2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 #define NO_THE_INDEX_COMPATIBILITY_MACROS
10 #include "cache-tree.h"
16 #include "resolve-undo.h"
19 #include "split-index.h"
23 static struct cache_entry
*refresh_cache_entry(struct cache_entry
*ce
,
24 unsigned int options
);
26 /* Mask for the name length in ce_flags in the on-disk index */
28 #define CE_NAMEMASK (0x0fff)
32 * The first letter should be 'A'..'Z' for extensions that are not
33 * necessary for a correct operation (i.e. optimization data).
34 * When new extensions are added that _needs_ to be understood in
35 * order to correctly interpret the index file, pick character that
36 * is outside the range, to cause the reader to abort.
39 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
40 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
41 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
42 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
43 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
45 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
46 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
47 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
48 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED)
50 struct index_state the_index
;
51 static const char *alternate_index_output
;
53 static void set_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
55 istate
->cache
[nr
] = ce
;
56 add_name_hash(istate
, ce
);
59 static void replace_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
61 struct cache_entry
*old
= istate
->cache
[nr
];
63 replace_index_entry_in_base(istate
, old
, ce
);
64 remove_name_hash(istate
, old
);
66 set_index_entry(istate
, nr
, ce
);
67 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
68 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
71 void rename_index_entry_at(struct index_state
*istate
, int nr
, const char *new_name
)
73 struct cache_entry
*old
= istate
->cache
[nr
], *new;
74 int namelen
= strlen(new_name
);
76 new = xmalloc(cache_entry_size(namelen
));
77 copy_cache_entry(new, old
);
78 new->ce_flags
&= ~CE_HASHED
;
79 new->ce_namelen
= namelen
;
81 memcpy(new->name
, new_name
, namelen
+ 1);
83 cache_tree_invalidate_path(istate
, old
->name
);
84 untracked_cache_remove_from_index(istate
, old
->name
);
85 remove_index_entry_at(istate
, nr
);
86 add_index_entry(istate
, new, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
89 void fill_stat_data(struct stat_data
*sd
, struct stat
*st
)
91 sd
->sd_ctime
.sec
= (unsigned int)st
->st_ctime
;
92 sd
->sd_mtime
.sec
= (unsigned int)st
->st_mtime
;
93 sd
->sd_ctime
.nsec
= ST_CTIME_NSEC(*st
);
94 sd
->sd_mtime
.nsec
= ST_MTIME_NSEC(*st
);
95 sd
->sd_dev
= st
->st_dev
;
96 sd
->sd_ino
= st
->st_ino
;
97 sd
->sd_uid
= st
->st_uid
;
98 sd
->sd_gid
= st
->st_gid
;
99 sd
->sd_size
= st
->st_size
;
102 int match_stat_data(const struct stat_data
*sd
, struct stat
*st
)
106 if (sd
->sd_mtime
.sec
!= (unsigned int)st
->st_mtime
)
107 changed
|= MTIME_CHANGED
;
108 if (trust_ctime
&& check_stat
&&
109 sd
->sd_ctime
.sec
!= (unsigned int)st
->st_ctime
)
110 changed
|= CTIME_CHANGED
;
113 if (check_stat
&& sd
->sd_mtime
.nsec
!= ST_MTIME_NSEC(*st
))
114 changed
|= MTIME_CHANGED
;
115 if (trust_ctime
&& check_stat
&&
116 sd
->sd_ctime
.nsec
!= ST_CTIME_NSEC(*st
))
117 changed
|= CTIME_CHANGED
;
121 if (sd
->sd_uid
!= (unsigned int) st
->st_uid
||
122 sd
->sd_gid
!= (unsigned int) st
->st_gid
)
123 changed
|= OWNER_CHANGED
;
124 if (sd
->sd_ino
!= (unsigned int) st
->st_ino
)
125 changed
|= INODE_CHANGED
;
130 * st_dev breaks on network filesystems where different
131 * clients will have different views of what "device"
132 * the filesystem is on
134 if (check_stat
&& sd
->sd_dev
!= (unsigned int) st
->st_dev
)
135 changed
|= INODE_CHANGED
;
138 if (sd
->sd_size
!= (unsigned int) st
->st_size
)
139 changed
|= DATA_CHANGED
;
145 * This only updates the "non-critical" parts of the directory
146 * cache, ie the parts that aren't tracked by GIT, and only used
147 * to validate the cache.
149 void fill_stat_cache_info(struct cache_entry
*ce
, struct stat
*st
)
151 fill_stat_data(&ce
->ce_stat_data
, st
);
153 if (assume_unchanged
)
154 ce
->ce_flags
|= CE_VALID
;
156 if (S_ISREG(st
->st_mode
))
157 ce_mark_uptodate(ce
);
160 static int ce_compare_data(const struct cache_entry
*ce
, struct stat
*st
)
163 int fd
= open(ce
->name
, O_RDONLY
);
166 unsigned char sha1
[20];
167 if (!index_fd(sha1
, fd
, st
, OBJ_BLOB
, ce
->name
, 0))
168 match
= hashcmp(sha1
, ce
->sha1
);
169 /* index_fd() closed the file descriptor already */
174 static int ce_compare_link(const struct cache_entry
*ce
, size_t expected_size
)
179 enum object_type type
;
180 struct strbuf sb
= STRBUF_INIT
;
182 if (strbuf_readlink(&sb
, ce
->name
, expected_size
))
185 buffer
= read_sha1_file(ce
->sha1
, &type
, &size
);
188 match
= memcmp(buffer
, sb
.buf
, size
);
195 static int ce_compare_gitlink(const struct cache_entry
*ce
)
197 unsigned char sha1
[20];
200 * We don't actually require that the .git directory
201 * under GITLINK directory be a valid git directory. It
202 * might even be missing (in case nobody populated that
205 * If so, we consider it always to match.
207 if (resolve_gitlink_ref(ce
->name
, "HEAD", sha1
) < 0)
209 return hashcmp(sha1
, ce
->sha1
);
212 static int ce_modified_check_fs(const struct cache_entry
*ce
, struct stat
*st
)
214 switch (st
->st_mode
& S_IFMT
) {
216 if (ce_compare_data(ce
, st
))
220 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
224 if (S_ISGITLINK(ce
->ce_mode
))
225 return ce_compare_gitlink(ce
) ? DATA_CHANGED
: 0;
232 static int ce_match_stat_basic(const struct cache_entry
*ce
, struct stat
*st
)
234 unsigned int changed
= 0;
236 if (ce
->ce_flags
& CE_REMOVE
)
237 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
239 switch (ce
->ce_mode
& S_IFMT
) {
241 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
242 /* We consider only the owner x bit to be relevant for
245 if (trust_executable_bit
&&
246 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
247 changed
|= MODE_CHANGED
;
250 if (!S_ISLNK(st
->st_mode
) &&
251 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
252 changed
|= TYPE_CHANGED
;
255 /* We ignore most of the st_xxx fields for gitlinks */
256 if (!S_ISDIR(st
->st_mode
))
257 changed
|= TYPE_CHANGED
;
258 else if (ce_compare_gitlink(ce
))
259 changed
|= DATA_CHANGED
;
262 die("internal error: ce_mode is %o", ce
->ce_mode
);
265 changed
|= match_stat_data(&ce
->ce_stat_data
, st
);
267 /* Racily smudged entry? */
268 if (!ce
->ce_stat_data
.sd_size
) {
269 if (!is_empty_blob_sha1(ce
->sha1
))
270 changed
|= DATA_CHANGED
;
276 static int is_racy_stat(const struct index_state
*istate
,
277 const struct stat_data
*sd
)
279 return (istate
->timestamp
.sec
&&
281 /* nanosecond timestamped files can also be racy! */
282 (istate
->timestamp
.sec
< sd
->sd_mtime
.sec
||
283 (istate
->timestamp
.sec
== sd
->sd_mtime
.sec
&&
284 istate
->timestamp
.nsec
<= sd
->sd_mtime
.nsec
))
286 istate
->timestamp
.sec
<= sd
->sd_mtime
.sec
291 static int is_racy_timestamp(const struct index_state
*istate
,
292 const struct cache_entry
*ce
)
294 return (!S_ISGITLINK(ce
->ce_mode
) &&
295 is_racy_stat(istate
, &ce
->ce_stat_data
));
298 int match_stat_data_racy(const struct index_state
*istate
,
299 const struct stat_data
*sd
, struct stat
*st
)
301 if (is_racy_stat(istate
, sd
))
302 return MTIME_CHANGED
;
303 return match_stat_data(sd
, st
);
306 int ie_match_stat(const struct index_state
*istate
,
307 const struct cache_entry
*ce
, struct stat
*st
,
308 unsigned int options
)
310 unsigned int changed
;
311 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
312 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
313 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
316 * If it's marked as always valid in the index, it's
317 * valid whatever the checked-out copy says.
319 * skip-worktree has the same effect with higher precedence
321 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
))
323 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
327 * Intent-to-add entries have not been added, so the index entry
328 * by definition never matches what is in the work tree until it
329 * actually gets added.
331 if (ce
->ce_flags
& CE_INTENT_TO_ADD
)
332 return DATA_CHANGED
| TYPE_CHANGED
| MODE_CHANGED
;
334 changed
= ce_match_stat_basic(ce
, st
);
337 * Within 1 second of this sequence:
338 * echo xyzzy >file && git-update-index --add file
339 * running this command:
341 * would give a falsely clean cache entry. The mtime and
342 * length match the cache, and other stat fields do not change.
344 * We could detect this at update-index time (the cache entry
345 * being registered/updated records the same time as "now")
346 * and delay the return from git-update-index, but that would
347 * effectively mean we can make at most one commit per second,
348 * which is not acceptable. Instead, we check cache entries
349 * whose mtime are the same as the index file timestamp more
350 * carefully than others.
352 if (!changed
&& is_racy_timestamp(istate
, ce
)) {
353 if (assume_racy_is_modified
)
354 changed
|= DATA_CHANGED
;
356 changed
|= ce_modified_check_fs(ce
, st
);
362 int ie_modified(const struct index_state
*istate
,
363 const struct cache_entry
*ce
,
364 struct stat
*st
, unsigned int options
)
366 int changed
, changed_fs
;
368 changed
= ie_match_stat(istate
, ce
, st
, options
);
372 * If the mode or type has changed, there's no point in trying
373 * to refresh the entry - it's not going to match
375 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
379 * Immediately after read-tree or update-index --cacheinfo,
380 * the length field is zero, as we have never even read the
381 * lstat(2) information once, and we cannot trust DATA_CHANGED
382 * returned by ie_match_stat() which in turn was returned by
383 * ce_match_stat_basic() to signal that the filesize of the
384 * blob changed. We have to actually go to the filesystem to
385 * see if the contents match, and if so, should answer "unchanged".
387 * The logic does not apply to gitlinks, as ce_match_stat_basic()
388 * already has checked the actual HEAD from the filesystem in the
389 * subproject. If ie_match_stat() already said it is different,
390 * then we know it is.
392 if ((changed
& DATA_CHANGED
) &&
393 (S_ISGITLINK(ce
->ce_mode
) || ce
->ce_stat_data
.sd_size
!= 0))
396 changed_fs
= ce_modified_check_fs(ce
, st
);
398 return changed
| changed_fs
;
402 int base_name_compare(const char *name1
, int len1
, int mode1
,
403 const char *name2
, int len2
, int mode2
)
405 unsigned char c1
, c2
;
406 int len
= len1
< len2
? len1
: len2
;
409 cmp
= memcmp(name1
, name2
, len
);
414 if (!c1
&& S_ISDIR(mode1
))
416 if (!c2
&& S_ISDIR(mode2
))
418 return (c1
< c2
) ? -1 : (c1
> c2
) ? 1 : 0;
422 * df_name_compare() is identical to base_name_compare(), except it
423 * compares conflicting directory/file entries as equal. Note that
424 * while a directory name compares as equal to a regular file, they
425 * then individually compare _differently_ to a filename that has
426 * a dot after the basename (because '\0' < '.' < '/').
428 * This is used by routines that want to traverse the git namespace
429 * but then handle conflicting entries together when possible.
431 int df_name_compare(const char *name1
, int len1
, int mode1
,
432 const char *name2
, int len2
, int mode2
)
434 int len
= len1
< len2
? len1
: len2
, cmp
;
435 unsigned char c1
, c2
;
437 cmp
= memcmp(name1
, name2
, len
);
440 /* Directories and files compare equal (same length, same name) */
444 if (!c1
&& S_ISDIR(mode1
))
447 if (!c2
&& S_ISDIR(mode2
))
449 if (c1
== '/' && !c2
)
451 if (c2
== '/' && !c1
)
456 int name_compare(const char *name1
, size_t len1
, const char *name2
, size_t len2
)
458 size_t min_len
= (len1
< len2
) ? len1
: len2
;
459 int cmp
= memcmp(name1
, name2
, min_len
);
469 int cache_name_stage_compare(const char *name1
, int len1
, int stage1
, const char *name2
, int len2
, int stage2
)
473 cmp
= name_compare(name1
, len1
, name2
, len2
);
484 static int index_name_stage_pos(const struct index_state
*istate
, const char *name
, int namelen
, int stage
)
489 last
= istate
->cache_nr
;
490 while (last
> first
) {
491 int next
= (last
+ first
) >> 1;
492 struct cache_entry
*ce
= istate
->cache
[next
];
493 int cmp
= cache_name_stage_compare(name
, namelen
, stage
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
505 int index_name_pos(const struct index_state
*istate
, const char *name
, int namelen
)
507 return index_name_stage_pos(istate
, name
, namelen
, 0);
510 /* Remove entry, return true if there are more entries to go.. */
511 int remove_index_entry_at(struct index_state
*istate
, int pos
)
513 struct cache_entry
*ce
= istate
->cache
[pos
];
515 record_resolve_undo(istate
, ce
);
516 remove_name_hash(istate
, ce
);
517 save_or_free_index_entry(istate
, ce
);
518 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
520 if (pos
>= istate
->cache_nr
)
522 memmove(istate
->cache
+ pos
,
523 istate
->cache
+ pos
+ 1,
524 (istate
->cache_nr
- pos
) * sizeof(struct cache_entry
*));
529 * Remove all cache entries marked for removal, that is where
530 * CE_REMOVE is set in ce_flags. This is much more effective than
531 * calling remove_index_entry_at() for each entry to be removed.
533 void remove_marked_cache_entries(struct index_state
*istate
)
535 struct cache_entry
**ce_array
= istate
->cache
;
538 for (i
= j
= 0; i
< istate
->cache_nr
; i
++) {
539 if (ce_array
[i
]->ce_flags
& CE_REMOVE
) {
540 remove_name_hash(istate
, ce_array
[i
]);
541 save_or_free_index_entry(istate
, ce_array
[i
]);
544 ce_array
[j
++] = ce_array
[i
];
546 if (j
== istate
->cache_nr
)
548 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
549 istate
->cache_nr
= j
;
552 int remove_file_from_index(struct index_state
*istate
, const char *path
)
554 int pos
= index_name_pos(istate
, path
, strlen(path
));
557 cache_tree_invalidate_path(istate
, path
);
558 untracked_cache_remove_from_index(istate
, path
);
559 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
560 remove_index_entry_at(istate
, pos
);
564 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
566 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
569 static int index_name_pos_also_unmerged(struct index_state
*istate
,
570 const char *path
, int namelen
)
572 int pos
= index_name_pos(istate
, path
, namelen
);
573 struct cache_entry
*ce
;
578 /* maybe unmerged? */
580 if (pos
>= istate
->cache_nr
||
581 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
584 /* order of preference: stage 2, 1, 3 */
585 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
586 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
587 !compare_name(ce
, path
, namelen
))
592 static int different_name(struct cache_entry
*ce
, struct cache_entry
*alias
)
594 int len
= ce_namelen(ce
);
595 return ce_namelen(alias
) != len
|| memcmp(ce
->name
, alias
->name
, len
);
599 * If we add a filename that aliases in the cache, we will use the
600 * name that we already have - but we don't want to update the same
601 * alias twice, because that implies that there were actually two
602 * different files with aliasing names!
604 * So we use the CE_ADDED flag to verify that the alias was an old
605 * one before we accept it as
607 static struct cache_entry
*create_alias_ce(struct index_state
*istate
,
608 struct cache_entry
*ce
,
609 struct cache_entry
*alias
)
612 struct cache_entry
*new;
614 if (alias
->ce_flags
& CE_ADDED
)
615 die("Will not add file alias '%s' ('%s' already exists in index)", ce
->name
, alias
->name
);
617 /* Ok, create the new entry using the name of the existing alias */
618 len
= ce_namelen(alias
);
619 new = xcalloc(1, cache_entry_size(len
));
620 memcpy(new->name
, alias
->name
, len
);
621 copy_cache_entry(new, ce
);
622 save_or_free_index_entry(istate
, ce
);
626 void set_object_name_for_intent_to_add_entry(struct cache_entry
*ce
)
628 unsigned char sha1
[20];
629 if (write_sha1_file("", 0, blob_type
, sha1
))
630 die("cannot create an empty blob in the object database");
631 hashcpy(ce
->sha1
, sha1
);
634 int add_to_index(struct index_state
*istate
, const char *path
, struct stat
*st
, int flags
)
636 int size
, namelen
, was_same
;
637 mode_t st_mode
= st
->st_mode
;
638 struct cache_entry
*ce
, *alias
;
639 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_IGNORE_SKIP_WORKTREE
|CE_MATCH_RACY_IS_DIRTY
;
640 int verbose
= flags
& (ADD_CACHE_VERBOSE
| ADD_CACHE_PRETEND
);
641 int pretend
= flags
& ADD_CACHE_PRETEND
;
642 int intent_only
= flags
& ADD_CACHE_INTENT
;
643 int add_option
= (ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
|
644 (intent_only
? ADD_CACHE_NEW_ONLY
: 0));
646 if (!S_ISREG(st_mode
) && !S_ISLNK(st_mode
) && !S_ISDIR(st_mode
))
647 return error("%s: can only add regular files, symbolic links or git-directories", path
);
649 namelen
= strlen(path
);
650 if (S_ISDIR(st_mode
)) {
651 while (namelen
&& path
[namelen
-1] == '/')
654 size
= cache_entry_size(namelen
);
655 ce
= xcalloc(1, size
);
656 memcpy(ce
->name
, path
, namelen
);
657 ce
->ce_namelen
= namelen
;
659 fill_stat_cache_info(ce
, st
);
661 ce
->ce_flags
|= CE_INTENT_TO_ADD
;
663 if (trust_executable_bit
&& has_symlinks
)
664 ce
->ce_mode
= create_ce_mode(st_mode
);
666 /* If there is an existing entry, pick the mode bits and type
667 * from it, otherwise assume unexecutable regular file.
669 struct cache_entry
*ent
;
670 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
672 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
673 ce
->ce_mode
= ce_mode_from_stat(ent
, st_mode
);
676 /* When core.ignorecase=true, determine if a directory of the same name but differing
677 * case already exists within the Git repository. If it does, ensure the directory
678 * case of the file being added to the repository matches (is folded into) the existing
679 * entry's directory case.
682 adjust_dirname_case(istate
, ce
->name
);
685 alias
= index_file_exists(istate
, ce
->name
, ce_namelen(ce
), ignore_case
);
686 if (alias
&& !ce_stage(alias
) && !ie_match_stat(istate
, alias
, st
, ce_option
)) {
687 /* Nothing changed, really */
688 if (!S_ISGITLINK(alias
->ce_mode
))
689 ce_mark_uptodate(alias
);
690 alias
->ce_flags
|= CE_ADDED
;
696 if (index_path(ce
->sha1
, path
, st
, HASH_WRITE_OBJECT
)) {
698 return error("unable to index file %s", path
);
701 set_object_name_for_intent_to_add_entry(ce
);
703 if (ignore_case
&& alias
&& different_name(ce
, alias
))
704 ce
= create_alias_ce(istate
, ce
, alias
);
705 ce
->ce_flags
|= CE_ADDED
;
707 /* It was suspected to be racily clean, but it turns out to be Ok */
710 !hashcmp(alias
->sha1
, ce
->sha1
) &&
711 ce
->ce_mode
== alias
->ce_mode
);
715 else if (add_index_entry(istate
, ce
, add_option
)) {
717 return error("unable to add %s to index", path
);
719 if (verbose
&& !was_same
)
720 printf("add '%s'\n", path
);
724 int add_file_to_index(struct index_state
*istate
, const char *path
, int flags
)
727 if (lstat(path
, &st
))
728 die_errno("unable to stat '%s'", path
);
729 return add_to_index(istate
, path
, &st
, flags
);
732 struct cache_entry
*make_cache_entry(unsigned int mode
,
733 const unsigned char *sha1
, const char *path
, int stage
,
734 unsigned int refresh_options
)
737 struct cache_entry
*ce
, *ret
;
739 if (!verify_path(path
)) {
740 error("Invalid path '%s'", path
);
745 size
= cache_entry_size(len
);
746 ce
= xcalloc(1, size
);
748 hashcpy(ce
->sha1
, sha1
);
749 memcpy(ce
->name
, path
, len
);
750 ce
->ce_flags
= create_ce_flags(stage
);
751 ce
->ce_namelen
= len
;
752 ce
->ce_mode
= create_ce_mode(mode
);
754 ret
= refresh_cache_entry(ce
, refresh_options
);
760 int ce_same_name(const struct cache_entry
*a
, const struct cache_entry
*b
)
762 int len
= ce_namelen(a
);
763 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
767 * We fundamentally don't like some paths: we don't want
768 * dot or dot-dot anywhere, and for obvious reasons don't
769 * want to recurse into ".git" either.
771 * Also, we don't want double slashes or slashes at the
772 * end that can make pathnames ambiguous.
774 static int verify_dotfile(const char *rest
)
777 * The first character was '.', but that
778 * has already been discarded, we now test
782 /* "." is not allowed */
783 if (*rest
== '\0' || is_dir_sep(*rest
))
788 * ".git" followed by NUL or slash is bad. This
789 * shares the path end test with the ".." case.
793 if (rest
[1] != 'i' && rest
[1] != 'I')
795 if (rest
[2] != 't' && rest
[2] != 'T')
800 if (rest
[1] == '\0' || is_dir_sep(rest
[1]))
806 int verify_path(const char *path
)
810 if (has_dos_drive_prefix(path
))
819 if (protect_hfs
&& is_hfs_dotgit(path
))
821 if (protect_ntfs
&& is_ntfs_dotgit(path
))
824 if ((c
== '.' && !verify_dotfile(path
)) ||
825 is_dir_sep(c
) || c
== '\0')
833 * Do we have another file that has the beginning components being a
834 * proper superset of the name we're trying to add?
836 static int has_file_name(struct index_state
*istate
,
837 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
840 int len
= ce_namelen(ce
);
841 int stage
= ce_stage(ce
);
842 const char *name
= ce
->name
;
844 while (pos
< istate
->cache_nr
) {
845 struct cache_entry
*p
= istate
->cache
[pos
++];
847 if (len
>= ce_namelen(p
))
849 if (memcmp(name
, p
->name
, len
))
851 if (ce_stage(p
) != stage
)
853 if (p
->name
[len
] != '/')
855 if (p
->ce_flags
& CE_REMOVE
)
860 remove_index_entry_at(istate
, --pos
);
866 * Do we have another file with a pathname that is a proper
867 * subset of the name we're trying to add?
869 static int has_dir_name(struct index_state
*istate
,
870 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
873 int stage
= ce_stage(ce
);
874 const char *name
= ce
->name
;
875 const char *slash
= name
+ ce_namelen(ce
);
883 if (slash
<= ce
->name
)
888 pos
= index_name_stage_pos(istate
, name
, len
, stage
);
891 * Found one, but not so fast. This could
892 * be a marker that says "I was here, but
893 * I am being removed". Such an entry is
894 * not a part of the resulting tree, and
895 * it is Ok to have a directory at the same
898 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
902 remove_index_entry_at(istate
, pos
);
910 * Trivial optimization: if we find an entry that
911 * already matches the sub-directory, then we know
912 * we're ok, and we can exit.
914 while (pos
< istate
->cache_nr
) {
915 struct cache_entry
*p
= istate
->cache
[pos
];
916 if ((ce_namelen(p
) <= len
) ||
917 (p
->name
[len
] != '/') ||
918 memcmp(p
->name
, name
, len
))
919 break; /* not our subdirectory */
920 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
922 * p is at the same stage as our entry, and
923 * is a subdirectory of what we are looking
924 * at, so we cannot have conflicts at our
925 * level or anything shorter.
934 /* We may be in a situation where we already have path/file and path
935 * is being added, or we already have path and path/file is being
936 * added. Either one would result in a nonsense tree that has path
937 * twice when git-write-tree tries to write it out. Prevent it.
939 * If ok-to-replace is specified, we remove the conflicting entries
940 * from the cache so the caller should recompute the insert position.
941 * When this happens, we return non-zero.
943 static int check_file_directory_conflict(struct index_state
*istate
,
944 const struct cache_entry
*ce
,
945 int pos
, int ok_to_replace
)
950 * When ce is an "I am going away" entry, we allow it to be added
952 if (ce
->ce_flags
& CE_REMOVE
)
956 * We check if the path is a sub-path of a subsequent pathname
957 * first, since removing those will not change the position
960 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
963 * Then check if the path might have a clashing sub-directory
966 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
969 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
972 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
973 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
974 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
975 int new_only
= option
& ADD_CACHE_NEW_ONLY
;
977 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
978 cache_tree_invalidate_path(istate
, ce
->name
);
979 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
981 /* existing match? Just replace it. */
984 replace_index_entry(istate
, pos
, ce
);
989 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
990 untracked_cache_add_to_index(istate
, ce
->name
);
993 * Inserting a merged entry ("stage 0") into the index
994 * will always replace all non-merged entries..
996 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
997 while (ce_same_name(istate
->cache
[pos
], ce
)) {
999 if (!remove_index_entry_at(istate
, pos
))
1006 if (!verify_path(ce
->name
))
1007 return error("Invalid path '%s'", ce
->name
);
1009 if (!skip_df_check
&&
1010 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
1012 return error("'%s' appears as both a file and as a directory",
1014 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1020 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1024 if (option
& ADD_CACHE_JUST_APPEND
)
1025 pos
= istate
->cache_nr
;
1028 ret
= add_index_entry_with_check(istate
, ce
, option
);
1034 /* Make sure the array is big enough .. */
1035 ALLOC_GROW(istate
->cache
, istate
->cache_nr
+ 1, istate
->cache_alloc
);
1039 if (istate
->cache_nr
> pos
+ 1)
1040 memmove(istate
->cache
+ pos
+ 1,
1041 istate
->cache
+ pos
,
1042 (istate
->cache_nr
- pos
- 1) * sizeof(ce
));
1043 set_index_entry(istate
, pos
, ce
);
1044 istate
->cache_changed
|= CE_ENTRY_ADDED
;
1049 * "refresh" does not calculate a new sha1 file or bring the
1050 * cache up-to-date for mode/content changes. But what it
1051 * _does_ do is to "re-match" the stat information of a file
1052 * with the cache, so that you can refresh the cache for a
1053 * file that hasn't been changed but where the stat entry is
1056 * For example, you'd want to do this after doing a "git-read-tree",
1057 * to link up the stat cache details with the proper files.
1059 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
1060 struct cache_entry
*ce
,
1061 unsigned int options
, int *err
,
1065 struct cache_entry
*updated
;
1067 int refresh
= options
& CE_MATCH_REFRESH
;
1068 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
1069 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
1070 int ignore_missing
= options
& CE_MATCH_IGNORE_MISSING
;
1072 if (!refresh
|| ce_uptodate(ce
))
1076 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1077 * that the change to the work tree does not matter and told
1080 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
)) {
1081 ce_mark_uptodate(ce
);
1084 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
)) {
1085 ce_mark_uptodate(ce
);
1089 if (has_symlink_leading_path(ce
->name
, ce_namelen(ce
))) {
1097 if (lstat(ce
->name
, &st
) < 0) {
1098 if (ignore_missing
&& errno
== ENOENT
)
1105 changed
= ie_match_stat(istate
, ce
, &st
, options
);
1107 *changed_ret
= changed
;
1110 * The path is unchanged. If we were told to ignore
1111 * valid bit, then we did the actual stat check and
1112 * found that the entry is unmodified. If the entry
1113 * is not marked VALID, this is the place to mark it
1114 * valid again, under "assume unchanged" mode.
1116 if (ignore_valid
&& assume_unchanged
&&
1117 !(ce
->ce_flags
& CE_VALID
))
1118 ; /* mark this one VALID again */
1121 * We do not mark the index itself "modified"
1122 * because CE_UPTODATE flag is in-core only;
1123 * we are not going to write this change out.
1125 if (!S_ISGITLINK(ce
->ce_mode
))
1126 ce_mark_uptodate(ce
);
1131 if (ie_modified(istate
, ce
, &st
, options
)) {
1138 updated
= xmalloc(size
);
1139 memcpy(updated
, ce
, size
);
1140 fill_stat_cache_info(updated
, &st
);
1142 * If ignore_valid is not set, we should leave CE_VALID bit
1143 * alone. Otherwise, paths marked with --no-assume-unchanged
1144 * (i.e. things to be edited) will reacquire CE_VALID bit
1145 * automatically, which is not really what we want.
1147 if (!ignore_valid
&& assume_unchanged
&&
1148 !(ce
->ce_flags
& CE_VALID
))
1149 updated
->ce_flags
&= ~CE_VALID
;
1151 /* istate->cache_changed is updated in the caller */
1155 static void show_file(const char * fmt
, const char * name
, int in_porcelain
,
1156 int * first
, const char *header_msg
)
1158 if (in_porcelain
&& *first
&& header_msg
) {
1159 printf("%s\n", header_msg
);
1165 int refresh_index(struct index_state
*istate
, unsigned int flags
,
1166 const struct pathspec
*pathspec
,
1167 char *seen
, const char *header_msg
)
1171 int really
= (flags
& REFRESH_REALLY
) != 0;
1172 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
1173 int quiet
= (flags
& REFRESH_QUIET
) != 0;
1174 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
1175 int ignore_submodules
= (flags
& REFRESH_IGNORE_SUBMODULES
) != 0;
1177 int in_porcelain
= (flags
& REFRESH_IN_PORCELAIN
);
1178 unsigned int options
= (CE_MATCH_REFRESH
|
1179 (really
? CE_MATCH_IGNORE_VALID
: 0) |
1180 (not_new
? CE_MATCH_IGNORE_MISSING
: 0));
1181 const char *modified_fmt
;
1182 const char *deleted_fmt
;
1183 const char *typechange_fmt
;
1184 const char *added_fmt
;
1185 const char *unmerged_fmt
;
1187 modified_fmt
= (in_porcelain
? "M\t%s\n" : "%s: needs update\n");
1188 deleted_fmt
= (in_porcelain
? "D\t%s\n" : "%s: needs update\n");
1189 typechange_fmt
= (in_porcelain
? "T\t%s\n" : "%s needs update\n");
1190 added_fmt
= (in_porcelain
? "A\t%s\n" : "%s needs update\n");
1191 unmerged_fmt
= (in_porcelain
? "U\t%s\n" : "%s: needs merge\n");
1192 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1193 struct cache_entry
*ce
, *new;
1194 int cache_errno
= 0;
1198 ce
= istate
->cache
[i
];
1199 if (ignore_submodules
&& S_ISGITLINK(ce
->ce_mode
))
1202 if (pathspec
&& !ce_path_match(ce
, pathspec
, seen
))
1206 while ((i
< istate
->cache_nr
) &&
1207 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
1213 show_file(unmerged_fmt
, ce
->name
, in_porcelain
,
1214 &first
, header_msg
);
1222 new = refresh_cache_ent(istate
, ce
, options
, &cache_errno
, &changed
);
1228 if (really
&& cache_errno
== EINVAL
) {
1229 /* If we are doing --really-refresh that
1230 * means the index is not valid anymore.
1232 ce
->ce_flags
&= ~CE_VALID
;
1233 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
1234 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
1239 if (cache_errno
== ENOENT
)
1241 else if (ce
->ce_flags
& CE_INTENT_TO_ADD
)
1242 fmt
= added_fmt
; /* must be before other checks */
1243 else if (changed
& TYPE_CHANGED
)
1244 fmt
= typechange_fmt
;
1248 ce
->name
, in_porcelain
, &first
, header_msg
);
1253 replace_index_entry(istate
, i
, new);
1258 static struct cache_entry
*refresh_cache_entry(struct cache_entry
*ce
,
1259 unsigned int options
)
1261 return refresh_cache_ent(&the_index
, ce
, options
, NULL
, NULL
);
1265 /*****************************************************************
1267 *****************************************************************/
1269 #define INDEX_FORMAT_DEFAULT 3
1271 static unsigned int get_index_format_default(void)
1273 char *envversion
= getenv("GIT_INDEX_VERSION");
1276 unsigned int version
= INDEX_FORMAT_DEFAULT
;
1279 if (!git_config_get_int("index.version", &value
))
1281 if (version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1282 warning(_("index.version set, but the value is invalid.\n"
1283 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1284 return INDEX_FORMAT_DEFAULT
;
1289 version
= strtoul(envversion
, &endp
, 10);
1291 version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1292 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1293 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1294 version
= INDEX_FORMAT_DEFAULT
;
1300 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1301 * Again - this is just a (very strong in practice) heuristic that
1302 * the inode hasn't changed.
1304 * We save the fields in big-endian order to allow using the
1305 * index file over NFS transparently.
1307 struct ondisk_cache_entry
{
1308 struct cache_time ctime
;
1309 struct cache_time mtime
;
1316 unsigned char sha1
[20];
1318 char name
[FLEX_ARRAY
]; /* more */
1322 * This struct is used when CE_EXTENDED bit is 1
1323 * The struct must match ondisk_cache_entry exactly from
1326 struct ondisk_cache_entry_extended
{
1327 struct cache_time ctime
;
1328 struct cache_time mtime
;
1335 unsigned char sha1
[20];
1338 char name
[FLEX_ARRAY
]; /* more */
1341 /* These are only used for v3 or lower */
1342 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1343 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1344 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1345 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1346 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1347 ondisk_cache_entry_size(ce_namelen(ce)))
1349 static int verify_hdr(struct cache_header
*hdr
, unsigned long size
)
1352 unsigned char sha1
[20];
1355 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
1356 return error("bad signature");
1357 hdr_version
= ntohl(hdr
->hdr_version
);
1358 if (hdr_version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< hdr_version
)
1359 return error("bad index version %d", hdr_version
);
1361 git_SHA1_Update(&c
, hdr
, size
- 20);
1362 git_SHA1_Final(sha1
, &c
);
1363 if (hashcmp(sha1
, (unsigned char *)hdr
+ size
- 20))
1364 return error("bad index file sha1 signature");
1368 static int read_index_extension(struct index_state
*istate
,
1369 const char *ext
, void *data
, unsigned long sz
)
1371 switch (CACHE_EXT(ext
)) {
1372 case CACHE_EXT_TREE
:
1373 istate
->cache_tree
= cache_tree_read(data
, sz
);
1375 case CACHE_EXT_RESOLVE_UNDO
:
1376 istate
->resolve_undo
= resolve_undo_read(data
, sz
);
1378 case CACHE_EXT_LINK
:
1379 if (read_link_extension(istate
, data
, sz
))
1382 case CACHE_EXT_UNTRACKED
:
1383 istate
->untracked
= read_untracked_extension(data
, sz
);
1386 if (*ext
< 'A' || 'Z' < *ext
)
1387 return error("index uses %.4s extension, which we do not understand",
1389 fprintf(stderr
, "ignoring %.4s extension\n", ext
);
1395 int hold_locked_index(struct lock_file
*lk
, int die_on_error
)
1397 return hold_lock_file_for_update(lk
, get_index_file(),
1403 int read_index(struct index_state
*istate
)
1405 return read_index_from(istate
, get_index_file());
1408 static struct cache_entry
*cache_entry_from_ondisk(struct ondisk_cache_entry
*ondisk
,
1413 struct cache_entry
*ce
= xmalloc(cache_entry_size(len
));
1415 ce
->ce_stat_data
.sd_ctime
.sec
= get_be32(&ondisk
->ctime
.sec
);
1416 ce
->ce_stat_data
.sd_mtime
.sec
= get_be32(&ondisk
->mtime
.sec
);
1417 ce
->ce_stat_data
.sd_ctime
.nsec
= get_be32(&ondisk
->ctime
.nsec
);
1418 ce
->ce_stat_data
.sd_mtime
.nsec
= get_be32(&ondisk
->mtime
.nsec
);
1419 ce
->ce_stat_data
.sd_dev
= get_be32(&ondisk
->dev
);
1420 ce
->ce_stat_data
.sd_ino
= get_be32(&ondisk
->ino
);
1421 ce
->ce_mode
= get_be32(&ondisk
->mode
);
1422 ce
->ce_stat_data
.sd_uid
= get_be32(&ondisk
->uid
);
1423 ce
->ce_stat_data
.sd_gid
= get_be32(&ondisk
->gid
);
1424 ce
->ce_stat_data
.sd_size
= get_be32(&ondisk
->size
);
1425 ce
->ce_flags
= flags
& ~CE_NAMEMASK
;
1426 ce
->ce_namelen
= len
;
1428 hashcpy(ce
->sha1
, ondisk
->sha1
);
1429 memcpy(ce
->name
, name
, len
);
1430 ce
->name
[len
] = '\0';
1435 * Adjacent cache entries tend to share the leading paths, so it makes
1436 * sense to only store the differences in later entries. In the v4
1437 * on-disk format of the index, each on-disk cache entry stores the
1438 * number of bytes to be stripped from the end of the previous name,
1439 * and the bytes to append to the result, to come up with its name.
1441 static unsigned long expand_name_field(struct strbuf
*name
, const char *cp_
)
1443 const unsigned char *ep
, *cp
= (const unsigned char *)cp_
;
1444 size_t len
= decode_varint(&cp
);
1446 if (name
->len
< len
)
1447 die("malformed name field in the index");
1448 strbuf_remove(name
, name
->len
- len
, len
);
1449 for (ep
= cp
; *ep
; ep
++)
1450 ; /* find the end */
1451 strbuf_add(name
, cp
, ep
- cp
);
1452 return (const char *)ep
+ 1 - cp_
;
1455 static struct cache_entry
*create_from_disk(struct ondisk_cache_entry
*ondisk
,
1456 unsigned long *ent_size
,
1457 struct strbuf
*previous_name
)
1459 struct cache_entry
*ce
;
1464 /* On-disk flags are just 16 bits */
1465 flags
= get_be16(&ondisk
->flags
);
1466 len
= flags
& CE_NAMEMASK
;
1468 if (flags
& CE_EXTENDED
) {
1469 struct ondisk_cache_entry_extended
*ondisk2
;
1471 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
1472 extended_flags
= get_be16(&ondisk2
->flags2
) << 16;
1473 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1474 if (extended_flags
& ~CE_EXTENDED_FLAGS
)
1475 die("Unknown index entry format %08x", extended_flags
);
1476 flags
|= extended_flags
;
1477 name
= ondisk2
->name
;
1480 name
= ondisk
->name
;
1482 if (!previous_name
) {
1483 /* v3 and earlier */
1484 if (len
== CE_NAMEMASK
)
1486 ce
= cache_entry_from_ondisk(ondisk
, flags
, name
, len
);
1488 *ent_size
= ondisk_ce_size(ce
);
1490 unsigned long consumed
;
1491 consumed
= expand_name_field(previous_name
, name
);
1492 ce
= cache_entry_from_ondisk(ondisk
, flags
,
1494 previous_name
->len
);
1496 *ent_size
= (name
- ((char *)ondisk
)) + consumed
;
1501 static void check_ce_order(struct index_state
*istate
)
1505 for (i
= 1; i
< istate
->cache_nr
; i
++) {
1506 struct cache_entry
*ce
= istate
->cache
[i
- 1];
1507 struct cache_entry
*next_ce
= istate
->cache
[i
];
1508 int name_compare
= strcmp(ce
->name
, next_ce
->name
);
1510 if (0 < name_compare
)
1511 die("unordered stage entries in index");
1512 if (!name_compare
) {
1514 die("multiple stage entries for merged file '%s'",
1516 if (ce_stage(ce
) > ce_stage(next_ce
))
1517 die("unordered stage entries for '%s'",
1523 /* remember to discard_cache() before reading a different cache! */
1524 int do_read_index(struct index_state
*istate
, const char *path
, int must_exist
)
1528 unsigned long src_offset
;
1529 struct cache_header
*hdr
;
1532 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
1534 if (istate
->initialized
)
1535 return istate
->cache_nr
;
1537 istate
->timestamp
.sec
= 0;
1538 istate
->timestamp
.nsec
= 0;
1539 fd
= open(path
, O_RDONLY
);
1541 if (!must_exist
&& errno
== ENOENT
)
1543 die_errno("%s: index file open failed", path
);
1547 die_errno("cannot stat the open index");
1549 mmap_size
= xsize_t(st
.st_size
);
1550 if (mmap_size
< sizeof(struct cache_header
) + 20)
1551 die("index file smaller than expected");
1553 mmap
= xmmap(NULL
, mmap_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1554 if (mmap
== MAP_FAILED
)
1555 die_errno("unable to map index file");
1559 if (verify_hdr(hdr
, mmap_size
) < 0)
1562 hashcpy(istate
->sha1
, (const unsigned char *)hdr
+ mmap_size
- 20);
1563 istate
->version
= ntohl(hdr
->hdr_version
);
1564 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
1565 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
1566 istate
->cache
= xcalloc(istate
->cache_alloc
, sizeof(*istate
->cache
));
1567 istate
->initialized
= 1;
1569 if (istate
->version
== 4)
1570 previous_name
= &previous_name_buf
;
1572 previous_name
= NULL
;
1574 src_offset
= sizeof(*hdr
);
1575 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1576 struct ondisk_cache_entry
*disk_ce
;
1577 struct cache_entry
*ce
;
1578 unsigned long consumed
;
1580 disk_ce
= (struct ondisk_cache_entry
*)((char *)mmap
+ src_offset
);
1581 ce
= create_from_disk(disk_ce
, &consumed
, previous_name
);
1582 set_index_entry(istate
, i
, ce
);
1584 src_offset
+= consumed
;
1586 strbuf_release(&previous_name_buf
);
1587 istate
->timestamp
.sec
= st
.st_mtime
;
1588 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
1590 while (src_offset
<= mmap_size
- 20 - 8) {
1591 /* After an array of active_nr index entries,
1592 * there can be arbitrary number of extended
1593 * sections, each of which is prefixed with
1594 * extension name (4-byte) and section length
1595 * in 4-byte network byte order.
1598 memcpy(&extsize
, (char *)mmap
+ src_offset
+ 4, 4);
1599 extsize
= ntohl(extsize
);
1600 if (read_index_extension(istate
,
1601 (const char *) mmap
+ src_offset
,
1602 (char *) mmap
+ src_offset
+ 8,
1606 src_offset
+= extsize
;
1608 munmap(mmap
, mmap_size
);
1609 return istate
->cache_nr
;
1612 munmap(mmap
, mmap_size
);
1613 die("index file corrupt");
1616 int read_index_from(struct index_state
*istate
, const char *path
)
1618 struct split_index
*split_index
;
1621 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1622 if (istate
->initialized
)
1623 return istate
->cache_nr
;
1625 ret
= do_read_index(istate
, path
, 0);
1626 split_index
= istate
->split_index
;
1627 if (!split_index
|| is_null_sha1(split_index
->base_sha1
)) {
1628 check_ce_order(istate
);
1632 if (split_index
->base
)
1633 discard_index(split_index
->base
);
1635 split_index
->base
= xcalloc(1, sizeof(*split_index
->base
));
1636 ret
= do_read_index(split_index
->base
,
1637 git_path("sharedindex.%s",
1638 sha1_to_hex(split_index
->base_sha1
)), 1);
1639 if (hashcmp(split_index
->base_sha1
, split_index
->base
->sha1
))
1640 die("broken index, expect %s in %s, got %s",
1641 sha1_to_hex(split_index
->base_sha1
),
1642 git_path("sharedindex.%s",
1643 sha1_to_hex(split_index
->base_sha1
)),
1644 sha1_to_hex(split_index
->base
->sha1
));
1645 merge_base_index(istate
);
1646 check_ce_order(istate
);
1650 int is_index_unborn(struct index_state
*istate
)
1652 return (!istate
->cache_nr
&& !istate
->timestamp
.sec
);
1655 int discard_index(struct index_state
*istate
)
1659 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1660 if (istate
->cache
[i
]->index
&&
1661 istate
->split_index
&&
1662 istate
->split_index
->base
&&
1663 istate
->cache
[i
]->index
<= istate
->split_index
->base
->cache_nr
&&
1664 istate
->cache
[i
] == istate
->split_index
->base
->cache
[istate
->cache
[i
]->index
- 1])
1666 free(istate
->cache
[i
]);
1668 resolve_undo_clear_index(istate
);
1669 istate
->cache_nr
= 0;
1670 istate
->cache_changed
= 0;
1671 istate
->timestamp
.sec
= 0;
1672 istate
->timestamp
.nsec
= 0;
1673 free_name_hash(istate
);
1674 cache_tree_free(&(istate
->cache_tree
));
1675 istate
->initialized
= 0;
1676 free(istate
->cache
);
1677 istate
->cache
= NULL
;
1678 istate
->cache_alloc
= 0;
1679 discard_split_index(istate
);
1680 free_untracked_cache(istate
->untracked
);
1681 istate
->untracked
= NULL
;
1685 int unmerged_index(const struct index_state
*istate
)
1688 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1689 if (ce_stage(istate
->cache
[i
]))
1695 #define WRITE_BUFFER_SIZE 8192
1696 static unsigned char write_buffer
[WRITE_BUFFER_SIZE
];
1697 static unsigned long write_buffer_len
;
1699 static int ce_write_flush(git_SHA_CTX
*context
, int fd
)
1701 unsigned int buffered
= write_buffer_len
;
1703 git_SHA1_Update(context
, write_buffer
, buffered
);
1704 if (write_in_full(fd
, write_buffer
, buffered
) != buffered
)
1706 write_buffer_len
= 0;
1711 static int ce_write(git_SHA_CTX
*context
, int fd
, void *data
, unsigned int len
)
1714 unsigned int buffered
= write_buffer_len
;
1715 unsigned int partial
= WRITE_BUFFER_SIZE
- buffered
;
1718 memcpy(write_buffer
+ buffered
, data
, partial
);
1719 buffered
+= partial
;
1720 if (buffered
== WRITE_BUFFER_SIZE
) {
1721 write_buffer_len
= buffered
;
1722 if (ce_write_flush(context
, fd
))
1726 write_buffer_len
= buffered
;
1728 data
= (char *) data
+ partial
;
1733 static int write_index_ext_header(git_SHA_CTX
*context
, int fd
,
1734 unsigned int ext
, unsigned int sz
)
1738 return ((ce_write(context
, fd
, &ext
, 4) < 0) ||
1739 (ce_write(context
, fd
, &sz
, 4) < 0)) ? -1 : 0;
1742 static int ce_flush(git_SHA_CTX
*context
, int fd
, unsigned char *sha1
)
1744 unsigned int left
= write_buffer_len
;
1747 write_buffer_len
= 0;
1748 git_SHA1_Update(context
, write_buffer
, left
);
1751 /* Flush first if not enough space for SHA1 signature */
1752 if (left
+ 20 > WRITE_BUFFER_SIZE
) {
1753 if (write_in_full(fd
, write_buffer
, left
) != left
)
1758 /* Append the SHA1 signature at the end */
1759 git_SHA1_Final(write_buffer
+ left
, context
);
1760 hashcpy(sha1
, write_buffer
+ left
);
1762 return (write_in_full(fd
, write_buffer
, left
) != left
) ? -1 : 0;
1765 static void ce_smudge_racily_clean_entry(struct cache_entry
*ce
)
1768 * The only thing we care about in this function is to smudge the
1769 * falsely clean entry due to touch-update-touch race, so we leave
1770 * everything else as they are. We are called for entries whose
1771 * ce_stat_data.sd_mtime match the index file mtime.
1773 * Note that this actually does not do much for gitlinks, for
1774 * which ce_match_stat_basic() always goes to the actual
1775 * contents. The caller checks with is_racy_timestamp() which
1776 * always says "no" for gitlinks, so we are not called for them ;-)
1780 if (lstat(ce
->name
, &st
) < 0)
1782 if (ce_match_stat_basic(ce
, &st
))
1784 if (ce_modified_check_fs(ce
, &st
)) {
1785 /* This is "racily clean"; smudge it. Note that this
1786 * is a tricky code. At first glance, it may appear
1787 * that it can break with this sequence:
1789 * $ echo xyzzy >frotz
1790 * $ git-update-index --add frotz
1793 * $ echo filfre >nitfol
1794 * $ git-update-index --add nitfol
1796 * but it does not. When the second update-index runs,
1797 * it notices that the entry "frotz" has the same timestamp
1798 * as index, and if we were to smudge it by resetting its
1799 * size to zero here, then the object name recorded
1800 * in index is the 6-byte file but the cached stat information
1801 * becomes zero --- which would then match what we would
1802 * obtain from the filesystem next time we stat("frotz").
1804 * However, the second update-index, before calling
1805 * this function, notices that the cached size is 6
1806 * bytes and what is on the filesystem is an empty
1807 * file, and never calls us, so the cached size information
1808 * for "frotz" stays 6 which does not match the filesystem.
1810 ce
->ce_stat_data
.sd_size
= 0;
1814 /* Copy miscellaneous fields but not the name */
1815 static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry
*ondisk
,
1816 struct cache_entry
*ce
)
1820 ondisk
->ctime
.sec
= htonl(ce
->ce_stat_data
.sd_ctime
.sec
);
1821 ondisk
->mtime
.sec
= htonl(ce
->ce_stat_data
.sd_mtime
.sec
);
1822 ondisk
->ctime
.nsec
= htonl(ce
->ce_stat_data
.sd_ctime
.nsec
);
1823 ondisk
->mtime
.nsec
= htonl(ce
->ce_stat_data
.sd_mtime
.nsec
);
1824 ondisk
->dev
= htonl(ce
->ce_stat_data
.sd_dev
);
1825 ondisk
->ino
= htonl(ce
->ce_stat_data
.sd_ino
);
1826 ondisk
->mode
= htonl(ce
->ce_mode
);
1827 ondisk
->uid
= htonl(ce
->ce_stat_data
.sd_uid
);
1828 ondisk
->gid
= htonl(ce
->ce_stat_data
.sd_gid
);
1829 ondisk
->size
= htonl(ce
->ce_stat_data
.sd_size
);
1830 hashcpy(ondisk
->sha1
, ce
->sha1
);
1832 flags
= ce
->ce_flags
& ~CE_NAMEMASK
;
1833 flags
|= (ce_namelen(ce
) >= CE_NAMEMASK
? CE_NAMEMASK
: ce_namelen(ce
));
1834 ondisk
->flags
= htons(flags
);
1835 if (ce
->ce_flags
& CE_EXTENDED
) {
1836 struct ondisk_cache_entry_extended
*ondisk2
;
1837 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
1838 ondisk2
->flags2
= htons((ce
->ce_flags
& CE_EXTENDED_FLAGS
) >> 16);
1839 return ondisk2
->name
;
1842 return ondisk
->name
;
1846 static int ce_write_entry(git_SHA_CTX
*c
, int fd
, struct cache_entry
*ce
,
1847 struct strbuf
*previous_name
)
1850 struct ondisk_cache_entry
*ondisk
;
1851 int saved_namelen
= saved_namelen
; /* compiler workaround */
1855 if (ce
->ce_flags
& CE_STRIP_NAME
) {
1856 saved_namelen
= ce_namelen(ce
);
1860 if (!previous_name
) {
1861 size
= ondisk_ce_size(ce
);
1862 ondisk
= xcalloc(1, size
);
1863 name
= copy_cache_entry_to_ondisk(ondisk
, ce
);
1864 memcpy(name
, ce
->name
, ce_namelen(ce
));
1866 int common
, to_remove
, prefix_size
;
1867 unsigned char to_remove_vi
[16];
1869 (ce
->name
[common
] &&
1870 common
< previous_name
->len
&&
1871 ce
->name
[common
] == previous_name
->buf
[common
]);
1873 ; /* still matching */
1874 to_remove
= previous_name
->len
- common
;
1875 prefix_size
= encode_varint(to_remove
, to_remove_vi
);
1877 if (ce
->ce_flags
& CE_EXTENDED
)
1878 size
= offsetof(struct ondisk_cache_entry_extended
, name
);
1880 size
= offsetof(struct ondisk_cache_entry
, name
);
1881 size
+= prefix_size
+ (ce_namelen(ce
) - common
+ 1);
1883 ondisk
= xcalloc(1, size
);
1884 name
= copy_cache_entry_to_ondisk(ondisk
, ce
);
1885 memcpy(name
, to_remove_vi
, prefix_size
);
1886 memcpy(name
+ prefix_size
, ce
->name
+ common
, ce_namelen(ce
) - common
);
1888 strbuf_splice(previous_name
, common
, to_remove
,
1889 ce
->name
+ common
, ce_namelen(ce
) - common
);
1891 if (ce
->ce_flags
& CE_STRIP_NAME
) {
1892 ce
->ce_namelen
= saved_namelen
;
1893 ce
->ce_flags
&= ~CE_STRIP_NAME
;
1896 result
= ce_write(c
, fd
, ondisk
, size
);
1902 * This function verifies if index_state has the correct sha1 of the
1903 * index file. Don't die if we have any other failure, just return 0.
1905 static int verify_index_from(const struct index_state
*istate
, const char *path
)
1910 unsigned char sha1
[20];
1912 if (!istate
->initialized
)
1915 fd
= open(path
, O_RDONLY
);
1922 if (st
.st_size
< sizeof(struct cache_header
) + 20)
1925 n
= pread_in_full(fd
, sha1
, 20, st
.st_size
- 20);
1929 if (hashcmp(istate
->sha1
, sha1
))
1940 static int verify_index(const struct index_state
*istate
)
1942 return verify_index_from(istate
, get_index_file());
1945 static int has_racy_timestamp(struct index_state
*istate
)
1947 int entries
= istate
->cache_nr
;
1950 for (i
= 0; i
< entries
; i
++) {
1951 struct cache_entry
*ce
= istate
->cache
[i
];
1952 if (is_racy_timestamp(istate
, ce
))
1959 * Opportunistically update the index but do not complain if we can't
1961 void update_index_if_able(struct index_state
*istate
, struct lock_file
*lockfile
)
1963 if ((istate
->cache_changed
|| has_racy_timestamp(istate
)) &&
1964 verify_index(istate
) &&
1965 write_locked_index(istate
, lockfile
, COMMIT_LOCK
))
1966 rollback_lock_file(lockfile
);
1969 static int do_write_index(struct index_state
*istate
, int newfd
,
1970 int strip_extensions
)
1973 struct cache_header hdr
;
1974 int i
, err
, removed
, extended
, hdr_version
;
1975 struct cache_entry
**cache
= istate
->cache
;
1976 int entries
= istate
->cache_nr
;
1978 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
1980 for (i
= removed
= extended
= 0; i
< entries
; i
++) {
1981 if (cache
[i
]->ce_flags
& CE_REMOVE
)
1984 /* reduce extended entries if possible */
1985 cache
[i
]->ce_flags
&= ~CE_EXTENDED
;
1986 if (cache
[i
]->ce_flags
& CE_EXTENDED_FLAGS
) {
1988 cache
[i
]->ce_flags
|= CE_EXTENDED
;
1992 if (!istate
->version
) {
1993 istate
->version
= get_index_format_default();
1994 if (getenv("GIT_TEST_SPLIT_INDEX"))
1995 init_split_index(istate
);
1998 /* demote version 3 to version 2 when the latter suffices */
1999 if (istate
->version
== 3 || istate
->version
== 2)
2000 istate
->version
= extended
? 3 : 2;
2002 hdr_version
= istate
->version
;
2004 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
2005 hdr
.hdr_version
= htonl(hdr_version
);
2006 hdr
.hdr_entries
= htonl(entries
- removed
);
2009 if (ce_write(&c
, newfd
, &hdr
, sizeof(hdr
)) < 0)
2012 previous_name
= (hdr_version
== 4) ? &previous_name_buf
: NULL
;
2013 for (i
= 0; i
< entries
; i
++) {
2014 struct cache_entry
*ce
= cache
[i
];
2015 if (ce
->ce_flags
& CE_REMOVE
)
2017 if (!ce_uptodate(ce
) && is_racy_timestamp(istate
, ce
))
2018 ce_smudge_racily_clean_entry(ce
);
2019 if (is_null_sha1(ce
->sha1
)) {
2020 static const char msg
[] = "cache entry has null sha1: %s";
2021 static int allow
= -1;
2024 allow
= git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2026 warning(msg
, ce
->name
);
2028 return error(msg
, ce
->name
);
2030 if (ce_write_entry(&c
, newfd
, ce
, previous_name
) < 0)
2033 strbuf_release(&previous_name_buf
);
2035 /* Write extension data here */
2036 if (!strip_extensions
&& istate
->split_index
) {
2037 struct strbuf sb
= STRBUF_INIT
;
2039 err
= write_link_extension(&sb
, istate
) < 0 ||
2040 write_index_ext_header(&c
, newfd
, CACHE_EXT_LINK
,
2042 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2043 strbuf_release(&sb
);
2047 if (!strip_extensions
&& istate
->cache_tree
) {
2048 struct strbuf sb
= STRBUF_INIT
;
2050 cache_tree_write(&sb
, istate
->cache_tree
);
2051 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_TREE
, sb
.len
) < 0
2052 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2053 strbuf_release(&sb
);
2057 if (!strip_extensions
&& istate
->resolve_undo
) {
2058 struct strbuf sb
= STRBUF_INIT
;
2060 resolve_undo_write(&sb
, istate
->resolve_undo
);
2061 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_RESOLVE_UNDO
,
2063 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2064 strbuf_release(&sb
);
2068 if (!strip_extensions
&& istate
->untracked
) {
2069 struct strbuf sb
= STRBUF_INIT
;
2071 write_untracked_extension(&sb
, istate
->untracked
);
2072 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_UNTRACKED
,
2074 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2075 strbuf_release(&sb
);
2080 if (ce_flush(&c
, newfd
, istate
->sha1
) || fstat(newfd
, &st
))
2082 istate
->timestamp
.sec
= (unsigned int)st
.st_mtime
;
2083 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
2087 void set_alternate_index_output(const char *name
)
2089 alternate_index_output
= name
;
2092 static int commit_locked_index(struct lock_file
*lk
)
2094 if (alternate_index_output
)
2095 return commit_lock_file_to(lk
, alternate_index_output
);
2097 return commit_lock_file(lk
);
2100 static int do_write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
2103 int ret
= do_write_index(istate
, get_lock_file_fd(lock
), 0);
2106 assert((flags
& (COMMIT_LOCK
| CLOSE_LOCK
)) !=
2107 (COMMIT_LOCK
| CLOSE_LOCK
));
2108 if (flags
& COMMIT_LOCK
)
2109 return commit_locked_index(lock
);
2110 else if (flags
& CLOSE_LOCK
)
2111 return close_lock_file(lock
);
2116 static int write_split_index(struct index_state
*istate
,
2117 struct lock_file
*lock
,
2121 prepare_to_write_split_index(istate
);
2122 ret
= do_write_locked_index(istate
, lock
, flags
);
2123 finish_writing_split_index(istate
);
2127 static struct tempfile temporary_sharedindex
;
2129 static int write_shared_index(struct index_state
*istate
,
2130 struct lock_file
*lock
, unsigned flags
)
2132 struct split_index
*si
= istate
->split_index
;
2135 fd
= mks_tempfile(&temporary_sharedindex
, git_path("sharedindex_XXXXXX"));
2137 hashclr(si
->base_sha1
);
2138 return do_write_locked_index(istate
, lock
, flags
);
2140 move_cache_to_base_index(istate
);
2141 ret
= do_write_index(si
->base
, fd
, 1);
2143 delete_tempfile(&temporary_sharedindex
);
2146 ret
= rename_tempfile(&temporary_sharedindex
,
2147 git_path("sharedindex.%s", sha1_to_hex(si
->base
->sha1
)));
2149 hashcpy(si
->base_sha1
, si
->base
->sha1
);
2153 int write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
2156 struct split_index
*si
= istate
->split_index
;
2158 if (!si
|| alternate_index_output
||
2159 (istate
->cache_changed
& ~EXTMASK
)) {
2161 hashclr(si
->base_sha1
);
2162 return do_write_locked_index(istate
, lock
, flags
);
2165 if (getenv("GIT_TEST_SPLIT_INDEX")) {
2166 int v
= si
->base_sha1
[0];
2168 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
2170 if (istate
->cache_changed
& SPLIT_INDEX_ORDERED
) {
2171 int ret
= write_shared_index(istate
, lock
, flags
);
2176 return write_split_index(istate
, lock
, flags
);
2180 * Read the index file that is potentially unmerged into given
2181 * index_state, dropping any unmerged entries. Returns true if
2182 * the index is unmerged. Callers who want to refuse to work
2183 * from an unmerged state can call this and check its return value,
2184 * instead of calling read_cache().
2186 int read_index_unmerged(struct index_state
*istate
)
2192 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2193 struct cache_entry
*ce
= istate
->cache
[i
];
2194 struct cache_entry
*new_ce
;
2200 len
= ce_namelen(ce
);
2201 size
= cache_entry_size(len
);
2202 new_ce
= xcalloc(1, size
);
2203 memcpy(new_ce
->name
, ce
->name
, len
);
2204 new_ce
->ce_flags
= create_ce_flags(0) | CE_CONFLICTED
;
2205 new_ce
->ce_namelen
= len
;
2206 new_ce
->ce_mode
= ce
->ce_mode
;
2207 if (add_index_entry(istate
, new_ce
, 0))
2208 return error("%s: cannot drop to stage #0",
2215 * Returns 1 if the path is an "other" path with respect to
2216 * the index; that is, the path is not mentioned in the index at all,
2217 * either as a file, a directory with some files in the index,
2218 * or as an unmerged entry.
2220 * We helpfully remove a trailing "/" from directories so that
2221 * the output of read_directory can be used as-is.
2223 int index_name_is_other(const struct index_state
*istate
, const char *name
,
2227 if (namelen
&& name
[namelen
- 1] == '/')
2229 pos
= index_name_pos(istate
, name
, namelen
);
2231 return 0; /* exact match */
2233 if (pos
< istate
->cache_nr
) {
2234 struct cache_entry
*ce
= istate
->cache
[pos
];
2235 if (ce_namelen(ce
) == namelen
&&
2236 !memcmp(ce
->name
, name
, namelen
))
2237 return 0; /* Yup, this one exists unmerged */
2242 void *read_blob_data_from_index(struct index_state
*istate
, const char *path
, unsigned long *size
)
2246 enum object_type type
;
2250 pos
= index_name_pos(istate
, path
, len
);
2253 * We might be in the middle of a merge, in which
2254 * case we would read stage #2 (ours).
2258 (pos
< 0 && i
< istate
->cache_nr
&&
2259 !strcmp(istate
->cache
[i
]->name
, path
));
2261 if (ce_stage(istate
->cache
[i
]) == 2)
2266 data
= read_sha1_file(istate
->cache
[pos
]->sha1
, &type
, &sz
);
2267 if (!data
|| type
!= OBJ_BLOB
) {
2276 void stat_validity_clear(struct stat_validity
*sv
)
2282 int stat_validity_check(struct stat_validity
*sv
, const char *path
)
2286 if (stat(path
, &st
) < 0)
2287 return sv
->sd
== NULL
;
2290 return S_ISREG(st
.st_mode
) && !match_stat_data(sv
->sd
, &st
);
2293 void stat_validity_update(struct stat_validity
*sv
, int fd
)
2297 if (fstat(fd
, &st
) < 0 || !S_ISREG(st
.st_mode
))
2298 stat_validity_clear(sv
);
2301 sv
->sd
= xcalloc(1, sizeof(struct stat_data
));
2302 fill_stat_data(sv
->sd
, &st
);