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" */
43 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
44 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
45 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
48 struct index_state the_index
;
49 static const char *alternate_index_output
;
51 static void set_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
53 istate
->cache
[nr
] = ce
;
54 add_name_hash(istate
, ce
);
57 static void replace_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
59 struct cache_entry
*old
= istate
->cache
[nr
];
61 replace_index_entry_in_base(istate
, old
, ce
);
62 remove_name_hash(istate
, old
);
64 set_index_entry(istate
, nr
, ce
);
65 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
66 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
69 void rename_index_entry_at(struct index_state
*istate
, int nr
, const char *new_name
)
71 struct cache_entry
*old
= istate
->cache
[nr
], *new;
72 int namelen
= strlen(new_name
);
74 new = xmalloc(cache_entry_size(namelen
));
75 copy_cache_entry(new, old
);
76 new->ce_flags
&= ~CE_HASHED
;
77 new->ce_namelen
= namelen
;
79 memcpy(new->name
, new_name
, namelen
+ 1);
81 cache_tree_invalidate_path(istate
, old
->name
);
82 remove_index_entry_at(istate
, nr
);
83 add_index_entry(istate
, new, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
86 void fill_stat_data(struct stat_data
*sd
, struct stat
*st
)
88 sd
->sd_ctime
.sec
= (unsigned int)st
->st_ctime
;
89 sd
->sd_mtime
.sec
= (unsigned int)st
->st_mtime
;
90 sd
->sd_ctime
.nsec
= ST_CTIME_NSEC(*st
);
91 sd
->sd_mtime
.nsec
= ST_MTIME_NSEC(*st
);
92 sd
->sd_dev
= st
->st_dev
;
93 sd
->sd_ino
= st
->st_ino
;
94 sd
->sd_uid
= st
->st_uid
;
95 sd
->sd_gid
= st
->st_gid
;
96 sd
->sd_size
= st
->st_size
;
99 int match_stat_data(const struct stat_data
*sd
, struct stat
*st
)
103 if (sd
->sd_mtime
.sec
!= (unsigned int)st
->st_mtime
)
104 changed
|= MTIME_CHANGED
;
105 if (trust_ctime
&& check_stat
&&
106 sd
->sd_ctime
.sec
!= (unsigned int)st
->st_ctime
)
107 changed
|= CTIME_CHANGED
;
110 if (check_stat
&& sd
->sd_mtime
.nsec
!= ST_MTIME_NSEC(*st
))
111 changed
|= MTIME_CHANGED
;
112 if (trust_ctime
&& check_stat
&&
113 sd
->sd_ctime
.nsec
!= ST_CTIME_NSEC(*st
))
114 changed
|= CTIME_CHANGED
;
118 if (sd
->sd_uid
!= (unsigned int) st
->st_uid
||
119 sd
->sd_gid
!= (unsigned int) st
->st_gid
)
120 changed
|= OWNER_CHANGED
;
121 if (sd
->sd_ino
!= (unsigned int) st
->st_ino
)
122 changed
|= INODE_CHANGED
;
127 * st_dev breaks on network filesystems where different
128 * clients will have different views of what "device"
129 * the filesystem is on
131 if (check_stat
&& sd
->sd_dev
!= (unsigned int) st
->st_dev
)
132 changed
|= INODE_CHANGED
;
135 if (sd
->sd_size
!= (unsigned int) st
->st_size
)
136 changed
|= DATA_CHANGED
;
142 * This only updates the "non-critical" parts of the directory
143 * cache, ie the parts that aren't tracked by GIT, and only used
144 * to validate the cache.
146 void fill_stat_cache_info(struct cache_entry
*ce
, struct stat
*st
)
148 fill_stat_data(&ce
->ce_stat_data
, st
);
150 if (assume_unchanged
)
151 ce
->ce_flags
|= CE_VALID
;
153 if (S_ISREG(st
->st_mode
))
154 ce_mark_uptodate(ce
);
157 static int ce_compare_data(const struct cache_entry
*ce
, struct stat
*st
)
160 int fd
= open(ce
->name
, O_RDONLY
);
163 unsigned char sha1
[20];
164 if (!index_fd(sha1
, fd
, st
, OBJ_BLOB
, ce
->name
, 0))
165 match
= hashcmp(sha1
, ce
->sha1
);
166 /* index_fd() closed the file descriptor already */
171 static int ce_compare_link(const struct cache_entry
*ce
, size_t expected_size
)
176 enum object_type type
;
177 struct strbuf sb
= STRBUF_INIT
;
179 if (strbuf_readlink(&sb
, ce
->name
, expected_size
))
182 buffer
= read_sha1_file(ce
->sha1
, &type
, &size
);
185 match
= memcmp(buffer
, sb
.buf
, size
);
192 static int ce_compare_gitlink(const struct cache_entry
*ce
)
194 unsigned char sha1
[20];
197 * We don't actually require that the .git directory
198 * under GITLINK directory be a valid git directory. It
199 * might even be missing (in case nobody populated that
202 * If so, we consider it always to match.
204 if (resolve_gitlink_ref(ce
->name
, "HEAD", sha1
) < 0)
206 return hashcmp(sha1
, ce
->sha1
);
209 static int ce_modified_check_fs(const struct cache_entry
*ce
, struct stat
*st
)
211 switch (st
->st_mode
& S_IFMT
) {
213 if (ce_compare_data(ce
, st
))
217 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
221 if (S_ISGITLINK(ce
->ce_mode
))
222 return ce_compare_gitlink(ce
) ? DATA_CHANGED
: 0;
229 static int ce_match_stat_basic(const struct cache_entry
*ce
, struct stat
*st
)
231 unsigned int changed
= 0;
233 if (ce
->ce_flags
& CE_REMOVE
)
234 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
236 switch (ce
->ce_mode
& S_IFMT
) {
238 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
239 /* We consider only the owner x bit to be relevant for
242 if (trust_executable_bit
&&
243 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
244 changed
|= MODE_CHANGED
;
247 if (!S_ISLNK(st
->st_mode
) &&
248 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
249 changed
|= TYPE_CHANGED
;
252 /* We ignore most of the st_xxx fields for gitlinks */
253 if (!S_ISDIR(st
->st_mode
))
254 changed
|= TYPE_CHANGED
;
255 else if (ce_compare_gitlink(ce
))
256 changed
|= DATA_CHANGED
;
259 die("internal error: ce_mode is %o", ce
->ce_mode
);
262 changed
|= match_stat_data(&ce
->ce_stat_data
, st
);
264 /* Racily smudged entry? */
265 if (!ce
->ce_stat_data
.sd_size
) {
266 if (!is_empty_blob_sha1(ce
->sha1
))
267 changed
|= DATA_CHANGED
;
273 static int is_racy_timestamp(const struct index_state
*istate
,
274 const struct cache_entry
*ce
)
276 return (!S_ISGITLINK(ce
->ce_mode
) &&
277 istate
->timestamp
.sec
&&
279 /* nanosecond timestamped files can also be racy! */
280 (istate
->timestamp
.sec
< ce
->ce_stat_data
.sd_mtime
.sec
||
281 (istate
->timestamp
.sec
== ce
->ce_stat_data
.sd_mtime
.sec
&&
282 istate
->timestamp
.nsec
<= ce
->ce_stat_data
.sd_mtime
.nsec
))
284 istate
->timestamp
.sec
<= ce
->ce_stat_data
.sd_mtime
.sec
289 int ie_match_stat(const struct index_state
*istate
,
290 const struct cache_entry
*ce
, struct stat
*st
,
291 unsigned int options
)
293 unsigned int changed
;
294 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
295 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
296 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
299 * If it's marked as always valid in the index, it's
300 * valid whatever the checked-out copy says.
302 * skip-worktree has the same effect with higher precedence
304 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
))
306 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
310 * Intent-to-add entries have not been added, so the index entry
311 * by definition never matches what is in the work tree until it
312 * actually gets added.
314 if (ce
->ce_flags
& CE_INTENT_TO_ADD
)
315 return DATA_CHANGED
| TYPE_CHANGED
| MODE_CHANGED
;
317 changed
= ce_match_stat_basic(ce
, st
);
320 * Within 1 second of this sequence:
321 * echo xyzzy >file && git-update-index --add file
322 * running this command:
324 * would give a falsely clean cache entry. The mtime and
325 * length match the cache, and other stat fields do not change.
327 * We could detect this at update-index time (the cache entry
328 * being registered/updated records the same time as "now")
329 * and delay the return from git-update-index, but that would
330 * effectively mean we can make at most one commit per second,
331 * which is not acceptable. Instead, we check cache entries
332 * whose mtime are the same as the index file timestamp more
333 * carefully than others.
335 if (!changed
&& is_racy_timestamp(istate
, ce
)) {
336 if (assume_racy_is_modified
)
337 changed
|= DATA_CHANGED
;
339 changed
|= ce_modified_check_fs(ce
, st
);
345 int ie_modified(const struct index_state
*istate
,
346 const struct cache_entry
*ce
,
347 struct stat
*st
, unsigned int options
)
349 int changed
, changed_fs
;
351 changed
= ie_match_stat(istate
, ce
, st
, options
);
355 * If the mode or type has changed, there's no point in trying
356 * to refresh the entry - it's not going to match
358 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
362 * Immediately after read-tree or update-index --cacheinfo,
363 * the length field is zero, as we have never even read the
364 * lstat(2) information once, and we cannot trust DATA_CHANGED
365 * returned by ie_match_stat() which in turn was returned by
366 * ce_match_stat_basic() to signal that the filesize of the
367 * blob changed. We have to actually go to the filesystem to
368 * see if the contents match, and if so, should answer "unchanged".
370 * The logic does not apply to gitlinks, as ce_match_stat_basic()
371 * already has checked the actual HEAD from the filesystem in the
372 * subproject. If ie_match_stat() already said it is different,
373 * then we know it is.
375 if ((changed
& DATA_CHANGED
) &&
376 (S_ISGITLINK(ce
->ce_mode
) || ce
->ce_stat_data
.sd_size
!= 0))
379 changed_fs
= ce_modified_check_fs(ce
, st
);
381 return changed
| changed_fs
;
385 int base_name_compare(const char *name1
, int len1
, int mode1
,
386 const char *name2
, int len2
, int mode2
)
388 unsigned char c1
, c2
;
389 int len
= len1
< len2
? len1
: len2
;
392 cmp
= memcmp(name1
, name2
, len
);
397 if (!c1
&& S_ISDIR(mode1
))
399 if (!c2
&& S_ISDIR(mode2
))
401 return (c1
< c2
) ? -1 : (c1
> c2
) ? 1 : 0;
405 * df_name_compare() is identical to base_name_compare(), except it
406 * compares conflicting directory/file entries as equal. Note that
407 * while a directory name compares as equal to a regular file, they
408 * then individually compare _differently_ to a filename that has
409 * a dot after the basename (because '\0' < '.' < '/').
411 * This is used by routines that want to traverse the git namespace
412 * but then handle conflicting entries together when possible.
414 int df_name_compare(const char *name1
, int len1
, int mode1
,
415 const char *name2
, int len2
, int mode2
)
417 int len
= len1
< len2
? len1
: len2
, cmp
;
418 unsigned char c1
, c2
;
420 cmp
= memcmp(name1
, name2
, len
);
423 /* Directories and files compare equal (same length, same name) */
427 if (!c1
&& S_ISDIR(mode1
))
430 if (!c2
&& S_ISDIR(mode2
))
432 if (c1
== '/' && !c2
)
434 if (c2
== '/' && !c1
)
439 int name_compare(const char *name1
, size_t len1
, const char *name2
, size_t len2
)
441 size_t min_len
= (len1
< len2
) ? len1
: len2
;
442 int cmp
= memcmp(name1
, name2
, min_len
);
452 int cache_name_stage_compare(const char *name1
, int len1
, int stage1
, const char *name2
, int len2
, int stage2
)
456 cmp
= name_compare(name1
, len1
, name2
, len2
);
467 static int index_name_stage_pos(const struct index_state
*istate
, const char *name
, int namelen
, int stage
)
472 last
= istate
->cache_nr
;
473 while (last
> first
) {
474 int next
= (last
+ first
) >> 1;
475 struct cache_entry
*ce
= istate
->cache
[next
];
476 int cmp
= cache_name_stage_compare(name
, namelen
, stage
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
488 int index_name_pos(const struct index_state
*istate
, const char *name
, int namelen
)
490 return index_name_stage_pos(istate
, name
, namelen
, 0);
493 /* Remove entry, return true if there are more entries to go.. */
494 int remove_index_entry_at(struct index_state
*istate
, int pos
)
496 struct cache_entry
*ce
= istate
->cache
[pos
];
498 record_resolve_undo(istate
, ce
);
499 remove_name_hash(istate
, ce
);
500 save_or_free_index_entry(istate
, ce
);
501 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
503 if (pos
>= istate
->cache_nr
)
505 memmove(istate
->cache
+ pos
,
506 istate
->cache
+ pos
+ 1,
507 (istate
->cache_nr
- pos
) * sizeof(struct cache_entry
*));
512 * Remove all cache entries marked for removal, that is where
513 * CE_REMOVE is set in ce_flags. This is much more effective than
514 * calling remove_index_entry_at() for each entry to be removed.
516 void remove_marked_cache_entries(struct index_state
*istate
)
518 struct cache_entry
**ce_array
= istate
->cache
;
521 for (i
= j
= 0; i
< istate
->cache_nr
; i
++) {
522 if (ce_array
[i
]->ce_flags
& CE_REMOVE
) {
523 remove_name_hash(istate
, ce_array
[i
]);
524 save_or_free_index_entry(istate
, ce_array
[i
]);
527 ce_array
[j
++] = ce_array
[i
];
529 if (j
== istate
->cache_nr
)
531 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
532 istate
->cache_nr
= j
;
535 int remove_file_from_index(struct index_state
*istate
, const char *path
)
537 int pos
= index_name_pos(istate
, path
, strlen(path
));
540 cache_tree_invalidate_path(istate
, path
);
541 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
542 remove_index_entry_at(istate
, pos
);
546 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
548 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
551 static int index_name_pos_also_unmerged(struct index_state
*istate
,
552 const char *path
, int namelen
)
554 int pos
= index_name_pos(istate
, path
, namelen
);
555 struct cache_entry
*ce
;
560 /* maybe unmerged? */
562 if (pos
>= istate
->cache_nr
||
563 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
566 /* order of preference: stage 2, 1, 3 */
567 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
568 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
569 !compare_name(ce
, path
, namelen
))
574 static int different_name(struct cache_entry
*ce
, struct cache_entry
*alias
)
576 int len
= ce_namelen(ce
);
577 return ce_namelen(alias
) != len
|| memcmp(ce
->name
, alias
->name
, len
);
581 * If we add a filename that aliases in the cache, we will use the
582 * name that we already have - but we don't want to update the same
583 * alias twice, because that implies that there were actually two
584 * different files with aliasing names!
586 * So we use the CE_ADDED flag to verify that the alias was an old
587 * one before we accept it as
589 static struct cache_entry
*create_alias_ce(struct index_state
*istate
,
590 struct cache_entry
*ce
,
591 struct cache_entry
*alias
)
594 struct cache_entry
*new;
596 if (alias
->ce_flags
& CE_ADDED
)
597 die("Will not add file alias '%s' ('%s' already exists in index)", ce
->name
, alias
->name
);
599 /* Ok, create the new entry using the name of the existing alias */
600 len
= ce_namelen(alias
);
601 new = xcalloc(1, cache_entry_size(len
));
602 memcpy(new->name
, alias
->name
, len
);
603 copy_cache_entry(new, ce
);
604 save_or_free_index_entry(istate
, ce
);
608 void set_object_name_for_intent_to_add_entry(struct cache_entry
*ce
)
610 unsigned char sha1
[20];
611 if (write_sha1_file("", 0, blob_type
, sha1
))
612 die("cannot create an empty blob in the object database");
613 hashcpy(ce
->sha1
, sha1
);
616 int add_to_index(struct index_state
*istate
, const char *path
, struct stat
*st
, int flags
)
618 int size
, namelen
, was_same
;
619 mode_t st_mode
= st
->st_mode
;
620 struct cache_entry
*ce
, *alias
;
621 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_IGNORE_SKIP_WORKTREE
|CE_MATCH_RACY_IS_DIRTY
;
622 int verbose
= flags
& (ADD_CACHE_VERBOSE
| ADD_CACHE_PRETEND
);
623 int pretend
= flags
& ADD_CACHE_PRETEND
;
624 int intent_only
= flags
& ADD_CACHE_INTENT
;
625 int add_option
= (ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
|
626 (intent_only
? ADD_CACHE_NEW_ONLY
: 0));
628 if (!S_ISREG(st_mode
) && !S_ISLNK(st_mode
) && !S_ISDIR(st_mode
))
629 return error("%s: can only add regular files, symbolic links or git-directories", path
);
631 namelen
= strlen(path
);
632 if (S_ISDIR(st_mode
)) {
633 while (namelen
&& path
[namelen
-1] == '/')
636 size
= cache_entry_size(namelen
);
637 ce
= xcalloc(1, size
);
638 memcpy(ce
->name
, path
, namelen
);
639 ce
->ce_namelen
= namelen
;
641 fill_stat_cache_info(ce
, st
);
643 ce
->ce_flags
|= CE_INTENT_TO_ADD
;
645 if (trust_executable_bit
&& has_symlinks
)
646 ce
->ce_mode
= create_ce_mode(st_mode
);
648 /* If there is an existing entry, pick the mode bits and type
649 * from it, otherwise assume unexecutable regular file.
651 struct cache_entry
*ent
;
652 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
654 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
655 ce
->ce_mode
= ce_mode_from_stat(ent
, st_mode
);
658 /* When core.ignorecase=true, determine if a directory of the same name but differing
659 * case already exists within the Git repository. If it does, ensure the directory
660 * case of the file being added to the repository matches (is folded into) the existing
661 * entry's directory case.
664 adjust_dirname_case(istate
, ce
->name
);
667 alias
= index_file_exists(istate
, ce
->name
, ce_namelen(ce
), ignore_case
);
668 if (alias
&& !ce_stage(alias
) && !ie_match_stat(istate
, alias
, st
, ce_option
)) {
669 /* Nothing changed, really */
671 if (!S_ISGITLINK(alias
->ce_mode
))
672 ce_mark_uptodate(alias
);
673 alias
->ce_flags
|= CE_ADDED
;
677 if (index_path(ce
->sha1
, path
, st
, HASH_WRITE_OBJECT
))
678 return error("unable to index file %s", path
);
680 set_object_name_for_intent_to_add_entry(ce
);
682 if (ignore_case
&& alias
&& different_name(ce
, alias
))
683 ce
= create_alias_ce(istate
, ce
, alias
);
684 ce
->ce_flags
|= CE_ADDED
;
686 /* It was suspected to be racily clean, but it turns out to be Ok */
689 !hashcmp(alias
->sha1
, ce
->sha1
) &&
690 ce
->ce_mode
== alias
->ce_mode
);
694 else if (add_index_entry(istate
, ce
, add_option
))
695 return error("unable to add %s to index",path
);
696 if (verbose
&& !was_same
)
697 printf("add '%s'\n", path
);
701 int add_file_to_index(struct index_state
*istate
, const char *path
, int flags
)
704 if (lstat(path
, &st
))
705 die_errno("unable to stat '%s'", path
);
706 return add_to_index(istate
, path
, &st
, flags
);
709 struct cache_entry
*make_cache_entry(unsigned int mode
,
710 const unsigned char *sha1
, const char *path
, int stage
,
711 unsigned int refresh_options
)
714 struct cache_entry
*ce
;
716 if (!verify_path(path
)) {
717 error("Invalid path '%s'", path
);
722 size
= cache_entry_size(len
);
723 ce
= xcalloc(1, size
);
725 hashcpy(ce
->sha1
, sha1
);
726 memcpy(ce
->name
, path
, len
);
727 ce
->ce_flags
= create_ce_flags(stage
);
728 ce
->ce_namelen
= len
;
729 ce
->ce_mode
= create_ce_mode(mode
);
731 return refresh_cache_entry(ce
, refresh_options
);
734 int ce_same_name(const struct cache_entry
*a
, const struct cache_entry
*b
)
736 int len
= ce_namelen(a
);
737 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
741 * We fundamentally don't like some paths: we don't want
742 * dot or dot-dot anywhere, and for obvious reasons don't
743 * want to recurse into ".git" either.
745 * Also, we don't want double slashes or slashes at the
746 * end that can make pathnames ambiguous.
748 static int verify_dotfile(const char *rest
)
751 * The first character was '.', but that
752 * has already been discarded, we now test
756 /* "." is not allowed */
757 if (*rest
== '\0' || is_dir_sep(*rest
))
762 * ".git" followed by NUL or slash is bad. This
763 * shares the path end test with the ".." case.
767 if (rest
[1] != 'i' && rest
[1] != 'I')
769 if (rest
[2] != 't' && rest
[2] != 'T')
774 if (rest
[1] == '\0' || is_dir_sep(rest
[1]))
780 int verify_path(const char *path
)
784 if (has_dos_drive_prefix(path
))
793 if (protect_hfs
&& is_hfs_dotgit(path
))
795 if (protect_ntfs
&& is_ntfs_dotgit(path
))
798 if ((c
== '.' && !verify_dotfile(path
)) ||
799 is_dir_sep(c
) || c
== '\0')
807 * Do we have another file that has the beginning components being a
808 * proper superset of the name we're trying to add?
810 static int has_file_name(struct index_state
*istate
,
811 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
814 int len
= ce_namelen(ce
);
815 int stage
= ce_stage(ce
);
816 const char *name
= ce
->name
;
818 while (pos
< istate
->cache_nr
) {
819 struct cache_entry
*p
= istate
->cache
[pos
++];
821 if (len
>= ce_namelen(p
))
823 if (memcmp(name
, p
->name
, len
))
825 if (ce_stage(p
) != stage
)
827 if (p
->name
[len
] != '/')
829 if (p
->ce_flags
& CE_REMOVE
)
834 remove_index_entry_at(istate
, --pos
);
840 * Do we have another file with a pathname that is a proper
841 * subset of the name we're trying to add?
843 static int has_dir_name(struct index_state
*istate
,
844 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
847 int stage
= ce_stage(ce
);
848 const char *name
= ce
->name
;
849 const char *slash
= name
+ ce_namelen(ce
);
857 if (slash
<= ce
->name
)
862 pos
= index_name_stage_pos(istate
, name
, len
, stage
);
865 * Found one, but not so fast. This could
866 * be a marker that says "I was here, but
867 * I am being removed". Such an entry is
868 * not a part of the resulting tree, and
869 * it is Ok to have a directory at the same
872 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
876 remove_index_entry_at(istate
, pos
);
884 * Trivial optimization: if we find an entry that
885 * already matches the sub-directory, then we know
886 * we're ok, and we can exit.
888 while (pos
< istate
->cache_nr
) {
889 struct cache_entry
*p
= istate
->cache
[pos
];
890 if ((ce_namelen(p
) <= len
) ||
891 (p
->name
[len
] != '/') ||
892 memcmp(p
->name
, name
, len
))
893 break; /* not our subdirectory */
894 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
896 * p is at the same stage as our entry, and
897 * is a subdirectory of what we are looking
898 * at, so we cannot have conflicts at our
899 * level or anything shorter.
908 /* We may be in a situation where we already have path/file and path
909 * is being added, or we already have path and path/file is being
910 * added. Either one would result in a nonsense tree that has path
911 * twice when git-write-tree tries to write it out. Prevent it.
913 * If ok-to-replace is specified, we remove the conflicting entries
914 * from the cache so the caller should recompute the insert position.
915 * When this happens, we return non-zero.
917 static int check_file_directory_conflict(struct index_state
*istate
,
918 const struct cache_entry
*ce
,
919 int pos
, int ok_to_replace
)
924 * When ce is an "I am going away" entry, we allow it to be added
926 if (ce
->ce_flags
& CE_REMOVE
)
930 * We check if the path is a sub-path of a subsequent pathname
931 * first, since removing those will not change the position
934 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
937 * Then check if the path might have a clashing sub-directory
940 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
943 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
946 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
947 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
948 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
949 int new_only
= option
& ADD_CACHE_NEW_ONLY
;
951 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
952 cache_tree_invalidate_path(istate
, ce
->name
);
953 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
955 /* existing match? Just replace it. */
958 replace_index_entry(istate
, pos
, ce
);
964 * Inserting a merged entry ("stage 0") into the index
965 * will always replace all non-merged entries..
967 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
968 while (ce_same_name(istate
->cache
[pos
], ce
)) {
970 if (!remove_index_entry_at(istate
, pos
))
977 if (!verify_path(ce
->name
))
978 return error("Invalid path '%s'", ce
->name
);
980 if (!skip_df_check
&&
981 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
983 return error("'%s' appears as both a file and as a directory",
985 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
991 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
995 if (option
& ADD_CACHE_JUST_APPEND
)
996 pos
= istate
->cache_nr
;
999 ret
= add_index_entry_with_check(istate
, ce
, option
);
1005 /* Make sure the array is big enough .. */
1006 ALLOC_GROW(istate
->cache
, istate
->cache_nr
+ 1, istate
->cache_alloc
);
1010 if (istate
->cache_nr
> pos
+ 1)
1011 memmove(istate
->cache
+ pos
+ 1,
1012 istate
->cache
+ pos
,
1013 (istate
->cache_nr
- pos
- 1) * sizeof(ce
));
1014 set_index_entry(istate
, pos
, ce
);
1015 istate
->cache_changed
|= CE_ENTRY_ADDED
;
1020 * "refresh" does not calculate a new sha1 file or bring the
1021 * cache up-to-date for mode/content changes. But what it
1022 * _does_ do is to "re-match" the stat information of a file
1023 * with the cache, so that you can refresh the cache for a
1024 * file that hasn't been changed but where the stat entry is
1027 * For example, you'd want to do this after doing a "git-read-tree",
1028 * to link up the stat cache details with the proper files.
1030 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
1031 struct cache_entry
*ce
,
1032 unsigned int options
, int *err
,
1036 struct cache_entry
*updated
;
1038 int refresh
= options
& CE_MATCH_REFRESH
;
1039 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
1040 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
1041 int ignore_missing
= options
& CE_MATCH_IGNORE_MISSING
;
1043 if (!refresh
|| ce_uptodate(ce
))
1047 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1048 * that the change to the work tree does not matter and told
1051 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
)) {
1052 ce_mark_uptodate(ce
);
1055 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
)) {
1056 ce_mark_uptodate(ce
);
1060 if (has_symlink_leading_path(ce
->name
, ce_namelen(ce
))) {
1068 if (lstat(ce
->name
, &st
) < 0) {
1069 if (ignore_missing
&& errno
== ENOENT
)
1076 changed
= ie_match_stat(istate
, ce
, &st
, options
);
1078 *changed_ret
= changed
;
1081 * The path is unchanged. If we were told to ignore
1082 * valid bit, then we did the actual stat check and
1083 * found that the entry is unmodified. If the entry
1084 * is not marked VALID, this is the place to mark it
1085 * valid again, under "assume unchanged" mode.
1087 if (ignore_valid
&& assume_unchanged
&&
1088 !(ce
->ce_flags
& CE_VALID
))
1089 ; /* mark this one VALID again */
1092 * We do not mark the index itself "modified"
1093 * because CE_UPTODATE flag is in-core only;
1094 * we are not going to write this change out.
1096 if (!S_ISGITLINK(ce
->ce_mode
))
1097 ce_mark_uptodate(ce
);
1102 if (ie_modified(istate
, ce
, &st
, options
)) {
1109 updated
= xmalloc(size
);
1110 memcpy(updated
, ce
, size
);
1111 fill_stat_cache_info(updated
, &st
);
1113 * If ignore_valid is not set, we should leave CE_VALID bit
1114 * alone. Otherwise, paths marked with --no-assume-unchanged
1115 * (i.e. things to be edited) will reacquire CE_VALID bit
1116 * automatically, which is not really what we want.
1118 if (!ignore_valid
&& assume_unchanged
&&
1119 !(ce
->ce_flags
& CE_VALID
))
1120 updated
->ce_flags
&= ~CE_VALID
;
1122 /* istate->cache_changed is updated in the caller */
1126 static void show_file(const char * fmt
, const char * name
, int in_porcelain
,
1127 int * first
, const char *header_msg
)
1129 if (in_porcelain
&& *first
&& header_msg
) {
1130 printf("%s\n", header_msg
);
1136 int refresh_index(struct index_state
*istate
, unsigned int flags
,
1137 const struct pathspec
*pathspec
,
1138 char *seen
, const char *header_msg
)
1142 int really
= (flags
& REFRESH_REALLY
) != 0;
1143 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
1144 int quiet
= (flags
& REFRESH_QUIET
) != 0;
1145 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
1146 int ignore_submodules
= (flags
& REFRESH_IGNORE_SUBMODULES
) != 0;
1148 int in_porcelain
= (flags
& REFRESH_IN_PORCELAIN
);
1149 unsigned int options
= (CE_MATCH_REFRESH
|
1150 (really
? CE_MATCH_IGNORE_VALID
: 0) |
1151 (not_new
? CE_MATCH_IGNORE_MISSING
: 0));
1152 const char *modified_fmt
;
1153 const char *deleted_fmt
;
1154 const char *typechange_fmt
;
1155 const char *added_fmt
;
1156 const char *unmerged_fmt
;
1158 modified_fmt
= (in_porcelain
? "M\t%s\n" : "%s: needs update\n");
1159 deleted_fmt
= (in_porcelain
? "D\t%s\n" : "%s: needs update\n");
1160 typechange_fmt
= (in_porcelain
? "T\t%s\n" : "%s needs update\n");
1161 added_fmt
= (in_porcelain
? "A\t%s\n" : "%s needs update\n");
1162 unmerged_fmt
= (in_porcelain
? "U\t%s\n" : "%s: needs merge\n");
1163 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1164 struct cache_entry
*ce
, *new;
1165 int cache_errno
= 0;
1169 ce
= istate
->cache
[i
];
1170 if (ignore_submodules
&& S_ISGITLINK(ce
->ce_mode
))
1173 if (pathspec
&& !ce_path_match(ce
, pathspec
, seen
))
1177 while ((i
< istate
->cache_nr
) &&
1178 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
1184 show_file(unmerged_fmt
, ce
->name
, in_porcelain
,
1185 &first
, header_msg
);
1193 new = refresh_cache_ent(istate
, ce
, options
, &cache_errno
, &changed
);
1199 if (really
&& cache_errno
== EINVAL
) {
1200 /* If we are doing --really-refresh that
1201 * means the index is not valid anymore.
1203 ce
->ce_flags
&= ~CE_VALID
;
1204 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
1205 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
1210 if (cache_errno
== ENOENT
)
1212 else if (ce
->ce_flags
& CE_INTENT_TO_ADD
)
1213 fmt
= added_fmt
; /* must be before other checks */
1214 else if (changed
& TYPE_CHANGED
)
1215 fmt
= typechange_fmt
;
1219 ce
->name
, in_porcelain
, &first
, header_msg
);
1224 replace_index_entry(istate
, i
, new);
1229 static struct cache_entry
*refresh_cache_entry(struct cache_entry
*ce
,
1230 unsigned int options
)
1232 return refresh_cache_ent(&the_index
, ce
, options
, NULL
, NULL
);
1236 /*****************************************************************
1238 *****************************************************************/
1240 #define INDEX_FORMAT_DEFAULT 3
1242 static unsigned int get_index_format_default(void)
1244 char *envversion
= getenv("GIT_INDEX_VERSION");
1247 unsigned int version
= INDEX_FORMAT_DEFAULT
;
1250 if (!git_config_get_int("index.version", &value
))
1252 if (version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1253 warning(_("index.version set, but the value is invalid.\n"
1254 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1255 return INDEX_FORMAT_DEFAULT
;
1260 version
= strtoul(envversion
, &endp
, 10);
1262 version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1263 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1264 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1265 version
= INDEX_FORMAT_DEFAULT
;
1271 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1272 * Again - this is just a (very strong in practice) heuristic that
1273 * the inode hasn't changed.
1275 * We save the fields in big-endian order to allow using the
1276 * index file over NFS transparently.
1278 struct ondisk_cache_entry
{
1279 struct cache_time ctime
;
1280 struct cache_time mtime
;
1287 unsigned char sha1
[20];
1289 char name
[FLEX_ARRAY
]; /* more */
1293 * This struct is used when CE_EXTENDED bit is 1
1294 * The struct must match ondisk_cache_entry exactly from
1297 struct ondisk_cache_entry_extended
{
1298 struct cache_time ctime
;
1299 struct cache_time mtime
;
1306 unsigned char sha1
[20];
1309 char name
[FLEX_ARRAY
]; /* more */
1312 /* These are only used for v3 or lower */
1313 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1314 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1315 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1316 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1317 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1318 ondisk_cache_entry_size(ce_namelen(ce)))
1320 static int verify_hdr(struct cache_header
*hdr
, unsigned long size
)
1323 unsigned char sha1
[20];
1326 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
1327 return error("bad signature");
1328 hdr_version
= ntohl(hdr
->hdr_version
);
1329 if (hdr_version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< hdr_version
)
1330 return error("bad index version %d", hdr_version
);
1332 git_SHA1_Update(&c
, hdr
, size
- 20);
1333 git_SHA1_Final(sha1
, &c
);
1334 if (hashcmp(sha1
, (unsigned char *)hdr
+ size
- 20))
1335 return error("bad index file sha1 signature");
1339 static int read_index_extension(struct index_state
*istate
,
1340 const char *ext
, void *data
, unsigned long sz
)
1342 switch (CACHE_EXT(ext
)) {
1343 case CACHE_EXT_TREE
:
1344 istate
->cache_tree
= cache_tree_read(data
, sz
);
1346 case CACHE_EXT_RESOLVE_UNDO
:
1347 istate
->resolve_undo
= resolve_undo_read(data
, sz
);
1349 case CACHE_EXT_LINK
:
1350 if (read_link_extension(istate
, data
, sz
))
1354 if (*ext
< 'A' || 'Z' < *ext
)
1355 return error("index uses %.4s extension, which we do not understand",
1357 fprintf(stderr
, "ignoring %.4s extension\n", ext
);
1363 int hold_locked_index(struct lock_file
*lk
, int die_on_error
)
1365 return hold_lock_file_for_update(lk
, get_index_file(),
1371 int read_index(struct index_state
*istate
)
1373 return read_index_from(istate
, get_index_file());
1376 static struct cache_entry
*cache_entry_from_ondisk(struct ondisk_cache_entry
*ondisk
,
1381 struct cache_entry
*ce
= xmalloc(cache_entry_size(len
));
1383 ce
->ce_stat_data
.sd_ctime
.sec
= get_be32(&ondisk
->ctime
.sec
);
1384 ce
->ce_stat_data
.sd_mtime
.sec
= get_be32(&ondisk
->mtime
.sec
);
1385 ce
->ce_stat_data
.sd_ctime
.nsec
= get_be32(&ondisk
->ctime
.nsec
);
1386 ce
->ce_stat_data
.sd_mtime
.nsec
= get_be32(&ondisk
->mtime
.nsec
);
1387 ce
->ce_stat_data
.sd_dev
= get_be32(&ondisk
->dev
);
1388 ce
->ce_stat_data
.sd_ino
= get_be32(&ondisk
->ino
);
1389 ce
->ce_mode
= get_be32(&ondisk
->mode
);
1390 ce
->ce_stat_data
.sd_uid
= get_be32(&ondisk
->uid
);
1391 ce
->ce_stat_data
.sd_gid
= get_be32(&ondisk
->gid
);
1392 ce
->ce_stat_data
.sd_size
= get_be32(&ondisk
->size
);
1393 ce
->ce_flags
= flags
& ~CE_NAMEMASK
;
1394 ce
->ce_namelen
= len
;
1396 hashcpy(ce
->sha1
, ondisk
->sha1
);
1397 memcpy(ce
->name
, name
, len
);
1398 ce
->name
[len
] = '\0';
1403 * Adjacent cache entries tend to share the leading paths, so it makes
1404 * sense to only store the differences in later entries. In the v4
1405 * on-disk format of the index, each on-disk cache entry stores the
1406 * number of bytes to be stripped from the end of the previous name,
1407 * and the bytes to append to the result, to come up with its name.
1409 static unsigned long expand_name_field(struct strbuf
*name
, const char *cp_
)
1411 const unsigned char *ep
, *cp
= (const unsigned char *)cp_
;
1412 size_t len
= decode_varint(&cp
);
1414 if (name
->len
< len
)
1415 die("malformed name field in the index");
1416 strbuf_remove(name
, name
->len
- len
, len
);
1417 for (ep
= cp
; *ep
; ep
++)
1418 ; /* find the end */
1419 strbuf_add(name
, cp
, ep
- cp
);
1420 return (const char *)ep
+ 1 - cp_
;
1423 static struct cache_entry
*create_from_disk(struct ondisk_cache_entry
*ondisk
,
1424 unsigned long *ent_size
,
1425 struct strbuf
*previous_name
)
1427 struct cache_entry
*ce
;
1432 /* On-disk flags are just 16 bits */
1433 flags
= get_be16(&ondisk
->flags
);
1434 len
= flags
& CE_NAMEMASK
;
1436 if (flags
& CE_EXTENDED
) {
1437 struct ondisk_cache_entry_extended
*ondisk2
;
1439 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
1440 extended_flags
= get_be16(&ondisk2
->flags2
) << 16;
1441 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1442 if (extended_flags
& ~CE_EXTENDED_FLAGS
)
1443 die("Unknown index entry format %08x", extended_flags
);
1444 flags
|= extended_flags
;
1445 name
= ondisk2
->name
;
1448 name
= ondisk
->name
;
1450 if (!previous_name
) {
1451 /* v3 and earlier */
1452 if (len
== CE_NAMEMASK
)
1454 ce
= cache_entry_from_ondisk(ondisk
, flags
, name
, len
);
1456 *ent_size
= ondisk_ce_size(ce
);
1458 unsigned long consumed
;
1459 consumed
= expand_name_field(previous_name
, name
);
1460 ce
= cache_entry_from_ondisk(ondisk
, flags
,
1462 previous_name
->len
);
1464 *ent_size
= (name
- ((char *)ondisk
)) + consumed
;
1469 static void check_ce_order(struct cache_entry
*ce
, struct cache_entry
*next_ce
)
1471 int name_compare
= strcmp(ce
->name
, next_ce
->name
);
1472 if (0 < name_compare
)
1473 die("unordered stage entries in index");
1474 if (!name_compare
) {
1476 die("multiple stage entries for merged file '%s'",
1478 if (ce_stage(ce
) > ce_stage(next_ce
))
1479 die("unordered stage entries for '%s'",
1484 /* remember to discard_cache() before reading a different cache! */
1485 int do_read_index(struct index_state
*istate
, const char *path
, int must_exist
)
1489 unsigned long src_offset
;
1490 struct cache_header
*hdr
;
1493 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
1495 if (istate
->initialized
)
1496 return istate
->cache_nr
;
1498 istate
->timestamp
.sec
= 0;
1499 istate
->timestamp
.nsec
= 0;
1500 fd
= open(path
, O_RDONLY
);
1502 if (!must_exist
&& errno
== ENOENT
)
1504 die_errno("%s: index file open failed", path
);
1508 die_errno("cannot stat the open index");
1510 mmap_size
= xsize_t(st
.st_size
);
1511 if (mmap_size
< sizeof(struct cache_header
) + 20)
1512 die("index file smaller than expected");
1514 mmap
= xmmap(NULL
, mmap_size
, PROT_READ
| PROT_WRITE
, MAP_PRIVATE
, fd
, 0);
1515 if (mmap
== MAP_FAILED
)
1516 die_errno("unable to map index file");
1520 if (verify_hdr(hdr
, mmap_size
) < 0)
1523 hashcpy(istate
->sha1
, (const unsigned char *)hdr
+ mmap_size
- 20);
1524 istate
->version
= ntohl(hdr
->hdr_version
);
1525 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
1526 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
1527 istate
->cache
= xcalloc(istate
->cache_alloc
, sizeof(*istate
->cache
));
1528 istate
->initialized
= 1;
1530 if (istate
->version
== 4)
1531 previous_name
= &previous_name_buf
;
1533 previous_name
= NULL
;
1535 src_offset
= sizeof(*hdr
);
1536 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1537 struct ondisk_cache_entry
*disk_ce
;
1538 struct cache_entry
*ce
;
1539 unsigned long consumed
;
1541 disk_ce
= (struct ondisk_cache_entry
*)((char *)mmap
+ src_offset
);
1542 ce
= create_from_disk(disk_ce
, &consumed
, previous_name
);
1543 set_index_entry(istate
, i
, ce
);
1546 check_ce_order(istate
->cache
[i
- 1], ce
);
1548 src_offset
+= consumed
;
1550 strbuf_release(&previous_name_buf
);
1551 istate
->timestamp
.sec
= st
.st_mtime
;
1552 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
1554 while (src_offset
<= mmap_size
- 20 - 8) {
1555 /* After an array of active_nr index entries,
1556 * there can be arbitrary number of extended
1557 * sections, each of which is prefixed with
1558 * extension name (4-byte) and section length
1559 * in 4-byte network byte order.
1562 memcpy(&extsize
, (char *)mmap
+ src_offset
+ 4, 4);
1563 extsize
= ntohl(extsize
);
1564 if (read_index_extension(istate
,
1565 (const char *) mmap
+ src_offset
,
1566 (char *) mmap
+ src_offset
+ 8,
1570 src_offset
+= extsize
;
1572 munmap(mmap
, mmap_size
);
1573 return istate
->cache_nr
;
1576 munmap(mmap
, mmap_size
);
1577 die("index file corrupt");
1580 int read_index_from(struct index_state
*istate
, const char *path
)
1582 struct split_index
*split_index
;
1585 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1586 if (istate
->initialized
)
1587 return istate
->cache_nr
;
1589 ret
= do_read_index(istate
, path
, 0);
1590 split_index
= istate
->split_index
;
1594 if (is_null_sha1(split_index
->base_sha1
))
1597 if (split_index
->base
)
1598 discard_index(split_index
->base
);
1600 split_index
->base
= xcalloc(1, sizeof(*split_index
->base
));
1601 ret
= do_read_index(split_index
->base
,
1602 git_path("sharedindex.%s",
1603 sha1_to_hex(split_index
->base_sha1
)), 1);
1604 if (hashcmp(split_index
->base_sha1
, split_index
->base
->sha1
))
1605 die("broken index, expect %s in %s, got %s",
1606 sha1_to_hex(split_index
->base_sha1
),
1607 git_path("sharedindex.%s",
1608 sha1_to_hex(split_index
->base_sha1
)),
1609 sha1_to_hex(split_index
->base
->sha1
));
1610 merge_base_index(istate
);
1614 int is_index_unborn(struct index_state
*istate
)
1616 return (!istate
->cache_nr
&& !istate
->timestamp
.sec
);
1619 int discard_index(struct index_state
*istate
)
1623 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1624 if (istate
->cache
[i
]->index
&&
1625 istate
->split_index
&&
1626 istate
->split_index
->base
&&
1627 istate
->cache
[i
]->index
<= istate
->split_index
->base
->cache_nr
&&
1628 istate
->cache
[i
] == istate
->split_index
->base
->cache
[istate
->cache
[i
]->index
- 1])
1630 free(istate
->cache
[i
]);
1632 resolve_undo_clear_index(istate
);
1633 istate
->cache_nr
= 0;
1634 istate
->cache_changed
= 0;
1635 istate
->timestamp
.sec
= 0;
1636 istate
->timestamp
.nsec
= 0;
1637 free_name_hash(istate
);
1638 cache_tree_free(&(istate
->cache_tree
));
1639 istate
->initialized
= 0;
1640 free(istate
->cache
);
1641 istate
->cache
= NULL
;
1642 istate
->cache_alloc
= 0;
1643 discard_split_index(istate
);
1647 int unmerged_index(const struct index_state
*istate
)
1650 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1651 if (ce_stage(istate
->cache
[i
]))
1657 #define WRITE_BUFFER_SIZE 8192
1658 static unsigned char write_buffer
[WRITE_BUFFER_SIZE
];
1659 static unsigned long write_buffer_len
;
1661 static int ce_write_flush(git_SHA_CTX
*context
, int fd
)
1663 unsigned int buffered
= write_buffer_len
;
1665 git_SHA1_Update(context
, write_buffer
, buffered
);
1666 if (write_in_full(fd
, write_buffer
, buffered
) != buffered
)
1668 write_buffer_len
= 0;
1673 static int ce_write(git_SHA_CTX
*context
, int fd
, void *data
, unsigned int len
)
1676 unsigned int buffered
= write_buffer_len
;
1677 unsigned int partial
= WRITE_BUFFER_SIZE
- buffered
;
1680 memcpy(write_buffer
+ buffered
, data
, partial
);
1681 buffered
+= partial
;
1682 if (buffered
== WRITE_BUFFER_SIZE
) {
1683 write_buffer_len
= buffered
;
1684 if (ce_write_flush(context
, fd
))
1688 write_buffer_len
= buffered
;
1690 data
= (char *) data
+ partial
;
1695 static int write_index_ext_header(git_SHA_CTX
*context
, int fd
,
1696 unsigned int ext
, unsigned int sz
)
1700 return ((ce_write(context
, fd
, &ext
, 4) < 0) ||
1701 (ce_write(context
, fd
, &sz
, 4) < 0)) ? -1 : 0;
1704 static int ce_flush(git_SHA_CTX
*context
, int fd
, unsigned char *sha1
)
1706 unsigned int left
= write_buffer_len
;
1709 write_buffer_len
= 0;
1710 git_SHA1_Update(context
, write_buffer
, left
);
1713 /* Flush first if not enough space for SHA1 signature */
1714 if (left
+ 20 > WRITE_BUFFER_SIZE
) {
1715 if (write_in_full(fd
, write_buffer
, left
) != left
)
1720 /* Append the SHA1 signature at the end */
1721 git_SHA1_Final(write_buffer
+ left
, context
);
1722 hashcpy(sha1
, write_buffer
+ left
);
1724 return (write_in_full(fd
, write_buffer
, left
) != left
) ? -1 : 0;
1727 static void ce_smudge_racily_clean_entry(struct cache_entry
*ce
)
1730 * The only thing we care about in this function is to smudge the
1731 * falsely clean entry due to touch-update-touch race, so we leave
1732 * everything else as they are. We are called for entries whose
1733 * ce_stat_data.sd_mtime match the index file mtime.
1735 * Note that this actually does not do much for gitlinks, for
1736 * which ce_match_stat_basic() always goes to the actual
1737 * contents. The caller checks with is_racy_timestamp() which
1738 * always says "no" for gitlinks, so we are not called for them ;-)
1742 if (lstat(ce
->name
, &st
) < 0)
1744 if (ce_match_stat_basic(ce
, &st
))
1746 if (ce_modified_check_fs(ce
, &st
)) {
1747 /* This is "racily clean"; smudge it. Note that this
1748 * is a tricky code. At first glance, it may appear
1749 * that it can break with this sequence:
1751 * $ echo xyzzy >frotz
1752 * $ git-update-index --add frotz
1755 * $ echo filfre >nitfol
1756 * $ git-update-index --add nitfol
1758 * but it does not. When the second update-index runs,
1759 * it notices that the entry "frotz" has the same timestamp
1760 * as index, and if we were to smudge it by resetting its
1761 * size to zero here, then the object name recorded
1762 * in index is the 6-byte file but the cached stat information
1763 * becomes zero --- which would then match what we would
1764 * obtain from the filesystem next time we stat("frotz").
1766 * However, the second update-index, before calling
1767 * this function, notices that the cached size is 6
1768 * bytes and what is on the filesystem is an empty
1769 * file, and never calls us, so the cached size information
1770 * for "frotz" stays 6 which does not match the filesystem.
1772 ce
->ce_stat_data
.sd_size
= 0;
1776 /* Copy miscellaneous fields but not the name */
1777 static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry
*ondisk
,
1778 struct cache_entry
*ce
)
1782 ondisk
->ctime
.sec
= htonl(ce
->ce_stat_data
.sd_ctime
.sec
);
1783 ondisk
->mtime
.sec
= htonl(ce
->ce_stat_data
.sd_mtime
.sec
);
1784 ondisk
->ctime
.nsec
= htonl(ce
->ce_stat_data
.sd_ctime
.nsec
);
1785 ondisk
->mtime
.nsec
= htonl(ce
->ce_stat_data
.sd_mtime
.nsec
);
1786 ondisk
->dev
= htonl(ce
->ce_stat_data
.sd_dev
);
1787 ondisk
->ino
= htonl(ce
->ce_stat_data
.sd_ino
);
1788 ondisk
->mode
= htonl(ce
->ce_mode
);
1789 ondisk
->uid
= htonl(ce
->ce_stat_data
.sd_uid
);
1790 ondisk
->gid
= htonl(ce
->ce_stat_data
.sd_gid
);
1791 ondisk
->size
= htonl(ce
->ce_stat_data
.sd_size
);
1792 hashcpy(ondisk
->sha1
, ce
->sha1
);
1794 flags
= ce
->ce_flags
& ~CE_NAMEMASK
;
1795 flags
|= (ce_namelen(ce
) >= CE_NAMEMASK
? CE_NAMEMASK
: ce_namelen(ce
));
1796 ondisk
->flags
= htons(flags
);
1797 if (ce
->ce_flags
& CE_EXTENDED
) {
1798 struct ondisk_cache_entry_extended
*ondisk2
;
1799 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
1800 ondisk2
->flags2
= htons((ce
->ce_flags
& CE_EXTENDED_FLAGS
) >> 16);
1801 return ondisk2
->name
;
1804 return ondisk
->name
;
1808 static int ce_write_entry(git_SHA_CTX
*c
, int fd
, struct cache_entry
*ce
,
1809 struct strbuf
*previous_name
)
1812 struct ondisk_cache_entry
*ondisk
;
1813 int saved_namelen
= saved_namelen
; /* compiler workaround */
1817 if (ce
->ce_flags
& CE_STRIP_NAME
) {
1818 saved_namelen
= ce_namelen(ce
);
1822 if (!previous_name
) {
1823 size
= ondisk_ce_size(ce
);
1824 ondisk
= xcalloc(1, size
);
1825 name
= copy_cache_entry_to_ondisk(ondisk
, ce
);
1826 memcpy(name
, ce
->name
, ce_namelen(ce
));
1828 int common
, to_remove
, prefix_size
;
1829 unsigned char to_remove_vi
[16];
1831 (ce
->name
[common
] &&
1832 common
< previous_name
->len
&&
1833 ce
->name
[common
] == previous_name
->buf
[common
]);
1835 ; /* still matching */
1836 to_remove
= previous_name
->len
- common
;
1837 prefix_size
= encode_varint(to_remove
, to_remove_vi
);
1839 if (ce
->ce_flags
& CE_EXTENDED
)
1840 size
= offsetof(struct ondisk_cache_entry_extended
, name
);
1842 size
= offsetof(struct ondisk_cache_entry
, name
);
1843 size
+= prefix_size
+ (ce_namelen(ce
) - common
+ 1);
1845 ondisk
= xcalloc(1, size
);
1846 name
= copy_cache_entry_to_ondisk(ondisk
, ce
);
1847 memcpy(name
, to_remove_vi
, prefix_size
);
1848 memcpy(name
+ prefix_size
, ce
->name
+ common
, ce_namelen(ce
) - common
);
1850 strbuf_splice(previous_name
, common
, to_remove
,
1851 ce
->name
+ common
, ce_namelen(ce
) - common
);
1853 if (ce
->ce_flags
& CE_STRIP_NAME
) {
1854 ce
->ce_namelen
= saved_namelen
;
1855 ce
->ce_flags
&= ~CE_STRIP_NAME
;
1858 result
= ce_write(c
, fd
, ondisk
, size
);
1864 * This function verifies if index_state has the correct sha1 of the
1865 * index file. Don't die if we have any other failure, just return 0.
1867 static int verify_index_from(const struct index_state
*istate
, const char *path
)
1872 unsigned char sha1
[20];
1874 if (!istate
->initialized
)
1877 fd
= open(path
, O_RDONLY
);
1884 if (st
.st_size
< sizeof(struct cache_header
) + 20)
1887 n
= pread_in_full(fd
, sha1
, 20, st
.st_size
- 20);
1891 if (hashcmp(istate
->sha1
, sha1
))
1902 static int verify_index(const struct index_state
*istate
)
1904 return verify_index_from(istate
, get_index_file());
1907 static int has_racy_timestamp(struct index_state
*istate
)
1909 int entries
= istate
->cache_nr
;
1912 for (i
= 0; i
< entries
; i
++) {
1913 struct cache_entry
*ce
= istate
->cache
[i
];
1914 if (is_racy_timestamp(istate
, ce
))
1921 * Opportunistically update the index but do not complain if we can't
1923 void update_index_if_able(struct index_state
*istate
, struct lock_file
*lockfile
)
1925 if ((istate
->cache_changed
|| has_racy_timestamp(istate
)) &&
1926 verify_index(istate
) &&
1927 write_locked_index(istate
, lockfile
, COMMIT_LOCK
))
1928 rollback_lock_file(lockfile
);
1931 static int do_write_index(struct index_state
*istate
, int newfd
,
1932 int strip_extensions
)
1935 struct cache_header hdr
;
1936 int i
, err
, removed
, extended
, hdr_version
;
1937 struct cache_entry
**cache
= istate
->cache
;
1938 int entries
= istate
->cache_nr
;
1940 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
1942 for (i
= removed
= extended
= 0; i
< entries
; i
++) {
1943 if (cache
[i
]->ce_flags
& CE_REMOVE
)
1946 /* reduce extended entries if possible */
1947 cache
[i
]->ce_flags
&= ~CE_EXTENDED
;
1948 if (cache
[i
]->ce_flags
& CE_EXTENDED_FLAGS
) {
1950 cache
[i
]->ce_flags
|= CE_EXTENDED
;
1954 if (!istate
->version
) {
1955 istate
->version
= get_index_format_default();
1956 if (getenv("GIT_TEST_SPLIT_INDEX"))
1957 init_split_index(istate
);
1960 /* demote version 3 to version 2 when the latter suffices */
1961 if (istate
->version
== 3 || istate
->version
== 2)
1962 istate
->version
= extended
? 3 : 2;
1964 hdr_version
= istate
->version
;
1966 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
1967 hdr
.hdr_version
= htonl(hdr_version
);
1968 hdr
.hdr_entries
= htonl(entries
- removed
);
1971 if (ce_write(&c
, newfd
, &hdr
, sizeof(hdr
)) < 0)
1974 previous_name
= (hdr_version
== 4) ? &previous_name_buf
: NULL
;
1975 for (i
= 0; i
< entries
; i
++) {
1976 struct cache_entry
*ce
= cache
[i
];
1977 if (ce
->ce_flags
& CE_REMOVE
)
1979 if (!ce_uptodate(ce
) && is_racy_timestamp(istate
, ce
))
1980 ce_smudge_racily_clean_entry(ce
);
1981 if (is_null_sha1(ce
->sha1
)) {
1982 static const char msg
[] = "cache entry has null sha1: %s";
1983 static int allow
= -1;
1986 allow
= git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
1988 warning(msg
, ce
->name
);
1990 return error(msg
, ce
->name
);
1992 if (ce_write_entry(&c
, newfd
, ce
, previous_name
) < 0)
1995 strbuf_release(&previous_name_buf
);
1997 /* Write extension data here */
1998 if (!strip_extensions
&& istate
->split_index
) {
1999 struct strbuf sb
= STRBUF_INIT
;
2001 err
= write_link_extension(&sb
, istate
) < 0 ||
2002 write_index_ext_header(&c
, newfd
, CACHE_EXT_LINK
,
2004 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2005 strbuf_release(&sb
);
2009 if (!strip_extensions
&& istate
->cache_tree
) {
2010 struct strbuf sb
= STRBUF_INIT
;
2012 cache_tree_write(&sb
, istate
->cache_tree
);
2013 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_TREE
, sb
.len
) < 0
2014 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2015 strbuf_release(&sb
);
2019 if (!strip_extensions
&& istate
->resolve_undo
) {
2020 struct strbuf sb
= STRBUF_INIT
;
2022 resolve_undo_write(&sb
, istate
->resolve_undo
);
2023 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_RESOLVE_UNDO
,
2025 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2026 strbuf_release(&sb
);
2031 if (ce_flush(&c
, newfd
, istate
->sha1
) || fstat(newfd
, &st
))
2033 istate
->timestamp
.sec
= (unsigned int)st
.st_mtime
;
2034 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
2038 void set_alternate_index_output(const char *name
)
2040 alternate_index_output
= name
;
2043 static int commit_locked_index(struct lock_file
*lk
)
2045 if (alternate_index_output
)
2046 return commit_lock_file_to(lk
, alternate_index_output
);
2048 return commit_lock_file(lk
);
2051 static int do_write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
2054 int ret
= do_write_index(istate
, lock
->fd
, 0);
2057 assert((flags
& (COMMIT_LOCK
| CLOSE_LOCK
)) !=
2058 (COMMIT_LOCK
| CLOSE_LOCK
));
2059 if (flags
& COMMIT_LOCK
)
2060 return commit_locked_index(lock
);
2061 else if (flags
& CLOSE_LOCK
)
2062 return close_lock_file(lock
);
2067 static int write_split_index(struct index_state
*istate
,
2068 struct lock_file
*lock
,
2072 prepare_to_write_split_index(istate
);
2073 ret
= do_write_locked_index(istate
, lock
, flags
);
2074 finish_writing_split_index(istate
);
2078 static char *temporary_sharedindex
;
2080 static void remove_temporary_sharedindex(void)
2082 if (temporary_sharedindex
) {
2083 unlink_or_warn(temporary_sharedindex
);
2084 free(temporary_sharedindex
);
2085 temporary_sharedindex
= NULL
;
2089 static void remove_temporary_sharedindex_on_signal(int signo
)
2091 remove_temporary_sharedindex();
2092 sigchain_pop(signo
);
2096 static int write_shared_index(struct index_state
*istate
,
2097 struct lock_file
*lock
, unsigned flags
)
2099 struct split_index
*si
= istate
->split_index
;
2100 static int installed_handler
;
2103 temporary_sharedindex
= git_pathdup("sharedindex_XXXXXX");
2104 fd
= mkstemp(temporary_sharedindex
);
2106 free(temporary_sharedindex
);
2107 temporary_sharedindex
= NULL
;
2108 hashclr(si
->base_sha1
);
2109 return do_write_locked_index(istate
, lock
, flags
);
2111 if (!installed_handler
) {
2112 atexit(remove_temporary_sharedindex
);
2113 sigchain_push_common(remove_temporary_sharedindex_on_signal
);
2115 move_cache_to_base_index(istate
);
2116 ret
= do_write_index(si
->base
, fd
, 1);
2119 remove_temporary_sharedindex();
2122 ret
= rename(temporary_sharedindex
,
2123 git_path("sharedindex.%s", sha1_to_hex(si
->base
->sha1
)));
2124 free(temporary_sharedindex
);
2125 temporary_sharedindex
= NULL
;
2127 hashcpy(si
->base_sha1
, si
->base
->sha1
);
2131 int write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
2134 struct split_index
*si
= istate
->split_index
;
2136 if (!si
|| alternate_index_output
||
2137 (istate
->cache_changed
& ~EXTMASK
)) {
2139 hashclr(si
->base_sha1
);
2140 return do_write_locked_index(istate
, lock
, flags
);
2143 if (getenv("GIT_TEST_SPLIT_INDEX")) {
2144 int v
= si
->base_sha1
[0];
2146 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
2148 if (istate
->cache_changed
& SPLIT_INDEX_ORDERED
) {
2149 int ret
= write_shared_index(istate
, lock
, flags
);
2154 return write_split_index(istate
, lock
, flags
);
2158 * Read the index file that is potentially unmerged into given
2159 * index_state, dropping any unmerged entries. Returns true if
2160 * the index is unmerged. Callers who want to refuse to work
2161 * from an unmerged state can call this and check its return value,
2162 * instead of calling read_cache().
2164 int read_index_unmerged(struct index_state
*istate
)
2170 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2171 struct cache_entry
*ce
= istate
->cache
[i
];
2172 struct cache_entry
*new_ce
;
2178 len
= ce_namelen(ce
);
2179 size
= cache_entry_size(len
);
2180 new_ce
= xcalloc(1, size
);
2181 memcpy(new_ce
->name
, ce
->name
, len
);
2182 new_ce
->ce_flags
= create_ce_flags(0) | CE_CONFLICTED
;
2183 new_ce
->ce_namelen
= len
;
2184 new_ce
->ce_mode
= ce
->ce_mode
;
2185 if (add_index_entry(istate
, new_ce
, 0))
2186 return error("%s: cannot drop to stage #0",
2193 * Returns 1 if the path is an "other" path with respect to
2194 * the index; that is, the path is not mentioned in the index at all,
2195 * either as a file, a directory with some files in the index,
2196 * or as an unmerged entry.
2198 * We helpfully remove a trailing "/" from directories so that
2199 * the output of read_directory can be used as-is.
2201 int index_name_is_other(const struct index_state
*istate
, const char *name
,
2205 if (namelen
&& name
[namelen
- 1] == '/')
2207 pos
= index_name_pos(istate
, name
, namelen
);
2209 return 0; /* exact match */
2211 if (pos
< istate
->cache_nr
) {
2212 struct cache_entry
*ce
= istate
->cache
[pos
];
2213 if (ce_namelen(ce
) == namelen
&&
2214 !memcmp(ce
->name
, name
, namelen
))
2215 return 0; /* Yup, this one exists unmerged */
2220 void *read_blob_data_from_index(struct index_state
*istate
, const char *path
, unsigned long *size
)
2224 enum object_type type
;
2228 pos
= index_name_pos(istate
, path
, len
);
2231 * We might be in the middle of a merge, in which
2232 * case we would read stage #2 (ours).
2236 (pos
< 0 && i
< istate
->cache_nr
&&
2237 !strcmp(istate
->cache
[i
]->name
, path
));
2239 if (ce_stage(istate
->cache
[i
]) == 2)
2244 data
= read_sha1_file(istate
->cache
[pos
]->sha1
, &type
, &sz
);
2245 if (!data
|| type
!= OBJ_BLOB
) {
2254 void stat_validity_clear(struct stat_validity
*sv
)
2260 int stat_validity_check(struct stat_validity
*sv
, const char *path
)
2264 if (stat(path
, &st
) < 0)
2265 return sv
->sd
== NULL
;
2268 return S_ISREG(st
.st_mode
) && !match_stat_data(sv
->sd
, &st
);
2271 void stat_validity_update(struct stat_validity
*sv
, int fd
)
2275 if (fstat(fd
, &st
) < 0 || !S_ISREG(st
.st_mode
))
2276 stat_validity_clear(sv
);
2279 sv
->sd
= xcalloc(1, sizeof(struct stat_data
));
2280 fill_stat_data(sv
->sd
, &st
);