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
= st
->st_ctime
;
34 ce
->ce_mtime
= st
->st_mtime
;
35 ce
->ce_dev
= st
->st_dev
;
36 ce
->ce_ino
= st
->st_ino
;
37 ce
->ce_uid
= st
->st_uid
;
38 ce
->ce_gid
= st
->st_gid
;
39 ce
->ce_size
= st
->st_size
;
42 ce
->ce_flags
|= CE_VALID
;
44 if (S_ISREG(st
->st_mode
))
48 static int ce_compare_data(struct cache_entry
*ce
, struct stat
*st
)
51 int fd
= open(ce
->name
, O_RDONLY
);
54 unsigned char sha1
[20];
55 if (!index_fd(sha1
, fd
, st
, 0, OBJ_BLOB
, ce
->name
))
56 match
= hashcmp(sha1
, ce
->sha1
);
57 /* index_fd() closed the file descriptor already */
62 static int ce_compare_link(struct cache_entry
*ce
, size_t expected_size
)
68 enum object_type type
;
71 target
= xmalloc(expected_size
);
72 len
= readlink(ce
->name
, target
, expected_size
);
73 if (len
!= expected_size
) {
77 buffer
= read_sha1_file(ce
->sha1
, &type
, &size
);
82 if (size
== expected_size
)
83 match
= memcmp(buffer
, target
, size
);
89 static int ce_compare_gitlink(struct cache_entry
*ce
)
91 unsigned char sha1
[20];
94 * We don't actually require that the .git directory
95 * under GITLINK directory be a valid git directory. It
96 * might even be missing (in case nobody populated that
99 * If so, we consider it always to match.
101 if (resolve_gitlink_ref(ce
->name
, "HEAD", sha1
) < 0)
103 return hashcmp(sha1
, ce
->sha1
);
106 static int ce_modified_check_fs(struct cache_entry
*ce
, struct stat
*st
)
108 switch (st
->st_mode
& S_IFMT
) {
110 if (ce_compare_data(ce
, st
))
114 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
118 if (S_ISGITLINK(ce
->ce_mode
))
126 static int ce_match_stat_basic(struct cache_entry
*ce
, struct stat
*st
)
128 unsigned int changed
= 0;
130 if (ce
->ce_flags
& CE_REMOVE
)
131 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
133 switch (ce
->ce_mode
& S_IFMT
) {
135 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
136 /* We consider only the owner x bit to be relevant for
139 if (trust_executable_bit
&&
140 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
141 changed
|= MODE_CHANGED
;
144 if (!S_ISLNK(st
->st_mode
) &&
145 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
146 changed
|= TYPE_CHANGED
;
149 if (!S_ISDIR(st
->st_mode
))
150 changed
|= TYPE_CHANGED
;
151 else if (ce_compare_gitlink(ce
))
152 changed
|= DATA_CHANGED
;
155 die("internal error: ce_mode is %o", ce
->ce_mode
);
157 if (ce
->ce_mtime
!= (unsigned int) st
->st_mtime
)
158 changed
|= MTIME_CHANGED
;
159 if (ce
->ce_ctime
!= (unsigned int) st
->st_ctime
)
160 changed
|= CTIME_CHANGED
;
162 if (ce
->ce_uid
!= (unsigned int) st
->st_uid
||
163 ce
->ce_gid
!= (unsigned int) st
->st_gid
)
164 changed
|= OWNER_CHANGED
;
165 if (ce
->ce_ino
!= (unsigned int) st
->st_ino
)
166 changed
|= INODE_CHANGED
;
170 * st_dev breaks on network filesystems where different
171 * clients will have different views of what "device"
172 * the filesystem is on
174 if (ce
->ce_dev
!= (unsigned int) st
->st_dev
)
175 changed
|= INODE_CHANGED
;
178 if (ce
->ce_size
!= (unsigned int) st
->st_size
)
179 changed
|= DATA_CHANGED
;
184 int ie_match_stat(struct index_state
*istate
,
185 struct cache_entry
*ce
, struct stat
*st
,
186 unsigned int options
)
188 unsigned int changed
;
189 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
190 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
193 * If it's marked as always valid in the index, it's
194 * valid whatever the checked-out copy says.
196 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
199 changed
= ce_match_stat_basic(ce
, st
);
202 * Within 1 second of this sequence:
203 * echo xyzzy >file && git-update-index --add file
204 * running this command:
206 * would give a falsely clean cache entry. The mtime and
207 * length match the cache, and other stat fields do not change.
209 * We could detect this at update-index time (the cache entry
210 * being registered/updated records the same time as "now")
211 * and delay the return from git-update-index, but that would
212 * effectively mean we can make at most one commit per second,
213 * which is not acceptable. Instead, we check cache entries
214 * whose mtime are the same as the index file timestamp more
215 * carefully than others.
219 istate
->timestamp
<= ce
->ce_mtime
) {
220 if (assume_racy_is_modified
)
221 changed
|= DATA_CHANGED
;
223 changed
|= ce_modified_check_fs(ce
, st
);
229 int ie_modified(struct index_state
*istate
,
230 struct cache_entry
*ce
, struct stat
*st
, unsigned int options
)
232 int changed
, changed_fs
;
234 changed
= ie_match_stat(istate
, ce
, st
, options
);
238 * If the mode or type has changed, there's no point in trying
239 * to refresh the entry - it's not going to match
241 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
244 /* Immediately after read-tree or update-index --cacheinfo,
245 * the length field is zero. For other cases the ce_size
246 * should match the SHA1 recorded in the index entry.
248 if ((changed
& DATA_CHANGED
) && ce
->ce_size
!= 0)
251 changed_fs
= ce_modified_check_fs(ce
, st
);
253 return changed
| changed_fs
;
257 int base_name_compare(const char *name1
, int len1
, int mode1
,
258 const char *name2
, int len2
, int mode2
)
260 unsigned char c1
, c2
;
261 int len
= len1
< len2
? len1
: len2
;
264 cmp
= memcmp(name1
, name2
, len
);
269 if (!c1
&& S_ISDIR(mode1
))
271 if (!c2
&& S_ISDIR(mode2
))
273 return (c1
< c2
) ? -1 : (c1
> c2
) ? 1 : 0;
276 int cache_name_compare(const char *name1
, int flags1
, const char *name2
, int flags2
)
278 int len1
= flags1
& CE_NAMEMASK
;
279 int len2
= flags2
& CE_NAMEMASK
;
280 int len
= len1
< len2
? len1
: len2
;
283 cmp
= memcmp(name1
, name2
, len
);
292 flags1
&= CE_STAGEMASK
;
293 flags2
&= CE_STAGEMASK
;
302 int index_name_pos(struct index_state
*istate
, const char *name
, int namelen
)
307 last
= istate
->cache_nr
;
308 while (last
> first
) {
309 int next
= (last
+ first
) >> 1;
310 struct cache_entry
*ce
= istate
->cache
[next
];
311 int cmp
= cache_name_compare(name
, namelen
, ce
->name
, ce
->ce_flags
);
323 /* Remove entry, return true if there are more entries to go.. */
324 int remove_index_entry_at(struct index_state
*istate
, int pos
)
326 istate
->cache_changed
= 1;
328 if (pos
>= istate
->cache_nr
)
330 memmove(istate
->cache
+ pos
,
331 istate
->cache
+ pos
+ 1,
332 (istate
->cache_nr
- pos
) * sizeof(struct cache_entry
*));
336 int remove_file_from_index(struct index_state
*istate
, const char *path
)
338 int pos
= index_name_pos(istate
, path
, strlen(path
));
341 cache_tree_invalidate_path(istate
->cache_tree
, path
);
342 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
343 remove_index_entry_at(istate
, pos
);
347 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
349 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
352 static int index_name_pos_also_unmerged(struct index_state
*istate
,
353 const char *path
, int namelen
)
355 int pos
= index_name_pos(istate
, path
, namelen
);
356 struct cache_entry
*ce
;
361 /* maybe unmerged? */
363 if (pos
>= istate
->cache_nr
||
364 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
367 /* order of preference: stage 2, 1, 3 */
368 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
369 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
370 !compare_name(ce
, path
, namelen
))
375 int add_file_to_index(struct index_state
*istate
, const char *path
, int verbose
)
377 int size
, namelen
, pos
;
379 struct cache_entry
*ce
;
380 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_RACY_IS_DIRTY
;
382 if (lstat(path
, &st
))
383 die("%s: unable to stat (%s)", path
, strerror(errno
));
385 if (!S_ISREG(st
.st_mode
) && !S_ISLNK(st
.st_mode
) && !S_ISDIR(st
.st_mode
))
386 die("%s: can only add regular files, symbolic links or git-directories", path
);
388 namelen
= strlen(path
);
389 if (S_ISDIR(st
.st_mode
)) {
390 while (namelen
&& path
[namelen
-1] == '/')
393 size
= cache_entry_size(namelen
);
394 ce
= xcalloc(1, size
);
395 memcpy(ce
->name
, path
, namelen
);
396 ce
->ce_flags
= namelen
;
397 fill_stat_cache_info(ce
, &st
);
399 if (trust_executable_bit
&& has_symlinks
)
400 ce
->ce_mode
= create_ce_mode(st
.st_mode
);
402 /* If there is an existing entry, pick the mode bits and type
403 * from it, otherwise assume unexecutable regular file.
405 struct cache_entry
*ent
;
406 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
408 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
409 ce
->ce_mode
= ce_mode_from_stat(ent
, st
.st_mode
);
412 pos
= index_name_pos(istate
, ce
->name
, namelen
);
414 !ce_stage(istate
->cache
[pos
]) &&
415 !ie_match_stat(istate
, istate
->cache
[pos
], &st
, ce_option
)) {
416 /* Nothing changed, really */
418 ce_mark_uptodate(istate
->cache
[pos
]);
422 if (index_path(ce
->sha1
, path
, &st
, 1))
423 die("unable to index file %s", path
);
424 if (add_index_entry(istate
, ce
, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
))
425 die("unable to add %s to index",path
);
427 printf("add '%s'\n", path
);
431 struct cache_entry
*make_cache_entry(unsigned int mode
,
432 const unsigned char *sha1
, const char *path
, int stage
,
436 struct cache_entry
*ce
;
438 if (!verify_path(path
))
442 size
= cache_entry_size(len
);
443 ce
= xcalloc(1, size
);
445 hashcpy(ce
->sha1
, sha1
);
446 memcpy(ce
->name
, path
, len
);
447 ce
->ce_flags
= create_ce_flags(len
, stage
);
448 ce
->ce_mode
= create_ce_mode(mode
);
451 return refresh_cache_entry(ce
, 0);
456 int ce_same_name(struct cache_entry
*a
, struct cache_entry
*b
)
458 int len
= ce_namelen(a
);
459 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
462 int ce_path_match(const struct cache_entry
*ce
, const char **pathspec
)
464 const char *match
, *name
;
470 len
= ce_namelen(ce
);
472 while ((match
= *pathspec
++) != NULL
) {
473 int matchlen
= strlen(match
);
476 if (memcmp(name
, match
, matchlen
))
478 if (matchlen
&& name
[matchlen
-1] == '/')
480 if (name
[matchlen
] == '/' || !name
[matchlen
])
489 * We fundamentally don't like some paths: we don't want
490 * dot or dot-dot anywhere, and for obvious reasons don't
491 * want to recurse into ".git" either.
493 * Also, we don't want double slashes or slashes at the
494 * end that can make pathnames ambiguous.
496 static int verify_dotfile(const char *rest
)
499 * The first character was '.', but that
500 * has already been discarded, we now test
504 /* "." is not allowed */
509 * ".git" followed by NUL or slash is bad. This
510 * shares the path end test with the ".." case.
520 if (rest
[1] == '\0' || rest
[1] == '/')
526 int verify_path(const char *path
)
543 if (verify_dotfile(path
))
553 * Do we have another file that has the beginning components being a
554 * proper superset of the name we're trying to add?
556 static int has_file_name(struct index_state
*istate
,
557 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
560 int len
= ce_namelen(ce
);
561 int stage
= ce_stage(ce
);
562 const char *name
= ce
->name
;
564 while (pos
< istate
->cache_nr
) {
565 struct cache_entry
*p
= istate
->cache
[pos
++];
567 if (len
>= ce_namelen(p
))
569 if (memcmp(name
, p
->name
, len
))
571 if (ce_stage(p
) != stage
)
573 if (p
->name
[len
] != '/')
575 if (p
->ce_flags
& CE_REMOVE
)
580 remove_index_entry_at(istate
, --pos
);
586 * Do we have another file with a pathname that is a proper
587 * subset of the name we're trying to add?
589 static int has_dir_name(struct index_state
*istate
,
590 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
593 int stage
= ce_stage(ce
);
594 const char *name
= ce
->name
;
595 const char *slash
= name
+ ce_namelen(ce
);
603 if (slash
<= ce
->name
)
608 pos
= index_name_pos(istate
, name
, create_ce_flags(len
, stage
));
611 * Found one, but not so fast. This could
612 * be a marker that says "I was here, but
613 * I am being removed". Such an entry is
614 * not a part of the resulting tree, and
615 * it is Ok to have a directory at the same
618 if (stage
|| istate
->cache
[pos
]->ce_mode
) {
622 remove_index_entry_at(istate
, pos
);
630 * Trivial optimization: if we find an entry that
631 * already matches the sub-directory, then we know
632 * we're ok, and we can exit.
634 while (pos
< istate
->cache_nr
) {
635 struct cache_entry
*p
= istate
->cache
[pos
];
636 if ((ce_namelen(p
) <= len
) ||
637 (p
->name
[len
] != '/') ||
638 memcmp(p
->name
, name
, len
))
639 break; /* not our subdirectory */
640 if (ce_stage(p
) == stage
&& (stage
|| p
->ce_mode
))
641 /* p is at the same stage as our entry, and
642 * is a subdirectory of what we are looking
643 * at, so we cannot have conflicts at our
644 * level or anything shorter.
653 /* We may be in a situation where we already have path/file and path
654 * is being added, or we already have path and path/file is being
655 * added. Either one would result in a nonsense tree that has path
656 * twice when git-write-tree tries to write it out. Prevent it.
658 * If ok-to-replace is specified, we remove the conflicting entries
659 * from the cache so the caller should recompute the insert position.
660 * When this happens, we return non-zero.
662 static int check_file_directory_conflict(struct index_state
*istate
,
663 const struct cache_entry
*ce
,
664 int pos
, int ok_to_replace
)
669 * When ce is an "I am going away" entry, we allow it to be added
671 if (ce
->ce_flags
& CE_REMOVE
)
675 * We check if the path is a sub-path of a subsequent pathname
676 * first, since removing those will not change the position
679 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
682 * Then check if the path might have a clashing sub-directory
685 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
688 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
691 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
692 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
693 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
695 cache_tree_invalidate_path(istate
->cache_tree
, ce
->name
);
696 pos
= index_name_pos(istate
, ce
->name
, ce
->ce_flags
);
698 /* existing match? Just replace it. */
700 istate
->cache_changed
= 1;
701 istate
->cache
[pos
] = ce
;
707 * Inserting a merged entry ("stage 0") into the index
708 * will always replace all non-merged entries..
710 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
711 while (ce_same_name(istate
->cache
[pos
], ce
)) {
713 if (!remove_index_entry_at(istate
, pos
))
720 if (!verify_path(ce
->name
))
723 if (!skip_df_check
&&
724 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
726 return error("'%s' appears as both a file and as a directory",
728 pos
= index_name_pos(istate
, ce
->name
, ce
->ce_flags
);
734 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
738 if (option
& ADD_CACHE_JUST_APPEND
)
739 pos
= istate
->cache_nr
;
742 ret
= add_index_entry_with_check(istate
, ce
, option
);
748 /* Make sure the array is big enough .. */
749 if (istate
->cache_nr
== istate
->cache_alloc
) {
750 istate
->cache_alloc
= alloc_nr(istate
->cache_alloc
);
751 istate
->cache
= xrealloc(istate
->cache
,
752 istate
->cache_alloc
* sizeof(struct cache_entry
*));
757 if (istate
->cache_nr
> pos
+ 1)
758 memmove(istate
->cache
+ pos
+ 1,
760 (istate
->cache_nr
- pos
- 1) * sizeof(ce
));
761 istate
->cache
[pos
] = ce
;
762 istate
->cache_changed
= 1;
767 * "refresh" does not calculate a new sha1 file or bring the
768 * cache up-to-date for mode/content changes. But what it
769 * _does_ do is to "re-match" the stat information of a file
770 * with the cache, so that you can refresh the cache for a
771 * file that hasn't been changed but where the stat entry is
774 * For example, you'd want to do this after doing a "git-read-tree",
775 * to link up the stat cache details with the proper files.
777 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
778 struct cache_entry
*ce
,
779 unsigned int options
, int *err
)
782 struct cache_entry
*updated
;
784 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
789 if (lstat(ce
->name
, &st
) < 0) {
795 changed
= ie_match_stat(istate
, ce
, &st
, options
);
798 * The path is unchanged. If we were told to ignore
799 * valid bit, then we did the actual stat check and
800 * found that the entry is unmodified. If the entry
801 * is not marked VALID, this is the place to mark it
802 * valid again, under "assume unchanged" mode.
804 if (ignore_valid
&& assume_unchanged
&&
805 !(ce
->ce_flags
& CE_VALID
))
806 ; /* mark this one VALID again */
809 * We do not mark the index itself "modified"
810 * because CE_UPTODATE flag is in-core only;
811 * we are not going to write this change out.
813 ce_mark_uptodate(ce
);
818 if (ie_modified(istate
, ce
, &st
, options
)) {
825 updated
= xmalloc(size
);
826 memcpy(updated
, ce
, size
);
827 fill_stat_cache_info(updated
, &st
);
829 * If ignore_valid is not set, we should leave CE_VALID bit
830 * alone. Otherwise, paths marked with --no-assume-unchanged
831 * (i.e. things to be edited) will reacquire CE_VALID bit
832 * automatically, which is not really what we want.
834 if (!ignore_valid
&& assume_unchanged
&&
835 !(ce
->ce_flags
& CE_VALID
))
836 updated
->ce_flags
&= ~CE_VALID
;
841 int refresh_index(struct index_state
*istate
, unsigned int flags
, const char **pathspec
, char *seen
)
845 int really
= (flags
& REFRESH_REALLY
) != 0;
846 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
847 int quiet
= (flags
& REFRESH_QUIET
) != 0;
848 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
849 unsigned int options
= really
? CE_MATCH_IGNORE_VALID
: 0;
851 for (i
= 0; i
< istate
->cache_nr
; i
++) {
852 struct cache_entry
*ce
, *new;
855 ce
= istate
->cache
[i
];
857 while ((i
< istate
->cache_nr
) &&
858 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
863 printf("%s: needs merge\n", ce
->name
);
868 if (pathspec
&& !match_pathspec(pathspec
, ce
->name
, strlen(ce
->name
), 0, seen
))
871 new = refresh_cache_ent(istate
, ce
, options
, &cache_errno
);
875 if (not_new
&& cache_errno
== ENOENT
)
877 if (really
&& cache_errno
== EINVAL
) {
878 /* If we are doing --really-refresh that
879 * means the index is not valid anymore.
881 ce
->ce_flags
&= ~CE_VALID
;
882 istate
->cache_changed
= 1;
886 printf("%s: needs update\n", ce
->name
);
890 istate
->cache_changed
= 1;
891 /* You can NOT just free istate->cache[i] here, since it
892 * might not be necessarily malloc()ed but can also come
894 istate
->cache
[i
] = new;
899 struct cache_entry
*refresh_cache_entry(struct cache_entry
*ce
, int really
)
901 return refresh_cache_ent(&the_index
, ce
, really
, NULL
);
904 static int verify_hdr(struct cache_header
*hdr
, unsigned long size
)
907 unsigned char sha1
[20];
909 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
910 return error("bad signature");
911 if (hdr
->hdr_version
!= htonl(2))
912 return error("bad index version");
914 SHA1_Update(&c
, hdr
, size
- 20);
915 SHA1_Final(sha1
, &c
);
916 if (hashcmp(sha1
, (unsigned char *)hdr
+ size
- 20))
917 return error("bad index file sha1 signature");
921 static int read_index_extension(struct index_state
*istate
,
922 const char *ext
, void *data
, unsigned long sz
)
924 switch (CACHE_EXT(ext
)) {
926 istate
->cache_tree
= cache_tree_read(data
, sz
);
929 if (*ext
< 'A' || 'Z' < *ext
)
930 return error("index uses %.4s extension, which we do not understand",
932 fprintf(stderr
, "ignoring %.4s extension\n", ext
);
938 int read_index(struct index_state
*istate
)
940 return read_index_from(istate
, get_index_file());
943 static void convert_from_disk(struct ondisk_cache_entry
*ondisk
, struct cache_entry
*ce
)
947 ce
->ce_ctime
= ntohl(ondisk
->ctime
.sec
);
948 ce
->ce_mtime
= ntohl(ondisk
->mtime
.sec
);
949 ce
->ce_dev
= ntohl(ondisk
->dev
);
950 ce
->ce_ino
= ntohl(ondisk
->ino
);
951 ce
->ce_mode
= ntohl(ondisk
->mode
);
952 ce
->ce_uid
= ntohl(ondisk
->uid
);
953 ce
->ce_gid
= ntohl(ondisk
->gid
);
954 ce
->ce_size
= ntohl(ondisk
->size
);
955 /* On-disk flags are just 16 bits */
956 ce
->ce_flags
= ntohs(ondisk
->flags
);
957 hashcpy(ce
->sha1
, ondisk
->sha1
);
959 len
= ce
->ce_flags
& CE_NAMEMASK
;
960 if (len
== CE_NAMEMASK
)
961 len
= strlen(ondisk
->name
);
963 * NEEDSWORK: If the original index is crafted, this copy could
966 memcpy(ce
->name
, ondisk
->name
, len
+ 1);
969 /* remember to discard_cache() before reading a different cache! */
970 int read_index_from(struct index_state
*istate
, const char *path
)
974 unsigned long src_offset
, dst_offset
;
975 struct cache_header
*hdr
;
981 return istate
->cache_nr
;
984 istate
->timestamp
= 0;
985 fd
= open(path
, O_RDONLY
);
989 die("index file open failed (%s)", strerror(errno
));
993 die("cannot stat the open index (%s)", strerror(errno
));
996 mmap_size
= xsize_t(st
.st_size
);
997 if (mmap_size
< sizeof(struct cache_header
) + 20)
998 die("index file smaller than expected");
1000 mmap
= xmmap(NULL
, mmap_size
, PROT_READ
| PROT_WRITE
, MAP_PRIVATE
, fd
, 0);
1002 if (mmap
== MAP_FAILED
)
1003 die("unable to map index file");
1006 if (verify_hdr(hdr
, mmap_size
) < 0)
1009 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
1010 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
1011 istate
->cache
= xcalloc(istate
->cache_alloc
, sizeof(struct cache_entry
*));
1014 * The disk format is actually larger than the in-memory format,
1015 * due to space for nsec etc, so even though the in-memory one
1016 * has room for a few more flags, we can allocate using the same
1019 istate
->alloc
= xmalloc(mmap_size
);
1021 src_offset
= sizeof(*hdr
);
1023 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1024 struct ondisk_cache_entry
*disk_ce
;
1025 struct cache_entry
*ce
;
1027 disk_ce
= (struct ondisk_cache_entry
*)((char *)mmap
+ src_offset
);
1028 ce
= (struct cache_entry
*)((char *)istate
->alloc
+ dst_offset
);
1029 convert_from_disk(disk_ce
, ce
);
1030 istate
->cache
[i
] = ce
;
1032 src_offset
+= ondisk_ce_size(ce
);
1033 dst_offset
+= ce_size(ce
);
1035 istate
->timestamp
= st
.st_mtime
;
1036 while (src_offset
<= mmap_size
- 20 - 8) {
1037 /* After an array of active_nr index entries,
1038 * there can be arbitrary number of extended
1039 * sections, each of which is prefixed with
1040 * extension name (4-byte) and section length
1041 * in 4-byte network byte order.
1043 unsigned long extsize
;
1044 memcpy(&extsize
, (char *)mmap
+ src_offset
+ 4, 4);
1045 extsize
= ntohl(extsize
);
1046 if (read_index_extension(istate
,
1047 (const char *) mmap
+ src_offset
,
1048 (char *) mmap
+ src_offset
+ 8,
1052 src_offset
+= extsize
;
1054 munmap(mmap
, mmap_size
);
1055 return istate
->cache_nr
;
1058 munmap(mmap
, mmap_size
);
1060 die("index file corrupt");
1063 int discard_index(struct index_state
*istate
)
1065 istate
->cache_nr
= 0;
1066 istate
->cache_changed
= 0;
1067 istate
->timestamp
= 0;
1068 cache_tree_free(&(istate
->cache_tree
));
1069 free(istate
->alloc
);
1070 istate
->alloc
= NULL
;
1072 /* no need to throw away allocated active_cache */
1076 #define WRITE_BUFFER_SIZE 8192
1077 static unsigned char write_buffer
[WRITE_BUFFER_SIZE
];
1078 static unsigned long write_buffer_len
;
1080 static int ce_write_flush(SHA_CTX
*context
, int fd
)
1082 unsigned int buffered
= write_buffer_len
;
1084 SHA1_Update(context
, write_buffer
, buffered
);
1085 if (write_in_full(fd
, write_buffer
, buffered
) != buffered
)
1087 write_buffer_len
= 0;
1092 static int ce_write(SHA_CTX
*context
, int fd
, void *data
, unsigned int len
)
1095 unsigned int buffered
= write_buffer_len
;
1096 unsigned int partial
= WRITE_BUFFER_SIZE
- buffered
;
1099 memcpy(write_buffer
+ buffered
, data
, partial
);
1100 buffered
+= partial
;
1101 if (buffered
== WRITE_BUFFER_SIZE
) {
1102 write_buffer_len
= buffered
;
1103 if (ce_write_flush(context
, fd
))
1107 write_buffer_len
= buffered
;
1109 data
= (char *) data
+ partial
;
1114 static int write_index_ext_header(SHA_CTX
*context
, int fd
,
1115 unsigned int ext
, unsigned int sz
)
1119 return ((ce_write(context
, fd
, &ext
, 4) < 0) ||
1120 (ce_write(context
, fd
, &sz
, 4) < 0)) ? -1 : 0;
1123 static int ce_flush(SHA_CTX
*context
, int fd
)
1125 unsigned int left
= write_buffer_len
;
1128 write_buffer_len
= 0;
1129 SHA1_Update(context
, write_buffer
, left
);
1132 /* Flush first if not enough space for SHA1 signature */
1133 if (left
+ 20 > WRITE_BUFFER_SIZE
) {
1134 if (write_in_full(fd
, write_buffer
, left
) != left
)
1139 /* Append the SHA1 signature at the end */
1140 SHA1_Final(write_buffer
+ left
, context
);
1142 return (write_in_full(fd
, write_buffer
, left
) != left
) ? -1 : 0;
1145 static void ce_smudge_racily_clean_entry(struct cache_entry
*ce
)
1148 * The only thing we care about in this function is to smudge the
1149 * falsely clean entry due to touch-update-touch race, so we leave
1150 * everything else as they are. We are called for entries whose
1151 * ce_mtime match the index file mtime.
1155 if (lstat(ce
->name
, &st
) < 0)
1157 if (ce_match_stat_basic(ce
, &st
))
1159 if (ce_modified_check_fs(ce
, &st
)) {
1160 /* This is "racily clean"; smudge it. Note that this
1161 * is a tricky code. At first glance, it may appear
1162 * that it can break with this sequence:
1164 * $ echo xyzzy >frotz
1165 * $ git-update-index --add frotz
1168 * $ echo filfre >nitfol
1169 * $ git-update-index --add nitfol
1171 * but it does not. When the second update-index runs,
1172 * it notices that the entry "frotz" has the same timestamp
1173 * as index, and if we were to smudge it by resetting its
1174 * size to zero here, then the object name recorded
1175 * in index is the 6-byte file but the cached stat information
1176 * becomes zero --- which would then match what we would
1177 * obtain from the filesystem next time we stat("frotz").
1179 * However, the second update-index, before calling
1180 * this function, notices that the cached size is 6
1181 * bytes and what is on the filesystem is an empty
1182 * file, and never calls us, so the cached size information
1183 * for "frotz" stays 6 which does not match the filesystem.
1189 static int ce_write_entry(SHA_CTX
*c
, int fd
, struct cache_entry
*ce
)
1191 int size
= ondisk_ce_size(ce
);
1192 struct ondisk_cache_entry
*ondisk
= xcalloc(1, size
);
1194 ondisk
->ctime
.sec
= htonl(ce
->ce_ctime
);
1195 ondisk
->ctime
.nsec
= 0;
1196 ondisk
->mtime
.sec
= htonl(ce
->ce_mtime
);
1197 ondisk
->mtime
.nsec
= 0;
1198 ondisk
->dev
= htonl(ce
->ce_dev
);
1199 ondisk
->ino
= htonl(ce
->ce_ino
);
1200 ondisk
->mode
= htonl(ce
->ce_mode
);
1201 ondisk
->uid
= htonl(ce
->ce_uid
);
1202 ondisk
->gid
= htonl(ce
->ce_gid
);
1203 ondisk
->size
= htonl(ce
->ce_size
);
1204 hashcpy(ondisk
->sha1
, ce
->sha1
);
1205 ondisk
->flags
= htons(ce
->ce_flags
);
1206 memcpy(ondisk
->name
, ce
->name
, ce_namelen(ce
));
1208 return ce_write(c
, fd
, ondisk
, size
);
1211 int write_index(struct index_state
*istate
, int newfd
)
1214 struct cache_header hdr
;
1215 int i
, err
, removed
;
1216 struct cache_entry
**cache
= istate
->cache
;
1217 int entries
= istate
->cache_nr
;
1219 for (i
= removed
= 0; i
< entries
; i
++)
1220 if (cache
[i
]->ce_flags
& CE_REMOVE
)
1223 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
1224 hdr
.hdr_version
= htonl(2);
1225 hdr
.hdr_entries
= htonl(entries
- removed
);
1228 if (ce_write(&c
, newfd
, &hdr
, sizeof(hdr
)) < 0)
1231 for (i
= 0; i
< entries
; i
++) {
1232 struct cache_entry
*ce
= cache
[i
];
1233 if (ce
->ce_flags
& CE_REMOVE
)
1235 if (istate
->timestamp
&&
1236 istate
->timestamp
<= ce
->ce_mtime
)
1237 ce_smudge_racily_clean_entry(ce
);
1238 if (ce_write_entry(&c
, newfd
, ce
) < 0)
1242 /* Write extension data here */
1243 if (istate
->cache_tree
) {
1246 strbuf_init(&sb
, 0);
1247 cache_tree_write(&sb
, istate
->cache_tree
);
1248 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_TREE
, sb
.len
) < 0
1249 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
1250 strbuf_release(&sb
);
1254 return ce_flush(&c
, newfd
);