2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 #define NO_THE_INDEX_COMPATIBILITY_MACROS
8 #include "cache-tree.h"
14 * The first letter should be 'A'..'Z' for extensions that are not
15 * necessary for a correct operation (i.e. optimization data).
16 * When new extensions are added that _needs_ to be understood in
17 * order to correctly interpret the index file, pick character that
18 * is outside the range, to cause the reader to abort.
21 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
22 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
24 struct index_state the_index
;
26 static unsigned int hash_name(const char *name
, int namelen
)
28 unsigned int hash
= 0x123;
31 unsigned char c
= *name
++;
37 static void hash_index_entry(struct index_state
*istate
, struct cache_entry
*ce
)
42 if (ce
->ce_flags
& CE_HASHED
)
44 ce
->ce_flags
|= CE_HASHED
;
46 hash
= hash_name(ce
->name
, ce_namelen(ce
));
47 pos
= insert_hash(hash
, ce
, &istate
->name_hash
);
54 static void lazy_init_name_hash(struct index_state
*istate
)
58 if (istate
->name_hash_initialized
)
60 for (nr
= 0; nr
< istate
->cache_nr
; nr
++)
61 hash_index_entry(istate
, istate
->cache
[nr
]);
62 istate
->name_hash_initialized
= 1;
65 static void set_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
67 ce
->ce_flags
&= ~CE_UNHASHED
;
68 istate
->cache
[nr
] = ce
;
69 if (istate
->name_hash_initialized
)
70 hash_index_entry(istate
, ce
);
73 static void replace_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
75 struct cache_entry
*old
= istate
->cache
[nr
];
77 remove_index_entry(old
);
78 set_index_entry(istate
, nr
, ce
);
79 istate
->cache_changed
= 1;
82 int index_name_exists(struct index_state
*istate
, const char *name
, int namelen
)
84 unsigned int hash
= hash_name(name
, namelen
);
85 struct cache_entry
*ce
;
87 lazy_init_name_hash(istate
);
88 ce
= lookup_hash(hash
, &istate
->name_hash
);
91 if (!(ce
->ce_flags
& CE_UNHASHED
)) {
92 if (!cache_name_compare(name
, namelen
, ce
->name
, ce
->ce_flags
))
101 * This only updates the "non-critical" parts of the directory
102 * cache, ie the parts that aren't tracked by GIT, and only used
103 * to validate the cache.
105 void fill_stat_cache_info(struct cache_entry
*ce
, struct stat
*st
)
107 ce
->ce_ctime
= st
->st_ctime
;
108 ce
->ce_mtime
= st
->st_mtime
;
109 ce
->ce_dev
= st
->st_dev
;
110 ce
->ce_ino
= st
->st_ino
;
111 ce
->ce_uid
= st
->st_uid
;
112 ce
->ce_gid
= st
->st_gid
;
113 ce
->ce_size
= st
->st_size
;
115 if (assume_unchanged
)
116 ce
->ce_flags
|= CE_VALID
;
118 if (S_ISREG(st
->st_mode
))
119 ce_mark_uptodate(ce
);
122 static int ce_compare_data(struct cache_entry
*ce
, struct stat
*st
)
125 int fd
= open(ce
->name
, O_RDONLY
);
128 unsigned char sha1
[20];
129 if (!index_fd(sha1
, fd
, st
, 0, OBJ_BLOB
, ce
->name
))
130 match
= hashcmp(sha1
, ce
->sha1
);
131 /* index_fd() closed the file descriptor already */
136 static int ce_compare_link(struct cache_entry
*ce
, size_t expected_size
)
142 enum object_type type
;
145 target
= xmalloc(expected_size
);
146 len
= readlink(ce
->name
, target
, expected_size
);
147 if (len
!= expected_size
) {
151 buffer
= read_sha1_file(ce
->sha1
, &type
, &size
);
156 if (size
== expected_size
)
157 match
= memcmp(buffer
, target
, size
);
163 static int ce_compare_gitlink(struct cache_entry
*ce
)
165 unsigned char sha1
[20];
168 * We don't actually require that the .git directory
169 * under GITLINK directory be a valid git directory. It
170 * might even be missing (in case nobody populated that
173 * If so, we consider it always to match.
175 if (resolve_gitlink_ref(ce
->name
, "HEAD", sha1
) < 0)
177 return hashcmp(sha1
, ce
->sha1
);
180 static int ce_modified_check_fs(struct cache_entry
*ce
, struct stat
*st
)
182 switch (st
->st_mode
& S_IFMT
) {
184 if (ce_compare_data(ce
, st
))
188 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
192 if (S_ISGITLINK(ce
->ce_mode
))
200 static int ce_match_stat_basic(struct cache_entry
*ce
, struct stat
*st
)
202 unsigned int changed
= 0;
204 if (ce
->ce_flags
& CE_REMOVE
)
205 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
207 switch (ce
->ce_mode
& S_IFMT
) {
209 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
210 /* We consider only the owner x bit to be relevant for
213 if (trust_executable_bit
&&
214 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
215 changed
|= MODE_CHANGED
;
218 if (!S_ISLNK(st
->st_mode
) &&
219 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
220 changed
|= TYPE_CHANGED
;
223 if (!S_ISDIR(st
->st_mode
))
224 changed
|= TYPE_CHANGED
;
225 else if (ce_compare_gitlink(ce
))
226 changed
|= DATA_CHANGED
;
229 die("internal error: ce_mode is %o", ce
->ce_mode
);
231 if (ce
->ce_mtime
!= (unsigned int) st
->st_mtime
)
232 changed
|= MTIME_CHANGED
;
233 if (ce
->ce_ctime
!= (unsigned int) st
->st_ctime
)
234 changed
|= CTIME_CHANGED
;
236 if (ce
->ce_uid
!= (unsigned int) st
->st_uid
||
237 ce
->ce_gid
!= (unsigned int) st
->st_gid
)
238 changed
|= OWNER_CHANGED
;
239 if (ce
->ce_ino
!= (unsigned int) st
->st_ino
)
240 changed
|= INODE_CHANGED
;
244 * st_dev breaks on network filesystems where different
245 * clients will have different views of what "device"
246 * the filesystem is on
248 if (ce
->ce_dev
!= (unsigned int) st
->st_dev
)
249 changed
|= INODE_CHANGED
;
252 if (ce
->ce_size
!= (unsigned int) st
->st_size
)
253 changed
|= DATA_CHANGED
;
258 static int is_racy_timestamp(const struct index_state
*istate
, struct cache_entry
*ce
)
260 return (!S_ISGITLINK(ce
->ce_mode
) &&
262 ((unsigned int)istate
->timestamp
) <= ce
->ce_mtime
);
265 int ie_match_stat(const struct index_state
*istate
,
266 struct cache_entry
*ce
, struct stat
*st
,
267 unsigned int options
)
269 unsigned int changed
;
270 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
271 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
274 * If it's marked as always valid in the index, it's
275 * valid whatever the checked-out copy says.
277 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
280 changed
= ce_match_stat_basic(ce
, st
);
283 * Within 1 second of this sequence:
284 * echo xyzzy >file && git-update-index --add file
285 * running this command:
287 * would give a falsely clean cache entry. The mtime and
288 * length match the cache, and other stat fields do not change.
290 * We could detect this at update-index time (the cache entry
291 * being registered/updated records the same time as "now")
292 * and delay the return from git-update-index, but that would
293 * effectively mean we can make at most one commit per second,
294 * which is not acceptable. Instead, we check cache entries
295 * whose mtime are the same as the index file timestamp more
296 * carefully than others.
298 if (!changed
&& is_racy_timestamp(istate
, ce
)) {
299 if (assume_racy_is_modified
)
300 changed
|= DATA_CHANGED
;
302 changed
|= ce_modified_check_fs(ce
, st
);
308 int ie_modified(const struct index_state
*istate
,
309 struct cache_entry
*ce
, struct stat
*st
, unsigned int options
)
311 int changed
, changed_fs
;
313 changed
= ie_match_stat(istate
, ce
, st
, options
);
317 * If the mode or type has changed, there's no point in trying
318 * to refresh the entry - it's not going to match
320 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
323 /* Immediately after read-tree or update-index --cacheinfo,
324 * the length field is zero. For other cases the ce_size
325 * should match the SHA1 recorded in the index entry.
327 if ((changed
& DATA_CHANGED
) && ce
->ce_size
!= 0)
330 changed_fs
= ce_modified_check_fs(ce
, st
);
332 return changed
| changed_fs
;
336 int base_name_compare(const char *name1
, int len1
, int mode1
,
337 const char *name2
, int len2
, int mode2
)
339 unsigned char c1
, c2
;
340 int len
= len1
< len2
? len1
: len2
;
343 cmp
= memcmp(name1
, name2
, len
);
348 if (!c1
&& S_ISDIR(mode1
))
350 if (!c2
&& S_ISDIR(mode2
))
352 return (c1
< c2
) ? -1 : (c1
> c2
) ? 1 : 0;
356 * df_name_compare() is identical to base_name_compare(), except it
357 * compares conflicting directory/file entries as equal. Note that
358 * while a directory name compares as equal to a regular file, they
359 * then individually compare _differently_ to a filename that has
360 * a dot after the basename (because '\0' < '.' < '/').
362 * This is used by routines that want to traverse the git namespace
363 * but then handle conflicting entries together when possible.
365 int df_name_compare(const char *name1
, int len1
, int mode1
,
366 const char *name2
, int len2
, int mode2
)
368 int len
= len1
< len2
? len1
: len2
, cmp
;
369 unsigned char c1
, c2
;
371 cmp
= memcmp(name1
, name2
, len
);
374 /* Directories and files compare equal (same length, same name) */
378 if (!c1
&& S_ISDIR(mode1
))
381 if (!c2
&& S_ISDIR(mode2
))
383 if (c1
== '/' && !c2
)
385 if (c2
== '/' && !c1
)
390 int cache_name_compare(const char *name1
, int flags1
, const char *name2
, int flags2
)
392 int len1
= flags1
& CE_NAMEMASK
;
393 int len2
= flags2
& CE_NAMEMASK
;
394 int len
= len1
< len2
? len1
: len2
;
397 cmp
= memcmp(name1
, name2
, len
);
406 flags1
&= CE_STAGEMASK
;
407 flags2
&= CE_STAGEMASK
;
416 int index_name_pos(const struct index_state
*istate
, const char *name
, int namelen
)
421 last
= istate
->cache_nr
;
422 while (last
> first
) {
423 int next
= (last
+ first
) >> 1;
424 struct cache_entry
*ce
= istate
->cache
[next
];
425 int cmp
= cache_name_compare(name
, namelen
, ce
->name
, ce
->ce_flags
);
437 /* Remove entry, return true if there are more entries to go.. */
438 int remove_index_entry_at(struct index_state
*istate
, int pos
)
440 struct cache_entry
*ce
= istate
->cache
[pos
];
442 remove_index_entry(ce
);
443 istate
->cache_changed
= 1;
445 if (pos
>= istate
->cache_nr
)
447 memmove(istate
->cache
+ pos
,
448 istate
->cache
+ pos
+ 1,
449 (istate
->cache_nr
- pos
) * sizeof(struct cache_entry
*));
453 int remove_file_from_index(struct index_state
*istate
, const char *path
)
455 int pos
= index_name_pos(istate
, path
, strlen(path
));
458 cache_tree_invalidate_path(istate
->cache_tree
, path
);
459 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
460 remove_index_entry_at(istate
, pos
);
464 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
466 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
469 static int index_name_pos_also_unmerged(struct index_state
*istate
,
470 const char *path
, int namelen
)
472 int pos
= index_name_pos(istate
, path
, namelen
);
473 struct cache_entry
*ce
;
478 /* maybe unmerged? */
480 if (pos
>= istate
->cache_nr
||
481 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
484 /* order of preference: stage 2, 1, 3 */
485 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
486 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
487 !compare_name(ce
, path
, namelen
))
492 int add_file_to_index(struct index_state
*istate
, const char *path
, int verbose
)
494 int size
, namelen
, pos
;
496 struct cache_entry
*ce
;
497 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_RACY_IS_DIRTY
;
499 if (lstat(path
, &st
))
500 die("%s: unable to stat (%s)", path
, strerror(errno
));
502 if (!S_ISREG(st
.st_mode
) && !S_ISLNK(st
.st_mode
) && !S_ISDIR(st
.st_mode
))
503 die("%s: can only add regular files, symbolic links or git-directories", path
);
505 namelen
= strlen(path
);
506 if (S_ISDIR(st
.st_mode
)) {
507 while (namelen
&& path
[namelen
-1] == '/')
510 size
= cache_entry_size(namelen
);
511 ce
= xcalloc(1, size
);
512 memcpy(ce
->name
, path
, namelen
);
513 ce
->ce_flags
= namelen
;
514 fill_stat_cache_info(ce
, &st
);
516 if (trust_executable_bit
&& has_symlinks
)
517 ce
->ce_mode
= create_ce_mode(st
.st_mode
);
519 /* If there is an existing entry, pick the mode bits and type
520 * from it, otherwise assume unexecutable regular file.
522 struct cache_entry
*ent
;
523 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
525 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
526 ce
->ce_mode
= ce_mode_from_stat(ent
, st
.st_mode
);
529 pos
= index_name_pos(istate
, ce
->name
, namelen
);
531 !ce_stage(istate
->cache
[pos
]) &&
532 !ie_match_stat(istate
, istate
->cache
[pos
], &st
, ce_option
)) {
533 /* Nothing changed, really */
535 ce_mark_uptodate(istate
->cache
[pos
]);
539 if (index_path(ce
->sha1
, path
, &st
, 1))
540 die("unable to index file %s", path
);
541 if (add_index_entry(istate
, ce
, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
))
542 die("unable to add %s to index",path
);
544 printf("add '%s'\n", path
);
548 struct cache_entry
*make_cache_entry(unsigned int mode
,
549 const unsigned char *sha1
, const char *path
, int stage
,
553 struct cache_entry
*ce
;
555 if (!verify_path(path
))
559 size
= cache_entry_size(len
);
560 ce
= xcalloc(1, size
);
562 hashcpy(ce
->sha1
, sha1
);
563 memcpy(ce
->name
, path
, len
);
564 ce
->ce_flags
= create_ce_flags(len
, stage
);
565 ce
->ce_mode
= create_ce_mode(mode
);
568 return refresh_cache_entry(ce
, 0);
573 int ce_same_name(struct cache_entry
*a
, struct cache_entry
*b
)
575 int len
= ce_namelen(a
);
576 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
579 int ce_path_match(const struct cache_entry
*ce
, const char **pathspec
)
581 const char *match
, *name
;
587 len
= ce_namelen(ce
);
589 while ((match
= *pathspec
++) != NULL
) {
590 int matchlen
= strlen(match
);
593 if (memcmp(name
, match
, matchlen
))
595 if (matchlen
&& name
[matchlen
-1] == '/')
597 if (name
[matchlen
] == '/' || !name
[matchlen
])
606 * We fundamentally don't like some paths: we don't want
607 * dot or dot-dot anywhere, and for obvious reasons don't
608 * want to recurse into ".git" either.
610 * Also, we don't want double slashes or slashes at the
611 * end that can make pathnames ambiguous.
613 static int verify_dotfile(const char *rest
)
616 * The first character was '.', but that
617 * has already been discarded, we now test
621 /* "." is not allowed */
626 * ".git" followed by NUL or slash is bad. This
627 * shares the path end test with the ".." case.
637 if (rest
[1] == '\0' || rest
[1] == '/')
643 int verify_path(const char *path
)
660 if (verify_dotfile(path
))
670 * Do we have another file that has the beginning components being a
671 * proper superset of the name we're trying to add?
673 static int has_file_name(struct index_state
*istate
,
674 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
677 int len
= ce_namelen(ce
);
678 int stage
= ce_stage(ce
);
679 const char *name
= ce
->name
;
681 while (pos
< istate
->cache_nr
) {
682 struct cache_entry
*p
= istate
->cache
[pos
++];
684 if (len
>= ce_namelen(p
))
686 if (memcmp(name
, p
->name
, len
))
688 if (ce_stage(p
) != stage
)
690 if (p
->name
[len
] != '/')
692 if (p
->ce_flags
& CE_REMOVE
)
697 remove_index_entry_at(istate
, --pos
);
703 * Do we have another file with a pathname that is a proper
704 * subset of the name we're trying to add?
706 static int has_dir_name(struct index_state
*istate
,
707 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
710 int stage
= ce_stage(ce
);
711 const char *name
= ce
->name
;
712 const char *slash
= name
+ ce_namelen(ce
);
720 if (slash
<= ce
->name
)
725 pos
= index_name_pos(istate
, name
, create_ce_flags(len
, stage
));
728 * Found one, but not so fast. This could
729 * be a marker that says "I was here, but
730 * I am being removed". Such an entry is
731 * not a part of the resulting tree, and
732 * it is Ok to have a directory at the same
735 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
739 remove_index_entry_at(istate
, pos
);
747 * Trivial optimization: if we find an entry that
748 * already matches the sub-directory, then we know
749 * we're ok, and we can exit.
751 while (pos
< istate
->cache_nr
) {
752 struct cache_entry
*p
= istate
->cache
[pos
];
753 if ((ce_namelen(p
) <= len
) ||
754 (p
->name
[len
] != '/') ||
755 memcmp(p
->name
, name
, len
))
756 break; /* not our subdirectory */
757 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
759 * p is at the same stage as our entry, and
760 * is a subdirectory of what we are looking
761 * at, so we cannot have conflicts at our
762 * level or anything shorter.
771 /* We may be in a situation where we already have path/file and path
772 * is being added, or we already have path and path/file is being
773 * added. Either one would result in a nonsense tree that has path
774 * twice when git-write-tree tries to write it out. Prevent it.
776 * If ok-to-replace is specified, we remove the conflicting entries
777 * from the cache so the caller should recompute the insert position.
778 * When this happens, we return non-zero.
780 static int check_file_directory_conflict(struct index_state
*istate
,
781 const struct cache_entry
*ce
,
782 int pos
, int ok_to_replace
)
787 * When ce is an "I am going away" entry, we allow it to be added
789 if (ce
->ce_flags
& CE_REMOVE
)
793 * We check if the path is a sub-path of a subsequent pathname
794 * first, since removing those will not change the position
797 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
800 * Then check if the path might have a clashing sub-directory
803 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
806 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
809 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
810 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
811 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
813 cache_tree_invalidate_path(istate
->cache_tree
, ce
->name
);
814 pos
= index_name_pos(istate
, ce
->name
, ce
->ce_flags
);
816 /* existing match? Just replace it. */
818 replace_index_entry(istate
, pos
, ce
);
824 * Inserting a merged entry ("stage 0") into the index
825 * will always replace all non-merged entries..
827 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
828 while (ce_same_name(istate
->cache
[pos
], ce
)) {
830 if (!remove_index_entry_at(istate
, pos
))
837 if (!verify_path(ce
->name
))
840 if (!skip_df_check
&&
841 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
843 return error("'%s' appears as both a file and as a directory",
845 pos
= index_name_pos(istate
, ce
->name
, ce
->ce_flags
);
851 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
855 if (option
& ADD_CACHE_JUST_APPEND
)
856 pos
= istate
->cache_nr
;
859 ret
= add_index_entry_with_check(istate
, ce
, option
);
865 /* Make sure the array is big enough .. */
866 if (istate
->cache_nr
== istate
->cache_alloc
) {
867 istate
->cache_alloc
= alloc_nr(istate
->cache_alloc
);
868 istate
->cache
= xrealloc(istate
->cache
,
869 istate
->cache_alloc
* sizeof(struct cache_entry
*));
874 if (istate
->cache_nr
> pos
+ 1)
875 memmove(istate
->cache
+ pos
+ 1,
877 (istate
->cache_nr
- pos
- 1) * sizeof(ce
));
878 set_index_entry(istate
, pos
, ce
);
879 istate
->cache_changed
= 1;
884 * "refresh" does not calculate a new sha1 file or bring the
885 * cache up-to-date for mode/content changes. But what it
886 * _does_ do is to "re-match" the stat information of a file
887 * with the cache, so that you can refresh the cache for a
888 * file that hasn't been changed but where the stat entry is
891 * For example, you'd want to do this after doing a "git-read-tree",
892 * to link up the stat cache details with the proper files.
894 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
895 struct cache_entry
*ce
,
896 unsigned int options
, int *err
)
899 struct cache_entry
*updated
;
901 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
906 if (lstat(ce
->name
, &st
) < 0) {
912 changed
= ie_match_stat(istate
, ce
, &st
, options
);
915 * The path is unchanged. If we were told to ignore
916 * valid bit, then we did the actual stat check and
917 * found that the entry is unmodified. If the entry
918 * is not marked VALID, this is the place to mark it
919 * valid again, under "assume unchanged" mode.
921 if (ignore_valid
&& assume_unchanged
&&
922 !(ce
->ce_flags
& CE_VALID
))
923 ; /* mark this one VALID again */
926 * We do not mark the index itself "modified"
927 * because CE_UPTODATE flag is in-core only;
928 * we are not going to write this change out.
930 ce_mark_uptodate(ce
);
935 if (ie_modified(istate
, ce
, &st
, options
)) {
942 updated
= xmalloc(size
);
943 memcpy(updated
, ce
, size
);
944 fill_stat_cache_info(updated
, &st
);
946 * If ignore_valid is not set, we should leave CE_VALID bit
947 * alone. Otherwise, paths marked with --no-assume-unchanged
948 * (i.e. things to be edited) will reacquire CE_VALID bit
949 * automatically, which is not really what we want.
951 if (!ignore_valid
&& assume_unchanged
&&
952 !(ce
->ce_flags
& CE_VALID
))
953 updated
->ce_flags
&= ~CE_VALID
;
958 int refresh_index(struct index_state
*istate
, unsigned int flags
, const char **pathspec
, char *seen
)
962 int really
= (flags
& REFRESH_REALLY
) != 0;
963 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
964 int quiet
= (flags
& REFRESH_QUIET
) != 0;
965 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
966 unsigned int options
= really
? CE_MATCH_IGNORE_VALID
: 0;
968 for (i
= 0; i
< istate
->cache_nr
; i
++) {
969 struct cache_entry
*ce
, *new;
972 ce
= istate
->cache
[i
];
974 while ((i
< istate
->cache_nr
) &&
975 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
980 printf("%s: needs merge\n", ce
->name
);
985 if (pathspec
&& !match_pathspec(pathspec
, ce
->name
, strlen(ce
->name
), 0, seen
))
988 new = refresh_cache_ent(istate
, ce
, options
, &cache_errno
);
992 if (not_new
&& cache_errno
== ENOENT
)
994 if (really
&& cache_errno
== EINVAL
) {
995 /* If we are doing --really-refresh that
996 * means the index is not valid anymore.
998 ce
->ce_flags
&= ~CE_VALID
;
999 istate
->cache_changed
= 1;
1003 printf("%s: needs update\n", ce
->name
);
1008 replace_index_entry(istate
, i
, new);
1013 struct cache_entry
*refresh_cache_entry(struct cache_entry
*ce
, int really
)
1015 return refresh_cache_ent(&the_index
, ce
, really
, NULL
);
1018 static int verify_hdr(struct cache_header
*hdr
, unsigned long size
)
1021 unsigned char sha1
[20];
1023 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
1024 return error("bad signature");
1025 if (hdr
->hdr_version
!= htonl(2))
1026 return error("bad index version");
1028 SHA1_Update(&c
, hdr
, size
- 20);
1029 SHA1_Final(sha1
, &c
);
1030 if (hashcmp(sha1
, (unsigned char *)hdr
+ size
- 20))
1031 return error("bad index file sha1 signature");
1035 static int read_index_extension(struct index_state
*istate
,
1036 const char *ext
, void *data
, unsigned long sz
)
1038 switch (CACHE_EXT(ext
)) {
1039 case CACHE_EXT_TREE
:
1040 istate
->cache_tree
= cache_tree_read(data
, sz
);
1043 if (*ext
< 'A' || 'Z' < *ext
)
1044 return error("index uses %.4s extension, which we do not understand",
1046 fprintf(stderr
, "ignoring %.4s extension\n", ext
);
1052 int read_index(struct index_state
*istate
)
1054 return read_index_from(istate
, get_index_file());
1057 static void convert_from_disk(struct ondisk_cache_entry
*ondisk
, struct cache_entry
*ce
)
1061 ce
->ce_ctime
= ntohl(ondisk
->ctime
.sec
);
1062 ce
->ce_mtime
= ntohl(ondisk
->mtime
.sec
);
1063 ce
->ce_dev
= ntohl(ondisk
->dev
);
1064 ce
->ce_ino
= ntohl(ondisk
->ino
);
1065 ce
->ce_mode
= ntohl(ondisk
->mode
);
1066 ce
->ce_uid
= ntohl(ondisk
->uid
);
1067 ce
->ce_gid
= ntohl(ondisk
->gid
);
1068 ce
->ce_size
= ntohl(ondisk
->size
);
1069 /* On-disk flags are just 16 bits */
1070 ce
->ce_flags
= ntohs(ondisk
->flags
);
1071 hashcpy(ce
->sha1
, ondisk
->sha1
);
1073 len
= ce
->ce_flags
& CE_NAMEMASK
;
1074 if (len
== CE_NAMEMASK
)
1075 len
= strlen(ondisk
->name
);
1077 * NEEDSWORK: If the original index is crafted, this copy could
1080 memcpy(ce
->name
, ondisk
->name
, len
+ 1);
1083 static inline size_t estimate_cache_size(size_t ondisk_size
, unsigned int entries
)
1087 per_entry
= sizeof(struct cache_entry
) - sizeof(struct ondisk_cache_entry
);
1090 * Alignment can cause differences. This should be "alignof", but
1091 * since that's a gcc'ism, just use the size of a pointer.
1093 per_entry
+= sizeof(void *);
1094 return ondisk_size
+ entries
*per_entry
;
1097 /* remember to discard_cache() before reading a different cache! */
1098 int read_index_from(struct index_state
*istate
, const char *path
)
1102 unsigned long src_offset
, dst_offset
;
1103 struct cache_header
*hdr
;
1109 return istate
->cache_nr
;
1112 istate
->timestamp
= 0;
1113 fd
= open(path
, O_RDONLY
);
1115 if (errno
== ENOENT
)
1117 die("index file open failed (%s)", strerror(errno
));
1121 die("cannot stat the open index (%s)", strerror(errno
));
1124 mmap_size
= xsize_t(st
.st_size
);
1125 if (mmap_size
< sizeof(struct cache_header
) + 20)
1126 die("index file smaller than expected");
1128 mmap
= xmmap(NULL
, mmap_size
, PROT_READ
| PROT_WRITE
, MAP_PRIVATE
, fd
, 0);
1130 if (mmap
== MAP_FAILED
)
1131 die("unable to map index file");
1134 if (verify_hdr(hdr
, mmap_size
) < 0)
1137 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
1138 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
1139 istate
->cache
= xcalloc(istate
->cache_alloc
, sizeof(struct cache_entry
*));
1142 * The disk format is actually larger than the in-memory format,
1143 * due to space for nsec etc, so even though the in-memory one
1144 * has room for a few more flags, we can allocate using the same
1147 istate
->alloc
= xmalloc(estimate_cache_size(mmap_size
, istate
->cache_nr
));
1149 src_offset
= sizeof(*hdr
);
1151 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1152 struct ondisk_cache_entry
*disk_ce
;
1153 struct cache_entry
*ce
;
1155 disk_ce
= (struct ondisk_cache_entry
*)((char *)mmap
+ src_offset
);
1156 ce
= (struct cache_entry
*)((char *)istate
->alloc
+ dst_offset
);
1157 convert_from_disk(disk_ce
, ce
);
1158 set_index_entry(istate
, i
, ce
);
1160 src_offset
+= ondisk_ce_size(ce
);
1161 dst_offset
+= ce_size(ce
);
1163 istate
->timestamp
= st
.st_mtime
;
1164 while (src_offset
<= mmap_size
- 20 - 8) {
1165 /* After an array of active_nr index entries,
1166 * there can be arbitrary number of extended
1167 * sections, each of which is prefixed with
1168 * extension name (4-byte) and section length
1169 * in 4-byte network byte order.
1171 unsigned long extsize
;
1172 memcpy(&extsize
, (char *)mmap
+ src_offset
+ 4, 4);
1173 extsize
= ntohl(extsize
);
1174 if (read_index_extension(istate
,
1175 (const char *) mmap
+ src_offset
,
1176 (char *) mmap
+ src_offset
+ 8,
1180 src_offset
+= extsize
;
1182 munmap(mmap
, mmap_size
);
1183 return istate
->cache_nr
;
1186 munmap(mmap
, mmap_size
);
1188 die("index file corrupt");
1191 int discard_index(struct index_state
*istate
)
1193 istate
->cache_nr
= 0;
1194 istate
->cache_changed
= 0;
1195 istate
->timestamp
= 0;
1196 free_hash(&istate
->name_hash
);
1197 cache_tree_free(&(istate
->cache_tree
));
1198 free(istate
->alloc
);
1199 istate
->alloc
= NULL
;
1201 /* no need to throw away allocated active_cache */
1205 int unmerged_index(const struct index_state
*istate
)
1208 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1209 if (ce_stage(istate
->cache
[i
]))
1215 #define WRITE_BUFFER_SIZE 8192
1216 static unsigned char write_buffer
[WRITE_BUFFER_SIZE
];
1217 static unsigned long write_buffer_len
;
1219 static int ce_write_flush(SHA_CTX
*context
, int fd
)
1221 unsigned int buffered
= write_buffer_len
;
1223 SHA1_Update(context
, write_buffer
, buffered
);
1224 if (write_in_full(fd
, write_buffer
, buffered
) != buffered
)
1226 write_buffer_len
= 0;
1231 static int ce_write(SHA_CTX
*context
, int fd
, void *data
, unsigned int len
)
1234 unsigned int buffered
= write_buffer_len
;
1235 unsigned int partial
= WRITE_BUFFER_SIZE
- buffered
;
1238 memcpy(write_buffer
+ buffered
, data
, partial
);
1239 buffered
+= partial
;
1240 if (buffered
== WRITE_BUFFER_SIZE
) {
1241 write_buffer_len
= buffered
;
1242 if (ce_write_flush(context
, fd
))
1246 write_buffer_len
= buffered
;
1248 data
= (char *) data
+ partial
;
1253 static int write_index_ext_header(SHA_CTX
*context
, int fd
,
1254 unsigned int ext
, unsigned int sz
)
1258 return ((ce_write(context
, fd
, &ext
, 4) < 0) ||
1259 (ce_write(context
, fd
, &sz
, 4) < 0)) ? -1 : 0;
1262 static int ce_flush(SHA_CTX
*context
, int fd
)
1264 unsigned int left
= write_buffer_len
;
1267 write_buffer_len
= 0;
1268 SHA1_Update(context
, write_buffer
, left
);
1271 /* Flush first if not enough space for SHA1 signature */
1272 if (left
+ 20 > WRITE_BUFFER_SIZE
) {
1273 if (write_in_full(fd
, write_buffer
, left
) != left
)
1278 /* Append the SHA1 signature at the end */
1279 SHA1_Final(write_buffer
+ left
, context
);
1281 return (write_in_full(fd
, write_buffer
, left
) != left
) ? -1 : 0;
1284 static void ce_smudge_racily_clean_entry(struct cache_entry
*ce
)
1287 * The only thing we care about in this function is to smudge the
1288 * falsely clean entry due to touch-update-touch race, so we leave
1289 * everything else as they are. We are called for entries whose
1290 * ce_mtime match the index file mtime.
1294 if (lstat(ce
->name
, &st
) < 0)
1296 if (ce_match_stat_basic(ce
, &st
))
1298 if (ce_modified_check_fs(ce
, &st
)) {
1299 /* This is "racily clean"; smudge it. Note that this
1300 * is a tricky code. At first glance, it may appear
1301 * that it can break with this sequence:
1303 * $ echo xyzzy >frotz
1304 * $ git-update-index --add frotz
1307 * $ echo filfre >nitfol
1308 * $ git-update-index --add nitfol
1310 * but it does not. When the second update-index runs,
1311 * it notices that the entry "frotz" has the same timestamp
1312 * as index, and if we were to smudge it by resetting its
1313 * size to zero here, then the object name recorded
1314 * in index is the 6-byte file but the cached stat information
1315 * becomes zero --- which would then match what we would
1316 * obtain from the filesystem next time we stat("frotz").
1318 * However, the second update-index, before calling
1319 * this function, notices that the cached size is 6
1320 * bytes and what is on the filesystem is an empty
1321 * file, and never calls us, so the cached size information
1322 * for "frotz" stays 6 which does not match the filesystem.
1328 static int ce_write_entry(SHA_CTX
*c
, int fd
, struct cache_entry
*ce
)
1330 int size
= ondisk_ce_size(ce
);
1331 struct ondisk_cache_entry
*ondisk
= xcalloc(1, size
);
1333 ondisk
->ctime
.sec
= htonl(ce
->ce_ctime
);
1334 ondisk
->ctime
.nsec
= 0;
1335 ondisk
->mtime
.sec
= htonl(ce
->ce_mtime
);
1336 ondisk
->mtime
.nsec
= 0;
1337 ondisk
->dev
= htonl(ce
->ce_dev
);
1338 ondisk
->ino
= htonl(ce
->ce_ino
);
1339 ondisk
->mode
= htonl(ce
->ce_mode
);
1340 ondisk
->uid
= htonl(ce
->ce_uid
);
1341 ondisk
->gid
= htonl(ce
->ce_gid
);
1342 ondisk
->size
= htonl(ce
->ce_size
);
1343 hashcpy(ondisk
->sha1
, ce
->sha1
);
1344 ondisk
->flags
= htons(ce
->ce_flags
);
1345 memcpy(ondisk
->name
, ce
->name
, ce_namelen(ce
));
1347 return ce_write(c
, fd
, ondisk
, size
);
1350 int write_index(const struct index_state
*istate
, int newfd
)
1353 struct cache_header hdr
;
1354 int i
, err
, removed
;
1355 struct cache_entry
**cache
= istate
->cache
;
1356 int entries
= istate
->cache_nr
;
1358 for (i
= removed
= 0; i
< entries
; i
++)
1359 if (cache
[i
]->ce_flags
& CE_REMOVE
)
1362 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
1363 hdr
.hdr_version
= htonl(2);
1364 hdr
.hdr_entries
= htonl(entries
- removed
);
1367 if (ce_write(&c
, newfd
, &hdr
, sizeof(hdr
)) < 0)
1370 for (i
= 0; i
< entries
; i
++) {
1371 struct cache_entry
*ce
= cache
[i
];
1372 if (ce
->ce_flags
& CE_REMOVE
)
1374 if (is_racy_timestamp(istate
, ce
))
1375 ce_smudge_racily_clean_entry(ce
);
1376 if (ce_write_entry(&c
, newfd
, ce
) < 0)
1380 /* Write extension data here */
1381 if (istate
->cache_tree
) {
1384 strbuf_init(&sb
, 0);
1385 cache_tree_write(&sb
, istate
->cache_tree
);
1386 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_TREE
, sb
.len
) < 0
1387 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
1388 strbuf_release(&sb
);
1392 return ce_flush(&c
, newfd
);