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 void set_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
28 istate
->cache
[nr
] = ce
;
29 add_name_hash(istate
, ce
);
32 static void replace_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
34 struct cache_entry
*old
= istate
->cache
[nr
];
36 remove_name_hash(old
);
37 set_index_entry(istate
, nr
, ce
);
38 istate
->cache_changed
= 1;
42 * This only updates the "non-critical" parts of the directory
43 * cache, ie the parts that aren't tracked by GIT, and only used
44 * to validate the cache.
46 void fill_stat_cache_info(struct cache_entry
*ce
, struct stat
*st
)
48 ce
->ce_ctime
= st
->st_ctime
;
49 ce
->ce_mtime
= st
->st_mtime
;
50 ce
->ce_dev
= st
->st_dev
;
51 ce
->ce_ino
= st
->st_ino
;
52 ce
->ce_uid
= st
->st_uid
;
53 ce
->ce_gid
= st
->st_gid
;
54 ce
->ce_size
= st
->st_size
;
57 ce
->ce_flags
|= CE_VALID
;
59 if (S_ISREG(st
->st_mode
))
63 static int ce_compare_data(struct cache_entry
*ce
, struct stat
*st
)
66 int fd
= open(ce
->name
, O_RDONLY
);
69 unsigned char sha1
[20];
70 if (!index_fd(sha1
, fd
, st
, 0, OBJ_BLOB
, ce
->name
))
71 match
= hashcmp(sha1
, ce
->sha1
);
72 /* index_fd() closed the file descriptor already */
77 static int ce_compare_link(struct cache_entry
*ce
, size_t expected_size
)
83 enum object_type type
;
86 target
= xmalloc(expected_size
);
87 len
= readlink(ce
->name
, target
, expected_size
);
88 if (len
!= expected_size
) {
92 buffer
= read_sha1_file(ce
->sha1
, &type
, &size
);
97 if (size
== expected_size
)
98 match
= memcmp(buffer
, target
, size
);
104 static int ce_compare_gitlink(struct cache_entry
*ce
)
106 unsigned char sha1
[20];
109 * We don't actually require that the .git directory
110 * under GITLINK directory be a valid git directory. It
111 * might even be missing (in case nobody populated that
114 * If so, we consider it always to match.
116 if (resolve_gitlink_ref(ce
->name
, "HEAD", sha1
) < 0)
118 return hashcmp(sha1
, ce
->sha1
);
121 static int ce_modified_check_fs(struct cache_entry
*ce
, struct stat
*st
)
123 switch (st
->st_mode
& S_IFMT
) {
125 if (ce_compare_data(ce
, st
))
129 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
133 if (S_ISGITLINK(ce
->ce_mode
))
141 static int ce_match_stat_basic(struct cache_entry
*ce
, struct stat
*st
)
143 unsigned int changed
= 0;
145 if (ce
->ce_flags
& CE_REMOVE
)
146 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
148 switch (ce
->ce_mode
& S_IFMT
) {
150 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
151 /* We consider only the owner x bit to be relevant for
154 if (trust_executable_bit
&&
155 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
156 changed
|= MODE_CHANGED
;
159 if (!S_ISLNK(st
->st_mode
) &&
160 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
161 changed
|= TYPE_CHANGED
;
164 if (!S_ISDIR(st
->st_mode
))
165 changed
|= TYPE_CHANGED
;
166 else if (ce_compare_gitlink(ce
))
167 changed
|= DATA_CHANGED
;
170 die("internal error: ce_mode is %o", ce
->ce_mode
);
172 if (ce
->ce_mtime
!= (unsigned int) st
->st_mtime
)
173 changed
|= MTIME_CHANGED
;
174 if (ce
->ce_ctime
!= (unsigned int) st
->st_ctime
)
175 changed
|= CTIME_CHANGED
;
177 if (ce
->ce_uid
!= (unsigned int) st
->st_uid
||
178 ce
->ce_gid
!= (unsigned int) st
->st_gid
)
179 changed
|= OWNER_CHANGED
;
180 if (ce
->ce_ino
!= (unsigned int) st
->st_ino
)
181 changed
|= INODE_CHANGED
;
185 * st_dev breaks on network filesystems where different
186 * clients will have different views of what "device"
187 * the filesystem is on
189 if (ce
->ce_dev
!= (unsigned int) st
->st_dev
)
190 changed
|= INODE_CHANGED
;
193 if (ce
->ce_size
!= (unsigned int) st
->st_size
)
194 changed
|= DATA_CHANGED
;
199 static int is_racy_timestamp(const struct index_state
*istate
, struct cache_entry
*ce
)
201 return (istate
->timestamp
&&
202 ((unsigned int)istate
->timestamp
) <= ce
->ce_mtime
);
205 int ie_match_stat(const struct index_state
*istate
,
206 struct cache_entry
*ce
, struct stat
*st
,
207 unsigned int options
)
209 unsigned int changed
;
210 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
211 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
214 * If it's marked as always valid in the index, it's
215 * valid whatever the checked-out copy says.
217 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
220 changed
= ce_match_stat_basic(ce
, st
);
223 * Within 1 second of this sequence:
224 * echo xyzzy >file && git-update-index --add file
225 * running this command:
227 * would give a falsely clean cache entry. The mtime and
228 * length match the cache, and other stat fields do not change.
230 * We could detect this at update-index time (the cache entry
231 * being registered/updated records the same time as "now")
232 * and delay the return from git-update-index, but that would
233 * effectively mean we can make at most one commit per second,
234 * which is not acceptable. Instead, we check cache entries
235 * whose mtime are the same as the index file timestamp more
236 * carefully than others.
238 if (!changed
&& is_racy_timestamp(istate
, ce
)) {
239 if (assume_racy_is_modified
)
240 changed
|= DATA_CHANGED
;
242 changed
|= ce_modified_check_fs(ce
, st
);
248 int ie_modified(const struct index_state
*istate
,
249 struct cache_entry
*ce
, struct stat
*st
, unsigned int options
)
251 int changed
, changed_fs
;
253 changed
= ie_match_stat(istate
, ce
, st
, options
);
257 * If the mode or type has changed, there's no point in trying
258 * to refresh the entry - it's not going to match
260 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
263 /* Immediately after read-tree or update-index --cacheinfo,
264 * the length field is zero. For other cases the ce_size
265 * should match the SHA1 recorded in the index entry.
267 if ((changed
& DATA_CHANGED
) && ce
->ce_size
!= 0)
270 changed_fs
= ce_modified_check_fs(ce
, st
);
272 return changed
| changed_fs
;
276 int base_name_compare(const char *name1
, int len1
, int mode1
,
277 const char *name2
, int len2
, int mode2
)
279 unsigned char c1
, c2
;
280 int len
= len1
< len2
? len1
: len2
;
283 cmp
= memcmp(name1
, name2
, len
);
288 if (!c1
&& S_ISDIR(mode1
))
290 if (!c2
&& S_ISDIR(mode2
))
292 return (c1
< c2
) ? -1 : (c1
> c2
) ? 1 : 0;
296 * df_name_compare() is identical to base_name_compare(), except it
297 * compares conflicting directory/file entries as equal. Note that
298 * while a directory name compares as equal to a regular file, they
299 * then individually compare _differently_ to a filename that has
300 * a dot after the basename (because '\0' < '.' < '/').
302 * This is used by routines that want to traverse the git namespace
303 * but then handle conflicting entries together when possible.
305 int df_name_compare(const char *name1
, int len1
, int mode1
,
306 const char *name2
, int len2
, int mode2
)
308 int len
= len1
< len2
? len1
: len2
, cmp
;
309 unsigned char c1
, c2
;
311 cmp
= memcmp(name1
, name2
, len
);
314 /* Directories and files compare equal (same length, same name) */
318 if (!c1
&& S_ISDIR(mode1
))
321 if (!c2
&& S_ISDIR(mode2
))
323 if (c1
== '/' && !c2
)
325 if (c2
== '/' && !c1
)
330 int cache_name_compare(const char *name1
, int flags1
, const char *name2
, int flags2
)
332 int len1
= flags1
& CE_NAMEMASK
;
333 int len2
= flags2
& CE_NAMEMASK
;
334 int len
= len1
< len2
? len1
: len2
;
337 cmp
= memcmp(name1
, name2
, len
);
346 flags1
&= CE_STAGEMASK
;
347 flags2
&= CE_STAGEMASK
;
356 int index_name_pos(const struct index_state
*istate
, const char *name
, int namelen
)
361 last
= istate
->cache_nr
;
362 while (last
> first
) {
363 int next
= (last
+ first
) >> 1;
364 struct cache_entry
*ce
= istate
->cache
[next
];
365 int cmp
= cache_name_compare(name
, namelen
, ce
->name
, ce
->ce_flags
);
377 /* Remove entry, return true if there are more entries to go.. */
378 int remove_index_entry_at(struct index_state
*istate
, int pos
)
380 struct cache_entry
*ce
= istate
->cache
[pos
];
382 remove_name_hash(ce
);
383 istate
->cache_changed
= 1;
385 if (pos
>= istate
->cache_nr
)
387 memmove(istate
->cache
+ pos
,
388 istate
->cache
+ pos
+ 1,
389 (istate
->cache_nr
- pos
) * sizeof(struct cache_entry
*));
393 int remove_file_from_index(struct index_state
*istate
, const char *path
)
395 int pos
= index_name_pos(istate
, path
, strlen(path
));
398 cache_tree_invalidate_path(istate
->cache_tree
, path
);
399 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
400 remove_index_entry_at(istate
, pos
);
404 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
406 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
409 static int index_name_pos_also_unmerged(struct index_state
*istate
,
410 const char *path
, int namelen
)
412 int pos
= index_name_pos(istate
, path
, namelen
);
413 struct cache_entry
*ce
;
418 /* maybe unmerged? */
420 if (pos
>= istate
->cache_nr
||
421 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
424 /* order of preference: stage 2, 1, 3 */
425 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
426 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
427 !compare_name(ce
, path
, namelen
))
432 int add_file_to_index(struct index_state
*istate
, const char *path
, int verbose
)
436 struct cache_entry
*ce
, *alias
;
437 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_RACY_IS_DIRTY
;
439 if (lstat(path
, &st
))
440 die("%s: unable to stat (%s)", path
, strerror(errno
));
442 if (!S_ISREG(st
.st_mode
) && !S_ISLNK(st
.st_mode
) && !S_ISDIR(st
.st_mode
))
443 die("%s: can only add regular files, symbolic links or git-directories", path
);
445 namelen
= strlen(path
);
446 if (S_ISDIR(st
.st_mode
)) {
447 while (namelen
&& path
[namelen
-1] == '/')
450 size
= cache_entry_size(namelen
);
451 ce
= xcalloc(1, size
);
452 memcpy(ce
->name
, path
, namelen
);
453 ce
->ce_flags
= namelen
;
454 fill_stat_cache_info(ce
, &st
);
456 if (trust_executable_bit
&& has_symlinks
)
457 ce
->ce_mode
= create_ce_mode(st
.st_mode
);
459 /* If there is an existing entry, pick the mode bits and type
460 * from it, otherwise assume unexecutable regular file.
462 struct cache_entry
*ent
;
463 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
465 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
466 ce
->ce_mode
= ce_mode_from_stat(ent
, st
.st_mode
);
469 alias
= index_name_exists(istate
, ce
->name
, ce_namelen(ce
), ignore_case
);
470 if (alias
&& !ce_stage(alias
) && !ie_match_stat(istate
, alias
, &st
, ce_option
)) {
471 /* Nothing changed, really */
473 ce_mark_uptodate(alias
);
477 if (index_path(ce
->sha1
, path
, &st
, 1))
478 die("unable to index file %s", path
);
479 if (add_index_entry(istate
, ce
, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
))
480 die("unable to add %s to index",path
);
482 printf("add '%s'\n", path
);
486 struct cache_entry
*make_cache_entry(unsigned int mode
,
487 const unsigned char *sha1
, const char *path
, int stage
,
491 struct cache_entry
*ce
;
493 if (!verify_path(path
))
497 size
= cache_entry_size(len
);
498 ce
= xcalloc(1, size
);
500 hashcpy(ce
->sha1
, sha1
);
501 memcpy(ce
->name
, path
, len
);
502 ce
->ce_flags
= create_ce_flags(len
, stage
);
503 ce
->ce_mode
= create_ce_mode(mode
);
506 return refresh_cache_entry(ce
, 0);
511 int ce_same_name(struct cache_entry
*a
, struct cache_entry
*b
)
513 int len
= ce_namelen(a
);
514 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
517 int ce_path_match(const struct cache_entry
*ce
, const char **pathspec
)
519 const char *match
, *name
;
525 len
= ce_namelen(ce
);
527 while ((match
= *pathspec
++) != NULL
) {
528 int matchlen
= strlen(match
);
531 if (memcmp(name
, match
, matchlen
))
533 if (matchlen
&& name
[matchlen
-1] == '/')
535 if (name
[matchlen
] == '/' || !name
[matchlen
])
544 * We fundamentally don't like some paths: we don't want
545 * dot or dot-dot anywhere, and for obvious reasons don't
546 * want to recurse into ".git" either.
548 * Also, we don't want double slashes or slashes at the
549 * end that can make pathnames ambiguous.
551 static int verify_dotfile(const char *rest
)
554 * The first character was '.', but that
555 * has already been discarded, we now test
559 /* "." is not allowed */
564 * ".git" followed by NUL or slash is bad. This
565 * shares the path end test with the ".." case.
575 if (rest
[1] == '\0' || rest
[1] == '/')
581 int verify_path(const char *path
)
598 if (verify_dotfile(path
))
608 * Do we have another file that has the beginning components being a
609 * proper superset of the name we're trying to add?
611 static int has_file_name(struct index_state
*istate
,
612 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
615 int len
= ce_namelen(ce
);
616 int stage
= ce_stage(ce
);
617 const char *name
= ce
->name
;
619 while (pos
< istate
->cache_nr
) {
620 struct cache_entry
*p
= istate
->cache
[pos
++];
622 if (len
>= ce_namelen(p
))
624 if (memcmp(name
, p
->name
, len
))
626 if (ce_stage(p
) != stage
)
628 if (p
->name
[len
] != '/')
630 if (p
->ce_flags
& CE_REMOVE
)
635 remove_index_entry_at(istate
, --pos
);
641 * Do we have another file with a pathname that is a proper
642 * subset of the name we're trying to add?
644 static int has_dir_name(struct index_state
*istate
,
645 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
648 int stage
= ce_stage(ce
);
649 const char *name
= ce
->name
;
650 const char *slash
= name
+ ce_namelen(ce
);
658 if (slash
<= ce
->name
)
663 pos
= index_name_pos(istate
, name
, create_ce_flags(len
, stage
));
666 * Found one, but not so fast. This could
667 * be a marker that says "I was here, but
668 * I am being removed". Such an entry is
669 * not a part of the resulting tree, and
670 * it is Ok to have a directory at the same
673 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
677 remove_index_entry_at(istate
, pos
);
685 * Trivial optimization: if we find an entry that
686 * already matches the sub-directory, then we know
687 * we're ok, and we can exit.
689 while (pos
< istate
->cache_nr
) {
690 struct cache_entry
*p
= istate
->cache
[pos
];
691 if ((ce_namelen(p
) <= len
) ||
692 (p
->name
[len
] != '/') ||
693 memcmp(p
->name
, name
, len
))
694 break; /* not our subdirectory */
695 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
697 * p is at the same stage as our entry, and
698 * is a subdirectory of what we are looking
699 * at, so we cannot have conflicts at our
700 * level or anything shorter.
709 /* We may be in a situation where we already have path/file and path
710 * is being added, or we already have path and path/file is being
711 * added. Either one would result in a nonsense tree that has path
712 * twice when git-write-tree tries to write it out. Prevent it.
714 * If ok-to-replace is specified, we remove the conflicting entries
715 * from the cache so the caller should recompute the insert position.
716 * When this happens, we return non-zero.
718 static int check_file_directory_conflict(struct index_state
*istate
,
719 const struct cache_entry
*ce
,
720 int pos
, int ok_to_replace
)
725 * When ce is an "I am going away" entry, we allow it to be added
727 if (ce
->ce_flags
& CE_REMOVE
)
731 * We check if the path is a sub-path of a subsequent pathname
732 * first, since removing those will not change the position
735 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
738 * Then check if the path might have a clashing sub-directory
741 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
744 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
747 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
748 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
749 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
751 cache_tree_invalidate_path(istate
->cache_tree
, ce
->name
);
752 pos
= index_name_pos(istate
, ce
->name
, ce
->ce_flags
);
754 /* existing match? Just replace it. */
756 replace_index_entry(istate
, pos
, ce
);
762 * Inserting a merged entry ("stage 0") into the index
763 * will always replace all non-merged entries..
765 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
766 while (ce_same_name(istate
->cache
[pos
], ce
)) {
768 if (!remove_index_entry_at(istate
, pos
))
775 if (!verify_path(ce
->name
))
778 if (!skip_df_check
&&
779 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
781 return error("'%s' appears as both a file and as a directory",
783 pos
= index_name_pos(istate
, ce
->name
, ce
->ce_flags
);
789 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
793 if (option
& ADD_CACHE_JUST_APPEND
)
794 pos
= istate
->cache_nr
;
797 ret
= add_index_entry_with_check(istate
, ce
, option
);
803 /* Make sure the array is big enough .. */
804 if (istate
->cache_nr
== istate
->cache_alloc
) {
805 istate
->cache_alloc
= alloc_nr(istate
->cache_alloc
);
806 istate
->cache
= xrealloc(istate
->cache
,
807 istate
->cache_alloc
* sizeof(struct cache_entry
*));
812 if (istate
->cache_nr
> pos
+ 1)
813 memmove(istate
->cache
+ pos
+ 1,
815 (istate
->cache_nr
- pos
- 1) * sizeof(ce
));
816 set_index_entry(istate
, pos
, ce
);
817 istate
->cache_changed
= 1;
822 * "refresh" does not calculate a new sha1 file or bring the
823 * cache up-to-date for mode/content changes. But what it
824 * _does_ do is to "re-match" the stat information of a file
825 * with the cache, so that you can refresh the cache for a
826 * file that hasn't been changed but where the stat entry is
829 * For example, you'd want to do this after doing a "git-read-tree",
830 * to link up the stat cache details with the proper files.
832 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
833 struct cache_entry
*ce
,
834 unsigned int options
, int *err
)
837 struct cache_entry
*updated
;
839 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
844 if (lstat(ce
->name
, &st
) < 0) {
850 changed
= ie_match_stat(istate
, ce
, &st
, options
);
853 * The path is unchanged. If we were told to ignore
854 * valid bit, then we did the actual stat check and
855 * found that the entry is unmodified. If the entry
856 * is not marked VALID, this is the place to mark it
857 * valid again, under "assume unchanged" mode.
859 if (ignore_valid
&& assume_unchanged
&&
860 !(ce
->ce_flags
& CE_VALID
))
861 ; /* mark this one VALID again */
864 * We do not mark the index itself "modified"
865 * because CE_UPTODATE flag is in-core only;
866 * we are not going to write this change out.
868 ce_mark_uptodate(ce
);
873 if (ie_modified(istate
, ce
, &st
, options
)) {
880 updated
= xmalloc(size
);
881 memcpy(updated
, ce
, size
);
882 fill_stat_cache_info(updated
, &st
);
884 * If ignore_valid is not set, we should leave CE_VALID bit
885 * alone. Otherwise, paths marked with --no-assume-unchanged
886 * (i.e. things to be edited) will reacquire CE_VALID bit
887 * automatically, which is not really what we want.
889 if (!ignore_valid
&& assume_unchanged
&&
890 !(ce
->ce_flags
& CE_VALID
))
891 updated
->ce_flags
&= ~CE_VALID
;
896 int refresh_index(struct index_state
*istate
, unsigned int flags
, const char **pathspec
, char *seen
)
900 int really
= (flags
& REFRESH_REALLY
) != 0;
901 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
902 int quiet
= (flags
& REFRESH_QUIET
) != 0;
903 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
904 unsigned int options
= really
? CE_MATCH_IGNORE_VALID
: 0;
906 for (i
= 0; i
< istate
->cache_nr
; i
++) {
907 struct cache_entry
*ce
, *new;
910 ce
= istate
->cache
[i
];
912 while ((i
< istate
->cache_nr
) &&
913 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
918 printf("%s: needs merge\n", ce
->name
);
923 if (pathspec
&& !match_pathspec(pathspec
, ce
->name
, strlen(ce
->name
), 0, seen
))
926 new = refresh_cache_ent(istate
, ce
, options
, &cache_errno
);
930 if (not_new
&& cache_errno
== ENOENT
)
932 if (really
&& cache_errno
== EINVAL
) {
933 /* If we are doing --really-refresh that
934 * means the index is not valid anymore.
936 ce
->ce_flags
&= ~CE_VALID
;
937 istate
->cache_changed
= 1;
941 printf("%s: needs update\n", ce
->name
);
946 replace_index_entry(istate
, i
, new);
951 struct cache_entry
*refresh_cache_entry(struct cache_entry
*ce
, int really
)
953 return refresh_cache_ent(&the_index
, ce
, really
, NULL
);
956 static int verify_hdr(struct cache_header
*hdr
, unsigned long size
)
959 unsigned char sha1
[20];
961 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
962 return error("bad signature");
963 if (hdr
->hdr_version
!= htonl(2))
964 return error("bad index version");
966 SHA1_Update(&c
, hdr
, size
- 20);
967 SHA1_Final(sha1
, &c
);
968 if (hashcmp(sha1
, (unsigned char *)hdr
+ size
- 20))
969 return error("bad index file sha1 signature");
973 static int read_index_extension(struct index_state
*istate
,
974 const char *ext
, void *data
, unsigned long sz
)
976 switch (CACHE_EXT(ext
)) {
978 istate
->cache_tree
= cache_tree_read(data
, sz
);
981 if (*ext
< 'A' || 'Z' < *ext
)
982 return error("index uses %.4s extension, which we do not understand",
984 fprintf(stderr
, "ignoring %.4s extension\n", ext
);
990 int read_index(struct index_state
*istate
)
992 return read_index_from(istate
, get_index_file());
995 static void convert_from_disk(struct ondisk_cache_entry
*ondisk
, struct cache_entry
*ce
)
999 ce
->ce_ctime
= ntohl(ondisk
->ctime
.sec
);
1000 ce
->ce_mtime
= ntohl(ondisk
->mtime
.sec
);
1001 ce
->ce_dev
= ntohl(ondisk
->dev
);
1002 ce
->ce_ino
= ntohl(ondisk
->ino
);
1003 ce
->ce_mode
= ntohl(ondisk
->mode
);
1004 ce
->ce_uid
= ntohl(ondisk
->uid
);
1005 ce
->ce_gid
= ntohl(ondisk
->gid
);
1006 ce
->ce_size
= ntohl(ondisk
->size
);
1007 /* On-disk flags are just 16 bits */
1008 ce
->ce_flags
= ntohs(ondisk
->flags
);
1009 hashcpy(ce
->sha1
, ondisk
->sha1
);
1011 len
= ce
->ce_flags
& CE_NAMEMASK
;
1012 if (len
== CE_NAMEMASK
)
1013 len
= strlen(ondisk
->name
);
1015 * NEEDSWORK: If the original index is crafted, this copy could
1018 memcpy(ce
->name
, ondisk
->name
, len
+ 1);
1021 static inline size_t estimate_cache_size(size_t ondisk_size
, unsigned int entries
)
1025 per_entry
= sizeof(struct cache_entry
) - sizeof(struct ondisk_cache_entry
);
1028 * Alignment can cause differences. This should be "alignof", but
1029 * since that's a gcc'ism, just use the size of a pointer.
1031 per_entry
+= sizeof(void *);
1032 return ondisk_size
+ entries
*per_entry
;
1035 /* remember to discard_cache() before reading a different cache! */
1036 int read_index_from(struct index_state
*istate
, const char *path
)
1040 unsigned long src_offset
, dst_offset
;
1041 struct cache_header
*hdr
;
1047 return istate
->cache_nr
;
1050 istate
->timestamp
= 0;
1051 fd
= open(path
, O_RDONLY
);
1053 if (errno
== ENOENT
)
1055 die("index file open failed (%s)", strerror(errno
));
1059 die("cannot stat the open index (%s)", strerror(errno
));
1062 mmap_size
= xsize_t(st
.st_size
);
1063 if (mmap_size
< sizeof(struct cache_header
) + 20)
1064 die("index file smaller than expected");
1066 mmap
= xmmap(NULL
, mmap_size
, PROT_READ
| PROT_WRITE
, MAP_PRIVATE
, fd
, 0);
1068 if (mmap
== MAP_FAILED
)
1069 die("unable to map index file");
1072 if (verify_hdr(hdr
, mmap_size
) < 0)
1075 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
1076 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
1077 istate
->cache
= xcalloc(istate
->cache_alloc
, sizeof(struct cache_entry
*));
1080 * The disk format is actually larger than the in-memory format,
1081 * due to space for nsec etc, so even though the in-memory one
1082 * has room for a few more flags, we can allocate using the same
1085 istate
->alloc
= xmalloc(estimate_cache_size(mmap_size
, istate
->cache_nr
));
1087 src_offset
= sizeof(*hdr
);
1089 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1090 struct ondisk_cache_entry
*disk_ce
;
1091 struct cache_entry
*ce
;
1093 disk_ce
= (struct ondisk_cache_entry
*)((char *)mmap
+ src_offset
);
1094 ce
= (struct cache_entry
*)((char *)istate
->alloc
+ dst_offset
);
1095 convert_from_disk(disk_ce
, ce
);
1096 set_index_entry(istate
, i
, ce
);
1098 src_offset
+= ondisk_ce_size(ce
);
1099 dst_offset
+= ce_size(ce
);
1101 istate
->timestamp
= st
.st_mtime
;
1102 while (src_offset
<= mmap_size
- 20 - 8) {
1103 /* After an array of active_nr index entries,
1104 * there can be arbitrary number of extended
1105 * sections, each of which is prefixed with
1106 * extension name (4-byte) and section length
1107 * in 4-byte network byte order.
1109 unsigned long extsize
;
1110 memcpy(&extsize
, (char *)mmap
+ src_offset
+ 4, 4);
1111 extsize
= ntohl(extsize
);
1112 if (read_index_extension(istate
,
1113 (const char *) mmap
+ src_offset
,
1114 (char *) mmap
+ src_offset
+ 8,
1118 src_offset
+= extsize
;
1120 munmap(mmap
, mmap_size
);
1121 return istate
->cache_nr
;
1124 munmap(mmap
, mmap_size
);
1126 die("index file corrupt");
1129 int discard_index(struct index_state
*istate
)
1131 istate
->cache_nr
= 0;
1132 istate
->cache_changed
= 0;
1133 istate
->timestamp
= 0;
1134 free_hash(&istate
->name_hash
);
1135 cache_tree_free(&(istate
->cache_tree
));
1136 free(istate
->alloc
);
1137 istate
->alloc
= NULL
;
1139 /* no need to throw away allocated active_cache */
1143 int unmerged_index(const struct index_state
*istate
)
1146 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1147 if (ce_stage(istate
->cache
[i
]))
1153 #define WRITE_BUFFER_SIZE 8192
1154 static unsigned char write_buffer
[WRITE_BUFFER_SIZE
];
1155 static unsigned long write_buffer_len
;
1157 static int ce_write_flush(SHA_CTX
*context
, int fd
)
1159 unsigned int buffered
= write_buffer_len
;
1161 SHA1_Update(context
, write_buffer
, buffered
);
1162 if (write_in_full(fd
, write_buffer
, buffered
) != buffered
)
1164 write_buffer_len
= 0;
1169 static int ce_write(SHA_CTX
*context
, int fd
, void *data
, unsigned int len
)
1172 unsigned int buffered
= write_buffer_len
;
1173 unsigned int partial
= WRITE_BUFFER_SIZE
- buffered
;
1176 memcpy(write_buffer
+ buffered
, data
, partial
);
1177 buffered
+= partial
;
1178 if (buffered
== WRITE_BUFFER_SIZE
) {
1179 write_buffer_len
= buffered
;
1180 if (ce_write_flush(context
, fd
))
1184 write_buffer_len
= buffered
;
1186 data
= (char *) data
+ partial
;
1191 static int write_index_ext_header(SHA_CTX
*context
, int fd
,
1192 unsigned int ext
, unsigned int sz
)
1196 return ((ce_write(context
, fd
, &ext
, 4) < 0) ||
1197 (ce_write(context
, fd
, &sz
, 4) < 0)) ? -1 : 0;
1200 static int ce_flush(SHA_CTX
*context
, int fd
)
1202 unsigned int left
= write_buffer_len
;
1205 write_buffer_len
= 0;
1206 SHA1_Update(context
, write_buffer
, left
);
1209 /* Flush first if not enough space for SHA1 signature */
1210 if (left
+ 20 > WRITE_BUFFER_SIZE
) {
1211 if (write_in_full(fd
, write_buffer
, left
) != left
)
1216 /* Append the SHA1 signature at the end */
1217 SHA1_Final(write_buffer
+ left
, context
);
1219 return (write_in_full(fd
, write_buffer
, left
) != left
) ? -1 : 0;
1222 static void ce_smudge_racily_clean_entry(struct cache_entry
*ce
)
1225 * The only thing we care about in this function is to smudge the
1226 * falsely clean entry due to touch-update-touch race, so we leave
1227 * everything else as they are. We are called for entries whose
1228 * ce_mtime match the index file mtime.
1232 if (lstat(ce
->name
, &st
) < 0)
1234 if (ce_match_stat_basic(ce
, &st
))
1236 if (ce_modified_check_fs(ce
, &st
)) {
1237 /* This is "racily clean"; smudge it. Note that this
1238 * is a tricky code. At first glance, it may appear
1239 * that it can break with this sequence:
1241 * $ echo xyzzy >frotz
1242 * $ git-update-index --add frotz
1245 * $ echo filfre >nitfol
1246 * $ git-update-index --add nitfol
1248 * but it does not. When the second update-index runs,
1249 * it notices that the entry "frotz" has the same timestamp
1250 * as index, and if we were to smudge it by resetting its
1251 * size to zero here, then the object name recorded
1252 * in index is the 6-byte file but the cached stat information
1253 * becomes zero --- which would then match what we would
1254 * obtain from the filesystem next time we stat("frotz").
1256 * However, the second update-index, before calling
1257 * this function, notices that the cached size is 6
1258 * bytes and what is on the filesystem is an empty
1259 * file, and never calls us, so the cached size information
1260 * for "frotz" stays 6 which does not match the filesystem.
1266 static int ce_write_entry(SHA_CTX
*c
, int fd
, struct cache_entry
*ce
)
1268 int size
= ondisk_ce_size(ce
);
1269 struct ondisk_cache_entry
*ondisk
= xcalloc(1, size
);
1271 ondisk
->ctime
.sec
= htonl(ce
->ce_ctime
);
1272 ondisk
->ctime
.nsec
= 0;
1273 ondisk
->mtime
.sec
= htonl(ce
->ce_mtime
);
1274 ondisk
->mtime
.nsec
= 0;
1275 ondisk
->dev
= htonl(ce
->ce_dev
);
1276 ondisk
->ino
= htonl(ce
->ce_ino
);
1277 ondisk
->mode
= htonl(ce
->ce_mode
);
1278 ondisk
->uid
= htonl(ce
->ce_uid
);
1279 ondisk
->gid
= htonl(ce
->ce_gid
);
1280 ondisk
->size
= htonl(ce
->ce_size
);
1281 hashcpy(ondisk
->sha1
, ce
->sha1
);
1282 ondisk
->flags
= htons(ce
->ce_flags
);
1283 memcpy(ondisk
->name
, ce
->name
, ce_namelen(ce
));
1285 return ce_write(c
, fd
, ondisk
, size
);
1288 int write_index(const struct index_state
*istate
, int newfd
)
1291 struct cache_header hdr
;
1292 int i
, err
, removed
;
1293 struct cache_entry
**cache
= istate
->cache
;
1294 int entries
= istate
->cache_nr
;
1296 for (i
= removed
= 0; i
< entries
; i
++)
1297 if (cache
[i
]->ce_flags
& CE_REMOVE
)
1300 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
1301 hdr
.hdr_version
= htonl(2);
1302 hdr
.hdr_entries
= htonl(entries
- removed
);
1305 if (ce_write(&c
, newfd
, &hdr
, sizeof(hdr
)) < 0)
1308 for (i
= 0; i
< entries
; i
++) {
1309 struct cache_entry
*ce
= cache
[i
];
1310 if (ce
->ce_flags
& CE_REMOVE
)
1312 if (is_racy_timestamp(istate
, ce
))
1313 ce_smudge_racily_clean_entry(ce
);
1314 if (ce_write_entry(&c
, newfd
, ce
) < 0)
1318 /* Write extension data here */
1319 if (istate
->cache_tree
) {
1322 strbuf_init(&sb
, 0);
1323 cache_tree_write(&sb
, istate
->cache_tree
);
1324 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_TREE
, sb
.len
) < 0
1325 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
1326 strbuf_release(&sb
);
1330 return ce_flush(&c
, newfd
);