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"
28 /* Mask for the name length in ce_flags in the on-disk index */
30 #define CE_NAMEMASK (0x0fff)
34 * The first letter should be 'A'..'Z' for extensions that are not
35 * necessary for a correct operation (i.e. optimization data).
36 * When new extensions are added that _needs_ to be understood in
37 * order to correctly interpret the index file, pick character that
38 * is outside the range, to cause the reader to abort.
41 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
42 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
43 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
44 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
45 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
46 #define CACHE_EXT_FSMONITOR 0x46534D4E /* "FSMN" */
48 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
49 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
50 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
51 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED | FSMONITOR_CHANGED)
55 * This is an estimate of the pathname length in the index. We use
56 * this for V4 index files to guess the un-deltafied size of the index
57 * in memory because of pathname deltafication. This is not required
58 * for V2/V3 index formats because their pathnames are not compressed.
59 * If the initial amount of memory set aside is not sufficient, the
60 * mem pool will allocate extra memory.
62 #define CACHE_ENTRY_PATH_LENGTH 80
64 static inline struct cache_entry
*mem_pool__ce_alloc(struct mem_pool
*mem_pool
, size_t len
)
66 struct cache_entry
*ce
;
67 ce
= mem_pool_alloc(mem_pool
, cache_entry_size(len
));
68 ce
->mem_pool_allocated
= 1;
72 static inline struct cache_entry
*mem_pool__ce_calloc(struct mem_pool
*mem_pool
, size_t len
)
74 struct cache_entry
* ce
;
75 ce
= mem_pool_calloc(mem_pool
, 1, cache_entry_size(len
));
76 ce
->mem_pool_allocated
= 1;
80 static struct mem_pool
*find_mem_pool(struct index_state
*istate
)
82 struct mem_pool
**pool_ptr
;
84 if (istate
->split_index
&& istate
->split_index
->base
)
85 pool_ptr
= &istate
->split_index
->base
->ce_mem_pool
;
87 pool_ptr
= &istate
->ce_mem_pool
;
90 mem_pool_init(pool_ptr
, 0);
95 struct index_state the_index
;
96 static const char *alternate_index_output
;
98 static void set_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
100 istate
->cache
[nr
] = ce
;
101 add_name_hash(istate
, ce
);
104 static void replace_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
106 struct cache_entry
*old
= istate
->cache
[nr
];
108 replace_index_entry_in_base(istate
, old
, ce
);
109 remove_name_hash(istate
, old
);
110 discard_cache_entry(old
);
111 ce
->ce_flags
&= ~CE_HASHED
;
112 set_index_entry(istate
, nr
, ce
);
113 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
114 mark_fsmonitor_invalid(istate
, ce
);
115 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
118 void rename_index_entry_at(struct index_state
*istate
, int nr
, const char *new_name
)
120 struct cache_entry
*old_entry
= istate
->cache
[nr
], *new_entry
;
121 int namelen
= strlen(new_name
);
123 new_entry
= make_empty_cache_entry(istate
, namelen
);
124 copy_cache_entry(new_entry
, old_entry
);
125 new_entry
->ce_flags
&= ~CE_HASHED
;
126 new_entry
->ce_namelen
= namelen
;
127 new_entry
->index
= 0;
128 memcpy(new_entry
->name
, new_name
, namelen
+ 1);
130 cache_tree_invalidate_path(istate
, old_entry
->name
);
131 untracked_cache_remove_from_index(istate
, old_entry
->name
);
132 remove_index_entry_at(istate
, nr
);
133 add_index_entry(istate
, new_entry
, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
136 void fill_stat_data(struct stat_data
*sd
, struct stat
*st
)
138 sd
->sd_ctime
.sec
= (unsigned int)st
->st_ctime
;
139 sd
->sd_mtime
.sec
= (unsigned int)st
->st_mtime
;
140 sd
->sd_ctime
.nsec
= ST_CTIME_NSEC(*st
);
141 sd
->sd_mtime
.nsec
= ST_MTIME_NSEC(*st
);
142 sd
->sd_dev
= st
->st_dev
;
143 sd
->sd_ino
= st
->st_ino
;
144 sd
->sd_uid
= st
->st_uid
;
145 sd
->sd_gid
= st
->st_gid
;
146 sd
->sd_size
= st
->st_size
;
149 int match_stat_data(const struct stat_data
*sd
, struct stat
*st
)
153 if (sd
->sd_mtime
.sec
!= (unsigned int)st
->st_mtime
)
154 changed
|= MTIME_CHANGED
;
155 if (trust_ctime
&& check_stat
&&
156 sd
->sd_ctime
.sec
!= (unsigned int)st
->st_ctime
)
157 changed
|= CTIME_CHANGED
;
160 if (check_stat
&& sd
->sd_mtime
.nsec
!= ST_MTIME_NSEC(*st
))
161 changed
|= MTIME_CHANGED
;
162 if (trust_ctime
&& check_stat
&&
163 sd
->sd_ctime
.nsec
!= ST_CTIME_NSEC(*st
))
164 changed
|= CTIME_CHANGED
;
168 if (sd
->sd_uid
!= (unsigned int) st
->st_uid
||
169 sd
->sd_gid
!= (unsigned int) st
->st_gid
)
170 changed
|= OWNER_CHANGED
;
171 if (sd
->sd_ino
!= (unsigned int) st
->st_ino
)
172 changed
|= INODE_CHANGED
;
177 * st_dev breaks on network filesystems where different
178 * clients will have different views of what "device"
179 * the filesystem is on
181 if (check_stat
&& sd
->sd_dev
!= (unsigned int) st
->st_dev
)
182 changed
|= INODE_CHANGED
;
185 if (sd
->sd_size
!= (unsigned int) st
->st_size
)
186 changed
|= DATA_CHANGED
;
192 * This only updates the "non-critical" parts of the directory
193 * cache, ie the parts that aren't tracked by GIT, and only used
194 * to validate the cache.
196 void fill_stat_cache_info(struct cache_entry
*ce
, struct stat
*st
)
198 fill_stat_data(&ce
->ce_stat_data
, st
);
200 if (assume_unchanged
)
201 ce
->ce_flags
|= CE_VALID
;
203 if (S_ISREG(st
->st_mode
)) {
204 ce_mark_uptodate(ce
);
205 mark_fsmonitor_valid(ce
);
209 static int ce_compare_data(const struct cache_entry
*ce
, struct stat
*st
)
212 int fd
= git_open_cloexec(ce
->name
, O_RDONLY
);
215 struct object_id oid
;
216 if (!index_fd(&oid
, fd
, st
, OBJ_BLOB
, ce
->name
, 0))
217 match
= oidcmp(&oid
, &ce
->oid
);
218 /* index_fd() closed the file descriptor already */
223 static int ce_compare_link(const struct cache_entry
*ce
, size_t expected_size
)
228 enum object_type type
;
229 struct strbuf sb
= STRBUF_INIT
;
231 if (strbuf_readlink(&sb
, ce
->name
, expected_size
))
234 buffer
= read_object_file(&ce
->oid
, &type
, &size
);
237 match
= memcmp(buffer
, sb
.buf
, size
);
244 static int ce_compare_gitlink(const struct cache_entry
*ce
)
246 struct object_id oid
;
249 * We don't actually require that the .git directory
250 * under GITLINK directory be a valid git directory. It
251 * might even be missing (in case nobody populated that
254 * If so, we consider it always to match.
256 if (resolve_gitlink_ref(ce
->name
, "HEAD", &oid
) < 0)
258 return oidcmp(&oid
, &ce
->oid
);
261 static int ce_modified_check_fs(const struct cache_entry
*ce
, struct stat
*st
)
263 switch (st
->st_mode
& S_IFMT
) {
265 if (ce_compare_data(ce
, st
))
269 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
273 if (S_ISGITLINK(ce
->ce_mode
))
274 return ce_compare_gitlink(ce
) ? DATA_CHANGED
: 0;
275 /* else fallthrough */
282 static int ce_match_stat_basic(const struct cache_entry
*ce
, struct stat
*st
)
284 unsigned int changed
= 0;
286 if (ce
->ce_flags
& CE_REMOVE
)
287 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
289 switch (ce
->ce_mode
& S_IFMT
) {
291 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
292 /* We consider only the owner x bit to be relevant for
295 if (trust_executable_bit
&&
296 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
297 changed
|= MODE_CHANGED
;
300 if (!S_ISLNK(st
->st_mode
) &&
301 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
302 changed
|= TYPE_CHANGED
;
305 /* We ignore most of the st_xxx fields for gitlinks */
306 if (!S_ISDIR(st
->st_mode
))
307 changed
|= TYPE_CHANGED
;
308 else if (ce_compare_gitlink(ce
))
309 changed
|= DATA_CHANGED
;
312 die("internal error: ce_mode is %o", ce
->ce_mode
);
315 changed
|= match_stat_data(&ce
->ce_stat_data
, st
);
317 /* Racily smudged entry? */
318 if (!ce
->ce_stat_data
.sd_size
) {
319 if (!is_empty_blob_sha1(ce
->oid
.hash
))
320 changed
|= DATA_CHANGED
;
326 static int is_racy_stat(const struct index_state
*istate
,
327 const struct stat_data
*sd
)
329 return (istate
->timestamp
.sec
&&
331 /* nanosecond timestamped files can also be racy! */
332 (istate
->timestamp
.sec
< sd
->sd_mtime
.sec
||
333 (istate
->timestamp
.sec
== sd
->sd_mtime
.sec
&&
334 istate
->timestamp
.nsec
<= sd
->sd_mtime
.nsec
))
336 istate
->timestamp
.sec
<= sd
->sd_mtime
.sec
341 static int is_racy_timestamp(const struct index_state
*istate
,
342 const struct cache_entry
*ce
)
344 return (!S_ISGITLINK(ce
->ce_mode
) &&
345 is_racy_stat(istate
, &ce
->ce_stat_data
));
348 int match_stat_data_racy(const struct index_state
*istate
,
349 const struct stat_data
*sd
, struct stat
*st
)
351 if (is_racy_stat(istate
, sd
))
352 return MTIME_CHANGED
;
353 return match_stat_data(sd
, st
);
356 int ie_match_stat(struct index_state
*istate
,
357 const struct cache_entry
*ce
, struct stat
*st
,
358 unsigned int options
)
360 unsigned int changed
;
361 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
362 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
363 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
364 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
366 if (!ignore_fsmonitor
)
367 refresh_fsmonitor(istate
);
369 * If it's marked as always valid in the index, it's
370 * valid whatever the checked-out copy says.
372 * skip-worktree has the same effect with higher precedence
374 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
))
376 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
378 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
))
382 * Intent-to-add entries have not been added, so the index entry
383 * by definition never matches what is in the work tree until it
384 * actually gets added.
386 if (ce_intent_to_add(ce
))
387 return DATA_CHANGED
| TYPE_CHANGED
| MODE_CHANGED
;
389 changed
= ce_match_stat_basic(ce
, st
);
392 * Within 1 second of this sequence:
393 * echo xyzzy >file && git-update-index --add file
394 * running this command:
396 * would give a falsely clean cache entry. The mtime and
397 * length match the cache, and other stat fields do not change.
399 * We could detect this at update-index time (the cache entry
400 * being registered/updated records the same time as "now")
401 * and delay the return from git-update-index, but that would
402 * effectively mean we can make at most one commit per second,
403 * which is not acceptable. Instead, we check cache entries
404 * whose mtime are the same as the index file timestamp more
405 * carefully than others.
407 if (!changed
&& is_racy_timestamp(istate
, ce
)) {
408 if (assume_racy_is_modified
)
409 changed
|= DATA_CHANGED
;
411 changed
|= ce_modified_check_fs(ce
, st
);
417 int ie_modified(struct index_state
*istate
,
418 const struct cache_entry
*ce
,
419 struct stat
*st
, unsigned int options
)
421 int changed
, changed_fs
;
423 changed
= ie_match_stat(istate
, ce
, st
, options
);
427 * If the mode or type has changed, there's no point in trying
428 * to refresh the entry - it's not going to match
430 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
434 * Immediately after read-tree or update-index --cacheinfo,
435 * the length field is zero, as we have never even read the
436 * lstat(2) information once, and we cannot trust DATA_CHANGED
437 * returned by ie_match_stat() which in turn was returned by
438 * ce_match_stat_basic() to signal that the filesize of the
439 * blob changed. We have to actually go to the filesystem to
440 * see if the contents match, and if so, should answer "unchanged".
442 * The logic does not apply to gitlinks, as ce_match_stat_basic()
443 * already has checked the actual HEAD from the filesystem in the
444 * subproject. If ie_match_stat() already said it is different,
445 * then we know it is.
447 if ((changed
& DATA_CHANGED
) &&
448 (S_ISGITLINK(ce
->ce_mode
) || ce
->ce_stat_data
.sd_size
!= 0))
451 changed_fs
= ce_modified_check_fs(ce
, st
);
453 return changed
| changed_fs
;
457 int base_name_compare(const char *name1
, int len1
, int mode1
,
458 const char *name2
, int len2
, int mode2
)
460 unsigned char c1
, c2
;
461 int len
= len1
< len2
? len1
: len2
;
464 cmp
= memcmp(name1
, name2
, len
);
469 if (!c1
&& S_ISDIR(mode1
))
471 if (!c2
&& S_ISDIR(mode2
))
473 return (c1
< c2
) ? -1 : (c1
> c2
) ? 1 : 0;
477 * df_name_compare() is identical to base_name_compare(), except it
478 * compares conflicting directory/file entries as equal. Note that
479 * while a directory name compares as equal to a regular file, they
480 * then individually compare _differently_ to a filename that has
481 * a dot after the basename (because '\0' < '.' < '/').
483 * This is used by routines that want to traverse the git namespace
484 * but then handle conflicting entries together when possible.
486 int df_name_compare(const char *name1
, int len1
, int mode1
,
487 const char *name2
, int len2
, int mode2
)
489 int len
= len1
< len2
? len1
: len2
, cmp
;
490 unsigned char c1
, c2
;
492 cmp
= memcmp(name1
, name2
, len
);
495 /* Directories and files compare equal (same length, same name) */
499 if (!c1
&& S_ISDIR(mode1
))
502 if (!c2
&& S_ISDIR(mode2
))
504 if (c1
== '/' && !c2
)
506 if (c2
== '/' && !c1
)
511 int name_compare(const char *name1
, size_t len1
, const char *name2
, size_t len2
)
513 size_t min_len
= (len1
< len2
) ? len1
: len2
;
514 int cmp
= memcmp(name1
, name2
, min_len
);
524 int cache_name_stage_compare(const char *name1
, int len1
, int stage1
, const char *name2
, int len2
, int stage2
)
528 cmp
= name_compare(name1
, len1
, name2
, len2
);
539 static int index_name_stage_pos(const struct index_state
*istate
, const char *name
, int namelen
, int stage
)
544 last
= istate
->cache_nr
;
545 while (last
> first
) {
546 int next
= (last
+ first
) >> 1;
547 struct cache_entry
*ce
= istate
->cache
[next
];
548 int cmp
= cache_name_stage_compare(name
, namelen
, stage
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
560 int index_name_pos(const struct index_state
*istate
, const char *name
, int namelen
)
562 return index_name_stage_pos(istate
, name
, namelen
, 0);
565 int remove_index_entry_at(struct index_state
*istate
, int pos
)
567 struct cache_entry
*ce
= istate
->cache
[pos
];
569 record_resolve_undo(istate
, ce
);
570 remove_name_hash(istate
, ce
);
571 save_or_free_index_entry(istate
, ce
);
572 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
574 if (pos
>= istate
->cache_nr
)
576 MOVE_ARRAY(istate
->cache
+ pos
, istate
->cache
+ pos
+ 1,
577 istate
->cache_nr
- pos
);
582 * Remove all cache entries marked for removal, that is where
583 * CE_REMOVE is set in ce_flags. This is much more effective than
584 * calling remove_index_entry_at() for each entry to be removed.
586 void remove_marked_cache_entries(struct index_state
*istate
)
588 struct cache_entry
**ce_array
= istate
->cache
;
591 for (i
= j
= 0; i
< istate
->cache_nr
; i
++) {
592 if (ce_array
[i
]->ce_flags
& CE_REMOVE
) {
593 remove_name_hash(istate
, ce_array
[i
]);
594 save_or_free_index_entry(istate
, ce_array
[i
]);
597 ce_array
[j
++] = ce_array
[i
];
599 if (j
== istate
->cache_nr
)
601 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
602 istate
->cache_nr
= j
;
605 int remove_file_from_index(struct index_state
*istate
, const char *path
)
607 int pos
= index_name_pos(istate
, path
, strlen(path
));
610 cache_tree_invalidate_path(istate
, path
);
611 untracked_cache_remove_from_index(istate
, path
);
612 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
613 remove_index_entry_at(istate
, pos
);
617 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
619 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
622 static int index_name_pos_also_unmerged(struct index_state
*istate
,
623 const char *path
, int namelen
)
625 int pos
= index_name_pos(istate
, path
, namelen
);
626 struct cache_entry
*ce
;
631 /* maybe unmerged? */
633 if (pos
>= istate
->cache_nr
||
634 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
637 /* order of preference: stage 2, 1, 3 */
638 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
639 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
640 !compare_name(ce
, path
, namelen
))
645 static int different_name(struct cache_entry
*ce
, struct cache_entry
*alias
)
647 int len
= ce_namelen(ce
);
648 return ce_namelen(alias
) != len
|| memcmp(ce
->name
, alias
->name
, len
);
652 * If we add a filename that aliases in the cache, we will use the
653 * name that we already have - but we don't want to update the same
654 * alias twice, because that implies that there were actually two
655 * different files with aliasing names!
657 * So we use the CE_ADDED flag to verify that the alias was an old
658 * one before we accept it as
660 static struct cache_entry
*create_alias_ce(struct index_state
*istate
,
661 struct cache_entry
*ce
,
662 struct cache_entry
*alias
)
665 struct cache_entry
*new_entry
;
667 if (alias
->ce_flags
& CE_ADDED
)
668 die("Will not add file alias '%s' ('%s' already exists in index)", ce
->name
, alias
->name
);
670 /* Ok, create the new entry using the name of the existing alias */
671 len
= ce_namelen(alias
);
672 new_entry
= make_empty_cache_entry(istate
, len
);
673 memcpy(new_entry
->name
, alias
->name
, len
);
674 copy_cache_entry(new_entry
, ce
);
675 save_or_free_index_entry(istate
, ce
);
679 void set_object_name_for_intent_to_add_entry(struct cache_entry
*ce
)
681 struct object_id oid
;
682 if (write_object_file("", 0, blob_type
, &oid
))
683 die("cannot create an empty blob in the object database");
684 oidcpy(&ce
->oid
, &oid
);
687 int add_to_index(struct index_state
*istate
, const char *path
, struct stat
*st
, int flags
)
689 int namelen
, was_same
;
690 mode_t st_mode
= st
->st_mode
;
691 struct cache_entry
*ce
, *alias
= NULL
;
692 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_IGNORE_SKIP_WORKTREE
|CE_MATCH_RACY_IS_DIRTY
;
693 int verbose
= flags
& (ADD_CACHE_VERBOSE
| ADD_CACHE_PRETEND
);
694 int pretend
= flags
& ADD_CACHE_PRETEND
;
695 int intent_only
= flags
& ADD_CACHE_INTENT
;
696 int add_option
= (ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
|
697 (intent_only
? ADD_CACHE_NEW_ONLY
: 0));
698 int newflags
= HASH_WRITE_OBJECT
;
700 if (flags
& HASH_RENORMALIZE
)
701 newflags
|= HASH_RENORMALIZE
;
703 if (!S_ISREG(st_mode
) && !S_ISLNK(st_mode
) && !S_ISDIR(st_mode
))
704 return error("%s: can only add regular files, symbolic links or git-directories", path
);
706 namelen
= strlen(path
);
707 if (S_ISDIR(st_mode
)) {
708 while (namelen
&& path
[namelen
-1] == '/')
711 ce
= make_empty_cache_entry(istate
, namelen
);
712 memcpy(ce
->name
, path
, namelen
);
713 ce
->ce_namelen
= namelen
;
715 fill_stat_cache_info(ce
, st
);
717 ce
->ce_flags
|= CE_INTENT_TO_ADD
;
720 if (trust_executable_bit
&& has_symlinks
) {
721 ce
->ce_mode
= create_ce_mode(st_mode
);
723 /* If there is an existing entry, pick the mode bits and type
724 * from it, otherwise assume unexecutable regular file.
726 struct cache_entry
*ent
;
727 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
729 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
730 ce
->ce_mode
= ce_mode_from_stat(ent
, st_mode
);
733 /* When core.ignorecase=true, determine if a directory of the same name but differing
734 * case already exists within the Git repository. If it does, ensure the directory
735 * case of the file being added to the repository matches (is folded into) the existing
736 * entry's directory case.
739 adjust_dirname_case(istate
, ce
->name
);
741 if (!(flags
& HASH_RENORMALIZE
)) {
742 alias
= index_file_exists(istate
, ce
->name
,
743 ce_namelen(ce
), ignore_case
);
746 !ie_match_stat(istate
, alias
, st
, ce_option
)) {
747 /* Nothing changed, really */
748 if (!S_ISGITLINK(alias
->ce_mode
))
749 ce_mark_uptodate(alias
);
750 alias
->ce_flags
|= CE_ADDED
;
752 discard_cache_entry(ce
);
757 if (index_path(&ce
->oid
, path
, st
, newflags
)) {
758 discard_cache_entry(ce
);
759 return error("unable to index file %s", path
);
762 set_object_name_for_intent_to_add_entry(ce
);
764 if (ignore_case
&& alias
&& different_name(ce
, alias
))
765 ce
= create_alias_ce(istate
, ce
, alias
);
766 ce
->ce_flags
|= CE_ADDED
;
768 /* It was suspected to be racily clean, but it turns out to be Ok */
771 !oidcmp(&alias
->oid
, &ce
->oid
) &&
772 ce
->ce_mode
== alias
->ce_mode
);
775 discard_cache_entry(ce
);
776 else if (add_index_entry(istate
, ce
, add_option
)) {
777 discard_cache_entry(ce
);
778 return error("unable to add %s to index", path
);
780 if (verbose
&& !was_same
)
781 printf("add '%s'\n", path
);
785 int add_file_to_index(struct index_state
*istate
, const char *path
, int flags
)
788 if (lstat(path
, &st
))
789 die_errno("unable to stat '%s'", path
);
790 return add_to_index(istate
, path
, &st
, flags
);
793 struct cache_entry
*make_empty_cache_entry(struct index_state
*istate
, size_t len
)
795 return mem_pool__ce_calloc(find_mem_pool(istate
), len
);
798 struct cache_entry
*make_empty_transient_cache_entry(size_t len
)
800 return xcalloc(1, cache_entry_size(len
));
803 struct cache_entry
*make_cache_entry(struct index_state
*istate
,
805 const struct object_id
*oid
,
808 unsigned int refresh_options
)
810 struct cache_entry
*ce
, *ret
;
813 if (!verify_path(path
, mode
)) {
814 error("Invalid path '%s'", path
);
819 ce
= make_empty_cache_entry(istate
, len
);
821 oidcpy(&ce
->oid
, oid
);
822 memcpy(ce
->name
, path
, len
);
823 ce
->ce_flags
= create_ce_flags(stage
);
824 ce
->ce_namelen
= len
;
825 ce
->ce_mode
= create_ce_mode(mode
);
827 ret
= refresh_cache_entry(&the_index
, ce
, refresh_options
);
829 discard_cache_entry(ce
);
833 struct cache_entry
*make_transient_cache_entry(unsigned int mode
, const struct object_id
*oid
,
834 const char *path
, int stage
)
836 struct cache_entry
*ce
;
839 if (!verify_path(path
, mode
)) {
840 error("Invalid path '%s'", path
);
845 ce
= make_empty_transient_cache_entry(len
);
847 oidcpy(&ce
->oid
, oid
);
848 memcpy(ce
->name
, path
, len
);
849 ce
->ce_flags
= create_ce_flags(stage
);
850 ce
->ce_namelen
= len
;
851 ce
->ce_mode
= create_ce_mode(mode
);
857 * Chmod an index entry with either +x or -x.
859 * Returns -1 if the chmod for the particular cache entry failed (if it's
860 * not a regular file), -2 if an invalid flip argument is passed in, 0
863 int chmod_index_entry(struct index_state
*istate
, struct cache_entry
*ce
,
866 if (!S_ISREG(ce
->ce_mode
))
873 ce
->ce_mode
&= ~0111;
878 cache_tree_invalidate_path(istate
, ce
->name
);
879 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
880 mark_fsmonitor_invalid(istate
, ce
);
881 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
886 int ce_same_name(const struct cache_entry
*a
, const struct cache_entry
*b
)
888 int len
= ce_namelen(a
);
889 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
893 * We fundamentally don't like some paths: we don't want
894 * dot or dot-dot anywhere, and for obvious reasons don't
895 * want to recurse into ".git" either.
897 * Also, we don't want double slashes or slashes at the
898 * end that can make pathnames ambiguous.
900 static int verify_dotfile(const char *rest
, unsigned mode
)
903 * The first character was '.', but that
904 * has already been discarded, we now test
908 /* "." is not allowed */
909 if (*rest
== '\0' || is_dir_sep(*rest
))
914 * ".git" followed by NUL or slash is bad. Note that we match
915 * case-insensitively here, even if ignore_case is not set.
916 * This outlaws ".GIT" everywhere out of an abundance of caution,
917 * since there's really no good reason to allow it.
919 * Once we've seen ".git", we can also find ".gitmodules", etc (also
920 * case-insensitively).
924 if (rest
[1] != 'i' && rest
[1] != 'I')
926 if (rest
[2] != 't' && rest
[2] != 'T')
928 if (rest
[3] == '\0' || is_dir_sep(rest
[3]))
932 if (skip_iprefix(rest
, "modules", &rest
) &&
933 (*rest
== '\0' || is_dir_sep(*rest
)))
938 if (rest
[1] == '\0' || is_dir_sep(rest
[1]))
944 int verify_path(const char *path
, unsigned mode
)
948 if (has_dos_drive_prefix(path
))
958 if (is_hfs_dotgit(path
))
961 if (is_hfs_dotgitmodules(path
))
966 if (is_ntfs_dotgit(path
))
969 if (is_ntfs_dotgitmodules(path
))
975 if ((c
== '.' && !verify_dotfile(path
, mode
)) ||
976 is_dir_sep(c
) || c
== '\0')
984 * Do we have another file that has the beginning components being a
985 * proper superset of the name we're trying to add?
987 static int has_file_name(struct index_state
*istate
,
988 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
991 int len
= ce_namelen(ce
);
992 int stage
= ce_stage(ce
);
993 const char *name
= ce
->name
;
995 while (pos
< istate
->cache_nr
) {
996 struct cache_entry
*p
= istate
->cache
[pos
++];
998 if (len
>= ce_namelen(p
))
1000 if (memcmp(name
, p
->name
, len
))
1002 if (ce_stage(p
) != stage
)
1004 if (p
->name
[len
] != '/')
1006 if (p
->ce_flags
& CE_REMOVE
)
1011 remove_index_entry_at(istate
, --pos
);
1018 * Like strcmp(), but also return the offset of the first change.
1019 * If strings are equal, return the length.
1021 int strcmp_offset(const char *s1
, const char *s2
, size_t *first_change
)
1026 return strcmp(s1
, s2
);
1028 for (k
= 0; s1
[k
] == s2
[k
]; k
++)
1033 return (unsigned char)s1
[k
] - (unsigned char)s2
[k
];
1037 * Do we have another file with a pathname that is a proper
1038 * subset of the name we're trying to add?
1040 * That is, is there another file in the index with a path
1041 * that matches a sub-directory in the given entry?
1043 static int has_dir_name(struct index_state
*istate
,
1044 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
1047 int stage
= ce_stage(ce
);
1048 const char *name
= ce
->name
;
1049 const char *slash
= name
+ ce_namelen(ce
);
1054 * We are frequently called during an iteration on a sorted
1055 * list of pathnames and while building a new index. Therefore,
1056 * there is a high probability that this entry will eventually
1057 * be appended to the index, rather than inserted in the middle.
1058 * If we can confirm that, we can avoid binary searches on the
1059 * components of the pathname.
1061 * Compare the entry's full path with the last path in the index.
1063 if (istate
->cache_nr
> 0) {
1064 cmp_last
= strcmp_offset(name
,
1065 istate
->cache
[istate
->cache_nr
- 1]->name
,
1068 if (len_eq_last
== 0) {
1070 * The entry sorts AFTER the last one in the
1071 * index and their paths have no common prefix,
1072 * so there cannot be a F/D conflict.
1077 * The entry sorts AFTER the last one in the
1078 * index, but has a common prefix. Fall through
1079 * to the loop below to disect the entry's path
1080 * and see where the difference is.
1083 } else if (cmp_last
== 0) {
1085 * The entry exactly matches the last one in the
1086 * index, but because of multiple stage and CE_REMOVE
1087 * items, we fall through and let the regular search
1097 if (*--slash
== '/')
1099 if (slash
<= ce
->name
)
1106 * (len + 1) is a directory boundary (including
1107 * the trailing slash). And since the loop is
1108 * decrementing "slash", the first iteration is
1109 * the longest directory prefix; subsequent
1110 * iterations consider parent directories.
1113 if (len
+ 1 <= len_eq_last
) {
1115 * The directory prefix (including the trailing
1116 * slash) also appears as a prefix in the last
1117 * entry, so the remainder cannot collide (because
1118 * strcmp said the whole path was greater).
1123 * LT: last: xxx/file_A
1129 if (len
> len_eq_last
) {
1131 * This part of the directory prefix (excluding
1132 * the trailing slash) is longer than the known
1133 * equal portions, so this sub-directory cannot
1134 * collide with a file.
1142 if (istate
->cache_nr
> 0 &&
1143 ce_namelen(istate
->cache
[istate
->cache_nr
- 1]) > len
) {
1145 * The directory prefix lines up with part of
1146 * a longer file or directory name, but sorts
1147 * after it, so this sub-directory cannot
1148 * collide with a file.
1150 * last: xxx/yy-file (because '-' sorts before '/')
1157 * This is a possible collision. Fall through and
1158 * let the regular search code handle it.
1165 pos
= index_name_stage_pos(istate
, name
, len
, stage
);
1168 * Found one, but not so fast. This could
1169 * be a marker that says "I was here, but
1170 * I am being removed". Such an entry is
1171 * not a part of the resulting tree, and
1172 * it is Ok to have a directory at the same
1175 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
1179 remove_index_entry_at(istate
, pos
);
1187 * Trivial optimization: if we find an entry that
1188 * already matches the sub-directory, then we know
1189 * we're ok, and we can exit.
1191 while (pos
< istate
->cache_nr
) {
1192 struct cache_entry
*p
= istate
->cache
[pos
];
1193 if ((ce_namelen(p
) <= len
) ||
1194 (p
->name
[len
] != '/') ||
1195 memcmp(p
->name
, name
, len
))
1196 break; /* not our subdirectory */
1197 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
1199 * p is at the same stage as our entry, and
1200 * is a subdirectory of what we are looking
1201 * at, so we cannot have conflicts at our
1202 * level or anything shorter.
1211 /* We may be in a situation where we already have path/file and path
1212 * is being added, or we already have path and path/file is being
1213 * added. Either one would result in a nonsense tree that has path
1214 * twice when git-write-tree tries to write it out. Prevent it.
1216 * If ok-to-replace is specified, we remove the conflicting entries
1217 * from the cache so the caller should recompute the insert position.
1218 * When this happens, we return non-zero.
1220 static int check_file_directory_conflict(struct index_state
*istate
,
1221 const struct cache_entry
*ce
,
1222 int pos
, int ok_to_replace
)
1227 * When ce is an "I am going away" entry, we allow it to be added
1229 if (ce
->ce_flags
& CE_REMOVE
)
1233 * We check if the path is a sub-path of a subsequent pathname
1234 * first, since removing those will not change the position
1237 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
1240 * Then check if the path might have a clashing sub-directory
1243 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
1246 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1249 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
1250 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
1251 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
1252 int new_only
= option
& ADD_CACHE_NEW_ONLY
;
1254 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1255 cache_tree_invalidate_path(istate
, ce
->name
);
1258 * If this entry's path sorts after the last entry in the index,
1259 * we can avoid searching for it.
1261 if (istate
->cache_nr
> 0 &&
1262 strcmp(ce
->name
, istate
->cache
[istate
->cache_nr
- 1]->name
) > 0)
1263 pos
= -istate
->cache_nr
- 1;
1265 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1267 /* existing match? Just replace it. */
1270 replace_index_entry(istate
, pos
, ce
);
1275 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1276 untracked_cache_add_to_index(istate
, ce
->name
);
1279 * Inserting a merged entry ("stage 0") into the index
1280 * will always replace all non-merged entries..
1282 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
1283 while (ce_same_name(istate
->cache
[pos
], ce
)) {
1285 if (!remove_index_entry_at(istate
, pos
))
1292 if (!verify_path(ce
->name
, ce
->ce_mode
))
1293 return error("Invalid path '%s'", ce
->name
);
1295 if (!skip_df_check
&&
1296 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
1298 return error("'%s' appears as both a file and as a directory",
1300 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1306 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1310 if (option
& ADD_CACHE_JUST_APPEND
)
1311 pos
= istate
->cache_nr
;
1314 ret
= add_index_entry_with_check(istate
, ce
, option
);
1320 /* Make sure the array is big enough .. */
1321 ALLOC_GROW(istate
->cache
, istate
->cache_nr
+ 1, istate
->cache_alloc
);
1325 if (istate
->cache_nr
> pos
+ 1)
1326 MOVE_ARRAY(istate
->cache
+ pos
+ 1, istate
->cache
+ pos
,
1327 istate
->cache_nr
- pos
- 1);
1328 set_index_entry(istate
, pos
, ce
);
1329 istate
->cache_changed
|= CE_ENTRY_ADDED
;
1334 * "refresh" does not calculate a new sha1 file or bring the
1335 * cache up-to-date for mode/content changes. But what it
1336 * _does_ do is to "re-match" the stat information of a file
1337 * with the cache, so that you can refresh the cache for a
1338 * file that hasn't been changed but where the stat entry is
1341 * For example, you'd want to do this after doing a "git-read-tree",
1342 * to link up the stat cache details with the proper files.
1344 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
1345 struct cache_entry
*ce
,
1346 unsigned int options
, int *err
,
1350 struct cache_entry
*updated
;
1352 int refresh
= options
& CE_MATCH_REFRESH
;
1353 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
1354 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
1355 int ignore_missing
= options
& CE_MATCH_IGNORE_MISSING
;
1356 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
1358 if (!refresh
|| ce_uptodate(ce
))
1361 if (!ignore_fsmonitor
)
1362 refresh_fsmonitor(istate
);
1364 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1365 * that the change to the work tree does not matter and told
1368 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
)) {
1369 ce_mark_uptodate(ce
);
1372 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
)) {
1373 ce_mark_uptodate(ce
);
1376 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
)) {
1377 ce_mark_uptodate(ce
);
1381 if (has_symlink_leading_path(ce
->name
, ce_namelen(ce
))) {
1389 if (lstat(ce
->name
, &st
) < 0) {
1390 if (ignore_missing
&& errno
== ENOENT
)
1397 changed
= ie_match_stat(istate
, ce
, &st
, options
);
1399 *changed_ret
= changed
;
1402 * The path is unchanged. If we were told to ignore
1403 * valid bit, then we did the actual stat check and
1404 * found that the entry is unmodified. If the entry
1405 * is not marked VALID, this is the place to mark it
1406 * valid again, under "assume unchanged" mode.
1408 if (ignore_valid
&& assume_unchanged
&&
1409 !(ce
->ce_flags
& CE_VALID
))
1410 ; /* mark this one VALID again */
1413 * We do not mark the index itself "modified"
1414 * because CE_UPTODATE flag is in-core only;
1415 * we are not going to write this change out.
1417 if (!S_ISGITLINK(ce
->ce_mode
)) {
1418 ce_mark_uptodate(ce
);
1419 mark_fsmonitor_valid(ce
);
1425 if (ie_modified(istate
, ce
, &st
, options
)) {
1431 updated
= make_empty_cache_entry(istate
, ce_namelen(ce
));
1432 copy_cache_entry(updated
, ce
);
1433 memcpy(updated
->name
, ce
->name
, ce
->ce_namelen
+ 1);
1434 fill_stat_cache_info(updated
, &st
);
1436 * If ignore_valid is not set, we should leave CE_VALID bit
1437 * alone. Otherwise, paths marked with --no-assume-unchanged
1438 * (i.e. things to be edited) will reacquire CE_VALID bit
1439 * automatically, which is not really what we want.
1441 if (!ignore_valid
&& assume_unchanged
&&
1442 !(ce
->ce_flags
& CE_VALID
))
1443 updated
->ce_flags
&= ~CE_VALID
;
1445 /* istate->cache_changed is updated in the caller */
1449 static void show_file(const char * fmt
, const char * name
, int in_porcelain
,
1450 int * first
, const char *header_msg
)
1452 if (in_porcelain
&& *first
&& header_msg
) {
1453 printf("%s\n", header_msg
);
1459 int refresh_index(struct index_state
*istate
, unsigned int flags
,
1460 const struct pathspec
*pathspec
,
1461 char *seen
, const char *header_msg
)
1465 int really
= (flags
& REFRESH_REALLY
) != 0;
1466 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
1467 int quiet
= (flags
& REFRESH_QUIET
) != 0;
1468 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
1469 int ignore_submodules
= (flags
& REFRESH_IGNORE_SUBMODULES
) != 0;
1471 int in_porcelain
= (flags
& REFRESH_IN_PORCELAIN
);
1472 unsigned int options
= (CE_MATCH_REFRESH
|
1473 (really
? CE_MATCH_IGNORE_VALID
: 0) |
1474 (not_new
? CE_MATCH_IGNORE_MISSING
: 0));
1475 const char *modified_fmt
;
1476 const char *deleted_fmt
;
1477 const char *typechange_fmt
;
1478 const char *added_fmt
;
1479 const char *unmerged_fmt
;
1480 uint64_t start
= getnanotime();
1481 struct progress
*progress
= NULL
;
1483 if (flags
& REFRESH_PROGRESS
&& isatty(2))
1484 progress
= start_delayed_progress(_("Refresh index"),
1487 modified_fmt
= (in_porcelain
? "M\t%s\n" : "%s: needs update\n");
1488 deleted_fmt
= (in_porcelain
? "D\t%s\n" : "%s: needs update\n");
1489 typechange_fmt
= (in_porcelain
? "T\t%s\n" : "%s needs update\n");
1490 added_fmt
= (in_porcelain
? "A\t%s\n" : "%s needs update\n");
1491 unmerged_fmt
= (in_porcelain
? "U\t%s\n" : "%s: needs merge\n");
1492 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1493 struct cache_entry
*ce
, *new_entry
;
1494 int cache_errno
= 0;
1498 ce
= istate
->cache
[i
];
1499 if (ignore_submodules
&& S_ISGITLINK(ce
->ce_mode
))
1502 if (pathspec
&& !ce_path_match(&the_index
, ce
, pathspec
, seen
))
1506 while ((i
< istate
->cache_nr
) &&
1507 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
1513 show_file(unmerged_fmt
, ce
->name
, in_porcelain
,
1514 &first
, header_msg
);
1522 new_entry
= refresh_cache_ent(istate
, ce
, options
, &cache_errno
, &changed
);
1523 if (new_entry
== ce
)
1526 display_progress(progress
, i
);
1530 if (really
&& cache_errno
== EINVAL
) {
1531 /* If we are doing --really-refresh that
1532 * means the index is not valid anymore.
1534 ce
->ce_flags
&= ~CE_VALID
;
1535 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
1536 mark_fsmonitor_invalid(istate
, ce
);
1537 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
1542 if (cache_errno
== ENOENT
)
1544 else if (ce_intent_to_add(ce
))
1545 fmt
= added_fmt
; /* must be before other checks */
1546 else if (changed
& TYPE_CHANGED
)
1547 fmt
= typechange_fmt
;
1551 ce
->name
, in_porcelain
, &first
, header_msg
);
1556 replace_index_entry(istate
, i
, new_entry
);
1559 display_progress(progress
, istate
->cache_nr
);
1560 stop_progress(&progress
);
1562 trace_performance_since(start
, "refresh index");
1566 struct cache_entry
*refresh_cache_entry(struct index_state
*istate
,
1567 struct cache_entry
*ce
,
1568 unsigned int options
)
1570 return refresh_cache_ent(istate
, ce
, options
, NULL
, NULL
);
1574 /*****************************************************************
1576 *****************************************************************/
1578 #define INDEX_FORMAT_DEFAULT 3
1580 static unsigned int get_index_format_default(void)
1582 char *envversion
= getenv("GIT_INDEX_VERSION");
1585 unsigned int version
= INDEX_FORMAT_DEFAULT
;
1588 if (!git_config_get_int("index.version", &value
))
1590 if (version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1591 warning(_("index.version set, but the value is invalid.\n"
1592 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1593 return INDEX_FORMAT_DEFAULT
;
1598 version
= strtoul(envversion
, &endp
, 10);
1600 version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1601 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1602 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1603 version
= INDEX_FORMAT_DEFAULT
;
1609 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1610 * Again - this is just a (very strong in practice) heuristic that
1611 * the inode hasn't changed.
1613 * We save the fields in big-endian order to allow using the
1614 * index file over NFS transparently.
1616 struct ondisk_cache_entry
{
1617 struct cache_time ctime
;
1618 struct cache_time mtime
;
1625 unsigned char sha1
[20];
1627 char name
[FLEX_ARRAY
]; /* more */
1631 * This struct is used when CE_EXTENDED bit is 1
1632 * The struct must match ondisk_cache_entry exactly from
1635 struct ondisk_cache_entry_extended
{
1636 struct cache_time ctime
;
1637 struct cache_time mtime
;
1644 unsigned char sha1
[20];
1647 char name
[FLEX_ARRAY
]; /* more */
1650 /* These are only used for v3 or lower */
1651 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1652 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1653 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1654 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1655 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1656 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1657 ondisk_cache_entry_size(ce_namelen(ce)))
1659 /* Allow fsck to force verification of the index checksum. */
1660 int verify_index_checksum
;
1662 /* Allow fsck to force verification of the cache entry order. */
1663 int verify_ce_order
;
1665 static int verify_hdr(struct cache_header
*hdr
, unsigned long size
)
1668 unsigned char hash
[GIT_MAX_RAWSZ
];
1671 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
1672 return error("bad signature");
1673 hdr_version
= ntohl(hdr
->hdr_version
);
1674 if (hdr_version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< hdr_version
)
1675 return error("bad index version %d", hdr_version
);
1677 if (!verify_index_checksum
)
1680 the_hash_algo
->init_fn(&c
);
1681 the_hash_algo
->update_fn(&c
, hdr
, size
- the_hash_algo
->rawsz
);
1682 the_hash_algo
->final_fn(hash
, &c
);
1683 if (hashcmp(hash
, (unsigned char *)hdr
+ size
- the_hash_algo
->rawsz
))
1684 return error("bad index file sha1 signature");
1688 static int read_index_extension(struct index_state
*istate
,
1689 const char *ext
, void *data
, unsigned long sz
)
1691 switch (CACHE_EXT(ext
)) {
1692 case CACHE_EXT_TREE
:
1693 istate
->cache_tree
= cache_tree_read(data
, sz
);
1695 case CACHE_EXT_RESOLVE_UNDO
:
1696 istate
->resolve_undo
= resolve_undo_read(data
, sz
);
1698 case CACHE_EXT_LINK
:
1699 if (read_link_extension(istate
, data
, sz
))
1702 case CACHE_EXT_UNTRACKED
:
1703 istate
->untracked
= read_untracked_extension(data
, sz
);
1705 case CACHE_EXT_FSMONITOR
:
1706 read_fsmonitor_extension(istate
, data
, sz
);
1709 if (*ext
< 'A' || 'Z' < *ext
)
1710 return error("index uses %.4s extension, which we do not understand",
1712 fprintf(stderr
, "ignoring %.4s extension\n", ext
);
1718 int hold_locked_index(struct lock_file
*lk
, int lock_flags
)
1720 return hold_lock_file_for_update(lk
, get_index_file(), lock_flags
);
1723 int read_index(struct index_state
*istate
)
1725 return read_index_from(istate
, get_index_file(), get_git_dir());
1728 static struct cache_entry
*cache_entry_from_ondisk(struct mem_pool
*mem_pool
,
1729 struct ondisk_cache_entry
*ondisk
,
1734 struct cache_entry
*ce
= mem_pool__ce_alloc(mem_pool
, len
);
1736 ce
->ce_stat_data
.sd_ctime
.sec
= get_be32(&ondisk
->ctime
.sec
);
1737 ce
->ce_stat_data
.sd_mtime
.sec
= get_be32(&ondisk
->mtime
.sec
);
1738 ce
->ce_stat_data
.sd_ctime
.nsec
= get_be32(&ondisk
->ctime
.nsec
);
1739 ce
->ce_stat_data
.sd_mtime
.nsec
= get_be32(&ondisk
->mtime
.nsec
);
1740 ce
->ce_stat_data
.sd_dev
= get_be32(&ondisk
->dev
);
1741 ce
->ce_stat_data
.sd_ino
= get_be32(&ondisk
->ino
);
1742 ce
->ce_mode
= get_be32(&ondisk
->mode
);
1743 ce
->ce_stat_data
.sd_uid
= get_be32(&ondisk
->uid
);
1744 ce
->ce_stat_data
.sd_gid
= get_be32(&ondisk
->gid
);
1745 ce
->ce_stat_data
.sd_size
= get_be32(&ondisk
->size
);
1746 ce
->ce_flags
= flags
& ~CE_NAMEMASK
;
1747 ce
->ce_namelen
= len
;
1749 hashcpy(ce
->oid
.hash
, ondisk
->sha1
);
1750 memcpy(ce
->name
, name
, len
);
1751 ce
->name
[len
] = '\0';
1756 * Adjacent cache entries tend to share the leading paths, so it makes
1757 * sense to only store the differences in later entries. In the v4
1758 * on-disk format of the index, each on-disk cache entry stores the
1759 * number of bytes to be stripped from the end of the previous name,
1760 * and the bytes to append to the result, to come up with its name.
1762 static unsigned long expand_name_field(struct strbuf
*name
, const char *cp_
)
1764 const unsigned char *ep
, *cp
= (const unsigned char *)cp_
;
1765 size_t len
= decode_varint(&cp
);
1767 if (name
->len
< len
)
1768 die("malformed name field in the index");
1769 strbuf_remove(name
, name
->len
- len
, len
);
1770 for (ep
= cp
; *ep
; ep
++)
1771 ; /* find the end */
1772 strbuf_add(name
, cp
, ep
- cp
);
1773 return (const char *)ep
+ 1 - cp_
;
1776 static struct cache_entry
*create_from_disk(struct mem_pool
*mem_pool
,
1777 struct ondisk_cache_entry
*ondisk
,
1778 unsigned long *ent_size
,
1779 struct strbuf
*previous_name
)
1781 struct cache_entry
*ce
;
1786 /* On-disk flags are just 16 bits */
1787 flags
= get_be16(&ondisk
->flags
);
1788 len
= flags
& CE_NAMEMASK
;
1790 if (flags
& CE_EXTENDED
) {
1791 struct ondisk_cache_entry_extended
*ondisk2
;
1793 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
1794 extended_flags
= get_be16(&ondisk2
->flags2
) << 16;
1795 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1796 if (extended_flags
& ~CE_EXTENDED_FLAGS
)
1797 die("Unknown index entry format %08x", extended_flags
);
1798 flags
|= extended_flags
;
1799 name
= ondisk2
->name
;
1802 name
= ondisk
->name
;
1804 if (!previous_name
) {
1805 /* v3 and earlier */
1806 if (len
== CE_NAMEMASK
)
1808 ce
= cache_entry_from_ondisk(mem_pool
, ondisk
, flags
, name
, len
);
1810 *ent_size
= ondisk_ce_size(ce
);
1812 unsigned long consumed
;
1813 consumed
= expand_name_field(previous_name
, name
);
1814 ce
= cache_entry_from_ondisk(mem_pool
, ondisk
, flags
,
1816 previous_name
->len
);
1818 *ent_size
= (name
- ((char *)ondisk
)) + consumed
;
1823 static void check_ce_order(struct index_state
*istate
)
1827 if (!verify_ce_order
)
1830 for (i
= 1; i
< istate
->cache_nr
; i
++) {
1831 struct cache_entry
*ce
= istate
->cache
[i
- 1];
1832 struct cache_entry
*next_ce
= istate
->cache
[i
];
1833 int name_compare
= strcmp(ce
->name
, next_ce
->name
);
1835 if (0 < name_compare
)
1836 die("unordered stage entries in index");
1837 if (!name_compare
) {
1839 die("multiple stage entries for merged file '%s'",
1841 if (ce_stage(ce
) > ce_stage(next_ce
))
1842 die("unordered stage entries for '%s'",
1848 static void tweak_untracked_cache(struct index_state
*istate
)
1850 switch (git_config_get_untracked_cache()) {
1851 case -1: /* keep: do nothing */
1854 remove_untracked_cache(istate
);
1857 add_untracked_cache(istate
);
1859 default: /* unknown value: do nothing */
1864 static void tweak_split_index(struct index_state
*istate
)
1866 switch (git_config_get_split_index()) {
1867 case -1: /* unset: do nothing */
1870 remove_split_index(istate
);
1873 add_split_index(istate
);
1875 default: /* unknown value: do nothing */
1880 static void post_read_index_from(struct index_state
*istate
)
1882 check_ce_order(istate
);
1883 tweak_untracked_cache(istate
);
1884 tweak_split_index(istate
);
1885 tweak_fsmonitor(istate
);
1888 static size_t estimate_cache_size_from_compressed(unsigned int entries
)
1890 return entries
* (sizeof(struct cache_entry
) + CACHE_ENTRY_PATH_LENGTH
);
1893 static size_t estimate_cache_size(size_t ondisk_size
, unsigned int entries
)
1895 long per_entry
= sizeof(struct cache_entry
) - sizeof(struct ondisk_cache_entry
);
1898 * Account for potential alignment differences.
1900 per_entry
+= align_padding_size(sizeof(struct cache_entry
), -sizeof(struct ondisk_cache_entry
));
1901 return ondisk_size
+ entries
* per_entry
;
1904 /* remember to discard_cache() before reading a different cache! */
1905 int do_read_index(struct index_state
*istate
, const char *path
, int must_exist
)
1909 unsigned long src_offset
;
1910 struct cache_header
*hdr
;
1913 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
1915 if (istate
->initialized
)
1916 return istate
->cache_nr
;
1918 istate
->timestamp
.sec
= 0;
1919 istate
->timestamp
.nsec
= 0;
1920 fd
= open(path
, O_RDONLY
);
1922 if (!must_exist
&& errno
== ENOENT
)
1924 die_errno("%s: index file open failed", path
);
1928 die_errno("cannot stat the open index");
1930 mmap_size
= xsize_t(st
.st_size
);
1931 if (mmap_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
1932 die("index file smaller than expected");
1934 mmap
= xmmap(NULL
, mmap_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1935 if (mmap
== MAP_FAILED
)
1936 die_errno("unable to map index file");
1940 if (verify_hdr(hdr
, mmap_size
) < 0)
1943 hashcpy(istate
->oid
.hash
, (const unsigned char *)hdr
+ mmap_size
- the_hash_algo
->rawsz
);
1944 istate
->version
= ntohl(hdr
->hdr_version
);
1945 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
1946 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
1947 istate
->cache
= xcalloc(istate
->cache_alloc
, sizeof(*istate
->cache
));
1948 istate
->initialized
= 1;
1950 if (istate
->version
== 4) {
1951 previous_name
= &previous_name_buf
;
1952 mem_pool_init(&istate
->ce_mem_pool
,
1953 estimate_cache_size_from_compressed(istate
->cache_nr
));
1955 previous_name
= NULL
;
1956 mem_pool_init(&istate
->ce_mem_pool
,
1957 estimate_cache_size(mmap_size
, istate
->cache_nr
));
1960 src_offset
= sizeof(*hdr
);
1961 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1962 struct ondisk_cache_entry
*disk_ce
;
1963 struct cache_entry
*ce
;
1964 unsigned long consumed
;
1966 disk_ce
= (struct ondisk_cache_entry
*)((char *)mmap
+ src_offset
);
1967 ce
= create_from_disk(istate
->ce_mem_pool
, disk_ce
, &consumed
, previous_name
);
1968 set_index_entry(istate
, i
, ce
);
1970 src_offset
+= consumed
;
1972 strbuf_release(&previous_name_buf
);
1973 istate
->timestamp
.sec
= st
.st_mtime
;
1974 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
1976 while (src_offset
<= mmap_size
- the_hash_algo
->rawsz
- 8) {
1977 /* After an array of active_nr index entries,
1978 * there can be arbitrary number of extended
1979 * sections, each of which is prefixed with
1980 * extension name (4-byte) and section length
1981 * in 4-byte network byte order.
1984 memcpy(&extsize
, (char *)mmap
+ src_offset
+ 4, 4);
1985 extsize
= ntohl(extsize
);
1986 if (read_index_extension(istate
,
1987 (const char *) mmap
+ src_offset
,
1988 (char *) mmap
+ src_offset
+ 8,
1992 src_offset
+= extsize
;
1994 munmap(mmap
, mmap_size
);
1995 return istate
->cache_nr
;
1998 munmap(mmap
, mmap_size
);
1999 die("index file corrupt");
2003 * Signal that the shared index is used by updating its mtime.
2005 * This way, shared index can be removed if they have not been used
2008 static void freshen_shared_index(const char *shared_index
, int warn
)
2010 if (!check_and_freshen_file(shared_index
, 1) && warn
)
2011 warning("could not freshen shared index '%s'", shared_index
);
2014 int read_index_from(struct index_state
*istate
, const char *path
,
2017 uint64_t start
= getnanotime();
2018 struct split_index
*split_index
;
2023 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2024 if (istate
->initialized
)
2025 return istate
->cache_nr
;
2027 ret
= do_read_index(istate
, path
, 0);
2028 trace_performance_since(start
, "read cache %s", path
);
2030 split_index
= istate
->split_index
;
2031 if (!split_index
|| is_null_oid(&split_index
->base_oid
)) {
2032 post_read_index_from(istate
);
2036 if (split_index
->base
)
2037 discard_index(split_index
->base
);
2039 split_index
->base
= xcalloc(1, sizeof(*split_index
->base
));
2041 base_oid_hex
= oid_to_hex(&split_index
->base_oid
);
2042 base_path
= xstrfmt("%s/sharedindex.%s", gitdir
, base_oid_hex
);
2043 ret
= do_read_index(split_index
->base
, base_path
, 1);
2044 if (oidcmp(&split_index
->base_oid
, &split_index
->base
->oid
))
2045 die("broken index, expect %s in %s, got %s",
2046 base_oid_hex
, base_path
,
2047 oid_to_hex(&split_index
->base
->oid
));
2049 freshen_shared_index(base_path
, 0);
2050 merge_base_index(istate
);
2051 post_read_index_from(istate
);
2052 trace_performance_since(start
, "read cache %s", base_path
);
2057 int is_index_unborn(struct index_state
*istate
)
2059 return (!istate
->cache_nr
&& !istate
->timestamp
.sec
);
2062 int discard_index(struct index_state
*istate
)
2065 * Cache entries in istate->cache[] should have been allocated
2066 * from the memory pool associated with this index, or from an
2067 * associated split_index. There is no need to free individual
2068 * cache entries. validate_cache_entries can detect when this
2069 * assertion does not hold.
2071 validate_cache_entries(istate
);
2073 resolve_undo_clear_index(istate
);
2074 istate
->cache_nr
= 0;
2075 istate
->cache_changed
= 0;
2076 istate
->timestamp
.sec
= 0;
2077 istate
->timestamp
.nsec
= 0;
2078 free_name_hash(istate
);
2079 cache_tree_free(&(istate
->cache_tree
));
2080 istate
->initialized
= 0;
2081 FREE_AND_NULL(istate
->cache
);
2082 istate
->cache_alloc
= 0;
2083 discard_split_index(istate
);
2084 free_untracked_cache(istate
->untracked
);
2085 istate
->untracked
= NULL
;
2087 if (istate
->ce_mem_pool
) {
2088 mem_pool_discard(istate
->ce_mem_pool
, should_validate_cache_entries());
2089 istate
->ce_mem_pool
= NULL
;
2096 * Validate the cache entries of this index.
2097 * All cache entries associated with this index
2098 * should have been allocated by the memory pool
2099 * associated with this index, or by a referenced
2102 void validate_cache_entries(const struct index_state
*istate
)
2106 if (!should_validate_cache_entries() ||!istate
|| !istate
->initialized
)
2109 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2111 die("internal error: cache entry is not allocated from expected memory pool");
2112 } else if (!istate
->ce_mem_pool
||
2113 !mem_pool_contains(istate
->ce_mem_pool
, istate
->cache
[i
])) {
2114 if (!istate
->split_index
||
2115 !istate
->split_index
->base
||
2116 !istate
->split_index
->base
->ce_mem_pool
||
2117 !mem_pool_contains(istate
->split_index
->base
->ce_mem_pool
, istate
->cache
[i
])) {
2118 die("internal error: cache entry is not allocated from expected memory pool");
2123 if (istate
->split_index
)
2124 validate_cache_entries(istate
->split_index
->base
);
2127 int unmerged_index(const struct index_state
*istate
)
2130 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2131 if (ce_stage(istate
->cache
[i
]))
2137 int index_has_changes(const struct index_state
*istate
,
2141 struct object_id cmp
;
2144 if (istate
!= &the_index
) {
2145 BUG("index_has_changes cannot yet accept istate != &the_index; do_diff_cache needs updating first.");
2148 cmp
= tree
->object
.oid
;
2149 if (tree
|| !get_oid_tree("HEAD", &cmp
)) {
2150 struct diff_options opt
;
2153 opt
.flags
.exit_with_status
= 1;
2155 opt
.flags
.quick
= 1;
2156 do_diff_cache(&cmp
, &opt
);
2158 for (i
= 0; sb
&& i
< diff_queued_diff
.nr
; i
++) {
2160 strbuf_addch(sb
, ' ');
2161 strbuf_addstr(sb
, diff_queued_diff
.queue
[i
]->two
->path
);
2164 return opt
.flags
.has_changes
!= 0;
2166 for (i
= 0; sb
&& i
< istate
->cache_nr
; i
++) {
2168 strbuf_addch(sb
, ' ');
2169 strbuf_addstr(sb
, istate
->cache
[i
]->name
);
2171 return !!istate
->cache_nr
;
2175 #define WRITE_BUFFER_SIZE 8192
2176 static unsigned char write_buffer
[WRITE_BUFFER_SIZE
];
2177 static unsigned long write_buffer_len
;
2179 static int ce_write_flush(git_hash_ctx
*context
, int fd
)
2181 unsigned int buffered
= write_buffer_len
;
2183 the_hash_algo
->update_fn(context
, write_buffer
, buffered
);
2184 if (write_in_full(fd
, write_buffer
, buffered
) < 0)
2186 write_buffer_len
= 0;
2191 static int ce_write(git_hash_ctx
*context
, int fd
, void *data
, unsigned int len
)
2194 unsigned int buffered
= write_buffer_len
;
2195 unsigned int partial
= WRITE_BUFFER_SIZE
- buffered
;
2198 memcpy(write_buffer
+ buffered
, data
, partial
);
2199 buffered
+= partial
;
2200 if (buffered
== WRITE_BUFFER_SIZE
) {
2201 write_buffer_len
= buffered
;
2202 if (ce_write_flush(context
, fd
))
2206 write_buffer_len
= buffered
;
2208 data
= (char *) data
+ partial
;
2213 static int write_index_ext_header(git_hash_ctx
*context
, int fd
,
2214 unsigned int ext
, unsigned int sz
)
2218 return ((ce_write(context
, fd
, &ext
, 4) < 0) ||
2219 (ce_write(context
, fd
, &sz
, 4) < 0)) ? -1 : 0;
2222 static int ce_flush(git_hash_ctx
*context
, int fd
, unsigned char *hash
)
2224 unsigned int left
= write_buffer_len
;
2227 write_buffer_len
= 0;
2228 the_hash_algo
->update_fn(context
, write_buffer
, left
);
2231 /* Flush first if not enough space for hash signature */
2232 if (left
+ the_hash_algo
->rawsz
> WRITE_BUFFER_SIZE
) {
2233 if (write_in_full(fd
, write_buffer
, left
) < 0)
2238 /* Append the hash signature at the end */
2239 the_hash_algo
->final_fn(write_buffer
+ left
, context
);
2240 hashcpy(hash
, write_buffer
+ left
);
2241 left
+= the_hash_algo
->rawsz
;
2242 return (write_in_full(fd
, write_buffer
, left
) < 0) ? -1 : 0;
2245 static void ce_smudge_racily_clean_entry(struct cache_entry
*ce
)
2248 * The only thing we care about in this function is to smudge the
2249 * falsely clean entry due to touch-update-touch race, so we leave
2250 * everything else as they are. We are called for entries whose
2251 * ce_stat_data.sd_mtime match the index file mtime.
2253 * Note that this actually does not do much for gitlinks, for
2254 * which ce_match_stat_basic() always goes to the actual
2255 * contents. The caller checks with is_racy_timestamp() which
2256 * always says "no" for gitlinks, so we are not called for them ;-)
2260 if (lstat(ce
->name
, &st
) < 0)
2262 if (ce_match_stat_basic(ce
, &st
))
2264 if (ce_modified_check_fs(ce
, &st
)) {
2265 /* This is "racily clean"; smudge it. Note that this
2266 * is a tricky code. At first glance, it may appear
2267 * that it can break with this sequence:
2269 * $ echo xyzzy >frotz
2270 * $ git-update-index --add frotz
2273 * $ echo filfre >nitfol
2274 * $ git-update-index --add nitfol
2276 * but it does not. When the second update-index runs,
2277 * it notices that the entry "frotz" has the same timestamp
2278 * as index, and if we were to smudge it by resetting its
2279 * size to zero here, then the object name recorded
2280 * in index is the 6-byte file but the cached stat information
2281 * becomes zero --- which would then match what we would
2282 * obtain from the filesystem next time we stat("frotz").
2284 * However, the second update-index, before calling
2285 * this function, notices that the cached size is 6
2286 * bytes and what is on the filesystem is an empty
2287 * file, and never calls us, so the cached size information
2288 * for "frotz" stays 6 which does not match the filesystem.
2290 ce
->ce_stat_data
.sd_size
= 0;
2294 /* Copy miscellaneous fields but not the name */
2295 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry
*ondisk
,
2296 struct cache_entry
*ce
)
2300 ondisk
->ctime
.sec
= htonl(ce
->ce_stat_data
.sd_ctime
.sec
);
2301 ondisk
->mtime
.sec
= htonl(ce
->ce_stat_data
.sd_mtime
.sec
);
2302 ondisk
->ctime
.nsec
= htonl(ce
->ce_stat_data
.sd_ctime
.nsec
);
2303 ondisk
->mtime
.nsec
= htonl(ce
->ce_stat_data
.sd_mtime
.nsec
);
2304 ondisk
->dev
= htonl(ce
->ce_stat_data
.sd_dev
);
2305 ondisk
->ino
= htonl(ce
->ce_stat_data
.sd_ino
);
2306 ondisk
->mode
= htonl(ce
->ce_mode
);
2307 ondisk
->uid
= htonl(ce
->ce_stat_data
.sd_uid
);
2308 ondisk
->gid
= htonl(ce
->ce_stat_data
.sd_gid
);
2309 ondisk
->size
= htonl(ce
->ce_stat_data
.sd_size
);
2310 hashcpy(ondisk
->sha1
, ce
->oid
.hash
);
2312 flags
= ce
->ce_flags
& ~CE_NAMEMASK
;
2313 flags
|= (ce_namelen(ce
) >= CE_NAMEMASK
? CE_NAMEMASK
: ce_namelen(ce
));
2314 ondisk
->flags
= htons(flags
);
2315 if (ce
->ce_flags
& CE_EXTENDED
) {
2316 struct ondisk_cache_entry_extended
*ondisk2
;
2317 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
2318 ondisk2
->flags2
= htons((ce
->ce_flags
& CE_EXTENDED_FLAGS
) >> 16);
2322 static int ce_write_entry(git_hash_ctx
*c
, int fd
, struct cache_entry
*ce
,
2323 struct strbuf
*previous_name
, struct ondisk_cache_entry
*ondisk
)
2327 unsigned int saved_namelen
;
2328 int stripped_name
= 0;
2329 static unsigned char padding
[8] = { 0x00 };
2331 if (ce
->ce_flags
& CE_STRIP_NAME
) {
2332 saved_namelen
= ce_namelen(ce
);
2337 if (ce
->ce_flags
& CE_EXTENDED
)
2338 size
= offsetof(struct ondisk_cache_entry_extended
, name
);
2340 size
= offsetof(struct ondisk_cache_entry
, name
);
2342 if (!previous_name
) {
2343 int len
= ce_namelen(ce
);
2344 copy_cache_entry_to_ondisk(ondisk
, ce
);
2345 result
= ce_write(c
, fd
, ondisk
, size
);
2347 result
= ce_write(c
, fd
, ce
->name
, len
);
2349 result
= ce_write(c
, fd
, padding
, align_padding_size(size
, len
));
2351 int common
, to_remove
, prefix_size
;
2352 unsigned char to_remove_vi
[16];
2354 (ce
->name
[common
] &&
2355 common
< previous_name
->len
&&
2356 ce
->name
[common
] == previous_name
->buf
[common
]);
2358 ; /* still matching */
2359 to_remove
= previous_name
->len
- common
;
2360 prefix_size
= encode_varint(to_remove
, to_remove_vi
);
2362 copy_cache_entry_to_ondisk(ondisk
, ce
);
2363 result
= ce_write(c
, fd
, ondisk
, size
);
2365 result
= ce_write(c
, fd
, to_remove_vi
, prefix_size
);
2367 result
= ce_write(c
, fd
, ce
->name
+ common
, ce_namelen(ce
) - common
);
2369 result
= ce_write(c
, fd
, padding
, 1);
2371 strbuf_splice(previous_name
, common
, to_remove
,
2372 ce
->name
+ common
, ce_namelen(ce
) - common
);
2374 if (stripped_name
) {
2375 ce
->ce_namelen
= saved_namelen
;
2376 ce
->ce_flags
&= ~CE_STRIP_NAME
;
2383 * This function verifies if index_state has the correct sha1 of the
2384 * index file. Don't die if we have any other failure, just return 0.
2386 static int verify_index_from(const struct index_state
*istate
, const char *path
)
2391 unsigned char hash
[GIT_MAX_RAWSZ
];
2393 if (!istate
->initialized
)
2396 fd
= open(path
, O_RDONLY
);
2403 if (st
.st_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2406 n
= pread_in_full(fd
, hash
, the_hash_algo
->rawsz
, st
.st_size
- the_hash_algo
->rawsz
);
2407 if (n
!= the_hash_algo
->rawsz
)
2410 if (hashcmp(istate
->oid
.hash
, hash
))
2421 static int verify_index(const struct index_state
*istate
)
2423 return verify_index_from(istate
, get_index_file());
2426 static int has_racy_timestamp(struct index_state
*istate
)
2428 int entries
= istate
->cache_nr
;
2431 for (i
= 0; i
< entries
; i
++) {
2432 struct cache_entry
*ce
= istate
->cache
[i
];
2433 if (is_racy_timestamp(istate
, ce
))
2439 void update_index_if_able(struct index_state
*istate
, struct lock_file
*lockfile
)
2441 if ((istate
->cache_changed
|| has_racy_timestamp(istate
)) &&
2442 verify_index(istate
))
2443 write_locked_index(istate
, lockfile
, COMMIT_LOCK
);
2445 rollback_lock_file(lockfile
);
2449 * On success, `tempfile` is closed. If it is the temporary file
2450 * of a `struct lock_file`, we will therefore effectively perform
2451 * a 'close_lock_file_gently()`. Since that is an implementation
2452 * detail of lockfiles, callers of `do_write_index()` should not
2455 static int do_write_index(struct index_state
*istate
, struct tempfile
*tempfile
,
2456 int strip_extensions
)
2458 uint64_t start
= getnanotime();
2459 int newfd
= tempfile
->fd
;
2461 struct cache_header hdr
;
2462 int i
, err
= 0, removed
, extended
, hdr_version
;
2463 struct cache_entry
**cache
= istate
->cache
;
2464 int entries
= istate
->cache_nr
;
2466 struct ondisk_cache_entry_extended ondisk
;
2467 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
2468 int drop_cache_tree
= istate
->drop_cache_tree
;
2470 for (i
= removed
= extended
= 0; i
< entries
; i
++) {
2471 if (cache
[i
]->ce_flags
& CE_REMOVE
)
2474 /* reduce extended entries if possible */
2475 cache
[i
]->ce_flags
&= ~CE_EXTENDED
;
2476 if (cache
[i
]->ce_flags
& CE_EXTENDED_FLAGS
) {
2478 cache
[i
]->ce_flags
|= CE_EXTENDED
;
2482 if (!istate
->version
) {
2483 istate
->version
= get_index_format_default();
2484 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0))
2485 init_split_index(istate
);
2488 /* demote version 3 to version 2 when the latter suffices */
2489 if (istate
->version
== 3 || istate
->version
== 2)
2490 istate
->version
= extended
? 3 : 2;
2492 hdr_version
= istate
->version
;
2494 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
2495 hdr
.hdr_version
= htonl(hdr_version
);
2496 hdr
.hdr_entries
= htonl(entries
- removed
);
2498 the_hash_algo
->init_fn(&c
);
2499 if (ce_write(&c
, newfd
, &hdr
, sizeof(hdr
)) < 0)
2502 previous_name
= (hdr_version
== 4) ? &previous_name_buf
: NULL
;
2504 for (i
= 0; i
< entries
; i
++) {
2505 struct cache_entry
*ce
= cache
[i
];
2506 if (ce
->ce_flags
& CE_REMOVE
)
2508 if (!ce_uptodate(ce
) && is_racy_timestamp(istate
, ce
))
2509 ce_smudge_racily_clean_entry(ce
);
2510 if (is_null_oid(&ce
->oid
)) {
2511 static const char msg
[] = "cache entry has null sha1: %s";
2512 static int allow
= -1;
2515 allow
= git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2517 warning(msg
, ce
->name
);
2519 err
= error(msg
, ce
->name
);
2521 drop_cache_tree
= 1;
2523 if (ce_write_entry(&c
, newfd
, ce
, previous_name
, (struct ondisk_cache_entry
*)&ondisk
) < 0)
2529 strbuf_release(&previous_name_buf
);
2534 /* Write extension data here */
2535 if (!strip_extensions
&& istate
->split_index
) {
2536 struct strbuf sb
= STRBUF_INIT
;
2538 err
= write_link_extension(&sb
, istate
) < 0 ||
2539 write_index_ext_header(&c
, newfd
, CACHE_EXT_LINK
,
2541 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2542 strbuf_release(&sb
);
2546 if (!strip_extensions
&& !drop_cache_tree
&& istate
->cache_tree
) {
2547 struct strbuf sb
= STRBUF_INIT
;
2549 cache_tree_write(&sb
, istate
->cache_tree
);
2550 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_TREE
, sb
.len
) < 0
2551 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2552 strbuf_release(&sb
);
2556 if (!strip_extensions
&& istate
->resolve_undo
) {
2557 struct strbuf sb
= STRBUF_INIT
;
2559 resolve_undo_write(&sb
, istate
->resolve_undo
);
2560 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_RESOLVE_UNDO
,
2562 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2563 strbuf_release(&sb
);
2567 if (!strip_extensions
&& istate
->untracked
) {
2568 struct strbuf sb
= STRBUF_INIT
;
2570 write_untracked_extension(&sb
, istate
->untracked
);
2571 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_UNTRACKED
,
2573 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2574 strbuf_release(&sb
);
2578 if (!strip_extensions
&& istate
->fsmonitor_last_update
) {
2579 struct strbuf sb
= STRBUF_INIT
;
2581 write_fsmonitor_extension(&sb
, istate
);
2582 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_FSMONITOR
, sb
.len
) < 0
2583 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2584 strbuf_release(&sb
);
2589 if (ce_flush(&c
, newfd
, istate
->oid
.hash
))
2591 if (close_tempfile_gently(tempfile
)) {
2592 error(_("could not close '%s'"), tempfile
->filename
.buf
);
2595 if (stat(tempfile
->filename
.buf
, &st
))
2597 istate
->timestamp
.sec
= (unsigned int)st
.st_mtime
;
2598 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
2599 trace_performance_since(start
, "write index, changed mask = %x", istate
->cache_changed
);
2603 void set_alternate_index_output(const char *name
)
2605 alternate_index_output
= name
;
2608 static int commit_locked_index(struct lock_file
*lk
)
2610 if (alternate_index_output
)
2611 return commit_lock_file_to(lk
, alternate_index_output
);
2613 return commit_lock_file(lk
);
2616 static int do_write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
2619 int ret
= do_write_index(istate
, lock
->tempfile
, 0);
2622 if (flags
& COMMIT_LOCK
)
2623 return commit_locked_index(lock
);
2624 return close_lock_file_gently(lock
);
2627 static int write_split_index(struct index_state
*istate
,
2628 struct lock_file
*lock
,
2632 prepare_to_write_split_index(istate
);
2633 ret
= do_write_locked_index(istate
, lock
, flags
);
2634 finish_writing_split_index(istate
);
2638 static const char *shared_index_expire
= "2.weeks.ago";
2640 static unsigned long get_shared_index_expire_date(void)
2642 static unsigned long shared_index_expire_date
;
2643 static int shared_index_expire_date_prepared
;
2645 if (!shared_index_expire_date_prepared
) {
2646 git_config_get_expiry("splitindex.sharedindexexpire",
2647 &shared_index_expire
);
2648 shared_index_expire_date
= approxidate(shared_index_expire
);
2649 shared_index_expire_date_prepared
= 1;
2652 return shared_index_expire_date
;
2655 static int should_delete_shared_index(const char *shared_index_path
)
2658 unsigned long expiration
;
2660 /* Check timestamp */
2661 expiration
= get_shared_index_expire_date();
2664 if (stat(shared_index_path
, &st
))
2665 return error_errno(_("could not stat '%s'"), shared_index_path
);
2666 if (st
.st_mtime
> expiration
)
2672 static int clean_shared_index_files(const char *current_hex
)
2675 DIR *dir
= opendir(get_git_dir());
2678 return error_errno(_("unable to open git dir: %s"), get_git_dir());
2680 while ((de
= readdir(dir
)) != NULL
) {
2681 const char *sha1_hex
;
2682 const char *shared_index_path
;
2683 if (!skip_prefix(de
->d_name
, "sharedindex.", &sha1_hex
))
2685 if (!strcmp(sha1_hex
, current_hex
))
2687 shared_index_path
= git_path("%s", de
->d_name
);
2688 if (should_delete_shared_index(shared_index_path
) > 0 &&
2689 unlink(shared_index_path
))
2690 warning_errno(_("unable to unlink: %s"), shared_index_path
);
2697 static int write_shared_index(struct index_state
*istate
,
2698 struct tempfile
**temp
)
2700 struct split_index
*si
= istate
->split_index
;
2703 move_cache_to_base_index(istate
);
2704 ret
= do_write_index(si
->base
, *temp
, 1);
2707 ret
= adjust_shared_perm(get_tempfile_path(*temp
));
2709 error("cannot fix permission bits on %s", get_tempfile_path(*temp
));
2712 ret
= rename_tempfile(temp
,
2713 git_path("sharedindex.%s", oid_to_hex(&si
->base
->oid
)));
2715 oidcpy(&si
->base_oid
, &si
->base
->oid
);
2716 clean_shared_index_files(oid_to_hex(&si
->base
->oid
));
2722 static const int default_max_percent_split_change
= 20;
2724 static int too_many_not_shared_entries(struct index_state
*istate
)
2726 int i
, not_shared
= 0;
2727 int max_split
= git_config_get_max_percent_split_change();
2729 switch (max_split
) {
2731 /* not or badly configured: use the default value */
2732 max_split
= default_max_percent_split_change
;
2735 return 1; /* 0% means always write a new shared index */
2737 return 0; /* 100% means never write a new shared index */
2739 break; /* just use the configured value */
2742 /* Count not shared entries */
2743 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2744 struct cache_entry
*ce
= istate
->cache
[i
];
2749 return (int64_t)istate
->cache_nr
* max_split
< (int64_t)not_shared
* 100;
2752 int write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
2755 int new_shared_index
, ret
;
2756 struct split_index
*si
= istate
->split_index
;
2758 if ((flags
& SKIP_IF_UNCHANGED
) && !istate
->cache_changed
) {
2759 if (flags
& COMMIT_LOCK
)
2760 rollback_lock_file(lock
);
2764 if (istate
->fsmonitor_last_update
)
2765 fill_fsmonitor_bitmap(istate
);
2767 if (!si
|| alternate_index_output
||
2768 (istate
->cache_changed
& ~EXTMASK
)) {
2770 oidclr(&si
->base_oid
);
2771 ret
= do_write_locked_index(istate
, lock
, flags
);
2775 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0)) {
2776 int v
= si
->base_oid
.hash
[0];
2778 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
2780 if (too_many_not_shared_entries(istate
))
2781 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
2783 new_shared_index
= istate
->cache_changed
& SPLIT_INDEX_ORDERED
;
2785 if (new_shared_index
) {
2786 struct tempfile
*temp
;
2789 temp
= mks_tempfile(git_path("sharedindex_XXXXXX"));
2791 oidclr(&si
->base_oid
);
2792 ret
= do_write_locked_index(istate
, lock
, flags
);
2795 ret
= write_shared_index(istate
, &temp
);
2797 saved_errno
= errno
;
2798 if (is_tempfile_active(temp
))
2799 delete_tempfile(&temp
);
2800 errno
= saved_errno
;
2806 ret
= write_split_index(istate
, lock
, flags
);
2808 /* Freshen the shared index only if the split-index was written */
2809 if (!ret
&& !new_shared_index
) {
2810 const char *shared_index
= git_path("sharedindex.%s",
2811 oid_to_hex(&si
->base_oid
));
2812 freshen_shared_index(shared_index
, 1);
2816 if (flags
& COMMIT_LOCK
)
2817 rollback_lock_file(lock
);
2822 * Read the index file that is potentially unmerged into given
2823 * index_state, dropping any unmerged entries to stage #0 (potentially
2824 * resulting in a path appearing as both a file and a directory in the
2825 * index; the caller is responsible to clear out the extra entries
2826 * before writing the index to a tree). Returns true if the index is
2827 * unmerged. Callers who want to refuse to work from an unmerged
2828 * state can call this and check its return value, instead of calling
2831 int read_index_unmerged(struct index_state
*istate
)
2837 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2838 struct cache_entry
*ce
= istate
->cache
[i
];
2839 struct cache_entry
*new_ce
;
2845 len
= ce_namelen(ce
);
2846 new_ce
= make_empty_cache_entry(istate
, len
);
2847 memcpy(new_ce
->name
, ce
->name
, len
);
2848 new_ce
->ce_flags
= create_ce_flags(0) | CE_CONFLICTED
;
2849 new_ce
->ce_namelen
= len
;
2850 new_ce
->ce_mode
= ce
->ce_mode
;
2851 if (add_index_entry(istate
, new_ce
, ADD_CACHE_SKIP_DFCHECK
))
2852 return error("%s: cannot drop to stage #0",
2859 * Returns 1 if the path is an "other" path with respect to
2860 * the index; that is, the path is not mentioned in the index at all,
2861 * either as a file, a directory with some files in the index,
2862 * or as an unmerged entry.
2864 * We helpfully remove a trailing "/" from directories so that
2865 * the output of read_directory can be used as-is.
2867 int index_name_is_other(const struct index_state
*istate
, const char *name
,
2871 if (namelen
&& name
[namelen
- 1] == '/')
2873 pos
= index_name_pos(istate
, name
, namelen
);
2875 return 0; /* exact match */
2877 if (pos
< istate
->cache_nr
) {
2878 struct cache_entry
*ce
= istate
->cache
[pos
];
2879 if (ce_namelen(ce
) == namelen
&&
2880 !memcmp(ce
->name
, name
, namelen
))
2881 return 0; /* Yup, this one exists unmerged */
2886 void *read_blob_data_from_index(const struct index_state
*istate
,
2887 const char *path
, unsigned long *size
)
2891 enum object_type type
;
2895 pos
= index_name_pos(istate
, path
, len
);
2898 * We might be in the middle of a merge, in which
2899 * case we would read stage #2 (ours).
2903 (pos
< 0 && i
< istate
->cache_nr
&&
2904 !strcmp(istate
->cache
[i
]->name
, path
));
2906 if (ce_stage(istate
->cache
[i
]) == 2)
2911 data
= read_object_file(&istate
->cache
[pos
]->oid
, &type
, &sz
);
2912 if (!data
|| type
!= OBJ_BLOB
) {
2921 void stat_validity_clear(struct stat_validity
*sv
)
2923 FREE_AND_NULL(sv
->sd
);
2926 int stat_validity_check(struct stat_validity
*sv
, const char *path
)
2930 if (stat(path
, &st
) < 0)
2931 return sv
->sd
== NULL
;
2934 return S_ISREG(st
.st_mode
) && !match_stat_data(sv
->sd
, &st
);
2937 void stat_validity_update(struct stat_validity
*sv
, int fd
)
2941 if (fstat(fd
, &st
) < 0 || !S_ISREG(st
.st_mode
))
2942 stat_validity_clear(sv
);
2945 sv
->sd
= xcalloc(1, sizeof(struct stat_data
));
2946 fill_stat_data(sv
->sd
, &st
);
2950 void move_index_extensions(struct index_state
*dst
, struct index_state
*src
)
2952 dst
->untracked
= src
->untracked
;
2953 src
->untracked
= NULL
;
2956 struct cache_entry
*dup_cache_entry(const struct cache_entry
*ce
,
2957 struct index_state
*istate
)
2959 unsigned int size
= ce_size(ce
);
2960 int mem_pool_allocated
;
2961 struct cache_entry
*new_entry
= make_empty_cache_entry(istate
, ce_namelen(ce
));
2962 mem_pool_allocated
= new_entry
->mem_pool_allocated
;
2964 memcpy(new_entry
, ce
, size
);
2965 new_entry
->mem_pool_allocated
= mem_pool_allocated
;
2969 void discard_cache_entry(struct cache_entry
*ce
)
2971 if (ce
&& should_validate_cache_entries())
2972 memset(ce
, 0xCD, cache_entry_size(ce
->ce_namelen
));
2974 if (ce
&& ce
->mem_pool_allocated
)
2980 int should_validate_cache_entries(void)
2982 static int validate_index_cache_entries
= -1;
2984 if (validate_index_cache_entries
< 0) {
2985 if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))
2986 validate_index_cache_entries
= 1;
2988 validate_index_cache_entries
= 0;
2991 return validate_index_cache_entries
;