->name
, path
, len
);
750 ce
->ce_flags
= create_ce_flags(stage
);
751 ce
->ce_namelen
= len
;
752 ce
->ce_mode
= create_ce_mode(mode
);
754 ret
= refresh_cache_entry(ce
, refresh_options
);
761 * Chmod an index entry with either +x or -x.
763 * Returns -1 if the chmod for the particular cache entry failed (if it's
764 * not a regular file), -2 if an invalid flip argument is passed in, 0
767 int chmod_index_entry(struct index_state
*istate
, struct cache_entry
*ce
,
770 if (!S_ISREG(ce
->ce_mode
))
777 ce
->ce_mode
&= ~0111;
782 cache_tree_invalidate_path(istate
, ce
->name
);
783 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
784 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
789 int ce_same_name(const struct cache_entry
*a
, const struct cache_entry
*b
)
791 int len
= ce_namelen(a
);
792 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
796 * We fundamentally don't like some paths: we don't want
797 * dot or dot-dot anywhere, and for obvious reasons don't
798 * want to recurse into ".git" either.
800 * Also, we don't want double slashes or slashes at the
801 * end that can make pathnames ambiguous.
803 static int verify_dotfile(const char *rest
)
806 * The first character was '.', but that
807 * has already been discarded, we now test
811 /* "." is not allowed */
812 if (*rest
== '\0' || is_dir_sep(*rest
))
817 * ".git" followed by NUL or slash is bad. This
818 * shares the path end test with the ".." case.
822 if (rest
[1] != 'i' && rest
[1] != 'I')
824 if (rest
[2] != 't' && rest
[2] != 'T')
829 if (rest
[1] == '\0' || is_dir_sep(rest
[1]))
835 int verify_path(const char *path
)
839 if (has_dos_drive_prefix(path
))
848 if (protect_hfs
&& is_hfs_dotgit(path
))
850 if (protect_ntfs
&& is_ntfs_dotgit(path
))
853 if ((c
== '.' && !verify_dotfile(path
)) ||
854 is_dir_sep(c
) || c
== '\0')
862 * Do we have another file that has the beginning components being a
863 * proper superset of the name we're trying to add?
865 static int has_file_name(struct index_state
*istate
,
866 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
869 int len
= ce_namelen(ce
);
870 int stage
= ce_stage(ce
);
871 const char *name
= ce
->name
;
873 while (pos
< istate
->cache_nr
) {
874 struct cache_entry
*p
= istate
->cache
[pos
++];
876 if (len
>= ce_namelen(p
))
878 if (memcmp(name
, p
->name
, len
))
880 if (ce_stage(p
) != stage
)
882 if (p
->name
[len
] != '/')
884 if (p
->ce_flags
& CE_REMOVE
)
889 remove_index_entry_at(istate
, --pos
);
895 * Do we have another file with a pathname that is a proper
896 * subset of the name we're trying to add?
898 static int has_dir_name(struct index_state
*istate
,
899 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
902 int stage
= ce_stage(ce
);
903 const char *name
= ce
->name
;
904 const char *slash
= name
+ ce_namelen(ce
);
912 if (slash
<= ce
->name
)
917 pos
= index_name_stage_pos(istate
, name
, len
, stage
);
920 * Found one, but not so fast. This could
921 * be a marker that says "I was here, but
922 * I am being removed". Such an entry is
923 * not a part of the resulting tree, and
924 * it is Ok to have a directory at the same
927 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
931 remove_index_entry_at(istate
, pos
);
939 * Trivial optimization: if we find an entry that
940 * already matches the sub-directory, then we know
941 * we're ok, and we can exit.
943 while (pos
< istate
->cache_nr
) {
944 struct cache_entry
*p
= istate
->cache
[pos
];
945 if ((ce_namelen(p
) <= len
) ||
946 (p
->name
[len
] != '/') ||
947 memcmp(p
->name
, name
, len
))
948 break; /* not our subdirectory */
949 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
951 * p is at the same stage as our entry, and
952 * is a subdirectory of what we are looking
953 * at, so we cannot have conflicts at our
954 * level or anything shorter.
963 /* We may be in a situation where we already have path/file and path
964 * is being added, or we already have path and path/file is being
965 * added. Either one would result in a nonsense tree that has path
966 * twice when git-write-tree tries to write it out. Prevent it.
968 * If ok-to-replace is specified, we remove the conflicting entries
969 * from the cache so the caller should recompute the insert position.
970 * When this happens, we return non-zero.
972 static int check_file_directory_conflict(struct index_state
*istate
,
973 const struct cache_entry
*ce
,
974 int pos
, int ok_to_replace
)
979 * When ce is an "I am going away" entry, we allow it to be added
981 if (ce
->ce_flags
& CE_REMOVE
)
985 * We check if the path is a sub-path of a subsequent pathname
986 * first, since removing those will not change the position
989 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
992 * Then check if the path might have a clashing sub-directory
995 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
998 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1001 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
1002 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
1003 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
1004 int new_only
= option
& ADD_CACHE_NEW_ONLY
;
1006 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1007 cache_tree_invalidate_path(istate
, ce
->name
);
1008 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1010 /* existing match? Just replace it. */
1013 replace_index_entry(istate
, pos
, ce
);
1018 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1019 untracked_cache_add_to_index(istate
, ce
->name
);
1022 * Inserting a merged entry ("stage 0") into the index
1023 * will always replace all non-merged entries..
1025 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
1026 while (ce_same_name(istate
->cache
[pos
], ce
)) {
1028 if (!remove_index_entry_at(istate
, pos
))
1035 if (!verify_path(ce
->name
))
1036 return error("Invalid path '%s'", ce
->name
);
1038 if (!skip_df_check
&&
1039 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
1041 return error("'%s' appears as both a file and as a directory",
1043 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1049 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1053 if (option
& ADD_CACHE_JUST_APPEND
)
1054 pos
= istate
->cache_nr
;
1057 ret
= add_index_entry_with_check(istate
, ce
, option
);
1063 /* Make sure the array is big enough .. */
1064 ALLOC_GROW(istate
->cache
, istate
->cache_nr
+ 1, istate
->cache_alloc
);
1068 if (istate
->cache_nr
> pos
+ 1)
1069 memmove(istate
->cache
+ pos
+ 1,
1070 istate
->cache
+ pos
,
1071 (istate
->cache_nr
- pos
- 1) * sizeof(ce
));
1072 set_index_entry(istate
, pos
, ce
);
1073 istate
->cache_changed
|= CE_ENTRY_ADDED
;
1078 * "refresh" does not calculate a new sha1 file or bring the
1079 * cache up-to-date for mode/content changes. But what it
1080 * _does_ do is to "re-match" the stat information of a file
1081 * with the cache, so that you can refresh the cache for a
1082 * file that hasn't been changed but where the stat entry is
1085 * For example, you'd want to do this after doing a "git-read-tree",
1086 * to link up the stat cache details with the proper files.
1088 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
1089 struct cache_entry
*ce
,
1090 unsigned int options
, int *err
,
1094 struct cache_entry
*updated
;
1096 int refresh
= options
& CE_MATCH_REFRESH
;
1097 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
1098 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
1099 int ignore_missing
= options
& CE_MATCH_IGNORE_MISSING
;
1101 if (!refresh
|| ce_uptodate(ce
))
1105 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1106 * that the change to the work tree does not matter and told
1109 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
)) {
1110 ce_mark_uptodate(ce
);
1113 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
)) {
1114 ce_mark_uptodate(ce
);
1118 if (has_symlink_leading_path(ce
->name
, ce_namelen(ce
))) {
1126 if (lstat(ce
->name
, &st
) < 0) {
1127 if (ignore_missing
&& errno
== ENOENT
)
1134 changed
= ie_match_stat(istate
, ce
, &st
, options
);
1136 *changed_ret
= changed
;
1139 * The path is unchanged. If we were told to ignore
1140 * valid bit, then we did the actual stat check and
1141 * found that the entry is unmodified. If the entry
1142 * is not marked VALID, this is the place to mark it
1143 * valid again, under "assume unchanged" mode.
1145 if (ignore_valid
&& assume_unchanged
&&
1146 !(ce
->ce_flags
& CE_VALID
))
1147 ; /* mark this one VALID again */
1150 * We do not mark the index itself "modified"
1151 * because CE_UPTODATE flag is in-core only;
1152 * we are not going to write this change out.
1154 if (!S_ISGITLINK(ce
->ce_mode
))
1155 ce_mark_uptodate(ce
);
1160 if (ie_modified(istate
, ce
, &st
, options
)) {
1167 updated
= xmalloc(size
);
1168 memcpy(updated
, ce
, size
);
1169 fill_stat_cache_info(updated
, &st
);
1171 * If ignore_valid is not set, we should leave CE_VALID bit
1172 * alone. Otherwise, paths marked with --no-assume-unchanged
1173 * (i.e. things to be edited) will reacquire CE_VALID bit
1174 * automatically, which is not really what we want.
1176 if (!ignore_valid
&& assume_unchanged
&&
1177 !(ce
->ce_flags
& CE_VALID
))
1178 updated
->ce_flags
&= ~CE_VALID
;
1180 /* istate->cache_changed is updated in the caller */
1184 static void show_file(const char * fmt
, const char * name
, int in_porcelain
,
1185 int * first
, const char *header_msg
)
1187 if (in_porcelain
&& *first
&& header_msg
) {
1188 printf("%s\n", header_msg
);
1194 int refresh_index(struct index_state
*istate
, unsigned int flags
,
1195 const struct pathspec
*pathspec
,
1196 char *seen
, const char *header_msg
)
1200 int really
= (flags
& REFRESH_REALLY
) != 0;
1201 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
1202 int quiet
= (flags
& REFRESH_QUIET
) != 0;
1203 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
1204 int ignore_submodules
= (flags
& REFRESH_IGNORE_SUBMODULES
) != 0;
1206 int in_porcelain
= (flags
& REFRESH_IN_PORCELAIN
);
1207 unsigned int options
= (CE_MATCH_REFRESH
|
1208 (really
? CE_MATCH_IGNORE_VALID
: 0) |
1209 (not_new
? CE_MATCH_IGNORE_MISSING
: 0));
1210 const char *modified_fmt
;
1211 const char *deleted_fmt
;
1212 const char *typechange_fmt
;
1213 const char *added_fmt
;
1214 const char *unmerged_fmt
;
1216 modified_fmt
= (in_porcelain
? "M\t%s\n" : "%s: needs update\n");
1217 deleted_fmt
= (in_porcelain
? "D\t%s\n" : "%s: needs update\n");
1218 typechange_fmt
= (in_porcelain
? "T\t%s\n" : "%s needs update\n");
1219 added_fmt
= (in_porcelain
? "A\t%s\n" : "%s needs update\n");
1220 unmerged_fmt
= (in_porcelain
? "U\t%s\n" : "%s: needs merge\n");
1221 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1222 struct cache_entry
*ce
, *new;
1223 int cache_errno
= 0;
1227 ce
= istate
->cache
[i
];
1228 if (ignore_submodules
&& S_ISGITLINK(ce
->ce_mode
))
1231 if (pathspec
&& !ce_path_match(ce
, pathspec
, seen
))
1235 while ((i
< istate
->cache_nr
) &&
1236 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
1242 show_file(unmerged_fmt
, ce
->name
, in_porcelain
,
1243 &first
, header_msg
);
1251 new = refresh_cache_ent(istate
, ce
, options
, &cache_errno
, &changed
);
1257 if (really
&& cache_errno
== EINVAL
) {
1258 /* If we are doing --really-refresh that
1259 * means the index is not valid anymore.
1261 ce
->ce_flags
&= ~CE_VALID
;
1262 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
1263 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
1268 if (cache_errno
== ENOENT
)
1270 else if (ce_intent_to_add(ce
))
1271 fmt
= added_fmt
; /* must be before other checks */
1272 else if (changed
& TYPE_CHANGED
)
1273 fmt
= typechange_fmt
;
1277 ce
->name
, in_porcelain
, &first
, header_msg
);
1282 replace_index_entry(istate
, i
, new);
1287 static struct cache_entry
*refresh_cache_entry(struct cache_entry
*ce
,
1288 unsigned int options
)
1290 return refresh_cache_ent(&the_index
, ce
, options
, NULL
, NULL
);
1294 /*****************************************************************
1296 *****************************************************************/
1298 #define INDEX_FORMAT_DEFAULT 3
1300 static unsigned int get_index_format_default(void)
1302 char *envversion
= getenv("GIT_INDEX_VERSION");
1305 unsigned int version
= INDEX_FORMAT_DEFAULT
;
1308 if (!git_config_get_int("index.version", &value
))
1310 if (version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1311 warning(_("index.version set, but the value is invalid.\n"
1312 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1313 return INDEX_FORMAT_DEFAULT
;
1318 version
= strtoul(envversion
, &endp
, 10);
1320 version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1321 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1322 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1323 version
= INDEX_FORMAT_DEFAULT
;
1329 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1330 * Again - this is just a (very strong in practice) heuristic that
1331 * the inode hasn't changed.
1333 * We save the fields in big-endian order to allow using the
1334 * index file over NFS transparently.
1336 struct ondisk_cache_entry
{
1337 struct cache_time ctime
;
1338 struct cache_time mtime
;
1345 unsigned char sha1
[20];
1347 char name
[FLEX_ARRAY
]; /* more */
1351 * This struct is used when CE_EXTENDED bit is 1
1352 * The struct must match ondisk_cache_entry exactly from
1355 struct ondisk_cache_entry_extended
{
1356 struct cache_time ctime
;
1357 struct cache_time mtime
;
1364 unsigned char sha1
[20];
1367 char name
[FLEX_ARRAY
]; /* more */
1370 /* These are only used for v3 or lower */
1371 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1372 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1373 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1374 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1375 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1376 ondisk_cache_entry_size(ce_namelen(ce)))
1378 static int verify_hdr(struct cache_header
*hdr
, unsigned long size
)
1381 unsigned char sha1
[20];
1384 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
1385 return error("bad signature");
1386 hdr_version
= ntohl(hdr
->hdr_version
);
1387 if (hdr_version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< hdr_version
)
1388 return error("bad index version %d", hdr_version
);
1390 git_SHA1_Update(&c
, hdr
, size
- 20);
1391 git_SHA1_Final(sha1
, &c
);
1392 if (hashcmp(sha1
, (unsigned char *)hdr
+ size
- 20))
1393 return error("bad index file sha1 signature");
1397 static int read_index_extension(struct index_state
*istate
,
1398 const char *ext
, void *data
, unsigned long sz
)
1400 switch (CACHE_EXT(ext
)) {
1401 case CACHE_EXT_TREE
:
1402 istate
->cache_tree
= cache_tree_read(data
, sz
);
1404 case CACHE_EXT_RESOLVE_UNDO
:
1405 istate
->resolve_undo
= resolve_undo_read(data
, sz
);
1407 case CACHE_EXT_LINK
:
1408 if (read_link_extension(istate
, data
, sz
))
1411 case CACHE_EXT_UNTRACKED
:
1412 istate
->untracked
= read_untracked_extension(data
, sz
);
1415 if (*ext
< 'A' || 'Z' < *ext
)
1416 return error("index uses %.4s extension, which we do not understand",
1418 fprintf(stderr
, "ignoring %.4s extension\n", ext
);
1424 int hold_locked_index(struct lock_file
*lk
, int die_on_error
)
1426 return hold_lock_file_for_update(lk
, get_index_file(),
1432 int read_index(struct index_state
*istate
)
1434 return read_index_from(istate
, get_index_file());
1437 static struct cache_entry
*cache_entry_from_ondisk(struct ondisk_cache_entry
*ondisk
,
1442 struct cache_entry
*ce
= xmalloc(cache_entry_size(len
));
1444 ce
->ce_stat_data
.sd_ctime
.sec
= get_be32(&ondisk
->ctime
.sec
);
1445 ce
->ce_stat_data
.sd_mtime
.sec
= get_be32(&ondisk
->mtime
.sec
);
1446 ce
->ce_stat_data
.sd_ctime
.nsec
= get_be32(&ondisk
->ctime
.nsec
);
1447 ce
->ce_stat_data
.sd_mtime
.nsec
= get_be32(&ondisk
->mtime
.nsec
);
1448 ce
->ce_stat_data
.sd_dev
= get_be32(&ondisk
->dev
);
1449 ce
->ce_stat_data
.sd_ino
= get_be32(&ondisk
->ino
);
1450 ce
->ce_mode
= get_be32(&ondisk
->mode
);
1451 ce
->ce_stat_data
.sd_uid
= get_be32(&ondisk
->uid
);
1452 ce
->ce_stat_data
.sd_gid
= get_be32(&ondisk
->gid
);
1453 ce
->ce_stat_data
.sd_size
= get_be32(&ondisk
->size
);
1454 ce
->ce_flags
= flags
& ~CE_NAMEMASK
;
1455 ce
->ce_namelen
= len
;
1457 hashcpy(ce
->sha1
, ondisk
->sha1
);
1458 memcpy(ce
->name
, name
, len
);
1459 ce
->name
[len
] = '\0';
1464 * Adjacent cache entries tend to share the leading paths, so it makes
1465 * sense to only store the differences in later entries. In the v4
1466 * on-disk format of the index, each on-disk cache entry stores the
1467 * number of bytes to be stripped from the end of the previous name,
1468 * and the bytes to append to the result, to come up with its name.
1470 static unsigned long expand_name_field(struct strbuf
*name
, const char *cp_
)
1472 const unsigned char *ep
, *cp
= (const unsigned char *)cp_
;
1473 size_t len
= decode_varint(&cp
);
1475 if (name
->len
< len
)
1476 die("malformed name field in the index");
1477 strbuf_remove(name
, name
->len
- len
, len
);
1478 for (ep
= cp
; *ep
; ep
++)
1479 ; /* find the end */
1480 strbuf_add(name
, cp
, ep
- cp
);
1481 return (const char *)ep
+ 1 - cp_
;
1484 static struct cache_entry
*create_from_disk(struct ondisk_cache_entry
*ondisk
,
1485 unsigned long *ent_size
,
1486 struct strbuf
*previous_name
)
1488 struct cache_entry
*ce
;
1493 /* On-disk flags are just 16 bits */
1494 flags
= get_be16(&ondisk
->flags
);
1495 len
= flags
& CE_NAMEMASK
;
1497 if (flags
& CE_EXTENDED
) {
1498 struct ondisk_cache_entry_extended
*ondisk2
;
1500 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
1501 extended_flags
= get_be16(&ondisk2
->flags2
) << 16;
1502 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1503 if (extended_flags
& ~CE_EXTENDED_FLAGS
)
1504 die("Unknown index entry format %08x", extended_flags
);
1505 flags
|= extended_flags
;
1506 name
= ondisk2
->name
;
1509 name
= ondisk
->name
;
1511 if (!previous_name
) {
1512 /* v3 and earlier */
1513 if (len
== CE_NAMEMASK
)
1515 ce
= cache_entry_from_ondisk(ondisk
, flags
, name
, len
);
1517 *ent_size
= ondisk_ce_size(ce
);
1519 unsigned long consumed
;
1520 consumed
= expand_name_field(previous_name
, name
);
1521 ce
= cache_entry_from_ondisk(ondisk
, flags
,
1523 previous_name
->len
);
1525 *ent_size
= (name
- ((char *)ondisk
)) + consumed
;
1530 static void check_ce_order(struct index_state
*istate
)
1534 for (i
= 1; i
< istate
->cache_nr
; i
++) {
1535 struct cache_entry
*ce
= istate
->cache
[i
- 1];
1536 struct cache_entry
*next_ce
= istate
->cache
[i
];
1537 int name_compare
= strcmp(ce
->name
, next_ce
->name
);
1539 if (0 < name_compare
)
1540 die("unordered stage entries in index");
1541 if (!name_compare
) {
1543 die("multiple stage entries for merged file '%s'",
1545 if (ce_stage(ce
) > ce_stage(next_ce
))
1546 die("unordered stage entries for '%s'",
1552 static void tweak_untracked_cache(struct index_state
*istate
)
1554 switch (git_config_get_untracked_cache()) {
1555 case -1: /* keep: do nothing */
1558 remove_untracked_cache(istate
);
1561 add_untracked_cache(istate
);
1563 default: /* unknown value: do nothing */
1568 static void post_read_index_from(struct index_state
*istate
)
1570 check_ce_order(istate
);
1571 tweak_untracked_cache(istate
);
1574 /* remember to discard_cache() before reading a different cache! */
1575 int do_read_index(struct index_state
*istate
, const char *path
, int must_exist
)
1579 unsigned long src_offset
;
1580 struct cache_header
*hdr
;
1583 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
1585 if (istate
->initialized
)
1586 return istate
->cache_nr
;
1588 istate
->timestamp
.sec
= 0;
1589 istate
->timestamp
.nsec
= 0;
1590 fd
= open(path
, O_RDONLY
);
1592 if (!must_exist
&& errno
== ENOENT
)
1594 die_errno("%s: index file open failed", path
);
1598 die_errno("cannot stat the open index");
1600 mmap_size
= xsize_t(st
.st_size
);
1601 if (mmap_size
< sizeof(struct cache_header
) + 20)
1602 die("index file smaller than expected");
1604 mmap
= xmmap(NULL
, mmap_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1605 if (mmap
== MAP_FAILED
)
1606 die_errno("unable to map index file");
1610 if (verify_hdr(hdr
, mmap_size
) < 0)
1613 hashcpy(istate
->sha1
, (const unsigned char *)hdr
+ mmap_size
- 20);
1614 istate
->version
= ntohl(hdr
->hdr_version
);
1615 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
1616 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
1617 istate
->cache
= xcalloc(istate
->cache_alloc
, sizeof(*istate
->cache
));
1618 istate
->initialized
= 1;
1620 if (istate
->version
== 4)
1621 previous_name
= &previous_name_buf
;
1623 previous_name
= NULL
;
1625 src_offset
= sizeof(*hdr
);
1626 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1627 struct ondisk_cache_entry
*disk_ce
;
1628 struct cache_entry
*ce
;
1629 unsigned long consumed
;
1631 disk_ce
= (struct ondisk_cache_entry
*)((char *)mmap
+ src_offset
);
1632 ce
= create_from_disk(disk_ce
, &consumed
, previous_name
);
1633 set_index_entry(istate
, i
, ce
);
1635 src_offset
+= consumed
;
1637 strbuf_release(&previous_name_buf
);
1638 istate
->timestamp
.sec
= st
.st_mtime
;
1639 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
1641 while (src_offset
<= mmap_size
- 20 - 8) {
1642 /* After an array of active_nr index entries,
1643 * there can be arbitrary number of extended
1644 * sections, each of which is prefixed with
1645 * extension name (4-byte) and section length
1646 * in 4-byte network byte order.
1649 memcpy(&extsize
, (char *)mmap
+ src_offset
+ 4, 4);
1650 extsize
= ntohl(extsize
);
1651 if (read_index_extension(istate
,
1652 (const char *) mmap
+ src_offset
,
1653 (char *) mmap
+ src_offset
+ 8,
1657 src_offset
+= extsize
;
1659 munmap(mmap
, mmap_size
);
1660 return istate
->cache_nr
;
1663 munmap(mmap
, mmap_size
);
1664 die("index file corrupt");
1667 int read_index_from(struct index_state
*istate
, const char *path
)
1669 struct split_index
*split_index
;
1672 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1673 if (istate
->initialized
)
1674 return istate
->cache_nr
;
1676 ret
= do_read_index(istate
, path
, 0);
1678 split_index
= istate
->split_index
;
1679 if (!split_index
|| is_null_sha1(split_index
->base_sha1
)) {
1680 post_read_index_from(istate
);
1684 if (split_index
->base
)
1685 discard_index(split_index
->base
);
1687 split_index
->base
= xcalloc(1, sizeof(*split_index
->base
));
1688 ret
= do_read_index(split_index
->base
,
1689 git_path("sharedindex.%s",
1690 sha1_to_hex(split_index
->base_sha1
)), 1);
1691 if (hashcmp(split_index
->base_sha1
, split_index
->base
->sha1
))
1692 die("broken index, expect %s in %s, got %s",
1693 sha1_to_hex(split_index
->base_sha1
),
1694 git_path("sharedindex.%s",
1695 sha1_to_hex(split_index
->base_sha1
)),
1696 sha1_to_hex(split_index
->base
->sha1
));
1697 merge_base_index(istate
);
1698 post_read_index_from(istate
);
1702 int is_index_unborn(struct index_state
*istate
)
1704 return (!istate
->cache_nr
&& !istate
->timestamp
.sec
);
1707 int discard_index(struct index_state
*istate
)
1711 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1712 if (istate
->cache
[i
]->index
&&
1713 istate
->split_index
&&
1714 istate
->split_index
->base
&&
1715 istate
->cache
[i
]->index
<= istate
->split_index
->base
->cache_nr
&&
1716 istate
->cache
[i
] == istate
->split_index
->base
->cache
[istate
->cache
[i
]->index
- 1])
1718 free(istate
->cache
[i
]);
1720 resolve_undo_clear_index(istate
);
1721 istate
->cache_nr
= 0;
1722 istate
->cache_changed
= 0;
1723 istate
->timestamp
.sec
= 0;
1724 istate
->timestamp
.nsec
= 0;
1725 free_name_hash(istate
);
1726 cache_tree_free(&(istate
->cache_tree
));
1727 istate
->initialized
= 0;
1728 free(istate
->cache
);
1729 istate
->cache
= NULL
;
1730 istate
->cache_alloc
= 0;
1731 discard_split_index(istate
);
1732 free_untracked_cache(istate
->untracked
);
1733 istate
->untracked
= NULL
;
1737 int unmerged_index(const struct index_state
*istate
)
1740 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1741 if (ce_stage(istate
->cache
[i
]))
1747 #define WRITE_BUFFER_SIZE 8192
1748 static unsigned char write_buffer
[WRITE_BUFFER_SIZE
];
1749 static unsigned long write_buffer_len
;
1751 static int ce_write_flush(git_SHA_CTX
*context
, int fd
)
1753 unsigned int buffered
= write_buffer_len
;
1755 git_SHA1_Update(context
, write_buffer
, buffered
);
1756 if (write_in_full(fd
, write_buffer
, buffered
) != buffered
)
1758 write_buffer_len
= 0;
1763 static int ce_write(git_SHA_CTX
*context
, int fd
, void *data
, unsigned int len
)
1766 unsigned int buffered
= write_buffer_len
;
1767 unsigned int partial
= WRITE_BUFFER_SIZE
- buffered
;
1770 memcpy(write_buffer
+ buffered
, data
, partial
);
1771 buffered
+= partial
;
1772 if (buffered
== WRITE_BUFFER_SIZE
) {
1773 write_buffer_len
= buffered
;
1774 if (ce_write_flush(context
, fd
))
1778 write_buffer_len
= buffered
;
1780 data
= (char *) data
+ partial
;
1785 static int write_index_ext_header(git_SHA_CTX
*context
, int fd
,
1786 unsigned int ext
, unsigned int sz
)
1790 return ((ce_write(context
, fd
, &ext
, 4) < 0) ||
1791 (ce_write(context
, fd
, &sz
, 4) < 0)) ? -1 : 0;
1794 static int ce_flush(git_SHA_CTX
*context
, int fd
, unsigned char *sha1
)
1796 unsigned int left
= write_buffer_len
;
1799 write_buffer_len
= 0;
1800 git_SHA1_Update(context
, write_buffer
, left
);
1803 /* Flush first if not enough space for SHA1 signature */
1804 if (left
+ 20 > WRITE_BUFFER_SIZE
) {
1805 if (write_in_full(fd
, write_buffer
, left
) != left
)
1810 /* Append the SHA1 signature at the end */
1811 git_SHA1_Final(write_buffer
+ left
, context
);
1812 hashcpy(sha1
, write_buffer
+ left
);
1814 return (write_in_full(fd
, write_buffer
, left
) != left
) ? -1 : 0;
1817 static void ce_smudge_racily_clean_entry(struct cache_entry
*ce
)
1820 * The only thing we care about in this function is to smudge the
1821 * falsely clean entry due to touch-update-touch race, so we leave
1822 * everything else as they are. We are called for entries whose
1823 * ce_stat_data.sd_mtime match the index file mtime.
1825 * Note that this actually does not do much for gitlinks, for
1826 * which ce_match_stat_basic() always goes to the actual
1827 * contents. The caller checks with is_racy_timestamp() which
1828 * always says "no" for gitlinks, so we are not called for them ;-)
1832 if (lstat(ce
->name
, &st
) < 0)
1834 if (ce_match_stat_basic(ce
, &st
))
1836 if (ce_modified_check_fs(ce
, &st
)) {
1837 /* This is "racily clean"; smudge it. Note that this
1838 * is a tricky code. At first glance, it may appear
1839 * that it can break with this sequence:
1841 * $ echo xyzzy >frotz
1842 * $ git-update-index --add frotz
1845 * $ echo filfre >nitfol
1846 * $ git-update-index --add nitfol
1848 * but it does not. When the second update-index runs,
1849 * it notices that the entry "frotz" has the same timestamp
1850 * as index, and if we were to smudge it by resetting its
1851 * size to zero here, then the object name recorded
1852 * in index is the 6-byte file but the cached stat information
1853 * becomes zero --- which would then match what we would
1854 * obtain from the filesystem next time we stat("frotz").
1856 * However, the second update-index, before calling
1857 * this function, notices that the cached size is 6
1858 * bytes and what is on the filesystem is an empty
1859 * file, and never calls us, so the cached size information
1860 * for "frotz" stays 6 which does not match the filesystem.
1862 ce
->ce_stat_data
.sd_size
= 0;
1866 /* Copy miscellaneous fields but not the name */
1867 static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry
*ondisk
,
1868 struct cache_entry
*ce
)
1872 ondisk
->ctime
.sec
= htonl(ce
->ce_stat_data
.sd_ctime
.sec
);
1873 ondisk
->mtime
.sec
= htonl(ce
->ce_stat_data
.sd_mtime
.sec
);
1874 ondisk
->ctime
.nsec
= htonl(ce
->ce_stat_data
.sd_ctime
.nsec
);
1875 ondisk
->mtime
.nsec
= htonl(ce
->ce_stat_data
.sd_mtime
.nsec
);
1876 ondisk
->dev
= htonl(ce
->ce_stat_data
.sd_dev
);
1877 ondisk
->ino
= htonl(ce
->ce_stat_data
.sd_ino
);
1878 ondisk
->mode
= htonl(ce
->ce_mode
);
1879 ondisk
->uid
= htonl(ce
->ce_stat_data
.sd_uid
);
1880 ondisk
->gid
= htonl(ce
->ce_stat_data
.sd_gid
);
1881 ondisk
->size
= htonl(ce
->ce_stat_data
.sd_size
);
1882 hashcpy(ondisk
->sha1
, ce
->sha1
);
1884 flags
= ce
->ce_flags
& ~CE_NAMEMASK
;
1885 flags
|= (ce_namelen(ce
) >= CE_NAMEMASK
? CE_NAMEMASK
: ce_namelen(ce
));
1886 ondisk
->flags
= htons(flags
);
1887 if (ce
->ce_flags
& CE_EXTENDED
) {
1888 struct ondisk_cache_entry_extended
*ondisk2
;
1889 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
1890 ondisk2
->flags2
= htons((ce
->ce_flags
& CE_EXTENDED_FLAGS
) >> 16);
1891 return ondisk2
->name
;
1894 return ondisk
->name
;
1898 static int ce_write_entry(git_SHA_CTX
*c
, int fd
, struct cache_entry
*ce
,
1899 struct strbuf
*previous_name
)
1902 struct ondisk_cache_entry
*ondisk
;
1903 int saved_namelen
= saved_namelen
; /* compiler workaround */
1907 if (ce
->ce_flags
& CE_STRIP_NAME
) {
1908 saved_namelen
= ce_namelen(ce
);
1912 if (!previous_name
) {
1913 size
= ondisk_ce_size(ce
);
1914 ondisk
= xcalloc(1, size
);
1915 name
= copy_cache_entry_to_ondisk(ondisk
, ce
);
1916 memcpy(name
, ce
->name
, ce_namelen(ce
));
1918 int common
, to_remove
, prefix_size
;
1919 unsigned char to_remove_vi
[16];
1921 (ce
->name
[common
] &&
1922 common
< previous_name
->len
&&
1923 ce
->name
[common
] == previous_name
->buf
[common
]);
1925 ; /* still matching */
1926 to_remove
= previous_name
->len
- common
;
1927 prefix_size
= encode_varint(to_remove
, to_remove_vi
);
1929 if (ce
->ce_flags
& CE_EXTENDED
)
1930 size
= offsetof(struct ondisk_cache_entry_extended
, name
);
1932 size
= offsetof(struct ondisk_cache_entry
, name
);
1933 size
+= prefix_size
+ (ce_namelen(ce
) - common
+ 1);
1935 ondisk
= xcalloc(1, size
);
1936 name
= copy_cache_entry_to_ondisk(ondisk
, ce
);
1937 memcpy(name
, to_remove_vi
, prefix_size
);
1938 memcpy(name
+ prefix_size
, ce
->name
+ common
, ce_namelen(ce
) - common
);
1940 strbuf_splice(previous_name
, common
, to_remove
,
1941 ce
->name
+ common
, ce_namelen(ce
) - common
);
1943 if (ce
->ce_flags
& CE_STRIP_NAME
) {
1944 ce
->ce_namelen
= saved_namelen
;
1945 ce
->ce_flags
&= ~CE_STRIP_NAME
;
1948 result
= ce_write(c
, fd
, ondisk
, size
);
1954 * This function verifies if index_state has the correct sha1 of the
1955 * index file. Don't die if we have any other failure, just return 0.
1957 static int verify_index_from(const struct index_state
*istate
, const char *path
)
1962 unsigned char sha1
[20];
1964 if (!istate
->initialized
)
1967 fd
= open(path
, O_RDONLY
);
1974 if (st
.st_size
< sizeof(struct cache_header
) + 20)
1977 n
= pread_in_full(fd
, sha1
, 20, st
.st_size
- 20);
1981 if (hashcmp(istate
->sha1
, sha1
))
1992 static int verify_index(const struct index_state
*istate
)
1994 return verify_index_from(istate
, get_index_file());
1997 static int has_racy_timestamp(struct index_state
*istate
)
1999 int entries
= istate
->cache_nr
;
2002 for (i
= 0; i
< entries
; i
++) {
2003 struct cache_entry
*ce
= istate
->cache
[i
];
2004 if (is_racy_timestamp(istate
, ce
))
2011 * Opportunistically update the index but do not complain if we can't
2013 void update_index_if_able(struct index_state
*istate
, struct lock_file
*lockfile
)
2015 if ((istate
->cache_changed
|| has_racy_timestamp(istate
)) &&
2016 verify_index(istate
) &&
2017 write_locked_index(istate
, lockfile
, COMMIT_LOCK
))
2018 rollback_lock_file(lockfile
);
2021 static int do_write_index(struct index_state
*istate
, int newfd
,
2022 int strip_extensions
)
2025 struct cache_header hdr
;
2026 int i
, err
, removed
, extended
, hdr_version
;
2027 struct cache_entry
**cache
= istate
->cache
;
2028 int entries
= istate
->cache_nr
;
2030 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
2032 for (i
= removed
= extended
= 0; i
< entries
; i
++) {
2033 if (cache
[i
]->ce_flags
& CE_REMOVE
)
2036 /* reduce extended entries if possible */
2037 cache
[i
]->ce_flags
&= ~CE_EXTENDED
;
2038 if (cache
[i
]->ce_flags
& CE_EXTENDED_FLAGS
) {
2040 cache
[i
]->ce_flags
|= CE_EXTENDED
;
2044 if (!istate
->version
) {
2045 istate
->version
= get_index_format_default();
2046 if (getenv("GIT_TEST_SPLIT_INDEX"))
2047 init_split_index(istate
);
2050 /* demote version 3 to version 2 when the latter suffices */
2051 if (istate
->version
== 3 || istate
->version
== 2)
2052 istate
->version
= extended
? 3 : 2;
2054 hdr_version
= istate
->version
;
2056 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
2057 hdr
.hdr_version
= htonl(hdr_version
);
2058 hdr
.hdr_entries
= htonl(entries
- removed
);
2061 if (ce_write(&c
, newfd
, &hdr
, sizeof(hdr
)) < 0)
2064 previous_name
= (hdr_version
== 4) ? &previous_name_buf
: NULL
;
2065 for (i
= 0; i
< entries
; i
++) {
2066 struct cache_entry
*ce
= cache
[i
];
2067 if (ce
->ce_flags
& CE_REMOVE
)
2069 if (!ce_uptodate(ce
) && is_racy_timestamp(istate
, ce
))
2070 ce_smudge_racily_clean_entry(ce
);
2071 if (is_null_sha1(ce
->sha1
)) {
2072 static const char msg
[] = "cache entry has null sha1: %s";
2073 static int allow
= -1;
2076 allow
= git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2078 warning(msg
, ce
->name
);
2080 return error(msg
, ce
->name
);
2082 if (ce_write_entry(&c
, newfd
, ce
, previous_name
) < 0)
2085 strbuf_release(&previous_name_buf
);
2087 /* Write extension data here */
2088 if (!strip_extensions
&& istate
->split_index
) {
2089 struct strbuf sb
= STRBUF_INIT
;
2091 err
= write_link_extension(&sb
, istate
) < 0 ||
2092 write_index_ext_header(&c
, newfd
, CACHE_EXT_LINK
,
2094 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2095 strbuf_release(&sb
);
2099 if (!strip_extensions
&& istate
->cache_tree
) {
2100 struct strbuf sb
= STRBUF_INIT
;
2102 cache_tree_write(&sb
, istate
->cache_tree
);
2103 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_TREE
, sb
.len
) < 0
2104 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2105 strbuf_release(&sb
);
2109 if (!strip_extensions
&& istate
->resolve_undo
) {
2110 struct strbuf sb
= STRBUF_INIT
;
2112 resolve_undo_write(&sb
, istate
->resolve_undo
);
2113 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_RESOLVE_UNDO
,
2115 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2116 strbuf_release(&sb
);
2120 if (!strip_extensions
&& istate
->untracked
) {
2121 struct strbuf sb
= STRBUF_INIT
;
2123 write_untracked_extension(&sb
, istate
->untracked
);
2124 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_UNTRACKED
,
2126 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2127 strbuf_release(&sb
);
2132 if (ce_flush(&c
, newfd
, istate
->sha1
) || fstat(newfd
, &st
))
2134 istate
->timestamp
.sec
= (unsigned int)st
.st_mtime
;
2135 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
2139 void set_alternate_index_output(const char *name
)
2141 alternate_index_output
= name
;
2144 static int commit_locked_index(struct lock_file
*lk
)
2146 if (alternate_index_output
)
2147 return commit_lock_file_to(lk
, alternate_index_output
);
2149 return commit_lock_file(lk
);
2152 static int do_write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
2155 int ret
= do_write_index(istate
, get_lock_file_fd(lock
), 0);
2158 assert((flags
& (COMMIT_LOCK
| CLOSE_LOCK
)) !=
2159 (COMMIT_LOCK
| CLOSE_LOCK
));
2160 if (flags
& COMMIT_LOCK
)
2161 return commit_locked_index(lock
);
2162 else if (flags
& CLOSE_LOCK
)
2163 return close_lock_file(lock
);
2168 static int write_split_index(struct index_state
*istate
,
2169 struct lock_file
*lock
,
2173 prepare_to_write_split_index(istate
);
2174 ret
= do_write_locked_index(istate
, lock
, flags
);
2175 finish_writing_split_index(istate
);
2179 static struct tempfile temporary_sharedindex
;
2181 static int write_shared_index(struct index_state
*istate
,
2182 struct lock_file
*lock
, unsigned flags
)
2184 struct split_index
*si
= istate
->split_index
;
2187 fd
= mks_tempfile(&temporary_sharedindex
, git_path("sharedindex_XXXXXX"));
2189 hashclr(si
->base_sha1
);
2190 return do_write_locked_index(istate
, lock
, flags
);
2192 move_cache_to_base_index(istate
);
2193 ret
= do_write_index(si
->base
, fd
, 1);
2195 delete_tempfile(&temporary_sharedindex
);
2198 ret
= rename_tempfile(&temporary_sharedindex
,
2199 git_path("sharedindex.%s", sha1_to_hex(si
->base
->sha1
)));
2201 hashcpy(si
->base_sha1
, si
->base
->sha1
);
2205 int write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
2208 struct split_index
*si
= istate
->split_index
;
2210 if (!si
|| alternate_index_output
||
2211 (istate
->cache_changed
& ~EXTMASK
)) {
2213 hashclr(si
->base_sha1
);
2214 return do_write_locked_index(istate
, lock
, flags
);
2217 if (getenv("GIT_TEST_SPLIT_INDEX")) {
2218 int v
= si
->base_sha1
[0];
2220 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
2222 if (istate
->cache_changed
& SPLIT_INDEX_ORDERED
) {
2223 int ret
= write_shared_index(istate
, lock
, flags
);
2228 return write_split_index(istate
, lock
, flags
);
2232 * Read the index file that is potentially unmerged into given
2233 * index_state, dropping any unmerged entries. Returns true if
2234 * the index is unmerged. Callers who want to refuse to work
2235 * from an unmerged state can call this and check its return value,
2236 * instead of calling read_cache().
2238 int read_index_unmerged(struct index_state
*istate
)
2244 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2245 struct cache_entry
*ce
= istate
->cache
[i
];
2246 struct cache_entry
*new_ce
;
2252 len
= ce_namelen(ce
);
2253 size
= cache_entry_size(len
);
2254 new_ce
= xcalloc(1, size
);
2255 memcpy(new_ce
->name
, ce
->name
, len
);
2256 new_ce
->ce_flags
= create_ce_flags(0) | CE_CONFLICTED
;
2257 new_ce
->ce_namelen
= len
;
2258 new_ce
->ce_mode
= ce
->ce_mode
;
2259 if (add_index_entry(istate
, new_ce
, 0))
2260 return error("%s: cannot drop to stage #0",
2267 * Returns 1 if the path is an "other" path with respect to
2268 * the index; that is, the path is not mentioned in the index at all,
2269 * either as a file, a directory with some files in the index,
2270 * or as an unmerged entry.
2272 * We helpfully remove a trailing "/" from directories so that
2273 * the output of read_directory can be used as-is.
2275 int index_name_is_other(const struct index_state
*istate
, const char *name
,
2279 if (namelen
&& name
[namelen
- 1] == '/')
2281 pos
= index_name_pos(istate
, name
, namelen
);
2283 return 0; /* exact match */
2285 if (pos
< istate
->cache_nr
) {
2286 struct cache_entry
*ce
= istate
->cache
[pos
];
2287 if (ce_namelen(ce
) == namelen
&&
2288 !memcmp(ce
->name
, name
, namelen
))
2289 return 0; /* Yup, this one exists unmerged */
2294 void *read_blob_data_from_index(struct index_state
*istate
, const char *path
, unsigned long *size
)
2298 enum object_type type
;
2302 pos
= index_name_pos(istate
, path
, len
);
2305 * We might be in the middle of a merge, in which
2306 * case we would read stage #2 (ours).
2310 (pos
< 0 && i
< istate
->cache_nr
&&
2311 !strcmp(istate
->cache
[i
]->name
, path
));
2313 if (ce_stage(istate
->cache
[i
]) == 2)
2318 data
= read_sha1_file(istate
->cache
[pos
]->sha1
, &type
, &sz
);
2319 if (!data
|| type
!= OBJ_BLOB
) {
2328 void stat_validity_clear(struct stat_validity
*sv
)
2334 int stat_validity_check(struct stat_validity
*sv
, const char *path
)
2338 if (stat(path
, &st
) < 0)
2339 return sv
->sd
== NULL
;
2342 return S_ISREG(st
.st_mode
) && !match_stat_data(sv
->sd
, &st
);
2345 void stat_validity_update(struct stat_validity
*sv
, int fd
)
2349 if (fstat(fd
, &st
) < 0 || !S_ISREG(st
.st_mode
))
2350 stat_validity_clear(sv
);
2353 sv
->sd
= xcalloc(1, sizeof(struct stat_data
));
2354 fill_stat_data(sv
->sd
, &st
);