2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 #define NO_THE_INDEX_COMPATIBILITY_MACROS
8 #include "cache-tree.h"
14 #include "resolve-undo.h"
18 static struct cache_entry
*refresh_cache_entry(struct cache_entry
*ce
,
19 unsigned int options
);
21 /* Mask for the name length in ce_flags in the on-disk index */
23 #define CE_NAMEMASK (0x0fff)
27 * The first letter should be 'A'..'Z' for extensions that are not
28 * necessary for a correct operation (i.e. optimization data).
29 * When new extensions are added that _needs_ to be understood in
30 * order to correctly interpret the index file, pick character that
31 * is outside the range, to cause the reader to abort.
34 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
35 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
36 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
38 struct index_state the_index
;
40 static void set_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
42 istate
->cache
[nr
] = ce
;
43 add_name_hash(istate
, ce
);
46 static void replace_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
48 struct cache_entry
*old
= istate
->cache
[nr
];
50 remove_name_hash(istate
, old
);
52 set_index_entry(istate
, nr
, ce
);
53 istate
->cache_changed
= 1;
56 void rename_index_entry_at(struct index_state
*istate
, int nr
, const char *new_name
)
58 struct cache_entry
*old
= istate
->cache
[nr
], *new;
59 int namelen
= strlen(new_name
);
61 new = xmalloc(cache_entry_size(namelen
));
62 copy_cache_entry(new, old
);
63 new->ce_flags
&= ~CE_HASHED
;
64 new->ce_namelen
= namelen
;
65 memcpy(new->name
, new_name
, namelen
+ 1);
67 cache_tree_invalidate_path(istate
->cache_tree
, old
->name
);
68 remove_index_entry_at(istate
, nr
);
69 add_index_entry(istate
, new, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
72 void fill_stat_data(struct stat_data
*sd
, struct stat
*st
)
74 sd
->sd_ctime
.sec
= (unsigned int)st
->st_ctime
;
75 sd
->sd_mtime
.sec
= (unsigned int)st
->st_mtime
;
76 sd
->sd_ctime
.nsec
= ST_CTIME_NSEC(*st
);
77 sd
->sd_mtime
.nsec
= ST_MTIME_NSEC(*st
);
78 sd
->sd_dev
= st
->st_dev
;
79 sd
->sd_ino
= st
->st_ino
;
80 sd
->sd_uid
= st
->st_uid
;
81 sd
->sd_gid
= st
->st_gid
;
82 sd
->sd_size
= st
->st_size
;
85 int match_stat_data(const struct stat_data
*sd
, struct stat
*st
)
89 if (sd
->sd_mtime
.sec
!= (unsigned int)st
->st_mtime
)
90 changed
|= MTIME_CHANGED
;
91 if (trust_ctime
&& check_stat
&&
92 sd
->sd_ctime
.sec
!= (unsigned int)st
->st_ctime
)
93 changed
|= CTIME_CHANGED
;
96 if (check_stat
&& sd
->sd_mtime
.nsec
!= ST_MTIME_NSEC(*st
))
97 changed
|= MTIME_CHANGED
;
98 if (trust_ctime
&& check_stat
&&
99 sd
->sd_ctime
.nsec
!= ST_CTIME_NSEC(*st
))
100 changed
|= CTIME_CHANGED
;
104 if (sd
->sd_uid
!= (unsigned int) st
->st_uid
||
105 sd
->sd_gid
!= (unsigned int) st
->st_gid
)
106 changed
|= OWNER_CHANGED
;
107 if (sd
->sd_ino
!= (unsigned int) st
->st_ino
)
108 changed
|= INODE_CHANGED
;
113 * st_dev breaks on network filesystems where different
114 * clients will have different views of what "device"
115 * the filesystem is on
117 if (check_stat
&& sd
->sd_dev
!= (unsigned int) st
->st_dev
)
118 changed
|= INODE_CHANGED
;
121 if (sd
->sd_size
!= (unsigned int) st
->st_size
)
122 changed
|= DATA_CHANGED
;
128 * This only updates the "non-critical" parts of the directory
129 * cache, ie the parts that aren't tracked by GIT, and only used
130 * to validate the cache.
132 void fill_stat_cache_info(struct cache_entry
*ce
, struct stat
*st
)
134 fill_stat_data(&ce
->ce_stat_data
, st
);
136 if (assume_unchanged
)
137 ce
->ce_flags
|= CE_VALID
;
139 if (S_ISREG(st
->st_mode
))
140 ce_mark_uptodate(ce
);
143 static int ce_compare_data(const struct cache_entry
*ce
, struct stat
*st
)
146 int fd
= open(ce
->name
, O_RDONLY
);
149 unsigned char sha1
[20];
150 if (!index_fd(sha1
, fd
, st
, OBJ_BLOB
, ce
->name
, 0))
151 match
= hashcmp(sha1
, ce
->sha1
);
152 /* index_fd() closed the file descriptor already */
157 static int ce_compare_link(const struct cache_entry
*ce
, size_t expected_size
)
162 enum object_type type
;
163 struct strbuf sb
= STRBUF_INIT
;
165 if (strbuf_readlink(&sb
, ce
->name
, expected_size
))
168 buffer
= read_sha1_file(ce
->sha1
, &type
, &size
);
171 match
= memcmp(buffer
, sb
.buf
, size
);
178 static int ce_compare_gitlink(const struct cache_entry
*ce
)
180 unsigned char sha1
[20];
183 * We don't actually require that the .git directory
184 * under GITLINK directory be a valid git directory. It
185 * might even be missing (in case nobody populated that
188 * If so, we consider it always to match.
190 if (resolve_gitlink_ref(ce
->name
, "HEAD", sha1
) < 0)
192 return hashcmp(sha1
, ce
->sha1
);
195 static int ce_modified_check_fs(const struct cache_entry
*ce
, struct stat
*st
)
197 switch (st
->st_mode
& S_IFMT
) {
199 if (ce_compare_data(ce
, st
))
203 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
207 if (S_ISGITLINK(ce
->ce_mode
))
208 return ce_compare_gitlink(ce
) ? DATA_CHANGED
: 0;
215 static int ce_match_stat_basic(const struct cache_entry
*ce
, struct stat
*st
)
217 unsigned int changed
= 0;
219 if (ce
->ce_flags
& CE_REMOVE
)
220 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
222 switch (ce
->ce_mode
& S_IFMT
) {
224 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
225 /* We consider only the owner x bit to be relevant for
228 if (trust_executable_bit
&&
229 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
230 changed
|= MODE_CHANGED
;
233 if (!S_ISLNK(st
->st_mode
) &&
234 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
235 changed
|= TYPE_CHANGED
;
238 /* We ignore most of the st_xxx fields for gitlinks */
239 if (!S_ISDIR(st
->st_mode
))
240 changed
|= TYPE_CHANGED
;
241 else if (ce_compare_gitlink(ce
))
242 changed
|= DATA_CHANGED
;
245 die("internal error: ce_mode is %o", ce
->ce_mode
);
248 changed
|= match_stat_data(&ce
->ce_stat_data
, st
);
250 /* Racily smudged entry? */
251 if (!ce
->ce_stat_data
.sd_size
) {
252 if (!is_empty_blob_sha1(ce
->sha1
))
253 changed
|= DATA_CHANGED
;
259 static int is_racy_timestamp(const struct index_state
*istate
,
260 const struct cache_entry
*ce
)
262 return (!S_ISGITLINK(ce
->ce_mode
) &&
263 istate
->timestamp
.sec
&&
265 /* nanosecond timestamped files can also be racy! */
266 (istate
->timestamp
.sec
< ce
->ce_stat_data
.sd_mtime
.sec
||
267 (istate
->timestamp
.sec
== ce
->ce_stat_data
.sd_mtime
.sec
&&
268 istate
->timestamp
.nsec
<= ce
->ce_stat_data
.sd_mtime
.nsec
))
270 istate
->timestamp
.sec
<= ce
->ce_stat_data
.sd_mtime
.sec
275 int ie_match_stat(const struct index_state
*istate
,
276 const struct cache_entry
*ce
, struct stat
*st
,
277 unsigned int options
)
279 unsigned int changed
;
280 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
281 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
282 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
285 * If it's marked as always valid in the index, it's
286 * valid whatever the checked-out copy says.
288 * skip-worktree has the same effect with higher precedence
290 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
))
292 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
296 * Intent-to-add entries have not been added, so the index entry
297 * by definition never matches what is in the work tree until it
298 * actually gets added.
300 if (ce
->ce_flags
& CE_INTENT_TO_ADD
)
301 return DATA_CHANGED
| TYPE_CHANGED
| MODE_CHANGED
;
303 changed
= ce_match_stat_basic(ce
, st
);
306 * Within 1 second of this sequence:
307 * echo xyzzy >file && git-update-index --add file
308 * running this command:
310 * would give a falsely clean cache entry. The mtime and
311 * length match the cache, and other stat fields do not change.
313 * We could detect this at update-index time (the cache entry
314 * being registered/updated records the same time as "now")
315 * and delay the return from git-update-index, but that would
316 * effectively mean we can make at most one commit per second,
317 * which is not acceptable. Instead, we check cache entries
318 * whose mtime are the same as the index file timestamp more
319 * carefully than others.
321 if (!changed
&& is_racy_timestamp(istate
, ce
)) {
322 if (assume_racy_is_modified
)
323 changed
|= DATA_CHANGED
;
325 changed
|= ce_modified_check_fs(ce
, st
);
331 int ie_modified(const struct index_state
*istate
,
332 const struct cache_entry
*ce
,
333 struct stat
*st
, unsigned int options
)
335 int changed
, changed_fs
;
337 changed
= ie_match_stat(istate
, ce
, st
, options
);
341 * If the mode or type has changed, there's no point in trying
342 * to refresh the entry - it's not going to match
344 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
348 * Immediately after read-tree or update-index --cacheinfo,
349 * the length field is zero, as we have never even read the
350 * lstat(2) information once, and we cannot trust DATA_CHANGED
351 * returned by ie_match_stat() which in turn was returned by
352 * ce_match_stat_basic() to signal that the filesize of the
353 * blob changed. We have to actually go to the filesystem to
354 * see if the contents match, and if so, should answer "unchanged".
356 * The logic does not apply to gitlinks, as ce_match_stat_basic()
357 * already has checked the actual HEAD from the filesystem in the
358 * subproject. If ie_match_stat() already said it is different,
359 * then we know it is.
361 if ((changed
& DATA_CHANGED
) &&
362 (S_ISGITLINK(ce
->ce_mode
) || ce
->ce_stat_data
.sd_size
!= 0))
365 changed_fs
= ce_modified_check_fs(ce
, st
);
367 return changed
| changed_fs
;
371 int base_name_compare(const char *name1
, int len1
, int mode1
,
372 const char *name2
, int len2
, int mode2
)
374 unsigned char c1
, c2
;
375 int len
= len1
< len2
? len1
: len2
;
378 cmp
= memcmp(name1
, name2
, len
);
383 if (!c1
&& S_ISDIR(mode1
))
385 if (!c2
&& S_ISDIR(mode2
))
387 return (c1
< c2
) ? -1 : (c1
> c2
) ? 1 : 0;
391 * df_name_compare() is identical to base_name_compare(), except it
392 * compares conflicting directory/file entries as equal. Note that
393 * while a directory name compares as equal to a regular file, they
394 * then individually compare _differently_ to a filename that has
395 * a dot after the basename (because '\0' < '.' < '/').
397 * This is used by routines that want to traverse the git namespace
398 * but then handle conflicting entries together when possible.
400 int df_name_compare(const char *name1
, int len1
, int mode1
,
401 const char *name2
, int len2
, int mode2
)
403 int len
= len1
< len2
? len1
: len2
, cmp
;
404 unsigned char c1
, c2
;
406 cmp
= memcmp(name1
, name2
, len
);
409 /* Directories and files compare equal (same length, same name) */
413 if (!c1
&& S_ISDIR(mode1
))
416 if (!c2
&& S_ISDIR(mode2
))
418 if (c1
== '/' && !c2
)
420 if (c2
== '/' && !c1
)
425 int cache_name_stage_compare(const char *name1
, int len1
, int stage1
, const char *name2
, int len2
, int stage2
)
427 int len
= len1
< len2
? len1
: len2
;
430 cmp
= memcmp(name1
, name2
, len
);
445 int cache_name_compare(const char *name1
, int len1
, const char *name2
, int len2
)
447 return cache_name_stage_compare(name1
, len1
, 0, name2
, len2
, 0);
450 static int index_name_stage_pos(const struct index_state
*istate
, const char *name
, int namelen
, int stage
)
455 last
= istate
->cache_nr
;
456 while (last
> first
) {
457 int next
= (last
+ first
) >> 1;
458 struct cache_entry
*ce
= istate
->cache
[next
];
459 int cmp
= cache_name_stage_compare(name
, namelen
, stage
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
471 int index_name_pos(const struct index_state
*istate
, const char *name
, int namelen
)
473 return index_name_stage_pos(istate
, name
, namelen
, 0);
476 /* Remove entry, return true if there are more entries to go.. */
477 int remove_index_entry_at(struct index_state
*istate
, int pos
)
479 struct cache_entry
*ce
= istate
->cache
[pos
];
481 record_resolve_undo(istate
, ce
);
482 remove_name_hash(istate
, ce
);
484 istate
->cache_changed
= 1;
486 if (pos
>= istate
->cache_nr
)
488 memmove(istate
->cache
+ pos
,
489 istate
->cache
+ pos
+ 1,
490 (istate
->cache_nr
- pos
) * sizeof(struct cache_entry
*));
495 * Remove all cache entries marked for removal, that is where
496 * CE_REMOVE is set in ce_flags. This is much more effective than
497 * calling remove_index_entry_at() for each entry to be removed.
499 void remove_marked_cache_entries(struct index_state
*istate
)
501 struct cache_entry
**ce_array
= istate
->cache
;
504 for (i
= j
= 0; i
< istate
->cache_nr
; i
++) {
505 if (ce_array
[i
]->ce_flags
& CE_REMOVE
) {
506 remove_name_hash(istate
, ce_array
[i
]);
510 ce_array
[j
++] = ce_array
[i
];
512 istate
->cache_changed
= 1;
513 istate
->cache_nr
= j
;
516 int remove_file_from_index(struct index_state
*istate
, const char *path
)
518 int pos
= index_name_pos(istate
, path
, strlen(path
));
521 cache_tree_invalidate_path(istate
->cache_tree
, path
);
522 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
523 remove_index_entry_at(istate
, pos
);
527 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
529 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
532 static int index_name_pos_also_unmerged(struct index_state
*istate
,
533 const char *path
, int namelen
)
535 int pos
= index_name_pos(istate
, path
, namelen
);
536 struct cache_entry
*ce
;
541 /* maybe unmerged? */
543 if (pos
>= istate
->cache_nr
||
544 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
547 /* order of preference: stage 2, 1, 3 */
548 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
549 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
550 !compare_name(ce
, path
, namelen
))
555 static int different_name(struct cache_entry
*ce
, struct cache_entry
*alias
)
557 int len
= ce_namelen(ce
);
558 return ce_namelen(alias
) != len
|| memcmp(ce
->name
, alias
->name
, len
);
562 * If we add a filename that aliases in the cache, we will use the
563 * name that we already have - but we don't want to update the same
564 * alias twice, because that implies that there were actually two
565 * different files with aliasing names!
567 * So we use the CE_ADDED flag to verify that the alias was an old
568 * one before we accept it as
570 static struct cache_entry
*create_alias_ce(struct cache_entry
*ce
, struct cache_entry
*alias
)
573 struct cache_entry
*new;
575 if (alias
->ce_flags
& CE_ADDED
)
576 die("Will not add file alias '%s' ('%s' already exists in index)", ce
->name
, alias
->name
);
578 /* Ok, create the new entry using the name of the existing alias */
579 len
= ce_namelen(alias
);
580 new = xcalloc(1, cache_entry_size(len
));
581 memcpy(new->name
, alias
->name
, len
);
582 copy_cache_entry(new, ce
);
587 static void record_intent_to_add(struct cache_entry
*ce
)
589 unsigned char sha1
[20];
590 if (write_sha1_file("", 0, blob_type
, sha1
))
591 die("cannot create an empty blob in the object database");
592 hashcpy(ce
->sha1
, sha1
);
595 int add_to_index(struct index_state
*istate
, const char *path
, struct stat
*st
, int flags
)
597 int size
, namelen
, was_same
;
598 mode_t st_mode
= st
->st_mode
;
599 struct cache_entry
*ce
, *alias
;
600 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_IGNORE_SKIP_WORKTREE
|CE_MATCH_RACY_IS_DIRTY
;
601 int verbose
= flags
& (ADD_CACHE_VERBOSE
| ADD_CACHE_PRETEND
);
602 int pretend
= flags
& ADD_CACHE_PRETEND
;
603 int intent_only
= flags
& ADD_CACHE_INTENT
;
604 int add_option
= (ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
|
605 (intent_only
? ADD_CACHE_NEW_ONLY
: 0));
607 if (!S_ISREG(st_mode
) && !S_ISLNK(st_mode
) && !S_ISDIR(st_mode
))
608 return error("%s: can only add regular files, symbolic links or git-directories", path
);
610 namelen
= strlen(path
);
611 if (S_ISDIR(st_mode
)) {
612 while (namelen
&& path
[namelen
-1] == '/')
615 size
= cache_entry_size(namelen
);
616 ce
= xcalloc(1, size
);
617 memcpy(ce
->name
, path
, namelen
);
618 ce
->ce_namelen
= namelen
;
620 fill_stat_cache_info(ce
, st
);
622 ce
->ce_flags
|= CE_INTENT_TO_ADD
;
624 if (trust_executable_bit
&& has_symlinks
)
625 ce
->ce_mode
= create_ce_mode(st_mode
);
627 /* If there is an existing entry, pick the mode bits and type
628 * from it, otherwise assume unexecutable regular file.
630 struct cache_entry
*ent
;
631 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
633 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
634 ce
->ce_mode
= ce_mode_from_stat(ent
, st_mode
);
637 /* When core.ignorecase=true, determine if a directory of the same name but differing
638 * case already exists within the Git repository. If it does, ensure the directory
639 * case of the file being added to the repository matches (is folded into) the existing
640 * entry's directory case.
643 const char *startPtr
= ce
->name
;
644 const char *ptr
= startPtr
;
646 while (*ptr
&& *ptr
!= '/')
649 struct cache_entry
*foundce
;
651 foundce
= index_dir_exists(istate
, ce
->name
, ptr
- ce
->name
- 1);
653 memcpy((void *)startPtr
, foundce
->name
+ (startPtr
- ce
->name
), ptr
- startPtr
);
660 alias
= index_file_exists(istate
, ce
->name
, ce_namelen(ce
), ignore_case
);
661 if (alias
&& !ce_stage(alias
) && !ie_match_stat(istate
, alias
, st
, ce_option
)) {
662 /* Nothing changed, really */
664 if (!S_ISGITLINK(alias
->ce_mode
))
665 ce_mark_uptodate(alias
);
666 alias
->ce_flags
|= CE_ADDED
;
670 if (index_path(ce
->sha1
, path
, st
, HASH_WRITE_OBJECT
))
671 return error("unable to index file %s", path
);
673 record_intent_to_add(ce
);
675 if (ignore_case
&& alias
&& different_name(ce
, alias
))
676 ce
= create_alias_ce(ce
, alias
);
677 ce
->ce_flags
|= CE_ADDED
;
679 /* It was suspected to be racily clean, but it turns out to be Ok */
682 !hashcmp(alias
->sha1
, ce
->sha1
) &&
683 ce
->ce_mode
== alias
->ce_mode
);
687 else if (add_index_entry(istate
, ce
, add_option
))
688 return error("unable to add %s to index",path
);
689 if (verbose
&& !was_same
)
690 printf("add '%s'\n", path
);
694 int add_file_to_index(struct index_state
*istate
, const char *path
, int flags
)
697 if (lstat(path
, &st
))
698 die_errno("unable to stat '%s'", path
);
699 return add_to_index(istate
, path
, &st
, flags
);
702 struct cache_entry
*make_cache_entry(unsigned int mode
,
703 const unsigned char *sha1
, const char *path
, int stage
,
704 unsigned int refresh_options
)
707 struct cache_entry
*ce
;
709 if (!verify_path(path
)) {
710 error("Invalid path '%s'", path
);
715 size
= cache_entry_size(len
);
716 ce
= xcalloc(1, size
);
718 hashcpy(ce
->sha1
, sha1
);
719 memcpy(ce
->name
, path
, len
);
720 ce
->ce_flags
= create_ce_flags(stage
);
721 ce
->ce_namelen
= len
;
722 ce
->ce_mode
= create_ce_mode(mode
);
724 return refresh_cache_entry(ce
, refresh_options
);
727 int ce_same_name(const struct cache_entry
*a
, const struct cache_entry
*b
)
729 int len
= ce_namelen(a
);
730 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
733 int ce_path_match(const struct cache_entry
*ce
, const struct pathspec
*pathspec
)
735 return match_pathspec_depth(pathspec
, ce
->name
, ce_namelen(ce
), 0, NULL
);
739 * We fundamentally don't like some paths: we don't want
740 * dot or dot-dot anywhere, and for obvious reasons don't
741 * want to recurse into ".git" either.
743 * Also, we don't want double slashes or slashes at the
744 * end that can make pathnames ambiguous.
746 static int verify_dotfile(const char *rest
)
749 * The first character was '.', but that
750 * has already been discarded, we now test
754 /* "." is not allowed */
755 if (*rest
== '\0' || is_dir_sep(*rest
))
760 * ".git" followed by NUL or slash is bad. This
761 * shares the path end test with the ".." case.
771 if (rest
[1] == '\0' || is_dir_sep(rest
[1]))
777 int verify_path(const char *path
)
781 if (has_dos_drive_prefix(path
))
791 if ((c
== '.' && !verify_dotfile(path
)) ||
792 is_dir_sep(c
) || c
== '\0')
800 * Do we have another file that has the beginning components being a
801 * proper superset of the name we're trying to add?
803 static int has_file_name(struct index_state
*istate
,
804 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
807 int len
= ce_namelen(ce
);
808 int stage
= ce_stage(ce
);
809 const char *name
= ce
->name
;
811 while (pos
< istate
->cache_nr
) {
812 struct cache_entry
*p
= istate
->cache
[pos
++];
814 if (len
>= ce_namelen(p
))
816 if (memcmp(name
, p
->name
, len
))
818 if (ce_stage(p
) != stage
)
820 if (p
->name
[len
] != '/')
822 if (p
->ce_flags
& CE_REMOVE
)
827 remove_index_entry_at(istate
, --pos
);
833 * Do we have another file with a pathname that is a proper
834 * subset of the name we're trying to add?
836 static int has_dir_name(struct index_state
*istate
,
837 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
840 int stage
= ce_stage(ce
);
841 const char *name
= ce
->name
;
842 const char *slash
= name
+ ce_namelen(ce
);
850 if (slash
<= ce
->name
)
855 pos
= index_name_stage_pos(istate
, name
, len
, stage
);
858 * Found one, but not so fast. This could
859 * be a marker that says "I was here, but
860 * I am being removed". Such an entry is
861 * not a part of the resulting tree, and
862 * it is Ok to have a directory at the same
865 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
869 remove_index_entry_at(istate
, pos
);
877 * Trivial optimization: if we find an entry that
878 * already matches the sub-directory, then we know
879 * we're ok, and we can exit.
881 while (pos
< istate
->cache_nr
) {
882 struct cache_entry
*p
= istate
->cache
[pos
];
883 if ((ce_namelen(p
) <= len
) ||
884 (p
->name
[len
] != '/') ||
885 memcmp(p
->name
, name
, len
))
886 break; /* not our subdirectory */
887 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
889 * p is at the same stage as our entry, and
890 * is a subdirectory of what we are looking
891 * at, so we cannot have conflicts at our
892 * level or anything shorter.
901 /* We may be in a situation where we already have path/file and path
902 * is being added, or we already have path and path/file is being
903 * added. Either one would result in a nonsense tree that has path
904 * twice when git-write-tree tries to write it out. Prevent it.
906 * If ok-to-replace is specified, we remove the conflicting entries
907 * from the cache so the caller should recompute the insert position.
908 * When this happens, we return non-zero.
910 static int check_file_directory_conflict(struct index_state
*istate
,
911 const struct cache_entry
*ce
,
912 int pos
, int ok_to_replace
)
917 * When ce is an "I am going away" entry, we allow it to be added
919 if (ce
->ce_flags
& CE_REMOVE
)
923 * We check if the path is a sub-path of a subsequent pathname
924 * first, since removing those will not change the position
927 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
930 * Then check if the path might have a clashing sub-directory
933 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
936 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
939 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
940 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
941 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
942 int new_only
= option
& ADD_CACHE_NEW_ONLY
;
944 cache_tree_invalidate_path(istate
->cache_tree
, ce
->name
);
945 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
947 /* existing match? Just replace it. */
950 replace_index_entry(istate
, pos
, ce
);
956 * Inserting a merged entry ("stage 0") into the index
957 * will always replace all non-merged entries..
959 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
960 while (ce_same_name(istate
->cache
[pos
], ce
)) {
962 if (!remove_index_entry_at(istate
, pos
))
969 if (!verify_path(ce
->name
))
970 return error("Invalid path '%s'", ce
->name
);
972 if (!skip_df_check
&&
973 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
975 return error("'%s' appears as both a file and as a directory",
977 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
983 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
987 if (option
& ADD_CACHE_JUST_APPEND
)
988 pos
= istate
->cache_nr
;
991 ret
= add_index_entry_with_check(istate
, ce
, option
);
997 /* Make sure the array is big enough .. */
998 if (istate
->cache_nr
== istate
->cache_alloc
) {
999 istate
->cache_alloc
= alloc_nr(istate
->cache_alloc
);
1000 istate
->cache
= xrealloc(istate
->cache
,
1001 istate
->cache_alloc
* sizeof(*istate
->cache
));
1006 if (istate
->cache_nr
> pos
+ 1)
1007 memmove(istate
->cache
+ pos
+ 1,
1008 istate
->cache
+ pos
,
1009 (istate
->cache_nr
- pos
- 1) * sizeof(ce
));
1010 set_index_entry(istate
, pos
, ce
);
1011 istate
->cache_changed
= 1;
1016 * "refresh" does not calculate a new sha1 file or bring the
1017 * cache up-to-date for mode/content changes. But what it
1018 * _does_ do is to "re-match" the stat information of a file
1019 * with the cache, so that you can refresh the cache for a
1020 * file that hasn't been changed but where the stat entry is
1023 * For example, you'd want to do this after doing a "git-read-tree",
1024 * to link up the stat cache details with the proper files.
1026 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
1027 struct cache_entry
*ce
,
1028 unsigned int options
, int *err
,
1032 struct cache_entry
*updated
;
1034 int refresh
= options
& CE_MATCH_REFRESH
;
1035 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
1036 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
1037 int ignore_missing
= options
& CE_MATCH_IGNORE_MISSING
;
1039 if (!refresh
|| ce_uptodate(ce
))
1043 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1044 * that the change to the work tree does not matter and told
1047 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
)) {
1048 ce_mark_uptodate(ce
);
1051 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
)) {
1052 ce_mark_uptodate(ce
);
1056 if (lstat(ce
->name
, &st
) < 0) {
1057 if (ignore_missing
&& errno
== ENOENT
)
1064 changed
= ie_match_stat(istate
, ce
, &st
, options
);
1066 *changed_ret
= changed
;
1069 * The path is unchanged. If we were told to ignore
1070 * valid bit, then we did the actual stat check and
1071 * found that the entry is unmodified. If the entry
1072 * is not marked VALID, this is the place to mark it
1073 * valid again, under "assume unchanged" mode.
1075 if (ignore_valid
&& assume_unchanged
&&
1076 !(ce
->ce_flags
& CE_VALID
))
1077 ; /* mark this one VALID again */
1080 * We do not mark the index itself "modified"
1081 * because CE_UPTODATE flag is in-core only;
1082 * we are not going to write this change out.
1084 if (!S_ISGITLINK(ce
->ce_mode
))
1085 ce_mark_uptodate(ce
);
1090 if (ie_modified(istate
, ce
, &st
, options
)) {
1097 updated
= xmalloc(size
);
1098 memcpy(updated
, ce
, size
);
1099 fill_stat_cache_info(updated
, &st
);
1101 * If ignore_valid is not set, we should leave CE_VALID bit
1102 * alone. Otherwise, paths marked with --no-assume-unchanged
1103 * (i.e. things to be edited) will reacquire CE_VALID bit
1104 * automatically, which is not really what we want.
1106 if (!ignore_valid
&& assume_unchanged
&&
1107 !(ce
->ce_flags
& CE_VALID
))
1108 updated
->ce_flags
&= ~CE_VALID
;
1113 static void show_file(const char * fmt
, const char * name
, int in_porcelain
,
1114 int * first
, const char *header_msg
)
1116 if (in_porcelain
&& *first
&& header_msg
) {
1117 printf("%s\n", header_msg
);
1123 int refresh_index(struct index_state
*istate
, unsigned int flags
,
1124 const struct pathspec
*pathspec
,
1125 char *seen
, const char *header_msg
)
1129 int really
= (flags
& REFRESH_REALLY
) != 0;
1130 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
1131 int quiet
= (flags
& REFRESH_QUIET
) != 0;
1132 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
1133 int ignore_submodules
= (flags
& REFRESH_IGNORE_SUBMODULES
) != 0;
1135 int in_porcelain
= (flags
& REFRESH_IN_PORCELAIN
);
1136 unsigned int options
= (CE_MATCH_REFRESH
|
1137 (really
? CE_MATCH_IGNORE_VALID
: 0) |
1138 (not_new
? CE_MATCH_IGNORE_MISSING
: 0));
1139 const char *modified_fmt
;
1140 const char *deleted_fmt
;
1141 const char *typechange_fmt
;
1142 const char *added_fmt
;
1143 const char *unmerged_fmt
;
1145 modified_fmt
= (in_porcelain
? "M\t%s\n" : "%s: needs update\n");
1146 deleted_fmt
= (in_porcelain
? "D\t%s\n" : "%s: needs update\n");
1147 typechange_fmt
= (in_porcelain
? "T\t%s\n" : "%s needs update\n");
1148 added_fmt
= (in_porcelain
? "A\t%s\n" : "%s needs update\n");
1149 unmerged_fmt
= (in_porcelain
? "U\t%s\n" : "%s: needs merge\n");
1150 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1151 struct cache_entry
*ce
, *new;
1152 int cache_errno
= 0;
1156 ce
= istate
->cache
[i
];
1157 if (ignore_submodules
&& S_ISGITLINK(ce
->ce_mode
))
1161 !match_pathspec_depth(pathspec
, ce
->name
, ce_namelen(ce
), 0, seen
))
1165 while ((i
< istate
->cache_nr
) &&
1166 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
1172 show_file(unmerged_fmt
, ce
->name
, in_porcelain
,
1173 &first
, header_msg
);
1181 new = refresh_cache_ent(istate
, ce
, options
, &cache_errno
, &changed
);
1187 if (really
&& cache_errno
== EINVAL
) {
1188 /* If we are doing --really-refresh that
1189 * means the index is not valid anymore.
1191 ce
->ce_flags
&= ~CE_VALID
;
1192 istate
->cache_changed
= 1;
1197 if (cache_errno
== ENOENT
)
1199 else if (ce
->ce_flags
& CE_INTENT_TO_ADD
)
1200 fmt
= added_fmt
; /* must be before other checks */
1201 else if (changed
& TYPE_CHANGED
)
1202 fmt
= typechange_fmt
;
1206 ce
->name
, in_porcelain
, &first
, header_msg
);
1211 replace_index_entry(istate
, i
, new);
1216 static struct cache_entry
*refresh_cache_entry(struct cache_entry
*ce
,
1217 unsigned int options
)
1219 return refresh_cache_ent(&the_index
, ce
, options
, NULL
, NULL
);
1223 /*****************************************************************
1225 *****************************************************************/
1227 #define INDEX_FORMAT_DEFAULT 3
1230 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1231 * Again - this is just a (very strong in practice) heuristic that
1232 * the inode hasn't changed.
1234 * We save the fields in big-endian order to allow using the
1235 * index file over NFS transparently.
1237 struct ondisk_cache_entry
{
1238 struct cache_time ctime
;
1239 struct cache_time mtime
;
1246 unsigned char sha1
[20];
1248 char name
[FLEX_ARRAY
]; /* more */
1252 * This struct is used when CE_EXTENDED bit is 1
1253 * The struct must match ondisk_cache_entry exactly from
1256 struct ondisk_cache_entry_extended
{
1257 struct cache_time ctime
;
1258 struct cache_time mtime
;
1265 unsigned char sha1
[20];
1268 char name
[FLEX_ARRAY
]; /* more */
1271 /* These are only used for v3 or lower */
1272 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1273 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1274 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1275 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1276 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1277 ondisk_cache_entry_size(ce_namelen(ce)))
1279 static int verify_hdr(struct cache_header
*hdr
, unsigned long size
)
1282 unsigned char sha1
[20];
1285 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
1286 return error("bad signature");
1287 hdr_version
= ntohl(hdr
->hdr_version
);
1288 if (hdr_version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< hdr_version
)
1289 return error("bad index version %d", hdr_version
);
1291 git_SHA1_Update(&c
, hdr
, size
- 20);
1292 git_SHA1_Final(sha1
, &c
);
1293 if (hashcmp(sha1
, (unsigned char *)hdr
+ size
- 20))
1294 return error("bad index file sha1 signature");
1298 static int read_index_extension(struct index_state
*istate
,
1299 const char *ext
, void *data
, unsigned long sz
)
1301 switch (CACHE_EXT(ext
)) {
1302 case CACHE_EXT_TREE
:
1303 istate
->cache_tree
= cache_tree_read(data
, sz
);
1305 case CACHE_EXT_RESOLVE_UNDO
:
1306 istate
->resolve_undo
= resolve_undo_read(data
, sz
);
1309 if (*ext
< 'A' || 'Z' < *ext
)
1310 return error("index uses %.4s extension, which we do not understand",
1312 fprintf(stderr
, "ignoring %.4s extension\n", ext
);
1318 int read_index(struct index_state
*istate
)
1320 return read_index_from(istate
, get_index_file());
1323 #ifndef NEEDS_ALIGNED_ACCESS
1324 #define ntoh_s(var) ntohs(var)
1325 #define ntoh_l(var) ntohl(var)
1327 static inline uint16_t ntoh_s_force_align(void *p
)
1330 memcpy(&x
, p
, sizeof(x
));
1333 static inline uint32_t ntoh_l_force_align(void *p
)
1336 memcpy(&x
, p
, sizeof(x
));
1339 #define ntoh_s(var) ntoh_s_force_align(&(var))
1340 #define ntoh_l(var) ntoh_l_force_align(&(var))
1343 static struct cache_entry
*cache_entry_from_ondisk(struct ondisk_cache_entry
*ondisk
,
1348 struct cache_entry
*ce
= xmalloc(cache_entry_size(len
));
1350 ce
->ce_stat_data
.sd_ctime
.sec
= ntoh_l(ondisk
->ctime
.sec
);
1351 ce
->ce_stat_data
.sd_mtime
.sec
= ntoh_l(ondisk
->mtime
.sec
);
1352 ce
->ce_stat_data
.sd_ctime
.nsec
= ntoh_l(ondisk
->ctime
.nsec
);
1353 ce
->ce_stat_data
.sd_mtime
.nsec
= ntoh_l(ondisk
->mtime
.nsec
);
1354 ce
->ce_stat_data
.sd_dev
= ntoh_l(ondisk
->dev
);
1355 ce
->ce_stat_data
.sd_ino
= ntoh_l(ondisk
->ino
);
1356 ce
->ce_mode
= ntoh_l(ondisk
->mode
);
1357 ce
->ce_stat_data
.sd_uid
= ntoh_l(ondisk
->uid
);
1358 ce
->ce_stat_data
.sd_gid
= ntoh_l(ondisk
->gid
);
1359 ce
->ce_stat_data
.sd_size
= ntoh_l(ondisk
->size
);
1360 ce
->ce_flags
= flags
& ~CE_NAMEMASK
;
1361 ce
->ce_namelen
= len
;
1362 hashcpy(ce
->sha1
, ondisk
->sha1
);
1363 memcpy(ce
->name
, name
, len
);
1364 ce
->name
[len
] = '\0';
1369 * Adjacent cache entries tend to share the leading paths, so it makes
1370 * sense to only store the differences in later entries. In the v4
1371 * on-disk format of the index, each on-disk cache entry stores the
1372 * number of bytes to be stripped from the end of the previous name,
1373 * and the bytes to append to the result, to come up with its name.
1375 static unsigned long expand_name_field(struct strbuf
*name
, const char *cp_
)
1377 const unsigned char *ep
, *cp
= (const unsigned char *)cp_
;
1378 size_t len
= decode_varint(&cp
);
1380 if (name
->len
< len
)
1381 die("malformed name field in the index");
1382 strbuf_remove(name
, name
->len
- len
, len
);
1383 for (ep
= cp
; *ep
; ep
++)
1384 ; /* find the end */
1385 strbuf_add(name
, cp
, ep
- cp
);
1386 return (const char *)ep
+ 1 - cp_
;
1389 static struct cache_entry
*create_from_disk(struct ondisk_cache_entry
*ondisk
,
1390 unsigned long *ent_size
,
1391 struct strbuf
*previous_name
)
1393 struct cache_entry
*ce
;
1398 /* On-disk flags are just 16 bits */
1399 flags
= ntoh_s(ondisk
->flags
);
1400 len
= flags
& CE_NAMEMASK
;
1402 if (flags
& CE_EXTENDED
) {
1403 struct ondisk_cache_entry_extended
*ondisk2
;
1405 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
1406 extended_flags
= ntoh_s(ondisk2
->flags2
) << 16;
1407 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1408 if (extended_flags
& ~CE_EXTENDED_FLAGS
)
1409 die("Unknown index entry format %08x", extended_flags
);
1410 flags
|= extended_flags
;
1411 name
= ondisk2
->name
;
1414 name
= ondisk
->name
;
1416 if (!previous_name
) {
1417 /* v3 and earlier */
1418 if (len
== CE_NAMEMASK
)
1420 ce
= cache_entry_from_ondisk(ondisk
, flags
, name
, len
);
1422 *ent_size
= ondisk_ce_size(ce
);
1424 unsigned long consumed
;
1425 consumed
= expand_name_field(previous_name
, name
);
1426 ce
= cache_entry_from_ondisk(ondisk
, flags
,
1428 previous_name
->len
);
1430 *ent_size
= (name
- ((char *)ondisk
)) + consumed
;
1435 /* remember to discard_cache() before reading a different cache! */
1436 int read_index_from(struct index_state
*istate
, const char *path
)
1440 unsigned long src_offset
;
1441 struct cache_header
*hdr
;
1444 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
1446 if (istate
->initialized
)
1447 return istate
->cache_nr
;
1449 istate
->timestamp
.sec
= 0;
1450 istate
->timestamp
.nsec
= 0;
1451 fd
= open(path
, O_RDONLY
);
1453 if (errno
== ENOENT
)
1455 die_errno("index file open failed");
1459 die_errno("cannot stat the open index");
1461 mmap_size
= xsize_t(st
.st_size
);
1462 if (mmap_size
< sizeof(struct cache_header
) + 20)
1463 die("index file smaller than expected");
1465 mmap
= xmmap(NULL
, mmap_size
, PROT_READ
| PROT_WRITE
, MAP_PRIVATE
, fd
, 0);
1466 if (mmap
== MAP_FAILED
)
1467 die_errno("unable to map index file");
1471 if (verify_hdr(hdr
, mmap_size
) < 0)
1474 istate
->version
= ntohl(hdr
->hdr_version
);
1475 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
1476 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
1477 istate
->cache
= xcalloc(istate
->cache_alloc
, sizeof(*istate
->cache
));
1478 istate
->initialized
= 1;
1480 if (istate
->version
== 4)
1481 previous_name
= &previous_name_buf
;
1483 previous_name
= NULL
;
1485 src_offset
= sizeof(*hdr
);
1486 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1487 struct ondisk_cache_entry
*disk_ce
;
1488 struct cache_entry
*ce
;
1489 unsigned long consumed
;
1491 disk_ce
= (struct ondisk_cache_entry
*)((char *)mmap
+ src_offset
);
1492 ce
= create_from_disk(disk_ce
, &consumed
, previous_name
);
1493 set_index_entry(istate
, i
, ce
);
1495 src_offset
+= consumed
;
1497 strbuf_release(&previous_name_buf
);
1498 istate
->timestamp
.sec
= st
.st_mtime
;
1499 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
1501 while (src_offset
<= mmap_size
- 20 - 8) {
1502 /* After an array of active_nr index entries,
1503 * there can be arbitrary number of extended
1504 * sections, each of which is prefixed with
1505 * extension name (4-byte) and section length
1506 * in 4-byte network byte order.
1509 memcpy(&extsize
, (char *)mmap
+ src_offset
+ 4, 4);
1510 extsize
= ntohl(extsize
);
1511 if (read_index_extension(istate
,
1512 (const char *) mmap
+ src_offset
,
1513 (char *) mmap
+ src_offset
+ 8,
1517 src_offset
+= extsize
;
1519 munmap(mmap
, mmap_size
);
1520 return istate
->cache_nr
;
1523 munmap(mmap
, mmap_size
);
1524 die("index file corrupt");
1527 int is_index_unborn(struct index_state
*istate
)
1529 return (!istate
->cache_nr
&& !istate
->timestamp
.sec
);
1532 int discard_index(struct index_state
*istate
)
1536 for (i
= 0; i
< istate
->cache_nr
; i
++)
1537 free(istate
->cache
[i
]);
1538 resolve_undo_clear_index(istate
);
1539 istate
->cache_nr
= 0;
1540 istate
->cache_changed
= 0;
1541 istate
->timestamp
.sec
= 0;
1542 istate
->timestamp
.nsec
= 0;
1543 free_name_hash(istate
);
1544 cache_tree_free(&(istate
->cache_tree
));
1545 istate
->initialized
= 0;
1546 free(istate
->cache
);
1547 istate
->cache
= NULL
;
1548 istate
->cache_alloc
= 0;
1552 int unmerged_index(const struct index_state
*istate
)
1555 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1556 if (ce_stage(istate
->cache
[i
]))
1562 #define WRITE_BUFFER_SIZE 8192
1563 static unsigned char write_buffer
[WRITE_BUFFER_SIZE
];
1564 static unsigned long write_buffer_len
;
1566 static int ce_write_flush(git_SHA_CTX
*context
, int fd
)
1568 unsigned int buffered
= write_buffer_len
;
1570 git_SHA1_Update(context
, write_buffer
, buffered
);
1571 if (write_in_full(fd
, write_buffer
, buffered
) != buffered
)
1573 write_buffer_len
= 0;
1578 static int ce_write(git_SHA_CTX
*context
, int fd
, void *data
, unsigned int len
)
1581 unsigned int buffered
= write_buffer_len
;
1582 unsigned int partial
= WRITE_BUFFER_SIZE
- buffered
;
1585 memcpy(write_buffer
+ buffered
, data
, partial
);
1586 buffered
+= partial
;
1587 if (buffered
== WRITE_BUFFER_SIZE
) {
1588 write_buffer_len
= buffered
;
1589 if (ce_write_flush(context
, fd
))
1593 write_buffer_len
= buffered
;
1595 data
= (char *) data
+ partial
;
1600 static int write_index_ext_header(git_SHA_CTX
*context
, int fd
,
1601 unsigned int ext
, unsigned int sz
)
1605 return ((ce_write(context
, fd
, &ext
, 4) < 0) ||
1606 (ce_write(context
, fd
, &sz
, 4) < 0)) ? -1 : 0;
1609 static int ce_flush(git_SHA_CTX
*context
, int fd
)
1611 unsigned int left
= write_buffer_len
;
1614 write_buffer_len
= 0;
1615 git_SHA1_Update(context
, write_buffer
, left
);
1618 /* Flush first if not enough space for SHA1 signature */
1619 if (left
+ 20 > WRITE_BUFFER_SIZE
) {
1620 if (write_in_full(fd
, write_buffer
, left
) != left
)
1625 /* Append the SHA1 signature at the end */
1626 git_SHA1_Final(write_buffer
+ left
, context
);
1628 return (write_in_full(fd
, write_buffer
, left
) != left
) ? -1 : 0;
1631 static void ce_smudge_racily_clean_entry(struct cache_entry
*ce
)
1634 * The only thing we care about in this function is to smudge the
1635 * falsely clean entry due to touch-update-touch race, so we leave
1636 * everything else as they are. We are called for entries whose
1637 * ce_stat_data.sd_mtime match the index file mtime.
1639 * Note that this actually does not do much for gitlinks, for
1640 * which ce_match_stat_basic() always goes to the actual
1641 * contents. The caller checks with is_racy_timestamp() which
1642 * always says "no" for gitlinks, so we are not called for them ;-)
1646 if (lstat(ce
->name
, &st
) < 0)
1648 if (ce_match_stat_basic(ce
, &st
))
1650 if (ce_modified_check_fs(ce
, &st
)) {
1651 /* This is "racily clean"; smudge it. Note that this
1652 * is a tricky code. At first glance, it may appear
1653 * that it can break with this sequence:
1655 * $ echo xyzzy >frotz
1656 * $ git-update-index --add frotz
1659 * $ echo filfre >nitfol
1660 * $ git-update-index --add nitfol
1662 * but it does not. When the second update-index runs,
1663 * it notices that the entry "frotz" has the same timestamp
1664 * as index, and if we were to smudge it by resetting its
1665 * size to zero here, then the object name recorded
1666 * in index is the 6-byte file but the cached stat information
1667 * becomes zero --- which would then match what we would
1668 * obtain from the filesystem next time we stat("frotz").
1670 * However, the second update-index, before calling
1671 * this function, notices that the cached size is 6
1672 * bytes and what is on the filesystem is an empty
1673 * file, and never calls us, so the cached size information
1674 * for "frotz" stays 6 which does not match the filesystem.
1676 ce
->ce_stat_data
.sd_size
= 0;
1680 /* Copy miscellaneous fields but not the name */
1681 static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry
*ondisk
,
1682 struct cache_entry
*ce
)
1686 ondisk
->ctime
.sec
= htonl(ce
->ce_stat_data
.sd_ctime
.sec
);
1687 ondisk
->mtime
.sec
= htonl(ce
->ce_stat_data
.sd_mtime
.sec
);
1688 ondisk
->ctime
.nsec
= htonl(ce
->ce_stat_data
.sd_ctime
.nsec
);
1689 ondisk
->mtime
.nsec
= htonl(ce
->ce_stat_data
.sd_mtime
.nsec
);
1690 ondisk
->dev
= htonl(ce
->ce_stat_data
.sd_dev
);
1691 ondisk
->ino
= htonl(ce
->ce_stat_data
.sd_ino
);
1692 ondisk
->mode
= htonl(ce
->ce_mode
);
1693 ondisk
->uid
= htonl(ce
->ce_stat_data
.sd_uid
);
1694 ondisk
->gid
= htonl(ce
->ce_stat_data
.sd_gid
);
1695 ondisk
->size
= htonl(ce
->ce_stat_data
.sd_size
);
1696 hashcpy(ondisk
->sha1
, ce
->sha1
);
1698 flags
= ce
->ce_flags
;
1699 flags
|= (ce_namelen(ce
) >= CE_NAMEMASK
? CE_NAMEMASK
: ce_namelen(ce
));
1700 ondisk
->flags
= htons(flags
);
1701 if (ce
->ce_flags
& CE_EXTENDED
) {
1702 struct ondisk_cache_entry_extended
*ondisk2
;
1703 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
1704 ondisk2
->flags2
= htons((ce
->ce_flags
& CE_EXTENDED_FLAGS
) >> 16);
1705 return ondisk2
->name
;
1708 return ondisk
->name
;
1712 static int ce_write_entry(git_SHA_CTX
*c
, int fd
, struct cache_entry
*ce
,
1713 struct strbuf
*previous_name
)
1716 struct ondisk_cache_entry
*ondisk
;
1720 if (!previous_name
) {
1721 size
= ondisk_ce_size(ce
);
1722 ondisk
= xcalloc(1, size
);
1723 name
= copy_cache_entry_to_ondisk(ondisk
, ce
);
1724 memcpy(name
, ce
->name
, ce_namelen(ce
));
1726 int common
, to_remove
, prefix_size
;
1727 unsigned char to_remove_vi
[16];
1729 (ce
->name
[common
] &&
1730 common
< previous_name
->len
&&
1731 ce
->name
[common
] == previous_name
->buf
[common
]);
1733 ; /* still matching */
1734 to_remove
= previous_name
->len
- common
;
1735 prefix_size
= encode_varint(to_remove
, to_remove_vi
);
1737 if (ce
->ce_flags
& CE_EXTENDED
)
1738 size
= offsetof(struct ondisk_cache_entry_extended
, name
);
1740 size
= offsetof(struct ondisk_cache_entry
, name
);
1741 size
+= prefix_size
+ (ce_namelen(ce
) - common
+ 1);
1743 ondisk
= xcalloc(1, size
);
1744 name
= copy_cache_entry_to_ondisk(ondisk
, ce
);
1745 memcpy(name
, to_remove_vi
, prefix_size
);
1746 memcpy(name
+ prefix_size
, ce
->name
+ common
, ce_namelen(ce
) - common
);
1748 strbuf_splice(previous_name
, common
, to_remove
,
1749 ce
->name
+ common
, ce_namelen(ce
) - common
);
1752 result
= ce_write(c
, fd
, ondisk
, size
);
1757 static int has_racy_timestamp(struct index_state
*istate
)
1759 int entries
= istate
->cache_nr
;
1762 for (i
= 0; i
< entries
; i
++) {
1763 struct cache_entry
*ce
= istate
->cache
[i
];
1764 if (is_racy_timestamp(istate
, ce
))
1771 * Opportunistically update the index but do not complain if we can't
1773 void update_index_if_able(struct index_state
*istate
, struct lock_file
*lockfile
)
1775 if ((istate
->cache_changed
|| has_racy_timestamp(istate
)) &&
1776 !write_index(istate
, lockfile
->fd
))
1777 commit_locked_index(lockfile
);
1779 rollback_lock_file(lockfile
);
1782 int write_index(struct index_state
*istate
, int newfd
)
1785 struct cache_header hdr
;
1786 int i
, err
, removed
, extended
, hdr_version
;
1787 struct cache_entry
**cache
= istate
->cache
;
1788 int entries
= istate
->cache_nr
;
1790 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
1792 for (i
= removed
= extended
= 0; i
< entries
; i
++) {
1793 if (cache
[i
]->ce_flags
& CE_REMOVE
)
1796 /* reduce extended entries if possible */
1797 cache
[i
]->ce_flags
&= ~CE_EXTENDED
;
1798 if (cache
[i
]->ce_flags
& CE_EXTENDED_FLAGS
) {
1800 cache
[i
]->ce_flags
|= CE_EXTENDED
;
1804 if (!istate
->version
)
1805 istate
->version
= INDEX_FORMAT_DEFAULT
;
1807 /* demote version 3 to version 2 when the latter suffices */
1808 if (istate
->version
== 3 || istate
->version
== 2)
1809 istate
->version
= extended
? 3 : 2;
1811 hdr_version
= istate
->version
;
1813 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
1814 hdr
.hdr_version
= htonl(hdr_version
);
1815 hdr
.hdr_entries
= htonl(entries
- removed
);
1818 if (ce_write(&c
, newfd
, &hdr
, sizeof(hdr
)) < 0)
1821 previous_name
= (hdr_version
== 4) ? &previous_name_buf
: NULL
;
1822 for (i
= 0; i
< entries
; i
++) {
1823 struct cache_entry
*ce
= cache
[i
];
1824 if (ce
->ce_flags
& CE_REMOVE
)
1826 if (!ce_uptodate(ce
) && is_racy_timestamp(istate
, ce
))
1827 ce_smudge_racily_clean_entry(ce
);
1828 if (is_null_sha1(ce
->sha1
)) {
1829 static const char msg
[] = "cache entry has null sha1: %s";
1830 static int allow
= -1;
1833 allow
= git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
1835 warning(msg
, ce
->name
);
1837 return error(msg
, ce
->name
);
1839 if (ce_write_entry(&c
, newfd
, ce
, previous_name
) < 0)
1842 strbuf_release(&previous_name_buf
);
1844 /* Write extension data here */
1845 if (istate
->cache_tree
) {
1846 struct strbuf sb
= STRBUF_INIT
;
1848 cache_tree_write(&sb
, istate
->cache_tree
);
1849 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_TREE
, sb
.len
) < 0
1850 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
1851 strbuf_release(&sb
);
1855 if (istate
->resolve_undo
) {
1856 struct strbuf sb
= STRBUF_INIT
;
1858 resolve_undo_write(&sb
, istate
->resolve_undo
);
1859 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_RESOLVE_UNDO
,
1861 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
1862 strbuf_release(&sb
);
1867 if (ce_flush(&c
, newfd
) || fstat(newfd
, &st
))
1869 istate
->timestamp
.sec
= (unsigned int)st
.st_mtime
;
1870 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
1875 * Read the index file that is potentially unmerged into given
1876 * index_state, dropping any unmerged entries. Returns true if
1877 * the index is unmerged. Callers who want to refuse to work
1878 * from an unmerged state can call this and check its return value,
1879 * instead of calling read_cache().
1881 int read_index_unmerged(struct index_state
*istate
)
1887 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1888 struct cache_entry
*ce
= istate
->cache
[i
];
1889 struct cache_entry
*new_ce
;
1895 len
= ce_namelen(ce
);
1896 size
= cache_entry_size(len
);
1897 new_ce
= xcalloc(1, size
);
1898 memcpy(new_ce
->name
, ce
->name
, len
);
1899 new_ce
->ce_flags
= create_ce_flags(0) | CE_CONFLICTED
;
1900 new_ce
->ce_namelen
= len
;
1901 new_ce
->ce_mode
= ce
->ce_mode
;
1902 if (add_index_entry(istate
, new_ce
, 0))
1903 return error("%s: cannot drop to stage #0",
1905 i
= index_name_pos(istate
, new_ce
->name
, len
);
1911 * Returns 1 if the path is an "other" path with respect to
1912 * the index; that is, the path is not mentioned in the index at all,
1913 * either as a file, a directory with some files in the index,
1914 * or as an unmerged entry.
1916 * We helpfully remove a trailing "/" from directories so that
1917 * the output of read_directory can be used as-is.
1919 int index_name_is_other(const struct index_state
*istate
, const char *name
,
1923 if (namelen
&& name
[namelen
- 1] == '/')
1925 pos
= index_name_pos(istate
, name
, namelen
);
1927 return 0; /* exact match */
1929 if (pos
< istate
->cache_nr
) {
1930 struct cache_entry
*ce
= istate
->cache
[pos
];
1931 if (ce_namelen(ce
) == namelen
&&
1932 !memcmp(ce
->name
, name
, namelen
))
1933 return 0; /* Yup, this one exists unmerged */
1938 void *read_blob_data_from_index(struct index_state
*istate
, const char *path
, unsigned long *size
)
1942 enum object_type type
;
1946 pos
= index_name_pos(istate
, path
, len
);
1949 * We might be in the middle of a merge, in which
1950 * case we would read stage #2 (ours).
1954 (pos
< 0 && i
< istate
->cache_nr
&&
1955 !strcmp(istate
->cache
[i
]->name
, path
));
1957 if (ce_stage(istate
->cache
[i
]) == 2)
1962 data
= read_sha1_file(istate
->cache
[pos
]->sha1
, &type
, &sz
);
1963 if (!data
|| type
!= OBJ_BLOB
) {
1972 void stat_validity_clear(struct stat_validity
*sv
)
1978 int stat_validity_check(struct stat_validity
*sv
, const char *path
)
1982 if (stat(path
, &st
) < 0)
1983 return sv
->sd
== NULL
;
1986 return S_ISREG(st
.st_mode
) && !match_stat_data(sv
->sd
, &st
);
1989 void stat_validity_update(struct stat_validity
*sv
, int fd
)
1993 if (fstat(fd
, &st
) < 0 || !S_ISREG(st
.st_mode
))
1994 stat_validity_clear(sv
);
1997 sv
->sd
= xcalloc(1, sizeof(struct stat_data
));
1998 fill_stat_data(sv
->sd
, &st
);