2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 #define NO_THE_INDEX_COMPATIBILITY_MACROS
13 #include "cache-tree.h"
19 #include "resolve-undo.h"
22 #include "split-index.h"
24 #include "fsmonitor.h"
26 /* Mask for the name length in ce_flags in the on-disk index */
28 #define CE_NAMEMASK (0x0fff)
32 * The first letter should be 'A'..'Z' for extensions that are not
33 * necessary for a correct operation (i.e. optimization data).
34 * When new extensions are added that _needs_ to be understood in
35 * order to correctly interpret the index file, pick character that
36 * is outside the range, to cause the reader to abort.
39 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
40 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
41 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
42 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
43 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
44 #define CACHE_EXT_FSMONITOR 0x46534D4E /* "FSMN" */
46 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
47 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
48 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
49 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED | FSMONITOR_CHANGED)
51 struct index_state the_index
;
52 static const char *alternate_index_output
;
54 static void set_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
56 istate
->cache
[nr
] = ce
;
57 add_name_hash(istate
, ce
);
60 static void replace_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
62 struct cache_entry
*old
= istate
->cache
[nr
];
64 replace_index_entry_in_base(istate
, old
, ce
);
65 remove_name_hash(istate
, old
);
67 ce
->ce_flags
&= ~CE_HASHED
;
68 set_index_entry(istate
, nr
, ce
);
69 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
70 mark_fsmonitor_invalid(istate
, ce
);
71 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
74 void rename_index_entry_at(struct index_state
*istate
, int nr
, const char *new_name
)
76 struct cache_entry
*old_entry
= istate
->cache
[nr
], *new_entry
;
77 int namelen
= strlen(new_name
);
79 new_entry
= xmalloc(cache_entry_size(namelen
));
80 copy_cache_entry(new_entry
, old_entry
);
81 new_entry
->ce_flags
&= ~CE_HASHED
;
82 new_entry
->ce_namelen
= namelen
;
84 memcpy(new_entry
->name
, new_name
, namelen
+ 1);
86 cache_tree_invalidate_path(istate
, old_entry
->name
);
87 untracked_cache_remove_from_index(istate
, old_entry
->name
);
88 remove_index_entry_at(istate
, nr
);
89 add_index_entry(istate
, new_entry
, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
92 void fill_stat_data(struct stat_data
*sd
, struct stat
*st
)
94 sd
->sd_ctime
.sec
= (unsigned int)st
->st_ctime
;
95 sd
->sd_mtime
.sec
= (unsigned int)st
->st_mtime
;
96 sd
->sd_ctime
.nsec
= ST_CTIME_NSEC(*st
);
97 sd
->sd_mtime
.nsec
= ST_MTIME_NSEC(*st
);
98 sd
->sd_dev
= st
->st_dev
;
99 sd
->sd_ino
= st
->st_ino
;
100 sd
->sd_uid
= st
->st_uid
;
101 sd
->sd_gid
= st
->st_gid
;
102 sd
->sd_size
= st
->st_size
;
105 int match_stat_data(const struct stat_data
*sd
, struct stat
*st
)
109 if (sd
->sd_mtime
.sec
!= (unsigned int)st
->st_mtime
)
110 changed
|= MTIME_CHANGED
;
111 if (trust_ctime
&& check_stat
&&
112 sd
->sd_ctime
.sec
!= (unsigned int)st
->st_ctime
)
113 changed
|= CTIME_CHANGED
;
116 if (check_stat
&& sd
->sd_mtime
.nsec
!= ST_MTIME_NSEC(*st
))
117 changed
|= MTIME_CHANGED
;
118 if (trust_ctime
&& check_stat
&&
119 sd
->sd_ctime
.nsec
!= ST_CTIME_NSEC(*st
))
120 changed
|= CTIME_CHANGED
;
124 if (sd
->sd_uid
!= (unsigned int) st
->st_uid
||
125 sd
->sd_gid
!= (unsigned int) st
->st_gid
)
126 changed
|= OWNER_CHANGED
;
127 if (sd
->sd_ino
!= (unsigned int) st
->st_ino
)
128 changed
|= INODE_CHANGED
;
133 * st_dev breaks on network filesystems where different
134 * clients will have different views of what "device"
135 * the filesystem is on
137 if (check_stat
&& sd
->sd_dev
!= (unsigned int) st
->st_dev
)
138 changed
|= INODE_CHANGED
;
141 if (sd
->sd_size
!= (unsigned int) st
->st_size
)
142 changed
|= DATA_CHANGED
;
148 * This only updates the "non-critical" parts of the directory
149 * cache, ie the parts that aren't tracked by GIT, and only used
150 * to validate the cache.
152 void fill_stat_cache_info(struct cache_entry
*ce
, struct stat
*st
)
154 fill_stat_data(&ce
->ce_stat_data
, st
);
156 if (assume_unchanged
)
157 ce
->ce_flags
|= CE_VALID
;
159 if (S_ISREG(st
->st_mode
)) {
160 ce_mark_uptodate(ce
);
161 mark_fsmonitor_valid(ce
);
165 static int ce_compare_data(const struct cache_entry
*ce
, struct stat
*st
)
168 int fd
= git_open_cloexec(ce
->name
, O_RDONLY
);
171 struct object_id oid
;
172 if (!index_fd(&oid
, fd
, st
, OBJ_BLOB
, ce
->name
, 0))
173 match
= oidcmp(&oid
, &ce
->oid
);
174 /* index_fd() closed the file descriptor already */
179 static int ce_compare_link(const struct cache_entry
*ce
, size_t expected_size
)
184 enum object_type type
;
185 struct strbuf sb
= STRBUF_INIT
;
187 if (strbuf_readlink(&sb
, ce
->name
, expected_size
))
190 buffer
= read_sha1_file(ce
->oid
.hash
, &type
, &size
);
193 match
= memcmp(buffer
, sb
.buf
, size
);
200 static int ce_compare_gitlink(const struct cache_entry
*ce
)
202 struct object_id oid
;
205 * We don't actually require that the .git directory
206 * under GITLINK directory be a valid git directory. It
207 * might even be missing (in case nobody populated that
210 * If so, we consider it always to match.
212 if (resolve_gitlink_ref(ce
->name
, "HEAD", &oid
) < 0)
214 return oidcmp(&oid
, &ce
->oid
);
217 static int ce_modified_check_fs(const struct cache_entry
*ce
, struct stat
*st
)
219 switch (st
->st_mode
& S_IFMT
) {
221 if (ce_compare_data(ce
, st
))
225 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
229 if (S_ISGITLINK(ce
->ce_mode
))
230 return ce_compare_gitlink(ce
) ? DATA_CHANGED
: 0;
231 /* else fallthrough */
238 static int ce_match_stat_basic(const struct cache_entry
*ce
, struct stat
*st
)
240 unsigned int changed
= 0;
242 if (ce
->ce_flags
& CE_REMOVE
)
243 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
245 switch (ce
->ce_mode
& S_IFMT
) {
247 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
248 /* We consider only the owner x bit to be relevant for
251 if (trust_executable_bit
&&
252 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
253 changed
|= MODE_CHANGED
;
256 if (!S_ISLNK(st
->st_mode
) &&
257 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
258 changed
|= TYPE_CHANGED
;
261 /* We ignore most of the st_xxx fields for gitlinks */
262 if (!S_ISDIR(st
->st_mode
))
263 changed
|= TYPE_CHANGED
;
264 else if (ce_compare_gitlink(ce
))
265 changed
|= DATA_CHANGED
;
268 die("internal error: ce_mode is %o", ce
->ce_mode
);
271 changed
|= match_stat_data(&ce
->ce_stat_data
, st
);
273 /* Racily smudged entry? */
274 if (!ce
->ce_stat_data
.sd_size
) {
275 if (!is_empty_blob_sha1(ce
->oid
.hash
))
276 changed
|= DATA_CHANGED
;
282 static int is_racy_stat(const struct index_state
*istate
,
283 const struct stat_data
*sd
)
285 return (istate
->timestamp
.sec
&&
287 /* nanosecond timestamped files can also be racy! */
288 (istate
->timestamp
.sec
< sd
->sd_mtime
.sec
||
289 (istate
->timestamp
.sec
== sd
->sd_mtime
.sec
&&
290 istate
->timestamp
.nsec
<= sd
->sd_mtime
.nsec
))
292 istate
->timestamp
.sec
<= sd
->sd_mtime
.sec
297 static int is_racy_timestamp(const struct index_state
*istate
,
298 const struct cache_entry
*ce
)
300 return (!S_ISGITLINK(ce
->ce_mode
) &&
301 is_racy_stat(istate
, &ce
->ce_stat_data
));
304 int match_stat_data_racy(const struct index_state
*istate
,
305 const struct stat_data
*sd
, struct stat
*st
)
307 if (is_racy_stat(istate
, sd
))
308 return MTIME_CHANGED
;
309 return match_stat_data(sd
, st
);
312 int ie_match_stat(struct index_state
*istate
,
313 const struct cache_entry
*ce
, struct stat
*st
,
314 unsigned int options
)
316 unsigned int changed
;
317 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
318 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
319 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
320 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
322 if (!ignore_fsmonitor
)
323 refresh_fsmonitor(istate
);
325 * If it's marked as always valid in the index, it's
326 * valid whatever the checked-out copy says.
328 * skip-worktree has the same effect with higher precedence
330 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
))
332 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
334 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
))
338 * Intent-to-add entries have not been added, so the index entry
339 * by definition never matches what is in the work tree until it
340 * actually gets added.
342 if (ce_intent_to_add(ce
))
343 return DATA_CHANGED
| TYPE_CHANGED
| MODE_CHANGED
;
345 changed
= ce_match_stat_basic(ce
, st
);
348 * Within 1 second of this sequence:
349 * echo xyzzy >file && git-update-index --add file
350 * running this command:
352 * would give a falsely clean cache entry. The mtime and
353 * length match the cache, and other stat fields do not change.
355 * We could detect this at update-index time (the cache entry
356 * being registered/updated records the same time as "now")
357 * and delay the return from git-update-index, but that would
358 * effectively mean we can make at most one commit per second,
359 * which is not acceptable. Instead, we check cache entries
360 * whose mtime are the same as the index file timestamp more
361 * carefully than others.
363 if (!changed
&& is_racy_timestamp(istate
, ce
)) {
364 if (assume_racy_is_modified
)
365 changed
|= DATA_CHANGED
;
367 changed
|= ce_modified_check_fs(ce
, st
);
373 int ie_modified(struct index_state
*istate
,
374 const struct cache_entry
*ce
,
375 struct stat
*st
, unsigned int options
)
377 int changed
, changed_fs
;
379 changed
= ie_match_stat(istate
, ce
, st
, options
);
383 * If the mode or type has changed, there's no point in trying
384 * to refresh the entry - it's not going to match
386 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
390 * Immediately after read-tree or update-index --cacheinfo,
391 * the length field is zero, as we have never even read the
392 * lstat(2) information once, and we cannot trust DATA_CHANGED
393 * returned by ie_match_stat() which in turn was returned by
394 * ce_match_stat_basic() to signal that the filesize of the
395 * blob changed. We have to actually go to the filesystem to
396 * see if the contents match, and if so, should answer "unchanged".
398 * The logic does not apply to gitlinks, as ce_match_stat_basic()
399 * already has checked the actual HEAD from the filesystem in the
400 * subproject. If ie_match_stat() already said it is different,
401 * then we know it is.
403 if ((changed
& DATA_CHANGED
) &&
404 (S_ISGITLINK(ce
->ce_mode
) || ce
->ce_stat_data
.sd_size
!= 0))
407 changed_fs
= ce_modified_check_fs(ce
, st
);
409 return changed
| changed_fs
;
413 int base_name_compare(const char *name1
, int len1
, int mode1
,
414 const char *name2
, int len2
, int mode2
)
416 unsigned char c1
, c2
;
417 int len
= len1
< len2
? len1
: len2
;
420 cmp
= memcmp(name1
, name2
, len
);
425 if (!c1
&& S_ISDIR(mode1
))
427 if (!c2
&& S_ISDIR(mode2
))
429 return (c1
< c2
) ? -1 : (c1
> c2
) ? 1 : 0;
433 * df_name_compare() is identical to base_name_compare(), except it
434 * compares conflicting directory/file entries as equal. Note that
435 * while a directory name compares as equal to a regular file, they
436 * then individually compare _differently_ to a filename that has
437 * a dot after the basename (because '\0' < '.' < '/').
439 * This is used by routines that want to traverse the git namespace
440 * but then handle conflicting entries together when possible.
442 int df_name_compare(const char *name1
, int len1
, int mode1
,
443 const char *name2
, int len2
, int mode2
)
445 int len
= len1
< len2
? len1
: len2
, cmp
;
446 unsigned char c1
, c2
;
448 cmp
= memcmp(name1
, name2
, len
);
451 /* Directories and files compare equal (same length, same name) */
455 if (!c1
&& S_ISDIR(mode1
))
458 if (!c2
&& S_ISDIR(mode2
))
460 if (c1
== '/' && !c2
)
462 if (c2
== '/' && !c1
)
467 int name_compare(const char *name1
, size_t len1
, const char *name2
, size_t len2
)
469 size_t min_len
= (len1
< len2
) ? len1
: len2
;
470 int cmp
= memcmp(name1
, name2
, min_len
);
480 int cache_name_stage_compare(const char *name1
, int len1
, int stage1
, const char *name2
, int len2
, int stage2
)
484 cmp
= name_compare(name1
, len1
, name2
, len2
);
495 static int index_name_stage_pos(const struct index_state
*istate
, const char *name
, int namelen
, int stage
)
500 last
= istate
->cache_nr
;
501 while (last
> first
) {
502 int next
= (last
+ first
) >> 1;
503 struct cache_entry
*ce
= istate
->cache
[next
];
504 int cmp
= cache_name_stage_compare(name
, namelen
, stage
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
516 int index_name_pos(const struct index_state
*istate
, const char *name
, int namelen
)
518 return index_name_stage_pos(istate
, name
, namelen
, 0);
521 int remove_index_entry_at(struct index_state
*istate
, int pos
)
523 struct cache_entry
*ce
= istate
->cache
[pos
];
525 record_resolve_undo(istate
, ce
);
526 remove_name_hash(istate
, ce
);
527 save_or_free_index_entry(istate
, ce
);
528 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
530 if (pos
>= istate
->cache_nr
)
532 MOVE_ARRAY(istate
->cache
+ pos
, istate
->cache
+ pos
+ 1,
533 istate
->cache_nr
- pos
);
538 * Remove all cache entries marked for removal, that is where
539 * CE_REMOVE is set in ce_flags. This is much more effective than
540 * calling remove_index_entry_at() for each entry to be removed.
542 void remove_marked_cache_entries(struct index_state
*istate
)
544 struct cache_entry
**ce_array
= istate
->cache
;
547 for (i
= j
= 0; i
< istate
->cache_nr
; i
++) {
548 if (ce_array
[i
]->ce_flags
& CE_REMOVE
) {
549 remove_name_hash(istate
, ce_array
[i
]);
550 save_or_free_index_entry(istate
, ce_array
[i
]);
553 ce_array
[j
++] = ce_array
[i
];
555 if (j
== istate
->cache_nr
)
557 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
558 istate
->cache_nr
= j
;
561 int remove_file_from_index(struct index_state
*istate
, const char *path
)
563 int pos
= index_name_pos(istate
, path
, strlen(path
));
566 cache_tree_invalidate_path(istate
, path
);
567 untracked_cache_remove_from_index(istate
, path
);
568 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
569 remove_index_entry_at(istate
, pos
);
573 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
575 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
578 static int index_name_pos_also_unmerged(struct index_state
*istate
,
579 const char *path
, int namelen
)
581 int pos
= index_name_pos(istate
, path
, namelen
);
582 struct cache_entry
*ce
;
587 /* maybe unmerged? */
589 if (pos
>= istate
->cache_nr
||
590 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
593 /* order of preference: stage 2, 1, 3 */
594 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
595 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
596 !compare_name(ce
, path
, namelen
))
601 static int different_name(struct cache_entry
*ce
, struct cache_entry
*alias
)
603 int len
= ce_namelen(ce
);
604 return ce_namelen(alias
) != len
|| memcmp(ce
->name
, alias
->name
, len
);
608 * If we add a filename that aliases in the cache, we will use the
609 * name that we already have - but we don't want to update the same
610 * alias twice, because that implies that there were actually two
611 * different files with aliasing names!
613 * So we use the CE_ADDED flag to verify that the alias was an old
614 * one before we accept it as
616 static struct cache_entry
*create_alias_ce(struct index_state
*istate
,
617 struct cache_entry
*ce
,
618 struct cache_entry
*alias
)
621 struct cache_entry
*new_entry
;
623 if (alias
->ce_flags
& CE_ADDED
)
624 die("Will not add file alias '%s' ('%s' already exists in index)", ce
->name
, alias
->name
);
626 /* Ok, create the new entry using the name of the existing alias */
627 len
= ce_namelen(alias
);
628 new_entry
= xcalloc(1, cache_entry_size(len
));
629 memcpy(new_entry
->name
, alias
->name
, len
);
630 copy_cache_entry(new_entry
, ce
);
631 save_or_free_index_entry(istate
, ce
);
635 void set_object_name_for_intent_to_add_entry(struct cache_entry
*ce
)
637 struct object_id oid
;
638 if (write_object_file("", 0, blob_type
, &oid
))
639 die("cannot create an empty blob in the object database");
640 oidcpy(&ce
->oid
, &oid
);
643 int add_to_index(struct index_state
*istate
, const char *path
, struct stat
*st
, int flags
)
645 int size
, namelen
, was_same
;
646 mode_t st_mode
= st
->st_mode
;
647 struct cache_entry
*ce
, *alias
= NULL
;
648 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_IGNORE_SKIP_WORKTREE
|CE_MATCH_RACY_IS_DIRTY
;
649 int verbose
= flags
& (ADD_CACHE_VERBOSE
| ADD_CACHE_PRETEND
);
650 int pretend
= flags
& ADD_CACHE_PRETEND
;
651 int intent_only
= flags
& ADD_CACHE_INTENT
;
652 int add_option
= (ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
|
653 (intent_only
? ADD_CACHE_NEW_ONLY
: 0));
654 int newflags
= HASH_WRITE_OBJECT
;
656 if (flags
& HASH_RENORMALIZE
)
657 newflags
|= HASH_RENORMALIZE
;
659 if (!S_ISREG(st_mode
) && !S_ISLNK(st_mode
) && !S_ISDIR(st_mode
))
660 return error("%s: can only add regular files, symbolic links or git-directories", path
);
662 namelen
= strlen(path
);
663 if (S_ISDIR(st_mode
)) {
664 while (namelen
&& path
[namelen
-1] == '/')
667 size
= cache_entry_size(namelen
);
668 ce
= xcalloc(1, size
);
669 memcpy(ce
->name
, path
, namelen
);
670 ce
->ce_namelen
= namelen
;
672 fill_stat_cache_info(ce
, st
);
674 ce
->ce_flags
|= CE_INTENT_TO_ADD
;
677 if (trust_executable_bit
&& has_symlinks
) {
678 ce
->ce_mode
= create_ce_mode(st_mode
);
680 /* If there is an existing entry, pick the mode bits and type
681 * from it, otherwise assume unexecutable regular file.
683 struct cache_entry
*ent
;
684 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
686 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
687 ce
->ce_mode
= ce_mode_from_stat(ent
, st_mode
);
690 /* When core.ignorecase=true, determine if a directory of the same name but differing
691 * case already exists within the Git repository. If it does, ensure the directory
692 * case of the file being added to the repository matches (is folded into) the existing
693 * entry's directory case.
696 adjust_dirname_case(istate
, ce
->name
);
698 if (!(flags
& HASH_RENORMALIZE
)) {
699 alias
= index_file_exists(istate
, ce
->name
,
700 ce_namelen(ce
), ignore_case
);
703 !ie_match_stat(istate
, alias
, st
, ce_option
)) {
704 /* Nothing changed, really */
705 if (!S_ISGITLINK(alias
->ce_mode
))
706 ce_mark_uptodate(alias
);
707 alias
->ce_flags
|= CE_ADDED
;
714 if (index_path(&ce
->oid
, path
, st
, newflags
)) {
716 return error("unable to index file %s", path
);
719 set_object_name_for_intent_to_add_entry(ce
);
721 if (ignore_case
&& alias
&& different_name(ce
, alias
))
722 ce
= create_alias_ce(istate
, ce
, alias
);
723 ce
->ce_flags
|= CE_ADDED
;
725 /* It was suspected to be racily clean, but it turns out to be Ok */
728 !oidcmp(&alias
->oid
, &ce
->oid
) &&
729 ce
->ce_mode
== alias
->ce_mode
);
733 else if (add_index_entry(istate
, ce
, add_option
)) {
735 return error("unable to add %s to index", path
);
737 if (verbose
&& !was_same
)
738 printf("add '%s'\n", path
);
742 int add_file_to_index(struct index_state
*istate
, const char *path
, int flags
)
745 if (lstat(path
, &st
))
746 die_errno("unable to stat '%s'", path
);
747 return add_to_index(istate
, path
, &st
, flags
);
750 struct cache_entry
*make_cache_entry(unsigned int mode
,
751 const unsigned char *sha1
, const char *path
, int stage
,
752 unsigned int refresh_options
)
755 struct cache_entry
*ce
, *ret
;
757 if (!verify_path(path
, mode
)) {
758 error("Invalid path '%s'", path
);
763 size
= cache_entry_size(len
);
764 ce
= xcalloc(1, size
);
766 hashcpy(ce
->oid
.hash
, sha1
);
767 memcpy(ce
->name
, path
, len
);
768 ce
->ce_flags
= create_ce_flags(stage
);
769 ce
->ce_namelen
= len
;
770 ce
->ce_mode
= create_ce_mode(mode
);
772 ret
= refresh_cache_entry(ce
, refresh_options
);
779 * Chmod an index entry with either +x or -x.
781 * Returns -1 if the chmod for the particular cache entry failed (if it's
782 * not a regular file), -2 if an invalid flip argument is passed in, 0
785 int chmod_index_entry(struct index_state
*istate
, struct cache_entry
*ce
,
788 if (!S_ISREG(ce
->ce_mode
))
795 ce
->ce_mode
&= ~0111;
800 cache_tree_invalidate_path(istate
, ce
->name
);
801 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
802 mark_fsmonitor_invalid(istate
, ce
);
803 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
808 int ce_same_name(const struct cache_entry
*a
, const struct cache_entry
*b
)
810 int len
= ce_namelen(a
);
811 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
815 * We fundamentally don't like some paths: we don't want
816 * dot or dot-dot anywhere, and for obvious reasons don't
817 * want to recurse into ".git" either.
819 * Also, we don't want double slashes or slashes at the
820 * end that can make pathnames ambiguous.
822 static int verify_dotfile(const char *rest
, unsigned mode
)
825 * The first character was '.', but that
826 * has already been discarded, we now test
830 /* "." is not allowed */
831 if (*rest
== '\0' || is_dir_sep(*rest
))
836 * ".git" followed by NUL or slash is bad. Note that we match
837 * case-insensitively here, even if ignore_case is not set.
838 * This outlaws ".GIT" everywhere out of an abundance of caution,
839 * since there's really no good reason to allow it.
841 * Once we've seen ".git", we can also find ".gitmodules", etc (also
842 * case-insensitively).
846 if (rest
[1] != 'i' && rest
[1] != 'I')
848 if (rest
[2] != 't' && rest
[2] != 'T')
850 if (rest
[3] == '\0' || is_dir_sep(rest
[3]))
854 if (skip_iprefix(rest
, "modules", &rest
) &&
855 (*rest
== '\0' || is_dir_sep(*rest
)))
860 if (rest
[1] == '\0' || is_dir_sep(rest
[1]))
866 int verify_path(const char *path
, unsigned mode
)
870 if (has_dos_drive_prefix(path
))
880 if (is_hfs_dotgit(path
))
883 if (is_hfs_dotgitmodules(path
))
888 if (is_ntfs_dotgit(path
))
891 if (is_ntfs_dotgitmodules(path
))
897 if ((c
== '.' && !verify_dotfile(path
, mode
)) ||
898 is_dir_sep(c
) || c
== '\0')
906 * Do we have another file that has the beginning components being a
907 * proper superset of the name we're trying to add?
909 static int has_file_name(struct index_state
*istate
,
910 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
913 int len
= ce_namelen(ce
);
914 int stage
= ce_stage(ce
);
915 const char *name
= ce
->name
;
917 while (pos
< istate
->cache_nr
) {
918 struct cache_entry
*p
= istate
->cache
[pos
++];
920 if (len
>= ce_namelen(p
))
922 if (memcmp(name
, p
->name
, len
))
924 if (ce_stage(p
) != stage
)
926 if (p
->name
[len
] != '/')
928 if (p
->ce_flags
& CE_REMOVE
)
933 remove_index_entry_at(istate
, --pos
);
940 * Like strcmp(), but also return the offset of the first change.
941 * If strings are equal, return the length.
943 int strcmp_offset(const char *s1
, const char *s2
, size_t *first_change
)
948 return strcmp(s1
, s2
);
950 for (k
= 0; s1
[k
] == s2
[k
]; k
++)
955 return (unsigned char)s1
[k
] - (unsigned char)s2
[k
];
959 * Do we have another file with a pathname that is a proper
960 * subset of the name we're trying to add?
962 * That is, is there another file in the index with a path
963 * that matches a sub-directory in the given entry?
965 static int has_dir_name(struct index_state
*istate
,
966 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
969 int stage
= ce_stage(ce
);
970 const char *name
= ce
->name
;
971 const char *slash
= name
+ ce_namelen(ce
);
976 * We are frequently called during an iteration on a sorted
977 * list of pathnames and while building a new index. Therefore,
978 * there is a high probability that this entry will eventually
979 * be appended to the index, rather than inserted in the middle.
980 * If we can confirm that, we can avoid binary searches on the
981 * components of the pathname.
983 * Compare the entry's full path with the last path in the index.
985 if (istate
->cache_nr
> 0) {
986 cmp_last
= strcmp_offset(name
,
987 istate
->cache
[istate
->cache_nr
- 1]->name
,
990 if (len_eq_last
== 0) {
992 * The entry sorts AFTER the last one in the
993 * index and their paths have no common prefix,
994 * so there cannot be a F/D conflict.
999 * The entry sorts AFTER the last one in the
1000 * index, but has a common prefix. Fall through
1001 * to the loop below to disect the entry's path
1002 * and see where the difference is.
1005 } else if (cmp_last
== 0) {
1007 * The entry exactly matches the last one in the
1008 * index, but because of multiple stage and CE_REMOVE
1009 * items, we fall through and let the regular search
1019 if (*--slash
== '/')
1021 if (slash
<= ce
->name
)
1028 * (len + 1) is a directory boundary (including
1029 * the trailing slash). And since the loop is
1030 * decrementing "slash", the first iteration is
1031 * the longest directory prefix; subsequent
1032 * iterations consider parent directories.
1035 if (len
+ 1 <= len_eq_last
) {
1037 * The directory prefix (including the trailing
1038 * slash) also appears as a prefix in the last
1039 * entry, so the remainder cannot collide (because
1040 * strcmp said the whole path was greater).
1045 * LT: last: xxx/file_A
1051 if (len
> len_eq_last
) {
1053 * This part of the directory prefix (excluding
1054 * the trailing slash) is longer than the known
1055 * equal portions, so this sub-directory cannot
1056 * collide with a file.
1064 if (istate
->cache_nr
> 0 &&
1065 ce_namelen(istate
->cache
[istate
->cache_nr
- 1]) > len
) {
1067 * The directory prefix lines up with part of
1068 * a longer file or directory name, but sorts
1069 * after it, so this sub-directory cannot
1070 * collide with a file.
1072 * last: xxx/yy-file (because '-' sorts before '/')
1079 * This is a possible collision. Fall through and
1080 * let the regular search code handle it.
1087 pos
= index_name_stage_pos(istate
, name
, len
, stage
);
1090 * Found one, but not so fast. This could
1091 * be a marker that says "I was here, but
1092 * I am being removed". Such an entry is
1093 * not a part of the resulting tree, and
1094 * it is Ok to have a directory at the same
1097 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
1101 remove_index_entry_at(istate
, pos
);
1109 * Trivial optimization: if we find an entry that
1110 * already matches the sub-directory, then we know
1111 * we're ok, and we can exit.
1113 while (pos
< istate
->cache_nr
) {
1114 struct cache_entry
*p
= istate
->cache
[pos
];
1115 if ((ce_namelen(p
) <= len
) ||
1116 (p
->name
[len
] != '/') ||
1117 memcmp(p
->name
, name
, len
))
1118 break; /* not our subdirectory */
1119 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
1121 * p is at the same stage as our entry, and
1122 * is a subdirectory of what we are looking
1123 * at, so we cannot have conflicts at our
1124 * level or anything shorter.
1133 /* We may be in a situation where we already have path/file and path
1134 * is being added, or we already have path and path/file is being
1135 * added. Either one would result in a nonsense tree that has path
1136 * twice when git-write-tree tries to write it out. Prevent it.
1138 * If ok-to-replace is specified, we remove the conflicting entries
1139 * from the cache so the caller should recompute the insert position.
1140 * When this happens, we return non-zero.
1142 static int check_file_directory_conflict(struct index_state
*istate
,
1143 const struct cache_entry
*ce
,
1144 int pos
, int ok_to_replace
)
1149 * When ce is an "I am going away" entry, we allow it to be added
1151 if (ce
->ce_flags
& CE_REMOVE
)
1155 * We check if the path is a sub-path of a subsequent pathname
1156 * first, since removing those will not change the position
1159 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
1162 * Then check if the path might have a clashing sub-directory
1165 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
1168 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1171 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
1172 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
1173 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
1174 int new_only
= option
& ADD_CACHE_NEW_ONLY
;
1176 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1177 cache_tree_invalidate_path(istate
, ce
->name
);
1180 * If this entry's path sorts after the last entry in the index,
1181 * we can avoid searching for it.
1183 if (istate
->cache_nr
> 0 &&
1184 strcmp(ce
->name
, istate
->cache
[istate
->cache_nr
- 1]->name
) > 0)
1185 pos
= -istate
->cache_nr
- 1;
1187 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1189 /* existing match? Just replace it. */
1192 replace_index_entry(istate
, pos
, ce
);
1197 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1198 untracked_cache_add_to_index(istate
, ce
->name
);
1201 * Inserting a merged entry ("stage 0") into the index
1202 * will always replace all non-merged entries..
1204 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
1205 while (ce_same_name(istate
->cache
[pos
], ce
)) {
1207 if (!remove_index_entry_at(istate
, pos
))
1214 if (!verify_path(ce
->name
, ce
->ce_mode
))
1215 return error("Invalid path '%s'", ce
->name
);
1217 if (!skip_df_check
&&
1218 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
1220 return error("'%s' appears as both a file and as a directory",
1222 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1228 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1232 if (option
& ADD_CACHE_JUST_APPEND
)
1233 pos
= istate
->cache_nr
;
1236 ret
= add_index_entry_with_check(istate
, ce
, option
);
1242 /* Make sure the array is big enough .. */
1243 ALLOC_GROW(istate
->cache
, istate
->cache_nr
+ 1, istate
->cache_alloc
);
1247 if (istate
->cache_nr
> pos
+ 1)
1248 MOVE_ARRAY(istate
->cache
+ pos
+ 1, istate
->cache
+ pos
,
1249 istate
->cache_nr
- pos
- 1);
1250 set_index_entry(istate
, pos
, ce
);
1251 istate
->cache_changed
|= CE_ENTRY_ADDED
;
1256 * "refresh" does not calculate a new sha1 file or bring the
1257 * cache up-to-date for mode/content changes. But what it
1258 * _does_ do is to "re-match" the stat information of a file
1259 * with the cache, so that you can refresh the cache for a
1260 * file that hasn't been changed but where the stat entry is
1263 * For example, you'd want to do this after doing a "git-read-tree",
1264 * to link up the stat cache details with the proper files.
1266 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
1267 struct cache_entry
*ce
,
1268 unsigned int options
, int *err
,
1272 struct cache_entry
*updated
;
1274 int refresh
= options
& CE_MATCH_REFRESH
;
1275 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
1276 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
1277 int ignore_missing
= options
& CE_MATCH_IGNORE_MISSING
;
1278 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
1280 if (!refresh
|| ce_uptodate(ce
))
1283 if (!ignore_fsmonitor
)
1284 refresh_fsmonitor(istate
);
1286 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1287 * that the change to the work tree does not matter and told
1290 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
)) {
1291 ce_mark_uptodate(ce
);
1294 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
)) {
1295 ce_mark_uptodate(ce
);
1298 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
)) {
1299 ce_mark_uptodate(ce
);
1303 if (has_symlink_leading_path(ce
->name
, ce_namelen(ce
))) {
1311 if (lstat(ce
->name
, &st
) < 0) {
1312 if (ignore_missing
&& errno
== ENOENT
)
1319 changed
= ie_match_stat(istate
, ce
, &st
, options
);
1321 *changed_ret
= changed
;
1324 * The path is unchanged. If we were told to ignore
1325 * valid bit, then we did the actual stat check and
1326 * found that the entry is unmodified. If the entry
1327 * is not marked VALID, this is the place to mark it
1328 * valid again, under "assume unchanged" mode.
1330 if (ignore_valid
&& assume_unchanged
&&
1331 !(ce
->ce_flags
& CE_VALID
))
1332 ; /* mark this one VALID again */
1335 * We do not mark the index itself "modified"
1336 * because CE_UPTODATE flag is in-core only;
1337 * we are not going to write this change out.
1339 if (!S_ISGITLINK(ce
->ce_mode
)) {
1340 ce_mark_uptodate(ce
);
1341 mark_fsmonitor_valid(ce
);
1347 if (ie_modified(istate
, ce
, &st
, options
)) {
1354 updated
= xmalloc(size
);
1355 copy_cache_entry(updated
, ce
);
1356 memcpy(updated
->name
, ce
->name
, ce
->ce_namelen
+ 1);
1357 fill_stat_cache_info(updated
, &st
);
1359 * If ignore_valid is not set, we should leave CE_VALID bit
1360 * alone. Otherwise, paths marked with --no-assume-unchanged
1361 * (i.e. things to be edited) will reacquire CE_VALID bit
1362 * automatically, which is not really what we want.
1364 if (!ignore_valid
&& assume_unchanged
&&
1365 !(ce
->ce_flags
& CE_VALID
))
1366 updated
->ce_flags
&= ~CE_VALID
;
1368 /* istate->cache_changed is updated in the caller */
1372 static void show_file(const char * fmt
, const char * name
, int in_porcelain
,
1373 int * first
, const char *header_msg
)
1375 if (in_porcelain
&& *first
&& header_msg
) {
1376 printf("%s\n", header_msg
);
1382 int refresh_index(struct index_state
*istate
, unsigned int flags
,
1383 const struct pathspec
*pathspec
,
1384 char *seen
, const char *header_msg
)
1388 int really
= (flags
& REFRESH_REALLY
) != 0;
1389 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
1390 int quiet
= (flags
& REFRESH_QUIET
) != 0;
1391 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
1392 int ignore_submodules
= (flags
& REFRESH_IGNORE_SUBMODULES
) != 0;
1394 int in_porcelain
= (flags
& REFRESH_IN_PORCELAIN
);
1395 unsigned int options
= (CE_MATCH_REFRESH
|
1396 (really
? CE_MATCH_IGNORE_VALID
: 0) |
1397 (not_new
? CE_MATCH_IGNORE_MISSING
: 0));
1398 const char *modified_fmt
;
1399 const char *deleted_fmt
;
1400 const char *typechange_fmt
;
1401 const char *added_fmt
;
1402 const char *unmerged_fmt
;
1403 uint64_t start
= getnanotime();
1405 modified_fmt
= (in_porcelain
? "M\t%s\n" : "%s: needs update\n");
1406 deleted_fmt
= (in_porcelain
? "D\t%s\n" : "%s: needs update\n");
1407 typechange_fmt
= (in_porcelain
? "T\t%s\n" : "%s needs update\n");
1408 added_fmt
= (in_porcelain
? "A\t%s\n" : "%s needs update\n");
1409 unmerged_fmt
= (in_porcelain
? "U\t%s\n" : "%s: needs merge\n");
1410 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1411 struct cache_entry
*ce
, *new_entry
;
1412 int cache_errno
= 0;
1416 ce
= istate
->cache
[i
];
1417 if (ignore_submodules
&& S_ISGITLINK(ce
->ce_mode
))
1420 if (pathspec
&& !ce_path_match(ce
, pathspec
, seen
))
1424 while ((i
< istate
->cache_nr
) &&
1425 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
1431 show_file(unmerged_fmt
, ce
->name
, in_porcelain
,
1432 &first
, header_msg
);
1440 new_entry
= refresh_cache_ent(istate
, ce
, options
, &cache_errno
, &changed
);
1441 if (new_entry
== ce
)
1446 if (really
&& cache_errno
== EINVAL
) {
1447 /* If we are doing --really-refresh that
1448 * means the index is not valid anymore.
1450 ce
->ce_flags
&= ~CE_VALID
;
1451 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
1452 mark_fsmonitor_invalid(istate
, ce
);
1453 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
1458 if (cache_errno
== ENOENT
)
1460 else if (ce_intent_to_add(ce
))
1461 fmt
= added_fmt
; /* must be before other checks */
1462 else if (changed
& TYPE_CHANGED
)
1463 fmt
= typechange_fmt
;
1467 ce
->name
, in_porcelain
, &first
, header_msg
);
1472 replace_index_entry(istate
, i
, new_entry
);
1474 trace_performance_since(start
, "refresh index");
1478 struct cache_entry
*refresh_cache_entry(struct cache_entry
*ce
,
1479 unsigned int options
)
1481 return refresh_cache_ent(&the_index
, ce
, options
, NULL
, NULL
);
1485 /*****************************************************************
1487 *****************************************************************/
1489 #define INDEX_FORMAT_DEFAULT 3
1491 static unsigned int get_index_format_default(void)
1493 char *envversion
= getenv("GIT_INDEX_VERSION");
1496 unsigned int version
= INDEX_FORMAT_DEFAULT
;
1499 if (!git_config_get_int("index.version", &value
))
1501 if (version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1502 warning(_("index.version set, but the value is invalid.\n"
1503 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1504 return INDEX_FORMAT_DEFAULT
;
1509 version
= strtoul(envversion
, &endp
, 10);
1511 version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1512 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1513 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1514 version
= INDEX_FORMAT_DEFAULT
;
1520 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1521 * Again - this is just a (very strong in practice) heuristic that
1522 * the inode hasn't changed.
1524 * We save the fields in big-endian order to allow using the
1525 * index file over NFS transparently.
1527 struct ondisk_cache_entry
{
1528 struct cache_time ctime
;
1529 struct cache_time mtime
;
1536 unsigned char sha1
[20];
1538 char name
[FLEX_ARRAY
]; /* more */
1542 * This struct is used when CE_EXTENDED bit is 1
1543 * The struct must match ondisk_cache_entry exactly from
1546 struct ondisk_cache_entry_extended
{
1547 struct cache_time ctime
;
1548 struct cache_time mtime
;
1555 unsigned char sha1
[20];
1558 char name
[FLEX_ARRAY
]; /* more */
1561 /* These are only used for v3 or lower */
1562 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1563 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1564 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1565 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1566 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1567 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1568 ondisk_cache_entry_size(ce_namelen(ce)))
1570 /* Allow fsck to force verification of the index checksum. */
1571 int verify_index_checksum
;
1573 /* Allow fsck to force verification of the cache entry order. */
1574 int verify_ce_order
;
1576 static int verify_hdr(struct cache_header
*hdr
, unsigned long size
)
1579 unsigned char hash
[GIT_MAX_RAWSZ
];
1582 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
1583 return error("bad signature");
1584 hdr_version
= ntohl(hdr
->hdr_version
);
1585 if (hdr_version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< hdr_version
)
1586 return error("bad index version %d", hdr_version
);
1588 if (!verify_index_checksum
)
1591 the_hash_algo
->init_fn(&c
);
1592 the_hash_algo
->update_fn(&c
, hdr
, size
- the_hash_algo
->rawsz
);
1593 the_hash_algo
->final_fn(hash
, &c
);
1594 if (hashcmp(hash
, (unsigned char *)hdr
+ size
- the_hash_algo
->rawsz
))
1595 return error("bad index file sha1 signature");
1599 static int read_index_extension(struct index_state
*istate
,
1600 const char *ext
, void *data
, unsigned long sz
)
1602 switch (CACHE_EXT(ext
)) {
1603 case CACHE_EXT_TREE
:
1604 istate
->cache_tree
= cache_tree_read(data
, sz
);
1606 case CACHE_EXT_RESOLVE_UNDO
:
1607 istate
->resolve_undo
= resolve_undo_read(data
, sz
);
1609 case CACHE_EXT_LINK
:
1610 if (read_link_extension(istate
, data
, sz
))
1613 case CACHE_EXT_UNTRACKED
:
1614 istate
->untracked
= read_untracked_extension(data
, sz
);
1616 case CACHE_EXT_FSMONITOR
:
1617 read_fsmonitor_extension(istate
, data
, sz
);
1620 if (*ext
< 'A' || 'Z' < *ext
)
1621 return error("index uses %.4s extension, which we do not understand",
1623 fprintf(stderr
, "ignoring %.4s extension\n", ext
);
1629 int hold_locked_index(struct lock_file
*lk
, int lock_flags
)
1631 return hold_lock_file_for_update(lk
, get_index_file(), lock_flags
);
1634 int read_index(struct index_state
*istate
)
1636 return read_index_from(istate
, get_index_file(), get_git_dir());
1639 static struct cache_entry
*cache_entry_from_ondisk(struct ondisk_cache_entry
*ondisk
,
1644 struct cache_entry
*ce
= xmalloc(cache_entry_size(len
));
1646 ce
->ce_stat_data
.sd_ctime
.sec
= get_be32(&ondisk
->ctime
.sec
);
1647 ce
->ce_stat_data
.sd_mtime
.sec
= get_be32(&ondisk
->mtime
.sec
);
1648 ce
->ce_stat_data
.sd_ctime
.nsec
= get_be32(&ondisk
->ctime
.nsec
);
1649 ce
->ce_stat_data
.sd_mtime
.nsec
= get_be32(&ondisk
->mtime
.nsec
);
1650 ce
->ce_stat_data
.sd_dev
= get_be32(&ondisk
->dev
);
1651 ce
->ce_stat_data
.sd_ino
= get_be32(&ondisk
->ino
);
1652 ce
->ce_mode
= get_be32(&ondisk
->mode
);
1653 ce
->ce_stat_data
.sd_uid
= get_be32(&ondisk
->uid
);
1654 ce
->ce_stat_data
.sd_gid
= get_be32(&ondisk
->gid
);
1655 ce
->ce_stat_data
.sd_size
= get_be32(&ondisk
->size
);
1656 ce
->ce_flags
= flags
& ~CE_NAMEMASK
;
1657 ce
->ce_namelen
= len
;
1659 hashcpy(ce
->oid
.hash
, ondisk
->sha1
);
1660 memcpy(ce
->name
, name
, len
);
1661 ce
->name
[len
] = '\0';
1666 * Adjacent cache entries tend to share the leading paths, so it makes
1667 * sense to only store the differences in later entries. In the v4
1668 * on-disk format of the index, each on-disk cache entry stores the
1669 * number of bytes to be stripped from the end of the previous name,
1670 * and the bytes to append to the result, to come up with its name.
1672 static unsigned long expand_name_field(struct strbuf
*name
, const char *cp_
)
1674 const unsigned char *ep
, *cp
= (const unsigned char *)cp_
;
1675 size_t len
= decode_varint(&cp
);
1677 if (name
->len
< len
)
1678 die("malformed name field in the index");
1679 strbuf_remove(name
, name
->len
- len
, len
);
1680 for (ep
= cp
; *ep
; ep
++)
1681 ; /* find the end */
1682 strbuf_add(name
, cp
, ep
- cp
);
1683 return (const char *)ep
+ 1 - cp_
;
1686 static struct cache_entry
*create_from_disk(struct ondisk_cache_entry
*ondisk
,
1687 unsigned long *ent_size
,
1688 struct strbuf
*previous_name
)
1690 struct cache_entry
*ce
;
1695 /* On-disk flags are just 16 bits */
1696 flags
= get_be16(&ondisk
->flags
);
1697 len
= flags
& CE_NAMEMASK
;
1699 if (flags
& CE_EXTENDED
) {
1700 struct ondisk_cache_entry_extended
*ondisk2
;
1702 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
1703 extended_flags
= get_be16(&ondisk2
->flags2
) << 16;
1704 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1705 if (extended_flags
& ~CE_EXTENDED_FLAGS
)
1706 die("Unknown index entry format %08x", extended_flags
);
1707 flags
|= extended_flags
;
1708 name
= ondisk2
->name
;
1711 name
= ondisk
->name
;
1713 if (!previous_name
) {
1714 /* v3 and earlier */
1715 if (len
== CE_NAMEMASK
)
1717 ce
= cache_entry_from_ondisk(ondisk
, flags
, name
, len
);
1719 *ent_size
= ondisk_ce_size(ce
);
1721 unsigned long consumed
;
1722 consumed
= expand_name_field(previous_name
, name
);
1723 ce
= cache_entry_from_ondisk(ondisk
, flags
,
1725 previous_name
->len
);
1727 *ent_size
= (name
- ((char *)ondisk
)) + consumed
;
1732 static void check_ce_order(struct index_state
*istate
)
1736 if (!verify_ce_order
)
1739 for (i
= 1; i
< istate
->cache_nr
; i
++) {
1740 struct cache_entry
*ce
= istate
->cache
[i
- 1];
1741 struct cache_entry
*next_ce
= istate
->cache
[i
];
1742 int name_compare
= strcmp(ce
->name
, next_ce
->name
);
1744 if (0 < name_compare
)
1745 die("unordered stage entries in index");
1746 if (!name_compare
) {
1748 die("multiple stage entries for merged file '%s'",
1750 if (ce_stage(ce
) > ce_stage(next_ce
))
1751 die("unordered stage entries for '%s'",
1757 static void tweak_untracked_cache(struct index_state
*istate
)
1759 switch (git_config_get_untracked_cache()) {
1760 case -1: /* keep: do nothing */
1763 remove_untracked_cache(istate
);
1766 add_untracked_cache(istate
);
1768 default: /* unknown value: do nothing */
1773 static void tweak_split_index(struct index_state
*istate
)
1775 switch (git_config_get_split_index()) {
1776 case -1: /* unset: do nothing */
1779 remove_split_index(istate
);
1782 add_split_index(istate
);
1784 default: /* unknown value: do nothing */
1789 static void post_read_index_from(struct index_state
*istate
)
1791 check_ce_order(istate
);
1792 tweak_untracked_cache(istate
);
1793 tweak_split_index(istate
);
1794 tweak_fsmonitor(istate
);
1797 /* remember to discard_cache() before reading a different cache! */
1798 int do_read_index(struct index_state
*istate
, const char *path
, int must_exist
)
1802 unsigned long src_offset
;
1803 struct cache_header
*hdr
;
1806 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
1808 if (istate
->initialized
)
1809 return istate
->cache_nr
;
1811 istate
->timestamp
.sec
= 0;
1812 istate
->timestamp
.nsec
= 0;
1813 fd
= open(path
, O_RDONLY
);
1815 if (!must_exist
&& errno
== ENOENT
)
1817 die_errno("%s: index file open failed", path
);
1821 die_errno("cannot stat the open index");
1823 mmap_size
= xsize_t(st
.st_size
);
1824 if (mmap_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
1825 die("index file smaller than expected");
1827 mmap
= xmmap(NULL
, mmap_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1828 if (mmap
== MAP_FAILED
)
1829 die_errno("unable to map index file");
1833 if (verify_hdr(hdr
, mmap_size
) < 0)
1836 hashcpy(istate
->sha1
, (const unsigned char *)hdr
+ mmap_size
- the_hash_algo
->rawsz
);
1837 istate
->version
= ntohl(hdr
->hdr_version
);
1838 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
1839 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
1840 istate
->cache
= xcalloc(istate
->cache_alloc
, sizeof(*istate
->cache
));
1841 istate
->initialized
= 1;
1843 if (istate
->version
== 4)
1844 previous_name
= &previous_name_buf
;
1846 previous_name
= NULL
;
1848 src_offset
= sizeof(*hdr
);
1849 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1850 struct ondisk_cache_entry
*disk_ce
;
1851 struct cache_entry
*ce
;
1852 unsigned long consumed
;
1854 disk_ce
= (struct ondisk_cache_entry
*)((char *)mmap
+ src_offset
);
1855 ce
= create_from_disk(disk_ce
, &consumed
, previous_name
);
1856 set_index_entry(istate
, i
, ce
);
1858 src_offset
+= consumed
;
1860 strbuf_release(&previous_name_buf
);
1861 istate
->timestamp
.sec
= st
.st_mtime
;
1862 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
1864 while (src_offset
<= mmap_size
- the_hash_algo
->rawsz
- 8) {
1865 /* After an array of active_nr index entries,
1866 * there can be arbitrary number of extended
1867 * sections, each of which is prefixed with
1868 * extension name (4-byte) and section length
1869 * in 4-byte network byte order.
1872 memcpy(&extsize
, (char *)mmap
+ src_offset
+ 4, 4);
1873 extsize
= ntohl(extsize
);
1874 if (read_index_extension(istate
,
1875 (const char *) mmap
+ src_offset
,
1876 (char *) mmap
+ src_offset
+ 8,
1880 src_offset
+= extsize
;
1882 munmap(mmap
, mmap_size
);
1883 return istate
->cache_nr
;
1886 munmap(mmap
, mmap_size
);
1887 die("index file corrupt");
1891 * Signal that the shared index is used by updating its mtime.
1893 * This way, shared index can be removed if they have not been used
1896 static void freshen_shared_index(const char *shared_index
, int warn
)
1898 if (!check_and_freshen_file(shared_index
, 1) && warn
)
1899 warning("could not freshen shared index '%s'", shared_index
);
1902 int read_index_from(struct index_state
*istate
, const char *path
,
1905 uint64_t start
= getnanotime();
1906 struct split_index
*split_index
;
1908 char *base_sha1_hex
;
1911 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1912 if (istate
->initialized
)
1913 return istate
->cache_nr
;
1915 ret
= do_read_index(istate
, path
, 0);
1916 trace_performance_since(start
, "read cache %s", path
);
1918 split_index
= istate
->split_index
;
1919 if (!split_index
|| is_null_sha1(split_index
->base_sha1
)) {
1920 post_read_index_from(istate
);
1924 if (split_index
->base
)
1925 discard_index(split_index
->base
);
1927 split_index
->base
= xcalloc(1, sizeof(*split_index
->base
));
1929 base_sha1_hex
= sha1_to_hex(split_index
->base_sha1
);
1930 base_path
= xstrfmt("%s/sharedindex.%s", gitdir
, base_sha1_hex
);
1931 ret
= do_read_index(split_index
->base
, base_path
, 1);
1932 if (hashcmp(split_index
->base_sha1
, split_index
->base
->sha1
))
1933 die("broken index, expect %s in %s, got %s",
1934 base_sha1_hex
, base_path
,
1935 sha1_to_hex(split_index
->base
->sha1
));
1937 freshen_shared_index(base_path
, 0);
1938 merge_base_index(istate
);
1939 post_read_index_from(istate
);
1940 trace_performance_since(start
, "read cache %s", base_path
);
1945 int is_index_unborn(struct index_state
*istate
)
1947 return (!istate
->cache_nr
&& !istate
->timestamp
.sec
);
1950 int discard_index(struct index_state
*istate
)
1954 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1955 if (istate
->cache
[i
]->index
&&
1956 istate
->split_index
&&
1957 istate
->split_index
->base
&&
1958 istate
->cache
[i
]->index
<= istate
->split_index
->base
->cache_nr
&&
1959 istate
->cache
[i
] == istate
->split_index
->base
->cache
[istate
->cache
[i
]->index
- 1])
1961 free(istate
->cache
[i
]);
1963 resolve_undo_clear_index(istate
);
1964 istate
->cache_nr
= 0;
1965 istate
->cache_changed
= 0;
1966 istate
->timestamp
.sec
= 0;
1967 istate
->timestamp
.nsec
= 0;
1968 free_name_hash(istate
);
1969 cache_tree_free(&(istate
->cache_tree
));
1970 istate
->initialized
= 0;
1971 FREE_AND_NULL(istate
->cache
);
1972 istate
->cache_alloc
= 0;
1973 discard_split_index(istate
);
1974 free_untracked_cache(istate
->untracked
);
1975 istate
->untracked
= NULL
;
1979 int unmerged_index(const struct index_state
*istate
)
1982 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1983 if (ce_stage(istate
->cache
[i
]))
1989 int index_has_changes(const struct index_state
*istate
,
1993 struct object_id cmp
;
1996 if (istate
!= &the_index
) {
1997 BUG("index_has_changes cannot yet accept istate != &the_index; do_diff_cache needs updating first.");
2000 cmp
= tree
->object
.oid
;
2001 if (tree
|| !get_oid_tree("HEAD", &cmp
)) {
2002 struct diff_options opt
;
2005 opt
.flags
.exit_with_status
= 1;
2007 opt
.flags
.quick
= 1;
2008 do_diff_cache(&cmp
, &opt
);
2010 for (i
= 0; sb
&& i
< diff_queued_diff
.nr
; i
++) {
2012 strbuf_addch(sb
, ' ');
2013 strbuf_addstr(sb
, diff_queued_diff
.queue
[i
]->two
->path
);
2016 return opt
.flags
.has_changes
!= 0;
2018 for (i
= 0; sb
&& i
< istate
->cache_nr
; i
++) {
2020 strbuf_addch(sb
, ' ');
2021 strbuf_addstr(sb
, istate
->cache
[i
]->name
);
2023 return !!istate
->cache_nr
;
2027 #define WRITE_BUFFER_SIZE 8192
2028 static unsigned char write_buffer
[WRITE_BUFFER_SIZE
];
2029 static unsigned long write_buffer_len
;
2031 static int ce_write_flush(git_hash_ctx
*context
, int fd
)
2033 unsigned int buffered
= write_buffer_len
;
2035 the_hash_algo
->update_fn(context
, write_buffer
, buffered
);
2036 if (write_in_full(fd
, write_buffer
, buffered
) < 0)
2038 write_buffer_len
= 0;
2043 static int ce_write(git_hash_ctx
*context
, int fd
, void *data
, unsigned int len
)
2046 unsigned int buffered
= write_buffer_len
;
2047 unsigned int partial
= WRITE_BUFFER_SIZE
- buffered
;
2050 memcpy(write_buffer
+ buffered
, data
, partial
);
2051 buffered
+= partial
;
2052 if (buffered
== WRITE_BUFFER_SIZE
) {
2053 write_buffer_len
= buffered
;
2054 if (ce_write_flush(context
, fd
))
2058 write_buffer_len
= buffered
;
2060 data
= (char *) data
+ partial
;
2065 static int write_index_ext_header(git_hash_ctx
*context
, int fd
,
2066 unsigned int ext
, unsigned int sz
)
2070 return ((ce_write(context
, fd
, &ext
, 4) < 0) ||
2071 (ce_write(context
, fd
, &sz
, 4) < 0)) ? -1 : 0;
2074 static int ce_flush(git_hash_ctx
*context
, int fd
, unsigned char *hash
)
2076 unsigned int left
= write_buffer_len
;
2079 write_buffer_len
= 0;
2080 the_hash_algo
->update_fn(context
, write_buffer
, left
);
2083 /* Flush first if not enough space for hash signature */
2084 if (left
+ the_hash_algo
->rawsz
> WRITE_BUFFER_SIZE
) {
2085 if (write_in_full(fd
, write_buffer
, left
) < 0)
2090 /* Append the hash signature at the end */
2091 the_hash_algo
->final_fn(write_buffer
+ left
, context
);
2092 hashcpy(hash
, write_buffer
+ left
);
2093 left
+= the_hash_algo
->rawsz
;
2094 return (write_in_full(fd
, write_buffer
, left
) < 0) ? -1 : 0;
2097 static void ce_smudge_racily_clean_entry(struct cache_entry
*ce
)
2100 * The only thing we care about in this function is to smudge the
2101 * falsely clean entry due to touch-update-touch race, so we leave
2102 * everything else as they are. We are called for entries whose
2103 * ce_stat_data.sd_mtime match the index file mtime.
2105 * Note that this actually does not do much for gitlinks, for
2106 * which ce_match_stat_basic() always goes to the actual
2107 * contents. The caller checks with is_racy_timestamp() which
2108 * always says "no" for gitlinks, so we are not called for them ;-)
2112 if (lstat(ce
->name
, &st
) < 0)
2114 if (ce_match_stat_basic(ce
, &st
))
2116 if (ce_modified_check_fs(ce
, &st
)) {
2117 /* This is "racily clean"; smudge it. Note that this
2118 * is a tricky code. At first glance, it may appear
2119 * that it can break with this sequence:
2121 * $ echo xyzzy >frotz
2122 * $ git-update-index --add frotz
2125 * $ echo filfre >nitfol
2126 * $ git-update-index --add nitfol
2128 * but it does not. When the second update-index runs,
2129 * it notices that the entry "frotz" has the same timestamp
2130 * as index, and if we were to smudge it by resetting its
2131 * size to zero here, then the object name recorded
2132 * in index is the 6-byte file but the cached stat information
2133 * becomes zero --- which would then match what we would
2134 * obtain from the filesystem next time we stat("frotz").
2136 * However, the second update-index, before calling
2137 * this function, notices that the cached size is 6
2138 * bytes and what is on the filesystem is an empty
2139 * file, and never calls us, so the cached size information
2140 * for "frotz" stays 6 which does not match the filesystem.
2142 ce
->ce_stat_data
.sd_size
= 0;
2146 /* Copy miscellaneous fields but not the name */
2147 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry
*ondisk
,
2148 struct cache_entry
*ce
)
2152 ondisk
->ctime
.sec
= htonl(ce
->ce_stat_data
.sd_ctime
.sec
);
2153 ondisk
->mtime
.sec
= htonl(ce
->ce_stat_data
.sd_mtime
.sec
);
2154 ondisk
->ctime
.nsec
= htonl(ce
->ce_stat_data
.sd_ctime
.nsec
);
2155 ondisk
->mtime
.nsec
= htonl(ce
->ce_stat_data
.sd_mtime
.nsec
);
2156 ondisk
->dev
= htonl(ce
->ce_stat_data
.sd_dev
);
2157 ondisk
->ino
= htonl(ce
->ce_stat_data
.sd_ino
);
2158 ondisk
->mode
= htonl(ce
->ce_mode
);
2159 ondisk
->uid
= htonl(ce
->ce_stat_data
.sd_uid
);
2160 ondisk
->gid
= htonl(ce
->ce_stat_data
.sd_gid
);
2161 ondisk
->size
= htonl(ce
->ce_stat_data
.sd_size
);
2162 hashcpy(ondisk
->sha1
, ce
->oid
.hash
);
2164 flags
= ce
->ce_flags
& ~CE_NAMEMASK
;
2165 flags
|= (ce_namelen(ce
) >= CE_NAMEMASK
? CE_NAMEMASK
: ce_namelen(ce
));
2166 ondisk
->flags
= htons(flags
);
2167 if (ce
->ce_flags
& CE_EXTENDED
) {
2168 struct ondisk_cache_entry_extended
*ondisk2
;
2169 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
2170 ondisk2
->flags2
= htons((ce
->ce_flags
& CE_EXTENDED_FLAGS
) >> 16);
2174 static int ce_write_entry(git_hash_ctx
*c
, int fd
, struct cache_entry
*ce
,
2175 struct strbuf
*previous_name
, struct ondisk_cache_entry
*ondisk
)
2179 unsigned int saved_namelen
;
2180 int stripped_name
= 0;
2181 static unsigned char padding
[8] = { 0x00 };
2183 if (ce
->ce_flags
& CE_STRIP_NAME
) {
2184 saved_namelen
= ce_namelen(ce
);
2189 if (ce
->ce_flags
& CE_EXTENDED
)
2190 size
= offsetof(struct ondisk_cache_entry_extended
, name
);
2192 size
= offsetof(struct ondisk_cache_entry
, name
);
2194 if (!previous_name
) {
2195 int len
= ce_namelen(ce
);
2196 copy_cache_entry_to_ondisk(ondisk
, ce
);
2197 result
= ce_write(c
, fd
, ondisk
, size
);
2199 result
= ce_write(c
, fd
, ce
->name
, len
);
2201 result
= ce_write(c
, fd
, padding
, align_padding_size(size
, len
));
2203 int common
, to_remove
, prefix_size
;
2204 unsigned char to_remove_vi
[16];
2206 (ce
->name
[common
] &&
2207 common
< previous_name
->len
&&
2208 ce
->name
[common
] == previous_name
->buf
[common
]);
2210 ; /* still matching */
2211 to_remove
= previous_name
->len
- common
;
2212 prefix_size
= encode_varint(to_remove
, to_remove_vi
);
2214 copy_cache_entry_to_ondisk(ondisk
, ce
);
2215 result
= ce_write(c
, fd
, ondisk
, size
);
2217 result
= ce_write(c
, fd
, to_remove_vi
, prefix_size
);
2219 result
= ce_write(c
, fd
, ce
->name
+ common
, ce_namelen(ce
) - common
);
2221 result
= ce_write(c
, fd
, padding
, 1);
2223 strbuf_splice(previous_name
, common
, to_remove
,
2224 ce
->name
+ common
, ce_namelen(ce
) - common
);
2226 if (stripped_name
) {
2227 ce
->ce_namelen
= saved_namelen
;
2228 ce
->ce_flags
&= ~CE_STRIP_NAME
;
2235 * This function verifies if index_state has the correct sha1 of the
2236 * index file. Don't die if we have any other failure, just return 0.
2238 static int verify_index_from(const struct index_state
*istate
, const char *path
)
2243 unsigned char hash
[GIT_MAX_RAWSZ
];
2245 if (!istate
->initialized
)
2248 fd
= open(path
, O_RDONLY
);
2255 if (st
.st_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2258 n
= pread_in_full(fd
, hash
, the_hash_algo
->rawsz
, st
.st_size
- the_hash_algo
->rawsz
);
2259 if (n
!= the_hash_algo
->rawsz
)
2262 if (hashcmp(istate
->sha1
, hash
))
2273 static int verify_index(const struct index_state
*istate
)
2275 return verify_index_from(istate
, get_index_file());
2278 static int has_racy_timestamp(struct index_state
*istate
)
2280 int entries
= istate
->cache_nr
;
2283 for (i
= 0; i
< entries
; i
++) {
2284 struct cache_entry
*ce
= istate
->cache
[i
];
2285 if (is_racy_timestamp(istate
, ce
))
2291 void update_index_if_able(struct index_state
*istate
, struct lock_file
*lockfile
)
2293 if ((istate
->cache_changed
|| has_racy_timestamp(istate
)) &&
2294 verify_index(istate
))
2295 write_locked_index(istate
, lockfile
, COMMIT_LOCK
);
2297 rollback_lock_file(lockfile
);
2301 * On success, `tempfile` is closed. If it is the temporary file
2302 * of a `struct lock_file`, we will therefore effectively perform
2303 * a 'close_lock_file_gently()`. Since that is an implementation
2304 * detail of lockfiles, callers of `do_write_index()` should not
2307 static int do_write_index(struct index_state
*istate
, struct tempfile
*tempfile
,
2308 int strip_extensions
)
2310 uint64_t start
= getnanotime();
2311 int newfd
= tempfile
->fd
;
2313 struct cache_header hdr
;
2314 int i
, err
= 0, removed
, extended
, hdr_version
;
2315 struct cache_entry
**cache
= istate
->cache
;
2316 int entries
= istate
->cache_nr
;
2318 struct ondisk_cache_entry_extended ondisk
;
2319 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
2320 int drop_cache_tree
= istate
->drop_cache_tree
;
2322 for (i
= removed
= extended
= 0; i
< entries
; i
++) {
2323 if (cache
[i
]->ce_flags
& CE_REMOVE
)
2326 /* reduce extended entries if possible */
2327 cache
[i
]->ce_flags
&= ~CE_EXTENDED
;
2328 if (cache
[i
]->ce_flags
& CE_EXTENDED_FLAGS
) {
2330 cache
[i
]->ce_flags
|= CE_EXTENDED
;
2334 if (!istate
->version
) {
2335 istate
->version
= get_index_format_default();
2336 if (getenv("GIT_TEST_SPLIT_INDEX"))
2337 init_split_index(istate
);
2340 /* demote version 3 to version 2 when the latter suffices */
2341 if (istate
->version
== 3 || istate
->version
== 2)
2342 istate
->version
= extended
? 3 : 2;
2344 hdr_version
= istate
->version
;
2346 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
2347 hdr
.hdr_version
= htonl(hdr_version
);
2348 hdr
.hdr_entries
= htonl(entries
- removed
);
2350 the_hash_algo
->init_fn(&c
);
2351 if (ce_write(&c
, newfd
, &hdr
, sizeof(hdr
)) < 0)
2354 previous_name
= (hdr_version
== 4) ? &previous_name_buf
: NULL
;
2356 for (i
= 0; i
< entries
; i
++) {
2357 struct cache_entry
*ce
= cache
[i
];
2358 if (ce
->ce_flags
& CE_REMOVE
)
2360 if (!ce_uptodate(ce
) && is_racy_timestamp(istate
, ce
))
2361 ce_smudge_racily_clean_entry(ce
);
2362 if (is_null_oid(&ce
->oid
)) {
2363 static const char msg
[] = "cache entry has null sha1: %s";
2364 static int allow
= -1;
2367 allow
= git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2369 warning(msg
, ce
->name
);
2371 err
= error(msg
, ce
->name
);
2373 drop_cache_tree
= 1;
2375 if (ce_write_entry(&c
, newfd
, ce
, previous_name
, (struct ondisk_cache_entry
*)&ondisk
) < 0)
2381 strbuf_release(&previous_name_buf
);
2386 /* Write extension data here */
2387 if (!strip_extensions
&& istate
->split_index
) {
2388 struct strbuf sb
= STRBUF_INIT
;
2390 err
= write_link_extension(&sb
, istate
) < 0 ||
2391 write_index_ext_header(&c
, newfd
, CACHE_EXT_LINK
,
2393 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2394 strbuf_release(&sb
);
2398 if (!strip_extensions
&& !drop_cache_tree
&& istate
->cache_tree
) {
2399 struct strbuf sb
= STRBUF_INIT
;
2401 cache_tree_write(&sb
, istate
->cache_tree
);
2402 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_TREE
, sb
.len
) < 0
2403 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2404 strbuf_release(&sb
);
2408 if (!strip_extensions
&& istate
->resolve_undo
) {
2409 struct strbuf sb
= STRBUF_INIT
;
2411 resolve_undo_write(&sb
, istate
->resolve_undo
);
2412 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_RESOLVE_UNDO
,
2414 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2415 strbuf_release(&sb
);
2419 if (!strip_extensions
&& istate
->untracked
) {
2420 struct strbuf sb
= STRBUF_INIT
;
2422 write_untracked_extension(&sb
, istate
->untracked
);
2423 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_UNTRACKED
,
2425 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2426 strbuf_release(&sb
);
2430 if (!strip_extensions
&& istate
->fsmonitor_last_update
) {
2431 struct strbuf sb
= STRBUF_INIT
;
2433 write_fsmonitor_extension(&sb
, istate
);
2434 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_FSMONITOR
, sb
.len
) < 0
2435 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2436 strbuf_release(&sb
);
2441 if (ce_flush(&c
, newfd
, istate
->sha1
))
2443 if (close_tempfile_gently(tempfile
)) {
2444 error(_("could not close '%s'"), tempfile
->filename
.buf
);
2447 if (stat(tempfile
->filename
.buf
, &st
))
2449 istate
->timestamp
.sec
= (unsigned int)st
.st_mtime
;
2450 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
2451 trace_performance_since(start
, "write index, changed mask = %x", istate
->cache_changed
);
2455 void set_alternate_index_output(const char *name
)
2457 alternate_index_output
= name
;
2460 static int commit_locked_index(struct lock_file
*lk
)
2462 if (alternate_index_output
)
2463 return commit_lock_file_to(lk
, alternate_index_output
);
2465 return commit_lock_file(lk
);
2468 static int do_write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
2471 int ret
= do_write_index(istate
, lock
->tempfile
, 0);
2474 if (flags
& COMMIT_LOCK
)
2475 return commit_locked_index(lock
);
2476 return close_lock_file_gently(lock
);
2479 static int write_split_index(struct index_state
*istate
,
2480 struct lock_file
*lock
,
2484 prepare_to_write_split_index(istate
);
2485 ret
= do_write_locked_index(istate
, lock
, flags
);
2486 finish_writing_split_index(istate
);
2490 static const char *shared_index_expire
= "2.weeks.ago";
2492 static unsigned long get_shared_index_expire_date(void)
2494 static unsigned long shared_index_expire_date
;
2495 static int shared_index_expire_date_prepared
;
2497 if (!shared_index_expire_date_prepared
) {
2498 git_config_get_expiry("splitindex.sharedindexexpire",
2499 &shared_index_expire
);
2500 shared_index_expire_date
= approxidate(shared_index_expire
);
2501 shared_index_expire_date_prepared
= 1;
2504 return shared_index_expire_date
;
2507 static int should_delete_shared_index(const char *shared_index_path
)
2510 unsigned long expiration
;
2512 /* Check timestamp */
2513 expiration
= get_shared_index_expire_date();
2516 if (stat(shared_index_path
, &st
))
2517 return error_errno(_("could not stat '%s'"), shared_index_path
);
2518 if (st
.st_mtime
> expiration
)
2524 static int clean_shared_index_files(const char *current_hex
)
2527 DIR *dir
= opendir(get_git_dir());
2530 return error_errno(_("unable to open git dir: %s"), get_git_dir());
2532 while ((de
= readdir(dir
)) != NULL
) {
2533 const char *sha1_hex
;
2534 const char *shared_index_path
;
2535 if (!skip_prefix(de
->d_name
, "sharedindex.", &sha1_hex
))
2537 if (!strcmp(sha1_hex
, current_hex
))
2539 shared_index_path
= git_path("%s", de
->d_name
);
2540 if (should_delete_shared_index(shared_index_path
) > 0 &&
2541 unlink(shared_index_path
))
2542 warning_errno(_("unable to unlink: %s"), shared_index_path
);
2549 static int write_shared_index(struct index_state
*istate
,
2550 struct tempfile
**temp
)
2552 struct split_index
*si
= istate
->split_index
;
2555 move_cache_to_base_index(istate
);
2556 ret
= do_write_index(si
->base
, *temp
, 1);
2559 ret
= adjust_shared_perm(get_tempfile_path(*temp
));
2561 error("cannot fix permission bits on %s", get_tempfile_path(*temp
));
2564 ret
= rename_tempfile(temp
,
2565 git_path("sharedindex.%s", sha1_to_hex(si
->base
->sha1
)));
2567 hashcpy(si
->base_sha1
, si
->base
->sha1
);
2568 clean_shared_index_files(sha1_to_hex(si
->base
->sha1
));
2574 static const int default_max_percent_split_change
= 20;
2576 static int too_many_not_shared_entries(struct index_state
*istate
)
2578 int i
, not_shared
= 0;
2579 int max_split
= git_config_get_max_percent_split_change();
2581 switch (max_split
) {
2583 /* not or badly configured: use the default value */
2584 max_split
= default_max_percent_split_change
;
2587 return 1; /* 0% means always write a new shared index */
2589 return 0; /* 100% means never write a new shared index */
2591 break; /* just use the configured value */
2594 /* Count not shared entries */
2595 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2596 struct cache_entry
*ce
= istate
->cache
[i
];
2601 return (int64_t)istate
->cache_nr
* max_split
< (int64_t)not_shared
* 100;
2604 int write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
2607 int new_shared_index
, ret
;
2608 struct split_index
*si
= istate
->split_index
;
2610 if ((flags
& SKIP_IF_UNCHANGED
) && !istate
->cache_changed
) {
2611 if (flags
& COMMIT_LOCK
)
2612 rollback_lock_file(lock
);
2616 if (istate
->fsmonitor_last_update
)
2617 fill_fsmonitor_bitmap(istate
);
2619 if (!si
|| alternate_index_output
||
2620 (istate
->cache_changed
& ~EXTMASK
)) {
2622 hashclr(si
->base_sha1
);
2623 ret
= do_write_locked_index(istate
, lock
, flags
);
2627 if (getenv("GIT_TEST_SPLIT_INDEX")) {
2628 int v
= si
->base_sha1
[0];
2630 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
2632 if (too_many_not_shared_entries(istate
))
2633 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
2635 new_shared_index
= istate
->cache_changed
& SPLIT_INDEX_ORDERED
;
2637 if (new_shared_index
) {
2638 struct tempfile
*temp
;
2641 temp
= mks_tempfile(git_path("sharedindex_XXXXXX"));
2643 hashclr(si
->base_sha1
);
2644 ret
= do_write_locked_index(istate
, lock
, flags
);
2647 ret
= write_shared_index(istate
, &temp
);
2649 saved_errno
= errno
;
2650 if (is_tempfile_active(temp
))
2651 delete_tempfile(&temp
);
2652 errno
= saved_errno
;
2658 ret
= write_split_index(istate
, lock
, flags
);
2660 /* Freshen the shared index only if the split-index was written */
2661 if (!ret
&& !new_shared_index
) {
2662 const char *shared_index
= git_path("sharedindex.%s",
2663 sha1_to_hex(si
->base_sha1
));
2664 freshen_shared_index(shared_index
, 1);
2668 if (flags
& COMMIT_LOCK
)
2669 rollback_lock_file(lock
);
2674 * Read the index file that is potentially unmerged into given
2675 * index_state, dropping any unmerged entries. Returns true if
2676 * the index is unmerged. Callers who want to refuse to work
2677 * from an unmerged state can call this and check its return value,
2678 * instead of calling read_cache().
2680 int read_index_unmerged(struct index_state
*istate
)
2686 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2687 struct cache_entry
*ce
= istate
->cache
[i
];
2688 struct cache_entry
*new_ce
;
2694 len
= ce_namelen(ce
);
2695 size
= cache_entry_size(len
);
2696 new_ce
= xcalloc(1, size
);
2697 memcpy(new_ce
->name
, ce
->name
, len
);
2698 new_ce
->ce_flags
= create_ce_flags(0) | CE_CONFLICTED
;
2699 new_ce
->ce_namelen
= len
;
2700 new_ce
->ce_mode
= ce
->ce_mode
;
2701 if (add_index_entry(istate
, new_ce
, 0))
2702 return error("%s: cannot drop to stage #0",
2709 * Returns 1 if the path is an "other" path with respect to
2710 * the index; that is, the path is not mentioned in the index at all,
2711 * either as a file, a directory with some files in the index,
2712 * or as an unmerged entry.
2714 * We helpfully remove a trailing "/" from directories so that
2715 * the output of read_directory can be used as-is.
2717 int index_name_is_other(const struct index_state
*istate
, const char *name
,
2721 if (namelen
&& name
[namelen
- 1] == '/')
2723 pos
= index_name_pos(istate
, name
, namelen
);
2725 return 0; /* exact match */
2727 if (pos
< istate
->cache_nr
) {
2728 struct cache_entry
*ce
= istate
->cache
[pos
];
2729 if (ce_namelen(ce
) == namelen
&&
2730 !memcmp(ce
->name
, name
, namelen
))
2731 return 0; /* Yup, this one exists unmerged */
2736 void *read_blob_data_from_index(const struct index_state
*istate
,
2737 const char *path
, unsigned long *size
)
2741 enum object_type type
;
2745 pos
= index_name_pos(istate
, path
, len
);
2748 * We might be in the middle of a merge, in which
2749 * case we would read stage #2 (ours).
2753 (pos
< 0 && i
< istate
->cache_nr
&&
2754 !strcmp(istate
->cache
[i
]->name
, path
));
2756 if (ce_stage(istate
->cache
[i
]) == 2)
2761 data
= read_sha1_file(istate
->cache
[pos
]->oid
.hash
, &type
, &sz
);
2762 if (!data
|| type
!= OBJ_BLOB
) {
2771 void stat_validity_clear(struct stat_validity
*sv
)
2773 FREE_AND_NULL(sv
->sd
);
2776 int stat_validity_check(struct stat_validity
*sv
, const char *path
)
2780 if (stat(path
, &st
) < 0)
2781 return sv
->sd
== NULL
;
2784 return S_ISREG(st
.st_mode
) && !match_stat_data(sv
->sd
, &st
);
2787 void stat_validity_update(struct stat_validity
*sv
, int fd
)
2791 if (fstat(fd
, &st
) < 0 || !S_ISREG(st
.st_mode
))
2792 stat_validity_clear(sv
);
2795 sv
->sd
= xcalloc(1, sizeof(struct stat_data
));
2796 fill_stat_data(sv
->sd
, &st
);
2800 void move_index_extensions(struct index_state
*dst
, struct index_state
*src
)
2802 dst
->untracked
= src
->untracked
;
2803 src
->untracked
= NULL
;