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"
20 * The first letter should be 'A'..'Z' for extensions that are not
21 * necessary for a correct operation (i.e. optimization data).
22 * When new extensions are added that _needs_ to be understood in
23 * order to correctly interpret the index file, pick character that
24 * is outside the range, to cause the reader to abort.
27 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
28 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
30 struct index_state the_index
;
32 static void set_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
34 istate
->cache
[nr
] = ce
;
35 add_name_hash(istate
, ce
);
38 static void replace_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
40 struct cache_entry
*old
= istate
->cache
[nr
];
42 remove_name_hash(old
);
43 set_index_entry(istate
, nr
, ce
);
44 istate
->cache_changed
= 1;
47 void rename_index_entry_at(struct index_state
*istate
, int nr
, const char *new_name
)
49 struct cache_entry
*old
= istate
->cache
[nr
], *new;
50 int namelen
= strlen(new_name
);
52 new = xmalloc(cache_entry_size(namelen
));
53 copy_cache_entry(new, old
);
54 new->ce_flags
&= ~(CE_STATE_MASK
| CE_NAMEMASK
);
55 new->ce_flags
|= (namelen
>= CE_NAMEMASK
? CE_NAMEMASK
: namelen
);
56 memcpy(new->name
, new_name
, namelen
+ 1);
58 cache_tree_invalidate_path(istate
->cache_tree
, old
->name
);
59 remove_index_entry_at(istate
, nr
);
60 add_index_entry(istate
, new, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
64 * This only updates the "non-critical" parts of the directory
65 * cache, ie the parts that aren't tracked by GIT, and only used
66 * to validate the cache.
68 void fill_stat_cache_info(struct cache_entry
*ce
, struct stat
*st
)
70 ce
->ce_ctime
= st
->st_ctime
;
71 ce
->ce_mtime
= st
->st_mtime
;
72 ce
->ce_dev
= st
->st_dev
;
73 ce
->ce_ino
= st
->st_ino
;
74 ce
->ce_uid
= st
->st_uid
;
75 ce
->ce_gid
= st
->st_gid
;
76 ce
->ce_size
= st
->st_size
;
79 ce
->ce_flags
|= CE_VALID
;
81 if (S_ISREG(st
->st_mode
))
85 static int ce_compare_data(struct cache_entry
*ce
, struct stat
*st
)
88 int fd
= open(ce
->name
, O_RDONLY
);
91 unsigned char sha1
[20];
92 if (!index_fd(sha1
, fd
, st
, 0, OBJ_BLOB
, ce
->name
))
93 match
= hashcmp(sha1
, ce
->sha1
);
94 /* index_fd() closed the file descriptor already */
99 static int ce_compare_link(struct cache_entry
*ce
, size_t expected_size
)
105 enum object_type type
;
108 target
= xmalloc(expected_size
);
109 len
= readlink(ce
->name
, target
, expected_size
);
110 if (len
!= expected_size
) {
114 buffer
= read_sha1_file(ce
->sha1
, &type
, &size
);
119 if (size
== expected_size
)
120 match
= memcmp(buffer
, target
, size
);
126 static int ce_compare_gitlink(struct cache_entry
*ce
)
128 unsigned char sha1
[20];
131 * We don't actually require that the .git directory
132 * under GITLINK directory be a valid git directory. It
133 * might even be missing (in case nobody populated that
136 * If so, we consider it always to match.
138 if (resolve_gitlink_ref(ce
->name
, "HEAD", sha1
) < 0)
140 return hashcmp(sha1
, ce
->sha1
);
143 static int ce_modified_check_fs(struct cache_entry
*ce
, struct stat
*st
)
145 switch (st
->st_mode
& S_IFMT
) {
147 if (ce_compare_data(ce
, st
))
151 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
155 if (S_ISGITLINK(ce
->ce_mode
))
156 return ce_compare_gitlink(ce
) ? DATA_CHANGED
: 0;
163 int is_empty_blob_sha1(const unsigned char *sha1
)
165 static const unsigned char empty_blob_sha1
[20] = {
166 0xe6,0x9d,0xe2,0x9b,0xb2,0xd1,0xd6,0x43,0x4b,0x8b,
167 0x29,0xae,0x77,0x5a,0xd8,0xc2,0xe4,0x8c,0x53,0x91
170 return !hashcmp(sha1
, empty_blob_sha1
);
173 static int ce_match_stat_basic(struct cache_entry
*ce
, struct stat
*st
)
175 unsigned int changed
= 0;
177 if (ce
->ce_flags
& CE_REMOVE
)
178 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
180 switch (ce
->ce_mode
& S_IFMT
) {
182 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
183 /* We consider only the owner x bit to be relevant for
186 if (trust_executable_bit
&&
187 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
188 changed
|= MODE_CHANGED
;
191 if (!S_ISLNK(st
->st_mode
) &&
192 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
193 changed
|= TYPE_CHANGED
;
196 /* We ignore most of the st_xxx fields for gitlinks */
197 if (!S_ISDIR(st
->st_mode
))
198 changed
|= TYPE_CHANGED
;
199 else if (ce_compare_gitlink(ce
))
200 changed
|= DATA_CHANGED
;
203 die("internal error: ce_mode is %o", ce
->ce_mode
);
205 if (ce
->ce_mtime
!= (unsigned int) st
->st_mtime
)
206 changed
|= MTIME_CHANGED
;
207 if (trust_ctime
&& ce
->ce_ctime
!= (unsigned int) st
->st_ctime
)
208 changed
|= CTIME_CHANGED
;
210 if (ce
->ce_uid
!= (unsigned int) st
->st_uid
||
211 ce
->ce_gid
!= (unsigned int) st
->st_gid
)
212 changed
|= OWNER_CHANGED
;
213 if (ce
->ce_ino
!= (unsigned int) st
->st_ino
)
214 changed
|= INODE_CHANGED
;
218 * st_dev breaks on network filesystems where different
219 * clients will have different views of what "device"
220 * the filesystem is on
222 if (ce
->ce_dev
!= (unsigned int) st
->st_dev
)
223 changed
|= INODE_CHANGED
;
226 if (ce
->ce_size
!= (unsigned int) st
->st_size
)
227 changed
|= DATA_CHANGED
;
229 /* Racily smudged entry? */
231 if (!is_empty_blob_sha1(ce
->sha1
))
232 changed
|= DATA_CHANGED
;
238 static int is_racy_timestamp(const struct index_state
*istate
, struct cache_entry
*ce
)
240 return (!S_ISGITLINK(ce
->ce_mode
) &&
242 ((unsigned int)istate
->timestamp
) <= ce
->ce_mtime
);
245 int ie_match_stat(const struct index_state
*istate
,
246 struct cache_entry
*ce
, struct stat
*st
,
247 unsigned int options
)
249 unsigned int changed
;
250 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
251 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
254 * If it's marked as always valid in the index, it's
255 * valid whatever the checked-out copy says.
257 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
261 * Intent-to-add entries have not been added, so the index entry
262 * by definition never matches what is in the work tree until it
263 * actually gets added.
265 if (ce
->ce_flags
& CE_INTENT_TO_ADD
)
266 return DATA_CHANGED
| TYPE_CHANGED
| MODE_CHANGED
;
268 changed
= ce_match_stat_basic(ce
, st
);
271 * Within 1 second of this sequence:
272 * echo xyzzy >file && git-update-index --add file
273 * running this command:
275 * would give a falsely clean cache entry. The mtime and
276 * length match the cache, and other stat fields do not change.
278 * We could detect this at update-index time (the cache entry
279 * being registered/updated records the same time as "now")
280 * and delay the return from git-update-index, but that would
281 * effectively mean we can make at most one commit per second,
282 * which is not acceptable. Instead, we check cache entries
283 * whose mtime are the same as the index file timestamp more
284 * carefully than others.
286 if (!changed
&& is_racy_timestamp(istate
, ce
)) {
287 if (assume_racy_is_modified
)
288 changed
|= DATA_CHANGED
;
290 changed
|= ce_modified_check_fs(ce
, st
);
296 int ie_modified(const struct index_state
*istate
,
297 struct cache_entry
*ce
, struct stat
*st
, unsigned int options
)
299 int changed
, changed_fs
;
301 changed
= ie_match_stat(istate
, ce
, st
, options
);
305 * If the mode or type has changed, there's no point in trying
306 * to refresh the entry - it's not going to match
308 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
312 * Immediately after read-tree or update-index --cacheinfo,
313 * the length field is zero, as we have never even read the
314 * lstat(2) information once, and we cannot trust DATA_CHANGED
315 * returned by ie_match_stat() which in turn was returned by
316 * ce_match_stat_basic() to signal that the filesize of the
317 * blob changed. We have to actually go to the filesystem to
318 * see if the contents match, and if so, should answer "unchanged".
320 * The logic does not apply to gitlinks, as ce_match_stat_basic()
321 * already has checked the actual HEAD from the filesystem in the
322 * subproject. If ie_match_stat() already said it is different,
323 * then we know it is.
325 if ((changed
& DATA_CHANGED
) &&
326 (S_ISGITLINK(ce
->ce_mode
) || ce
->ce_size
!= 0))
329 changed_fs
= ce_modified_check_fs(ce
, st
);
331 return changed
| changed_fs
;
335 int base_name_compare(const char *name1
, int len1
, int mode1
,
336 const char *name2
, int len2
, int mode2
)
338 unsigned char c1
, c2
;
339 int len
= len1
< len2
? len1
: len2
;
342 cmp
= memcmp(name1
, name2
, len
);
347 if (!c1
&& S_ISDIR(mode1
))
349 if (!c2
&& S_ISDIR(mode2
))
351 return (c1
< c2
) ? -1 : (c1
> c2
) ? 1 : 0;
355 * df_name_compare() is identical to base_name_compare(), except it
356 * compares conflicting directory/file entries as equal. Note that
357 * while a directory name compares as equal to a regular file, they
358 * then individually compare _differently_ to a filename that has
359 * a dot after the basename (because '\0' < '.' < '/').
361 * This is used by routines that want to traverse the git namespace
362 * but then handle conflicting entries together when possible.
364 int df_name_compare(const char *name1
, int len1
, int mode1
,
365 const char *name2
, int len2
, int mode2
)
367 int len
= len1
< len2
? len1
: len2
, cmp
;
368 unsigned char c1
, c2
;
370 cmp
= memcmp(name1
, name2
, len
);
373 /* Directories and files compare equal (same length, same name) */
377 if (!c1
&& S_ISDIR(mode1
))
380 if (!c2
&& S_ISDIR(mode2
))
382 if (c1
== '/' && !c2
)
384 if (c2
== '/' && !c1
)
389 int cache_name_compare(const char *name1
, int flags1
, const char *name2
, int flags2
)
391 int len1
= flags1
& CE_NAMEMASK
;
392 int len2
= flags2
& CE_NAMEMASK
;
393 int len
= len1
< len2
? len1
: len2
;
396 cmp
= memcmp(name1
, name2
, len
);
405 flags1
&= CE_STAGEMASK
;
406 flags2
&= CE_STAGEMASK
;
415 int index_name_pos(const struct index_state
*istate
, const char *name
, int namelen
)
420 last
= istate
->cache_nr
;
421 while (last
> first
) {
422 int next
= (last
+ first
) >> 1;
423 struct cache_entry
*ce
= istate
->cache
[next
];
424 int cmp
= cache_name_compare(name
, namelen
, ce
->name
, ce
->ce_flags
);
436 /* Remove entry, return true if there are more entries to go.. */
437 int remove_index_entry_at(struct index_state
*istate
, int pos
)
439 struct cache_entry
*ce
= istate
->cache
[pos
];
441 remove_name_hash(ce
);
442 istate
->cache_changed
= 1;
444 if (pos
>= istate
->cache_nr
)
446 memmove(istate
->cache
+ pos
,
447 istate
->cache
+ pos
+ 1,
448 (istate
->cache_nr
- pos
) * sizeof(struct cache_entry
*));
452 int remove_file_from_index(struct index_state
*istate
, const char *path
)
454 int pos
= index_name_pos(istate
, path
, strlen(path
));
457 cache_tree_invalidate_path(istate
->cache_tree
, path
);
458 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
459 remove_index_entry_at(istate
, pos
);
463 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
465 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
468 static int index_name_pos_also_unmerged(struct index_state
*istate
,
469 const char *path
, int namelen
)
471 int pos
= index_name_pos(istate
, path
, namelen
);
472 struct cache_entry
*ce
;
477 /* maybe unmerged? */
479 if (pos
>= istate
->cache_nr
||
480 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
483 /* order of preference: stage 2, 1, 3 */
484 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
485 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
486 !compare_name(ce
, path
, namelen
))
491 static int different_name(struct cache_entry
*ce
, struct cache_entry
*alias
)
493 int len
= ce_namelen(ce
);
494 return ce_namelen(alias
) != len
|| memcmp(ce
->name
, alias
->name
, len
);
498 * If we add a filename that aliases in the cache, we will use the
499 * name that we already have - but we don't want to update the same
500 * alias twice, because that implies that there were actually two
501 * different files with aliasing names!
503 * So we use the CE_ADDED flag to verify that the alias was an old
504 * one before we accept it as
506 static struct cache_entry
*create_alias_ce(struct cache_entry
*ce
, struct cache_entry
*alias
)
509 struct cache_entry
*new;
511 if (alias
->ce_flags
& CE_ADDED
)
512 die("Will not add file alias '%s' ('%s' already exists in index)", ce
->name
, alias
->name
);
514 /* Ok, create the new entry using the name of the existing alias */
515 len
= ce_namelen(alias
);
516 new = xcalloc(1, cache_entry_size(len
));
517 memcpy(new->name
, alias
->name
, len
);
518 copy_cache_entry(new, ce
);
523 static void record_intent_to_add(struct cache_entry
*ce
)
525 unsigned char sha1
[20];
526 if (write_sha1_file("", 0, blob_type
, sha1
))
527 die("cannot create an empty blob in the object database");
528 hashcpy(ce
->sha1
, sha1
);
531 int add_to_index(struct index_state
*istate
, const char *path
, struct stat
*st
, int flags
)
533 int size
, namelen
, was_same
;
534 mode_t st_mode
= st
->st_mode
;
535 struct cache_entry
*ce
, *alias
;
536 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_RACY_IS_DIRTY
;
537 int verbose
= flags
& (ADD_CACHE_VERBOSE
| ADD_CACHE_PRETEND
);
538 int pretend
= flags
& ADD_CACHE_PRETEND
;
539 int intent_only
= flags
& ADD_CACHE_INTENT
;
540 int add_option
= (ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
|
541 (intent_only
? ADD_CACHE_NEW_ONLY
: 0));
543 if (!S_ISREG(st_mode
) && !S_ISLNK(st_mode
) && !S_ISDIR(st_mode
))
544 return error("%s: can only add regular files, symbolic links or git-directories", path
);
546 namelen
= strlen(path
);
547 if (S_ISDIR(st_mode
)) {
548 while (namelen
&& path
[namelen
-1] == '/')
551 size
= cache_entry_size(namelen
);
552 ce
= xcalloc(1, size
);
553 memcpy(ce
->name
, path
, namelen
);
554 ce
->ce_flags
= namelen
;
556 fill_stat_cache_info(ce
, st
);
558 ce
->ce_flags
|= CE_INTENT_TO_ADD
;
560 if (trust_executable_bit
&& has_symlinks
)
561 ce
->ce_mode
= create_ce_mode(st_mode
);
563 /* If there is an existing entry, pick the mode bits and type
564 * from it, otherwise assume unexecutable regular file.
566 struct cache_entry
*ent
;
567 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
569 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
570 ce
->ce_mode
= ce_mode_from_stat(ent
, st_mode
);
573 alias
= index_name_exists(istate
, ce
->name
, ce_namelen(ce
), ignore_case
);
574 if (alias
&& !ce_stage(alias
) && !ie_match_stat(istate
, alias
, st
, ce_option
)) {
575 /* Nothing changed, really */
577 ce_mark_uptodate(alias
);
578 alias
->ce_flags
|= CE_ADDED
;
582 if (index_path(ce
->sha1
, path
, st
, 1))
583 return error("unable to index file %s", path
);
585 record_intent_to_add(ce
);
587 if (ignore_case
&& alias
&& different_name(ce
, alias
))
588 ce
= create_alias_ce(ce
, alias
);
589 ce
->ce_flags
|= CE_ADDED
;
591 /* It was suspected to be racily clean, but it turns out to be Ok */
594 !hashcmp(alias
->sha1
, ce
->sha1
) &&
595 ce
->ce_mode
== alias
->ce_mode
);
599 else if (add_index_entry(istate
, ce
, add_option
))
600 return error("unable to add %s to index",path
);
601 if (verbose
&& !was_same
)
602 printf("add '%s'\n", path
);
606 int add_file_to_index(struct index_state
*istate
, const char *path
, int flags
)
609 if (lstat(path
, &st
))
610 die("%s: unable to stat (%s)", path
, strerror(errno
));
611 return add_to_index(istate
, path
, &st
, flags
);
614 struct cache_entry
*make_cache_entry(unsigned int mode
,
615 const unsigned char *sha1
, const char *path
, int stage
,
619 struct cache_entry
*ce
;
621 if (!verify_path(path
)) {
622 error("Invalid path '%s'", path
);
627 size
= cache_entry_size(len
);
628 ce
= xcalloc(1, size
);
630 hashcpy(ce
->sha1
, sha1
);
631 memcpy(ce
->name
, path
, len
);
632 ce
->ce_flags
= create_ce_flags(len
, stage
);
633 ce
->ce_mode
= create_ce_mode(mode
);
636 return refresh_cache_entry(ce
, 0);
641 int ce_same_name(struct cache_entry
*a
, struct cache_entry
*b
)
643 int len
= ce_namelen(a
);
644 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
647 int ce_path_match(const struct cache_entry
*ce
, const char **pathspec
)
649 const char *match
, *name
;
655 len
= ce_namelen(ce
);
657 while ((match
= *pathspec
++) != NULL
) {
658 int matchlen
= strlen(match
);
661 if (memcmp(name
, match
, matchlen
))
663 if (matchlen
&& name
[matchlen
-1] == '/')
665 if (name
[matchlen
] == '/' || !name
[matchlen
])
674 * We fundamentally don't like some paths: we don't want
675 * dot or dot-dot anywhere, and for obvious reasons don't
676 * want to recurse into ".git" either.
678 * Also, we don't want double slashes or slashes at the
679 * end that can make pathnames ambiguous.
681 static int verify_dotfile(const char *rest
)
684 * The first character was '.', but that
685 * has already been discarded, we now test
689 /* "." is not allowed */
694 * ".git" followed by NUL or slash is bad. This
695 * shares the path end test with the ".." case.
705 if (rest
[1] == '\0' || rest
[1] == '/')
711 int verify_path(const char *path
)
728 if (verify_dotfile(path
))
738 * Do we have another file that has the beginning components being a
739 * proper superset of the name we're trying to add?
741 static int has_file_name(struct index_state
*istate
,
742 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
745 int len
= ce_namelen(ce
);
746 int stage
= ce_stage(ce
);
747 const char *name
= ce
->name
;
749 while (pos
< istate
->cache_nr
) {
750 struct cache_entry
*p
= istate
->cache
[pos
++];
752 if (len
>= ce_namelen(p
))
754 if (memcmp(name
, p
->name
, len
))
756 if (ce_stage(p
) != stage
)
758 if (p
->name
[len
] != '/')
760 if (p
->ce_flags
& CE_REMOVE
)
765 remove_index_entry_at(istate
, --pos
);
771 * Do we have another file with a pathname that is a proper
772 * subset of the name we're trying to add?
774 static int has_dir_name(struct index_state
*istate
,
775 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
778 int stage
= ce_stage(ce
);
779 const char *name
= ce
->name
;
780 const char *slash
= name
+ ce_namelen(ce
);
788 if (slash
<= ce
->name
)
793 pos
= index_name_pos(istate
, name
, create_ce_flags(len
, stage
));
796 * Found one, but not so fast. This could
797 * be a marker that says "I was here, but
798 * I am being removed". Such an entry is
799 * not a part of the resulting tree, and
800 * it is Ok to have a directory at the same
803 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
807 remove_index_entry_at(istate
, pos
);
815 * Trivial optimization: if we find an entry that
816 * already matches the sub-directory, then we know
817 * we're ok, and we can exit.
819 while (pos
< istate
->cache_nr
) {
820 struct cache_entry
*p
= istate
->cache
[pos
];
821 if ((ce_namelen(p
) <= len
) ||
822 (p
->name
[len
] != '/') ||
823 memcmp(p
->name
, name
, len
))
824 break; /* not our subdirectory */
825 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
827 * p is at the same stage as our entry, and
828 * is a subdirectory of what we are looking
829 * at, so we cannot have conflicts at our
830 * level or anything shorter.
839 /* We may be in a situation where we already have path/file and path
840 * is being added, or we already have path and path/file is being
841 * added. Either one would result in a nonsense tree that has path
842 * twice when git-write-tree tries to write it out. Prevent it.
844 * If ok-to-replace is specified, we remove the conflicting entries
845 * from the cache so the caller should recompute the insert position.
846 * When this happens, we return non-zero.
848 static int check_file_directory_conflict(struct index_state
*istate
,
849 const struct cache_entry
*ce
,
850 int pos
, int ok_to_replace
)
855 * When ce is an "I am going away" entry, we allow it to be added
857 if (ce
->ce_flags
& CE_REMOVE
)
861 * We check if the path is a sub-path of a subsequent pathname
862 * first, since removing those will not change the position
865 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
868 * Then check if the path might have a clashing sub-directory
871 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
874 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
877 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
878 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
879 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
880 int new_only
= option
& ADD_CACHE_NEW_ONLY
;
882 cache_tree_invalidate_path(istate
->cache_tree
, ce
->name
);
883 pos
= index_name_pos(istate
, ce
->name
, ce
->ce_flags
);
885 /* existing match? Just replace it. */
888 replace_index_entry(istate
, pos
, ce
);
894 * Inserting a merged entry ("stage 0") into the index
895 * will always replace all non-merged entries..
897 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
898 while (ce_same_name(istate
->cache
[pos
], ce
)) {
900 if (!remove_index_entry_at(istate
, pos
))
907 if (!verify_path(ce
->name
))
908 return error("Invalid path '%s'", ce
->name
);
910 if (!skip_df_check
&&
911 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
913 return error("'%s' appears as both a file and as a directory",
915 pos
= index_name_pos(istate
, ce
->name
, ce
->ce_flags
);
921 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
925 if (option
& ADD_CACHE_JUST_APPEND
)
926 pos
= istate
->cache_nr
;
929 ret
= add_index_entry_with_check(istate
, ce
, option
);
935 /* Make sure the array is big enough .. */
936 if (istate
->cache_nr
== istate
->cache_alloc
) {
937 istate
->cache_alloc
= alloc_nr(istate
->cache_alloc
);
938 istate
->cache
= xrealloc(istate
->cache
,
939 istate
->cache_alloc
* sizeof(struct cache_entry
*));
944 if (istate
->cache_nr
> pos
+ 1)
945 memmove(istate
->cache
+ pos
+ 1,
947 (istate
->cache_nr
- pos
- 1) * sizeof(ce
));
948 set_index_entry(istate
, pos
, ce
);
949 istate
->cache_changed
= 1;
954 * "refresh" does not calculate a new sha1 file or bring the
955 * cache up-to-date for mode/content changes. But what it
956 * _does_ do is to "re-match" the stat information of a file
957 * with the cache, so that you can refresh the cache for a
958 * file that hasn't been changed but where the stat entry is
961 * For example, you'd want to do this after doing a "git-read-tree",
962 * to link up the stat cache details with the proper files.
964 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
965 struct cache_entry
*ce
,
966 unsigned int options
, int *err
)
969 struct cache_entry
*updated
;
971 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
977 * CE_VALID means the user promised us that the change to
978 * the work tree does not matter and told us not to worry.
980 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
)) {
981 ce_mark_uptodate(ce
);
985 if (lstat(ce
->name
, &st
) < 0) {
991 changed
= ie_match_stat(istate
, ce
, &st
, options
);
994 * The path is unchanged. If we were told to ignore
995 * valid bit, then we did the actual stat check and
996 * found that the entry is unmodified. If the entry
997 * is not marked VALID, this is the place to mark it
998 * valid again, under "assume unchanged" mode.
1000 if (ignore_valid
&& assume_unchanged
&&
1001 !(ce
->ce_flags
& CE_VALID
))
1002 ; /* mark this one VALID again */
1005 * We do not mark the index itself "modified"
1006 * because CE_UPTODATE flag is in-core only;
1007 * we are not going to write this change out.
1009 ce_mark_uptodate(ce
);
1014 if (ie_modified(istate
, ce
, &st
, options
)) {
1021 updated
= xmalloc(size
);
1022 memcpy(updated
, ce
, size
);
1023 fill_stat_cache_info(updated
, &st
);
1025 * If ignore_valid is not set, we should leave CE_VALID bit
1026 * alone. Otherwise, paths marked with --no-assume-unchanged
1027 * (i.e. things to be edited) will reacquire CE_VALID bit
1028 * automatically, which is not really what we want.
1030 if (!ignore_valid
&& assume_unchanged
&&
1031 !(ce
->ce_flags
& CE_VALID
))
1032 updated
->ce_flags
&= ~CE_VALID
;
1037 int refresh_index(struct index_state
*istate
, unsigned int flags
, const char **pathspec
, char *seen
)
1041 int really
= (flags
& REFRESH_REALLY
) != 0;
1042 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
1043 int quiet
= (flags
& REFRESH_QUIET
) != 0;
1044 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
1045 int ignore_submodules
= (flags
& REFRESH_IGNORE_SUBMODULES
) != 0;
1046 unsigned int options
= really
? CE_MATCH_IGNORE_VALID
: 0;
1047 const char *needs_update_message
;
1049 needs_update_message
= ((flags
& REFRESH_SAY_CHANGED
)
1050 ? "locally modified" : "needs update");
1051 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1052 struct cache_entry
*ce
, *new;
1053 int cache_errno
= 0;
1055 ce
= istate
->cache
[i
];
1056 if (ignore_submodules
&& S_ISGITLINK(ce
->ce_mode
))
1060 while ((i
< istate
->cache_nr
) &&
1061 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
1066 printf("%s: needs merge\n", ce
->name
);
1071 if (pathspec
&& !match_pathspec(pathspec
, ce
->name
, strlen(ce
->name
), 0, seen
))
1074 new = refresh_cache_ent(istate
, ce
, options
, &cache_errno
);
1078 if (not_new
&& cache_errno
== ENOENT
)
1080 if (really
&& cache_errno
== EINVAL
) {
1081 /* If we are doing --really-refresh that
1082 * means the index is not valid anymore.
1084 ce
->ce_flags
&= ~CE_VALID
;
1085 istate
->cache_changed
= 1;
1089 printf("%s: %s\n", ce
->name
, needs_update_message
);
1094 replace_index_entry(istate
, i
, new);
1099 struct cache_entry
*refresh_cache_entry(struct cache_entry
*ce
, int really
)
1101 return refresh_cache_ent(&the_index
, ce
, really
, NULL
);
1104 static int verify_hdr(struct cache_header
*hdr
, unsigned long size
)
1107 unsigned char sha1
[20];
1109 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
1110 return error("bad signature");
1111 if (hdr
->hdr_version
!= htonl(2) && hdr
->hdr_version
!= htonl(3))
1112 return error("bad index version");
1114 git_SHA1_Update(&c
, hdr
, size
- 20);
1115 git_SHA1_Final(sha1
, &c
);
1116 if (hashcmp(sha1
, (unsigned char *)hdr
+ size
- 20))
1117 return error("bad index file sha1 signature");
1121 static int read_index_extension(struct index_state
*istate
,
1122 const char *ext
, void *data
, unsigned long sz
)
1124 switch (CACHE_EXT(ext
)) {
1125 case CACHE_EXT_TREE
:
1126 istate
->cache_tree
= cache_tree_read(data
, sz
);
1129 if (*ext
< 'A' || 'Z' < *ext
)
1130 return error("index uses %.4s extension, which we do not understand",
1132 fprintf(stderr
, "ignoring %.4s extension\n", ext
);
1138 int read_index(struct index_state
*istate
)
1140 return read_index_from(istate
, get_index_file());
1143 static void convert_from_disk(struct ondisk_cache_entry
*ondisk
, struct cache_entry
*ce
)
1148 ce
->ce_ctime
= ntohl(ondisk
->ctime
.sec
);
1149 ce
->ce_mtime
= ntohl(ondisk
->mtime
.sec
);
1150 ce
->ce_dev
= ntohl(ondisk
->dev
);
1151 ce
->ce_ino
= ntohl(ondisk
->ino
);
1152 ce
->ce_mode
= ntohl(ondisk
->mode
);
1153 ce
->ce_uid
= ntohl(ondisk
->uid
);
1154 ce
->ce_gid
= ntohl(ondisk
->gid
);
1155 ce
->ce_size
= ntohl(ondisk
->size
);
1156 /* On-disk flags are just 16 bits */
1157 ce
->ce_flags
= ntohs(ondisk
->flags
);
1159 hashcpy(ce
->sha1
, ondisk
->sha1
);
1161 len
= ce
->ce_flags
& CE_NAMEMASK
;
1163 if (ce
->ce_flags
& CE_EXTENDED
) {
1164 struct ondisk_cache_entry_extended
*ondisk2
;
1166 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
1167 extended_flags
= ntohs(ondisk2
->flags2
) << 16;
1168 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1169 if (extended_flags
& ~CE_EXTENDED_FLAGS
)
1170 die("Unknown index entry format %08x", extended_flags
);
1171 ce
->ce_flags
|= extended_flags
;
1172 name
= ondisk2
->name
;
1175 name
= ondisk
->name
;
1177 if (len
== CE_NAMEMASK
)
1180 * NEEDSWORK: If the original index is crafted, this copy could
1183 memcpy(ce
->name
, name
, len
+ 1);
1186 static inline size_t estimate_cache_size(size_t ondisk_size
, unsigned int entries
)
1190 per_entry
= sizeof(struct cache_entry
) - sizeof(struct ondisk_cache_entry
);
1193 * Alignment can cause differences. This should be "alignof", but
1194 * since that's a gcc'ism, just use the size of a pointer.
1196 per_entry
+= sizeof(void *);
1197 return ondisk_size
+ entries
*per_entry
;
1200 /* remember to discard_cache() before reading a different cache! */
1201 int read_index_from(struct index_state
*istate
, const char *path
)
1205 unsigned long src_offset
, dst_offset
;
1206 struct cache_header
*hdr
;
1211 if (istate
->initialized
)
1212 return istate
->cache_nr
;
1215 istate
->timestamp
= 0;
1216 fd
= open(path
, O_RDONLY
);
1218 if (errno
== ENOENT
)
1220 die("index file open failed (%s)", strerror(errno
));
1224 die("cannot stat the open index (%s)", strerror(errno
));
1227 mmap_size
= xsize_t(st
.st_size
);
1228 if (mmap_size
< sizeof(struct cache_header
) + 20)
1229 die("index file smaller than expected");
1231 mmap
= xmmap(NULL
, mmap_size
, PROT_READ
| PROT_WRITE
, MAP_PRIVATE
, fd
, 0);
1233 if (mmap
== MAP_FAILED
)
1234 die("unable to map index file");
1237 if (verify_hdr(hdr
, mmap_size
) < 0)
1240 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
1241 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
1242 istate
->cache
= xcalloc(istate
->cache_alloc
, sizeof(struct cache_entry
*));
1245 * The disk format is actually larger than the in-memory format,
1246 * due to space for nsec etc, so even though the in-memory one
1247 * has room for a few more flags, we can allocate using the same
1250 istate
->alloc
= xmalloc(estimate_cache_size(mmap_size
, istate
->cache_nr
));
1251 istate
->initialized
= 1;
1253 src_offset
= sizeof(*hdr
);
1255 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1256 struct ondisk_cache_entry
*disk_ce
;
1257 struct cache_entry
*ce
;
1259 disk_ce
= (struct ondisk_cache_entry
*)((char *)mmap
+ src_offset
);
1260 ce
= (struct cache_entry
*)((char *)istate
->alloc
+ dst_offset
);
1261 convert_from_disk(disk_ce
, ce
);
1262 set_index_entry(istate
, i
, ce
);
1264 src_offset
+= ondisk_ce_size(ce
);
1265 dst_offset
+= ce_size(ce
);
1267 istate
->timestamp
= st
.st_mtime
;
1268 while (src_offset
<= mmap_size
- 20 - 8) {
1269 /* After an array of active_nr index entries,
1270 * there can be arbitrary number of extended
1271 * sections, each of which is prefixed with
1272 * extension name (4-byte) and section length
1273 * in 4-byte network byte order.
1275 unsigned long extsize
;
1276 memcpy(&extsize
, (char *)mmap
+ src_offset
+ 4, 4);
1277 extsize
= ntohl(extsize
);
1278 if (read_index_extension(istate
,
1279 (const char *) mmap
+ src_offset
,
1280 (char *) mmap
+ src_offset
+ 8,
1284 src_offset
+= extsize
;
1286 munmap(mmap
, mmap_size
);
1287 return istate
->cache_nr
;
1290 munmap(mmap
, mmap_size
);
1292 die("index file corrupt");
1295 int is_index_unborn(struct index_state
*istate
)
1297 return (!istate
->cache_nr
&& !istate
->alloc
&& !istate
->timestamp
);
1300 int discard_index(struct index_state
*istate
)
1302 istate
->cache_nr
= 0;
1303 istate
->cache_changed
= 0;
1304 istate
->timestamp
= 0;
1305 istate
->name_hash_initialized
= 0;
1306 free_hash(&istate
->name_hash
);
1307 cache_tree_free(&(istate
->cache_tree
));
1308 free(istate
->alloc
);
1309 istate
->alloc
= NULL
;
1310 istate
->initialized
= 0;
1312 /* no need to throw away allocated active_cache */
1316 int unmerged_index(const struct index_state
*istate
)
1319 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1320 if (ce_stage(istate
->cache
[i
]))
1326 #define WRITE_BUFFER_SIZE 8192
1327 static unsigned char write_buffer
[WRITE_BUFFER_SIZE
];
1328 static unsigned long write_buffer_len
;
1330 static int ce_write_flush(git_SHA_CTX
*context
, int fd
)
1332 unsigned int buffered
= write_buffer_len
;
1334 git_SHA1_Update(context
, write_buffer
, buffered
);
1335 if (write_in_full(fd
, write_buffer
, buffered
) != buffered
)
1337 write_buffer_len
= 0;
1342 static int ce_write(git_SHA_CTX
*context
, int fd
, void *data
, unsigned int len
)
1345 unsigned int buffered
= write_buffer_len
;
1346 unsigned int partial
= WRITE_BUFFER_SIZE
- buffered
;
1349 memcpy(write_buffer
+ buffered
, data
, partial
);
1350 buffered
+= partial
;
1351 if (buffered
== WRITE_BUFFER_SIZE
) {
1352 write_buffer_len
= buffered
;
1353 if (ce_write_flush(context
, fd
))
1357 write_buffer_len
= buffered
;
1359 data
= (char *) data
+ partial
;
1364 static int write_index_ext_header(git_SHA_CTX
*context
, int fd
,
1365 unsigned int ext
, unsigned int sz
)
1369 return ((ce_write(context
, fd
, &ext
, 4) < 0) ||
1370 (ce_write(context
, fd
, &sz
, 4) < 0)) ? -1 : 0;
1373 static int ce_flush(git_SHA_CTX
*context
, int fd
)
1375 unsigned int left
= write_buffer_len
;
1378 write_buffer_len
= 0;
1379 git_SHA1_Update(context
, write_buffer
, left
);
1382 /* Flush first if not enough space for SHA1 signature */
1383 if (left
+ 20 > WRITE_BUFFER_SIZE
) {
1384 if (write_in_full(fd
, write_buffer
, left
) != left
)
1389 /* Append the SHA1 signature at the end */
1390 git_SHA1_Final(write_buffer
+ left
, context
);
1392 return (write_in_full(fd
, write_buffer
, left
) != left
) ? -1 : 0;
1395 static void ce_smudge_racily_clean_entry(struct cache_entry
*ce
)
1398 * The only thing we care about in this function is to smudge the
1399 * falsely clean entry due to touch-update-touch race, so we leave
1400 * everything else as they are. We are called for entries whose
1401 * ce_mtime match the index file mtime.
1403 * Note that this actually does not do much for gitlinks, for
1404 * which ce_match_stat_basic() always goes to the actual
1405 * contents. The caller checks with is_racy_timestamp() which
1406 * always says "no" for gitlinks, so we are not called for them ;-)
1410 if (lstat(ce
->name
, &st
) < 0)
1412 if (ce_match_stat_basic(ce
, &st
))
1414 if (ce_modified_check_fs(ce
, &st
)) {
1415 /* This is "racily clean"; smudge it. Note that this
1416 * is a tricky code. At first glance, it may appear
1417 * that it can break with this sequence:
1419 * $ echo xyzzy >frotz
1420 * $ git-update-index --add frotz
1423 * $ echo filfre >nitfol
1424 * $ git-update-index --add nitfol
1426 * but it does not. When the second update-index runs,
1427 * it notices that the entry "frotz" has the same timestamp
1428 * as index, and if we were to smudge it by resetting its
1429 * size to zero here, then the object name recorded
1430 * in index is the 6-byte file but the cached stat information
1431 * becomes zero --- which would then match what we would
1432 * obtain from the filesystem next time we stat("frotz").
1434 * However, the second update-index, before calling
1435 * this function, notices that the cached size is 6
1436 * bytes and what is on the filesystem is an empty
1437 * file, and never calls us, so the cached size information
1438 * for "frotz" stays 6 which does not match the filesystem.
1444 static int ce_write_entry(git_SHA_CTX
*c
, int fd
, struct cache_entry
*ce
)
1446 int size
= ondisk_ce_size(ce
);
1447 struct ondisk_cache_entry
*ondisk
= xcalloc(1, size
);
1450 ondisk
->ctime
.sec
= htonl(ce
->ce_ctime
);
1451 ondisk
->ctime
.nsec
= 0;
1452 ondisk
->mtime
.sec
= htonl(ce
->ce_mtime
);
1453 ondisk
->mtime
.nsec
= 0;
1454 ondisk
->dev
= htonl(ce
->ce_dev
);
1455 ondisk
->ino
= htonl(ce
->ce_ino
);
1456 ondisk
->mode
= htonl(ce
->ce_mode
);
1457 ondisk
->uid
= htonl(ce
->ce_uid
);
1458 ondisk
->gid
= htonl(ce
->ce_gid
);
1459 ondisk
->size
= htonl(ce
->ce_size
);
1460 hashcpy(ondisk
->sha1
, ce
->sha1
);
1461 ondisk
->flags
= htons(ce
->ce_flags
);
1462 if (ce
->ce_flags
& CE_EXTENDED
) {
1463 struct ondisk_cache_entry_extended
*ondisk2
;
1464 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
1465 ondisk2
->flags2
= htons((ce
->ce_flags
& CE_EXTENDED_FLAGS
) >> 16);
1466 name
= ondisk2
->name
;
1469 name
= ondisk
->name
;
1470 memcpy(name
, ce
->name
, ce_namelen(ce
));
1472 return ce_write(c
, fd
, ondisk
, size
);
1475 int write_index(const struct index_state
*istate
, int newfd
)
1478 struct cache_header hdr
;
1479 int i
, err
, removed
, extended
;
1480 struct cache_entry
**cache
= istate
->cache
;
1481 int entries
= istate
->cache_nr
;
1483 for (i
= removed
= extended
= 0; i
< entries
; i
++) {
1484 if (cache
[i
]->ce_flags
& CE_REMOVE
)
1487 /* reduce extended entries if possible */
1488 cache
[i
]->ce_flags
&= ~CE_EXTENDED
;
1489 if (cache
[i
]->ce_flags
& CE_EXTENDED_FLAGS
) {
1491 cache
[i
]->ce_flags
|= CE_EXTENDED
;
1495 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
1496 /* for extended format, increase version so older git won't try to read it */
1497 hdr
.hdr_version
= htonl(extended
? 3 : 2);
1498 hdr
.hdr_entries
= htonl(entries
- removed
);
1501 if (ce_write(&c
, newfd
, &hdr
, sizeof(hdr
)) < 0)
1504 for (i
= 0; i
< entries
; i
++) {
1505 struct cache_entry
*ce
= cache
[i
];
1506 if (ce
->ce_flags
& CE_REMOVE
)
1508 if (!ce_uptodate(ce
) && is_racy_timestamp(istate
, ce
))
1509 ce_smudge_racily_clean_entry(ce
);
1510 if (ce_write_entry(&c
, newfd
, ce
) < 0)
1514 /* Write extension data here */
1515 if (istate
->cache_tree
) {
1516 struct strbuf sb
= STRBUF_INIT
;
1518 cache_tree_write(&sb
, istate
->cache_tree
);
1519 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_TREE
, sb
.len
) < 0
1520 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
1521 strbuf_release(&sb
);
1525 return ce_flush(&c
, newfd
);
1529 * Read the index file that is potentially unmerged into given
1530 * index_state, dropping any unmerged entries. Returns true is
1531 * the index is unmerged. Callers who want to refuse to work
1532 * from an unmerged state can call this and check its return value,
1533 * instead of calling read_cache().
1535 int read_index_unmerged(struct index_state
*istate
)
1541 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1542 struct cache_entry
*ce
= istate
->cache
[i
];
1543 struct cache_entry
*new_ce
;
1549 len
= strlen(ce
->name
);
1550 size
= cache_entry_size(len
);
1551 new_ce
= xcalloc(1, size
);
1552 hashcpy(new_ce
->sha1
, ce
->sha1
);
1553 memcpy(new_ce
->name
, ce
->name
, len
);
1554 new_ce
->ce_flags
= create_ce_flags(len
, 0);
1555 new_ce
->ce_mode
= ce
->ce_mode
;
1556 if (add_index_entry(istate
, new_ce
, 0))
1557 return error("%s: cannot drop to stage #0",
1559 i
= index_name_pos(istate
, new_ce
->name
, len
);
1564 struct update_callback_data
1570 static void update_callback(struct diff_queue_struct
*q
,
1571 struct diff_options
*opt
, void *cbdata
)
1574 struct update_callback_data
*data
= cbdata
;
1576 for (i
= 0; i
< q
->nr
; i
++) {
1577 struct diff_filepair
*p
= q
->queue
[i
];
1578 const char *path
= p
->one
->path
;
1579 switch (p
->status
) {
1581 die("unexpected diff status %c", p
->status
);
1582 case DIFF_STATUS_UNMERGED
:
1583 case DIFF_STATUS_MODIFIED
:
1584 case DIFF_STATUS_TYPE_CHANGED
:
1585 if (add_file_to_index(&the_index
, path
, data
->flags
)) {
1586 if (!(data
->flags
& ADD_CACHE_IGNORE_ERRORS
))
1587 die("updating files failed");
1591 case DIFF_STATUS_DELETED
:
1592 if (data
->flags
& ADD_CACHE_IGNORE_REMOVAL
)
1594 if (!(data
->flags
& ADD_CACHE_PRETEND
))
1595 remove_file_from_index(&the_index
, path
);
1596 if (data
->flags
& (ADD_CACHE_PRETEND
|ADD_CACHE_VERBOSE
))
1597 printf("remove '%s'\n", path
);
1603 int add_files_to_cache(const char *prefix
, const char **pathspec
, int flags
)
1605 struct update_callback_data data
;
1606 struct rev_info rev
;
1607 init_revisions(&rev
, prefix
);
1608 setup_revisions(0, NULL
, &rev
, NULL
);
1609 rev
.prune_data
= pathspec
;
1610 rev
.diffopt
.output_format
= DIFF_FORMAT_CALLBACK
;
1611 rev
.diffopt
.format_callback
= update_callback
;
1613 data
.add_errors
= 0;
1614 rev
.diffopt
.format_callback_data
= &data
;
1615 run_diff_files(&rev
, DIFF_RACY_IS_MODIFIED
);
1616 return !!data
.add_errors
;
1620 * Returns 1 if the path is an "other" path with respect to
1621 * the index; that is, the path is not mentioned in the index at all,
1622 * either as a file, a directory with some files in the index,
1623 * or as an unmerged entry.
1625 * We helpfully remove a trailing "/" from directories so that
1626 * the output of read_directory can be used as-is.
1628 int index_name_is_other(const struct index_state
*istate
, const char *name
,
1632 if (namelen
&& name
[namelen
- 1] == '/')
1634 pos
= index_name_pos(istate
, name
, namelen
);
1636 return 0; /* exact match */
1638 if (pos
< istate
->cache_nr
) {
1639 struct cache_entry
*ce
= istate
->cache
[pos
];
1640 if (ce_namelen(ce
) == namelen
&&
1641 !memcmp(ce
->name
, name
, namelen
))
1642 return 0; /* Yup, this one exists unmerged */