2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 #define NO_THE_INDEX_COMPATIBILITY_MACROS
11 #include "cache-tree.h"
14 #include "object-store.h"
18 #include "resolve-undo.h"
21 #include "split-index.h"
23 #include "fsmonitor.h"
25 /* Mask for the name length in ce_flags in the on-disk index */
27 #define CE_NAMEMASK (0x0fff)
31 * The first letter should be 'A'..'Z' for extensions that are not
32 * necessary for a correct operation (i.e. optimization data).
33 * When new extensions are added that _needs_ to be understood in
34 * order to correctly interpret the index file, pick character that
35 * is outside the range, to cause the reader to abort.
38 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
39 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
40 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
41 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
42 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
43 #define CACHE_EXT_FSMONITOR 0x46534D4E /* "FSMN" */
45 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
46 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
47 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
48 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED | FSMONITOR_CHANGED)
52 * This is an estimate of the pathname length in the index. We use
53 * this for V4 index files to guess the un-deltafied size of the index
54 * in memory because of pathname deltafication. This is not required
55 * for V2/V3 index formats because their pathnames are not compressed.
56 * If the initial amount of memory set aside is not sufficient, the
57 * mem pool will allocate extra memory.
59 #define CACHE_ENTRY_PATH_LENGTH 80
61 static inline struct cache_entry
*mem_pool__ce_alloc(struct mem_pool
*mem_pool
, size_t len
)
63 struct cache_entry
*ce
;
64 ce
= mem_pool_alloc(mem_pool
, cache_entry_size(len
));
65 ce
->mem_pool_allocated
= 1;
69 static inline struct cache_entry
*mem_pool__ce_calloc(struct mem_pool
*mem_pool
, size_t len
)
71 struct cache_entry
* ce
;
72 ce
= mem_pool_calloc(mem_pool
, 1, cache_entry_size(len
));
73 ce
->mem_pool_allocated
= 1;
77 static struct mem_pool
*find_mem_pool(struct index_state
*istate
)
79 struct mem_pool
**pool_ptr
;
81 if (istate
->split_index
&& istate
->split_index
->base
)
82 pool_ptr
= &istate
->split_index
->base
->ce_mem_pool
;
84 pool_ptr
= &istate
->ce_mem_pool
;
87 mem_pool_init(pool_ptr
, 0);
92 struct index_state the_index
;
93 static const char *alternate_index_output
;
95 static void set_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
97 istate
->cache
[nr
] = ce
;
98 add_name_hash(istate
, ce
);
101 static void replace_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
103 struct cache_entry
*old
= istate
->cache
[nr
];
105 replace_index_entry_in_base(istate
, old
, ce
);
106 remove_name_hash(istate
, old
);
107 discard_cache_entry(old
);
108 ce
->ce_flags
&= ~CE_HASHED
;
109 set_index_entry(istate
, nr
, ce
);
110 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
111 mark_fsmonitor_invalid(istate
, ce
);
112 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
115 void rename_index_entry_at(struct index_state
*istate
, int nr
, const char *new_name
)
117 struct cache_entry
*old_entry
= istate
->cache
[nr
], *new_entry
;
118 int namelen
= strlen(new_name
);
120 new_entry
= make_empty_cache_entry(istate
, namelen
);
121 copy_cache_entry(new_entry
, old_entry
);
122 new_entry
->ce_flags
&= ~CE_HASHED
;
123 new_entry
->ce_namelen
= namelen
;
124 new_entry
->index
= 0;
125 memcpy(new_entry
->name
, new_name
, namelen
+ 1);
127 cache_tree_invalidate_path(istate
, old_entry
->name
);
128 untracked_cache_remove_from_index(istate
, old_entry
->name
);
129 remove_index_entry_at(istate
, nr
);
130 add_index_entry(istate
, new_entry
, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
133 void fill_stat_data(struct stat_data
*sd
, struct stat
*st
)
135 sd
->sd_ctime
.sec
= (unsigned int)st
->st_ctime
;
136 sd
->sd_mtime
.sec
= (unsigned int)st
->st_mtime
;
137 sd
->sd_ctime
.nsec
= ST_CTIME_NSEC(*st
);
138 sd
->sd_mtime
.nsec
= ST_MTIME_NSEC(*st
);
139 sd
->sd_dev
= st
->st_dev
;
140 sd
->sd_ino
= st
->st_ino
;
141 sd
->sd_uid
= st
->st_uid
;
142 sd
->sd_gid
= st
->st_gid
;
143 sd
->sd_size
= st
->st_size
;
146 int match_stat_data(const struct stat_data
*sd
, struct stat
*st
)
150 if (sd
->sd_mtime
.sec
!= (unsigned int)st
->st_mtime
)
151 changed
|= MTIME_CHANGED
;
152 if (trust_ctime
&& check_stat
&&
153 sd
->sd_ctime
.sec
!= (unsigned int)st
->st_ctime
)
154 changed
|= CTIME_CHANGED
;
157 if (check_stat
&& sd
->sd_mtime
.nsec
!= ST_MTIME_NSEC(*st
))
158 changed
|= MTIME_CHANGED
;
159 if (trust_ctime
&& check_stat
&&
160 sd
->sd_ctime
.nsec
!= ST_CTIME_NSEC(*st
))
161 changed
|= CTIME_CHANGED
;
165 if (sd
->sd_uid
!= (unsigned int) st
->st_uid
||
166 sd
->sd_gid
!= (unsigned int) st
->st_gid
)
167 changed
|= OWNER_CHANGED
;
168 if (sd
->sd_ino
!= (unsigned int) st
->st_ino
)
169 changed
|= INODE_CHANGED
;
174 * st_dev breaks on network filesystems where different
175 * clients will have different views of what "device"
176 * the filesystem is on
178 if (check_stat
&& sd
->sd_dev
!= (unsigned int) st
->st_dev
)
179 changed
|= INODE_CHANGED
;
182 if (sd
->sd_size
!= (unsigned int) st
->st_size
)
183 changed
|= DATA_CHANGED
;
189 * This only updates the "non-critical" parts of the directory
190 * cache, ie the parts that aren't tracked by GIT, and only used
191 * to validate the cache.
193 void fill_stat_cache_info(struct cache_entry
*ce
, struct stat
*st
)
195 fill_stat_data(&ce
->ce_stat_data
, st
);
197 if (assume_unchanged
)
198 ce
->ce_flags
|= CE_VALID
;
200 if (S_ISREG(st
->st_mode
)) {
201 ce_mark_uptodate(ce
);
202 mark_fsmonitor_valid(ce
);
206 static int ce_compare_data(const struct cache_entry
*ce
, struct stat
*st
)
209 int fd
= git_open_cloexec(ce
->name
, O_RDONLY
);
212 struct object_id oid
;
213 if (!index_fd(&oid
, fd
, st
, OBJ_BLOB
, ce
->name
, 0))
214 match
= oidcmp(&oid
, &ce
->oid
);
215 /* index_fd() closed the file descriptor already */
220 static int ce_compare_link(const struct cache_entry
*ce
, size_t expected_size
)
225 enum object_type type
;
226 struct strbuf sb
= STRBUF_INIT
;
228 if (strbuf_readlink(&sb
, ce
->name
, expected_size
))
231 buffer
= read_object_file(&ce
->oid
, &type
, &size
);
234 match
= memcmp(buffer
, sb
.buf
, size
);
241 static int ce_compare_gitlink(const struct cache_entry
*ce
)
243 struct object_id oid
;
246 * We don't actually require that the .git directory
247 * under GITLINK directory be a valid git directory. It
248 * might even be missing (in case nobody populated that
251 * If so, we consider it always to match.
253 if (resolve_gitlink_ref(ce
->name
, "HEAD", &oid
) < 0)
255 return oidcmp(&oid
, &ce
->oid
);
258 static int ce_modified_check_fs(const struct cache_entry
*ce
, struct stat
*st
)
260 switch (st
->st_mode
& S_IFMT
) {
262 if (ce_compare_data(ce
, st
))
266 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
270 if (S_ISGITLINK(ce
->ce_mode
))
271 return ce_compare_gitlink(ce
) ? DATA_CHANGED
: 0;
272 /* else fallthrough */
279 static int ce_match_stat_basic(const struct cache_entry
*ce
, struct stat
*st
)
281 unsigned int changed
= 0;
283 if (ce
->ce_flags
& CE_REMOVE
)
284 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
286 switch (ce
->ce_mode
& S_IFMT
) {
288 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
289 /* We consider only the owner x bit to be relevant for
292 if (trust_executable_bit
&&
293 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
294 changed
|= MODE_CHANGED
;
297 if (!S_ISLNK(st
->st_mode
) &&
298 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
299 changed
|= TYPE_CHANGED
;
302 /* We ignore most of the st_xxx fields for gitlinks */
303 if (!S_ISDIR(st
->st_mode
))
304 changed
|= TYPE_CHANGED
;
305 else if (ce_compare_gitlink(ce
))
306 changed
|= DATA_CHANGED
;
309 die("internal error: ce_mode is %o", ce
->ce_mode
);
312 changed
|= match_stat_data(&ce
->ce_stat_data
, st
);
314 /* Racily smudged entry? */
315 if (!ce
->ce_stat_data
.sd_size
) {
316 if (!is_empty_blob_sha1(ce
->oid
.hash
))
317 changed
|= DATA_CHANGED
;
323 static int is_racy_stat(const struct index_state
*istate
,
324 const struct stat_data
*sd
)
326 return (istate
->timestamp
.sec
&&
328 /* nanosecond timestamped files can also be racy! */
329 (istate
->timestamp
.sec
< sd
->sd_mtime
.sec
||
330 (istate
->timestamp
.sec
== sd
->sd_mtime
.sec
&&
331 istate
->timestamp
.nsec
<= sd
->sd_mtime
.nsec
))
333 istate
->timestamp
.sec
<= sd
->sd_mtime
.sec
338 static int is_racy_timestamp(const struct index_state
*istate
,
339 const struct cache_entry
*ce
)
341 return (!S_ISGITLINK(ce
->ce_mode
) &&
342 is_racy_stat(istate
, &ce
->ce_stat_data
));
345 int match_stat_data_racy(const struct index_state
*istate
,
346 const struct stat_data
*sd
, struct stat
*st
)
348 if (is_racy_stat(istate
, sd
))
349 return MTIME_CHANGED
;
350 return match_stat_data(sd
, st
);
353 int ie_match_stat(struct index_state
*istate
,
354 const struct cache_entry
*ce
, struct stat
*st
,
355 unsigned int options
)
357 unsigned int changed
;
358 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
359 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
360 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
361 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
363 if (!ignore_fsmonitor
)
364 refresh_fsmonitor(istate
);
366 * If it's marked as always valid in the index, it's
367 * valid whatever the checked-out copy says.
369 * skip-worktree has the same effect with higher precedence
371 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
))
373 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
375 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
))
379 * Intent-to-add entries have not been added, so the index entry
380 * by definition never matches what is in the work tree until it
381 * actually gets added.
383 if (ce_intent_to_add(ce
))
384 return DATA_CHANGED
| TYPE_CHANGED
| MODE_CHANGED
;
386 changed
= ce_match_stat_basic(ce
, st
);
389 * Within 1 second of this sequence:
390 * echo xyzzy >file && git-update-index --add file
391 * running this command:
393 * would give a falsely clean cache entry. The mtime and
394 * length match the cache, and other stat fields do not change.
396 * We could detect this at update-index time (the cache entry
397 * being registered/updated records the same time as "now")
398 * and delay the return from git-update-index, but that would
399 * effectively mean we can make at most one commit per second,
400 * which is not acceptable. Instead, we check cache entries
401 * whose mtime are the same as the index file timestamp more
402 * carefully than others.
404 if (!changed
&& is_racy_timestamp(istate
, ce
)) {
405 if (assume_racy_is_modified
)
406 changed
|= DATA_CHANGED
;
408 changed
|= ce_modified_check_fs(ce
, st
);
414 int ie_modified(struct index_state
*istate
,
415 const struct cache_entry
*ce
,
416 struct stat
*st
, unsigned int options
)
418 int changed
, changed_fs
;
420 changed
= ie_match_stat(istate
, ce
, st
, options
);
424 * If the mode or type has changed, there's no point in trying
425 * to refresh the entry - it's not going to match
427 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
431 * Immediately after read-tree or update-index --cacheinfo,
432 * the length field is zero, as we have never even read the
433 * lstat(2) information once, and we cannot trust DATA_CHANGED
434 * returned by ie_match_stat() which in turn was returned by
435 * ce_match_stat_basic() to signal that the filesize of the
436 * blob changed. We have to actually go to the filesystem to
437 * see if the contents match, and if so, should answer "unchanged".
439 * The logic does not apply to gitlinks, as ce_match_stat_basic()
440 * already has checked the actual HEAD from the filesystem in the
441 * subproject. If ie_match_stat() already said it is different,
442 * then we know it is.
444 if ((changed
& DATA_CHANGED
) &&
445 (S_ISGITLINK(ce
->ce_mode
) || ce
->ce_stat_data
.sd_size
!= 0))
448 changed_fs
= ce_modified_check_fs(ce
, st
);
450 return changed
| changed_fs
;
454 int base_name_compare(const char *name1
, int len1
, int mode1
,
455 const char *name2
, int len2
, int mode2
)
457 unsigned char c1
, c2
;
458 int len
= len1
< len2
? len1
: len2
;
461 cmp
= memcmp(name1
, name2
, len
);
466 if (!c1
&& S_ISDIR(mode1
))
468 if (!c2
&& S_ISDIR(mode2
))
470 return (c1
< c2
) ? -1 : (c1
> c2
) ? 1 : 0;
474 * df_name_compare() is identical to base_name_compare(), except it
475 * compares conflicting directory/file entries as equal. Note that
476 * while a directory name compares as equal to a regular file, they
477 * then individually compare _differently_ to a filename that has
478 * a dot after the basename (because '\0' < '.' < '/').
480 * This is used by routines that want to traverse the git namespace
481 * but then handle conflicting entries together when possible.
483 int df_name_compare(const char *name1
, int len1
, int mode1
,
484 const char *name2
, int len2
, int mode2
)
486 int len
= len1
< len2
? len1
: len2
, cmp
;
487 unsigned char c1
, c2
;
489 cmp
= memcmp(name1
, name2
, len
);
492 /* Directories and files compare equal (same length, same name) */
496 if (!c1
&& S_ISDIR(mode1
))
499 if (!c2
&& S_ISDIR(mode2
))
501 if (c1
== '/' && !c2
)
503 if (c2
== '/' && !c1
)
508 int name_compare(const char *name1
, size_t len1
, const char *name2
, size_t len2
)
510 size_t min_len
= (len1
< len2
) ? len1
: len2
;
511 int cmp
= memcmp(name1
, name2
, min_len
);
521 int cache_name_stage_compare(const char *name1
, int len1
, int stage1
, const char *name2
, int len2
, int stage2
)
525 cmp
= name_compare(name1
, len1
, name2
, len2
);
536 static int index_name_stage_pos(const struct index_state
*istate
, const char *name
, int namelen
, int stage
)
541 last
= istate
->cache_nr
;
542 while (last
> first
) {
543 int next
= (last
+ first
) >> 1;
544 struct cache_entry
*ce
= istate
->cache
[next
];
545 int cmp
= cache_name_stage_compare(name
, namelen
, stage
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
557 int index_name_pos(const struct index_state
*istate
, const char *name
, int namelen
)
559 return index_name_stage_pos(istate
, name
, namelen
, 0);
562 int remove_index_entry_at(struct index_state
*istate
, int pos
)
564 struct cache_entry
*ce
= istate
->cache
[pos
];
566 record_resolve_undo(istate
, ce
);
567 remove_name_hash(istate
, ce
);
568 save_or_free_index_entry(istate
, ce
);
569 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
571 if (pos
>= istate
->cache_nr
)
573 MOVE_ARRAY(istate
->cache
+ pos
, istate
->cache
+ pos
+ 1,
574 istate
->cache_nr
- pos
);
579 * Remove all cache entries marked for removal, that is where
580 * CE_REMOVE is set in ce_flags. This is much more effective than
581 * calling remove_index_entry_at() for each entry to be removed.
583 void remove_marked_cache_entries(struct index_state
*istate
)
585 struct cache_entry
**ce_array
= istate
->cache
;
588 for (i
= j
= 0; i
< istate
->cache_nr
; i
++) {
589 if (ce_array
[i
]->ce_flags
& CE_REMOVE
) {
590 remove_name_hash(istate
, ce_array
[i
]);
591 save_or_free_index_entry(istate
, ce_array
[i
]);
594 ce_array
[j
++] = ce_array
[i
];
596 if (j
== istate
->cache_nr
)
598 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
599 istate
->cache_nr
= j
;
602 int remove_file_from_index(struct index_state
*istate
, const char *path
)
604 int pos
= index_name_pos(istate
, path
, strlen(path
));
607 cache_tree_invalidate_path(istate
, path
);
608 untracked_cache_remove_from_index(istate
, path
);
609 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
610 remove_index_entry_at(istate
, pos
);
614 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
616 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
619 static int index_name_pos_also_unmerged(struct index_state
*istate
,
620 const char *path
, int namelen
)
622 int pos
= index_name_pos(istate
, path
, namelen
);
623 struct cache_entry
*ce
;
628 /* maybe unmerged? */
630 if (pos
>= istate
->cache_nr
||
631 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
634 /* order of preference: stage 2, 1, 3 */
635 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
636 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
637 !compare_name(ce
, path
, namelen
))
642 static int different_name(struct cache_entry
*ce
, struct cache_entry
*alias
)
644 int len
= ce_namelen(ce
);
645 return ce_namelen(alias
) != len
|| memcmp(ce
->name
, alias
->name
, len
);
649 * If we add a filename that aliases in the cache, we will use the
650 * name that we already have - but we don't want to update the same
651 * alias twice, because that implies that there were actually two
652 * different files with aliasing names!
654 * So we use the CE_ADDED flag to verify that the alias was an old
655 * one before we accept it as
657 static struct cache_entry
*create_alias_ce(struct index_state
*istate
,
658 struct cache_entry
*ce
,
659 struct cache_entry
*alias
)
662 struct cache_entry
*new_entry
;
664 if (alias
->ce_flags
& CE_ADDED
)
665 die("Will not add file alias '%s' ('%s' already exists in index)", ce
->name
, alias
->name
);
667 /* Ok, create the new entry using the name of the existing alias */
668 len
= ce_namelen(alias
);
669 new_entry
= make_empty_cache_entry(istate
, len
);
670 memcpy(new_entry
->name
, alias
->name
, len
);
671 copy_cache_entry(new_entry
, ce
);
672 save_or_free_index_entry(istate
, ce
);
676 void set_object_name_for_intent_to_add_entry(struct cache_entry
*ce
)
678 struct object_id oid
;
679 if (write_object_file("", 0, blob_type
, &oid
))
680 die("cannot create an empty blob in the object database");
681 oidcpy(&ce
->oid
, &oid
);
684 int add_to_index(struct index_state
*istate
, const char *path
, struct stat
*st
, int flags
)
686 int namelen
, was_same
;
687 mode_t st_mode
= st
->st_mode
;
688 struct cache_entry
*ce
, *alias
= NULL
;
689 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_IGNORE_SKIP_WORKTREE
|CE_MATCH_RACY_IS_DIRTY
;
690 int verbose
= flags
& (ADD_CACHE_VERBOSE
| ADD_CACHE_PRETEND
);
691 int pretend
= flags
& ADD_CACHE_PRETEND
;
692 int intent_only
= flags
& ADD_CACHE_INTENT
;
693 int add_option
= (ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
|
694 (intent_only
? ADD_CACHE_NEW_ONLY
: 0));
695 int newflags
= HASH_WRITE_OBJECT
;
697 if (flags
& HASH_RENORMALIZE
)
698 newflags
|= HASH_RENORMALIZE
;
700 if (!S_ISREG(st_mode
) && !S_ISLNK(st_mode
) && !S_ISDIR(st_mode
))
701 return error("%s: can only add regular files, symbolic links or git-directories", path
);
703 namelen
= strlen(path
);
704 if (S_ISDIR(st_mode
)) {
705 while (namelen
&& path
[namelen
-1] == '/')
708 ce
= make_empty_cache_entry(istate
, namelen
);
709 memcpy(ce
->name
, path
, namelen
);
710 ce
->ce_namelen
= namelen
;
712 fill_stat_cache_info(ce
, st
);
714 ce
->ce_flags
|= CE_INTENT_TO_ADD
;
717 if (trust_executable_bit
&& has_symlinks
) {
718 ce
->ce_mode
= create_ce_mode(st_mode
);
720 /* If there is an existing entry, pick the mode bits and type
721 * from it, otherwise assume unexecutable regular file.
723 struct cache_entry
*ent
;
724 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
726 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
727 ce
->ce_mode
= ce_mode_from_stat(ent
, st_mode
);
730 /* When core.ignorecase=true, determine if a directory of the same name but differing
731 * case already exists within the Git repository. If it does, ensure the directory
732 * case of the file being added to the repository matches (is folded into) the existing
733 * entry's directory case.
736 adjust_dirname_case(istate
, ce
->name
);
738 if (!(flags
& HASH_RENORMALIZE
)) {
739 alias
= index_file_exists(istate
, ce
->name
,
740 ce_namelen(ce
), ignore_case
);
743 !ie_match_stat(istate
, alias
, st
, ce_option
)) {
744 /* Nothing changed, really */
745 if (!S_ISGITLINK(alias
->ce_mode
))
746 ce_mark_uptodate(alias
);
747 alias
->ce_flags
|= CE_ADDED
;
749 discard_cache_entry(ce
);
754 if (index_path(&ce
->oid
, path
, st
, newflags
)) {
755 discard_cache_entry(ce
);
756 return error("unable to index file %s", path
);
759 set_object_name_for_intent_to_add_entry(ce
);
761 if (ignore_case
&& alias
&& different_name(ce
, alias
))
762 ce
= create_alias_ce(istate
, ce
, alias
);
763 ce
->ce_flags
|= CE_ADDED
;
765 /* It was suspected to be racily clean, but it turns out to be Ok */
768 !oidcmp(&alias
->oid
, &ce
->oid
) &&
769 ce
->ce_mode
== alias
->ce_mode
);
772 discard_cache_entry(ce
);
773 else if (add_index_entry(istate
, ce
, add_option
)) {
774 discard_cache_entry(ce
);
775 return error("unable to add %s to index", path
);
777 if (verbose
&& !was_same
)
778 printf("add '%s'\n", path
);
782 int add_file_to_index(struct index_state
*istate
, const char *path
, int flags
)
785 if (lstat(path
, &st
))
786 die_errno("unable to stat '%s'", path
);
787 return add_to_index(istate
, path
, &st
, flags
);
790 struct cache_entry
*make_empty_cache_entry(struct index_state
*istate
, size_t len
)
792 return mem_pool__ce_calloc(find_mem_pool(istate
), len
);
795 struct cache_entry
*make_empty_transient_cache_entry(size_t len
)
797 return xcalloc(1, cache_entry_size(len
));
800 struct cache_entry
*make_cache_entry(struct index_state
*istate
,
802 const struct object_id
*oid
,
805 unsigned int refresh_options
)
807 struct cache_entry
*ce
, *ret
;
810 if (!verify_path(path
, mode
)) {
811 error("Invalid path '%s'", path
);
816 ce
= make_empty_cache_entry(istate
, len
);
818 oidcpy(&ce
->oid
, oid
);
819 memcpy(ce
->name
, path
, len
);
820 ce
->ce_flags
= create_ce_flags(stage
);
821 ce
->ce_namelen
= len
;
822 ce
->ce_mode
= create_ce_mode(mode
);
824 ret
= refresh_cache_entry(&the_index
, ce
, refresh_options
);
826 discard_cache_entry(ce
);
830 struct cache_entry
*make_transient_cache_entry(unsigned int mode
, const struct object_id
*oid
,
831 const char *path
, int stage
)
833 struct cache_entry
*ce
;
836 if (!verify_path(path
, mode
)) {
837 error("Invalid path '%s'", path
);
842 ce
= make_empty_transient_cache_entry(len
);
844 oidcpy(&ce
->oid
, oid
);
845 memcpy(ce
->name
, path
, len
);
846 ce
->ce_flags
= create_ce_flags(stage
);
847 ce
->ce_namelen
= len
;
848 ce
->ce_mode
= create_ce_mode(mode
);
854 * Chmod an index entry with either +x or -x.
856 * Returns -1 if the chmod for the particular cache entry failed (if it's
857 * not a regular file), -2 if an invalid flip argument is passed in, 0
860 int chmod_index_entry(struct index_state
*istate
, struct cache_entry
*ce
,
863 if (!S_ISREG(ce
->ce_mode
))
870 ce
->ce_mode
&= ~0111;
875 cache_tree_invalidate_path(istate
, ce
->name
);
876 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
877 mark_fsmonitor_invalid(istate
, ce
);
878 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
883 int ce_same_name(const struct cache_entry
*a
, const struct cache_entry
*b
)
885 int len
= ce_namelen(a
);
886 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
890 * We fundamentally don't like some paths: we don't want
891 * dot or dot-dot anywhere, and for obvious reasons don't
892 * want to recurse into ".git" either.
894 * Also, we don't want double slashes or slashes at the
895 * end that can make pathnames ambiguous.
897 static int verify_dotfile(const char *rest
, unsigned mode
)
900 * The first character was '.', but that
901 * has already been discarded, we now test
905 /* "." is not allowed */
906 if (*rest
== '\0' || is_dir_sep(*rest
))
911 * ".git" followed by NUL or slash is bad. Note that we match
912 * case-insensitively here, even if ignore_case is not set.
913 * This outlaws ".GIT" everywhere out of an abundance of caution,
914 * since there's really no good reason to allow it.
916 * Once we've seen ".git", we can also find ".gitmodules", etc (also
917 * case-insensitively).
921 if (rest
[1] != 'i' && rest
[1] != 'I')
923 if (rest
[2] != 't' && rest
[2] != 'T')
925 if (rest
[3] == '\0' || is_dir_sep(rest
[3]))
929 if (skip_iprefix(rest
, "modules", &rest
) &&
930 (*rest
== '\0' || is_dir_sep(*rest
)))
935 if (rest
[1] == '\0' || is_dir_sep(rest
[1]))
941 int verify_path(const char *path
, unsigned mode
)
945 if (has_dos_drive_prefix(path
))
955 if (is_hfs_dotgit(path
))
958 if (is_hfs_dotgitmodules(path
))
963 if (is_ntfs_dotgit(path
))
966 if (is_ntfs_dotgitmodules(path
))
972 if ((c
== '.' && !verify_dotfile(path
, mode
)) ||
973 is_dir_sep(c
) || c
== '\0')
981 * Do we have another file that has the beginning components being a
982 * proper superset of the name we're trying to add?
984 static int has_file_name(struct index_state
*istate
,
985 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
988 int len
= ce_namelen(ce
);
989 int stage
= ce_stage(ce
);
990 const char *name
= ce
->name
;
992 while (pos
< istate
->cache_nr
) {
993 struct cache_entry
*p
= istate
->cache
[pos
++];
995 if (len
>= ce_namelen(p
))
997 if (memcmp(name
, p
->name
, len
))
999 if (ce_stage(p
) != stage
)
1001 if (p
->name
[len
] != '/')
1003 if (p
->ce_flags
& CE_REMOVE
)
1008 remove_index_entry_at(istate
, --pos
);
1015 * Like strcmp(), but also return the offset of the first change.
1016 * If strings are equal, return the length.
1018 int strcmp_offset(const char *s1
, const char *s2
, size_t *first_change
)
1023 return strcmp(s1
, s2
);
1025 for (k
= 0; s1
[k
] == s2
[k
]; k
++)
1030 return (unsigned char)s1
[k
] - (unsigned char)s2
[k
];
1034 * Do we have another file with a pathname that is a proper
1035 * subset of the name we're trying to add?
1037 * That is, is there another file in the index with a path
1038 * that matches a sub-directory in the given entry?
1040 static int has_dir_name(struct index_state
*istate
,
1041 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
1044 int stage
= ce_stage(ce
);
1045 const char *name
= ce
->name
;
1046 const char *slash
= name
+ ce_namelen(ce
);
1051 * We are frequently called during an iteration on a sorted
1052 * list of pathnames and while building a new index. Therefore,
1053 * there is a high probability that this entry will eventually
1054 * be appended to the index, rather than inserted in the middle.
1055 * If we can confirm that, we can avoid binary searches on the
1056 * components of the pathname.
1058 * Compare the entry's full path with the last path in the index.
1060 if (istate
->cache_nr
> 0) {
1061 cmp_last
= strcmp_offset(name
,
1062 istate
->cache
[istate
->cache_nr
- 1]->name
,
1065 if (len_eq_last
== 0) {
1067 * The entry sorts AFTER the last one in the
1068 * index and their paths have no common prefix,
1069 * so there cannot be a F/D conflict.
1074 * The entry sorts AFTER the last one in the
1075 * index, but has a common prefix. Fall through
1076 * to the loop below to disect the entry's path
1077 * and see where the difference is.
1080 } else if (cmp_last
== 0) {
1082 * The entry exactly matches the last one in the
1083 * index, but because of multiple stage and CE_REMOVE
1084 * items, we fall through and let the regular search
1094 if (*--slash
== '/')
1096 if (slash
<= ce
->name
)
1103 * (len + 1) is a directory boundary (including
1104 * the trailing slash). And since the loop is
1105 * decrementing "slash", the first iteration is
1106 * the longest directory prefix; subsequent
1107 * iterations consider parent directories.
1110 if (len
+ 1 <= len_eq_last
) {
1112 * The directory prefix (including the trailing
1113 * slash) also appears as a prefix in the last
1114 * entry, so the remainder cannot collide (because
1115 * strcmp said the whole path was greater).
1120 * LT: last: xxx/file_A
1126 if (len
> len_eq_last
) {
1128 * This part of the directory prefix (excluding
1129 * the trailing slash) is longer than the known
1130 * equal portions, so this sub-directory cannot
1131 * collide with a file.
1139 if (istate
->cache_nr
> 0 &&
1140 ce_namelen(istate
->cache
[istate
->cache_nr
- 1]) > len
) {
1142 * The directory prefix lines up with part of
1143 * a longer file or directory name, but sorts
1144 * after it, so this sub-directory cannot
1145 * collide with a file.
1147 * last: xxx/yy-file (because '-' sorts before '/')
1154 * This is a possible collision. Fall through and
1155 * let the regular search code handle it.
1162 pos
= index_name_stage_pos(istate
, name
, len
, stage
);
1165 * Found one, but not so fast. This could
1166 * be a marker that says "I was here, but
1167 * I am being removed". Such an entry is
1168 * not a part of the resulting tree, and
1169 * it is Ok to have a directory at the same
1172 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
1176 remove_index_entry_at(istate
, pos
);
1184 * Trivial optimization: if we find an entry that
1185 * already matches the sub-directory, then we know
1186 * we're ok, and we can exit.
1188 while (pos
< istate
->cache_nr
) {
1189 struct cache_entry
*p
= istate
->cache
[pos
];
1190 if ((ce_namelen(p
) <= len
) ||
1191 (p
->name
[len
] != '/') ||
1192 memcmp(p
->name
, name
, len
))
1193 break; /* not our subdirectory */
1194 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
1196 * p is at the same stage as our entry, and
1197 * is a subdirectory of what we are looking
1198 * at, so we cannot have conflicts at our
1199 * level or anything shorter.
1208 /* We may be in a situation where we already have path/file and path
1209 * is being added, or we already have path and path/file is being
1210 * added. Either one would result in a nonsense tree that has path
1211 * twice when git-write-tree tries to write it out. Prevent it.
1213 * If ok-to-replace is specified, we remove the conflicting entries
1214 * from the cache so the caller should recompute the insert position.
1215 * When this happens, we return non-zero.
1217 static int check_file_directory_conflict(struct index_state
*istate
,
1218 const struct cache_entry
*ce
,
1219 int pos
, int ok_to_replace
)
1224 * When ce is an "I am going away" entry, we allow it to be added
1226 if (ce
->ce_flags
& CE_REMOVE
)
1230 * We check if the path is a sub-path of a subsequent pathname
1231 * first, since removing those will not change the position
1234 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
1237 * Then check if the path might have a clashing sub-directory
1240 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
1243 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1246 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
1247 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
1248 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
1249 int new_only
= option
& ADD_CACHE_NEW_ONLY
;
1251 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1252 cache_tree_invalidate_path(istate
, ce
->name
);
1255 * If this entry's path sorts after the last entry in the index,
1256 * we can avoid searching for it.
1258 if (istate
->cache_nr
> 0 &&
1259 strcmp(ce
->name
, istate
->cache
[istate
->cache_nr
- 1]->name
) > 0)
1260 pos
= -istate
->cache_nr
- 1;
1262 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1264 /* existing match? Just replace it. */
1267 replace_index_entry(istate
, pos
, ce
);
1272 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1273 untracked_cache_add_to_index(istate
, ce
->name
);
1276 * Inserting a merged entry ("stage 0") into the index
1277 * will always replace all non-merged entries..
1279 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
1280 while (ce_same_name(istate
->cache
[pos
], ce
)) {
1282 if (!remove_index_entry_at(istate
, pos
))
1289 if (!verify_path(ce
->name
, ce
->ce_mode
))
1290 return error("Invalid path '%s'", ce
->name
);
1292 if (!skip_df_check
&&
1293 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
1295 return error("'%s' appears as both a file and as a directory",
1297 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1303 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1307 if (option
& ADD_CACHE_JUST_APPEND
)
1308 pos
= istate
->cache_nr
;
1311 ret
= add_index_entry_with_check(istate
, ce
, option
);
1317 /* Make sure the array is big enough .. */
1318 ALLOC_GROW(istate
->cache
, istate
->cache_nr
+ 1, istate
->cache_alloc
);
1322 if (istate
->cache_nr
> pos
+ 1)
1323 MOVE_ARRAY(istate
->cache
+ pos
+ 1, istate
->cache
+ pos
,
1324 istate
->cache_nr
- pos
- 1);
1325 set_index_entry(istate
, pos
, ce
);
1326 istate
->cache_changed
|= CE_ENTRY_ADDED
;
1331 * "refresh" does not calculate a new sha1 file or bring the
1332 * cache up-to-date for mode/content changes. But what it
1333 * _does_ do is to "re-match" the stat information of a file
1334 * with the cache, so that you can refresh the cache for a
1335 * file that hasn't been changed but where the stat entry is
1338 * For example, you'd want to do this after doing a "git-read-tree",
1339 * to link up the stat cache details with the proper files.
1341 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
1342 struct cache_entry
*ce
,
1343 unsigned int options
, int *err
,
1347 struct cache_entry
*updated
;
1349 int refresh
= options
& CE_MATCH_REFRESH
;
1350 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
1351 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
1352 int ignore_missing
= options
& CE_MATCH_IGNORE_MISSING
;
1353 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
1355 if (!refresh
|| ce_uptodate(ce
))
1358 if (!ignore_fsmonitor
)
1359 refresh_fsmonitor(istate
);
1361 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1362 * that the change to the work tree does not matter and told
1365 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
)) {
1366 ce_mark_uptodate(ce
);
1369 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
)) {
1370 ce_mark_uptodate(ce
);
1373 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
)) {
1374 ce_mark_uptodate(ce
);
1378 if (has_symlink_leading_path(ce
->name
, ce_namelen(ce
))) {
1386 if (lstat(ce
->name
, &st
) < 0) {
1387 if (ignore_missing
&& errno
== ENOENT
)
1394 changed
= ie_match_stat(istate
, ce
, &st
, options
);
1396 *changed_ret
= changed
;
1399 * The path is unchanged. If we were told to ignore
1400 * valid bit, then we did the actual stat check and
1401 * found that the entry is unmodified. If the entry
1402 * is not marked VALID, this is the place to mark it
1403 * valid again, under "assume unchanged" mode.
1405 if (ignore_valid
&& assume_unchanged
&&
1406 !(ce
->ce_flags
& CE_VALID
))
1407 ; /* mark this one VALID again */
1410 * We do not mark the index itself "modified"
1411 * because CE_UPTODATE flag is in-core only;
1412 * we are not going to write this change out.
1414 if (!S_ISGITLINK(ce
->ce_mode
)) {
1415 ce_mark_uptodate(ce
);
1416 mark_fsmonitor_valid(ce
);
1422 if (ie_modified(istate
, ce
, &st
, options
)) {
1428 updated
= make_empty_cache_entry(istate
, ce_namelen(ce
));
1429 copy_cache_entry(updated
, ce
);
1430 memcpy(updated
->name
, ce
->name
, ce
->ce_namelen
+ 1);
1431 fill_stat_cache_info(updated
, &st
);
1433 * If ignore_valid is not set, we should leave CE_VALID bit
1434 * alone. Otherwise, paths marked with --no-assume-unchanged
1435 * (i.e. things to be edited) will reacquire CE_VALID bit
1436 * automatically, which is not really what we want.
1438 if (!ignore_valid
&& assume_unchanged
&&
1439 !(ce
->ce_flags
& CE_VALID
))
1440 updated
->ce_flags
&= ~CE_VALID
;
1442 /* istate->cache_changed is updated in the caller */
1446 static void show_file(const char * fmt
, const char * name
, int in_porcelain
,
1447 int * first
, const char *header_msg
)
1449 if (in_porcelain
&& *first
&& header_msg
) {
1450 printf("%s\n", header_msg
);
1456 int refresh_index(struct index_state
*istate
, unsigned int flags
,
1457 const struct pathspec
*pathspec
,
1458 char *seen
, const char *header_msg
)
1462 int really
= (flags
& REFRESH_REALLY
) != 0;
1463 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
1464 int quiet
= (flags
& REFRESH_QUIET
) != 0;
1465 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
1466 int ignore_submodules
= (flags
& REFRESH_IGNORE_SUBMODULES
) != 0;
1468 int in_porcelain
= (flags
& REFRESH_IN_PORCELAIN
);
1469 unsigned int options
= (CE_MATCH_REFRESH
|
1470 (really
? CE_MATCH_IGNORE_VALID
: 0) |
1471 (not_new
? CE_MATCH_IGNORE_MISSING
: 0));
1472 const char *modified_fmt
;
1473 const char *deleted_fmt
;
1474 const char *typechange_fmt
;
1475 const char *added_fmt
;
1476 const char *unmerged_fmt
;
1477 uint64_t start
= getnanotime();
1479 modified_fmt
= (in_porcelain
? "M\t%s\n" : "%s: needs update\n");
1480 deleted_fmt
= (in_porcelain
? "D\t%s\n" : "%s: needs update\n");
1481 typechange_fmt
= (in_porcelain
? "T\t%s\n" : "%s needs update\n");
1482 added_fmt
= (in_porcelain
? "A\t%s\n" : "%s needs update\n");
1483 unmerged_fmt
= (in_porcelain
? "U\t%s\n" : "%s: needs merge\n");
1484 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1485 struct cache_entry
*ce
, *new_entry
;
1486 int cache_errno
= 0;
1490 ce
= istate
->cache
[i
];
1491 if (ignore_submodules
&& S_ISGITLINK(ce
->ce_mode
))
1494 if (pathspec
&& !ce_path_match(ce
, pathspec
, seen
))
1498 while ((i
< istate
->cache_nr
) &&
1499 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
1505 show_file(unmerged_fmt
, ce
->name
, in_porcelain
,
1506 &first
, header_msg
);
1514 new_entry
= refresh_cache_ent(istate
, ce
, options
, &cache_errno
, &changed
);
1515 if (new_entry
== ce
)
1520 if (really
&& cache_errno
== EINVAL
) {
1521 /* If we are doing --really-refresh that
1522 * means the index is not valid anymore.
1524 ce
->ce_flags
&= ~CE_VALID
;
1525 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
1526 mark_fsmonitor_invalid(istate
, ce
);
1527 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
1532 if (cache_errno
== ENOENT
)
1534 else if (ce_intent_to_add(ce
))
1535 fmt
= added_fmt
; /* must be before other checks */
1536 else if (changed
& TYPE_CHANGED
)
1537 fmt
= typechange_fmt
;
1541 ce
->name
, in_porcelain
, &first
, header_msg
);
1546 replace_index_entry(istate
, i
, new_entry
);
1548 trace_performance_since(start
, "refresh index");
1552 struct cache_entry
*refresh_cache_entry(struct index_state
*istate
,
1553 struct cache_entry
*ce
,
1554 unsigned int options
)
1556 return refresh_cache_ent(istate
, ce
, options
, NULL
, NULL
);
1560 /*****************************************************************
1562 *****************************************************************/
1564 #define INDEX_FORMAT_DEFAULT 3
1566 static unsigned int get_index_format_default(void)
1568 char *envversion
= getenv("GIT_INDEX_VERSION");
1571 unsigned int version
= INDEX_FORMAT_DEFAULT
;
1574 if (!git_config_get_int("index.version", &value
))
1576 if (version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1577 warning(_("index.version set, but the value is invalid.\n"
1578 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1579 return INDEX_FORMAT_DEFAULT
;
1584 version
= strtoul(envversion
, &endp
, 10);
1586 version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1587 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1588 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1589 version
= INDEX_FORMAT_DEFAULT
;
1595 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1596 * Again - this is just a (very strong in practice) heuristic that
1597 * the inode hasn't changed.
1599 * We save the fields in big-endian order to allow using the
1600 * index file over NFS transparently.
1602 struct ondisk_cache_entry
{
1603 struct cache_time ctime
;
1604 struct cache_time mtime
;
1611 unsigned char sha1
[20];
1613 char name
[FLEX_ARRAY
]; /* more */
1617 * This struct is used when CE_EXTENDED bit is 1
1618 * The struct must match ondisk_cache_entry exactly from
1621 struct ondisk_cache_entry_extended
{
1622 struct cache_time ctime
;
1623 struct cache_time mtime
;
1630 unsigned char sha1
[20];
1633 char name
[FLEX_ARRAY
]; /* more */
1636 /* These are only used for v3 or lower */
1637 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1638 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1639 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1640 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1641 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1642 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1643 ondisk_cache_entry_size(ce_namelen(ce)))
1645 /* Allow fsck to force verification of the index checksum. */
1646 int verify_index_checksum
;
1648 /* Allow fsck to force verification of the cache entry order. */
1649 int verify_ce_order
;
1651 static int verify_hdr(struct cache_header
*hdr
, unsigned long size
)
1654 unsigned char hash
[GIT_MAX_RAWSZ
];
1657 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
1658 return error("bad signature");
1659 hdr_version
= ntohl(hdr
->hdr_version
);
1660 if (hdr_version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< hdr_version
)
1661 return error("bad index version %d", hdr_version
);
1663 if (!verify_index_checksum
)
1666 the_hash_algo
->init_fn(&c
);
1667 the_hash_algo
->update_fn(&c
, hdr
, size
- the_hash_algo
->rawsz
);
1668 the_hash_algo
->final_fn(hash
, &c
);
1669 if (hashcmp(hash
, (unsigned char *)hdr
+ size
- the_hash_algo
->rawsz
))
1670 return error("bad index file sha1 signature");
1674 static int read_index_extension(struct index_state
*istate
,
1675 const char *ext
, void *data
, unsigned long sz
)
1677 switch (CACHE_EXT(ext
)) {
1678 case CACHE_EXT_TREE
:
1679 istate
->cache_tree
= cache_tree_read(data
, sz
);
1681 case CACHE_EXT_RESOLVE_UNDO
:
1682 istate
->resolve_undo
= resolve_undo_read(data
, sz
);
1684 case CACHE_EXT_LINK
:
1685 if (read_link_extension(istate
, data
, sz
))
1688 case CACHE_EXT_UNTRACKED
:
1689 istate
->untracked
= read_untracked_extension(data
, sz
);
1691 case CACHE_EXT_FSMONITOR
:
1692 read_fsmonitor_extension(istate
, data
, sz
);
1695 if (*ext
< 'A' || 'Z' < *ext
)
1696 return error("index uses %.4s extension, which we do not understand",
1698 fprintf(stderr
, "ignoring %.4s extension\n", ext
);
1704 int hold_locked_index(struct lock_file
*lk
, int lock_flags
)
1706 return hold_lock_file_for_update(lk
, get_index_file(), lock_flags
);
1709 int read_index(struct index_state
*istate
)
1711 return read_index_from(istate
, get_index_file(), get_git_dir());
1714 static struct cache_entry
*cache_entry_from_ondisk(struct mem_pool
*mem_pool
,
1715 struct ondisk_cache_entry
*ondisk
,
1720 struct cache_entry
*ce
= mem_pool__ce_alloc(mem_pool
, len
);
1722 ce
->ce_stat_data
.sd_ctime
.sec
= get_be32(&ondisk
->ctime
.sec
);
1723 ce
->ce_stat_data
.sd_mtime
.sec
= get_be32(&ondisk
->mtime
.sec
);
1724 ce
->ce_stat_data
.sd_ctime
.nsec
= get_be32(&ondisk
->ctime
.nsec
);
1725 ce
->ce_stat_data
.sd_mtime
.nsec
= get_be32(&ondisk
->mtime
.nsec
);
1726 ce
->ce_stat_data
.sd_dev
= get_be32(&ondisk
->dev
);
1727 ce
->ce_stat_data
.sd_ino
= get_be32(&ondisk
->ino
);
1728 ce
->ce_mode
= get_be32(&ondisk
->mode
);
1729 ce
->ce_stat_data
.sd_uid
= get_be32(&ondisk
->uid
);
1730 ce
->ce_stat_data
.sd_gid
= get_be32(&ondisk
->gid
);
1731 ce
->ce_stat_data
.sd_size
= get_be32(&ondisk
->size
);
1732 ce
->ce_flags
= flags
& ~CE_NAMEMASK
;
1733 ce
->ce_namelen
= len
;
1735 hashcpy(ce
->oid
.hash
, ondisk
->sha1
);
1736 memcpy(ce
->name
, name
, len
);
1737 ce
->name
[len
] = '\0';
1742 * Adjacent cache entries tend to share the leading paths, so it makes
1743 * sense to only store the differences in later entries. In the v4
1744 * on-disk format of the index, each on-disk cache entry stores the
1745 * number of bytes to be stripped from the end of the previous name,
1746 * and the bytes to append to the result, to come up with its name.
1748 static unsigned long expand_name_field(struct strbuf
*name
, const char *cp_
)
1750 const unsigned char *ep
, *cp
= (const unsigned char *)cp_
;
1751 size_t len
= decode_varint(&cp
);
1753 if (name
->len
< len
)
1754 die("malformed name field in the index");
1755 strbuf_remove(name
, name
->len
- len
, len
);
1756 for (ep
= cp
; *ep
; ep
++)
1757 ; /* find the end */
1758 strbuf_add(name
, cp
, ep
- cp
);
1759 return (const char *)ep
+ 1 - cp_
;
1762 static struct cache_entry
*create_from_disk(struct mem_pool
*mem_pool
,
1763 struct ondisk_cache_entry
*ondisk
,
1764 unsigned long *ent_size
,
1765 struct strbuf
*previous_name
)
1767 struct cache_entry
*ce
;
1772 /* On-disk flags are just 16 bits */
1773 flags
= get_be16(&ondisk
->flags
);
1774 len
= flags
& CE_NAMEMASK
;
1776 if (flags
& CE_EXTENDED
) {
1777 struct ondisk_cache_entry_extended
*ondisk2
;
1779 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
1780 extended_flags
= get_be16(&ondisk2
->flags2
) << 16;
1781 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1782 if (extended_flags
& ~CE_EXTENDED_FLAGS
)
1783 die("Unknown index entry format %08x", extended_flags
);
1784 flags
|= extended_flags
;
1785 name
= ondisk2
->name
;
1788 name
= ondisk
->name
;
1790 if (!previous_name
) {
1791 /* v3 and earlier */
1792 if (len
== CE_NAMEMASK
)
1794 ce
= cache_entry_from_ondisk(mem_pool
, ondisk
, flags
, name
, len
);
1796 *ent_size
= ondisk_ce_size(ce
);
1798 unsigned long consumed
;
1799 consumed
= expand_name_field(previous_name
, name
);
1800 ce
= cache_entry_from_ondisk(mem_pool
, ondisk
, flags
,
1802 previous_name
->len
);
1804 *ent_size
= (name
- ((char *)ondisk
)) + consumed
;
1809 static void check_ce_order(struct index_state
*istate
)
1813 if (!verify_ce_order
)
1816 for (i
= 1; i
< istate
->cache_nr
; i
++) {
1817 struct cache_entry
*ce
= istate
->cache
[i
- 1];
1818 struct cache_entry
*next_ce
= istate
->cache
[i
];
1819 int name_compare
= strcmp(ce
->name
, next_ce
->name
);
1821 if (0 < name_compare
)
1822 die("unordered stage entries in index");
1823 if (!name_compare
) {
1825 die("multiple stage entries for merged file '%s'",
1827 if (ce_stage(ce
) > ce_stage(next_ce
))
1828 die("unordered stage entries for '%s'",
1834 static void tweak_untracked_cache(struct index_state
*istate
)
1836 switch (git_config_get_untracked_cache()) {
1837 case -1: /* keep: do nothing */
1840 remove_untracked_cache(istate
);
1843 add_untracked_cache(istate
);
1845 default: /* unknown value: do nothing */
1850 static void tweak_split_index(struct index_state
*istate
)
1852 switch (git_config_get_split_index()) {
1853 case -1: /* unset: do nothing */
1856 remove_split_index(istate
);
1859 add_split_index(istate
);
1861 default: /* unknown value: do nothing */
1866 static void post_read_index_from(struct index_state
*istate
)
1868 check_ce_order(istate
);
1869 tweak_untracked_cache(istate
);
1870 tweak_split_index(istate
);
1871 tweak_fsmonitor(istate
);
1874 static size_t estimate_cache_size_from_compressed(unsigned int entries
)
1876 return entries
* (sizeof(struct cache_entry
) + CACHE_ENTRY_PATH_LENGTH
);
1879 static size_t estimate_cache_size(size_t ondisk_size
, unsigned int entries
)
1881 long per_entry
= sizeof(struct cache_entry
) - sizeof(struct ondisk_cache_entry
);
1884 * Account for potential alignment differences.
1886 per_entry
+= align_padding_size(sizeof(struct cache_entry
), -sizeof(struct ondisk_cache_entry
));
1887 return ondisk_size
+ entries
* per_entry
;
1890 /* remember to discard_cache() before reading a different cache! */
1891 int do_read_index(struct index_state
*istate
, const char *path
, int must_exist
)
1895 unsigned long src_offset
;
1896 struct cache_header
*hdr
;
1899 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
1901 if (istate
->initialized
)
1902 return istate
->cache_nr
;
1904 istate
->timestamp
.sec
= 0;
1905 istate
->timestamp
.nsec
= 0;
1906 fd
= open(path
, O_RDONLY
);
1908 if (!must_exist
&& errno
== ENOENT
)
1910 die_errno("%s: index file open failed", path
);
1914 die_errno("cannot stat the open index");
1916 mmap_size
= xsize_t(st
.st_size
);
1917 if (mmap_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
1918 die("index file smaller than expected");
1920 mmap
= xmmap(NULL
, mmap_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1921 if (mmap
== MAP_FAILED
)
1922 die_errno("unable to map index file");
1926 if (verify_hdr(hdr
, mmap_size
) < 0)
1929 hashcpy(istate
->oid
.hash
, (const unsigned char *)hdr
+ mmap_size
- the_hash_algo
->rawsz
);
1930 istate
->version
= ntohl(hdr
->hdr_version
);
1931 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
1932 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
1933 istate
->cache
= xcalloc(istate
->cache_alloc
, sizeof(*istate
->cache
));
1934 istate
->initialized
= 1;
1936 if (istate
->version
== 4) {
1937 previous_name
= &previous_name_buf
;
1938 mem_pool_init(&istate
->ce_mem_pool
,
1939 estimate_cache_size_from_compressed(istate
->cache_nr
));
1941 previous_name
= NULL
;
1942 mem_pool_init(&istate
->ce_mem_pool
,
1943 estimate_cache_size(mmap_size
, istate
->cache_nr
));
1946 src_offset
= sizeof(*hdr
);
1947 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1948 struct ondisk_cache_entry
*disk_ce
;
1949 struct cache_entry
*ce
;
1950 unsigned long consumed
;
1952 disk_ce
= (struct ondisk_cache_entry
*)((char *)mmap
+ src_offset
);
1953 ce
= create_from_disk(istate
->ce_mem_pool
, disk_ce
, &consumed
, previous_name
);
1954 set_index_entry(istate
, i
, ce
);
1956 src_offset
+= consumed
;
1958 strbuf_release(&previous_name_buf
);
1959 istate
->timestamp
.sec
= st
.st_mtime
;
1960 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
1962 while (src_offset
<= mmap_size
- the_hash_algo
->rawsz
- 8) {
1963 /* After an array of active_nr index entries,
1964 * there can be arbitrary number of extended
1965 * sections, each of which is prefixed with
1966 * extension name (4-byte) and section length
1967 * in 4-byte network byte order.
1970 memcpy(&extsize
, (char *)mmap
+ src_offset
+ 4, 4);
1971 extsize
= ntohl(extsize
);
1972 if (read_index_extension(istate
,
1973 (const char *) mmap
+ src_offset
,
1974 (char *) mmap
+ src_offset
+ 8,
1978 src_offset
+= extsize
;
1980 munmap(mmap
, mmap_size
);
1981 return istate
->cache_nr
;
1984 munmap(mmap
, mmap_size
);
1985 die("index file corrupt");
1989 * Signal that the shared index is used by updating its mtime.
1991 * This way, shared index can be removed if they have not been used
1994 static void freshen_shared_index(const char *shared_index
, int warn
)
1996 if (!check_and_freshen_file(shared_index
, 1) && warn
)
1997 warning("could not freshen shared index '%s'", shared_index
);
2000 int read_index_from(struct index_state
*istate
, const char *path
,
2003 uint64_t start
= getnanotime();
2004 struct split_index
*split_index
;
2009 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2010 if (istate
->initialized
)
2011 return istate
->cache_nr
;
2013 ret
= do_read_index(istate
, path
, 0);
2014 trace_performance_since(start
, "read cache %s", path
);
2016 split_index
= istate
->split_index
;
2017 if (!split_index
|| is_null_oid(&split_index
->base_oid
)) {
2018 post_read_index_from(istate
);
2022 if (split_index
->base
)
2023 discard_index(split_index
->base
);
2025 split_index
->base
= xcalloc(1, sizeof(*split_index
->base
));
2027 base_oid_hex
= oid_to_hex(&split_index
->base_oid
);
2028 base_path
= xstrfmt("%s/sharedindex.%s", gitdir
, base_oid_hex
);
2029 ret
= do_read_index(split_index
->base
, base_path
, 1);
2030 if (oidcmp(&split_index
->base_oid
, &split_index
->base
->oid
))
2031 die("broken index, expect %s in %s, got %s",
2032 base_oid_hex
, base_path
,
2033 oid_to_hex(&split_index
->base
->oid
));
2035 freshen_shared_index(base_path
, 0);
2036 merge_base_index(istate
);
2037 post_read_index_from(istate
);
2038 trace_performance_since(start
, "read cache %s", base_path
);
2043 int is_index_unborn(struct index_state
*istate
)
2045 return (!istate
->cache_nr
&& !istate
->timestamp
.sec
);
2048 int discard_index(struct index_state
*istate
)
2051 * Cache entries in istate->cache[] should have been allocated
2052 * from the memory pool associated with this index, or from an
2053 * associated split_index. There is no need to free individual
2054 * cache entries. validate_cache_entries can detect when this
2055 * assertion does not hold.
2057 validate_cache_entries(istate
);
2059 resolve_undo_clear_index(istate
);
2060 istate
->cache_nr
= 0;
2061 istate
->cache_changed
= 0;
2062 istate
->timestamp
.sec
= 0;
2063 istate
->timestamp
.nsec
= 0;
2064 free_name_hash(istate
);
2065 cache_tree_free(&(istate
->cache_tree
));
2066 istate
->initialized
= 0;
2067 FREE_AND_NULL(istate
->cache
);
2068 istate
->cache_alloc
= 0;
2069 discard_split_index(istate
);
2070 free_untracked_cache(istate
->untracked
);
2071 istate
->untracked
= NULL
;
2073 if (istate
->ce_mem_pool
) {
2074 mem_pool_discard(istate
->ce_mem_pool
, should_validate_cache_entries());
2075 istate
->ce_mem_pool
= NULL
;
2082 * Validate the cache entries of this index.
2083 * All cache entries associated with this index
2084 * should have been allocated by the memory pool
2085 * associated with this index, or by a referenced
2088 void validate_cache_entries(const struct index_state
*istate
)
2092 if (!should_validate_cache_entries() ||!istate
|| !istate
->initialized
)
2095 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2097 die("internal error: cache entry is not allocated from expected memory pool");
2098 } else if (!istate
->ce_mem_pool
||
2099 !mem_pool_contains(istate
->ce_mem_pool
, istate
->cache
[i
])) {
2100 if (!istate
->split_index
||
2101 !istate
->split_index
->base
||
2102 !istate
->split_index
->base
->ce_mem_pool
||
2103 !mem_pool_contains(istate
->split_index
->base
->ce_mem_pool
, istate
->cache
[i
])) {
2104 die("internal error: cache entry is not allocated from expected memory pool");
2109 if (istate
->split_index
)
2110 validate_cache_entries(istate
->split_index
->base
);
2113 int unmerged_index(const struct index_state
*istate
)
2116 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2117 if (ce_stage(istate
->cache
[i
]))
2123 #define WRITE_BUFFER_SIZE 8192
2124 static unsigned char write_buffer
[WRITE_BUFFER_SIZE
];
2125 static unsigned long write_buffer_len
;
2127 static int ce_write_flush(git_hash_ctx
*context
, int fd
)
2129 unsigned int buffered
= write_buffer_len
;
2131 the_hash_algo
->update_fn(context
, write_buffer
, buffered
);
2132 if (write_in_full(fd
, write_buffer
, buffered
) < 0)
2134 write_buffer_len
= 0;
2139 static int ce_write(git_hash_ctx
*context
, int fd
, void *data
, unsigned int len
)
2142 unsigned int buffered
= write_buffer_len
;
2143 unsigned int partial
= WRITE_BUFFER_SIZE
- buffered
;
2146 memcpy(write_buffer
+ buffered
, data
, partial
);
2147 buffered
+= partial
;
2148 if (buffered
== WRITE_BUFFER_SIZE
) {
2149 write_buffer_len
= buffered
;
2150 if (ce_write_flush(context
, fd
))
2154 write_buffer_len
= buffered
;
2156 data
= (char *) data
+ partial
;
2161 static int write_index_ext_header(git_hash_ctx
*context
, int fd
,
2162 unsigned int ext
, unsigned int sz
)
2166 return ((ce_write(context
, fd
, &ext
, 4) < 0) ||
2167 (ce_write(context
, fd
, &sz
, 4) < 0)) ? -1 : 0;
2170 static int ce_flush(git_hash_ctx
*context
, int fd
, unsigned char *hash
)
2172 unsigned int left
= write_buffer_len
;
2175 write_buffer_len
= 0;
2176 the_hash_algo
->update_fn(context
, write_buffer
, left
);
2179 /* Flush first if not enough space for hash signature */
2180 if (left
+ the_hash_algo
->rawsz
> WRITE_BUFFER_SIZE
) {
2181 if (write_in_full(fd
, write_buffer
, left
) < 0)
2186 /* Append the hash signature at the end */
2187 the_hash_algo
->final_fn(write_buffer
+ left
, context
);
2188 hashcpy(hash
, write_buffer
+ left
);
2189 left
+= the_hash_algo
->rawsz
;
2190 return (write_in_full(fd
, write_buffer
, left
) < 0) ? -1 : 0;
2193 static void ce_smudge_racily_clean_entry(struct cache_entry
*ce
)
2196 * The only thing we care about in this function is to smudge the
2197 * falsely clean entry due to touch-update-touch race, so we leave
2198 * everything else as they are. We are called for entries whose
2199 * ce_stat_data.sd_mtime match the index file mtime.
2201 * Note that this actually does not do much for gitlinks, for
2202 * which ce_match_stat_basic() always goes to the actual
2203 * contents. The caller checks with is_racy_timestamp() which
2204 * always says "no" for gitlinks, so we are not called for them ;-)
2208 if (lstat(ce
->name
, &st
) < 0)
2210 if (ce_match_stat_basic(ce
, &st
))
2212 if (ce_modified_check_fs(ce
, &st
)) {
2213 /* This is "racily clean"; smudge it. Note that this
2214 * is a tricky code. At first glance, it may appear
2215 * that it can break with this sequence:
2217 * $ echo xyzzy >frotz
2218 * $ git-update-index --add frotz
2221 * $ echo filfre >nitfol
2222 * $ git-update-index --add nitfol
2224 * but it does not. When the second update-index runs,
2225 * it notices that the entry "frotz" has the same timestamp
2226 * as index, and if we were to smudge it by resetting its
2227 * size to zero here, then the object name recorded
2228 * in index is the 6-byte file but the cached stat information
2229 * becomes zero --- which would then match what we would
2230 * obtain from the filesystem next time we stat("frotz").
2232 * However, the second update-index, before calling
2233 * this function, notices that the cached size is 6
2234 * bytes and what is on the filesystem is an empty
2235 * file, and never calls us, so the cached size information
2236 * for "frotz" stays 6 which does not match the filesystem.
2238 ce
->ce_stat_data
.sd_size
= 0;
2242 /* Copy miscellaneous fields but not the name */
2243 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry
*ondisk
,
2244 struct cache_entry
*ce
)
2248 ondisk
->ctime
.sec
= htonl(ce
->ce_stat_data
.sd_ctime
.sec
);
2249 ondisk
->mtime
.sec
= htonl(ce
->ce_stat_data
.sd_mtime
.sec
);
2250 ondisk
->ctime
.nsec
= htonl(ce
->ce_stat_data
.sd_ctime
.nsec
);
2251 ondisk
->mtime
.nsec
= htonl(ce
->ce_stat_data
.sd_mtime
.nsec
);
2252 ondisk
->dev
= htonl(ce
->ce_stat_data
.sd_dev
);
2253 ondisk
->ino
= htonl(ce
->ce_stat_data
.sd_ino
);
2254 ondisk
->mode
= htonl(ce
->ce_mode
);
2255 ondisk
->uid
= htonl(ce
->ce_stat_data
.sd_uid
);
2256 ondisk
->gid
= htonl(ce
->ce_stat_data
.sd_gid
);
2257 ondisk
->size
= htonl(ce
->ce_stat_data
.sd_size
);
2258 hashcpy(ondisk
->sha1
, ce
->oid
.hash
);
2260 flags
= ce
->ce_flags
& ~CE_NAMEMASK
;
2261 flags
|= (ce_namelen(ce
) >= CE_NAMEMASK
? CE_NAMEMASK
: ce_namelen(ce
));
2262 ondisk
->flags
= htons(flags
);
2263 if (ce
->ce_flags
& CE_EXTENDED
) {
2264 struct ondisk_cache_entry_extended
*ondisk2
;
2265 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
2266 ondisk2
->flags2
= htons((ce
->ce_flags
& CE_EXTENDED_FLAGS
) >> 16);
2270 static int ce_write_entry(git_hash_ctx
*c
, int fd
, struct cache_entry
*ce
,
2271 struct strbuf
*previous_name
, struct ondisk_cache_entry
*ondisk
)
2275 unsigned int saved_namelen
;
2276 int stripped_name
= 0;
2277 static unsigned char padding
[8] = { 0x00 };
2279 if (ce
->ce_flags
& CE_STRIP_NAME
) {
2280 saved_namelen
= ce_namelen(ce
);
2285 if (ce
->ce_flags
& CE_EXTENDED
)
2286 size
= offsetof(struct ondisk_cache_entry_extended
, name
);
2288 size
= offsetof(struct ondisk_cache_entry
, name
);
2290 if (!previous_name
) {
2291 int len
= ce_namelen(ce
);
2292 copy_cache_entry_to_ondisk(ondisk
, ce
);
2293 result
= ce_write(c
, fd
, ondisk
, size
);
2295 result
= ce_write(c
, fd
, ce
->name
, len
);
2297 result
= ce_write(c
, fd
, padding
, align_padding_size(size
, len
));
2299 int common
, to_remove
, prefix_size
;
2300 unsigned char to_remove_vi
[16];
2302 (ce
->name
[common
] &&
2303 common
< previous_name
->len
&&
2304 ce
->name
[common
] == previous_name
->buf
[common
]);
2306 ; /* still matching */
2307 to_remove
= previous_name
->len
- common
;
2308 prefix_size
= encode_varint(to_remove
, to_remove_vi
);
2310 copy_cache_entry_to_ondisk(ondisk
, ce
);
2311 result
= ce_write(c
, fd
, ondisk
, size
);
2313 result
= ce_write(c
, fd
, to_remove_vi
, prefix_size
);
2315 result
= ce_write(c
, fd
, ce
->name
+ common
, ce_namelen(ce
) - common
);
2317 result
= ce_write(c
, fd
, padding
, 1);
2319 strbuf_splice(previous_name
, common
, to_remove
,
2320 ce
->name
+ common
, ce_namelen(ce
) - common
);
2322 if (stripped_name
) {
2323 ce
->ce_namelen
= saved_namelen
;
2324 ce
->ce_flags
&= ~CE_STRIP_NAME
;
2331 * This function verifies if index_state has the correct sha1 of the
2332 * index file. Don't die if we have any other failure, just return 0.
2334 static int verify_index_from(const struct index_state
*istate
, const char *path
)
2339 unsigned char hash
[GIT_MAX_RAWSZ
];
2341 if (!istate
->initialized
)
2344 fd
= open(path
, O_RDONLY
);
2351 if (st
.st_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2354 n
= pread_in_full(fd
, hash
, the_hash_algo
->rawsz
, st
.st_size
- the_hash_algo
->rawsz
);
2355 if (n
!= the_hash_algo
->rawsz
)
2358 if (hashcmp(istate
->oid
.hash
, hash
))
2369 static int verify_index(const struct index_state
*istate
)
2371 return verify_index_from(istate
, get_index_file());
2374 static int has_racy_timestamp(struct index_state
*istate
)
2376 int entries
= istate
->cache_nr
;
2379 for (i
= 0; i
< entries
; i
++) {
2380 struct cache_entry
*ce
= istate
->cache
[i
];
2381 if (is_racy_timestamp(istate
, ce
))
2387 void update_index_if_able(struct index_state
*istate
, struct lock_file
*lockfile
)
2389 if ((istate
->cache_changed
|| has_racy_timestamp(istate
)) &&
2390 verify_index(istate
))
2391 write_locked_index(istate
, lockfile
, COMMIT_LOCK
);
2393 rollback_lock_file(lockfile
);
2397 * On success, `tempfile` is closed. If it is the temporary file
2398 * of a `struct lock_file`, we will therefore effectively perform
2399 * a 'close_lock_file_gently()`. Since that is an implementation
2400 * detail of lockfiles, callers of `do_write_index()` should not
2403 static int do_write_index(struct index_state
*istate
, struct tempfile
*tempfile
,
2404 int strip_extensions
)
2406 uint64_t start
= getnanotime();
2407 int newfd
= tempfile
->fd
;
2409 struct cache_header hdr
;
2410 int i
, err
= 0, removed
, extended
, hdr_version
;
2411 struct cache_entry
**cache
= istate
->cache
;
2412 int entries
= istate
->cache_nr
;
2414 struct ondisk_cache_entry_extended ondisk
;
2415 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
2416 int drop_cache_tree
= istate
->drop_cache_tree
;
2418 for (i
= removed
= extended
= 0; i
< entries
; i
++) {
2419 if (cache
[i
]->ce_flags
& CE_REMOVE
)
2422 /* reduce extended entries if possible */
2423 cache
[i
]->ce_flags
&= ~CE_EXTENDED
;
2424 if (cache
[i
]->ce_flags
& CE_EXTENDED_FLAGS
) {
2426 cache
[i
]->ce_flags
|= CE_EXTENDED
;
2430 if (!istate
->version
) {
2431 istate
->version
= get_index_format_default();
2432 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0))
2433 init_split_index(istate
);
2436 /* demote version 3 to version 2 when the latter suffices */
2437 if (istate
->version
== 3 || istate
->version
== 2)
2438 istate
->version
= extended
? 3 : 2;
2440 hdr_version
= istate
->version
;
2442 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
2443 hdr
.hdr_version
= htonl(hdr_version
);
2444 hdr
.hdr_entries
= htonl(entries
- removed
);
2446 the_hash_algo
->init_fn(&c
);
2447 if (ce_write(&c
, newfd
, &hdr
, sizeof(hdr
)) < 0)
2450 previous_name
= (hdr_version
== 4) ? &previous_name_buf
: NULL
;
2452 for (i
= 0; i
< entries
; i
++) {
2453 struct cache_entry
*ce
= cache
[i
];
2454 if (ce
->ce_flags
& CE_REMOVE
)
2456 if (!ce_uptodate(ce
) && is_racy_timestamp(istate
, ce
))
2457 ce_smudge_racily_clean_entry(ce
);
2458 if (is_null_oid(&ce
->oid
)) {
2459 static const char msg
[] = "cache entry has null sha1: %s";
2460 static int allow
= -1;
2463 allow
= git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2465 warning(msg
, ce
->name
);
2467 err
= error(msg
, ce
->name
);
2469 drop_cache_tree
= 1;
2471 if (ce_write_entry(&c
, newfd
, ce
, previous_name
, (struct ondisk_cache_entry
*)&ondisk
) < 0)
2477 strbuf_release(&previous_name_buf
);
2482 /* Write extension data here */
2483 if (!strip_extensions
&& istate
->split_index
) {
2484 struct strbuf sb
= STRBUF_INIT
;
2486 err
= write_link_extension(&sb
, istate
) < 0 ||
2487 write_index_ext_header(&c
, newfd
, CACHE_EXT_LINK
,
2489 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2490 strbuf_release(&sb
);
2494 if (!strip_extensions
&& !drop_cache_tree
&& istate
->cache_tree
) {
2495 struct strbuf sb
= STRBUF_INIT
;
2497 cache_tree_write(&sb
, istate
->cache_tree
);
2498 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_TREE
, sb
.len
) < 0
2499 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2500 strbuf_release(&sb
);
2504 if (!strip_extensions
&& istate
->resolve_undo
) {
2505 struct strbuf sb
= STRBUF_INIT
;
2507 resolve_undo_write(&sb
, istate
->resolve_undo
);
2508 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_RESOLVE_UNDO
,
2510 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2511 strbuf_release(&sb
);
2515 if (!strip_extensions
&& istate
->untracked
) {
2516 struct strbuf sb
= STRBUF_INIT
;
2518 write_untracked_extension(&sb
, istate
->untracked
);
2519 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_UNTRACKED
,
2521 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2522 strbuf_release(&sb
);
2526 if (!strip_extensions
&& istate
->fsmonitor_last_update
) {
2527 struct strbuf sb
= STRBUF_INIT
;
2529 write_fsmonitor_extension(&sb
, istate
);
2530 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_FSMONITOR
, sb
.len
) < 0
2531 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2532 strbuf_release(&sb
);
2537 if (ce_flush(&c
, newfd
, istate
->oid
.hash
))
2539 if (close_tempfile_gently(tempfile
)) {
2540 error(_("could not close '%s'"), tempfile
->filename
.buf
);
2543 if (stat(tempfile
->filename
.buf
, &st
))
2545 istate
->timestamp
.sec
= (unsigned int)st
.st_mtime
;
2546 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
2547 trace_performance_since(start
, "write index, changed mask = %x", istate
->cache_changed
);
2551 void set_alternate_index_output(const char *name
)
2553 alternate_index_output
= name
;
2556 static int commit_locked_index(struct lock_file
*lk
)
2558 if (alternate_index_output
)
2559 return commit_lock_file_to(lk
, alternate_index_output
);
2561 return commit_lock_file(lk
);
2564 static int do_write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
2567 int ret
= do_write_index(istate
, lock
->tempfile
, 0);
2570 if (flags
& COMMIT_LOCK
)
2571 return commit_locked_index(lock
);
2572 return close_lock_file_gently(lock
);
2575 static int write_split_index(struct index_state
*istate
,
2576 struct lock_file
*lock
,
2580 prepare_to_write_split_index(istate
);
2581 ret
= do_write_locked_index(istate
, lock
, flags
);
2582 finish_writing_split_index(istate
);
2586 static const char *shared_index_expire
= "2.weeks.ago";
2588 static unsigned long get_shared_index_expire_date(void)
2590 static unsigned long shared_index_expire_date
;
2591 static int shared_index_expire_date_prepared
;
2593 if (!shared_index_expire_date_prepared
) {
2594 git_config_get_expiry("splitindex.sharedindexexpire",
2595 &shared_index_expire
);
2596 shared_index_expire_date
= approxidate(shared_index_expire
);
2597 shared_index_expire_date_prepared
= 1;
2600 return shared_index_expire_date
;
2603 static int should_delete_shared_index(const char *shared_index_path
)
2606 unsigned long expiration
;
2608 /* Check timestamp */
2609 expiration
= get_shared_index_expire_date();
2612 if (stat(shared_index_path
, &st
))
2613 return error_errno(_("could not stat '%s'"), shared_index_path
);
2614 if (st
.st_mtime
> expiration
)
2620 static int clean_shared_index_files(const char *current_hex
)
2623 DIR *dir
= opendir(get_git_dir());
2626 return error_errno(_("unable to open git dir: %s"), get_git_dir());
2628 while ((de
= readdir(dir
)) != NULL
) {
2629 const char *sha1_hex
;
2630 const char *shared_index_path
;
2631 if (!skip_prefix(de
->d_name
, "sharedindex.", &sha1_hex
))
2633 if (!strcmp(sha1_hex
, current_hex
))
2635 shared_index_path
= git_path("%s", de
->d_name
);
2636 if (should_delete_shared_index(shared_index_path
) > 0 &&
2637 unlink(shared_index_path
))
2638 warning_errno(_("unable to unlink: %s"), shared_index_path
);
2645 static int write_shared_index(struct index_state
*istate
,
2646 struct tempfile
**temp
)
2648 struct split_index
*si
= istate
->split_index
;
2651 move_cache_to_base_index(istate
);
2652 ret
= do_write_index(si
->base
, *temp
, 1);
2655 ret
= adjust_shared_perm(get_tempfile_path(*temp
));
2657 error("cannot fix permission bits on %s", get_tempfile_path(*temp
));
2660 ret
= rename_tempfile(temp
,
2661 git_path("sharedindex.%s", oid_to_hex(&si
->base
->oid
)));
2663 oidcpy(&si
->base_oid
, &si
->base
->oid
);
2664 clean_shared_index_files(oid_to_hex(&si
->base
->oid
));
2670 static const int default_max_percent_split_change
= 20;
2672 static int too_many_not_shared_entries(struct index_state
*istate
)
2674 int i
, not_shared
= 0;
2675 int max_split
= git_config_get_max_percent_split_change();
2677 switch (max_split
) {
2679 /* not or badly configured: use the default value */
2680 max_split
= default_max_percent_split_change
;
2683 return 1; /* 0% means always write a new shared index */
2685 return 0; /* 100% means never write a new shared index */
2687 break; /* just use the configured value */
2690 /* Count not shared entries */
2691 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2692 struct cache_entry
*ce
= istate
->cache
[i
];
2697 return (int64_t)istate
->cache_nr
* max_split
< (int64_t)not_shared
* 100;
2700 int write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
2703 int new_shared_index
, ret
;
2704 struct split_index
*si
= istate
->split_index
;
2706 if ((flags
& SKIP_IF_UNCHANGED
) && !istate
->cache_changed
) {
2707 if (flags
& COMMIT_LOCK
)
2708 rollback_lock_file(lock
);
2712 if (istate
->fsmonitor_last_update
)
2713 fill_fsmonitor_bitmap(istate
);
2715 if (!si
|| alternate_index_output
||
2716 (istate
->cache_changed
& ~EXTMASK
)) {
2718 oidclr(&si
->base_oid
);
2719 ret
= do_write_locked_index(istate
, lock
, flags
);
2723 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0)) {
2724 int v
= si
->base_oid
.hash
[0];
2726 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
2728 if (too_many_not_shared_entries(istate
))
2729 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
2731 new_shared_index
= istate
->cache_changed
& SPLIT_INDEX_ORDERED
;
2733 if (new_shared_index
) {
2734 struct tempfile
*temp
;
2737 temp
= mks_tempfile(git_path("sharedindex_XXXXXX"));
2739 oidclr(&si
->base_oid
);
2740 ret
= do_write_locked_index(istate
, lock
, flags
);
2743 ret
= write_shared_index(istate
, &temp
);
2745 saved_errno
= errno
;
2746 if (is_tempfile_active(temp
))
2747 delete_tempfile(&temp
);
2748 errno
= saved_errno
;
2754 ret
= write_split_index(istate
, lock
, flags
);
2756 /* Freshen the shared index only if the split-index was written */
2757 if (!ret
&& !new_shared_index
) {
2758 const char *shared_index
= git_path("sharedindex.%s",
2759 oid_to_hex(&si
->base_oid
));
2760 freshen_shared_index(shared_index
, 1);
2764 if (flags
& COMMIT_LOCK
)
2765 rollback_lock_file(lock
);
2770 * Read the index file that is potentially unmerged into given
2771 * index_state, dropping any unmerged entries. Returns true if
2772 * the index is unmerged. Callers who want to refuse to work
2773 * from an unmerged state can call this and check its return value,
2774 * instead of calling read_cache().
2776 int read_index_unmerged(struct index_state
*istate
)
2782 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2783 struct cache_entry
*ce
= istate
->cache
[i
];
2784 struct cache_entry
*new_ce
;
2790 len
= ce_namelen(ce
);
2791 new_ce
= make_empty_cache_entry(istate
, len
);
2792 memcpy(new_ce
->name
, ce
->name
, len
);
2793 new_ce
->ce_flags
= create_ce_flags(0) | CE_CONFLICTED
;
2794 new_ce
->ce_namelen
= len
;
2795 new_ce
->ce_mode
= ce
->ce_mode
;
2796 if (add_index_entry(istate
, new_ce
, 0))
2797 return error("%s: cannot drop to stage #0",
2804 * Returns 1 if the path is an "other" path with respect to
2805 * the index; that is, the path is not mentioned in the index at all,
2806 * either as a file, a directory with some files in the index,
2807 * or as an unmerged entry.
2809 * We helpfully remove a trailing "/" from directories so that
2810 * the output of read_directory can be used as-is.
2812 int index_name_is_other(const struct index_state
*istate
, const char *name
,
2816 if (namelen
&& name
[namelen
- 1] == '/')
2818 pos
= index_name_pos(istate
, name
, namelen
);
2820 return 0; /* exact match */
2822 if (pos
< istate
->cache_nr
) {
2823 struct cache_entry
*ce
= istate
->cache
[pos
];
2824 if (ce_namelen(ce
) == namelen
&&
2825 !memcmp(ce
->name
, name
, namelen
))
2826 return 0; /* Yup, this one exists unmerged */
2831 void *read_blob_data_from_index(const struct index_state
*istate
,
2832 const char *path
, unsigned long *size
)
2836 enum object_type type
;
2840 pos
= index_name_pos(istate
, path
, len
);
2843 * We might be in the middle of a merge, in which
2844 * case we would read stage #2 (ours).
2848 (pos
< 0 && i
< istate
->cache_nr
&&
2849 !strcmp(istate
->cache
[i
]->name
, path
));
2851 if (ce_stage(istate
->cache
[i
]) == 2)
2856 data
= read_object_file(&istate
->cache
[pos
]->oid
, &type
, &sz
);
2857 if (!data
|| type
!= OBJ_BLOB
) {
2866 void stat_validity_clear(struct stat_validity
*sv
)
2868 FREE_AND_NULL(sv
->sd
);
2871 int stat_validity_check(struct stat_validity
*sv
, const char *path
)
2875 if (stat(path
, &st
) < 0)
2876 return sv
->sd
== NULL
;
2879 return S_ISREG(st
.st_mode
) && !match_stat_data(sv
->sd
, &st
);
2882 void stat_validity_update(struct stat_validity
*sv
, int fd
)
2886 if (fstat(fd
, &st
) < 0 || !S_ISREG(st
.st_mode
))
2887 stat_validity_clear(sv
);
2890 sv
->sd
= xcalloc(1, sizeof(struct stat_data
));
2891 fill_stat_data(sv
->sd
, &st
);
2895 void move_index_extensions(struct index_state
*dst
, struct index_state
*src
)
2897 dst
->untracked
= src
->untracked
;
2898 src
->untracked
= NULL
;
2901 struct cache_entry
*dup_cache_entry(const struct cache_entry
*ce
,
2902 struct index_state
*istate
)
2904 unsigned int size
= ce_size(ce
);
2905 int mem_pool_allocated
;
2906 struct cache_entry
*new_entry
= make_empty_cache_entry(istate
, ce_namelen(ce
));
2907 mem_pool_allocated
= new_entry
->mem_pool_allocated
;
2909 memcpy(new_entry
, ce
, size
);
2910 new_entry
->mem_pool_allocated
= mem_pool_allocated
;
2914 void discard_cache_entry(struct cache_entry
*ce
)
2916 if (ce
&& should_validate_cache_entries())
2917 memset(ce
, 0xCD, cache_entry_size(ce
->ce_namelen
));
2919 if (ce
&& ce
->mem_pool_allocated
)
2925 int should_validate_cache_entries(void)
2927 static int validate_index_cache_entries
= -1;
2929 if (validate_index_cache_entries
< 0) {
2930 if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))
2931 validate_index_cache_entries
= 1;
2933 validate_index_cache_entries
= 0;
2936 return validate_index_cache_entries
;