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 #include "resolve-undo.h"
18 static struct cache_entry
*refresh_cache_entry(struct cache_entry
*ce
,
19 unsigned int options
);
21 /* Mask for the name length in ce_flags in the on-disk index */
23 #define CE_NAMEMASK (0x0fff)
27 * The first letter should be 'A'..'Z' for extensions that are not
28 * necessary for a correct operation (i.e. optimization data).
29 * When new extensions are added that _needs_ to be understood in
30 * order to correctly interpret the index file, pick character that
31 * is outside the range, to cause the reader to abort.
34 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
35 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
36 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
38 struct index_state the_index
;
40 static void set_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
42 istate
->cache
[nr
] = ce
;
43 add_name_hash(istate
, ce
);
46 static void replace_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
48 struct cache_entry
*old
= istate
->cache
[nr
];
50 remove_name_hash(istate
, old
);
51 set_index_entry(istate
, nr
, ce
);
52 istate
->cache_changed
= 1;
55 void rename_index_entry_at(struct index_state
*istate
, int nr
, const char *new_name
)
57 struct cache_entry
*old
= istate
->cache
[nr
], *new;
58 int namelen
= strlen(new_name
);
60 new = xmalloc(cache_entry_size(namelen
));
61 copy_cache_entry(new, old
);
62 new->ce_flags
&= ~CE_STATE_MASK
;
63 new->ce_namelen
= namelen
;
64 memcpy(new->name
, new_name
, namelen
+ 1);
66 cache_tree_invalidate_path(istate
->cache_tree
, old
->name
);
67 remove_index_entry_at(istate
, nr
);
68 add_index_entry(istate
, new, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
71 void fill_stat_data(struct stat_data
*sd
, struct stat
*st
)
73 sd
->sd_ctime
.sec
= (unsigned int)st
->st_ctime
;
74 sd
->sd_mtime
.sec
= (unsigned int)st
->st_mtime
;
75 sd
->sd_ctime
.nsec
= ST_CTIME_NSEC(*st
);
76 sd
->sd_mtime
.nsec
= ST_MTIME_NSEC(*st
);
77 sd
->sd_dev
= st
->st_dev
;
78 sd
->sd_ino
= st
->st_ino
;
79 sd
->sd_uid
= st
->st_uid
;
80 sd
->sd_gid
= st
->st_gid
;
81 sd
->sd_size
= st
->st_size
;
84 int match_stat_data(const struct stat_data
*sd
, struct stat
*st
)
88 if (sd
->sd_mtime
.sec
!= (unsigned int)st
->st_mtime
)
89 changed
|= MTIME_CHANGED
;
90 if (trust_ctime
&& check_stat
&&
91 sd
->sd_ctime
.sec
!= (unsigned int)st
->st_ctime
)
92 changed
|= CTIME_CHANGED
;
95 if (check_stat
&& sd
->sd_mtime
.nsec
!= ST_MTIME_NSEC(*st
))
96 changed
|= MTIME_CHANGED
;
97 if (trust_ctime
&& check_stat
&&
98 sd
->sd_ctime
.nsec
!= ST_CTIME_NSEC(*st
))
99 changed
|= CTIME_CHANGED
;
103 if (sd
->sd_uid
!= (unsigned int) st
->st_uid
||
104 sd
->sd_gid
!= (unsigned int) st
->st_gid
)
105 changed
|= OWNER_CHANGED
;
106 if (sd
->sd_ino
!= (unsigned int) st
->st_ino
)
107 changed
|= INODE_CHANGED
;
112 * st_dev breaks on network filesystems where different
113 * clients will have different views of what "device"
114 * the filesystem is on
116 if (check_stat
&& sd
->sd_dev
!= (unsigned int) st
->st_dev
)
117 changed
|= INODE_CHANGED
;
120 if (sd
->sd_size
!= (unsigned int) st
->st_size
)
121 changed
|= DATA_CHANGED
;
127 * This only updates the "non-critical" parts of the directory
128 * cache, ie the parts that aren't tracked by GIT, and only used
129 * to validate the cache.
131 void fill_stat_cache_info(struct cache_entry
*ce
, struct stat
*st
)
133 fill_stat_data(&ce
->ce_stat_data
, st
);
135 if (assume_unchanged
)
136 ce
->ce_flags
|= CE_VALID
;
138 if (S_ISREG(st
->st_mode
))
139 ce_mark_uptodate(ce
);
142 static int ce_compare_data(const struct cache_entry
*ce
, struct stat
*st
)
145 int fd
= open(ce
->name
, O_RDONLY
);
148 unsigned char sha1
[20];
149 if (!index_fd(sha1
, fd
, st
, OBJ_BLOB
, ce
->name
, 0))
150 match
= hashcmp(sha1
, ce
->sha1
);
151 /* index_fd() closed the file descriptor already */
156 static int ce_compare_link(const struct cache_entry
*ce
, size_t expected_size
)
161 enum object_type type
;
162 struct strbuf sb
= STRBUF_INIT
;
164 if (strbuf_readlink(&sb
, ce
->name
, expected_size
))
167 buffer
= read_sha1_file(ce
->sha1
, &type
, &size
);
170 match
= memcmp(buffer
, sb
.buf
, size
);
177 static int ce_compare_gitlink(const struct cache_entry
*ce
)
179 unsigned char sha1
[20];
182 * We don't actually require that the .git directory
183 * under GITLINK directory be a valid git directory. It
184 * might even be missing (in case nobody populated that
187 * If so, we consider it always to match.
189 if (resolve_gitlink_ref(ce
->name
, "HEAD", sha1
) < 0)
191 return hashcmp(sha1
, ce
->sha1
);
194 static int ce_modified_check_fs(const struct cache_entry
*ce
, struct stat
*st
)
196 switch (st
->st_mode
& S_IFMT
) {
198 if (ce_compare_data(ce
, st
))
202 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
206 if (S_ISGITLINK(ce
->ce_mode
))
207 return ce_compare_gitlink(ce
) ? DATA_CHANGED
: 0;
214 static int ce_match_stat_basic(const struct cache_entry
*ce
, struct stat
*st
)
216 unsigned int changed
= 0;
218 if (ce
->ce_flags
& CE_REMOVE
)
219 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
221 switch (ce
->ce_mode
& S_IFMT
) {
223 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
224 /* We consider only the owner x bit to be relevant for
227 if (trust_executable_bit
&&
228 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
229 changed
|= MODE_CHANGED
;
232 if (!S_ISLNK(st
->st_mode
) &&
233 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
234 changed
|= TYPE_CHANGED
;
237 /* We ignore most of the st_xxx fields for gitlinks */
238 if (!S_ISDIR(st
->st_mode
))
239 changed
|= TYPE_CHANGED
;
240 else if (ce_compare_gitlink(ce
))
241 changed
|= DATA_CHANGED
;
244 die("internal error: ce_mode is %o", ce
->ce_mode
);
247 changed
|= match_stat_data(&ce
->ce_stat_data
, st
);
249 /* Racily smudged entry? */
250 if (!ce
->ce_stat_data
.sd_size
) {
251 if (!is_empty_blob_sha1(ce
->sha1
))
252 changed
|= DATA_CHANGED
;
258 static int is_racy_timestamp(const struct index_state
*istate
,
259 const struct cache_entry
*ce
)
261 return (!S_ISGITLINK(ce
->ce_mode
) &&
262 istate
->timestamp
.sec
&&
264 /* nanosecond timestamped files can also be racy! */
265 (istate
->timestamp
.sec
< ce
->ce_stat_data
.sd_mtime
.sec
||
266 (istate
->timestamp
.sec
== ce
->ce_stat_data
.sd_mtime
.sec
&&
267 istate
->timestamp
.nsec
<= ce
->ce_stat_data
.sd_mtime
.nsec
))
269 istate
->timestamp
.sec
<= ce
->ce_stat_data
.sd_mtime
.sec
274 int ie_match_stat(const struct index_state
*istate
,
275 const struct cache_entry
*ce
, struct stat
*st
,
276 unsigned int options
)
278 unsigned int changed
;
279 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
280 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
281 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
284 * If it's marked as always valid in the index, it's
285 * valid whatever the checked-out copy says.
287 * skip-worktree has the same effect with higher precedence
289 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
))
291 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
295 * Intent-to-add entries have not been added, so the index entry
296 * by definition never matches what is in the work tree until it
297 * actually gets added.
299 if (ce
->ce_flags
& CE_INTENT_TO_ADD
)
300 return DATA_CHANGED
| TYPE_CHANGED
| MODE_CHANGED
;
302 changed
= ce_match_stat_basic(ce
, st
);
305 * Within 1 second of this sequence:
306 * echo xyzzy >file && git-update-index --add file
307 * running this command:
309 * would give a falsely clean cache entry. The mtime and
310 * length match the cache, and other stat fields do not change.
312 * We could detect this at update-index time (the cache entry
313 * being registered/updated records the same time as "now")
314 * and delay the return from git-update-index, but that would
315 * effectively mean we can make at most one commit per second,
316 * which is not acceptable. Instead, we check cache entries
317 * whose mtime are the same as the index file timestamp more
318 * carefully than others.
320 if (!changed
&& is_racy_timestamp(istate
, ce
)) {
321 if (assume_racy_is_modified
)
322 changed
|= DATA_CHANGED
;
324 changed
|= ce_modified_check_fs(ce
, st
);
330 int ie_modified(const struct index_state
*istate
,
331 const struct cache_entry
*ce
,
332 struct stat
*st
, unsigned int options
)
334 int changed
, changed_fs
;
336 changed
= ie_match_stat(istate
, ce
, st
, options
);
340 * If the mode or type has changed, there's no point in trying
341 * to refresh the entry - it's not going to match
343 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
347 * Immediately after read-tree or update-index --cacheinfo,
348 * the length field is zero, as we have never even read the
349 * lstat(2) information once, and we cannot trust DATA_CHANGED
350 * returned by ie_match_stat() which in turn was returned by
351 * ce_match_stat_basic() to signal that the filesize of the
352 * blob changed. We have to actually go to the filesystem to
353 * see if the contents match, and if so, should answer "unchanged".
355 * The logic does not apply to gitlinks, as ce_match_stat_basic()
356 * already has checked the actual HEAD from the filesystem in the
357 * subproject. If ie_match_stat() already said it is different,
358 * then we know it is.
360 if ((changed
& DATA_CHANGED
) &&
361 (S_ISGITLINK(ce
->ce_mode
) || ce
->ce_stat_data
.sd_size
!= 0))
364 changed_fs
= ce_modified_check_fs(ce
, st
);
366 return changed
| changed_fs
;
370 int base_name_compare(const char *name1
, int len1
, int mode1
,
371 const char *name2
, int len2
, int mode2
)
373 unsigned char c1
, c2
;
374 int len
= len1
< len2
? len1
: len2
;
377 cmp
= memcmp(name1
, name2
, len
);
382 if (!c1
&& S_ISDIR(mode1
))
384 if (!c2
&& S_ISDIR(mode2
))
386 return (c1
< c2
) ? -1 : (c1
> c2
) ? 1 : 0;
390 * df_name_compare() is identical to base_name_compare(), except it
391 * compares conflicting directory/file entries as equal. Note that
392 * while a directory name compares as equal to a regular file, they
393 * then individually compare _differently_ to a filename that has
394 * a dot after the basename (because '\0' < '.' < '/').
396 * This is used by routines that want to traverse the git namespace
397 * but then handle conflicting entries together when possible.
399 int df_name_compare(const char *name1
, int len1
, int mode1
,
400 const char *name2
, int len2
, int mode2
)
402 int len
= len1
< len2
? len1
: len2
, cmp
;
403 unsigned char c1
, c2
;
405 cmp
= memcmp(name1
, name2
, len
);
408 /* Directories and files compare equal (same length, same name) */
412 if (!c1
&& S_ISDIR(mode1
))
415 if (!c2
&& S_ISDIR(mode2
))
417 if (c1
== '/' && !c2
)
419 if (c2
== '/' && !c1
)
424 int cache_name_stage_compare(const char *name1
, int len1
, int stage1
, const char *name2
, int len2
, int stage2
)
426 int len
= len1
< len2
? len1
: len2
;
429 cmp
= memcmp(name1
, name2
, len
);
444 int cache_name_compare(const char *name1
, int len1
, const char *name2
, int len2
)
446 return cache_name_stage_compare(name1
, len1
, 0, name2
, len2
, 0);
449 static int index_name_stage_pos(const struct index_state
*istate
, const char *name
, int namelen
, int stage
)
454 last
= istate
->cache_nr
;
455 while (last
> first
) {
456 int next
= (last
+ first
) >> 1;
457 struct cache_entry
*ce
= istate
->cache
[next
];
458 int cmp
= cache_name_stage_compare(name
, namelen
, stage
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
470 int index_name_pos(const struct index_state
*istate
, const char *name
, int namelen
)
472 return index_name_stage_pos(istate
, name
, namelen
, 0);
475 /* Remove entry, return true if there are more entries to go.. */
476 int remove_index_entry_at(struct index_state
*istate
, int pos
)
478 struct cache_entry
*ce
= istate
->cache
[pos
];
480 record_resolve_undo(istate
, ce
);
481 remove_name_hash(istate
, ce
);
482 istate
->cache_changed
= 1;
484 if (pos
>= istate
->cache_nr
)
486 memmove(istate
->cache
+ pos
,
487 istate
->cache
+ pos
+ 1,
488 (istate
->cache_nr
- pos
) * sizeof(struct cache_entry
*));
493 * Remove all cache entries marked for removal, that is where
494 * CE_REMOVE is set in ce_flags. This is much more effective than
495 * calling remove_index_entry_at() for each entry to be removed.
497 void remove_marked_cache_entries(struct index_state
*istate
)
499 struct cache_entry
**ce_array
= istate
->cache
;
502 for (i
= j
= 0; i
< istate
->cache_nr
; i
++) {
503 if (ce_array
[i
]->ce_flags
& CE_REMOVE
)
504 remove_name_hash(istate
, ce_array
[i
]);
506 ce_array
[j
++] = ce_array
[i
];
508 istate
->cache_changed
= 1;
509 istate
->cache_nr
= j
;
512 int remove_file_from_index(struct index_state
*istate
, const char *path
)
514 int pos
= index_name_pos(istate
, path
, strlen(path
));
517 cache_tree_invalidate_path(istate
->cache_tree
, path
);
518 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
519 remove_index_entry_at(istate
, pos
);
523 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
525 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
528 static int index_name_pos_also_unmerged(struct index_state
*istate
,
529 const char *path
, int namelen
)
531 int pos
= index_name_pos(istate
, path
, namelen
);
532 struct cache_entry
*ce
;
537 /* maybe unmerged? */
539 if (pos
>= istate
->cache_nr
||
540 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
543 /* order of preference: stage 2, 1, 3 */
544 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
545 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
546 !compare_name(ce
, path
, namelen
))
551 static int different_name(struct cache_entry
*ce
, struct cache_entry
*alias
)
553 int len
= ce_namelen(ce
);
554 return ce_namelen(alias
) != len
|| memcmp(ce
->name
, alias
->name
, len
);
558 * If we add a filename that aliases in the cache, we will use the
559 * name that we already have - but we don't want to update the same
560 * alias twice, because that implies that there were actually two
561 * different files with aliasing names!
563 * So we use the CE_ADDED flag to verify that the alias was an old
564 * one before we accept it as
566 static struct cache_entry
*create_alias_ce(struct cache_entry
*ce
, struct cache_entry
*alias
)
569 struct cache_entry
*new;
571 if (alias
->ce_flags
& CE_ADDED
)
572 die("Will not add file alias '%s' ('%s' already exists in index)", ce
->name
, alias
->name
);
574 /* Ok, create the new entry using the name of the existing alias */
575 len
= ce_namelen(alias
);
576 new = xcalloc(1, cache_entry_size(len
));
577 memcpy(new->name
, alias
->name
, len
);
578 copy_cache_entry(new, ce
);
583 static void record_intent_to_add(struct cache_entry
*ce
)
585 unsigned char sha1
[20];
586 if (write_sha1_file("", 0, blob_type
, sha1
))
587 die("cannot create an empty blob in the object database");
588 hashcpy(ce
->sha1
, sha1
);
591 int add_to_index(struct index_state
*istate
, const char *path
, struct stat
*st
, int flags
)
593 int size
, namelen
, was_same
;
594 mode_t st_mode
= st
->st_mode
;
595 struct cache_entry
*ce
, *alias
;
596 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_IGNORE_SKIP_WORKTREE
|CE_MATCH_RACY_IS_DIRTY
;
597 int verbose
= flags
& (ADD_CACHE_VERBOSE
| ADD_CACHE_PRETEND
);
598 int pretend
= flags
& ADD_CACHE_PRETEND
;
599 int intent_only
= flags
& ADD_CACHE_INTENT
;
600 int add_option
= (ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
|
601 (intent_only
? ADD_CACHE_NEW_ONLY
: 0));
603 if (!S_ISREG(st_mode
) && !S_ISLNK(st_mode
) && !S_ISDIR(st_mode
))
604 return error("%s: can only add regular files, symbolic links or git-directories", path
);
606 namelen
= strlen(path
);
607 if (S_ISDIR(st_mode
)) {
608 while (namelen
&& path
[namelen
-1] == '/')
611 size
= cache_entry_size(namelen
);
612 ce
= xcalloc(1, size
);
613 memcpy(ce
->name
, path
, namelen
);
614 ce
->ce_namelen
= namelen
;
616 fill_stat_cache_info(ce
, st
);
618 ce
->ce_flags
|= CE_INTENT_TO_ADD
;
620 if (trust_executable_bit
&& has_symlinks
)
621 ce
->ce_mode
= create_ce_mode(st_mode
);
623 /* If there is an existing entry, pick the mode bits and type
624 * from it, otherwise assume unexecutable regular file.
626 struct cache_entry
*ent
;
627 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
629 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
630 ce
->ce_mode
= ce_mode_from_stat(ent
, st_mode
);
633 /* When core.ignorecase=true, determine if a directory of the same name but differing
634 * case already exists within the Git repository. If it does, ensure the directory
635 * case of the file being added to the repository matches (is folded into) the existing
636 * entry's directory case.
639 const char *startPtr
= ce
->name
;
640 const char *ptr
= startPtr
;
642 while (*ptr
&& *ptr
!= '/')
645 struct cache_entry
*foundce
;
647 foundce
= index_name_exists(istate
, ce
->name
, ptr
- ce
->name
, ignore_case
);
649 memcpy((void *)startPtr
, foundce
->name
+ (startPtr
- ce
->name
), ptr
- startPtr
);
656 alias
= index_name_exists(istate
, ce
->name
, ce_namelen(ce
), ignore_case
);
657 if (alias
&& !ce_stage(alias
) && !ie_match_stat(istate
, alias
, st
, ce_option
)) {
658 /* Nothing changed, really */
660 if (!S_ISGITLINK(alias
->ce_mode
))
661 ce_mark_uptodate(alias
);
662 alias
->ce_flags
|= CE_ADDED
;
666 if (index_path(ce
->sha1
, path
, st
, HASH_WRITE_OBJECT
))
667 return error("unable to index file %s", path
);
669 record_intent_to_add(ce
);
671 if (ignore_case
&& alias
&& different_name(ce
, alias
))
672 ce
= create_alias_ce(ce
, alias
);
673 ce
->ce_flags
|= CE_ADDED
;
675 /* It was suspected to be racily clean, but it turns out to be Ok */
678 !hashcmp(alias
->sha1
, ce
->sha1
) &&
679 ce
->ce_mode
== alias
->ce_mode
);
683 else if (add_index_entry(istate
, ce
, add_option
))
684 return error("unable to add %s to index",path
);
685 if (verbose
&& !was_same
)
686 printf("add '%s'\n", path
);
690 int add_file_to_index(struct index_state
*istate
, const char *path
, int flags
)
693 if (lstat(path
, &st
))
694 die_errno("unable to stat '%s'", path
);
695 return add_to_index(istate
, path
, &st
, flags
);
698 struct cache_entry
*make_cache_entry(unsigned int mode
,
699 const unsigned char *sha1
, const char *path
, int stage
,
700 unsigned int refresh_options
)
703 struct cache_entry
*ce
;
705 if (!verify_path(path
)) {
706 error("Invalid path '%s'", path
);
711 size
= cache_entry_size(len
);
712 ce
= xcalloc(1, size
);
714 hashcpy(ce
->sha1
, sha1
);
715 memcpy(ce
->name
, path
, len
);
716 ce
->ce_flags
= create_ce_flags(stage
);
717 ce
->ce_namelen
= len
;
718 ce
->ce_mode
= create_ce_mode(mode
);
720 return refresh_cache_entry(ce
, refresh_options
);
723 int ce_same_name(const struct cache_entry
*a
, const struct cache_entry
*b
)
725 int len
= ce_namelen(a
);
726 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
729 int ce_path_match(const struct cache_entry
*ce
, const struct pathspec
*pathspec
)
731 return match_pathspec_depth(pathspec
, ce
->name
, ce_namelen(ce
), 0, NULL
);
735 * We fundamentally don't like some paths: we don't want
736 * dot or dot-dot anywhere, and for obvious reasons don't
737 * want to recurse into ".git" either.
739 * Also, we don't want double slashes or slashes at the
740 * end that can make pathnames ambiguous.
742 static int verify_dotfile(const char *rest
)
745 * The first character was '.', but that
746 * has already been discarded, we now test
750 /* "." is not allowed */
751 if (*rest
== '\0' || is_dir_sep(*rest
))
756 * ".git" followed by NUL or slash is bad. This
757 * shares the path end test with the ".." case.
767 if (rest
[1] == '\0' || is_dir_sep(rest
[1]))
773 int verify_path(const char *path
)
777 if (has_dos_drive_prefix(path
))
787 if ((c
== '.' && !verify_dotfile(path
)) ||
788 is_dir_sep(c
) || c
== '\0')
796 * Do we have another file that has the beginning components being a
797 * proper superset of the name we're trying to add?
799 static int has_file_name(struct index_state
*istate
,
800 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
803 int len
= ce_namelen(ce
);
804 int stage
= ce_stage(ce
);
805 const char *name
= ce
->name
;
807 while (pos
< istate
->cache_nr
) {
808 struct cache_entry
*p
= istate
->cache
[pos
++];
810 if (len
>= ce_namelen(p
))
812 if (memcmp(name
, p
->name
, len
))
814 if (ce_stage(p
) != stage
)
816 if (p
->name
[len
] != '/')
818 if (p
->ce_flags
& CE_REMOVE
)
823 remove_index_entry_at(istate
, --pos
);
829 * Do we have another file with a pathname that is a proper
830 * subset of the name we're trying to add?
832 static int has_dir_name(struct index_state
*istate
,
833 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
836 int stage
= ce_stage(ce
);
837 const char *name
= ce
->name
;
838 const char *slash
= name
+ ce_namelen(ce
);
846 if (slash
<= ce
->name
)
851 pos
= index_name_stage_pos(istate
, name
, len
, stage
);
854 * Found one, but not so fast. This could
855 * be a marker that says "I was here, but
856 * I am being removed". Such an entry is
857 * not a part of the resulting tree, and
858 * it is Ok to have a directory at the same
861 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
865 remove_index_entry_at(istate
, pos
);
873 * Trivial optimization: if we find an entry that
874 * already matches the sub-directory, then we know
875 * we're ok, and we can exit.
877 while (pos
< istate
->cache_nr
) {
878 struct cache_entry
*p
= istate
->cache
[pos
];
879 if ((ce_namelen(p
) <= len
) ||
880 (p
->name
[len
] != '/') ||
881 memcmp(p
->name
, name
, len
))
882 break; /* not our subdirectory */
883 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
885 * p is at the same stage as our entry, and
886 * is a subdirectory of what we are looking
887 * at, so we cannot have conflicts at our
888 * level or anything shorter.
897 /* We may be in a situation where we already have path/file and path
898 * is being added, or we already have path and path/file is being
899 * added. Either one would result in a nonsense tree that has path
900 * twice when git-write-tree tries to write it out. Prevent it.
902 * If ok-to-replace is specified, we remove the conflicting entries
903 * from the cache so the caller should recompute the insert position.
904 * When this happens, we return non-zero.
906 static int check_file_directory_conflict(struct index_state
*istate
,
907 const struct cache_entry
*ce
,
908 int pos
, int ok_to_replace
)
913 * When ce is an "I am going away" entry, we allow it to be added
915 if (ce
->ce_flags
& CE_REMOVE
)
919 * We check if the path is a sub-path of a subsequent pathname
920 * first, since removing those will not change the position
923 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
926 * Then check if the path might have a clashing sub-directory
929 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
932 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
935 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
936 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
937 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
938 int new_only
= option
& ADD_CACHE_NEW_ONLY
;
940 cache_tree_invalidate_path(istate
->cache_tree
, ce
->name
);
941 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
943 /* existing match? Just replace it. */
946 replace_index_entry(istate
, pos
, ce
);
952 * Inserting a merged entry ("stage 0") into the index
953 * will always replace all non-merged entries..
955 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
956 while (ce_same_name(istate
->cache
[pos
], ce
)) {
958 if (!remove_index_entry_at(istate
, pos
))
965 if (!verify_path(ce
->name
))
966 return error("Invalid path '%s'", ce
->name
);
968 if (!skip_df_check
&&
969 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
971 return error("'%s' appears as both a file and as a directory",
973 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
979 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
983 if (option
& ADD_CACHE_JUST_APPEND
)
984 pos
= istate
->cache_nr
;
987 ret
= add_index_entry_with_check(istate
, ce
, option
);
993 /* Make sure the array is big enough .. */
994 if (istate
->cache_nr
== istate
->cache_alloc
) {
995 istate
->cache_alloc
= alloc_nr(istate
->cache_alloc
);
996 istate
->cache
= xrealloc(istate
->cache
,
997 istate
->cache_alloc
* sizeof(*istate
->cache
));
1002 if (istate
->cache_nr
> pos
+ 1)
1003 memmove(istate
->cache
+ pos
+ 1,
1004 istate
->cache
+ pos
,
1005 (istate
->cache_nr
- pos
- 1) * sizeof(ce
));
1006 set_index_entry(istate
, pos
, ce
);
1007 istate
->cache_changed
= 1;
1012 * "refresh" does not calculate a new sha1 file or bring the
1013 * cache up-to-date for mode/content changes. But what it
1014 * _does_ do is to "re-match" the stat information of a file
1015 * with the cache, so that you can refresh the cache for a
1016 * file that hasn't been changed but where the stat entry is
1019 * For example, you'd want to do this after doing a "git-read-tree",
1020 * to link up the stat cache details with the proper files.
1022 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
1023 struct cache_entry
*ce
,
1024 unsigned int options
, int *err
,
1028 struct cache_entry
*updated
;
1030 int refresh
= options
& CE_MATCH_REFRESH
;
1031 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
1032 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
1033 int ignore_missing
= options
& CE_MATCH_IGNORE_MISSING
;
1035 if (!refresh
|| ce_uptodate(ce
))
1039 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1040 * that the change to the work tree does not matter and told
1043 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
)) {
1044 ce_mark_uptodate(ce
);
1047 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
)) {
1048 ce_mark_uptodate(ce
);
1052 if (lstat(ce
->name
, &st
) < 0) {
1053 if (ignore_missing
&& errno
== ENOENT
)
1060 changed
= ie_match_stat(istate
, ce
, &st
, options
);
1062 *changed_ret
= changed
;
1065 * The path is unchanged. If we were told to ignore
1066 * valid bit, then we did the actual stat check and
1067 * found that the entry is unmodified. If the entry
1068 * is not marked VALID, this is the place to mark it
1069 * valid again, under "assume unchanged" mode.
1071 if (ignore_valid
&& assume_unchanged
&&
1072 !(ce
->ce_flags
& CE_VALID
))
1073 ; /* mark this one VALID again */
1076 * We do not mark the index itself "modified"
1077 * because CE_UPTODATE flag is in-core only;
1078 * we are not going to write this change out.
1080 if (!S_ISGITLINK(ce
->ce_mode
))
1081 ce_mark_uptodate(ce
);
1086 if (ie_modified(istate
, ce
, &st
, options
)) {
1093 updated
= xmalloc(size
);
1094 memcpy(updated
, ce
, size
);
1095 fill_stat_cache_info(updated
, &st
);
1097 * If ignore_valid is not set, we should leave CE_VALID bit
1098 * alone. Otherwise, paths marked with --no-assume-unchanged
1099 * (i.e. things to be edited) will reacquire CE_VALID bit
1100 * automatically, which is not really what we want.
1102 if (!ignore_valid
&& assume_unchanged
&&
1103 !(ce
->ce_flags
& CE_VALID
))
1104 updated
->ce_flags
&= ~CE_VALID
;
1109 static void show_file(const char * fmt
, const char * name
, int in_porcelain
,
1110 int * first
, const char *header_msg
)
1112 if (in_porcelain
&& *first
&& header_msg
) {
1113 printf("%s\n", header_msg
);
1119 int refresh_index(struct index_state
*istate
, unsigned int flags
, const char **pathspec
,
1120 char *seen
, const char *header_msg
)
1124 int really
= (flags
& REFRESH_REALLY
) != 0;
1125 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
1126 int quiet
= (flags
& REFRESH_QUIET
) != 0;
1127 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
1128 int ignore_submodules
= (flags
& REFRESH_IGNORE_SUBMODULES
) != 0;
1130 int in_porcelain
= (flags
& REFRESH_IN_PORCELAIN
);
1131 unsigned int options
= (CE_MATCH_REFRESH
|
1132 (really
? CE_MATCH_IGNORE_VALID
: 0) |
1133 (not_new
? CE_MATCH_IGNORE_MISSING
: 0));
1134 const char *modified_fmt
;
1135 const char *deleted_fmt
;
1136 const char *typechange_fmt
;
1137 const char *added_fmt
;
1138 const char *unmerged_fmt
;
1140 modified_fmt
= (in_porcelain
? "M\t%s\n" : "%s: needs update\n");
1141 deleted_fmt
= (in_porcelain
? "D\t%s\n" : "%s: needs update\n");
1142 typechange_fmt
= (in_porcelain
? "T\t%s\n" : "%s needs update\n");
1143 added_fmt
= (in_porcelain
? "A\t%s\n" : "%s needs update\n");
1144 unmerged_fmt
= (in_porcelain
? "U\t%s\n" : "%s: needs merge\n");
1145 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1146 struct cache_entry
*ce
, *new;
1147 int cache_errno
= 0;
1151 ce
= istate
->cache
[i
];
1152 if (ignore_submodules
&& S_ISGITLINK(ce
->ce_mode
))
1156 !match_pathspec(pathspec
, ce
->name
, ce_namelen(ce
), 0, seen
))
1160 while ((i
< istate
->cache_nr
) &&
1161 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
1167 show_file(unmerged_fmt
, ce
->name
, in_porcelain
,
1168 &first
, header_msg
);
1176 new = refresh_cache_ent(istate
, ce
, options
, &cache_errno
, &changed
);
1182 if (really
&& cache_errno
== EINVAL
) {
1183 /* If we are doing --really-refresh that
1184 * means the index is not valid anymore.
1186 ce
->ce_flags
&= ~CE_VALID
;
1187 istate
->cache_changed
= 1;
1192 if (cache_errno
== ENOENT
)
1194 else if (ce
->ce_flags
& CE_INTENT_TO_ADD
)
1195 fmt
= added_fmt
; /* must be before other checks */
1196 else if (changed
& TYPE_CHANGED
)
1197 fmt
= typechange_fmt
;
1201 ce
->name
, in_porcelain
, &first
, header_msg
);
1206 replace_index_entry(istate
, i
, new);
1211 static struct cache_entry
*refresh_cache_entry(struct cache_entry
*ce
,
1212 unsigned int options
)
1214 return refresh_cache_ent(&the_index
, ce
, options
, NULL
, NULL
);
1218 /*****************************************************************
1220 *****************************************************************/
1222 #define INDEX_FORMAT_DEFAULT 3
1225 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1226 * Again - this is just a (very strong in practice) heuristic that
1227 * the inode hasn't changed.
1229 * We save the fields in big-endian order to allow using the
1230 * index file over NFS transparently.
1232 struct ondisk_cache_entry
{
1233 struct cache_time ctime
;
1234 struct cache_time mtime
;
1241 unsigned char sha1
[20];
1242 unsigned short flags
;
1243 char name
[FLEX_ARRAY
]; /* more */
1247 * This struct is used when CE_EXTENDED bit is 1
1248 * The struct must match ondisk_cache_entry exactly from
1251 struct ondisk_cache_entry_extended
{
1252 struct cache_time ctime
;
1253 struct cache_time mtime
;
1260 unsigned char sha1
[20];
1261 unsigned short flags
;
1262 unsigned short flags2
;
1263 char name
[FLEX_ARRAY
]; /* more */
1266 /* These are only used for v3 or lower */
1267 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1268 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1269 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1270 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1271 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1272 ondisk_cache_entry_size(ce_namelen(ce)))
1274 static int verify_hdr(struct cache_header
*hdr
, unsigned long size
)
1277 unsigned char sha1
[20];
1280 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
1281 return error("bad signature");
1282 hdr_version
= ntohl(hdr
->hdr_version
);
1283 if (hdr_version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< hdr_version
)
1284 return error("bad index version %d", hdr_version
);
1286 git_SHA1_Update(&c
, hdr
, size
- 20);
1287 git_SHA1_Final(sha1
, &c
);
1288 if (hashcmp(sha1
, (unsigned char *)hdr
+ size
- 20))
1289 return error("bad index file sha1 signature");
1293 static int read_index_extension(struct index_state
*istate
,
1294 const char *ext
, void *data
, unsigned long sz
)
1296 switch (CACHE_EXT(ext
)) {
1297 case CACHE_EXT_TREE
:
1298 istate
->cache_tree
= cache_tree_read(data
, sz
);
1300 case CACHE_EXT_RESOLVE_UNDO
:
1301 istate
->resolve_undo
= resolve_undo_read(data
, sz
);
1304 if (*ext
< 'A' || 'Z' < *ext
)
1305 return error("index uses %.4s extension, which we do not understand",
1307 fprintf(stderr
, "ignoring %.4s extension\n", ext
);
1313 int read_index(struct index_state
*istate
)
1315 return read_index_from(istate
, get_index_file());
1318 #ifndef NEEDS_ALIGNED_ACCESS
1319 #define ntoh_s(var) ntohs(var)
1320 #define ntoh_l(var) ntohl(var)
1322 static inline uint16_t ntoh_s_force_align(void *p
)
1325 memcpy(&x
, p
, sizeof(x
));
1328 static inline uint32_t ntoh_l_force_align(void *p
)
1331 memcpy(&x
, p
, sizeof(x
));
1334 #define ntoh_s(var) ntoh_s_force_align(&(var))
1335 #define ntoh_l(var) ntoh_l_force_align(&(var))
1338 static struct cache_entry
*cache_entry_from_ondisk(struct ondisk_cache_entry
*ondisk
,
1343 struct cache_entry
*ce
= xmalloc(cache_entry_size(len
));
1345 ce
->ce_stat_data
.sd_ctime
.sec
= ntoh_l(ondisk
->ctime
.sec
);
1346 ce
->ce_stat_data
.sd_mtime
.sec
= ntoh_l(ondisk
->mtime
.sec
);
1347 ce
->ce_stat_data
.sd_ctime
.nsec
= ntoh_l(ondisk
->ctime
.nsec
);
1348 ce
->ce_stat_data
.sd_mtime
.nsec
= ntoh_l(ondisk
->mtime
.nsec
);
1349 ce
->ce_stat_data
.sd_dev
= ntoh_l(ondisk
->dev
);
1350 ce
->ce_stat_data
.sd_ino
= ntoh_l(ondisk
->ino
);
1351 ce
->ce_mode
= ntoh_l(ondisk
->mode
);
1352 ce
->ce_stat_data
.sd_uid
= ntoh_l(ondisk
->uid
);
1353 ce
->ce_stat_data
.sd_gid
= ntoh_l(ondisk
->gid
);
1354 ce
->ce_stat_data
.sd_size
= ntoh_l(ondisk
->size
);
1355 ce
->ce_flags
= flags
& ~CE_NAMEMASK
;
1356 ce
->ce_namelen
= len
;
1357 hashcpy(ce
->sha1
, ondisk
->sha1
);
1358 memcpy(ce
->name
, name
, len
);
1359 ce
->name
[len
] = '\0';
1364 * Adjacent cache entries tend to share the leading paths, so it makes
1365 * sense to only store the differences in later entries. In the v4
1366 * on-disk format of the index, each on-disk cache entry stores the
1367 * number of bytes to be stripped from the end of the previous name,
1368 * and the bytes to append to the result, to come up with its name.
1370 static unsigned long expand_name_field(struct strbuf
*name
, const char *cp_
)
1372 const unsigned char *ep
, *cp
= (const unsigned char *)cp_
;
1373 size_t len
= decode_varint(&cp
);
1375 if (name
->len
< len
)
1376 die("malformed name field in the index");
1377 strbuf_remove(name
, name
->len
- len
, len
);
1378 for (ep
= cp
; *ep
; ep
++)
1379 ; /* find the end */
1380 strbuf_add(name
, cp
, ep
- cp
);
1381 return (const char *)ep
+ 1 - cp_
;
1384 static struct cache_entry
*create_from_disk(struct ondisk_cache_entry
*ondisk
,
1385 unsigned long *ent_size
,
1386 struct strbuf
*previous_name
)
1388 struct cache_entry
*ce
;
1393 /* On-disk flags are just 16 bits */
1394 flags
= ntoh_s(ondisk
->flags
);
1395 len
= flags
& CE_NAMEMASK
;
1397 if (flags
& CE_EXTENDED
) {
1398 struct ondisk_cache_entry_extended
*ondisk2
;
1400 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
1401 extended_flags
= ntoh_s(ondisk2
->flags2
) << 16;
1402 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1403 if (extended_flags
& ~CE_EXTENDED_FLAGS
)
1404 die("Unknown index entry format %08x", extended_flags
);
1405 flags
|= extended_flags
;
1406 name
= ondisk2
->name
;
1409 name
= ondisk
->name
;
1411 if (!previous_name
) {
1412 /* v3 and earlier */
1413 if (len
== CE_NAMEMASK
)
1415 ce
= cache_entry_from_ondisk(ondisk
, flags
, name
, len
);
1417 *ent_size
= ondisk_ce_size(ce
);
1419 unsigned long consumed
;
1420 consumed
= expand_name_field(previous_name
, name
);
1421 ce
= cache_entry_from_ondisk(ondisk
, flags
,
1423 previous_name
->len
);
1425 *ent_size
= (name
- ((char *)ondisk
)) + consumed
;
1430 /* remember to discard_cache() before reading a different cache! */
1431 int read_index_from(struct index_state
*istate
, const char *path
)
1435 unsigned long src_offset
;
1436 struct cache_header
*hdr
;
1439 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
1441 if (istate
->initialized
)
1442 return istate
->cache_nr
;
1444 istate
->timestamp
.sec
= 0;
1445 istate
->timestamp
.nsec
= 0;
1446 fd
= open(path
, O_RDONLY
);
1448 if (errno
== ENOENT
)
1450 die_errno("index file open failed");
1454 die_errno("cannot stat the open index");
1456 mmap_size
= xsize_t(st
.st_size
);
1457 if (mmap_size
< sizeof(struct cache_header
) + 20)
1458 die("index file smaller than expected");
1460 mmap
= xmmap(NULL
, mmap_size
, PROT_READ
| PROT_WRITE
, MAP_PRIVATE
, fd
, 0);
1461 if (mmap
== MAP_FAILED
)
1462 die_errno("unable to map index file");
1466 if (verify_hdr(hdr
, mmap_size
) < 0)
1469 istate
->version
= ntohl(hdr
->hdr_version
);
1470 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
1471 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
1472 istate
->cache
= xcalloc(istate
->cache_alloc
, sizeof(*istate
->cache
));
1473 istate
->initialized
= 1;
1475 if (istate
->version
== 4)
1476 previous_name
= &previous_name_buf
;
1478 previous_name
= NULL
;
1480 src_offset
= sizeof(*hdr
);
1481 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1482 struct ondisk_cache_entry
*disk_ce
;
1483 struct cache_entry
*ce
;
1484 unsigned long consumed
;
1486 disk_ce
= (struct ondisk_cache_entry
*)((char *)mmap
+ src_offset
);
1487 ce
= create_from_disk(disk_ce
, &consumed
, previous_name
);
1488 set_index_entry(istate
, i
, ce
);
1490 src_offset
+= consumed
;
1492 strbuf_release(&previous_name_buf
);
1493 istate
->timestamp
.sec
= st
.st_mtime
;
1494 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
1496 while (src_offset
<= mmap_size
- 20 - 8) {
1497 /* After an array of active_nr index entries,
1498 * there can be arbitrary number of extended
1499 * sections, each of which is prefixed with
1500 * extension name (4-byte) and section length
1501 * in 4-byte network byte order.
1504 memcpy(&extsize
, (char *)mmap
+ src_offset
+ 4, 4);
1505 extsize
= ntohl(extsize
);
1506 if (read_index_extension(istate
,
1507 (const char *) mmap
+ src_offset
,
1508 (char *) mmap
+ src_offset
+ 8,
1512 src_offset
+= extsize
;
1514 munmap(mmap
, mmap_size
);
1515 return istate
->cache_nr
;
1518 munmap(mmap
, mmap_size
);
1519 die("index file corrupt");
1522 int is_index_unborn(struct index_state
*istate
)
1524 return (!istate
->cache_nr
&& !istate
->timestamp
.sec
);
1527 int discard_index(struct index_state
*istate
)
1531 for (i
= 0; i
< istate
->cache_nr
; i
++)
1532 free(istate
->cache
[i
]);
1533 resolve_undo_clear_index(istate
);
1534 istate
->cache_nr
= 0;
1535 istate
->cache_changed
= 0;
1536 istate
->timestamp
.sec
= 0;
1537 istate
->timestamp
.nsec
= 0;
1538 free_name_hash(istate
);
1539 cache_tree_free(&(istate
->cache_tree
));
1540 istate
->initialized
= 0;
1541 free(istate
->cache
);
1542 istate
->cache
= NULL
;
1543 istate
->cache_alloc
= 0;
1547 int unmerged_index(const struct index_state
*istate
)
1550 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1551 if (ce_stage(istate
->cache
[i
]))
1557 #define WRITE_BUFFER_SIZE 8192
1558 static unsigned char write_buffer
[WRITE_BUFFER_SIZE
];
1559 static unsigned long write_buffer_len
;
1561 static int ce_write_flush(git_SHA_CTX
*context
, int fd
)
1563 unsigned int buffered
= write_buffer_len
;
1565 git_SHA1_Update(context
, write_buffer
, buffered
);
1566 if (write_in_full(fd
, write_buffer
, buffered
) != buffered
)
1568 write_buffer_len
= 0;
1573 static int ce_write(git_SHA_CTX
*context
, int fd
, void *data
, unsigned int len
)
1576 unsigned int buffered
= write_buffer_len
;
1577 unsigned int partial
= WRITE_BUFFER_SIZE
- buffered
;
1580 memcpy(write_buffer
+ buffered
, data
, partial
);
1581 buffered
+= partial
;
1582 if (buffered
== WRITE_BUFFER_SIZE
) {
1583 write_buffer_len
= buffered
;
1584 if (ce_write_flush(context
, fd
))
1588 write_buffer_len
= buffered
;
1590 data
= (char *) data
+ partial
;
1595 static int write_index_ext_header(git_SHA_CTX
*context
, int fd
,
1596 unsigned int ext
, unsigned int sz
)
1600 return ((ce_write(context
, fd
, &ext
, 4) < 0) ||
1601 (ce_write(context
, fd
, &sz
, 4) < 0)) ? -1 : 0;
1604 static int ce_flush(git_SHA_CTX
*context
, int fd
)
1606 unsigned int left
= write_buffer_len
;
1609 write_buffer_len
= 0;
1610 git_SHA1_Update(context
, write_buffer
, left
);
1613 /* Flush first if not enough space for SHA1 signature */
1614 if (left
+ 20 > WRITE_BUFFER_SIZE
) {
1615 if (write_in_full(fd
, write_buffer
, left
) != left
)
1620 /* Append the SHA1 signature at the end */
1621 git_SHA1_Final(write_buffer
+ left
, context
);
1623 return (write_in_full(fd
, write_buffer
, left
) != left
) ? -1 : 0;
1626 static void ce_smudge_racily_clean_entry(struct cache_entry
*ce
)
1629 * The only thing we care about in this function is to smudge the
1630 * falsely clean entry due to touch-update-touch race, so we leave
1631 * everything else as they are. We are called for entries whose
1632 * ce_stat_data.sd_mtime match the index file mtime.
1634 * Note that this actually does not do much for gitlinks, for
1635 * which ce_match_stat_basic() always goes to the actual
1636 * contents. The caller checks with is_racy_timestamp() which
1637 * always says "no" for gitlinks, so we are not called for them ;-)
1641 if (lstat(ce
->name
, &st
) < 0)
1643 if (ce_match_stat_basic(ce
, &st
))
1645 if (ce_modified_check_fs(ce
, &st
)) {
1646 /* This is "racily clean"; smudge it. Note that this
1647 * is a tricky code. At first glance, it may appear
1648 * that it can break with this sequence:
1650 * $ echo xyzzy >frotz
1651 * $ git-update-index --add frotz
1654 * $ echo filfre >nitfol
1655 * $ git-update-index --add nitfol
1657 * but it does not. When the second update-index runs,
1658 * it notices that the entry "frotz" has the same timestamp
1659 * as index, and if we were to smudge it by resetting its
1660 * size to zero here, then the object name recorded
1661 * in index is the 6-byte file but the cached stat information
1662 * becomes zero --- which would then match what we would
1663 * obtain from the filesystem next time we stat("frotz").
1665 * However, the second update-index, before calling
1666 * this function, notices that the cached size is 6
1667 * bytes and what is on the filesystem is an empty
1668 * file, and never calls us, so the cached size information
1669 * for "frotz" stays 6 which does not match the filesystem.
1671 ce
->ce_stat_data
.sd_size
= 0;
1675 /* Copy miscellaneous fields but not the name */
1676 static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry
*ondisk
,
1677 struct cache_entry
*ce
)
1681 ondisk
->ctime
.sec
= htonl(ce
->ce_stat_data
.sd_ctime
.sec
);
1682 ondisk
->mtime
.sec
= htonl(ce
->ce_stat_data
.sd_mtime
.sec
);
1683 ondisk
->ctime
.nsec
= htonl(ce
->ce_stat_data
.sd_ctime
.nsec
);
1684 ondisk
->mtime
.nsec
= htonl(ce
->ce_stat_data
.sd_mtime
.nsec
);
1685 ondisk
->dev
= htonl(ce
->ce_stat_data
.sd_dev
);
1686 ondisk
->ino
= htonl(ce
->ce_stat_data
.sd_ino
);
1687 ondisk
->mode
= htonl(ce
->ce_mode
);
1688 ondisk
->uid
= htonl(ce
->ce_stat_data
.sd_uid
);
1689 ondisk
->gid
= htonl(ce
->ce_stat_data
.sd_gid
);
1690 ondisk
->size
= htonl(ce
->ce_stat_data
.sd_size
);
1691 hashcpy(ondisk
->sha1
, ce
->sha1
);
1693 flags
= ce
->ce_flags
;
1694 flags
|= (ce_namelen(ce
) >= CE_NAMEMASK
? CE_NAMEMASK
: ce_namelen(ce
));
1695 ondisk
->flags
= htons(flags
);
1696 if (ce
->ce_flags
& CE_EXTENDED
) {
1697 struct ondisk_cache_entry_extended
*ondisk2
;
1698 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
1699 ondisk2
->flags2
= htons((ce
->ce_flags
& CE_EXTENDED_FLAGS
) >> 16);
1700 return ondisk2
->name
;
1703 return ondisk
->name
;
1707 static int ce_write_entry(git_SHA_CTX
*c
, int fd
, struct cache_entry
*ce
,
1708 struct strbuf
*previous_name
)
1711 struct ondisk_cache_entry
*ondisk
;
1715 if (!previous_name
) {
1716 size
= ondisk_ce_size(ce
);
1717 ondisk
= xcalloc(1, size
);
1718 name
= copy_cache_entry_to_ondisk(ondisk
, ce
);
1719 memcpy(name
, ce
->name
, ce_namelen(ce
));
1721 int common
, to_remove
, prefix_size
;
1722 unsigned char to_remove_vi
[16];
1724 (ce
->name
[common
] &&
1725 common
< previous_name
->len
&&
1726 ce
->name
[common
] == previous_name
->buf
[common
]);
1728 ; /* still matching */
1729 to_remove
= previous_name
->len
- common
;
1730 prefix_size
= encode_varint(to_remove
, to_remove_vi
);
1732 if (ce
->ce_flags
& CE_EXTENDED
)
1733 size
= offsetof(struct ondisk_cache_entry_extended
, name
);
1735 size
= offsetof(struct ondisk_cache_entry
, name
);
1736 size
+= prefix_size
+ (ce_namelen(ce
) - common
+ 1);
1738 ondisk
= xcalloc(1, size
);
1739 name
= copy_cache_entry_to_ondisk(ondisk
, ce
);
1740 memcpy(name
, to_remove_vi
, prefix_size
);
1741 memcpy(name
+ prefix_size
, ce
->name
+ common
, ce_namelen(ce
) - common
);
1743 strbuf_splice(previous_name
, common
, to_remove
,
1744 ce
->name
+ common
, ce_namelen(ce
) - common
);
1747 result
= ce_write(c
, fd
, ondisk
, size
);
1752 static int has_racy_timestamp(struct index_state
*istate
)
1754 int entries
= istate
->cache_nr
;
1757 for (i
= 0; i
< entries
; i
++) {
1758 struct cache_entry
*ce
= istate
->cache
[i
];
1759 if (is_racy_timestamp(istate
, ce
))
1766 * Opportunistically update the index but do not complain if we can't
1768 void update_index_if_able(struct index_state
*istate
, struct lock_file
*lockfile
)
1770 if ((istate
->cache_changed
|| has_racy_timestamp(istate
)) &&
1771 !write_index(istate
, lockfile
->fd
))
1772 commit_locked_index(lockfile
);
1774 rollback_lock_file(lockfile
);
1777 int write_index(struct index_state
*istate
, int newfd
)
1780 struct cache_header hdr
;
1781 int i
, err
, removed
, extended
, hdr_version
;
1782 struct cache_entry
**cache
= istate
->cache
;
1783 int entries
= istate
->cache_nr
;
1785 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
1787 for (i
= removed
= extended
= 0; i
< entries
; i
++) {
1788 if (cache
[i
]->ce_flags
& CE_REMOVE
)
1791 /* reduce extended entries if possible */
1792 cache
[i
]->ce_flags
&= ~CE_EXTENDED
;
1793 if (cache
[i
]->ce_flags
& CE_EXTENDED_FLAGS
) {
1795 cache
[i
]->ce_flags
|= CE_EXTENDED
;
1799 if (!istate
->version
)
1800 istate
->version
= INDEX_FORMAT_DEFAULT
;
1802 /* demote version 3 to version 2 when the latter suffices */
1803 if (istate
->version
== 3 || istate
->version
== 2)
1804 istate
->version
= extended
? 3 : 2;
1806 hdr_version
= istate
->version
;
1808 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
1809 hdr
.hdr_version
= htonl(hdr_version
);
1810 hdr
.hdr_entries
= htonl(entries
- removed
);
1813 if (ce_write(&c
, newfd
, &hdr
, sizeof(hdr
)) < 0)
1816 previous_name
= (hdr_version
== 4) ? &previous_name_buf
: NULL
;
1817 for (i
= 0; i
< entries
; i
++) {
1818 struct cache_entry
*ce
= cache
[i
];
1819 if (ce
->ce_flags
& CE_REMOVE
)
1821 if (!ce_uptodate(ce
) && is_racy_timestamp(istate
, ce
))
1822 ce_smudge_racily_clean_entry(ce
);
1823 if (is_null_sha1(ce
->sha1
))
1824 return error("cache entry has null sha1: %s", ce
->name
);
1825 if (ce_write_entry(&c
, newfd
, ce
, previous_name
) < 0)
1828 strbuf_release(&previous_name_buf
);
1830 /* Write extension data here */
1831 if (istate
->cache_tree
) {
1832 struct strbuf sb
= STRBUF_INIT
;
1834 cache_tree_write(&sb
, istate
->cache_tree
);
1835 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_TREE
, sb
.len
) < 0
1836 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
1837 strbuf_release(&sb
);
1841 if (istate
->resolve_undo
) {
1842 struct strbuf sb
= STRBUF_INIT
;
1844 resolve_undo_write(&sb
, istate
->resolve_undo
);
1845 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_RESOLVE_UNDO
,
1847 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
1848 strbuf_release(&sb
);
1853 if (ce_flush(&c
, newfd
) || fstat(newfd
, &st
))
1855 istate
->timestamp
.sec
= (unsigned int)st
.st_mtime
;
1856 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
1861 * Read the index file that is potentially unmerged into given
1862 * index_state, dropping any unmerged entries. Returns true if
1863 * the index is unmerged. Callers who want to refuse to work
1864 * from an unmerged state can call this and check its return value,
1865 * instead of calling read_cache().
1867 int read_index_unmerged(struct index_state
*istate
)
1873 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1874 struct cache_entry
*ce
= istate
->cache
[i
];
1875 struct cache_entry
*new_ce
;
1881 len
= ce_namelen(ce
);
1882 size
= cache_entry_size(len
);
1883 new_ce
= xcalloc(1, size
);
1884 memcpy(new_ce
->name
, ce
->name
, len
);
1885 new_ce
->ce_flags
= create_ce_flags(0) | CE_CONFLICTED
;
1886 new_ce
->ce_namelen
= len
;
1887 new_ce
->ce_mode
= ce
->ce_mode
;
1888 if (add_index_entry(istate
, new_ce
, 0))
1889 return error("%s: cannot drop to stage #0",
1891 i
= index_name_pos(istate
, new_ce
->name
, len
);
1897 * Returns 1 if the path is an "other" path with respect to
1898 * the index; that is, the path is not mentioned in the index at all,
1899 * either as a file, a directory with some files in the index,
1900 * or as an unmerged entry.
1902 * We helpfully remove a trailing "/" from directories so that
1903 * the output of read_directory can be used as-is.
1905 int index_name_is_other(const struct index_state
*istate
, const char *name
,
1909 if (namelen
&& name
[namelen
- 1] == '/')
1911 pos
= index_name_pos(istate
, name
, namelen
);
1913 return 0; /* exact match */
1915 if (pos
< istate
->cache_nr
) {
1916 struct cache_entry
*ce
= istate
->cache
[pos
];
1917 if (ce_namelen(ce
) == namelen
&&
1918 !memcmp(ce
->name
, name
, namelen
))
1919 return 0; /* Yup, this one exists unmerged */
1924 void *read_blob_data_from_index(struct index_state
*istate
, const char *path
, unsigned long *size
)
1928 enum object_type type
;
1932 pos
= index_name_pos(istate
, path
, len
);
1935 * We might be in the middle of a merge, in which
1936 * case we would read stage #2 (ours).
1940 (pos
< 0 && i
< istate
->cache_nr
&&
1941 !strcmp(istate
->cache
[i
]->name
, path
));
1943 if (ce_stage(istate
->cache
[i
]) == 2)
1948 data
= read_sha1_file(istate
->cache
[pos
]->sha1
, &type
, &sz
);
1949 if (!data
|| type
!= OBJ_BLOB
) {
1958 void stat_validity_clear(struct stat_validity
*sv
)
1964 int stat_validity_check(struct stat_validity
*sv
, const char *path
)
1968 if (stat(path
, &st
) < 0)
1969 return sv
->sd
== NULL
;
1972 return S_ISREG(st
.st_mode
) && !match_stat_data(sv
->sd
, &st
);
1975 void stat_validity_update(struct stat_validity
*sv
, int fd
)
1979 if (fstat(fd
, &st
) < 0 || !S_ISREG(st
.st_mode
))
1980 stat_validity_clear(sv
);
1983 sv
->sd
= xcalloc(1, sizeof(struct stat_data
));
1984 fill_stat_data(sv
->sd
, &st
);