2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 #define NO_THE_INDEX_COMPATIBILITY_MACROS
9 #include "cache-tree.h"
15 #include "resolve-undo.h"
18 #include "split-index.h"
22 static struct cache_entry
*refresh_cache_entry(struct cache_entry
*ce
,
23 unsigned int options
);
25 /* Mask for the name length in ce_flags in the on-disk index */
27 #define CE_NAMEMASK (0x0fff)
31 * The first letter should be 'A'..'Z' for extensions that are not
32 * necessary for a correct operation (i.e. optimization data).
33 * When new extensions are added that _needs_ to be understood in
34 * order to correctly interpret the index file, pick character that
35 * is outside the range, to cause the reader to abort.
38 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
39 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
40 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
41 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
42 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
44 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
45 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
46 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
49 struct index_state the_index
;
50 static const char *alternate_index_output
;
52 static void set_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
54 istate
->cache
[nr
] = ce
;
55 add_name_hash(istate
, ce
);
58 static void replace_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
60 struct cache_entry
*old
= istate
->cache
[nr
];
62 replace_index_entry_in_base(istate
, old
, ce
);
63 remove_name_hash(istate
, old
);
65 set_index_entry(istate
, nr
, ce
);
66 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
67 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
70 void rename_index_entry_at(struct index_state
*istate
, int nr
, const char *new_name
)
72 struct cache_entry
*old
= istate
->cache
[nr
], *new;
73 int namelen
= strlen(new_name
);
75 new = xmalloc(cache_entry_size(namelen
));
76 copy_cache_entry(new, old
);
77 new->ce_flags
&= ~CE_HASHED
;
78 new->ce_namelen
= namelen
;
80 memcpy(new->name
, new_name
, namelen
+ 1);
82 cache_tree_invalidate_path(istate
, old
->name
);
83 remove_index_entry_at(istate
, nr
);
84 add_index_entry(istate
, new, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
87 void fill_stat_data(struct stat_data
*sd
, struct stat
*st
)
89 sd
->sd_ctime
.sec
= (unsigned int)st
->st_ctime
;
90 sd
->sd_mtime
.sec
= (unsigned int)st
->st_mtime
;
91 sd
->sd_ctime
.nsec
= ST_CTIME_NSEC(*st
);
92 sd
->sd_mtime
.nsec
= ST_MTIME_NSEC(*st
);
93 sd
->sd_dev
= st
->st_dev
;
94 sd
->sd_ino
= st
->st_ino
;
95 sd
->sd_uid
= st
->st_uid
;
96 sd
->sd_gid
= st
->st_gid
;
97 sd
->sd_size
= st
->st_size
;
100 int match_stat_data(const struct stat_data
*sd
, struct stat
*st
)
104 if (sd
->sd_mtime
.sec
!= (unsigned int)st
->st_mtime
)
105 changed
|= MTIME_CHANGED
;
106 if (trust_ctime
&& check_stat
&&
107 sd
->sd_ctime
.sec
!= (unsigned int)st
->st_ctime
)
108 changed
|= CTIME_CHANGED
;
111 if (check_stat
&& sd
->sd_mtime
.nsec
!= ST_MTIME_NSEC(*st
))
112 changed
|= MTIME_CHANGED
;
113 if (trust_ctime
&& check_stat
&&
114 sd
->sd_ctime
.nsec
!= ST_CTIME_NSEC(*st
))
115 changed
|= CTIME_CHANGED
;
119 if (sd
->sd_uid
!= (unsigned int) st
->st_uid
||
120 sd
->sd_gid
!= (unsigned int) st
->st_gid
)
121 changed
|= OWNER_CHANGED
;
122 if (sd
->sd_ino
!= (unsigned int) st
->st_ino
)
123 changed
|= INODE_CHANGED
;
128 * st_dev breaks on network filesystems where different
129 * clients will have different views of what "device"
130 * the filesystem is on
132 if (check_stat
&& sd
->sd_dev
!= (unsigned int) st
->st_dev
)
133 changed
|= INODE_CHANGED
;
136 if (sd
->sd_size
!= (unsigned int) st
->st_size
)
137 changed
|= DATA_CHANGED
;
143 * This only updates the "non-critical" parts of the directory
144 * cache, ie the parts that aren't tracked by GIT, and only used
145 * to validate the cache.
147 void fill_stat_cache_info(struct cache_entry
*ce
, struct stat
*st
)
149 fill_stat_data(&ce
->ce_stat_data
, st
);
151 if (assume_unchanged
)
152 ce
->ce_flags
|= CE_VALID
;
154 if (S_ISREG(st
->st_mode
))
155 ce_mark_uptodate(ce
);
158 static int ce_compare_data(const struct cache_entry
*ce
, struct stat
*st
)
161 int fd
= open(ce
->name
, O_RDONLY
);
164 unsigned char sha1
[20];
165 if (!index_fd(sha1
, fd
, st
, OBJ_BLOB
, ce
->name
, 0))
166 match
= hashcmp(sha1
, ce
->sha1
);
167 /* index_fd() closed the file descriptor already */
172 static int ce_compare_link(const struct cache_entry
*ce
, size_t expected_size
)
177 enum object_type type
;
178 struct strbuf sb
= STRBUF_INIT
;
180 if (strbuf_readlink(&sb
, ce
->name
, expected_size
))
183 buffer
= read_sha1_file(ce
->sha1
, &type
, &size
);
186 match
= memcmp(buffer
, sb
.buf
, size
);
193 static int ce_compare_gitlink(const struct cache_entry
*ce
)
195 unsigned char sha1
[20];
198 * We don't actually require that the .git directory
199 * under GITLINK directory be a valid git directory. It
200 * might even be missing (in case nobody populated that
203 * If so, we consider it always to match.
205 if (resolve_gitlink_ref(ce
->name
, "HEAD", sha1
) < 0)
207 return hashcmp(sha1
, ce
->sha1
);
210 static int ce_modified_check_fs(const struct cache_entry
*ce
, struct stat
*st
)
212 switch (st
->st_mode
& S_IFMT
) {
214 if (ce_compare_data(ce
, st
))
218 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
222 if (S_ISGITLINK(ce
->ce_mode
))
223 return ce_compare_gitlink(ce
) ? DATA_CHANGED
: 0;
230 static int ce_match_stat_basic(const struct cache_entry
*ce
, struct stat
*st
)
232 unsigned int changed
= 0;
234 if (ce
->ce_flags
& CE_REMOVE
)
235 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
237 switch (ce
->ce_mode
& S_IFMT
) {
239 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
240 /* We consider only the owner x bit to be relevant for
243 if (trust_executable_bit
&&
244 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
245 changed
|= MODE_CHANGED
;
248 if (!S_ISLNK(st
->st_mode
) &&
249 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
250 changed
|= TYPE_CHANGED
;
253 /* We ignore most of the st_xxx fields for gitlinks */
254 if (!S_ISDIR(st
->st_mode
))
255 changed
|= TYPE_CHANGED
;
256 else if (ce_compare_gitlink(ce
))
257 changed
|= DATA_CHANGED
;
260 die("internal error: ce_mode is %o", ce
->ce_mode
);
263 changed
|= match_stat_data(&ce
->ce_stat_data
, st
);
265 /* Racily smudged entry? */
266 if (!ce
->ce_stat_data
.sd_size
) {
267 if (!is_empty_blob_sha1(ce
->sha1
))
268 changed
|= DATA_CHANGED
;
274 static int is_racy_timestamp(const struct index_state
*istate
,
275 const struct cache_entry
*ce
)
277 return (!S_ISGITLINK(ce
->ce_mode
) &&
278 istate
->timestamp
.sec
&&
280 /* nanosecond timestamped files can also be racy! */
281 (istate
->timestamp
.sec
< ce
->ce_stat_data
.sd_mtime
.sec
||
282 (istate
->timestamp
.sec
== ce
->ce_stat_data
.sd_mtime
.sec
&&
283 istate
->timestamp
.nsec
<= ce
->ce_stat_data
.sd_mtime
.nsec
))
285 istate
->timestamp
.sec
<= ce
->ce_stat_data
.sd_mtime
.sec
290 int ie_match_stat(const struct index_state
*istate
,
291 const struct cache_entry
*ce
, struct stat
*st
,
292 unsigned int options
)
294 unsigned int changed
;
295 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
296 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
297 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
300 * If it's marked as always valid in the index, it's
301 * valid whatever the checked-out copy says.
303 * skip-worktree has the same effect with higher precedence
305 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
))
307 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
311 * Intent-to-add entries have not been added, so the index entry
312 * by definition never matches what is in the work tree until it
313 * actually gets added.
315 if (ce
->ce_flags
& CE_INTENT_TO_ADD
)
316 return DATA_CHANGED
| TYPE_CHANGED
| MODE_CHANGED
;
318 changed
= ce_match_stat_basic(ce
, st
);
321 * Within 1 second of this sequence:
322 * echo xyzzy >file && git-update-index --add file
323 * running this command:
325 * would give a falsely clean cache entry. The mtime and
326 * length match the cache, and other stat fields do not change.
328 * We could detect this at update-index time (the cache entry
329 * being registered/updated records the same time as "now")
330 * and delay the return from git-update-index, but that would
331 * effectively mean we can make at most one commit per second,
332 * which is not acceptable. Instead, we check cache entries
333 * whose mtime are the same as the index file timestamp more
334 * carefully than others.
336 if (!changed
&& is_racy_timestamp(istate
, ce
)) {
337 if (assume_racy_is_modified
)
338 changed
|= DATA_CHANGED
;
340 changed
|= ce_modified_check_fs(ce
, st
);
346 int ie_modified(const struct index_state
*istate
,
347 const struct cache_entry
*ce
,
348 struct stat
*st
, unsigned int options
)
350 int changed
, changed_fs
;
352 changed
= ie_match_stat(istate
, ce
, st
, options
);
356 * If the mode or type has changed, there's no point in trying
357 * to refresh the entry - it's not going to match
359 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
363 * Immediately after read-tree or update-index --cacheinfo,
364 * the length field is zero, as we have never even read the
365 * lstat(2) information once, and we cannot trust DATA_CHANGED
366 * returned by ie_match_stat() which in turn was returned by
367 * ce_match_stat_basic() to signal that the filesize of the
368 * blob changed. We have to actually go to the filesystem to
369 * see if the contents match, and if so, should answer "unchanged".
371 * The logic does not apply to gitlinks, as ce_match_stat_basic()
372 * already has checked the actual HEAD from the filesystem in the
373 * subproject. If ie_match_stat() already said it is different,
374 * then we know it is.
376 if ((changed
& DATA_CHANGED
) &&
377 (S_ISGITLINK(ce
->ce_mode
) || ce
->ce_stat_data
.sd_size
!= 0))
380 changed_fs
= ce_modified_check_fs(ce
, st
);
382 return changed
| changed_fs
;
386 int base_name_compare(const char *name1
, int len1
, int mode1
,
387 const char *name2
, int len2
, int mode2
)
389 unsigned char c1
, c2
;
390 int len
= len1
< len2
? len1
: len2
;
393 cmp
= memcmp(name1
, name2
, len
);
398 if (!c1
&& S_ISDIR(mode1
))
400 if (!c2
&& S_ISDIR(mode2
))
402 return (c1
< c2
) ? -1 : (c1
> c2
) ? 1 : 0;
406 * df_name_compare() is identical to base_name_compare(), except it
407 * compares conflicting directory/file entries as equal. Note that
408 * while a directory name compares as equal to a regular file, they
409 * then individually compare _differently_ to a filename that has
410 * a dot after the basename (because '\0' < '.' < '/').
412 * This is used by routines that want to traverse the git namespace
413 * but then handle conflicting entries together when possible.
415 int df_name_compare(const char *name1
, int len1
, int mode1
,
416 const char *name2
, int len2
, int mode2
)
418 int len
= len1
< len2
? len1
: len2
, cmp
;
419 unsigned char c1
, c2
;
421 cmp
= memcmp(name1
, name2
, len
);
424 /* Directories and files compare equal (same length, same name) */
428 if (!c1
&& S_ISDIR(mode1
))
431 if (!c2
&& S_ISDIR(mode2
))
433 if (c1
== '/' && !c2
)
435 if (c2
== '/' && !c1
)
440 int name_compare(const char *name1
, size_t len1
, const char *name2
, size_t len2
)
442 size_t min_len
= (len1
< len2
) ? len1
: len2
;
443 int cmp
= memcmp(name1
, name2
, min_len
);
453 int cache_name_stage_compare(const char *name1
, int len1
, int stage1
, const char *name2
, int len2
, int stage2
)
457 cmp
= name_compare(name1
, len1
, name2
, len2
);
468 static int index_name_stage_pos(const struct index_state
*istate
, const char *name
, int namelen
, int stage
)
473 last
= istate
->cache_nr
;
474 while (last
> first
) {
475 int next
= (last
+ first
) >> 1;
476 struct cache_entry
*ce
= istate
->cache
[next
];
477 int cmp
= cache_name_stage_compare(name
, namelen
, stage
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
489 int index_name_pos(const struct index_state
*istate
, const char *name
, int namelen
)
491 return index_name_stage_pos(istate
, name
, namelen
, 0);
494 /* Remove entry, return true if there are more entries to go.. */
495 int remove_index_entry_at(struct index_state
*istate
, int pos
)
497 struct cache_entry
*ce
= istate
->cache
[pos
];
499 record_resolve_undo(istate
, ce
);
500 remove_name_hash(istate
, ce
);
501 save_or_free_index_entry(istate
, ce
);
502 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
504 if (pos
>= istate
->cache_nr
)
506 memmove(istate
->cache
+ pos
,
507 istate
->cache
+ pos
+ 1,
508 (istate
->cache_nr
- pos
) * sizeof(struct cache_entry
*));
513 * Remove all cache entries marked for removal, that is where
514 * CE_REMOVE is set in ce_flags. This is much more effective than
515 * calling remove_index_entry_at() for each entry to be removed.
517 void remove_marked_cache_entries(struct index_state
*istate
)
519 struct cache_entry
**ce_array
= istate
->cache
;
522 for (i
= j
= 0; i
< istate
->cache_nr
; i
++) {
523 if (ce_array
[i
]->ce_flags
& CE_REMOVE
) {
524 remove_name_hash(istate
, ce_array
[i
]);
525 save_or_free_index_entry(istate
, ce_array
[i
]);
528 ce_array
[j
++] = ce_array
[i
];
530 if (j
== istate
->cache_nr
)
532 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
533 istate
->cache_nr
= j
;
536 int remove_file_from_index(struct index_state
*istate
, const char *path
)
538 int pos
= index_name_pos(istate
, path
, strlen(path
));
541 cache_tree_invalidate_path(istate
, path
);
542 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
543 remove_index_entry_at(istate
, pos
);
547 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
549 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
552 static int index_name_pos_also_unmerged(struct index_state
*istate
,
553 const char *path
, int namelen
)
555 int pos
= index_name_pos(istate
, path
, namelen
);
556 struct cache_entry
*ce
;
561 /* maybe unmerged? */
563 if (pos
>= istate
->cache_nr
||
564 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
567 /* order of preference: stage 2, 1, 3 */
568 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
569 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
570 !compare_name(ce
, path
, namelen
))
575 static int different_name(struct cache_entry
*ce
, struct cache_entry
*alias
)
577 int len
= ce_namelen(ce
);
578 return ce_namelen(alias
) != len
|| memcmp(ce
->name
, alias
->name
, len
);
582 * If we add a filename that aliases in the cache, we will use the
583 * name that we already have - but we don't want to update the same
584 * alias twice, because that implies that there were actually two
585 * different files with aliasing names!
587 * So we use the CE_ADDED flag to verify that the alias was an old
588 * one before we accept it as
590 static struct cache_entry
*create_alias_ce(struct index_state
*istate
,
591 struct cache_entry
*ce
,
592 struct cache_entry
*alias
)
595 struct cache_entry
*new;
597 if (alias
->ce_flags
& CE_ADDED
)
598 die("Will not add file alias '%s' ('%s' already exists in index)", ce
->name
, alias
->name
);
600 /* Ok, create the new entry using the name of the existing alias */
601 len
= ce_namelen(alias
);
602 new = xcalloc(1, cache_entry_size(len
));
603 memcpy(new->name
, alias
->name
, len
);
604 copy_cache_entry(new, ce
);
605 save_or_free_index_entry(istate
, ce
);
609 void set_object_name_for_intent_to_add_entry(struct cache_entry
*ce
)
611 unsigned char sha1
[20];
612 if (write_sha1_file("", 0, blob_type
, sha1
))
613 die("cannot create an empty blob in the object database");
614 hashcpy(ce
->sha1
, sha1
);
617 int add_to_index(struct index_state
*istate
, const char *path
, struct stat
*st
, int flags
)
619 int size
, namelen
, was_same
;
620 mode_t st_mode
= st
->st_mode
;
621 struct cache_entry
*ce
, *alias
;
622 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_IGNORE_SKIP_WORKTREE
|CE_MATCH_RACY_IS_DIRTY
;
623 int verbose
= flags
& (ADD_CACHE_VERBOSE
| ADD_CACHE_PRETEND
);
624 int pretend
= flags
& ADD_CACHE_PRETEND
;
625 int intent_only
= flags
& ADD_CACHE_INTENT
;
626 int add_option
= (ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
|
627 (intent_only
? ADD_CACHE_NEW_ONLY
: 0));
629 if (!S_ISREG(st_mode
) && !S_ISLNK(st_mode
) && !S_ISDIR(st_mode
))
630 return error("%s: can only add regular files, symbolic links or git-directories", path
);
632 namelen
= strlen(path
);
633 if (S_ISDIR(st_mode
)) {
634 while (namelen
&& path
[namelen
-1] == '/')
637 size
= cache_entry_size(namelen
);
638 ce
= xcalloc(1, size
);
639 memcpy(ce
->name
, path
, namelen
);
640 ce
->ce_namelen
= namelen
;
642 fill_stat_cache_info(ce
, st
);
644 ce
->ce_flags
|= CE_INTENT_TO_ADD
;
646 if (trust_executable_bit
&& has_symlinks
)
647 ce
->ce_mode
= create_ce_mode(st_mode
);
649 /* If there is an existing entry, pick the mode bits and type
650 * from it, otherwise assume unexecutable regular file.
652 struct cache_entry
*ent
;
653 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
655 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
656 ce
->ce_mode
= ce_mode_from_stat(ent
, st_mode
);
659 /* When core.ignorecase=true, determine if a directory of the same name but differing
660 * case already exists within the Git repository. If it does, ensure the directory
661 * case of the file being added to the repository matches (is folded into) the existing
662 * entry's directory case.
665 const char *startPtr
= ce
->name
;
666 const char *ptr
= startPtr
;
668 while (*ptr
&& *ptr
!= '/')
671 struct cache_entry
*foundce
;
673 foundce
= index_dir_exists(istate
, ce
->name
, ptr
- ce
->name
- 1);
675 memcpy((void *)startPtr
, foundce
->name
+ (startPtr
- ce
->name
), ptr
- startPtr
);
682 alias
= index_file_exists(istate
, ce
->name
, ce_namelen(ce
), ignore_case
);
683 if (alias
&& !ce_stage(alias
) && !ie_match_stat(istate
, alias
, st
, ce_option
)) {
684 /* Nothing changed, really */
686 if (!S_ISGITLINK(alias
->ce_mode
))
687 ce_mark_uptodate(alias
);
688 alias
->ce_flags
|= CE_ADDED
;
692 if (index_path(ce
->sha1
, path
, st
, HASH_WRITE_OBJECT
))
693 return error("unable to index file %s", path
);
695 set_object_name_for_intent_to_add_entry(ce
);
697 if (ignore_case
&& alias
&& different_name(ce
, alias
))
698 ce
= create_alias_ce(istate
, ce
, alias
);
699 ce
->ce_flags
|= CE_ADDED
;
701 /* It was suspected to be racily clean, but it turns out to be Ok */
704 !hashcmp(alias
->sha1
, ce
->sha1
) &&
705 ce
->ce_mode
== alias
->ce_mode
);
709 else if (add_index_entry(istate
, ce
, add_option
))
710 return error("unable to add %s to index",path
);
711 if (verbose
&& !was_same
)
712 printf("add '%s'\n", path
);
716 int add_file_to_index(struct index_state
*istate
, const char *path
, int flags
)
719 if (lstat(path
, &st
))
720 die_errno("unable to stat '%s'", path
);
721 return add_to_index(istate
, path
, &st
, flags
);
724 struct cache_entry
*make_cache_entry(unsigned int mode
,
725 const unsigned char *sha1
, const char *path
, int stage
,
726 unsigned int refresh_options
)
729 struct cache_entry
*ce
, *ret
;
731 if (!verify_path(path
)) {
732 error("Invalid path '%s'", path
);
737 size
= cache_entry_size(len
);
738 ce
= xcalloc(1, size
);
740 hashcpy(ce
->sha1
, sha1
);
741 memcpy(ce
->name
, path
, len
);
742 ce
->ce_flags
= create_ce_flags(stage
);
743 ce
->ce_namelen
= len
;
744 ce
->ce_mode
= create_ce_mode(mode
);
746 ret
= refresh_cache_entry(ce
, refresh_options
);
755 int ce_same_name(const struct cache_entry
*a
, const struct cache_entry
*b
)
757 int len
= ce_namelen(a
);
758 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
762 * We fundamentally don't like some paths: we don't want
763 * dot or dot-dot anywhere, and for obvious reasons don't
764 * want to recurse into ".git" either.
766 * Also, we don't want double slashes or slashes at the
767 * end that can make pathnames ambiguous.
769 static int verify_dotfile(const char *rest
)
772 * The first character was '.', but that
773 * has already been discarded, we now test
777 /* "." is not allowed */
778 if (*rest
== '\0' || is_dir_sep(*rest
))
783 * ".git" followed by NUL or slash is bad. This
784 * shares the path end test with the ".." case.
788 if (rest
[1] != 'i' && rest
[1] != 'I')
790 if (rest
[2] != 't' && rest
[2] != 'T')
795 if (rest
[1] == '\0' || is_dir_sep(rest
[1]))
801 int verify_path(const char *path
)
805 if (has_dos_drive_prefix(path
))
814 if (protect_hfs
&& is_hfs_dotgit(path
))
816 if (protect_ntfs
&& is_ntfs_dotgit(path
))
819 if ((c
== '.' && !verify_dotfile(path
)) ||
820 is_dir_sep(c
) || c
== '\0')
828 * Do we have another file that has the beginning components being a
829 * proper superset of the name we're trying to add?
831 static int has_file_name(struct index_state
*istate
,
832 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
835 int len
= ce_namelen(ce
);
836 int stage
= ce_stage(ce
);
837 const char *name
= ce
->name
;
839 while (pos
< istate
->cache_nr
) {
840 struct cache_entry
*p
= istate
->cache
[pos
++];
842 if (len
>= ce_namelen(p
))
844 if (memcmp(name
, p
->name
, len
))
846 if (ce_stage(p
) != stage
)
848 if (p
->name
[len
] != '/')
850 if (p
->ce_flags
& CE_REMOVE
)
855 remove_index_entry_at(istate
, --pos
);
861 * Do we have another file with a pathname that is a proper
862 * subset of the name we're trying to add?
864 static int has_dir_name(struct index_state
*istate
,
865 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
868 int stage
= ce_stage(ce
);
869 const char *name
= ce
->name
;
870 const char *slash
= name
+ ce_namelen(ce
);
878 if (slash
<= ce
->name
)
883 pos
= index_name_stage_pos(istate
, name
, len
, stage
);
886 * Found one, but not so fast. This could
887 * be a marker that says "I was here, but
888 * I am being removed". Such an entry is
889 * not a part of the resulting tree, and
890 * it is Ok to have a directory at the same
893 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
897 remove_index_entry_at(istate
, pos
);
905 * Trivial optimization: if we find an entry that
906 * already matches the sub-directory, then we know
907 * we're ok, and we can exit.
909 while (pos
< istate
->cache_nr
) {
910 struct cache_entry
*p
= istate
->cache
[pos
];
911 if ((ce_namelen(p
) <= len
) ||
912 (p
->name
[len
] != '/') ||
913 memcmp(p
->name
, name
, len
))
914 break; /* not our subdirectory */
915 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
917 * p is at the same stage as our entry, and
918 * is a subdirectory of what we are looking
919 * at, so we cannot have conflicts at our
920 * level or anything shorter.
929 /* We may be in a situation where we already have path/file and path
930 * is being added, or we already have path and path/file is being
931 * added. Either one would result in a nonsense tree that has path
932 * twice when git-write-tree tries to write it out. Prevent it.
934 * If ok-to-replace is specified, we remove the conflicting entries
935 * from the cache so the caller should recompute the insert position.
936 * When this happens, we return non-zero.
938 static int check_file_directory_conflict(struct index_state
*istate
,
939 const struct cache_entry
*ce
,
940 int pos
, int ok_to_replace
)
945 * When ce is an "I am going away" entry, we allow it to be added
947 if (ce
->ce_flags
& CE_REMOVE
)
951 * We check if the path is a sub-path of a subsequent pathname
952 * first, since removing those will not change the position
955 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
958 * Then check if the path might have a clashing sub-directory
961 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
964 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
967 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
968 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
969 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
970 int new_only
= option
& ADD_CACHE_NEW_ONLY
;
972 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
973 cache_tree_invalidate_path(istate
, ce
->name
);
974 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
976 /* existing match? Just replace it. */
979 replace_index_entry(istate
, pos
, ce
);
985 * Inserting a merged entry ("stage 0") into the index
986 * will always replace all non-merged entries..
988 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
989 while (ce_same_name(istate
->cache
[pos
], ce
)) {
991 if (!remove_index_entry_at(istate
, pos
))
998 if (!verify_path(ce
->name
))
999 return error("Invalid path '%s'", ce
->name
);
1001 if (!skip_df_check
&&
1002 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
1004 return error("'%s' appears as both a file and as a directory",
1006 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1012 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1016 if (option
& ADD_CACHE_JUST_APPEND
)
1017 pos
= istate
->cache_nr
;
1020 ret
= add_index_entry_with_check(istate
, ce
, option
);
1026 /* Make sure the array is big enough .. */
1027 ALLOC_GROW(istate
->cache
, istate
->cache_nr
+ 1, istate
->cache_alloc
);
1031 if (istate
->cache_nr
> pos
+ 1)
1032 memmove(istate
->cache
+ pos
+ 1,
1033 istate
->cache
+ pos
,
1034 (istate
->cache_nr
- pos
- 1) * sizeof(ce
));
1035 set_index_entry(istate
, pos
, ce
);
1036 istate
->cache_changed
|= CE_ENTRY_ADDED
;
1041 * "refresh" does not calculate a new sha1 file or bring the
1042 * cache up-to-date for mode/content changes. But what it
1043 * _does_ do is to "re-match" the stat information of a file
1044 * with the cache, so that you can refresh the cache for a
1045 * file that hasn't been changed but where the stat entry is
1048 * For example, you'd want to do this after doing a "git-read-tree",
1049 * to link up the stat cache details with the proper files.
1051 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
1052 struct cache_entry
*ce
,
1053 unsigned int options
, int *err
,
1057 struct cache_entry
*updated
;
1059 int refresh
= options
& CE_MATCH_REFRESH
;
1060 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
1061 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
1062 int ignore_missing
= options
& CE_MATCH_IGNORE_MISSING
;
1064 if (!refresh
|| ce_uptodate(ce
))
1068 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1069 * that the change to the work tree does not matter and told
1072 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
)) {
1073 ce_mark_uptodate(ce
);
1076 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
)) {
1077 ce_mark_uptodate(ce
);
1081 if (has_symlink_leading_path(ce
->name
, ce_namelen(ce
))) {
1089 if (lstat(ce
->name
, &st
) < 0) {
1090 if (ignore_missing
&& errno
== ENOENT
)
1097 changed
= ie_match_stat(istate
, ce
, &st
, options
);
1099 *changed_ret
= changed
;
1102 * The path is unchanged. If we were told to ignore
1103 * valid bit, then we did the actual stat check and
1104 * found that the entry is unmodified. If the entry
1105 * is not marked VALID, this is the place to mark it
1106 * valid again, under "assume unchanged" mode.
1108 if (ignore_valid
&& assume_unchanged
&&
1109 !(ce
->ce_flags
& CE_VALID
))
1110 ; /* mark this one VALID again */
1113 * We do not mark the index itself "modified"
1114 * because CE_UPTODATE flag is in-core only;
1115 * we are not going to write this change out.
1117 if (!S_ISGITLINK(ce
->ce_mode
))
1118 ce_mark_uptodate(ce
);
1123 if (ie_modified(istate
, ce
, &st
, options
)) {
1130 updated
= xmalloc(size
);
1131 memcpy(updated
, ce
, size
);
1132 fill_stat_cache_info(updated
, &st
);
1134 * If ignore_valid is not set, we should leave CE_VALID bit
1135 * alone. Otherwise, paths marked with --no-assume-unchanged
1136 * (i.e. things to be edited) will reacquire CE_VALID bit
1137 * automatically, which is not really what we want.
1139 if (!ignore_valid
&& assume_unchanged
&&
1140 !(ce
->ce_flags
& CE_VALID
))
1141 updated
->ce_flags
&= ~CE_VALID
;
1143 /* istate->cache_changed is updated in the caller */
1147 static void show_file(const char * fmt
, const char * name
, int in_porcelain
,
1148 int * first
, const char *header_msg
)
1150 if (in_porcelain
&& *first
&& header_msg
) {
1151 printf("%s\n", header_msg
);
1157 int refresh_index(struct index_state
*istate
, unsigned int flags
,
1158 const struct pathspec
*pathspec
,
1159 char *seen
, const char *header_msg
)
1163 int really
= (flags
& REFRESH_REALLY
) != 0;
1164 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
1165 int quiet
= (flags
& REFRESH_QUIET
) != 0;
1166 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
1167 int ignore_submodules
= (flags
& REFRESH_IGNORE_SUBMODULES
) != 0;
1169 int in_porcelain
= (flags
& REFRESH_IN_PORCELAIN
);
1170 unsigned int options
= (CE_MATCH_REFRESH
|
1171 (really
? CE_MATCH_IGNORE_VALID
: 0) |
1172 (not_new
? CE_MATCH_IGNORE_MISSING
: 0));
1173 const char *modified_fmt
;
1174 const char *deleted_fmt
;
1175 const char *typechange_fmt
;
1176 const char *added_fmt
;
1177 const char *unmerged_fmt
;
1179 modified_fmt
= (in_porcelain
? "M\t%s\n" : "%s: needs update\n");
1180 deleted_fmt
= (in_porcelain
? "D\t%s\n" : "%s: needs update\n");
1181 typechange_fmt
= (in_porcelain
? "T\t%s\n" : "%s needs update\n");
1182 added_fmt
= (in_porcelain
? "A\t%s\n" : "%s needs update\n");
1183 unmerged_fmt
= (in_porcelain
? "U\t%s\n" : "%s: needs merge\n");
1184 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1185 struct cache_entry
*ce
, *new;
1186 int cache_errno
= 0;
1190 ce
= istate
->cache
[i
];
1191 if (ignore_submodules
&& S_ISGITLINK(ce
->ce_mode
))
1194 if (pathspec
&& !ce_path_match(ce
, pathspec
, seen
))
1198 while ((i
< istate
->cache_nr
) &&
1199 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
1205 show_file(unmerged_fmt
, ce
->name
, in_porcelain
,
1206 &first
, header_msg
);
1214 new = refresh_cache_ent(istate
, ce
, options
, &cache_errno
, &changed
);
1220 if (really
&& cache_errno
== EINVAL
) {
1221 /* If we are doing --really-refresh that
1222 * means the index is not valid anymore.
1224 ce
->ce_flags
&= ~CE_VALID
;
1225 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
1226 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
1231 if (cache_errno
== ENOENT
)
1233 else if (ce
->ce_flags
& CE_INTENT_TO_ADD
)
1234 fmt
= added_fmt
; /* must be before other checks */
1235 else if (changed
& TYPE_CHANGED
)
1236 fmt
= typechange_fmt
;
1240 ce
->name
, in_porcelain
, &first
, header_msg
);
1245 replace_index_entry(istate
, i
, new);
1250 static struct cache_entry
*refresh_cache_entry(struct cache_entry
*ce
,
1251 unsigned int options
)
1253 return refresh_cache_ent(&the_index
, ce
, options
, NULL
, NULL
);
1257 /*****************************************************************
1259 *****************************************************************/
1261 #define INDEX_FORMAT_DEFAULT 3
1263 static unsigned int get_index_format_default(void)
1265 char *envversion
= getenv("GIT_INDEX_VERSION");
1268 unsigned int version
= INDEX_FORMAT_DEFAULT
;
1271 if (!git_config_get_int("index.version", &value
))
1273 if (version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1274 warning(_("index.version set, but the value is invalid.\n"
1275 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1276 return INDEX_FORMAT_DEFAULT
;
1281 version
= strtoul(envversion
, &endp
, 10);
1283 version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1284 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1285 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1286 version
= INDEX_FORMAT_DEFAULT
;
1292 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1293 * Again - this is just a (very strong in practice) heuristic that
1294 * the inode hasn't changed.
1296 * We save the fields in big-endian order to allow using the
1297 * index file over NFS transparently.
1299 struct ondisk_cache_entry
{
1300 struct cache_time ctime
;
1301 struct cache_time mtime
;
1308 unsigned char sha1
[20];
1310 char name
[FLEX_ARRAY
]; /* more */
1314 * This struct is used when CE_EXTENDED bit is 1
1315 * The struct must match ondisk_cache_entry exactly from
1318 struct ondisk_cache_entry_extended
{
1319 struct cache_time ctime
;
1320 struct cache_time mtime
;
1327 unsigned char sha1
[20];
1330 char name
[FLEX_ARRAY
]; /* more */
1333 /* These are only used for v3 or lower */
1334 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1335 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1336 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1337 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1338 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1339 ondisk_cache_entry_size(ce_namelen(ce)))
1341 static int verify_hdr(struct cache_header
*hdr
, unsigned long size
)
1344 unsigned char sha1
[20];
1347 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
1348 return error("bad signature");
1349 hdr_version
= ntohl(hdr
->hdr_version
);
1350 if (hdr_version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< hdr_version
)
1351 return error("bad index version %d", hdr_version
);
1353 git_SHA1_Update(&c
, hdr
, size
- 20);
1354 git_SHA1_Final(sha1
, &c
);
1355 if (hashcmp(sha1
, (unsigned char *)hdr
+ size
- 20))
1356 return error("bad index file sha1 signature");
1360 static int read_index_extension(struct index_state
*istate
,
1361 const char *ext
, void *data
, unsigned long sz
)
1363 switch (CACHE_EXT(ext
)) {
1364 case CACHE_EXT_TREE
:
1365 istate
->cache_tree
= cache_tree_read(data
, sz
);
1367 case CACHE_EXT_RESOLVE_UNDO
:
1368 istate
->resolve_undo
= resolve_undo_read(data
, sz
);
1370 case CACHE_EXT_LINK
:
1371 if (read_link_extension(istate
, data
, sz
))
1375 if (*ext
< 'A' || 'Z' < *ext
)
1376 return error("index uses %.4s extension, which we do not understand",
1378 fprintf(stderr
, "ignoring %.4s extension\n", ext
);
1384 int hold_locked_index(struct lock_file
*lk
, int die_on_error
)
1386 return hold_lock_file_for_update(lk
, get_index_file(),
1392 int read_index(struct index_state
*istate
)
1394 return read_index_from(istate
, get_index_file());
1397 static struct cache_entry
*cache_entry_from_ondisk(struct ondisk_cache_entry
*ondisk
,
1402 struct cache_entry
*ce
= xmalloc(cache_entry_size(len
));
1404 ce
->ce_stat_data
.sd_ctime
.sec
= get_be32(&ondisk
->ctime
.sec
);
1405 ce
->ce_stat_data
.sd_mtime
.sec
= get_be32(&ondisk
->mtime
.sec
);
1406 ce
->ce_stat_data
.sd_ctime
.nsec
= get_be32(&ondisk
->ctime
.nsec
);
1407 ce
->ce_stat_data
.sd_mtime
.nsec
= get_be32(&ondisk
->mtime
.nsec
);
1408 ce
->ce_stat_data
.sd_dev
= get_be32(&ondisk
->dev
);
1409 ce
->ce_stat_data
.sd_ino
= get_be32(&ondisk
->ino
);
1410 ce
->ce_mode
= get_be32(&ondisk
->mode
);
1411 ce
->ce_stat_data
.sd_uid
= get_be32(&ondisk
->uid
);
1412 ce
->ce_stat_data
.sd_gid
= get_be32(&ondisk
->gid
);
1413 ce
->ce_stat_data
.sd_size
= get_be32(&ondisk
->size
);
1414 ce
->ce_flags
= flags
& ~CE_NAMEMASK
;
1415 ce
->ce_namelen
= len
;
1417 hashcpy(ce
->sha1
, ondisk
->sha1
);
1418 memcpy(ce
->name
, name
, len
);
1419 ce
->name
[len
] = '\0';
1424 * Adjacent cache entries tend to share the leading paths, so it makes
1425 * sense to only store the differences in later entries. In the v4
1426 * on-disk format of the index, each on-disk cache entry stores the
1427 * number of bytes to be stripped from the end of the previous name,
1428 * and the bytes to append to the result, to come up with its name.
1430 static unsigned long expand_name_field(struct strbuf
*name
, const char *cp_
)
1432 const unsigned char *ep
, *cp
= (const unsigned char *)cp_
;
1433 size_t len
= decode_varint(&cp
);
1435 if (name
->len
< len
)
1436 die("malformed name field in the index");
1437 strbuf_remove(name
, name
->len
- len
, len
);
1438 for (ep
= cp
; *ep
; ep
++)
1439 ; /* find the end */
1440 strbuf_add(name
, cp
, ep
- cp
);
1441 return (const char *)ep
+ 1 - cp_
;
1444 static struct cache_entry
*create_from_disk(struct ondisk_cache_entry
*ondisk
,
1445 unsigned long *ent_size
,
1446 struct strbuf
*previous_name
)
1448 struct cache_entry
*ce
;
1453 /* On-disk flags are just 16 bits */
1454 flags
= get_be16(&ondisk
->flags
);
1455 len
= flags
& CE_NAMEMASK
;
1457 if (flags
& CE_EXTENDED
) {
1458 struct ondisk_cache_entry_extended
*ondisk2
;
1460 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
1461 extended_flags
= get_be16(&ondisk2
->flags2
) << 16;
1462 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1463 if (extended_flags
& ~CE_EXTENDED_FLAGS
)
1464 die("Unknown index entry format %08x", extended_flags
);
1465 flags
|= extended_flags
;
1466 name
= ondisk2
->name
;
1469 name
= ondisk
->name
;
1471 if (!previous_name
) {
1472 /* v3 and earlier */
1473 if (len
== CE_NAMEMASK
)
1475 ce
= cache_entry_from_ondisk(ondisk
, flags
, name
, len
);
1477 *ent_size
= ondisk_ce_size(ce
);
1479 unsigned long consumed
;
1480 consumed
= expand_name_field(previous_name
, name
);
1481 ce
= cache_entry_from_ondisk(ondisk
, flags
,
1483 previous_name
->len
);
1485 *ent_size
= (name
- ((char *)ondisk
)) + consumed
;
1490 static void check_ce_order(struct cache_entry
*ce
, struct cache_entry
*next_ce
)
1492 int name_compare
= strcmp(ce
->name
, next_ce
->name
);
1493 if (0 < name_compare
)
1494 die("unordered stage entries in index");
1495 if (!name_compare
) {
1497 die("multiple stage entries for merged file '%s'",
1499 if (ce_stage(ce
) > ce_stage(next_ce
))
1500 die("unordered stage entries for '%s'",
1505 /* remember to discard_cache() before reading a different cache! */
1506 int do_read_index(struct index_state
*istate
, const char *path
, int must_exist
)
1510 unsigned long src_offset
;
1511 struct cache_header
*hdr
;
1514 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
1516 if (istate
->initialized
)
1517 return istate
->cache_nr
;
1519 istate
->timestamp
.sec
= 0;
1520 istate
->timestamp
.nsec
= 0;
1521 fd
= open(path
, O_RDONLY
);
1523 if (!must_exist
&& errno
== ENOENT
)
1525 die_errno("%s: index file open failed", path
);
1529 die_errno("cannot stat the open index");
1531 mmap_size
= xsize_t(st
.st_size
);
1532 if (mmap_size
< sizeof(struct cache_header
) + 20)
1533 die("index file smaller than expected");
1535 mmap
= xmmap(NULL
, mmap_size
, PROT_READ
| PROT_WRITE
, MAP_PRIVATE
, fd
, 0);
1536 if (mmap
== MAP_FAILED
)
1537 die_errno("unable to map index file");
1541 if (verify_hdr(hdr
, mmap_size
) < 0)
1544 hashcpy(istate
->sha1
, (const unsigned char *)hdr
+ mmap_size
- 20);
1545 istate
->version
= ntohl(hdr
->hdr_version
);
1546 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
1547 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
1548 istate
->cache
= xcalloc(istate
->cache_alloc
, sizeof(*istate
->cache
));
1549 istate
->initialized
= 1;
1551 if (istate
->version
== 4)
1552 previous_name
= &previous_name_buf
;
1554 previous_name
= NULL
;
1556 src_offset
= sizeof(*hdr
);
1557 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1558 struct ondisk_cache_entry
*disk_ce
;
1559 struct cache_entry
*ce
;
1560 unsigned long consumed
;
1562 disk_ce
= (struct ondisk_cache_entry
*)((char *)mmap
+ src_offset
);
1563 ce
= create_from_disk(disk_ce
, &consumed
, previous_name
);
1564 set_index_entry(istate
, i
, ce
);
1567 check_ce_order(istate
->cache
[i
- 1], ce
);
1569 src_offset
+= consumed
;
1571 strbuf_release(&previous_name_buf
);
1572 istate
->timestamp
.sec
= st
.st_mtime
;
1573 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
1575 while (src_offset
<= mmap_size
- 20 - 8) {
1576 /* After an array of active_nr index entries,
1577 * there can be arbitrary number of extended
1578 * sections, each of which is prefixed with
1579 * extension name (4-byte) and section length
1580 * in 4-byte network byte order.
1583 memcpy(&extsize
, (char *)mmap
+ src_offset
+ 4, 4);
1584 extsize
= ntohl(extsize
);
1585 if (read_index_extension(istate
,
1586 (const char *) mmap
+ src_offset
,
1587 (char *) mmap
+ src_offset
+ 8,
1591 src_offset
+= extsize
;
1593 munmap(mmap
, mmap_size
);
1594 return istate
->cache_nr
;
1597 munmap(mmap
, mmap_size
);
1598 die("index file corrupt");
1601 int read_index_from(struct index_state
*istate
, const char *path
)
1603 struct split_index
*split_index
;
1606 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1607 if (istate
->initialized
)
1608 return istate
->cache_nr
;
1610 ret
= do_read_index(istate
, path
, 0);
1611 split_index
= istate
->split_index
;
1615 if (is_null_sha1(split_index
->base_sha1
))
1618 if (split_index
->base
)
1619 discard_index(split_index
->base
);
1621 split_index
->base
= xcalloc(1, sizeof(*split_index
->base
));
1622 ret
= do_read_index(split_index
->base
,
1623 git_path("sharedindex.%s",
1624 sha1_to_hex(split_index
->base_sha1
)), 1);
1625 if (hashcmp(split_index
->base_sha1
, split_index
->base
->sha1
))
1626 die("broken index, expect %s in %s, got %s",
1627 sha1_to_hex(split_index
->base_sha1
),
1628 git_path("sharedindex.%s",
1629 sha1_to_hex(split_index
->base_sha1
)),
1630 sha1_to_hex(split_index
->base
->sha1
));
1631 merge_base_index(istate
);
1635 int is_index_unborn(struct index_state
*istate
)
1637 return (!istate
->cache_nr
&& !istate
->timestamp
.sec
);
1640 int discard_index(struct index_state
*istate
)
1644 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1645 if (istate
->cache
[i
]->index
&&
1646 istate
->split_index
&&
1647 istate
->split_index
->base
&&
1648 istate
->cache
[i
]->index
<= istate
->split_index
->base
->cache_nr
&&
1649 istate
->cache
[i
] == istate
->split_index
->base
->cache
[istate
->cache
[i
]->index
- 1])
1651 free(istate
->cache
[i
]);
1653 resolve_undo_clear_index(istate
);
1654 istate
->cache_nr
= 0;
1655 istate
->cache_changed
= 0;
1656 istate
->timestamp
.sec
= 0;
1657 istate
->timestamp
.nsec
= 0;
1658 free_name_hash(istate
);
1659 cache_tree_free(&(istate
->cache_tree
));
1660 istate
->initialized
= 0;
1661 free(istate
->cache
);
1662 istate
->cache
= NULL
;
1663 istate
->cache_alloc
= 0;
1664 discard_split_index(istate
);
1668 int unmerged_index(const struct index_state
*istate
)
1671 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1672 if (ce_stage(istate
->cache
[i
]))
1678 #define WRITE_BUFFER_SIZE 8192
1679 static unsigned char write_buffer
[WRITE_BUFFER_SIZE
];
1680 static unsigned long write_buffer_len
;
1682 static int ce_write_flush(git_SHA_CTX
*context
, int fd
)
1684 unsigned int buffered
= write_buffer_len
;
1686 git_SHA1_Update(context
, write_buffer
, buffered
);
1687 if (write_in_full(fd
, write_buffer
, buffered
) != buffered
)
1689 write_buffer_len
= 0;
1694 static int ce_write(git_SHA_CTX
*context
, int fd
, void *data
, unsigned int len
)
1697 unsigned int buffered
= write_buffer_len
;
1698 unsigned int partial
= WRITE_BUFFER_SIZE
- buffered
;
1701 memcpy(write_buffer
+ buffered
, data
, partial
);
1702 buffered
+= partial
;
1703 if (buffered
== WRITE_BUFFER_SIZE
) {
1704 write_buffer_len
= buffered
;
1705 if (ce_write_flush(context
, fd
))
1709 write_buffer_len
= buffered
;
1711 data
= (char *) data
+ partial
;
1716 static int write_index_ext_header(git_SHA_CTX
*context
, int fd
,
1717 unsigned int ext
, unsigned int sz
)
1721 return ((ce_write(context
, fd
, &ext
, 4) < 0) ||
1722 (ce_write(context
, fd
, &sz
, 4) < 0)) ? -1 : 0;
1725 static int ce_flush(git_SHA_CTX
*context
, int fd
, unsigned char *sha1
)
1727 unsigned int left
= write_buffer_len
;
1730 write_buffer_len
= 0;
1731 git_SHA1_Update(context
, write_buffer
, left
);
1734 /* Flush first if not enough space for SHA1 signature */
1735 if (left
+ 20 > WRITE_BUFFER_SIZE
) {
1736 if (write_in_full(fd
, write_buffer
, left
) != left
)
1741 /* Append the SHA1 signature at the end */
1742 git_SHA1_Final(write_buffer
+ left
, context
);
1743 hashcpy(sha1
, write_buffer
+ left
);
1745 return (write_in_full(fd
, write_buffer
, left
) != left
) ? -1 : 0;
1748 static void ce_smudge_racily_clean_entry(struct cache_entry
*ce
)
1751 * The only thing we care about in this function is to smudge the
1752 * falsely clean entry due to touch-update-touch race, so we leave
1753 * everything else as they are. We are called for entries whose
1754 * ce_stat_data.sd_mtime match the index file mtime.
1756 * Note that this actually does not do much for gitlinks, for
1757 * which ce_match_stat_basic() always goes to the actual
1758 * contents. The caller checks with is_racy_timestamp() which
1759 * always says "no" for gitlinks, so we are not called for them ;-)
1763 if (lstat(ce
->name
, &st
) < 0)
1765 if (ce_match_stat_basic(ce
, &st
))
1767 if (ce_modified_check_fs(ce
, &st
)) {
1768 /* This is "racily clean"; smudge it. Note that this
1769 * is a tricky code. At first glance, it may appear
1770 * that it can break with this sequence:
1772 * $ echo xyzzy >frotz
1773 * $ git-update-index --add frotz
1776 * $ echo filfre >nitfol
1777 * $ git-update-index --add nitfol
1779 * but it does not. When the second update-index runs,
1780 * it notices that the entry "frotz" has the same timestamp
1781 * as index, and if we were to smudge it by resetting its
1782 * size to zero here, then the object name recorded
1783 * in index is the 6-byte file but the cached stat information
1784 * becomes zero --- which would then match what we would
1785 * obtain from the filesystem next time we stat("frotz").
1787 * However, the second update-index, before calling
1788 * this function, notices that the cached size is 6
1789 * bytes and what is on the filesystem is an empty
1790 * file, and never calls us, so the cached size information
1791 * for "frotz" stays 6 which does not match the filesystem.
1793 ce
->ce_stat_data
.sd_size
= 0;
1797 /* Copy miscellaneous fields but not the name */
1798 static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry
*ondisk
,
1799 struct cache_entry
*ce
)
1803 ondisk
->ctime
.sec
= htonl(ce
->ce_stat_data
.sd_ctime
.sec
);
1804 ondisk
->mtime
.sec
= htonl(ce
->ce_stat_data
.sd_mtime
.sec
);
1805 ondisk
->ctime
.nsec
= htonl(ce
->ce_stat_data
.sd_ctime
.nsec
);
1806 ondisk
->mtime
.nsec
= htonl(ce
->ce_stat_data
.sd_mtime
.nsec
);
1807 ondisk
->dev
= htonl(ce
->ce_stat_data
.sd_dev
);
1808 ondisk
->ino
= htonl(ce
->ce_stat_data
.sd_ino
);
1809 ondisk
->mode
= htonl(ce
->ce_mode
);
1810 ondisk
->uid
= htonl(ce
->ce_stat_data
.sd_uid
);
1811 ondisk
->gid
= htonl(ce
->ce_stat_data
.sd_gid
);
1812 ondisk
->size
= htonl(ce
->ce_stat_data
.sd_size
);
1813 hashcpy(ondisk
->sha1
, ce
->sha1
);
1815 flags
= ce
->ce_flags
& ~CE_NAMEMASK
;
1816 flags
|= (ce_namelen(ce
) >= CE_NAMEMASK
? CE_NAMEMASK
: ce_namelen(ce
));
1817 ondisk
->flags
= htons(flags
);
1818 if (ce
->ce_flags
& CE_EXTENDED
) {
1819 struct ondisk_cache_entry_extended
*ondisk2
;
1820 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
1821 ondisk2
->flags2
= htons((ce
->ce_flags
& CE_EXTENDED_FLAGS
) >> 16);
1822 return ondisk2
->name
;
1825 return ondisk
->name
;
1829 static int ce_write_entry(git_SHA_CTX
*c
, int fd
, struct cache_entry
*ce
,
1830 struct strbuf
*previous_name
)
1833 struct ondisk_cache_entry
*ondisk
;
1834 int saved_namelen
= saved_namelen
; /* compiler workaround */
1838 if (ce
->ce_flags
& CE_STRIP_NAME
) {
1839 saved_namelen
= ce_namelen(ce
);
1843 if (!previous_name
) {
1844 size
= ondisk_ce_size(ce
);
1845 ondisk
= xcalloc(1, size
);
1846 name
= copy_cache_entry_to_ondisk(ondisk
, ce
);
1847 memcpy(name
, ce
->name
, ce_namelen(ce
));
1849 int common
, to_remove
, prefix_size
;
1850 unsigned char to_remove_vi
[16];
1852 (ce
->name
[common
] &&
1853 common
< previous_name
->len
&&
1854 ce
->name
[common
] == previous_name
->buf
[common
]);
1856 ; /* still matching */
1857 to_remove
= previous_name
->len
- common
;
1858 prefix_size
= encode_varint(to_remove
, to_remove_vi
);
1860 if (ce
->ce_flags
& CE_EXTENDED
)
1861 size
= offsetof(struct ondisk_cache_entry_extended
, name
);
1863 size
= offsetof(struct ondisk_cache_entry
, name
);
1864 size
+= prefix_size
+ (ce_namelen(ce
) - common
+ 1);
1866 ondisk
= xcalloc(1, size
);
1867 name
= copy_cache_entry_to_ondisk(ondisk
, ce
);
1868 memcpy(name
, to_remove_vi
, prefix_size
);
1869 memcpy(name
+ prefix_size
, ce
->name
+ common
, ce_namelen(ce
) - common
);
1871 strbuf_splice(previous_name
, common
, to_remove
,
1872 ce
->name
+ common
, ce_namelen(ce
) - common
);
1874 if (ce
->ce_flags
& CE_STRIP_NAME
) {
1875 ce
->ce_namelen
= saved_namelen
;
1876 ce
->ce_flags
&= ~CE_STRIP_NAME
;
1879 result
= ce_write(c
, fd
, ondisk
, size
);
1885 * This function verifies if index_state has the correct sha1 of the
1886 * index file. Don't die if we have any other failure, just return 0.
1888 static int verify_index_from(const struct index_state
*istate
, const char *path
)
1893 unsigned char sha1
[20];
1895 if (!istate
->initialized
)
1898 fd
= open(path
, O_RDONLY
);
1905 if (st
.st_size
< sizeof(struct cache_header
) + 20)
1908 n
= pread_in_full(fd
, sha1
, 20, st
.st_size
- 20);
1912 if (hashcmp(istate
->sha1
, sha1
))
1923 static int verify_index(const struct index_state
*istate
)
1925 return verify_index_from(istate
, get_index_file());
1928 static int has_racy_timestamp(struct index_state
*istate
)
1930 int entries
= istate
->cache_nr
;
1933 for (i
= 0; i
< entries
; i
++) {
1934 struct cache_entry
*ce
= istate
->cache
[i
];
1935 if (is_racy_timestamp(istate
, ce
))
1942 * Opportunistically update the index but do not complain if we can't
1944 void update_index_if_able(struct index_state
*istate
, struct lock_file
*lockfile
)
1946 if ((istate
->cache_changed
|| has_racy_timestamp(istate
)) &&
1947 verify_index(istate
) &&
1948 write_locked_index(istate
, lockfile
, COMMIT_LOCK
))
1949 rollback_lock_file(lockfile
);
1952 static int do_write_index(struct index_state
*istate
, int newfd
,
1953 int strip_extensions
)
1956 struct cache_header hdr
;
1957 int i
, err
, removed
, extended
, hdr_version
;
1958 struct cache_entry
**cache
= istate
->cache
;
1959 int entries
= istate
->cache_nr
;
1961 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
1963 for (i
= removed
= extended
= 0; i
< entries
; i
++) {
1964 if (cache
[i
]->ce_flags
& CE_REMOVE
)
1967 /* reduce extended entries if possible */
1968 cache
[i
]->ce_flags
&= ~CE_EXTENDED
;
1969 if (cache
[i
]->ce_flags
& CE_EXTENDED_FLAGS
) {
1971 cache
[i
]->ce_flags
|= CE_EXTENDED
;
1975 if (!istate
->version
) {
1976 istate
->version
= get_index_format_default();
1977 if (getenv("GIT_TEST_SPLIT_INDEX"))
1978 init_split_index(istate
);
1981 /* demote version 3 to version 2 when the latter suffices */
1982 if (istate
->version
== 3 || istate
->version
== 2)
1983 istate
->version
= extended
? 3 : 2;
1985 hdr_version
= istate
->version
;
1987 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
1988 hdr
.hdr_version
= htonl(hdr_version
);
1989 hdr
.hdr_entries
= htonl(entries
- removed
);
1992 if (ce_write(&c
, newfd
, &hdr
, sizeof(hdr
)) < 0)
1995 previous_name
= (hdr_version
== 4) ? &previous_name_buf
: NULL
;
1996 for (i
= 0; i
< entries
; i
++) {
1997 struct cache_entry
*ce
= cache
[i
];
1998 if (ce
->ce_flags
& CE_REMOVE
)
2000 if (!ce_uptodate(ce
) && is_racy_timestamp(istate
, ce
))
2001 ce_smudge_racily_clean_entry(ce
);
2002 if (is_null_sha1(ce
->sha1
)) {
2003 static const char msg
[] = "cache entry has null sha1: %s";
2004 static int allow
= -1;
2007 allow
= git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2009 warning(msg
, ce
->name
);
2011 return error(msg
, ce
->name
);
2013 if (ce_write_entry(&c
, newfd
, ce
, previous_name
) < 0)
2016 strbuf_release(&previous_name_buf
);
2018 /* Write extension data here */
2019 if (!strip_extensions
&& istate
->split_index
) {
2020 struct strbuf sb
= STRBUF_INIT
;
2022 err
= write_link_extension(&sb
, istate
) < 0 ||
2023 write_index_ext_header(&c
, newfd
, CACHE_EXT_LINK
,
2025 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2026 strbuf_release(&sb
);
2030 if (!strip_extensions
&& istate
->cache_tree
) {
2031 struct strbuf sb
= STRBUF_INIT
;
2033 cache_tree_write(&sb
, istate
->cache_tree
);
2034 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_TREE
, sb
.len
) < 0
2035 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2036 strbuf_release(&sb
);
2040 if (!strip_extensions
&& istate
->resolve_undo
) {
2041 struct strbuf sb
= STRBUF_INIT
;
2043 resolve_undo_write(&sb
, istate
->resolve_undo
);
2044 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_RESOLVE_UNDO
,
2046 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2047 strbuf_release(&sb
);
2051 if (!strip_extensions
&& istate
->untracked
) {
2052 struct strbuf sb
= STRBUF_INIT
;
2054 write_untracked_extension(&sb
, istate
->untracked
);
2055 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_UNTRACKED
,
2057 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2058 strbuf_release(&sb
);
2063 if (ce_flush(&c
, newfd
, istate
->sha1
) || fstat(newfd
, &st
))
2065 istate
->timestamp
.sec
= (unsigned int)st
.st_mtime
;
2066 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
2070 void set_alternate_index_output(const char *name
)
2072 alternate_index_output
= name
;
2075 static int commit_locked_index(struct lock_file
*lk
)
2077 if (alternate_index_output
)
2078 return commit_lock_file_to(lk
, alternate_index_output
);
2080 return commit_lock_file(lk
);
2083 static int do_write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
2086 int ret
= do_write_index(istate
, lock
->fd
, 0);
2089 assert((flags
& (COMMIT_LOCK
| CLOSE_LOCK
)) !=
2090 (COMMIT_LOCK
| CLOSE_LOCK
));
2091 if (flags
& COMMIT_LOCK
)
2092 return commit_locked_index(lock
);
2093 else if (flags
& CLOSE_LOCK
)
2094 return close_lock_file(lock
);
2099 static int write_split_index(struct index_state
*istate
,
2100 struct lock_file
*lock
,
2104 prepare_to_write_split_index(istate
);
2105 ret
= do_write_locked_index(istate
, lock
, flags
);
2106 finish_writing_split_index(istate
);
2110 static char *temporary_sharedindex
;
2112 static void remove_temporary_sharedindex(void)
2114 if (temporary_sharedindex
) {
2115 unlink_or_warn(temporary_sharedindex
);
2116 free(temporary_sharedindex
);
2117 temporary_sharedindex
= NULL
;
2121 static void remove_temporary_sharedindex_on_signal(int signo
)
2123 remove_temporary_sharedindex();
2124 sigchain_pop(signo
);
2128 static int write_shared_index(struct index_state
*istate
,
2129 struct lock_file
*lock
, unsigned flags
)
2131 struct split_index
*si
= istate
->split_index
;
2132 static int installed_handler
;
2135 temporary_sharedindex
= git_pathdup("sharedindex_XXXXXX");
2136 fd
= mkstemp(temporary_sharedindex
);
2138 free(temporary_sharedindex
);
2139 temporary_sharedindex
= NULL
;
2140 hashclr(si
->base_sha1
);
2141 return do_write_locked_index(istate
, lock
, flags
);
2143 if (!installed_handler
) {
2144 atexit(remove_temporary_sharedindex
);
2145 sigchain_push_common(remove_temporary_sharedindex_on_signal
);
2147 move_cache_to_base_index(istate
);
2148 ret
= do_write_index(si
->base
, fd
, 1);
2151 remove_temporary_sharedindex();
2154 ret
= rename(temporary_sharedindex
,
2155 git_path("sharedindex.%s", sha1_to_hex(si
->base
->sha1
)));
2156 free(temporary_sharedindex
);
2157 temporary_sharedindex
= NULL
;
2159 hashcpy(si
->base_sha1
, si
->base
->sha1
);
2163 int write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
2166 struct split_index
*si
= istate
->split_index
;
2168 if (!si
|| alternate_index_output
||
2169 (istate
->cache_changed
& ~EXTMASK
)) {
2171 hashclr(si
->base_sha1
);
2172 return do_write_locked_index(istate
, lock
, flags
);
2175 if (getenv("GIT_TEST_SPLIT_INDEX")) {
2176 int v
= si
->base_sha1
[0];
2178 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
2180 if (istate
->cache_changed
& SPLIT_INDEX_ORDERED
) {
2181 int ret
= write_shared_index(istate
, lock
, flags
);
2186 return write_split_index(istate
, lock
, flags
);
2190 * Read the index file that is potentially unmerged into given
2191 * index_state, dropping any unmerged entries. Returns true if
2192 * the index is unmerged. Callers who want to refuse to work
2193 * from an unmerged state can call this and check its return value,
2194 * instead of calling read_cache().
2196 int read_index_unmerged(struct index_state
*istate
)
2202 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2203 struct cache_entry
*ce
= istate
->cache
[i
];
2204 struct cache_entry
*new_ce
;
2210 len
= ce_namelen(ce
);
2211 size
= cache_entry_size(len
);
2212 new_ce
= xcalloc(1, size
);
2213 memcpy(new_ce
->name
, ce
->name
, len
);
2214 new_ce
->ce_flags
= create_ce_flags(0) | CE_CONFLICTED
;
2215 new_ce
->ce_namelen
= len
;
2216 new_ce
->ce_mode
= ce
->ce_mode
;
2217 if (add_index_entry(istate
, new_ce
, 0))
2218 return error("%s: cannot drop to stage #0",
2225 * Returns 1 if the path is an "other" path with respect to
2226 * the index; that is, the path is not mentioned in the index at all,
2227 * either as a file, a directory with some files in the index,
2228 * or as an unmerged entry.
2230 * We helpfully remove a trailing "/" from directories so that
2231 * the output of read_directory can be used as-is.
2233 int index_name_is_other(const struct index_state
*istate
, const char *name
,
2237 if (namelen
&& name
[namelen
- 1] == '/')
2239 pos
= index_name_pos(istate
, name
, namelen
);
2241 return 0; /* exact match */
2243 if (pos
< istate
->cache_nr
) {
2244 struct cache_entry
*ce
= istate
->cache
[pos
];
2245 if (ce_namelen(ce
) == namelen
&&
2246 !memcmp(ce
->name
, name
, namelen
))
2247 return 0; /* Yup, this one exists unmerged */
2252 void *read_blob_data_from_index(struct index_state
*istate
, const char *path
, unsigned long *size
)
2256 enum object_type type
;
2260 pos
= index_name_pos(istate
, path
, len
);
2263 * We might be in the middle of a merge, in which
2264 * case we would read stage #2 (ours).
2268 (pos
< 0 && i
< istate
->cache_nr
&&
2269 !strcmp(istate
->cache
[i
]->name
, path
));
2271 if (ce_stage(istate
->cache
[i
]) == 2)
2276 data
= read_sha1_file(istate
->cache
[pos
]->sha1
, &type
, &sz
);
2277 if (!data
|| type
!= OBJ_BLOB
) {
2286 void stat_validity_clear(struct stat_validity
*sv
)
2292 int stat_validity_check(struct stat_validity
*sv
, const char *path
)
2296 if (stat(path
, &st
) < 0)
2297 return sv
->sd
== NULL
;
2300 return S_ISREG(st
.st_mode
) && !match_stat_data(sv
->sd
, &st
);
2303 void stat_validity_update(struct stat_validity
*sv
, int fd
)
2307 if (fstat(fd
, &st
) < 0 || !S_ISREG(st
.st_mode
))
2308 stat_validity_clear(sv
);
2311 sv
->sd
= xcalloc(1, sizeof(struct stat_data
));
2312 fill_stat_data(sv
->sd
, &st
);