2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 #define NO_THE_INDEX_COMPATIBILITY_MACROS
13 #include "cache-tree.h"
16 #include "object-store.h"
20 #include "resolve-undo.h"
23 #include "split-index.h"
25 #include "fsmonitor.h"
27 /* Mask for the name length in ce_flags in the on-disk index */
29 #define CE_NAMEMASK (0x0fff)
33 * The first letter should be 'A'..'Z' for extensions that are not
34 * necessary for a correct operation (i.e. optimization data).
35 * When new extensions are added that _needs_ to be understood in
36 * order to correctly interpret the index file, pick character that
37 * is outside the range, to cause the reader to abort.
40 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
41 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
42 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
43 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
44 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
45 #define CACHE_EXT_FSMONITOR 0x46534D4E /* "FSMN" */
47 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
48 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
49 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
50 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED | FSMONITOR_CHANGED)
54 * This is an estimate of the pathname length in the index. We use
55 * this for V4 index files to guess the un-deltafied size of the index
56 * in memory because of pathname deltafication. This is not required
57 * for V2/V3 index formats because their pathnames are not compressed.
58 * If the initial amount of memory set aside is not sufficient, the
59 * mem pool will allocate extra memory.
61 #define CACHE_ENTRY_PATH_LENGTH 80
63 static inline struct cache_entry
*mem_pool__ce_alloc(struct mem_pool
*mem_pool
, size_t len
)
65 struct cache_entry
*ce
;
66 ce
= mem_pool_alloc(mem_pool
, cache_entry_size(len
));
67 ce
->mem_pool_allocated
= 1;
71 static inline struct cache_entry
*mem_pool__ce_calloc(struct mem_pool
*mem_pool
, size_t len
)
73 struct cache_entry
* ce
;
74 ce
= mem_pool_calloc(mem_pool
, 1, cache_entry_size(len
));
75 ce
->mem_pool_allocated
= 1;
79 static struct mem_pool
*find_mem_pool(struct index_state
*istate
)
81 struct mem_pool
**pool_ptr
;
83 if (istate
->split_index
&& istate
->split_index
->base
)
84 pool_ptr
= &istate
->split_index
->base
->ce_mem_pool
;
86 pool_ptr
= &istate
->ce_mem_pool
;
89 mem_pool_init(pool_ptr
, 0);
94 struct index_state the_index
;
95 static const char *alternate_index_output
;
97 static void set_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
99 istate
->cache
[nr
] = ce
;
100 add_name_hash(istate
, ce
);
103 static void replace_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
105 struct cache_entry
*old
= istate
->cache
[nr
];
107 replace_index_entry_in_base(istate
, old
, ce
);
108 remove_name_hash(istate
, old
);
109 discard_cache_entry(old
);
110 ce
->ce_flags
&= ~CE_HASHED
;
111 set_index_entry(istate
, nr
, ce
);
112 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
113 mark_fsmonitor_invalid(istate
, ce
);
114 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
117 void rename_index_entry_at(struct index_state
*istate
, int nr
, const char *new_name
)
119 struct cache_entry
*old_entry
= istate
->cache
[nr
], *new_entry
;
120 int namelen
= strlen(new_name
);
122 new_entry
= make_empty_cache_entry(istate
, namelen
);
123 copy_cache_entry(new_entry
, old_entry
);
124 new_entry
->ce_flags
&= ~CE_HASHED
;
125 new_entry
->ce_namelen
= namelen
;
126 new_entry
->index
= 0;
127 memcpy(new_entry
->name
, new_name
, namelen
+ 1);
129 cache_tree_invalidate_path(istate
, old_entry
->name
);
130 untracked_cache_remove_from_index(istate
, old_entry
->name
);
131 remove_index_entry_at(istate
, nr
);
132 add_index_entry(istate
, new_entry
, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
135 void fill_stat_data(struct stat_data
*sd
, struct stat
*st
)
137 sd
->sd_ctime
.sec
= (unsigned int)st
->st_ctime
;
138 sd
->sd_mtime
.sec
= (unsigned int)st
->st_mtime
;
139 sd
->sd_ctime
.nsec
= ST_CTIME_NSEC(*st
);
140 sd
->sd_mtime
.nsec
= ST_MTIME_NSEC(*st
);
141 sd
->sd_dev
= st
->st_dev
;
142 sd
->sd_ino
= st
->st_ino
;
143 sd
->sd_uid
= st
->st_uid
;
144 sd
->sd_gid
= st
->st_gid
;
145 sd
->sd_size
= st
->st_size
;
148 int match_stat_data(const struct stat_data
*sd
, struct stat
*st
)
152 if (sd
->sd_mtime
.sec
!= (unsigned int)st
->st_mtime
)
153 changed
|= MTIME_CHANGED
;
154 if (trust_ctime
&& check_stat
&&
155 sd
->sd_ctime
.sec
!= (unsigned int)st
->st_ctime
)
156 changed
|= CTIME_CHANGED
;
159 if (check_stat
&& sd
->sd_mtime
.nsec
!= ST_MTIME_NSEC(*st
))
160 changed
|= MTIME_CHANGED
;
161 if (trust_ctime
&& check_stat
&&
162 sd
->sd_ctime
.nsec
!= ST_CTIME_NSEC(*st
))
163 changed
|= CTIME_CHANGED
;
167 if (sd
->sd_uid
!= (unsigned int) st
->st_uid
||
168 sd
->sd_gid
!= (unsigned int) st
->st_gid
)
169 changed
|= OWNER_CHANGED
;
170 if (sd
->sd_ino
!= (unsigned int) st
->st_ino
)
171 changed
|= INODE_CHANGED
;
176 * st_dev breaks on network filesystems where different
177 * clients will have different views of what "device"
178 * the filesystem is on
180 if (check_stat
&& sd
->sd_dev
!= (unsigned int) st
->st_dev
)
181 changed
|= INODE_CHANGED
;
184 if (sd
->sd_size
!= (unsigned int) st
->st_size
)
185 changed
|= DATA_CHANGED
;
191 * This only updates the "non-critical" parts of the directory
192 * cache, ie the parts that aren't tracked by GIT, and only used
193 * to validate the cache.
195 void fill_stat_cache_info(struct cache_entry
*ce
, struct stat
*st
)
197 fill_stat_data(&ce
->ce_stat_data
, st
);
199 if (assume_unchanged
)
200 ce
->ce_flags
|= CE_VALID
;
202 if (S_ISREG(st
->st_mode
)) {
203 ce_mark_uptodate(ce
);
204 mark_fsmonitor_valid(ce
);
208 static int ce_compare_data(const struct cache_entry
*ce
, struct stat
*st
)
211 int fd
= git_open_cloexec(ce
->name
, O_RDONLY
);
214 struct object_id oid
;
215 if (!index_fd(&oid
, fd
, st
, OBJ_BLOB
, ce
->name
, 0))
216 match
= oidcmp(&oid
, &ce
->oid
);
217 /* index_fd() closed the file descriptor already */
222 static int ce_compare_link(const struct cache_entry
*ce
, size_t expected_size
)
227 enum object_type type
;
228 struct strbuf sb
= STRBUF_INIT
;
230 if (strbuf_readlink(&sb
, ce
->name
, expected_size
))
233 buffer
= read_object_file(&ce
->oid
, &type
, &size
);
236 match
= memcmp(buffer
, sb
.buf
, size
);
243 static int ce_compare_gitlink(const struct cache_entry
*ce
)
245 struct object_id oid
;
248 * We don't actually require that the .git directory
249 * under GITLINK directory be a valid git directory. It
250 * might even be missing (in case nobody populated that
253 * If so, we consider it always to match.
255 if (resolve_gitlink_ref(ce
->name
, "HEAD", &oid
) < 0)
257 return oidcmp(&oid
, &ce
->oid
);
260 static int ce_modified_check_fs(const struct cache_entry
*ce
, struct stat
*st
)
262 switch (st
->st_mode
& S_IFMT
) {
264 if (ce_compare_data(ce
, st
))
268 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
272 if (S_ISGITLINK(ce
->ce_mode
))
273 return ce_compare_gitlink(ce
) ? DATA_CHANGED
: 0;
274 /* else fallthrough */
281 static int ce_match_stat_basic(const struct cache_entry
*ce
, struct stat
*st
)
283 unsigned int changed
= 0;
285 if (ce
->ce_flags
& CE_REMOVE
)
286 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
288 switch (ce
->ce_mode
& S_IFMT
) {
290 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
291 /* We consider only the owner x bit to be relevant for
294 if (trust_executable_bit
&&
295 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
296 changed
|= MODE_CHANGED
;
299 if (!S_ISLNK(st
->st_mode
) &&
300 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
301 changed
|= TYPE_CHANGED
;
304 /* We ignore most of the st_xxx fields for gitlinks */
305 if (!S_ISDIR(st
->st_mode
))
306 changed
|= TYPE_CHANGED
;
307 else if (ce_compare_gitlink(ce
))
308 changed
|= DATA_CHANGED
;
311 die("internal error: ce_mode is %o", ce
->ce_mode
);
314 changed
|= match_stat_data(&ce
->ce_stat_data
, st
);
316 /* Racily smudged entry? */
317 if (!ce
->ce_stat_data
.sd_size
) {
318 if (!is_empty_blob_sha1(ce
->oid
.hash
))
319 changed
|= DATA_CHANGED
;
325 static int is_racy_stat(const struct index_state
*istate
,
326 const struct stat_data
*sd
)
328 return (istate
->timestamp
.sec
&&
330 /* nanosecond timestamped files can also be racy! */
331 (istate
->timestamp
.sec
< sd
->sd_mtime
.sec
||
332 (istate
->timestamp
.sec
== sd
->sd_mtime
.sec
&&
333 istate
->timestamp
.nsec
<= sd
->sd_mtime
.nsec
))
335 istate
->timestamp
.sec
<= sd
->sd_mtime
.sec
340 static int is_racy_timestamp(const struct index_state
*istate
,
341 const struct cache_entry
*ce
)
343 return (!S_ISGITLINK(ce
->ce_mode
) &&
344 is_racy_stat(istate
, &ce
->ce_stat_data
));
347 int match_stat_data_racy(const struct index_state
*istate
,
348 const struct stat_data
*sd
, struct stat
*st
)
350 if (is_racy_stat(istate
, sd
))
351 return MTIME_CHANGED
;
352 return match_stat_data(sd
, st
);
355 int ie_match_stat(struct index_state
*istate
,
356 const struct cache_entry
*ce
, struct stat
*st
,
357 unsigned int options
)
359 unsigned int changed
;
360 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
361 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
362 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
363 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
365 if (!ignore_fsmonitor
)
366 refresh_fsmonitor(istate
);
368 * If it's marked as always valid in the index, it's
369 * valid whatever the checked-out copy says.
371 * skip-worktree has the same effect with higher precedence
373 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
))
375 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
377 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
))
381 * Intent-to-add entries have not been added, so the index entry
382 * by definition never matches what is in the work tree until it
383 * actually gets added.
385 if (ce_intent_to_add(ce
))
386 return DATA_CHANGED
| TYPE_CHANGED
| MODE_CHANGED
;
388 changed
= ce_match_stat_basic(ce
, st
);
391 * Within 1 second of this sequence:
392 * echo xyzzy >file && git-update-index --add file
393 * running this command:
395 * would give a falsely clean cache entry. The mtime and
396 * length match the cache, and other stat fields do not change.
398 * We could detect this at update-index time (the cache entry
399 * being registered/updated records the same time as "now")
400 * and delay the return from git-update-index, but that would
401 * effectively mean we can make at most one commit per second,
402 * which is not acceptable. Instead, we check cache entries
403 * whose mtime are the same as the index file timestamp more
404 * carefully than others.
406 if (!changed
&& is_racy_timestamp(istate
, ce
)) {
407 if (assume_racy_is_modified
)
408 changed
|= DATA_CHANGED
;
410 changed
|= ce_modified_check_fs(ce
, st
);
416 int ie_modified(struct index_state
*istate
,
417 const struct cache_entry
*ce
,
418 struct stat
*st
, unsigned int options
)
420 int changed
, changed_fs
;
422 changed
= ie_match_stat(istate
, ce
, st
, options
);
426 * If the mode or type has changed, there's no point in trying
427 * to refresh the entry - it's not going to match
429 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
433 * Immediately after read-tree or update-index --cacheinfo,
434 * the length field is zero, as we have never even read the
435 * lstat(2) information once, and we cannot trust DATA_CHANGED
436 * returned by ie_match_stat() which in turn was returned by
437 * ce_match_stat_basic() to signal that the filesize of the
438 * blob changed. We have to actually go to the filesystem to
439 * see if the contents match, and if so, should answer "unchanged".
441 * The logic does not apply to gitlinks, as ce_match_stat_basic()
442 * already has checked the actual HEAD from the filesystem in the
443 * subproject. If ie_match_stat() already said it is different,
444 * then we know it is.
446 if ((changed
& DATA_CHANGED
) &&
447 (S_ISGITLINK(ce
->ce_mode
) || ce
->ce_stat_data
.sd_size
!= 0))
450 changed_fs
= ce_modified_check_fs(ce
, st
);
452 return changed
| changed_fs
;
456 int base_name_compare(const char *name1
, int len1
, int mode1
,
457 const char *name2
, int len2
, int mode2
)
459 unsigned char c1
, c2
;
460 int len
= len1
< len2
? len1
: len2
;
463 cmp
= memcmp(name1
, name2
, len
);
468 if (!c1
&& S_ISDIR(mode1
))
470 if (!c2
&& S_ISDIR(mode2
))
472 return (c1
< c2
) ? -1 : (c1
> c2
) ? 1 : 0;
476 * df_name_compare() is identical to base_name_compare(), except it
477 * compares conflicting directory/file entries as equal. Note that
478 * while a directory name compares as equal to a regular file, they
479 * then individually compare _differently_ to a filename that has
480 * a dot after the basename (because '\0' < '.' < '/').
482 * This is used by routines that want to traverse the git namespace
483 * but then handle conflicting entries together when possible.
485 int df_name_compare(const char *name1
, int len1
, int mode1
,
486 const char *name2
, int len2
, int mode2
)
488 int len
= len1
< len2
? len1
: len2
, cmp
;
489 unsigned char c1
, c2
;
491 cmp
= memcmp(name1
, name2
, len
);
494 /* Directories and files compare equal (same length, same name) */
498 if (!c1
&& S_ISDIR(mode1
))
501 if (!c2
&& S_ISDIR(mode2
))
503 if (c1
== '/' && !c2
)
505 if (c2
== '/' && !c1
)
510 int name_compare(const char *name1
, size_t len1
, const char *name2
, size_t len2
)
512 size_t min_len
= (len1
< len2
) ? len1
: len2
;
513 int cmp
= memcmp(name1
, name2
, min_len
);
523 int cache_name_stage_compare(const char *name1
, int len1
, int stage1
, const char *name2
, int len2
, int stage2
)
527 cmp
= name_compare(name1
, len1
, name2
, len2
);
538 static int index_name_stage_pos(const struct index_state
*istate
, const char *name
, int namelen
, int stage
)
543 last
= istate
->cache_nr
;
544 while (last
> first
) {
545 int next
= (last
+ first
) >> 1;
546 struct cache_entry
*ce
= istate
->cache
[next
];
547 int cmp
= cache_name_stage_compare(name
, namelen
, stage
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
559 int index_name_pos(const struct index_state
*istate
, const char *name
, int namelen
)
561 return index_name_stage_pos(istate
, name
, namelen
, 0);
564 int remove_index_entry_at(struct index_state
*istate
, int pos
)
566 struct cache_entry
*ce
= istate
->cache
[pos
];
568 record_resolve_undo(istate
, ce
);
569 remove_name_hash(istate
, ce
);
570 save_or_free_index_entry(istate
, ce
);
571 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
573 if (pos
>= istate
->cache_nr
)
575 MOVE_ARRAY(istate
->cache
+ pos
, istate
->cache
+ pos
+ 1,
576 istate
->cache_nr
- pos
);
581 * Remove all cache entries marked for removal, that is where
582 * CE_REMOVE is set in ce_flags. This is much more effective than
583 * calling remove_index_entry_at() for each entry to be removed.
585 void remove_marked_cache_entries(struct index_state
*istate
)
587 struct cache_entry
**ce_array
= istate
->cache
;
590 for (i
= j
= 0; i
< istate
->cache_nr
; i
++) {
591 if (ce_array
[i
]->ce_flags
& CE_REMOVE
) {
592 remove_name_hash(istate
, ce_array
[i
]);
593 save_or_free_index_entry(istate
, ce_array
[i
]);
596 ce_array
[j
++] = ce_array
[i
];
598 if (j
== istate
->cache_nr
)
600 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
601 istate
->cache_nr
= j
;
604 int remove_file_from_index(struct index_state
*istate
, const char *path
)
606 int pos
= index_name_pos(istate
, path
, strlen(path
));
609 cache_tree_invalidate_path(istate
, path
);
610 untracked_cache_remove_from_index(istate
, path
);
611 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
612 remove_index_entry_at(istate
, pos
);
616 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
618 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
621 static int index_name_pos_also_unmerged(struct index_state
*istate
,
622 const char *path
, int namelen
)
624 int pos
= index_name_pos(istate
, path
, namelen
);
625 struct cache_entry
*ce
;
630 /* maybe unmerged? */
632 if (pos
>= istate
->cache_nr
||
633 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
636 /* order of preference: stage 2, 1, 3 */
637 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
638 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
639 !compare_name(ce
, path
, namelen
))
644 static int different_name(struct cache_entry
*ce
, struct cache_entry
*alias
)
646 int len
= ce_namelen(ce
);
647 return ce_namelen(alias
) != len
|| memcmp(ce
->name
, alias
->name
, len
);
651 * If we add a filename that aliases in the cache, we will use the
652 * name that we already have - but we don't want to update the same
653 * alias twice, because that implies that there were actually two
654 * different files with aliasing names!
656 * So we use the CE_ADDED flag to verify that the alias was an old
657 * one before we accept it as
659 static struct cache_entry
*create_alias_ce(struct index_state
*istate
,
660 struct cache_entry
*ce
,
661 struct cache_entry
*alias
)
664 struct cache_entry
*new_entry
;
666 if (alias
->ce_flags
& CE_ADDED
)
667 die("Will not add file alias '%s' ('%s' already exists in index)", ce
->name
, alias
->name
);
669 /* Ok, create the new entry using the name of the existing alias */
670 len
= ce_namelen(alias
);
671 new_entry
= make_empty_cache_entry(istate
, len
);
672 memcpy(new_entry
->name
, alias
->name
, len
);
673 copy_cache_entry(new_entry
, ce
);
674 save_or_free_index_entry(istate
, ce
);
678 void set_object_name_for_intent_to_add_entry(struct cache_entry
*ce
)
680 struct object_id oid
;
681 if (write_object_file("", 0, blob_type
, &oid
))
682 die("cannot create an empty blob in the object database");
683 oidcpy(&ce
->oid
, &oid
);
686 int add_to_index(struct index_state
*istate
, const char *path
, struct stat
*st
, int flags
)
688 int namelen
, was_same
;
689 mode_t st_mode
= st
->st_mode
;
690 struct cache_entry
*ce
, *alias
= NULL
;
691 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_IGNORE_SKIP_WORKTREE
|CE_MATCH_RACY_IS_DIRTY
;
692 int verbose
= flags
& (ADD_CACHE_VERBOSE
| ADD_CACHE_PRETEND
);
693 int pretend
= flags
& ADD_CACHE_PRETEND
;
694 int intent_only
= flags
& ADD_CACHE_INTENT
;
695 int add_option
= (ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
|
696 (intent_only
? ADD_CACHE_NEW_ONLY
: 0));
697 int newflags
= HASH_WRITE_OBJECT
;
699 if (flags
& HASH_RENORMALIZE
)
700 newflags
|= HASH_RENORMALIZE
;
702 if (!S_ISREG(st_mode
) && !S_ISLNK(st_mode
) && !S_ISDIR(st_mode
))
703 return error("%s: can only add regular files, symbolic links or git-directories", path
);
705 namelen
= strlen(path
);
706 if (S_ISDIR(st_mode
)) {
707 while (namelen
&& path
[namelen
-1] == '/')
710 ce
= make_empty_cache_entry(istate
, namelen
);
711 memcpy(ce
->name
, path
, namelen
);
712 ce
->ce_namelen
= namelen
;
714 fill_stat_cache_info(ce
, st
);
716 ce
->ce_flags
|= CE_INTENT_TO_ADD
;
719 if (trust_executable_bit
&& has_symlinks
) {
720 ce
->ce_mode
= create_ce_mode(st_mode
);
722 /* If there is an existing entry, pick the mode bits and type
723 * from it, otherwise assume unexecutable regular file.
725 struct cache_entry
*ent
;
726 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
728 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
729 ce
->ce_mode
= ce_mode_from_stat(ent
, st_mode
);
732 /* When core.ignorecase=true, determine if a directory of the same name but differing
733 * case already exists within the Git repository. If it does, ensure the directory
734 * case of the file being added to the repository matches (is folded into) the existing
735 * entry's directory case.
738 adjust_dirname_case(istate
, ce
->name
);
740 if (!(flags
& HASH_RENORMALIZE
)) {
741 alias
= index_file_exists(istate
, ce
->name
,
742 ce_namelen(ce
), ignore_case
);
745 !ie_match_stat(istate
, alias
, st
, ce_option
)) {
746 /* Nothing changed, really */
747 if (!S_ISGITLINK(alias
->ce_mode
))
748 ce_mark_uptodate(alias
);
749 alias
->ce_flags
|= CE_ADDED
;
751 discard_cache_entry(ce
);
756 if (index_path(&ce
->oid
, path
, st
, newflags
)) {
757 discard_cache_entry(ce
);
758 return error("unable to index file %s", path
);
761 set_object_name_for_intent_to_add_entry(ce
);
763 if (ignore_case
&& alias
&& different_name(ce
, alias
))
764 ce
= create_alias_ce(istate
, ce
, alias
);
765 ce
->ce_flags
|= CE_ADDED
;
767 /* It was suspected to be racily clean, but it turns out to be Ok */
770 !oidcmp(&alias
->oid
, &ce
->oid
) &&
771 ce
->ce_mode
== alias
->ce_mode
);
774 discard_cache_entry(ce
);
775 else if (add_index_entry(istate
, ce
, add_option
)) {
776 discard_cache_entry(ce
);
777 return error("unable to add %s to index", path
);
779 if (verbose
&& !was_same
)
780 printf("add '%s'\n", path
);
784 int add_file_to_index(struct index_state
*istate
, const char *path
, int flags
)
787 if (lstat(path
, &st
))
788 die_errno("unable to stat '%s'", path
);
789 return add_to_index(istate
, path
, &st
, flags
);
792 struct cache_entry
*make_empty_cache_entry(struct index_state
*istate
, size_t len
)
794 return mem_pool__ce_calloc(find_mem_pool(istate
), len
);
797 struct cache_entry
*make_empty_transient_cache_entry(size_t len
)
799 return xcalloc(1, cache_entry_size(len
));
802 struct cache_entry
*make_cache_entry(struct index_state
*istate
,
804 const struct object_id
*oid
,
807 unsigned int refresh_options
)
809 struct cache_entry
*ce
, *ret
;
812 if (!verify_path(path
, mode
)) {
813 error("Invalid path '%s'", path
);
818 ce
= make_empty_cache_entry(istate
, len
);
820 oidcpy(&ce
->oid
, oid
);
821 memcpy(ce
->name
, path
, len
);
822 ce
->ce_flags
= create_ce_flags(stage
);
823 ce
->ce_namelen
= len
;
824 ce
->ce_mode
= create_ce_mode(mode
);
826 ret
= refresh_cache_entry(&the_index
, ce
, refresh_options
);
828 discard_cache_entry(ce
);
832 struct cache_entry
*make_transient_cache_entry(unsigned int mode
, const struct object_id
*oid
,
833 const char *path
, int stage
)
835 struct cache_entry
*ce
;
838 if (!verify_path(path
, mode
)) {
839 error("Invalid path '%s'", path
);
844 ce
= make_empty_transient_cache_entry(len
);
846 oidcpy(&ce
->oid
, oid
);
847 memcpy(ce
->name
, path
, len
);
848 ce
->ce_flags
= create_ce_flags(stage
);
849 ce
->ce_namelen
= len
;
850 ce
->ce_mode
= create_ce_mode(mode
);
856 * Chmod an index entry with either +x or -x.
858 * Returns -1 if the chmod for the particular cache entry failed (if it's
859 * not a regular file), -2 if an invalid flip argument is passed in, 0
862 int chmod_index_entry(struct index_state
*istate
, struct cache_entry
*ce
,
865 if (!S_ISREG(ce
->ce_mode
))
872 ce
->ce_mode
&= ~0111;
877 cache_tree_invalidate_path(istate
, ce
->name
);
878 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
879 mark_fsmonitor_invalid(istate
, ce
);
880 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
885 int ce_same_name(const struct cache_entry
*a
, const struct cache_entry
*b
)
887 int len
= ce_namelen(a
);
888 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
892 * We fundamentally don't like some paths: we don't want
893 * dot or dot-dot anywhere, and for obvious reasons don't
894 * want to recurse into ".git" either.
896 * Also, we don't want double slashes or slashes at the
897 * end that can make pathnames ambiguous.
899 static int verify_dotfile(const char *rest
, unsigned mode
)
902 * The first character was '.', but that
903 * has already been discarded, we now test
907 /* "." is not allowed */
908 if (*rest
== '\0' || is_dir_sep(*rest
))
913 * ".git" followed by NUL or slash is bad. Note that we match
914 * case-insensitively here, even if ignore_case is not set.
915 * This outlaws ".GIT" everywhere out of an abundance of caution,
916 * since there's really no good reason to allow it.
918 * Once we've seen ".git", we can also find ".gitmodules", etc (also
919 * case-insensitively).
923 if (rest
[1] != 'i' && rest
[1] != 'I')
925 if (rest
[2] != 't' && rest
[2] != 'T')
927 if (rest
[3] == '\0' || is_dir_sep(rest
[3]))
931 if (skip_iprefix(rest
, "modules", &rest
) &&
932 (*rest
== '\0' || is_dir_sep(*rest
)))
937 if (rest
[1] == '\0' || is_dir_sep(rest
[1]))
943 int verify_path(const char *path
, unsigned mode
)
947 if (has_dos_drive_prefix(path
))
957 if (is_hfs_dotgit(path
))
960 if (is_hfs_dotgitmodules(path
))
965 if (is_ntfs_dotgit(path
))
968 if (is_ntfs_dotgitmodules(path
))
974 if ((c
== '.' && !verify_dotfile(path
, mode
)) ||
975 is_dir_sep(c
) || c
== '\0')
983 * Do we have another file that has the beginning components being a
984 * proper superset of the name we're trying to add?
986 static int has_file_name(struct index_state
*istate
,
987 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
990 int len
= ce_namelen(ce
);
991 int stage
= ce_stage(ce
);
992 const char *name
= ce
->name
;
994 while (pos
< istate
->cache_nr
) {
995 struct cache_entry
*p
= istate
->cache
[pos
++];
997 if (len
>= ce_namelen(p
))
999 if (memcmp(name
, p
->name
, len
))
1001 if (ce_stage(p
) != stage
)
1003 if (p
->name
[len
] != '/')
1005 if (p
->ce_flags
& CE_REMOVE
)
1010 remove_index_entry_at(istate
, --pos
);
1017 * Like strcmp(), but also return the offset of the first change.
1018 * If strings are equal, return the length.
1020 int strcmp_offset(const char *s1
, const char *s2
, size_t *first_change
)
1025 return strcmp(s1
, s2
);
1027 for (k
= 0; s1
[k
] == s2
[k
]; k
++)
1032 return (unsigned char)s1
[k
] - (unsigned char)s2
[k
];
1036 * Do we have another file with a pathname that is a proper
1037 * subset of the name we're trying to add?
1039 * That is, is there another file in the index with a path
1040 * that matches a sub-directory in the given entry?
1042 static int has_dir_name(struct index_state
*istate
,
1043 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
1046 int stage
= ce_stage(ce
);
1047 const char *name
= ce
->name
;
1048 const char *slash
= name
+ ce_namelen(ce
);
1053 * We are frequently called during an iteration on a sorted
1054 * list of pathnames and while building a new index. Therefore,
1055 * there is a high probability that this entry will eventually
1056 * be appended to the index, rather than inserted in the middle.
1057 * If we can confirm that, we can avoid binary searches on the
1058 * components of the pathname.
1060 * Compare the entry's full path with the last path in the index.
1062 if (istate
->cache_nr
> 0) {
1063 cmp_last
= strcmp_offset(name
,
1064 istate
->cache
[istate
->cache_nr
- 1]->name
,
1067 if (len_eq_last
== 0) {
1069 * The entry sorts AFTER the last one in the
1070 * index and their paths have no common prefix,
1071 * so there cannot be a F/D conflict.
1076 * The entry sorts AFTER the last one in the
1077 * index, but has a common prefix. Fall through
1078 * to the loop below to disect the entry's path
1079 * and see where the difference is.
1082 } else if (cmp_last
== 0) {
1084 * The entry exactly matches the last one in the
1085 * index, but because of multiple stage and CE_REMOVE
1086 * items, we fall through and let the regular search
1096 if (*--slash
== '/')
1098 if (slash
<= ce
->name
)
1105 * (len + 1) is a directory boundary (including
1106 * the trailing slash). And since the loop is
1107 * decrementing "slash", the first iteration is
1108 * the longest directory prefix; subsequent
1109 * iterations consider parent directories.
1112 if (len
+ 1 <= len_eq_last
) {
1114 * The directory prefix (including the trailing
1115 * slash) also appears as a prefix in the last
1116 * entry, so the remainder cannot collide (because
1117 * strcmp said the whole path was greater).
1122 * LT: last: xxx/file_A
1128 if (len
> len_eq_last
) {
1130 * This part of the directory prefix (excluding
1131 * the trailing slash) is longer than the known
1132 * equal portions, so this sub-directory cannot
1133 * collide with a file.
1141 if (istate
->cache_nr
> 0 &&
1142 ce_namelen(istate
->cache
[istate
->cache_nr
- 1]) > len
) {
1144 * The directory prefix lines up with part of
1145 * a longer file or directory name, but sorts
1146 * after it, so this sub-directory cannot
1147 * collide with a file.
1149 * last: xxx/yy-file (because '-' sorts before '/')
1156 * This is a possible collision. Fall through and
1157 * let the regular search code handle it.
1164 pos
= index_name_stage_pos(istate
, name
, len
, stage
);
1167 * Found one, but not so fast. This could
1168 * be a marker that says "I was here, but
1169 * I am being removed". Such an entry is
1170 * not a part of the resulting tree, and
1171 * it is Ok to have a directory at the same
1174 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
1178 remove_index_entry_at(istate
, pos
);
1186 * Trivial optimization: if we find an entry that
1187 * already matches the sub-directory, then we know
1188 * we're ok, and we can exit.
1190 while (pos
< istate
->cache_nr
) {
1191 struct cache_entry
*p
= istate
->cache
[pos
];
1192 if ((ce_namelen(p
) <= len
) ||
1193 (p
->name
[len
] != '/') ||
1194 memcmp(p
->name
, name
, len
))
1195 break; /* not our subdirectory */
1196 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
1198 * p is at the same stage as our entry, and
1199 * is a subdirectory of what we are looking
1200 * at, so we cannot have conflicts at our
1201 * level or anything shorter.
1210 /* We may be in a situation where we already have path/file and path
1211 * is being added, or we already have path and path/file is being
1212 * added. Either one would result in a nonsense tree that has path
1213 * twice when git-write-tree tries to write it out. Prevent it.
1215 * If ok-to-replace is specified, we remove the conflicting entries
1216 * from the cache so the caller should recompute the insert position.
1217 * When this happens, we return non-zero.
1219 static int check_file_directory_conflict(struct index_state
*istate
,
1220 const struct cache_entry
*ce
,
1221 int pos
, int ok_to_replace
)
1226 * When ce is an "I am going away" entry, we allow it to be added
1228 if (ce
->ce_flags
& CE_REMOVE
)
1232 * We check if the path is a sub-path of a subsequent pathname
1233 * first, since removing those will not change the position
1236 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
1239 * Then check if the path might have a clashing sub-directory
1242 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
1245 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1248 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
1249 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
1250 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
1251 int new_only
= option
& ADD_CACHE_NEW_ONLY
;
1253 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1254 cache_tree_invalidate_path(istate
, ce
->name
);
1257 * If this entry's path sorts after the last entry in the index,
1258 * we can avoid searching for it.
1260 if (istate
->cache_nr
> 0 &&
1261 strcmp(ce
->name
, istate
->cache
[istate
->cache_nr
- 1]->name
) > 0)
1262 pos
= -istate
->cache_nr
- 1;
1264 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1266 /* existing match? Just replace it. */
1269 replace_index_entry(istate
, pos
, ce
);
1274 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1275 untracked_cache_add_to_index(istate
, ce
->name
);
1278 * Inserting a merged entry ("stage 0") into the index
1279 * will always replace all non-merged entries..
1281 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
1282 while (ce_same_name(istate
->cache
[pos
], ce
)) {
1284 if (!remove_index_entry_at(istate
, pos
))
1291 if (!verify_path(ce
->name
, ce
->ce_mode
))
1292 return error("Invalid path '%s'", ce
->name
);
1294 if (!skip_df_check
&&
1295 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
1297 return error("'%s' appears as both a file and as a directory",
1299 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1305 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1309 if (option
& ADD_CACHE_JUST_APPEND
)
1310 pos
= istate
->cache_nr
;
1313 ret
= add_index_entry_with_check(istate
, ce
, option
);
1319 /* Make sure the array is big enough .. */
1320 ALLOC_GROW(istate
->cache
, istate
->cache_nr
+ 1, istate
->cache_alloc
);
1324 if (istate
->cache_nr
> pos
+ 1)
1325 MOVE_ARRAY(istate
->cache
+ pos
+ 1, istate
->cache
+ pos
,
1326 istate
->cache_nr
- pos
- 1);
1327 set_index_entry(istate
, pos
, ce
);
1328 istate
->cache_changed
|= CE_ENTRY_ADDED
;
1333 * "refresh" does not calculate a new sha1 file or bring the
1334 * cache up-to-date for mode/content changes. But what it
1335 * _does_ do is to "re-match" the stat information of a file
1336 * with the cache, so that you can refresh the cache for a
1337 * file that hasn't been changed but where the stat entry is
1340 * For example, you'd want to do this after doing a "git-read-tree",
1341 * to link up the stat cache details with the proper files.
1343 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
1344 struct cache_entry
*ce
,
1345 unsigned int options
, int *err
,
1349 struct cache_entry
*updated
;
1351 int refresh
= options
& CE_MATCH_REFRESH
;
1352 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
1353 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
1354 int ignore_missing
= options
& CE_MATCH_IGNORE_MISSING
;
1355 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
1357 if (!refresh
|| ce_uptodate(ce
))
1360 if (!ignore_fsmonitor
)
1361 refresh_fsmonitor(istate
);
1363 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1364 * that the change to the work tree does not matter and told
1367 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
)) {
1368 ce_mark_uptodate(ce
);
1371 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
)) {
1372 ce_mark_uptodate(ce
);
1375 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
)) {
1376 ce_mark_uptodate(ce
);
1380 if (has_symlink_leading_path(ce
->name
, ce_namelen(ce
))) {
1388 if (lstat(ce
->name
, &st
) < 0) {
1389 if (ignore_missing
&& errno
== ENOENT
)
1396 changed
= ie_match_stat(istate
, ce
, &st
, options
);
1398 *changed_ret
= changed
;
1401 * The path is unchanged. If we were told to ignore
1402 * valid bit, then we did the actual stat check and
1403 * found that the entry is unmodified. If the entry
1404 * is not marked VALID, this is the place to mark it
1405 * valid again, under "assume unchanged" mode.
1407 if (ignore_valid
&& assume_unchanged
&&
1408 !(ce
->ce_flags
& CE_VALID
))
1409 ; /* mark this one VALID again */
1412 * We do not mark the index itself "modified"
1413 * because CE_UPTODATE flag is in-core only;
1414 * we are not going to write this change out.
1416 if (!S_ISGITLINK(ce
->ce_mode
)) {
1417 ce_mark_uptodate(ce
);
1418 mark_fsmonitor_valid(ce
);
1424 if (ie_modified(istate
, ce
, &st
, options
)) {
1430 updated
= make_empty_cache_entry(istate
, ce_namelen(ce
));
1431 copy_cache_entry(updated
, ce
);
1432 memcpy(updated
->name
, ce
->name
, ce
->ce_namelen
+ 1);
1433 fill_stat_cache_info(updated
, &st
);
1435 * If ignore_valid is not set, we should leave CE_VALID bit
1436 * alone. Otherwise, paths marked with --no-assume-unchanged
1437 * (i.e. things to be edited) will reacquire CE_VALID bit
1438 * automatically, which is not really what we want.
1440 if (!ignore_valid
&& assume_unchanged
&&
1441 !(ce
->ce_flags
& CE_VALID
))
1442 updated
->ce_flags
&= ~CE_VALID
;
1444 /* istate->cache_changed is updated in the caller */
1448 static void show_file(const char * fmt
, const char * name
, int in_porcelain
,
1449 int * first
, const char *header_msg
)
1451 if (in_porcelain
&& *first
&& header_msg
) {
1452 printf("%s\n", header_msg
);
1458 int refresh_index(struct index_state
*istate
, unsigned int flags
,
1459 const struct pathspec
*pathspec
,
1460 char *seen
, const char *header_msg
)
1464 int really
= (flags
& REFRESH_REALLY
) != 0;
1465 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
1466 int quiet
= (flags
& REFRESH_QUIET
) != 0;
1467 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
1468 int ignore_submodules
= (flags
& REFRESH_IGNORE_SUBMODULES
) != 0;
1470 int in_porcelain
= (flags
& REFRESH_IN_PORCELAIN
);
1471 unsigned int options
= (CE_MATCH_REFRESH
|
1472 (really
? CE_MATCH_IGNORE_VALID
: 0) |
1473 (not_new
? CE_MATCH_IGNORE_MISSING
: 0));
1474 const char *modified_fmt
;
1475 const char *deleted_fmt
;
1476 const char *typechange_fmt
;
1477 const char *added_fmt
;
1478 const char *unmerged_fmt
;
1479 uint64_t start
= getnanotime();
1481 modified_fmt
= (in_porcelain
? "M\t%s\n" : "%s: needs update\n");
1482 deleted_fmt
= (in_porcelain
? "D\t%s\n" : "%s: needs update\n");
1483 typechange_fmt
= (in_porcelain
? "T\t%s\n" : "%s needs update\n");
1484 added_fmt
= (in_porcelain
? "A\t%s\n" : "%s needs update\n");
1485 unmerged_fmt
= (in_porcelain
? "U\t%s\n" : "%s: needs merge\n");
1486 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1487 struct cache_entry
*ce
, *new_entry
;
1488 int cache_errno
= 0;
1492 ce
= istate
->cache
[i
];
1493 if (ignore_submodules
&& S_ISGITLINK(ce
->ce_mode
))
1496 if (pathspec
&& !ce_path_match(&the_index
, ce
, pathspec
, seen
))
1500 while ((i
< istate
->cache_nr
) &&
1501 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
1507 show_file(unmerged_fmt
, ce
->name
, in_porcelain
,
1508 &first
, header_msg
);
1516 new_entry
= refresh_cache_ent(istate
, ce
, options
, &cache_errno
, &changed
);
1517 if (new_entry
== ce
)
1522 if (really
&& cache_errno
== EINVAL
) {
1523 /* If we are doing --really-refresh that
1524 * means the index is not valid anymore.
1526 ce
->ce_flags
&= ~CE_VALID
;
1527 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
1528 mark_fsmonitor_invalid(istate
, ce
);
1529 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
1534 if (cache_errno
== ENOENT
)
1536 else if (ce_intent_to_add(ce
))
1537 fmt
= added_fmt
; /* must be before other checks */
1538 else if (changed
& TYPE_CHANGED
)
1539 fmt
= typechange_fmt
;
1543 ce
->name
, in_porcelain
, &first
, header_msg
);
1548 replace_index_entry(istate
, i
, new_entry
);
1550 trace_performance_since(start
, "refresh index");
1554 struct cache_entry
*refresh_cache_entry(struct index_state
*istate
,
1555 struct cache_entry
*ce
,
1556 unsigned int options
)
1558 return refresh_cache_ent(istate
, ce
, options
, NULL
, NULL
);
1562 /*****************************************************************
1564 *****************************************************************/
1566 #define INDEX_FORMAT_DEFAULT 3
1568 static unsigned int get_index_format_default(void)
1570 char *envversion
= getenv("GIT_INDEX_VERSION");
1573 unsigned int version
= INDEX_FORMAT_DEFAULT
;
1576 if (!git_config_get_int("index.version", &value
))
1578 if (version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1579 warning(_("index.version set, but the value is invalid.\n"
1580 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1581 return INDEX_FORMAT_DEFAULT
;
1586 version
= strtoul(envversion
, &endp
, 10);
1588 version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1589 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1590 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1591 version
= INDEX_FORMAT_DEFAULT
;
1597 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1598 * Again - this is just a (very strong in practice) heuristic that
1599 * the inode hasn't changed.
1601 * We save the fields in big-endian order to allow using the
1602 * index file over NFS transparently.
1604 struct ondisk_cache_entry
{
1605 struct cache_time ctime
;
1606 struct cache_time mtime
;
1613 unsigned char sha1
[20];
1615 char name
[FLEX_ARRAY
]; /* more */
1619 * This struct is used when CE_EXTENDED bit is 1
1620 * The struct must match ondisk_cache_entry exactly from
1623 struct ondisk_cache_entry_extended
{
1624 struct cache_time ctime
;
1625 struct cache_time mtime
;
1632 unsigned char sha1
[20];
1635 char name
[FLEX_ARRAY
]; /* more */
1638 /* These are only used for v3 or lower */
1639 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1640 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1641 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1642 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1643 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1644 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1645 ondisk_cache_entry_size(ce_namelen(ce)))
1647 /* Allow fsck to force verification of the index checksum. */
1648 int verify_index_checksum
;
1650 /* Allow fsck to force verification of the cache entry order. */
1651 int verify_ce_order
;
1653 static int verify_hdr(struct cache_header
*hdr
, unsigned long size
)
1656 unsigned char hash
[GIT_MAX_RAWSZ
];
1659 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
1660 return error("bad signature");
1661 hdr_version
= ntohl(hdr
->hdr_version
);
1662 if (hdr_version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< hdr_version
)
1663 return error("bad index version %d", hdr_version
);
1665 if (!verify_index_checksum
)
1668 the_hash_algo
->init_fn(&c
);
1669 the_hash_algo
->update_fn(&c
, hdr
, size
- the_hash_algo
->rawsz
);
1670 the_hash_algo
->final_fn(hash
, &c
);
1671 if (hashcmp(hash
, (unsigned char *)hdr
+ size
- the_hash_algo
->rawsz
))
1672 return error("bad index file sha1 signature");
1676 static int read_index_extension(struct index_state
*istate
,
1677 const char *ext
, void *data
, unsigned long sz
)
1679 switch (CACHE_EXT(ext
)) {
1680 case CACHE_EXT_TREE
:
1681 istate
->cache_tree
= cache_tree_read(data
, sz
);
1683 case CACHE_EXT_RESOLVE_UNDO
:
1684 istate
->resolve_undo
= resolve_undo_read(data
, sz
);
1686 case CACHE_EXT_LINK
:
1687 if (read_link_extension(istate
, data
, sz
))
1690 case CACHE_EXT_UNTRACKED
:
1691 istate
->untracked
= read_untracked_extension(data
, sz
);
1693 case CACHE_EXT_FSMONITOR
:
1694 read_fsmonitor_extension(istate
, data
, sz
);
1697 if (*ext
< 'A' || 'Z' < *ext
)
1698 return error("index uses %.4s extension, which we do not understand",
1700 fprintf(stderr
, "ignoring %.4s extension\n", ext
);
1706 int hold_locked_index(struct lock_file
*lk
, int lock_flags
)
1708 return hold_lock_file_for_update(lk
, get_index_file(), lock_flags
);
1711 int read_index(struct index_state
*istate
)
1713 return read_index_from(istate
, get_index_file(), get_git_dir());
1716 static struct cache_entry
*cache_entry_from_ondisk(struct mem_pool
*mem_pool
,
1717 struct ondisk_cache_entry
*ondisk
,
1722 struct cache_entry
*ce
= mem_pool__ce_alloc(mem_pool
, len
);
1724 ce
->ce_stat_data
.sd_ctime
.sec
= get_be32(&ondisk
->ctime
.sec
);
1725 ce
->ce_stat_data
.sd_mtime
.sec
= get_be32(&ondisk
->mtime
.sec
);
1726 ce
->ce_stat_data
.sd_ctime
.nsec
= get_be32(&ondisk
->ctime
.nsec
);
1727 ce
->ce_stat_data
.sd_mtime
.nsec
= get_be32(&ondisk
->mtime
.nsec
);
1728 ce
->ce_stat_data
.sd_dev
= get_be32(&ondisk
->dev
);
1729 ce
->ce_stat_data
.sd_ino
= get_be32(&ondisk
->ino
);
1730 ce
->ce_mode
= get_be32(&ondisk
->mode
);
1731 ce
->ce_stat_data
.sd_uid
= get_be32(&ondisk
->uid
);
1732 ce
->ce_stat_data
.sd_gid
= get_be32(&ondisk
->gid
);
1733 ce
->ce_stat_data
.sd_size
= get_be32(&ondisk
->size
);
1734 ce
->ce_flags
= flags
& ~CE_NAMEMASK
;
1735 ce
->ce_namelen
= len
;
1737 hashcpy(ce
->oid
.hash
, ondisk
->sha1
);
1738 memcpy(ce
->name
, name
, len
);
1739 ce
->name
[len
] = '\0';
1744 * Adjacent cache entries tend to share the leading paths, so it makes
1745 * sense to only store the differences in later entries. In the v4
1746 * on-disk format of the index, each on-disk cache entry stores the
1747 * number of bytes to be stripped from the end of the previous name,
1748 * and the bytes to append to the result, to come up with its name.
1750 static unsigned long expand_name_field(struct strbuf
*name
, const char *cp_
)
1752 const unsigned char *ep
, *cp
= (const unsigned char *)cp_
;
1753 size_t len
= decode_varint(&cp
);
1755 if (name
->len
< len
)
1756 die("malformed name field in the index");
1757 strbuf_remove(name
, name
->len
- len
, len
);
1758 for (ep
= cp
; *ep
; ep
++)
1759 ; /* find the end */
1760 strbuf_add(name
, cp
, ep
- cp
);
1761 return (const char *)ep
+ 1 - cp_
;
1764 static struct cache_entry
*create_from_disk(struct mem_pool
*mem_pool
,
1765 struct ondisk_cache_entry
*ondisk
,
1766 unsigned long *ent_size
,
1767 struct strbuf
*previous_name
)
1769 struct cache_entry
*ce
;
1774 /* On-disk flags are just 16 bits */
1775 flags
= get_be16(&ondisk
->flags
);
1776 len
= flags
& CE_NAMEMASK
;
1778 if (flags
& CE_EXTENDED
) {
1779 struct ondisk_cache_entry_extended
*ondisk2
;
1781 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
1782 extended_flags
= get_be16(&ondisk2
->flags2
) << 16;
1783 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1784 if (extended_flags
& ~CE_EXTENDED_FLAGS
)
1785 die("Unknown index entry format %08x", extended_flags
);
1786 flags
|= extended_flags
;
1787 name
= ondisk2
->name
;
1790 name
= ondisk
->name
;
1792 if (!previous_name
) {
1793 /* v3 and earlier */
1794 if (len
== CE_NAMEMASK
)
1796 ce
= cache_entry_from_ondisk(mem_pool
, ondisk
, flags
, name
, len
);
1798 *ent_size
= ondisk_ce_size(ce
);
1800 unsigned long consumed
;
1801 consumed
= expand_name_field(previous_name
, name
);
1802 ce
= cache_entry_from_ondisk(mem_pool
, ondisk
, flags
,
1804 previous_name
->len
);
1806 *ent_size
= (name
- ((char *)ondisk
)) + consumed
;
1811 static void check_ce_order(struct index_state
*istate
)
1815 if (!verify_ce_order
)
1818 for (i
= 1; i
< istate
->cache_nr
; i
++) {
1819 struct cache_entry
*ce
= istate
->cache
[i
- 1];
1820 struct cache_entry
*next_ce
= istate
->cache
[i
];
1821 int name_compare
= strcmp(ce
->name
, next_ce
->name
);
1823 if (0 < name_compare
)
1824 die("unordered stage entries in index");
1825 if (!name_compare
) {
1827 die("multiple stage entries for merged file '%s'",
1829 if (ce_stage(ce
) > ce_stage(next_ce
))
1830 die("unordered stage entries for '%s'",
1836 static void tweak_untracked_cache(struct index_state
*istate
)
1838 switch (git_config_get_untracked_cache()) {
1839 case -1: /* keep: do nothing */
1842 remove_untracked_cache(istate
);
1845 add_untracked_cache(istate
);
1847 default: /* unknown value: do nothing */
1852 static void tweak_split_index(struct index_state
*istate
)
1854 switch (git_config_get_split_index()) {
1855 case -1: /* unset: do nothing */
1858 remove_split_index(istate
);
1861 add_split_index(istate
);
1863 default: /* unknown value: do nothing */
1868 static void post_read_index_from(struct index_state
*istate
)
1870 check_ce_order(istate
);
1871 tweak_untracked_cache(istate
);
1872 tweak_split_index(istate
);
1873 tweak_fsmonitor(istate
);
1876 static size_t estimate_cache_size_from_compressed(unsigned int entries
)
1878 return entries
* (sizeof(struct cache_entry
) + CACHE_ENTRY_PATH_LENGTH
);
1881 static size_t estimate_cache_size(size_t ondisk_size
, unsigned int entries
)
1883 long per_entry
= sizeof(struct cache_entry
) - sizeof(struct ondisk_cache_entry
);
1886 * Account for potential alignment differences.
1888 per_entry
+= align_padding_size(sizeof(struct cache_entry
), -sizeof(struct ondisk_cache_entry
));
1889 return ondisk_size
+ entries
* per_entry
;
1892 /* remember to discard_cache() before reading a different cache! */
1893 int do_read_index(struct index_state
*istate
, const char *path
, int must_exist
)
1897 unsigned long src_offset
;
1898 struct cache_header
*hdr
;
1901 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
1903 if (istate
->initialized
)
1904 return istate
->cache_nr
;
1906 istate
->timestamp
.sec
= 0;
1907 istate
->timestamp
.nsec
= 0;
1908 fd
= open(path
, O_RDONLY
);
1910 if (!must_exist
&& errno
== ENOENT
)
1912 die_errno("%s: index file open failed", path
);
1916 die_errno("cannot stat the open index");
1918 mmap_size
= xsize_t(st
.st_size
);
1919 if (mmap_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
1920 die("index file smaller than expected");
1922 mmap
= xmmap(NULL
, mmap_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1923 if (mmap
== MAP_FAILED
)
1924 die_errno("unable to map index file");
1928 if (verify_hdr(hdr
, mmap_size
) < 0)
1931 hashcpy(istate
->oid
.hash
, (const unsigned char *)hdr
+ mmap_size
- the_hash_algo
->rawsz
);
1932 istate
->version
= ntohl(hdr
->hdr_version
);
1933 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
1934 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
1935 istate
->cache
= xcalloc(istate
->cache_alloc
, sizeof(*istate
->cache
));
1936 istate
->initialized
= 1;
1938 if (istate
->version
== 4) {
1939 previous_name
= &previous_name_buf
;
1940 mem_pool_init(&istate
->ce_mem_pool
,
1941 estimate_cache_size_from_compressed(istate
->cache_nr
));
1943 previous_name
= NULL
;
1944 mem_pool_init(&istate
->ce_mem_pool
,
1945 estimate_cache_size(mmap_size
, istate
->cache_nr
));
1948 src_offset
= sizeof(*hdr
);
1949 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1950 struct ondisk_cache_entry
*disk_ce
;
1951 struct cache_entry
*ce
;
1952 unsigned long consumed
;
1954 disk_ce
= (struct ondisk_cache_entry
*)((char *)mmap
+ src_offset
);
1955 ce
= create_from_disk(istate
->ce_mem_pool
, disk_ce
, &consumed
, previous_name
);
1956 set_index_entry(istate
, i
, ce
);
1958 src_offset
+= consumed
;
1960 strbuf_release(&previous_name_buf
);
1961 istate
->timestamp
.sec
= st
.st_mtime
;
1962 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
1964 while (src_offset
<= mmap_size
- the_hash_algo
->rawsz
- 8) {
1965 /* After an array of active_nr index entries,
1966 * there can be arbitrary number of extended
1967 * sections, each of which is prefixed with
1968 * extension name (4-byte) and section length
1969 * in 4-byte network byte order.
1972 memcpy(&extsize
, (char *)mmap
+ src_offset
+ 4, 4);
1973 extsize
= ntohl(extsize
);
1974 if (read_index_extension(istate
,
1975 (const char *) mmap
+ src_offset
,
1976 (char *) mmap
+ src_offset
+ 8,
1980 src_offset
+= extsize
;
1982 munmap(mmap
, mmap_size
);
1983 return istate
->cache_nr
;
1986 munmap(mmap
, mmap_size
);
1987 die("index file corrupt");
1991 * Signal that the shared index is used by updating its mtime.
1993 * This way, shared index can be removed if they have not been used
1996 static void freshen_shared_index(const char *shared_index
, int warn
)
1998 if (!check_and_freshen_file(shared_index
, 1) && warn
)
1999 warning("could not freshen shared index '%s'", shared_index
);
2002 int read_index_from(struct index_state
*istate
, const char *path
,
2005 uint64_t start
= getnanotime();
2006 struct split_index
*split_index
;
2011 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2012 if (istate
->initialized
)
2013 return istate
->cache_nr
;
2015 ret
= do_read_index(istate
, path
, 0);
2016 trace_performance_since(start
, "read cache %s", path
);
2018 split_index
= istate
->split_index
;
2019 if (!split_index
|| is_null_oid(&split_index
->base_oid
)) {
2020 post_read_index_from(istate
);
2024 if (split_index
->base
)
2025 discard_index(split_index
->base
);
2027 split_index
->base
= xcalloc(1, sizeof(*split_index
->base
));
2029 base_oid_hex
= oid_to_hex(&split_index
->base_oid
);
2030 base_path
= xstrfmt("%s/sharedindex.%s", gitdir
, base_oid_hex
);
2031 ret
= do_read_index(split_index
->base
, base_path
, 1);
2032 if (oidcmp(&split_index
->base_oid
, &split_index
->base
->oid
))
2033 die("broken index, expect %s in %s, got %s",
2034 base_oid_hex
, base_path
,
2035 oid_to_hex(&split_index
->base
->oid
));
2037 freshen_shared_index(base_path
, 0);
2038 merge_base_index(istate
);
2039 post_read_index_from(istate
);
2040 trace_performance_since(start
, "read cache %s", base_path
);
2045 int is_index_unborn(struct index_state
*istate
)
2047 return (!istate
->cache_nr
&& !istate
->timestamp
.sec
);
2050 int discard_index(struct index_state
*istate
)
2053 * Cache entries in istate->cache[] should have been allocated
2054 * from the memory pool associated with this index, or from an
2055 * associated split_index. There is no need to free individual
2056 * cache entries. validate_cache_entries can detect when this
2057 * assertion does not hold.
2059 validate_cache_entries(istate
);
2061 resolve_undo_clear_index(istate
);
2062 istate
->cache_nr
= 0;
2063 istate
->cache_changed
= 0;
2064 istate
->timestamp
.sec
= 0;
2065 istate
->timestamp
.nsec
= 0;
2066 free_name_hash(istate
);
2067 cache_tree_free(&(istate
->cache_tree
));
2068 istate
->initialized
= 0;
2069 FREE_AND_NULL(istate
->cache
);
2070 istate
->cache_alloc
= 0;
2071 discard_split_index(istate
);
2072 free_untracked_cache(istate
->untracked
);
2073 istate
->untracked
= NULL
;
2075 if (istate
->ce_mem_pool
) {
2076 mem_pool_discard(istate
->ce_mem_pool
, should_validate_cache_entries());
2077 istate
->ce_mem_pool
= NULL
;
2084 * Validate the cache entries of this index.
2085 * All cache entries associated with this index
2086 * should have been allocated by the memory pool
2087 * associated with this index, or by a referenced
2090 void validate_cache_entries(const struct index_state
*istate
)
2094 if (!should_validate_cache_entries() ||!istate
|| !istate
->initialized
)
2097 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2099 die("internal error: cache entry is not allocated from expected memory pool");
2100 } else if (!istate
->ce_mem_pool
||
2101 !mem_pool_contains(istate
->ce_mem_pool
, istate
->cache
[i
])) {
2102 if (!istate
->split_index
||
2103 !istate
->split_index
->base
||
2104 !istate
->split_index
->base
->ce_mem_pool
||
2105 !mem_pool_contains(istate
->split_index
->base
->ce_mem_pool
, istate
->cache
[i
])) {
2106 die("internal error: cache entry is not allocated from expected memory pool");
2111 if (istate
->split_index
)
2112 validate_cache_entries(istate
->split_index
->base
);
2115 int unmerged_index(const struct index_state
*istate
)
2118 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2119 if (ce_stage(istate
->cache
[i
]))
2125 int index_has_changes(const struct index_state
*istate
,
2129 struct object_id cmp
;
2132 if (istate
!= &the_index
) {
2133 BUG("index_has_changes cannot yet accept istate != &the_index; do_diff_cache needs updating first.");
2136 cmp
= tree
->object
.oid
;
2137 if (tree
|| !get_oid_tree("HEAD", &cmp
)) {
2138 struct diff_options opt
;
2141 opt
.flags
.exit_with_status
= 1;
2143 opt
.flags
.quick
= 1;
2144 do_diff_cache(&cmp
, &opt
);
2146 for (i
= 0; sb
&& i
< diff_queued_diff
.nr
; i
++) {
2148 strbuf_addch(sb
, ' ');
2149 strbuf_addstr(sb
, diff_queued_diff
.queue
[i
]->two
->path
);
2152 return opt
.flags
.has_changes
!= 0;
2154 for (i
= 0; sb
&& i
< istate
->cache_nr
; i
++) {
2156 strbuf_addch(sb
, ' ');
2157 strbuf_addstr(sb
, istate
->cache
[i
]->name
);
2159 return !!istate
->cache_nr
;
2163 #define WRITE_BUFFER_SIZE 8192
2164 static unsigned char write_buffer
[WRITE_BUFFER_SIZE
];
2165 static unsigned long write_buffer_len
;
2167 static int ce_write_flush(git_hash_ctx
*context
, int fd
)
2169 unsigned int buffered
= write_buffer_len
;
2171 the_hash_algo
->update_fn(context
, write_buffer
, buffered
);
2172 if (write_in_full(fd
, write_buffer
, buffered
) < 0)
2174 write_buffer_len
= 0;
2179 static int ce_write(git_hash_ctx
*context
, int fd
, void *data
, unsigned int len
)
2182 unsigned int buffered
= write_buffer_len
;
2183 unsigned int partial
= WRITE_BUFFER_SIZE
- buffered
;
2186 memcpy(write_buffer
+ buffered
, data
, partial
);
2187 buffered
+= partial
;
2188 if (buffered
== WRITE_BUFFER_SIZE
) {
2189 write_buffer_len
= buffered
;
2190 if (ce_write_flush(context
, fd
))
2194 write_buffer_len
= buffered
;
2196 data
= (char *) data
+ partial
;
2201 static int write_index_ext_header(git_hash_ctx
*context
, int fd
,
2202 unsigned int ext
, unsigned int sz
)
2206 return ((ce_write(context
, fd
, &ext
, 4) < 0) ||
2207 (ce_write(context
, fd
, &sz
, 4) < 0)) ? -1 : 0;
2210 static int ce_flush(git_hash_ctx
*context
, int fd
, unsigned char *hash
)
2212 unsigned int left
= write_buffer_len
;
2215 write_buffer_len
= 0;
2216 the_hash_algo
->update_fn(context
, write_buffer
, left
);
2219 /* Flush first if not enough space for hash signature */
2220 if (left
+ the_hash_algo
->rawsz
> WRITE_BUFFER_SIZE
) {
2221 if (write_in_full(fd
, write_buffer
, left
) < 0)
2226 /* Append the hash signature at the end */
2227 the_hash_algo
->final_fn(write_buffer
+ left
, context
);
2228 hashcpy(hash
, write_buffer
+ left
);
2229 left
+= the_hash_algo
->rawsz
;
2230 return (write_in_full(fd
, write_buffer
, left
) < 0) ? -1 : 0;
2233 static void ce_smudge_racily_clean_entry(struct cache_entry
*ce
)
2236 * The only thing we care about in this function is to smudge the
2237 * falsely clean entry due to touch-update-touch race, so we leave
2238 * everything else as they are. We are called for entries whose
2239 * ce_stat_data.sd_mtime match the index file mtime.
2241 * Note that this actually does not do much for gitlinks, for
2242 * which ce_match_stat_basic() always goes to the actual
2243 * contents. The caller checks with is_racy_timestamp() which
2244 * always says "no" for gitlinks, so we are not called for them ;-)
2248 if (lstat(ce
->name
, &st
) < 0)
2250 if (ce_match_stat_basic(ce
, &st
))
2252 if (ce_modified_check_fs(ce
, &st
)) {
2253 /* This is "racily clean"; smudge it. Note that this
2254 * is a tricky code. At first glance, it may appear
2255 * that it can break with this sequence:
2257 * $ echo xyzzy >frotz
2258 * $ git-update-index --add frotz
2261 * $ echo filfre >nitfol
2262 * $ git-update-index --add nitfol
2264 * but it does not. When the second update-index runs,
2265 * it notices that the entry "frotz" has the same timestamp
2266 * as index, and if we were to smudge it by resetting its
2267 * size to zero here, then the object name recorded
2268 * in index is the 6-byte file but the cached stat information
2269 * becomes zero --- which would then match what we would
2270 * obtain from the filesystem next time we stat("frotz").
2272 * However, the second update-index, before calling
2273 * this function, notices that the cached size is 6
2274 * bytes and what is on the filesystem is an empty
2275 * file, and never calls us, so the cached size information
2276 * for "frotz" stays 6 which does not match the filesystem.
2278 ce
->ce_stat_data
.sd_size
= 0;
2282 /* Copy miscellaneous fields but not the name */
2283 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry
*ondisk
,
2284 struct cache_entry
*ce
)
2288 ondisk
->ctime
.sec
= htonl(ce
->ce_stat_data
.sd_ctime
.sec
);
2289 ondisk
->mtime
.sec
= htonl(ce
->ce_stat_data
.sd_mtime
.sec
);
2290 ondisk
->ctime
.nsec
= htonl(ce
->ce_stat_data
.sd_ctime
.nsec
);
2291 ondisk
->mtime
.nsec
= htonl(ce
->ce_stat_data
.sd_mtime
.nsec
);
2292 ondisk
->dev
= htonl(ce
->ce_stat_data
.sd_dev
);
2293 ondisk
->ino
= htonl(ce
->ce_stat_data
.sd_ino
);
2294 ondisk
->mode
= htonl(ce
->ce_mode
);
2295 ondisk
->uid
= htonl(ce
->ce_stat_data
.sd_uid
);
2296 ondisk
->gid
= htonl(ce
->ce_stat_data
.sd_gid
);
2297 ondisk
->size
= htonl(ce
->ce_stat_data
.sd_size
);
2298 hashcpy(ondisk
->sha1
, ce
->oid
.hash
);
2300 flags
= ce
->ce_flags
& ~CE_NAMEMASK
;
2301 flags
|= (ce_namelen(ce
) >= CE_NAMEMASK
? CE_NAMEMASK
: ce_namelen(ce
));
2302 ondisk
->flags
= htons(flags
);
2303 if (ce
->ce_flags
& CE_EXTENDED
) {
2304 struct ondisk_cache_entry_extended
*ondisk2
;
2305 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
2306 ondisk2
->flags2
= htons((ce
->ce_flags
& CE_EXTENDED_FLAGS
) >> 16);
2310 static int ce_write_entry(git_hash_ctx
*c
, int fd
, struct cache_entry
*ce
,
2311 struct strbuf
*previous_name
, struct ondisk_cache_entry
*ondisk
)
2315 unsigned int saved_namelen
;
2316 int stripped_name
= 0;
2317 static unsigned char padding
[8] = { 0x00 };
2319 if (ce
->ce_flags
& CE_STRIP_NAME
) {
2320 saved_namelen
= ce_namelen(ce
);
2325 if (ce
->ce_flags
& CE_EXTENDED
)
2326 size
= offsetof(struct ondisk_cache_entry_extended
, name
);
2328 size
= offsetof(struct ondisk_cache_entry
, name
);
2330 if (!previous_name
) {
2331 int len
= ce_namelen(ce
);
2332 copy_cache_entry_to_ondisk(ondisk
, ce
);
2333 result
= ce_write(c
, fd
, ondisk
, size
);
2335 result
= ce_write(c
, fd
, ce
->name
, len
);
2337 result
= ce_write(c
, fd
, padding
, align_padding_size(size
, len
));
2339 int common
, to_remove
, prefix_size
;
2340 unsigned char to_remove_vi
[16];
2342 (ce
->name
[common
] &&
2343 common
< previous_name
->len
&&
2344 ce
->name
[common
] == previous_name
->buf
[common
]);
2346 ; /* still matching */
2347 to_remove
= previous_name
->len
- common
;
2348 prefix_size
= encode_varint(to_remove
, to_remove_vi
);
2350 copy_cache_entry_to_ondisk(ondisk
, ce
);
2351 result
= ce_write(c
, fd
, ondisk
, size
);
2353 result
= ce_write(c
, fd
, to_remove_vi
, prefix_size
);
2355 result
= ce_write(c
, fd
, ce
->name
+ common
, ce_namelen(ce
) - common
);
2357 result
= ce_write(c
, fd
, padding
, 1);
2359 strbuf_splice(previous_name
, common
, to_remove
,
2360 ce
->name
+ common
, ce_namelen(ce
) - common
);
2362 if (stripped_name
) {
2363 ce
->ce_namelen
= saved_namelen
;
2364 ce
->ce_flags
&= ~CE_STRIP_NAME
;
2371 * This function verifies if index_state has the correct sha1 of the
2372 * index file. Don't die if we have any other failure, just return 0.
2374 static int verify_index_from(const struct index_state
*istate
, const char *path
)
2379 unsigned char hash
[GIT_MAX_RAWSZ
];
2381 if (!istate
->initialized
)
2384 fd
= open(path
, O_RDONLY
);
2391 if (st
.st_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2394 n
= pread_in_full(fd
, hash
, the_hash_algo
->rawsz
, st
.st_size
- the_hash_algo
->rawsz
);
2395 if (n
!= the_hash_algo
->rawsz
)
2398 if (hashcmp(istate
->oid
.hash
, hash
))
2409 static int verify_index(const struct index_state
*istate
)
2411 return verify_index_from(istate
, get_index_file());
2414 static int has_racy_timestamp(struct index_state
*istate
)
2416 int entries
= istate
->cache_nr
;
2419 for (i
= 0; i
< entries
; i
++) {
2420 struct cache_entry
*ce
= istate
->cache
[i
];
2421 if (is_racy_timestamp(istate
, ce
))
2427 void update_index_if_able(struct index_state
*istate
, struct lock_file
*lockfile
)
2429 if ((istate
->cache_changed
|| has_racy_timestamp(istate
)) &&
2430 verify_index(istate
))
2431 write_locked_index(istate
, lockfile
, COMMIT_LOCK
);
2433 rollback_lock_file(lockfile
);
2437 * On success, `tempfile` is closed. If it is the temporary file
2438 * of a `struct lock_file`, we will therefore effectively perform
2439 * a 'close_lock_file_gently()`. Since that is an implementation
2440 * detail of lockfiles, callers of `do_write_index()` should not
2443 static int do_write_index(struct index_state
*istate
, struct tempfile
*tempfile
,
2444 int strip_extensions
)
2446 uint64_t start
= getnanotime();
2447 int newfd
= tempfile
->fd
;
2449 struct cache_header hdr
;
2450 int i
, err
= 0, removed
, extended
, hdr_version
;
2451 struct cache_entry
**cache
= istate
->cache
;
2452 int entries
= istate
->cache_nr
;
2454 struct ondisk_cache_entry_extended ondisk
;
2455 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
2456 int drop_cache_tree
= istate
->drop_cache_tree
;
2458 for (i
= removed
= extended
= 0; i
< entries
; i
++) {
2459 if (cache
[i
]->ce_flags
& CE_REMOVE
)
2462 /* reduce extended entries if possible */
2463 cache
[i
]->ce_flags
&= ~CE_EXTENDED
;
2464 if (cache
[i
]->ce_flags
& CE_EXTENDED_FLAGS
) {
2466 cache
[i
]->ce_flags
|= CE_EXTENDED
;
2470 if (!istate
->version
) {
2471 istate
->version
= get_index_format_default();
2472 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0))
2473 init_split_index(istate
);
2476 /* demote version 3 to version 2 when the latter suffices */
2477 if (istate
->version
== 3 || istate
->version
== 2)
2478 istate
->version
= extended
? 3 : 2;
2480 hdr_version
= istate
->version
;
2482 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
2483 hdr
.hdr_version
= htonl(hdr_version
);
2484 hdr
.hdr_entries
= htonl(entries
- removed
);
2486 the_hash_algo
->init_fn(&c
);
2487 if (ce_write(&c
, newfd
, &hdr
, sizeof(hdr
)) < 0)
2490 previous_name
= (hdr_version
== 4) ? &previous_name_buf
: NULL
;
2492 for (i
= 0; i
< entries
; i
++) {
2493 struct cache_entry
*ce
= cache
[i
];
2494 if (ce
->ce_flags
& CE_REMOVE
)
2496 if (!ce_uptodate(ce
) && is_racy_timestamp(istate
, ce
))
2497 ce_smudge_racily_clean_entry(ce
);
2498 if (is_null_oid(&ce
->oid
)) {
2499 static const char msg
[] = "cache entry has null sha1: %s";
2500 static int allow
= -1;
2503 allow
= git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2505 warning(msg
, ce
->name
);
2507 err
= error(msg
, ce
->name
);
2509 drop_cache_tree
= 1;
2511 if (ce_write_entry(&c
, newfd
, ce
, previous_name
, (struct ondisk_cache_entry
*)&ondisk
) < 0)
2517 strbuf_release(&previous_name_buf
);
2522 /* Write extension data here */
2523 if (!strip_extensions
&& istate
->split_index
) {
2524 struct strbuf sb
= STRBUF_INIT
;
2526 err
= write_link_extension(&sb
, istate
) < 0 ||
2527 write_index_ext_header(&c
, newfd
, CACHE_EXT_LINK
,
2529 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2530 strbuf_release(&sb
);
2534 if (!strip_extensions
&& !drop_cache_tree
&& istate
->cache_tree
) {
2535 struct strbuf sb
= STRBUF_INIT
;
2537 cache_tree_write(&sb
, istate
->cache_tree
);
2538 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_TREE
, sb
.len
) < 0
2539 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2540 strbuf_release(&sb
);
2544 if (!strip_extensions
&& istate
->resolve_undo
) {
2545 struct strbuf sb
= STRBUF_INIT
;
2547 resolve_undo_write(&sb
, istate
->resolve_undo
);
2548 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_RESOLVE_UNDO
,
2550 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2551 strbuf_release(&sb
);
2555 if (!strip_extensions
&& istate
->untracked
) {
2556 struct strbuf sb
= STRBUF_INIT
;
2558 write_untracked_extension(&sb
, istate
->untracked
);
2559 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_UNTRACKED
,
2561 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2562 strbuf_release(&sb
);
2566 if (!strip_extensions
&& istate
->fsmonitor_last_update
) {
2567 struct strbuf sb
= STRBUF_INIT
;
2569 write_fsmonitor_extension(&sb
, istate
);
2570 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_FSMONITOR
, sb
.len
) < 0
2571 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2572 strbuf_release(&sb
);
2577 if (ce_flush(&c
, newfd
, istate
->oid
.hash
))
2579 if (close_tempfile_gently(tempfile
)) {
2580 error(_("could not close '%s'"), tempfile
->filename
.buf
);
2583 if (stat(tempfile
->filename
.buf
, &st
))
2585 istate
->timestamp
.sec
= (unsigned int)st
.st_mtime
;
2586 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
2587 trace_performance_since(start
, "write index, changed mask = %x", istate
->cache_changed
);
2591 void set_alternate_index_output(const char *name
)
2593 alternate_index_output
= name
;
2596 static int commit_locked_index(struct lock_file
*lk
)
2598 if (alternate_index_output
)
2599 return commit_lock_file_to(lk
, alternate_index_output
);
2601 return commit_lock_file(lk
);
2604 static int do_write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
2607 int ret
= do_write_index(istate
, lock
->tempfile
, 0);
2610 if (flags
& COMMIT_LOCK
)
2611 return commit_locked_index(lock
);
2612 return close_lock_file_gently(lock
);
2615 static int write_split_index(struct index_state
*istate
,
2616 struct lock_file
*lock
,
2620 prepare_to_write_split_index(istate
);
2621 ret
= do_write_locked_index(istate
, lock
, flags
);
2622 finish_writing_split_index(istate
);
2626 static const char *shared_index_expire
= "2.weeks.ago";
2628 static unsigned long get_shared_index_expire_date(void)
2630 static unsigned long shared_index_expire_date
;
2631 static int shared_index_expire_date_prepared
;
2633 if (!shared_index_expire_date_prepared
) {
2634 git_config_get_expiry("splitindex.sharedindexexpire",
2635 &shared_index_expire
);
2636 shared_index_expire_date
= approxidate(shared_index_expire
);
2637 shared_index_expire_date_prepared
= 1;
2640 return shared_index_expire_date
;
2643 static int should_delete_shared_index(const char *shared_index_path
)
2646 unsigned long expiration
;
2648 /* Check timestamp */
2649 expiration
= get_shared_index_expire_date();
2652 if (stat(shared_index_path
, &st
))
2653 return error_errno(_("could not stat '%s'"), shared_index_path
);
2654 if (st
.st_mtime
> expiration
)
2660 static int clean_shared_index_files(const char *current_hex
)
2663 DIR *dir
= opendir(get_git_dir());
2666 return error_errno(_("unable to open git dir: %s"), get_git_dir());
2668 while ((de
= readdir(dir
)) != NULL
) {
2669 const char *sha1_hex
;
2670 const char *shared_index_path
;
2671 if (!skip_prefix(de
->d_name
, "sharedindex.", &sha1_hex
))
2673 if (!strcmp(sha1_hex
, current_hex
))
2675 shared_index_path
= git_path("%s", de
->d_name
);
2676 if (should_delete_shared_index(shared_index_path
) > 0 &&
2677 unlink(shared_index_path
))
2678 warning_errno(_("unable to unlink: %s"), shared_index_path
);
2685 static int write_shared_index(struct index_state
*istate
,
2686 struct tempfile
**temp
)
2688 struct split_index
*si
= istate
->split_index
;
2691 move_cache_to_base_index(istate
);
2692 ret
= do_write_index(si
->base
, *temp
, 1);
2695 ret
= adjust_shared_perm(get_tempfile_path(*temp
));
2697 error("cannot fix permission bits on %s", get_tempfile_path(*temp
));
2700 ret
= rename_tempfile(temp
,
2701 git_path("sharedindex.%s", oid_to_hex(&si
->base
->oid
)));
2703 oidcpy(&si
->base_oid
, &si
->base
->oid
);
2704 clean_shared_index_files(oid_to_hex(&si
->base
->oid
));
2710 static const int default_max_percent_split_change
= 20;
2712 static int too_many_not_shared_entries(struct index_state
*istate
)
2714 int i
, not_shared
= 0;
2715 int max_split
= git_config_get_max_percent_split_change();
2717 switch (max_split
) {
2719 /* not or badly configured: use the default value */
2720 max_split
= default_max_percent_split_change
;
2723 return 1; /* 0% means always write a new shared index */
2725 return 0; /* 100% means never write a new shared index */
2727 break; /* just use the configured value */
2730 /* Count not shared entries */
2731 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2732 struct cache_entry
*ce
= istate
->cache
[i
];
2737 return (int64_t)istate
->cache_nr
* max_split
< (int64_t)not_shared
* 100;
2740 int write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
2743 int new_shared_index
, ret
;
2744 struct split_index
*si
= istate
->split_index
;
2746 if ((flags
& SKIP_IF_UNCHANGED
) && !istate
->cache_changed
) {
2747 if (flags
& COMMIT_LOCK
)
2748 rollback_lock_file(lock
);
2752 if (istate
->fsmonitor_last_update
)
2753 fill_fsmonitor_bitmap(istate
);
2755 if (!si
|| alternate_index_output
||
2756 (istate
->cache_changed
& ~EXTMASK
)) {
2758 oidclr(&si
->base_oid
);
2759 ret
= do_write_locked_index(istate
, lock
, flags
);
2763 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0)) {
2764 int v
= si
->base_oid
.hash
[0];
2766 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
2768 if (too_many_not_shared_entries(istate
))
2769 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
2771 new_shared_index
= istate
->cache_changed
& SPLIT_INDEX_ORDERED
;
2773 if (new_shared_index
) {
2774 struct tempfile
*temp
;
2777 temp
= mks_tempfile(git_path("sharedindex_XXXXXX"));
2779 oidclr(&si
->base_oid
);
2780 ret
= do_write_locked_index(istate
, lock
, flags
);
2783 ret
= write_shared_index(istate
, &temp
);
2785 saved_errno
= errno
;
2786 if (is_tempfile_active(temp
))
2787 delete_tempfile(&temp
);
2788 errno
= saved_errno
;
2794 ret
= write_split_index(istate
, lock
, flags
);
2796 /* Freshen the shared index only if the split-index was written */
2797 if (!ret
&& !new_shared_index
) {
2798 const char *shared_index
= git_path("sharedindex.%s",
2799 oid_to_hex(&si
->base_oid
));
2800 freshen_shared_index(shared_index
, 1);
2804 if (flags
& COMMIT_LOCK
)
2805 rollback_lock_file(lock
);
2810 * Read the index file that is potentially unmerged into given
2811 * index_state, dropping any unmerged entries to stage #0 (potentially
2812 * resulting in a path appearing as both a file and a directory in the
2813 * index; the caller is responsible to clear out the extra entries
2814 * before writing the index to a tree). Returns true if the index is
2815 * unmerged. Callers who want to refuse to work from an unmerged
2816 * state can call this and check its return value, instead of calling
2819 int read_index_unmerged(struct index_state
*istate
)
2825 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2826 struct cache_entry
*ce
= istate
->cache
[i
];
2827 struct cache_entry
*new_ce
;
2833 len
= ce_namelen(ce
);
2834 new_ce
= make_empty_cache_entry(istate
, len
);
2835 memcpy(new_ce
->name
, ce
->name
, len
);
2836 new_ce
->ce_flags
= create_ce_flags(0) | CE_CONFLICTED
;
2837 new_ce
->ce_namelen
= len
;
2838 new_ce
->ce_mode
= ce
->ce_mode
;
2839 if (add_index_entry(istate
, new_ce
, ADD_CACHE_SKIP_DFCHECK
))
2840 return error("%s: cannot drop to stage #0",
2847 * Returns 1 if the path is an "other" path with respect to
2848 * the index; that is, the path is not mentioned in the index at all,
2849 * either as a file, a directory with some files in the index,
2850 * or as an unmerged entry.
2852 * We helpfully remove a trailing "/" from directories so that
2853 * the output of read_directory can be used as-is.
2855 int index_name_is_other(const struct index_state
*istate
, const char *name
,
2859 if (namelen
&& name
[namelen
- 1] == '/')
2861 pos
= index_name_pos(istate
, name
, namelen
);
2863 return 0; /* exact match */
2865 if (pos
< istate
->cache_nr
) {
2866 struct cache_entry
*ce
= istate
->cache
[pos
];
2867 if (ce_namelen(ce
) == namelen
&&
2868 !memcmp(ce
->name
, name
, namelen
))
2869 return 0; /* Yup, this one exists unmerged */
2874 void *read_blob_data_from_index(const struct index_state
*istate
,
2875 const char *path
, unsigned long *size
)
2879 enum object_type type
;
2883 pos
= index_name_pos(istate
, path
, len
);
2886 * We might be in the middle of a merge, in which
2887 * case we would read stage #2 (ours).
2891 (pos
< 0 && i
< istate
->cache_nr
&&
2892 !strcmp(istate
->cache
[i
]->name
, path
));
2894 if (ce_stage(istate
->cache
[i
]) == 2)
2899 data
= read_object_file(&istate
->cache
[pos
]->oid
, &type
, &sz
);
2900 if (!data
|| type
!= OBJ_BLOB
) {
2909 void stat_validity_clear(struct stat_validity
*sv
)
2911 FREE_AND_NULL(sv
->sd
);
2914 int stat_validity_check(struct stat_validity
*sv
, const char *path
)
2918 if (stat(path
, &st
) < 0)
2919 return sv
->sd
== NULL
;
2922 return S_ISREG(st
.st_mode
) && !match_stat_data(sv
->sd
, &st
);
2925 void stat_validity_update(struct stat_validity
*sv
, int fd
)
2929 if (fstat(fd
, &st
) < 0 || !S_ISREG(st
.st_mode
))
2930 stat_validity_clear(sv
);
2933 sv
->sd
= xcalloc(1, sizeof(struct stat_data
));
2934 fill_stat_data(sv
->sd
, &st
);
2938 void move_index_extensions(struct index_state
*dst
, struct index_state
*src
)
2940 dst
->untracked
= src
->untracked
;
2941 src
->untracked
= NULL
;
2944 struct cache_entry
*dup_cache_entry(const struct cache_entry
*ce
,
2945 struct index_state
*istate
)
2947 unsigned int size
= ce_size(ce
);
2948 int mem_pool_allocated
;
2949 struct cache_entry
*new_entry
= make_empty_cache_entry(istate
, ce_namelen(ce
));
2950 mem_pool_allocated
= new_entry
->mem_pool_allocated
;
2952 memcpy(new_entry
, ce
, size
);
2953 new_entry
->mem_pool_allocated
= mem_pool_allocated
;
2957 void discard_cache_entry(struct cache_entry
*ce
)
2959 if (ce
&& should_validate_cache_entries())
2960 memset(ce
, 0xCD, cache_entry_size(ce
->ce_namelen
));
2962 if (ce
&& ce
->mem_pool_allocated
)
2968 int should_validate_cache_entries(void)
2970 static int validate_index_cache_entries
= -1;
2972 if (validate_index_cache_entries
< 0) {
2973 if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))
2974 validate_index_cache_entries
= 1;
2976 validate_index_cache_entries
= 0;
2979 return validate_index_cache_entries
;