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
;
27 * This only updates the "non-critical" parts of the directory
28 * cache, ie the parts that aren't tracked by GIT, and only used
29 * to validate the cache.
31 void fill_stat_cache_info(struct cache_entry
*ce
, struct stat
*st
)
33 ce
->ce_ctime
.sec
= htonl(st
->st_ctime
);
34 ce
->ce_mtime
.sec
= htonl(st
->st_mtime
);
36 ce
->ce_ctime
.nsec
= htonl(st
->st_ctim
.tv_nsec
);
37 ce
->ce_mtime
.nsec
= htonl(st
->st_mtim
.tv_nsec
);
39 ce
->ce_dev
= htonl(st
->st_dev
);
40 ce
->ce_ino
= htonl(st
->st_ino
);
41 ce
->ce_uid
= htonl(st
->st_uid
);
42 ce
->ce_gid
= htonl(st
->st_gid
);
43 ce
->ce_size
= htonl(st
->st_size
);
46 ce
->ce_flags
|= htons(CE_VALID
);
49 static int ce_compare_data(struct cache_entry
*ce
, struct stat
*st
)
52 int fd
= open(ce
->name
, O_RDONLY
);
55 unsigned char sha1
[20];
56 if (!index_fd(sha1
, fd
, st
, 0, OBJ_BLOB
, ce
->name
))
57 match
= hashcmp(sha1
, ce
->sha1
);
58 /* index_fd() closed the file descriptor already */
63 static int ce_compare_link(struct cache_entry
*ce
, size_t expected_size
)
69 enum object_type type
;
72 target
= xmalloc(expected_size
);
73 len
= readlink(ce
->name
, target
, expected_size
);
74 if (len
!= expected_size
) {
78 buffer
= read_sha1_file(ce
->sha1
, &type
, &size
);
83 if (size
== expected_size
)
84 match
= memcmp(buffer
, target
, size
);
90 static int ce_compare_gitlink(struct cache_entry
*ce
)
92 unsigned char sha1
[20];
95 * We don't actually require that the .git directory
96 * under GITLINK directory be a valid git directory. It
97 * might even be missing (in case nobody populated that
100 * If so, we consider it always to match.
102 if (resolve_gitlink_ref(ce
->name
, "HEAD", sha1
) < 0)
104 return hashcmp(sha1
, ce
->sha1
);
107 static int ce_modified_check_fs(struct cache_entry
*ce
, struct stat
*st
)
109 switch (st
->st_mode
& S_IFMT
) {
111 if (ce_compare_data(ce
, st
))
115 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
119 if (S_ISGITLINK(ntohl(ce
->ce_mode
)))
127 static int ce_match_stat_basic(struct cache_entry
*ce
, struct stat
*st
)
129 unsigned int changed
= 0;
131 switch (ntohl(ce
->ce_mode
) & S_IFMT
) {
133 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
134 /* We consider only the owner x bit to be relevant for
137 if (trust_executable_bit
&&
138 (0100 & (ntohl(ce
->ce_mode
) ^ st
->st_mode
)))
139 changed
|= MODE_CHANGED
;
142 if (!S_ISLNK(st
->st_mode
) &&
143 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
144 changed
|= TYPE_CHANGED
;
147 if (!S_ISDIR(st
->st_mode
))
148 changed
|= TYPE_CHANGED
;
149 else if (ce_compare_gitlink(ce
))
150 changed
|= DATA_CHANGED
;
152 case 0: /* Special case: unmerged file in index */
153 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
155 die("internal error: ce_mode is %o", ntohl(ce
->ce_mode
));
157 if (ce
->ce_mtime
.sec
!= htonl(st
->st_mtime
))
158 changed
|= MTIME_CHANGED
;
159 if (ce
->ce_ctime
.sec
!= htonl(st
->st_ctime
))
160 changed
|= CTIME_CHANGED
;
164 * nsec seems unreliable - not all filesystems support it, so
165 * as long as it is in the inode cache you get right nsec
166 * but after it gets flushed, you get zero nsec.
168 if (ce
->ce_mtime
.nsec
!= htonl(st
->st_mtim
.tv_nsec
))
169 changed
|= MTIME_CHANGED
;
170 if (ce
->ce_ctime
.nsec
!= htonl(st
->st_ctim
.tv_nsec
))
171 changed
|= CTIME_CHANGED
;
174 if (ce
->ce_uid
!= htonl(st
->st_uid
) ||
175 ce
->ce_gid
!= htonl(st
->st_gid
))
176 changed
|= OWNER_CHANGED
;
177 if (ce
->ce_ino
!= htonl(st
->st_ino
))
178 changed
|= INODE_CHANGED
;
182 * st_dev breaks on network filesystems where different
183 * clients will have different views of what "device"
184 * the filesystem is on
186 if (ce
->ce_dev
!= htonl(st
->st_dev
))
187 changed
|= INODE_CHANGED
;
190 if (ce
->ce_size
!= htonl(st
->st_size
))
191 changed
|= DATA_CHANGED
;
196 int ie_match_stat(struct index_state
*istate
,
197 struct cache_entry
*ce
, struct stat
*st
,
198 unsigned int options
)
200 unsigned int changed
;
201 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
202 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
205 * If it's marked as always valid in the index, it's
206 * valid whatever the checked-out copy says.
208 if (!ignore_valid
&& (ce
->ce_flags
& htons(CE_VALID
)))
211 changed
= ce_match_stat_basic(ce
, st
);
214 * Within 1 second of this sequence:
215 * echo xyzzy >file && git-update-index --add file
216 * running this command:
218 * would give a falsely clean cache entry. The mtime and
219 * length match the cache, and other stat fields do not change.
221 * We could detect this at update-index time (the cache entry
222 * being registered/updated records the same time as "now")
223 * and delay the return from git-update-index, but that would
224 * effectively mean we can make at most one commit per second,
225 * which is not acceptable. Instead, we check cache entries
226 * whose mtime are the same as the index file timestamp more
227 * carefully than others.
231 istate
->timestamp
<= ntohl(ce
->ce_mtime
.sec
)) {
232 if (assume_racy_is_modified
)
233 changed
|= DATA_CHANGED
;
235 changed
|= ce_modified_check_fs(ce
, st
);
241 int ie_modified(struct index_state
*istate
,
242 struct cache_entry
*ce
, struct stat
*st
, unsigned int options
)
244 int changed
, changed_fs
;
246 changed
= ie_match_stat(istate
, ce
, st
, options
);
250 * If the mode or type has changed, there's no point in trying
251 * to refresh the entry - it's not going to match
253 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
256 /* Immediately after read-tree or update-index --cacheinfo,
257 * the length field is zero. For other cases the ce_size
258 * should match the SHA1 recorded in the index entry.
260 if ((changed
& DATA_CHANGED
) && ce
->ce_size
!= htonl(0))
263 changed_fs
= ce_modified_check_fs(ce
, st
);
265 return changed
| changed_fs
;
269 int base_name_compare(const char *name1
, int len1
, int mode1
,
270 const char *name2
, int len2
, int mode2
)
272 unsigned char c1
, c2
;
273 int len
= len1
< len2
? len1
: len2
;
276 cmp
= memcmp(name1
, name2
, len
);
281 if (!c1
&& S_ISDIR(mode1
))
283 if (!c2
&& S_ISDIR(mode2
))
285 return (c1
< c2
) ? -1 : (c1
> c2
) ? 1 : 0;
288 int cache_name_compare(const char *name1
, int flags1
, const char *name2
, int flags2
)
290 int len1
= flags1
& CE_NAMEMASK
;
291 int len2
= flags2
& CE_NAMEMASK
;
292 int len
= len1
< len2
? len1
: len2
;
295 cmp
= memcmp(name1
, name2
, len
);
304 flags1
&= CE_STAGEMASK
;
305 flags2
&= CE_STAGEMASK
;
314 int index_name_pos(struct index_state
*istate
, const char *name
, int namelen
)
319 last
= istate
->cache_nr
;
320 while (last
> first
) {
321 int next
= (last
+ first
) >> 1;
322 struct cache_entry
*ce
= istate
->cache
[next
];
323 int cmp
= cache_name_compare(name
, namelen
, ce
->name
, ntohs(ce
->ce_flags
));
335 /* Remove entry, return true if there are more entries to go.. */
336 int remove_index_entry_at(struct index_state
*istate
, int pos
)
338 istate
->cache_changed
= 1;
340 if (pos
>= istate
->cache_nr
)
342 memmove(istate
->cache
+ pos
,
343 istate
->cache
+ pos
+ 1,
344 (istate
->cache_nr
- pos
) * sizeof(struct cache_entry
*));
348 int remove_file_from_index(struct index_state
*istate
, const char *path
)
350 int pos
= index_name_pos(istate
, path
, strlen(path
));
353 cache_tree_invalidate_path(istate
->cache_tree
, path
);
354 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
355 remove_index_entry_at(istate
, pos
);
359 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
361 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
364 static int index_name_pos_also_unmerged(struct index_state
*istate
,
365 const char *path
, int namelen
)
367 int pos
= index_name_pos(istate
, path
, namelen
);
368 struct cache_entry
*ce
;
373 /* maybe unmerged? */
375 if (pos
>= istate
->cache_nr
||
376 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
379 /* order of preference: stage 2, 1, 3 */
380 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
381 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
382 !compare_name(ce
, path
, namelen
))
387 int add_file_to_index(struct index_state
*istate
, const char *path
, int verbose
)
389 int size
, namelen
, pos
;
391 struct cache_entry
*ce
;
392 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_RACY_IS_DIRTY
;
394 if (lstat(path
, &st
))
395 die("%s: unable to stat (%s)", path
, strerror(errno
));
397 if (!S_ISREG(st
.st_mode
) && !S_ISLNK(st
.st_mode
) && !S_ISDIR(st
.st_mode
))
398 die("%s: can only add regular files, symbolic links or git-directories", path
);
400 namelen
= strlen(path
);
401 if (S_ISDIR(st
.st_mode
)) {
402 while (namelen
&& path
[namelen
-1] == '/')
405 size
= cache_entry_size(namelen
);
406 ce
= xcalloc(1, size
);
407 memcpy(ce
->name
, path
, namelen
);
408 ce
->ce_flags
= htons(namelen
);
409 fill_stat_cache_info(ce
, &st
);
411 if (trust_executable_bit
&& has_symlinks
)
412 ce
->ce_mode
= create_ce_mode(st
.st_mode
);
414 /* If there is an existing entry, pick the mode bits and type
415 * from it, otherwise assume unexecutable regular file.
417 struct cache_entry
*ent
;
418 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
420 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
421 ce
->ce_mode
= ce_mode_from_stat(ent
, st
.st_mode
);
424 pos
= index_name_pos(istate
, ce
->name
, namelen
);
426 !ce_stage(istate
->cache
[pos
]) &&
427 !ie_match_stat(istate
, istate
->cache
[pos
], &st
, ce_option
)) {
428 /* Nothing changed, really */
433 if (index_path(ce
->sha1
, path
, &st
, 1))
434 die("unable to index file %s", path
);
435 if (add_index_entry(istate
, ce
, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
))
436 die("unable to add %s to index",path
);
438 printf("add '%s'\n", path
);
442 struct cache_entry
*make_cache_entry(unsigned int mode
,
443 const unsigned char *sha1
, const char *path
, int stage
,
447 struct cache_entry
*ce
;
449 if (!verify_path(path
))
453 size
= cache_entry_size(len
);
454 ce
= xcalloc(1, size
);
456 hashcpy(ce
->sha1
, sha1
);
457 memcpy(ce
->name
, path
, len
);
458 ce
->ce_flags
= create_ce_flags(len
, stage
);
459 ce
->ce_mode
= create_ce_mode(mode
);
462 return refresh_cache_entry(ce
, 0);
467 int ce_same_name(struct cache_entry
*a
, struct cache_entry
*b
)
469 int len
= ce_namelen(a
);
470 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
473 int ce_path_match(const struct cache_entry
*ce
, const char **pathspec
)
475 const char *match
, *name
;
481 len
= ce_namelen(ce
);
483 while ((match
= *pathspec
++) != NULL
) {
484 int matchlen
= strlen(match
);
487 if (memcmp(name
, match
, matchlen
))
489 if (matchlen
&& name
[matchlen
-1] == '/')
491 if (name
[matchlen
] == '/' || !name
[matchlen
])
500 * We fundamentally don't like some paths: we don't want
501 * dot or dot-dot anywhere, and for obvious reasons don't
502 * want to recurse into ".git" either.
504 * Also, we don't want double slashes or slashes at the
505 * end that can make pathnames ambiguous.
507 static int verify_dotfile(const char *rest
)
510 * The first character was '.', but that
511 * has already been discarded, we now test
515 /* "." is not allowed */
520 * ".git" followed by NUL or slash is bad. This
521 * shares the path end test with the ".." case.
531 if (rest
[1] == '\0' || rest
[1] == '/')
537 int verify_path(const char *path
)
542 if (is_absolute_path(path
))
543 return error("Cannot handle absolute path: %s", path
);
559 if (verify_dotfile(path
))
569 * Do we have another file that has the beginning components being a
570 * proper superset of the name we're trying to add?
572 static int has_file_name(struct index_state
*istate
,
573 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
576 int len
= ce_namelen(ce
);
577 int stage
= ce_stage(ce
);
578 const char *name
= ce
->name
;
580 while (pos
< istate
->cache_nr
) {
581 struct cache_entry
*p
= istate
->cache
[pos
++];
583 if (len
>= ce_namelen(p
))
585 if (memcmp(name
, p
->name
, len
))
587 if (ce_stage(p
) != stage
)
589 if (p
->name
[len
] != '/')
591 if (!ce_stage(p
) && !p
->ce_mode
)
596 remove_index_entry_at(istate
, --pos
);
602 * Do we have another file with a pathname that is a proper
603 * subset of the name we're trying to add?
605 static int has_dir_name(struct index_state
*istate
,
606 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
609 int stage
= ce_stage(ce
);
610 const char *name
= ce
->name
;
611 const char *slash
= name
+ ce_namelen(ce
);
619 if (slash
<= ce
->name
)
624 pos
= index_name_pos(istate
, name
, ntohs(create_ce_flags(len
, stage
)));
627 * Found one, but not so fast. This could
628 * be a marker that says "I was here, but
629 * I am being removed". Such an entry is
630 * not a part of the resulting tree, and
631 * it is Ok to have a directory at the same
634 if (stage
|| istate
->cache
[pos
]->ce_mode
) {
638 remove_index_entry_at(istate
, pos
);
646 * Trivial optimization: if we find an entry that
647 * already matches the sub-directory, then we know
648 * we're ok, and we can exit.
650 while (pos
< istate
->cache_nr
) {
651 struct cache_entry
*p
= istate
->cache
[pos
];
652 if ((ce_namelen(p
) <= len
) ||
653 (p
->name
[len
] != '/') ||
654 memcmp(p
->name
, name
, len
))
655 break; /* not our subdirectory */
656 if (ce_stage(p
) == stage
&& (stage
|| p
->ce_mode
))
657 /* p is at the same stage as our entry, and
658 * is a subdirectory of what we are looking
659 * at, so we cannot have conflicts at our
660 * level or anything shorter.
669 /* We may be in a situation where we already have path/file and path
670 * is being added, or we already have path and path/file is being
671 * added. Either one would result in a nonsense tree that has path
672 * twice when git-write-tree tries to write it out. Prevent it.
674 * If ok-to-replace is specified, we remove the conflicting entries
675 * from the cache so the caller should recompute the insert position.
676 * When this happens, we return non-zero.
678 static int check_file_directory_conflict(struct index_state
*istate
,
679 const struct cache_entry
*ce
,
680 int pos
, int ok_to_replace
)
685 * When ce is an "I am going away" entry, we allow it to be added
687 if (!ce_stage(ce
) && !ce
->ce_mode
)
691 * We check if the path is a sub-path of a subsequent pathname
692 * first, since removing those will not change the position
695 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
698 * Then check if the path might have a clashing sub-directory
701 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
704 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
707 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
708 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
709 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
711 cache_tree_invalidate_path(istate
->cache_tree
, ce
->name
);
712 pos
= index_name_pos(istate
, ce
->name
, ntohs(ce
->ce_flags
));
714 /* existing match? Just replace it. */
716 istate
->cache_changed
= 1;
717 istate
->cache
[pos
] = ce
;
723 * Inserting a merged entry ("stage 0") into the index
724 * will always replace all non-merged entries..
726 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
727 while (ce_same_name(istate
->cache
[pos
], ce
)) {
729 if (!remove_index_entry_at(istate
, pos
))
736 if (!verify_path(ce
->name
))
739 if (!skip_df_check
&&
740 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
742 return error("'%s' appears as both a file and as a directory",
744 pos
= index_name_pos(istate
, ce
->name
, ntohs(ce
->ce_flags
));
750 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
754 if (option
& ADD_CACHE_JUST_APPEND
)
755 pos
= istate
->cache_nr
;
758 ret
= add_index_entry_with_check(istate
, ce
, option
);
764 /* Make sure the array is big enough .. */
765 if (istate
->cache_nr
== istate
->cache_alloc
) {
766 istate
->cache_alloc
= alloc_nr(istate
->cache_alloc
);
767 istate
->cache
= xrealloc(istate
->cache
,
768 istate
->cache_alloc
* sizeof(struct cache_entry
*));
773 if (istate
->cache_nr
> pos
+ 1)
774 memmove(istate
->cache
+ pos
+ 1,
776 (istate
->cache_nr
- pos
- 1) * sizeof(ce
));
777 istate
->cache
[pos
] = ce
;
778 istate
->cache_changed
= 1;
783 * "refresh" does not calculate a new sha1 file or bring the
784 * cache up-to-date for mode/content changes. But what it
785 * _does_ do is to "re-match" the stat information of a file
786 * with the cache, so that you can refresh the cache for a
787 * file that hasn't been changed but where the stat entry is
790 * For example, you'd want to do this after doing a "git-read-tree",
791 * to link up the stat cache details with the proper files.
793 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
794 struct cache_entry
*ce
,
795 unsigned int options
, int *err
)
798 struct cache_entry
*updated
;
800 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
802 if (lstat(ce
->name
, &st
) < 0) {
808 changed
= ie_match_stat(istate
, ce
, &st
, options
);
811 * The path is unchanged. If we were told to ignore
812 * valid bit, then we did the actual stat check and
813 * found that the entry is unmodified. If the entry
814 * is not marked VALID, this is the place to mark it
815 * valid again, under "assume unchanged" mode.
817 if (ignore_valid
&& assume_unchanged
&&
818 !(ce
->ce_flags
& htons(CE_VALID
)))
819 ; /* mark this one VALID again */
824 if (ie_modified(istate
, ce
, &st
, options
)) {
831 updated
= xmalloc(size
);
832 memcpy(updated
, ce
, size
);
833 fill_stat_cache_info(updated
, &st
);
836 * If ignore_valid is not set, we should leave CE_VALID bit
837 * alone. Otherwise, paths marked with --no-assume-unchanged
838 * (i.e. things to be edited) will reacquire CE_VALID bit
839 * automatically, which is not really what we want.
841 if (!ignore_valid
&& assume_unchanged
&&
842 !(ce
->ce_flags
& htons(CE_VALID
)))
843 updated
->ce_flags
&= ~htons(CE_VALID
);
848 int refresh_index(struct index_state
*istate
, unsigned int flags
, const char **pathspec
, char *seen
)
852 int really
= (flags
& REFRESH_REALLY
) != 0;
853 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
854 int quiet
= (flags
& REFRESH_QUIET
) != 0;
855 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
856 unsigned int options
= really
? CE_MATCH_IGNORE_VALID
: 0;
858 for (i
= 0; i
< istate
->cache_nr
; i
++) {
859 struct cache_entry
*ce
, *new;
862 ce
= istate
->cache
[i
];
864 while ((i
< istate
->cache_nr
) &&
865 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
870 printf("%s: needs merge\n", ce
->name
);
875 if (pathspec
&& !match_pathspec(pathspec
, ce
->name
, strlen(ce
->name
), 0, seen
))
878 new = refresh_cache_ent(istate
, ce
, options
, &cache_errno
);
882 if (not_new
&& cache_errno
== ENOENT
)
884 if (really
&& cache_errno
== EINVAL
) {
885 /* If we are doing --really-refresh that
886 * means the index is not valid anymore.
888 ce
->ce_flags
&= ~htons(CE_VALID
);
889 istate
->cache_changed
= 1;
893 printf("%s: needs update\n", ce
->name
);
897 istate
->cache_changed
= 1;
898 /* You can NOT just free istate->cache[i] here, since it
899 * might not be necessarily malloc()ed but can also come
901 istate
->cache
[i
] = new;
906 struct cache_entry
*refresh_cache_entry(struct cache_entry
*ce
, int really
)
908 return refresh_cache_ent(&the_index
, ce
, really
, NULL
);
911 static int verify_hdr(struct cache_header
*hdr
, unsigned long size
)
914 unsigned char sha1
[20];
916 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
917 return error("bad signature");
918 if (hdr
->hdr_version
!= htonl(2))
919 return error("bad index version");
921 SHA1_Update(&c
, hdr
, size
- 20);
922 SHA1_Final(sha1
, &c
);
923 if (hashcmp(sha1
, (unsigned char *)hdr
+ size
- 20))
924 return error("bad index file sha1 signature");
928 static int read_index_extension(struct index_state
*istate
,
929 const char *ext
, void *data
, unsigned long sz
)
931 switch (CACHE_EXT(ext
)) {
933 istate
->cache_tree
= cache_tree_read(data
, sz
);
936 if (*ext
< 'A' || 'Z' < *ext
)
937 return error("index uses %.4s extension, which we do not understand",
939 fprintf(stderr
, "ignoring %.4s extension\n", ext
);
945 int read_index(struct index_state
*istate
)
947 return read_index_from(istate
, get_index_file());
950 /* remember to discard_cache() before reading a different cache! */
951 int read_index_from(struct index_state
*istate
, const char *path
)
955 unsigned long offset
;
956 struct cache_header
*hdr
;
960 return istate
->cache_nr
;
963 istate
->timestamp
= 0;
964 fd
= open(path
, O_RDONLY
);
968 die("index file open failed (%s)", strerror(errno
));
972 die("cannot stat the open index (%s)", strerror(errno
));
975 istate
->mmap_size
= xsize_t(st
.st_size
);
976 if (istate
->mmap_size
< sizeof(struct cache_header
) + 20)
977 die("index file smaller than expected");
979 istate
->mmap
= xmmap(NULL
, istate
->mmap_size
, PROT_READ
| PROT_WRITE
, MAP_PRIVATE
, fd
, 0);
983 if (verify_hdr(hdr
, istate
->mmap_size
) < 0)
986 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
987 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
988 istate
->cache
= xcalloc(istate
->cache_alloc
, sizeof(struct cache_entry
*));
990 offset
= sizeof(*hdr
);
991 for (i
= 0; i
< istate
->cache_nr
; i
++) {
992 struct cache_entry
*ce
;
994 ce
= (struct cache_entry
*)((char *)(istate
->mmap
) + offset
);
995 offset
= offset
+ ce_size(ce
);
996 istate
->cache
[i
] = ce
;
998 istate
->timestamp
= st
.st_mtime
;
999 while (offset
<= istate
->mmap_size
- 20 - 8) {
1000 /* After an array of active_nr index entries,
1001 * there can be arbitrary number of extended
1002 * sections, each of which is prefixed with
1003 * extension name (4-byte) and section length
1004 * in 4-byte network byte order.
1006 unsigned long extsize
;
1007 memcpy(&extsize
, (char *)(istate
->mmap
) + offset
+ 4, 4);
1008 extsize
= ntohl(extsize
);
1009 if (read_index_extension(istate
,
1010 ((const char *) (istate
->mmap
)) + offset
,
1011 (char *) (istate
->mmap
) + offset
+ 8,
1017 return istate
->cache_nr
;
1020 munmap(istate
->mmap
, istate
->mmap_size
);
1022 die("index file corrupt");
1025 int discard_index(struct index_state
*istate
)
1029 istate
->cache_nr
= 0;
1030 istate
->cache_changed
= 0;
1031 istate
->timestamp
= 0;
1032 cache_tree_free(&(istate
->cache_tree
));
1033 if (istate
->mmap
== NULL
)
1035 ret
= munmap(istate
->mmap
, istate
->mmap_size
);
1036 istate
->mmap
= NULL
;
1037 istate
->mmap_size
= 0;
1039 /* no need to throw away allocated active_cache */
1043 #define WRITE_BUFFER_SIZE 8192
1044 static unsigned char write_buffer
[WRITE_BUFFER_SIZE
];
1045 static unsigned long write_buffer_len
;
1047 static int ce_write_flush(SHA_CTX
*context
, int fd
)
1049 unsigned int buffered
= write_buffer_len
;
1051 SHA1_Update(context
, write_buffer
, buffered
);
1052 if (write_in_full(fd
, write_buffer
, buffered
) != buffered
)
1054 write_buffer_len
= 0;
1059 static int ce_write(SHA_CTX
*context
, int fd
, void *data
, unsigned int len
)
1062 unsigned int buffered
= write_buffer_len
;
1063 unsigned int partial
= WRITE_BUFFER_SIZE
- buffered
;
1066 memcpy(write_buffer
+ buffered
, data
, partial
);
1067 buffered
+= partial
;
1068 if (buffered
== WRITE_BUFFER_SIZE
) {
1069 write_buffer_len
= buffered
;
1070 if (ce_write_flush(context
, fd
))
1074 write_buffer_len
= buffered
;
1076 data
= (char *) data
+ partial
;
1081 static int write_index_ext_header(SHA_CTX
*context
, int fd
,
1082 unsigned int ext
, unsigned int sz
)
1086 return ((ce_write(context
, fd
, &ext
, 4) < 0) ||
1087 (ce_write(context
, fd
, &sz
, 4) < 0)) ? -1 : 0;
1090 static int ce_flush(SHA_CTX
*context
, int fd
)
1092 unsigned int left
= write_buffer_len
;
1095 write_buffer_len
= 0;
1096 SHA1_Update(context
, write_buffer
, left
);
1099 /* Flush first if not enough space for SHA1 signature */
1100 if (left
+ 20 > WRITE_BUFFER_SIZE
) {
1101 if (write_in_full(fd
, write_buffer
, left
) != left
)
1106 /* Append the SHA1 signature at the end */
1107 SHA1_Final(write_buffer
+ left
, context
);
1109 return (write_in_full(fd
, write_buffer
, left
) != left
) ? -1 : 0;
1112 static void ce_smudge_racily_clean_entry(struct cache_entry
*ce
)
1115 * The only thing we care about in this function is to smudge the
1116 * falsely clean entry due to touch-update-touch race, so we leave
1117 * everything else as they are. We are called for entries whose
1118 * ce_mtime match the index file mtime.
1122 if (lstat(ce
->name
, &st
) < 0)
1124 if (ce_match_stat_basic(ce
, &st
))
1126 if (ce_modified_check_fs(ce
, &st
)) {
1127 /* This is "racily clean"; smudge it. Note that this
1128 * is a tricky code. At first glance, it may appear
1129 * that it can break with this sequence:
1131 * $ echo xyzzy >frotz
1132 * $ git-update-index --add frotz
1135 * $ echo filfre >nitfol
1136 * $ git-update-index --add nitfol
1138 * but it does not. When the second update-index runs,
1139 * it notices that the entry "frotz" has the same timestamp
1140 * as index, and if we were to smudge it by resetting its
1141 * size to zero here, then the object name recorded
1142 * in index is the 6-byte file but the cached stat information
1143 * becomes zero --- which would then match what we would
1144 * obtain from the filesystem next time we stat("frotz").
1146 * However, the second update-index, before calling
1147 * this function, notices that the cached size is 6
1148 * bytes and what is on the filesystem is an empty
1149 * file, and never calls us, so the cached size information
1150 * for "frotz" stays 6 which does not match the filesystem.
1152 ce
->ce_size
= htonl(0);
1153 trace_printf("trace: index: Index object for '%s' smudged due to being racily clean\n", ce
->name
);
1157 int write_index(struct index_state
*istate
, int newfd
)
1160 struct cache_header hdr
;
1161 int i
, err
, removed
;
1162 struct cache_entry
**cache
= istate
->cache
;
1163 int entries
= istate
->cache_nr
;
1165 for (i
= removed
= 0; i
< entries
; i
++)
1166 if (!cache
[i
]->ce_mode
)
1169 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
1170 hdr
.hdr_version
= htonl(2);
1171 hdr
.hdr_entries
= htonl(entries
- removed
);
1174 if (ce_write(&c
, newfd
, &hdr
, sizeof(hdr
)) < 0)
1177 for (i
= 0; i
< entries
; i
++) {
1178 struct cache_entry
*ce
= cache
[i
];
1181 if (istate
->timestamp
&&
1182 istate
->timestamp
<= ntohl(ce
->ce_mtime
.sec
))
1183 ce_smudge_racily_clean_entry(ce
);
1184 if (ce_write(&c
, newfd
, ce
, ce_size(ce
)) < 0)
1188 /* Write extension data here */
1189 if (istate
->cache_tree
) {
1192 strbuf_init(&sb
, 0);
1193 cache_tree_write(&sb
, istate
->cache_tree
);
1194 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_TREE
, sb
.len
) < 0
1195 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
1196 strbuf_release(&sb
);
1200 return ce_flush(&c
, newfd
);