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
)
40 unsigned int hash
= hash_name(ce
->name
, ce_namelen(ce
));
42 pos
= insert_hash(hash
, ce
, &istate
->name_hash
);
49 static void lazy_init_name_hash(struct index_state
*istate
)
53 if (istate
->name_hash_initialized
)
55 for (nr
= 0; nr
< istate
->cache_nr
; nr
++)
56 hash_index_entry(istate
, istate
->cache
[nr
]);
57 istate
->name_hash_initialized
= 1;
60 static void set_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
62 istate
->cache
[nr
] = ce
;
63 if (istate
->name_hash_initialized
)
64 hash_index_entry(istate
, ce
);
68 * We don't actually *remove* it, we can just mark it invalid so that
69 * we won't find it in lookups.
71 * Not only would we have to search the lists (simple enough), but
72 * we'd also have to rehash other hash buckets in case this makes the
73 * hash bucket empty (common). So it's much better to just mark
76 static void remove_hash_entry(struct index_state
*istate
, struct cache_entry
*ce
)
78 ce
->ce_flags
|= CE_UNHASHED
;
81 static void replace_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
83 struct cache_entry
*old
= istate
->cache
[nr
];
86 remove_hash_entry(istate
, old
);
87 set_index_entry(istate
, nr
, ce
);
89 istate
->cache_changed
= 1;
92 int index_name_exists(struct index_state
*istate
, const char *name
, int namelen
)
94 unsigned int hash
= hash_name(name
, namelen
);
95 struct cache_entry
*ce
;
97 lazy_init_name_hash(istate
);
98 ce
= lookup_hash(hash
, &istate
->name_hash
);
101 if (!(ce
->ce_flags
& CE_UNHASHED
)) {
102 if (!cache_name_compare(name
, namelen
, ce
->name
, ce
->ce_flags
))
111 * This only updates the "non-critical" parts of the directory
112 * cache, ie the parts that aren't tracked by GIT, and only used
113 * to validate the cache.
115 void fill_stat_cache_info(struct cache_entry
*ce
, struct stat
*st
)
117 ce
->ce_ctime
= st
->st_ctime
;
118 ce
->ce_mtime
= st
->st_mtime
;
119 ce
->ce_dev
= st
->st_dev
;
120 ce
->ce_ino
= st
->st_ino
;
121 ce
->ce_uid
= st
->st_uid
;
122 ce
->ce_gid
= st
->st_gid
;
123 ce
->ce_size
= st
->st_size
;
125 if (assume_unchanged
)
126 ce
->ce_flags
|= CE_VALID
;
128 if (S_ISREG(st
->st_mode
))
129 ce_mark_uptodate(ce
);
132 static int ce_compare_data(struct cache_entry
*ce
, struct stat
*st
)
135 int fd
= open(ce
->name
, O_RDONLY
);
138 unsigned char sha1
[20];
139 if (!index_fd(sha1
, fd
, st
, 0, OBJ_BLOB
, ce
->name
))
140 match
= hashcmp(sha1
, ce
->sha1
);
141 /* index_fd() closed the file descriptor already */
146 static int ce_compare_link(struct cache_entry
*ce
, size_t expected_size
)
152 enum object_type type
;
155 target
= xmalloc(expected_size
);
156 len
= readlink(ce
->name
, target
, expected_size
);
157 if (len
!= expected_size
) {
161 buffer
= read_sha1_file(ce
->sha1
, &type
, &size
);
166 if (size
== expected_size
)
167 match
= memcmp(buffer
, target
, size
);
173 static int ce_compare_gitlink(struct cache_entry
*ce
)
175 unsigned char sha1
[20];
178 * We don't actually require that the .git directory
179 * under GITLINK directory be a valid git directory. It
180 * might even be missing (in case nobody populated that
183 * If so, we consider it always to match.
185 if (resolve_gitlink_ref(ce
->name
, "HEAD", sha1
) < 0)
187 return hashcmp(sha1
, ce
->sha1
);
190 static int ce_modified_check_fs(struct cache_entry
*ce
, struct stat
*st
)
192 switch (st
->st_mode
& S_IFMT
) {
194 if (ce_compare_data(ce
, st
))
198 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
202 if (S_ISGITLINK(ce
->ce_mode
))
210 static int ce_match_stat_basic(struct cache_entry
*ce
, struct stat
*st
)
212 unsigned int changed
= 0;
214 if (ce
->ce_flags
& CE_REMOVE
)
215 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
217 switch (ce
->ce_mode
& S_IFMT
) {
219 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
220 /* We consider only the owner x bit to be relevant for
223 if (trust_executable_bit
&&
224 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
225 changed
|= MODE_CHANGED
;
228 if (!S_ISLNK(st
->st_mode
) &&
229 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
230 changed
|= TYPE_CHANGED
;
233 if (!S_ISDIR(st
->st_mode
))
234 changed
|= TYPE_CHANGED
;
235 else if (ce_compare_gitlink(ce
))
236 changed
|= DATA_CHANGED
;
239 die("internal error: ce_mode is %o", ce
->ce_mode
);
241 if (ce
->ce_mtime
!= (unsigned int) st
->st_mtime
)
242 changed
|= MTIME_CHANGED
;
243 if (ce
->ce_ctime
!= (unsigned int) st
->st_ctime
)
244 changed
|= CTIME_CHANGED
;
246 if (ce
->ce_uid
!= (unsigned int) st
->st_uid
||
247 ce
->ce_gid
!= (unsigned int) st
->st_gid
)
248 changed
|= OWNER_CHANGED
;
249 if (ce
->ce_ino
!= (unsigned int) st
->st_ino
)
250 changed
|= INODE_CHANGED
;
254 * st_dev breaks on network filesystems where different
255 * clients will have different views of what "device"
256 * the filesystem is on
258 if (ce
->ce_dev
!= (unsigned int) st
->st_dev
)
259 changed
|= INODE_CHANGED
;
262 if (ce
->ce_size
!= (unsigned int) st
->st_size
)
263 changed
|= DATA_CHANGED
;
268 static int is_racy_timestamp(struct index_state
*istate
, struct cache_entry
*ce
)
270 return (istate
->timestamp
&&
271 ((unsigned int)istate
->timestamp
) <= ce
->ce_mtime
);
274 int ie_match_stat(struct index_state
*istate
,
275 struct cache_entry
*ce
, struct stat
*st
,
276 unsigned int options
)
278 unsigned int changed
;
279 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
280 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
283 * If it's marked as always valid in the index, it's
284 * valid whatever the checked-out copy says.
286 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
289 changed
= ce_match_stat_basic(ce
, st
);
292 * Within 1 second of this sequence:
293 * echo xyzzy >file && git-update-index --add file
294 * running this command:
296 * would give a falsely clean cache entry. The mtime and
297 * length match the cache, and other stat fields do not change.
299 * We could detect this at update-index time (the cache entry
300 * being registered/updated records the same time as "now")
301 * and delay the return from git-update-index, but that would
302 * effectively mean we can make at most one commit per second,
303 * which is not acceptable. Instead, we check cache entries
304 * whose mtime are the same as the index file timestamp more
305 * carefully than others.
307 if (!changed
&& is_racy_timestamp(istate
, ce
)) {
308 if (assume_racy_is_modified
)
309 changed
|= DATA_CHANGED
;
311 changed
|= ce_modified_check_fs(ce
, st
);
317 int ie_modified(struct index_state
*istate
,
318 struct cache_entry
*ce
, struct stat
*st
, unsigned int options
)
320 int changed
, changed_fs
;
322 changed
= ie_match_stat(istate
, ce
, st
, options
);
326 * If the mode or type has changed, there's no point in trying
327 * to refresh the entry - it's not going to match
329 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
332 /* Immediately after read-tree or update-index --cacheinfo,
333 * the length field is zero. For other cases the ce_size
334 * should match the SHA1 recorded in the index entry.
336 if ((changed
& DATA_CHANGED
) && ce
->ce_size
!= 0)
339 changed_fs
= ce_modified_check_fs(ce
, st
);
341 return changed
| changed_fs
;
345 int base_name_compare(const char *name1
, int len1
, int mode1
,
346 const char *name2
, int len2
, int mode2
)
348 unsigned char c1
, c2
;
349 int len
= len1
< len2
? len1
: len2
;
352 cmp
= memcmp(name1
, name2
, len
);
357 if (!c1
&& S_ISDIR(mode1
))
359 if (!c2
&& S_ISDIR(mode2
))
361 return (c1
< c2
) ? -1 : (c1
> c2
) ? 1 : 0;
364 int cache_name_compare(const char *name1
, int flags1
, const char *name2
, int flags2
)
366 int len1
= flags1
& CE_NAMEMASK
;
367 int len2
= flags2
& CE_NAMEMASK
;
368 int len
= len1
< len2
? len1
: len2
;
371 cmp
= memcmp(name1
, name2
, len
);
380 flags1
&= CE_STAGEMASK
;
381 flags2
&= CE_STAGEMASK
;
390 int index_name_pos(struct index_state
*istate
, const char *name
, int namelen
)
395 last
= istate
->cache_nr
;
396 while (last
> first
) {
397 int next
= (last
+ first
) >> 1;
398 struct cache_entry
*ce
= istate
->cache
[next
];
399 int cmp
= cache_name_compare(name
, namelen
, ce
->name
, ce
->ce_flags
);
411 /* Remove entry, return true if there are more entries to go.. */
412 int remove_index_entry_at(struct index_state
*istate
, int pos
)
414 struct cache_entry
*ce
= istate
->cache
[pos
];
416 remove_hash_entry(istate
, ce
);
417 istate
->cache_changed
= 1;
419 if (pos
>= istate
->cache_nr
)
421 memmove(istate
->cache
+ pos
,
422 istate
->cache
+ pos
+ 1,
423 (istate
->cache_nr
- pos
) * sizeof(struct cache_entry
*));
427 int remove_file_from_index(struct index_state
*istate
, const char *path
)
429 int pos
= index_name_pos(istate
, path
, strlen(path
));
432 cache_tree_invalidate_path(istate
->cache_tree
, path
);
433 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
434 remove_index_entry_at(istate
, pos
);
438 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
440 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
443 static int index_name_pos_also_unmerged(struct index_state
*istate
,
444 const char *path
, int namelen
)
446 int pos
= index_name_pos(istate
, path
, namelen
);
447 struct cache_entry
*ce
;
452 /* maybe unmerged? */
454 if (pos
>= istate
->cache_nr
||
455 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
458 /* order of preference: stage 2, 1, 3 */
459 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
460 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
461 !compare_name(ce
, path
, namelen
))
466 int add_file_to_index(struct index_state
*istate
, const char *path
, int verbose
)
468 int size
, namelen
, pos
;
470 struct cache_entry
*ce
;
471 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_RACY_IS_DIRTY
;
473 if (lstat(path
, &st
))
474 die("%s: unable to stat (%s)", path
, strerror(errno
));
476 if (!S_ISREG(st
.st_mode
) && !S_ISLNK(st
.st_mode
) && !S_ISDIR(st
.st_mode
))
477 die("%s: can only add regular files, symbolic links or git-directories", path
);
479 namelen
= strlen(path
);
480 if (S_ISDIR(st
.st_mode
)) {
481 while (namelen
&& path
[namelen
-1] == '/')
484 size
= cache_entry_size(namelen
);
485 ce
= xcalloc(1, size
);
486 memcpy(ce
->name
, path
, namelen
);
487 ce
->ce_flags
= namelen
;
488 fill_stat_cache_info(ce
, &st
);
490 if (trust_executable_bit
&& has_symlinks
)
491 ce
->ce_mode
= create_ce_mode(st
.st_mode
);
493 /* If there is an existing entry, pick the mode bits and type
494 * from it, otherwise assume unexecutable regular file.
496 struct cache_entry
*ent
;
497 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
499 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
500 ce
->ce_mode
= ce_mode_from_stat(ent
, st
.st_mode
);
503 pos
= index_name_pos(istate
, ce
->name
, namelen
);
505 !ce_stage(istate
->cache
[pos
]) &&
506 !ie_match_stat(istate
, istate
->cache
[pos
], &st
, ce_option
)) {
507 /* Nothing changed, really */
509 ce_mark_uptodate(istate
->cache
[pos
]);
513 if (index_path(ce
->sha1
, path
, &st
, 1))
514 die("unable to index file %s", path
);
515 if (add_index_entry(istate
, ce
, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
))
516 die("unable to add %s to index",path
);
518 printf("add '%s'\n", path
);
522 struct cache_entry
*make_cache_entry(unsigned int mode
,
523 const unsigned char *sha1
, const char *path
, int stage
,
527 struct cache_entry
*ce
;
529 if (!verify_path(path
))
533 size
= cache_entry_size(len
);
534 ce
= xcalloc(1, size
);
536 hashcpy(ce
->sha1
, sha1
);
537 memcpy(ce
->name
, path
, len
);
538 ce
->ce_flags
= create_ce_flags(len
, stage
);
539 ce
->ce_mode
= create_ce_mode(mode
);
542 return refresh_cache_entry(ce
, 0);
547 int ce_same_name(struct cache_entry
*a
, struct cache_entry
*b
)
549 int len
= ce_namelen(a
);
550 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
553 int ce_path_match(const struct cache_entry
*ce
, const char **pathspec
)
555 const char *match
, *name
;
561 len
= ce_namelen(ce
);
563 while ((match
= *pathspec
++) != NULL
) {
564 int matchlen
= strlen(match
);
567 if (memcmp(name
, match
, matchlen
))
569 if (matchlen
&& name
[matchlen
-1] == '/')
571 if (name
[matchlen
] == '/' || !name
[matchlen
])
580 * We fundamentally don't like some paths: we don't want
581 * dot or dot-dot anywhere, and for obvious reasons don't
582 * want to recurse into ".git" either.
584 * Also, we don't want double slashes or slashes at the
585 * end that can make pathnames ambiguous.
587 static int verify_dotfile(const char *rest
)
590 * The first character was '.', but that
591 * has already been discarded, we now test
595 /* "." is not allowed */
600 * ".git" followed by NUL or slash is bad. This
601 * shares the path end test with the ".." case.
611 if (rest
[1] == '\0' || rest
[1] == '/')
617 int verify_path(const char *path
)
634 if (verify_dotfile(path
))
644 * Do we have another file that has the beginning components being a
645 * proper superset of the name we're trying to add?
647 static int has_file_name(struct index_state
*istate
,
648 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
651 int len
= ce_namelen(ce
);
652 int stage
= ce_stage(ce
);
653 const char *name
= ce
->name
;
655 while (pos
< istate
->cache_nr
) {
656 struct cache_entry
*p
= istate
->cache
[pos
++];
658 if (len
>= ce_namelen(p
))
660 if (memcmp(name
, p
->name
, len
))
662 if (ce_stage(p
) != stage
)
664 if (p
->name
[len
] != '/')
666 if (p
->ce_flags
& CE_REMOVE
)
671 remove_index_entry_at(istate
, --pos
);
677 * Do we have another file with a pathname that is a proper
678 * subset of the name we're trying to add?
680 static int has_dir_name(struct index_state
*istate
,
681 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
684 int stage
= ce_stage(ce
);
685 const char *name
= ce
->name
;
686 const char *slash
= name
+ ce_namelen(ce
);
694 if (slash
<= ce
->name
)
699 pos
= index_name_pos(istate
, name
, create_ce_flags(len
, stage
));
702 * Found one, but not so fast. This could
703 * be a marker that says "I was here, but
704 * I am being removed". Such an entry is
705 * not a part of the resulting tree, and
706 * it is Ok to have a directory at the same
709 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
713 remove_index_entry_at(istate
, pos
);
721 * Trivial optimization: if we find an entry that
722 * already matches the sub-directory, then we know
723 * we're ok, and we can exit.
725 while (pos
< istate
->cache_nr
) {
726 struct cache_entry
*p
= istate
->cache
[pos
];
727 if ((ce_namelen(p
) <= len
) ||
728 (p
->name
[len
] != '/') ||
729 memcmp(p
->name
, name
, len
))
730 break; /* not our subdirectory */
731 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
733 * p is at the same stage as our entry, and
734 * is a subdirectory of what we are looking
735 * at, so we cannot have conflicts at our
736 * level or anything shorter.
745 /* We may be in a situation where we already have path/file and path
746 * is being added, or we already have path and path/file is being
747 * added. Either one would result in a nonsense tree that has path
748 * twice when git-write-tree tries to write it out. Prevent it.
750 * If ok-to-replace is specified, we remove the conflicting entries
751 * from the cache so the caller should recompute the insert position.
752 * When this happens, we return non-zero.
754 static int check_file_directory_conflict(struct index_state
*istate
,
755 const struct cache_entry
*ce
,
756 int pos
, int ok_to_replace
)
761 * When ce is an "I am going away" entry, we allow it to be added
763 if (ce
->ce_flags
& CE_REMOVE
)
767 * We check if the path is a sub-path of a subsequent pathname
768 * first, since removing those will not change the position
771 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
774 * Then check if the path might have a clashing sub-directory
777 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
780 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
783 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
784 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
785 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
787 cache_tree_invalidate_path(istate
->cache_tree
, ce
->name
);
788 pos
= index_name_pos(istate
, ce
->name
, ce
->ce_flags
);
790 /* existing match? Just replace it. */
792 replace_index_entry(istate
, pos
, ce
);
798 * Inserting a merged entry ("stage 0") into the index
799 * will always replace all non-merged entries..
801 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
802 while (ce_same_name(istate
->cache
[pos
], ce
)) {
804 if (!remove_index_entry_at(istate
, pos
))
811 if (!verify_path(ce
->name
))
814 if (!skip_df_check
&&
815 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
817 return error("'%s' appears as both a file and as a directory",
819 pos
= index_name_pos(istate
, ce
->name
, ce
->ce_flags
);
825 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
829 if (option
& ADD_CACHE_JUST_APPEND
)
830 pos
= istate
->cache_nr
;
833 ret
= add_index_entry_with_check(istate
, ce
, option
);
839 /* Make sure the array is big enough .. */
840 if (istate
->cache_nr
== istate
->cache_alloc
) {
841 istate
->cache_alloc
= alloc_nr(istate
->cache_alloc
);
842 istate
->cache
= xrealloc(istate
->cache
,
843 istate
->cache_alloc
* sizeof(struct cache_entry
*));
848 if (istate
->cache_nr
> pos
+ 1)
849 memmove(istate
->cache
+ pos
+ 1,
851 (istate
->cache_nr
- pos
- 1) * sizeof(ce
));
852 set_index_entry(istate
, pos
, ce
);
853 istate
->cache_changed
= 1;
858 * "refresh" does not calculate a new sha1 file or bring the
859 * cache up-to-date for mode/content changes. But what it
860 * _does_ do is to "re-match" the stat information of a file
861 * with the cache, so that you can refresh the cache for a
862 * file that hasn't been changed but where the stat entry is
865 * For example, you'd want to do this after doing a "git-read-tree",
866 * to link up the stat cache details with the proper files.
868 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
869 struct cache_entry
*ce
,
870 unsigned int options
, int *err
)
873 struct cache_entry
*updated
;
875 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
880 if (lstat(ce
->name
, &st
) < 0) {
886 changed
= ie_match_stat(istate
, ce
, &st
, options
);
889 * The path is unchanged. If we were told to ignore
890 * valid bit, then we did the actual stat check and
891 * found that the entry is unmodified. If the entry
892 * is not marked VALID, this is the place to mark it
893 * valid again, under "assume unchanged" mode.
895 if (ignore_valid
&& assume_unchanged
&&
896 !(ce
->ce_flags
& CE_VALID
))
897 ; /* mark this one VALID again */
900 * We do not mark the index itself "modified"
901 * because CE_UPTODATE flag is in-core only;
902 * we are not going to write this change out.
904 ce_mark_uptodate(ce
);
909 if (ie_modified(istate
, ce
, &st
, options
)) {
916 updated
= xmalloc(size
);
917 memcpy(updated
, ce
, size
);
918 fill_stat_cache_info(updated
, &st
);
920 * If ignore_valid is not set, we should leave CE_VALID bit
921 * alone. Otherwise, paths marked with --no-assume-unchanged
922 * (i.e. things to be edited) will reacquire CE_VALID bit
923 * automatically, which is not really what we want.
925 if (!ignore_valid
&& assume_unchanged
&&
926 !(ce
->ce_flags
& CE_VALID
))
927 updated
->ce_flags
&= ~CE_VALID
;
932 int refresh_index(struct index_state
*istate
, unsigned int flags
, const char **pathspec
, char *seen
)
936 int really
= (flags
& REFRESH_REALLY
) != 0;
937 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
938 int quiet
= (flags
& REFRESH_QUIET
) != 0;
939 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
940 unsigned int options
= really
? CE_MATCH_IGNORE_VALID
: 0;
942 for (i
= 0; i
< istate
->cache_nr
; i
++) {
943 struct cache_entry
*ce
, *new;
946 ce
= istate
->cache
[i
];
948 while ((i
< istate
->cache_nr
) &&
949 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
954 printf("%s: needs merge\n", ce
->name
);
959 if (pathspec
&& !match_pathspec(pathspec
, ce
->name
, strlen(ce
->name
), 0, seen
))
962 new = refresh_cache_ent(istate
, ce
, options
, &cache_errno
);
966 if (not_new
&& cache_errno
== ENOENT
)
968 if (really
&& cache_errno
== EINVAL
) {
969 /* If we are doing --really-refresh that
970 * means the index is not valid anymore.
972 ce
->ce_flags
&= ~CE_VALID
;
973 istate
->cache_changed
= 1;
977 printf("%s: needs update\n", ce
->name
);
982 replace_index_entry(istate
, i
, new);
987 struct cache_entry
*refresh_cache_entry(struct cache_entry
*ce
, int really
)
989 return refresh_cache_ent(&the_index
, ce
, really
, NULL
);
992 static int verify_hdr(struct cache_header
*hdr
, unsigned long size
)
995 unsigned char sha1
[20];
997 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
998 return error("bad signature");
999 if (hdr
->hdr_version
!= htonl(2))
1000 return error("bad index version");
1002 SHA1_Update(&c
, hdr
, size
- 20);
1003 SHA1_Final(sha1
, &c
);
1004 if (hashcmp(sha1
, (unsigned char *)hdr
+ size
- 20))
1005 return error("bad index file sha1 signature");
1009 static int read_index_extension(struct index_state
*istate
,
1010 const char *ext
, void *data
, unsigned long sz
)
1012 switch (CACHE_EXT(ext
)) {
1013 case CACHE_EXT_TREE
:
1014 istate
->cache_tree
= cache_tree_read(data
, sz
);
1017 if (*ext
< 'A' || 'Z' < *ext
)
1018 return error("index uses %.4s extension, which we do not understand",
1020 fprintf(stderr
, "ignoring %.4s extension\n", ext
);
1026 int read_index(struct index_state
*istate
)
1028 return read_index_from(istate
, get_index_file());
1031 static void convert_from_disk(struct ondisk_cache_entry
*ondisk
, struct cache_entry
*ce
)
1035 ce
->ce_ctime
= ntohl(ondisk
->ctime
.sec
);
1036 ce
->ce_mtime
= ntohl(ondisk
->mtime
.sec
);
1037 ce
->ce_dev
= ntohl(ondisk
->dev
);
1038 ce
->ce_ino
= ntohl(ondisk
->ino
);
1039 ce
->ce_mode
= ntohl(ondisk
->mode
);
1040 ce
->ce_uid
= ntohl(ondisk
->uid
);
1041 ce
->ce_gid
= ntohl(ondisk
->gid
);
1042 ce
->ce_size
= ntohl(ondisk
->size
);
1043 /* On-disk flags are just 16 bits */
1044 ce
->ce_flags
= ntohs(ondisk
->flags
);
1045 hashcpy(ce
->sha1
, ondisk
->sha1
);
1047 len
= ce
->ce_flags
& CE_NAMEMASK
;
1048 if (len
== CE_NAMEMASK
)
1049 len
= strlen(ondisk
->name
);
1051 * NEEDSWORK: If the original index is crafted, this copy could
1054 memcpy(ce
->name
, ondisk
->name
, len
+ 1);
1057 static inline size_t estimate_cache_size(size_t ondisk_size
, unsigned int entries
)
1061 per_entry
= sizeof(struct cache_entry
) - sizeof(struct ondisk_cache_entry
);
1064 * Alignment can cause differences. This should be "alignof", but
1065 * since that's a gcc'ism, just use the size of a pointer.
1067 per_entry
+= sizeof(void *);
1068 return ondisk_size
+ entries
*per_entry
;
1071 /* remember to discard_cache() before reading a different cache! */
1072 int read_index_from(struct index_state
*istate
, const char *path
)
1076 unsigned long src_offset
, dst_offset
;
1077 struct cache_header
*hdr
;
1083 return istate
->cache_nr
;
1086 istate
->timestamp
= 0;
1087 fd
= open(path
, O_RDONLY
);
1089 if (errno
== ENOENT
)
1091 die("index file open failed (%s)", strerror(errno
));
1095 die("cannot stat the open index (%s)", strerror(errno
));
1098 mmap_size
= xsize_t(st
.st_size
);
1099 if (mmap_size
< sizeof(struct cache_header
) + 20)
1100 die("index file smaller than expected");
1102 mmap
= xmmap(NULL
, mmap_size
, PROT_READ
| PROT_WRITE
, MAP_PRIVATE
, fd
, 0);
1104 if (mmap
== MAP_FAILED
)
1105 die("unable to map index file");
1108 if (verify_hdr(hdr
, mmap_size
) < 0)
1111 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
1112 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
1113 istate
->cache
= xcalloc(istate
->cache_alloc
, sizeof(struct cache_entry
*));
1116 * The disk format is actually larger than the in-memory format,
1117 * due to space for nsec etc, so even though the in-memory one
1118 * has room for a few more flags, we can allocate using the same
1121 istate
->alloc
= xmalloc(estimate_cache_size(mmap_size
, istate
->cache_nr
));
1123 src_offset
= sizeof(*hdr
);
1125 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1126 struct ondisk_cache_entry
*disk_ce
;
1127 struct cache_entry
*ce
;
1129 disk_ce
= (struct ondisk_cache_entry
*)((char *)mmap
+ src_offset
);
1130 ce
= (struct cache_entry
*)((char *)istate
->alloc
+ dst_offset
);
1131 convert_from_disk(disk_ce
, ce
);
1132 set_index_entry(istate
, i
, ce
);
1134 src_offset
+= ondisk_ce_size(ce
);
1135 dst_offset
+= ce_size(ce
);
1137 istate
->timestamp
= st
.st_mtime
;
1138 while (src_offset
<= mmap_size
- 20 - 8) {
1139 /* After an array of active_nr index entries,
1140 * there can be arbitrary number of extended
1141 * sections, each of which is prefixed with
1142 * extension name (4-byte) and section length
1143 * in 4-byte network byte order.
1145 unsigned long extsize
;
1146 memcpy(&extsize
, (char *)mmap
+ src_offset
+ 4, 4);
1147 extsize
= ntohl(extsize
);
1148 if (read_index_extension(istate
,
1149 (const char *) mmap
+ src_offset
,
1150 (char *) mmap
+ src_offset
+ 8,
1154 src_offset
+= extsize
;
1156 munmap(mmap
, mmap_size
);
1157 return istate
->cache_nr
;
1160 munmap(mmap
, mmap_size
);
1162 die("index file corrupt");
1165 int discard_index(struct index_state
*istate
)
1167 istate
->cache_nr
= 0;
1168 istate
->cache_changed
= 0;
1169 istate
->timestamp
= 0;
1170 free_hash(&istate
->name_hash
);
1171 cache_tree_free(&(istate
->cache_tree
));
1172 free(istate
->alloc
);
1173 istate
->alloc
= NULL
;
1175 /* no need to throw away allocated active_cache */
1179 int unmerged_index(struct index_state
*istate
)
1182 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1183 if (ce_stage(istate
->cache
[i
]))
1189 #define WRITE_BUFFER_SIZE 8192
1190 static unsigned char write_buffer
[WRITE_BUFFER_SIZE
];
1191 static unsigned long write_buffer_len
;
1193 static int ce_write_flush(SHA_CTX
*context
, int fd
)
1195 unsigned int buffered
= write_buffer_len
;
1197 SHA1_Update(context
, write_buffer
, buffered
);
1198 if (write_in_full(fd
, write_buffer
, buffered
) != buffered
)
1200 write_buffer_len
= 0;
1205 static int ce_write(SHA_CTX
*context
, int fd
, void *data
, unsigned int len
)
1208 unsigned int buffered
= write_buffer_len
;
1209 unsigned int partial
= WRITE_BUFFER_SIZE
- buffered
;
1212 memcpy(write_buffer
+ buffered
, data
, partial
);
1213 buffered
+= partial
;
1214 if (buffered
== WRITE_BUFFER_SIZE
) {
1215 write_buffer_len
= buffered
;
1216 if (ce_write_flush(context
, fd
))
1220 write_buffer_len
= buffered
;
1222 data
= (char *) data
+ partial
;
1227 static int write_index_ext_header(SHA_CTX
*context
, int fd
,
1228 unsigned int ext
, unsigned int sz
)
1232 return ((ce_write(context
, fd
, &ext
, 4) < 0) ||
1233 (ce_write(context
, fd
, &sz
, 4) < 0)) ? -1 : 0;
1236 static int ce_flush(SHA_CTX
*context
, int fd
)
1238 unsigned int left
= write_buffer_len
;
1241 write_buffer_len
= 0;
1242 SHA1_Update(context
, write_buffer
, left
);
1245 /* Flush first if not enough space for SHA1 signature */
1246 if (left
+ 20 > WRITE_BUFFER_SIZE
) {
1247 if (write_in_full(fd
, write_buffer
, left
) != left
)
1252 /* Append the SHA1 signature at the end */
1253 SHA1_Final(write_buffer
+ left
, context
);
1255 return (write_in_full(fd
, write_buffer
, left
) != left
) ? -1 : 0;
1258 static void ce_smudge_racily_clean_entry(struct cache_entry
*ce
)
1261 * The only thing we care about in this function is to smudge the
1262 * falsely clean entry due to touch-update-touch race, so we leave
1263 * everything else as they are. We are called for entries whose
1264 * ce_mtime match the index file mtime.
1268 if (lstat(ce
->name
, &st
) < 0)
1270 if (ce_match_stat_basic(ce
, &st
))
1272 if (ce_modified_check_fs(ce
, &st
)) {
1273 /* This is "racily clean"; smudge it. Note that this
1274 * is a tricky code. At first glance, it may appear
1275 * that it can break with this sequence:
1277 * $ echo xyzzy >frotz
1278 * $ git-update-index --add frotz
1281 * $ echo filfre >nitfol
1282 * $ git-update-index --add nitfol
1284 * but it does not. When the second update-index runs,
1285 * it notices that the entry "frotz" has the same timestamp
1286 * as index, and if we were to smudge it by resetting its
1287 * size to zero here, then the object name recorded
1288 * in index is the 6-byte file but the cached stat information
1289 * becomes zero --- which would then match what we would
1290 * obtain from the filesystem next time we stat("frotz").
1292 * However, the second update-index, before calling
1293 * this function, notices that the cached size is 6
1294 * bytes and what is on the filesystem is an empty
1295 * file, and never calls us, so the cached size information
1296 * for "frotz" stays 6 which does not match the filesystem.
1302 static int ce_write_entry(SHA_CTX
*c
, int fd
, struct cache_entry
*ce
)
1304 int size
= ondisk_ce_size(ce
);
1305 struct ondisk_cache_entry
*ondisk
= xcalloc(1, size
);
1307 ondisk
->ctime
.sec
= htonl(ce
->ce_ctime
);
1308 ondisk
->ctime
.nsec
= 0;
1309 ondisk
->mtime
.sec
= htonl(ce
->ce_mtime
);
1310 ondisk
->mtime
.nsec
= 0;
1311 ondisk
->dev
= htonl(ce
->ce_dev
);
1312 ondisk
->ino
= htonl(ce
->ce_ino
);
1313 ondisk
->mode
= htonl(ce
->ce_mode
);
1314 ondisk
->uid
= htonl(ce
->ce_uid
);
1315 ondisk
->gid
= htonl(ce
->ce_gid
);
1316 ondisk
->size
= htonl(ce
->ce_size
);
1317 hashcpy(ondisk
->sha1
, ce
->sha1
);
1318 ondisk
->flags
= htons(ce
->ce_flags
);
1319 memcpy(ondisk
->name
, ce
->name
, ce_namelen(ce
));
1321 return ce_write(c
, fd
, ondisk
, size
);
1324 int write_index(struct index_state
*istate
, int newfd
)
1327 struct cache_header hdr
;
1328 int i
, err
, removed
;
1329 struct cache_entry
**cache
= istate
->cache
;
1330 int entries
= istate
->cache_nr
;
1332 for (i
= removed
= 0; i
< entries
; i
++)
1333 if (cache
[i
]->ce_flags
& CE_REMOVE
)
1336 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
1337 hdr
.hdr_version
= htonl(2);
1338 hdr
.hdr_entries
= htonl(entries
- removed
);
1341 if (ce_write(&c
, newfd
, &hdr
, sizeof(hdr
)) < 0)
1344 for (i
= 0; i
< entries
; i
++) {
1345 struct cache_entry
*ce
= cache
[i
];
1346 if (ce
->ce_flags
& CE_REMOVE
)
1348 if (is_racy_timestamp(istate
, ce
))
1349 ce_smudge_racily_clean_entry(ce
);
1350 if (ce_write_entry(&c
, newfd
, ce
) < 0)
1354 /* Write extension data here */
1355 if (istate
->cache_tree
) {
1358 strbuf_init(&sb
, 0);
1359 cache_tree_write(&sb
, istate
->cache_tree
);
1360 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_TREE
, sb
.len
) < 0
1361 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
1362 strbuf_release(&sb
);
1366 return ce_flush(&c
, newfd
);