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"
17 #include "resolve-undo.h"
20 #include "split-index.h"
22 #include "fsmonitor.h"
24 /* Mask for the name length in ce_flags in the on-disk index */
26 #define CE_NAMEMASK (0x0fff)
30 * The first letter should be 'A'..'Z' for extensions that are not
31 * necessary for a correct operation (i.e. optimization data).
32 * When new extensions are added that _needs_ to be understood in
33 * order to correctly interpret the index file, pick character that
34 * is outside the range, to cause the reader to abort.
37 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
38 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
39 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
40 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
41 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
42 #define CACHE_EXT_FSMONITOR 0x46534D4E /* "FSMN" */
44 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
45 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
46 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
47 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED | FSMONITOR_CHANGED)
51 * This is an estimate of the pathname length in the index. We use
52 * this for V4 index files to guess the un-deltafied size of the index
53 * in memory because of pathname deltafication. This is not required
54 * for V2/V3 index formats because their pathnames are not compressed.
55 * If the initial amount of memory set aside is not sufficient, the
56 * mem pool will allocate extra memory.
58 #define CACHE_ENTRY_PATH_LENGTH 80
60 static inline struct cache_entry
*mem_pool__ce_alloc(struct mem_pool
*mem_pool
, size_t len
)
62 struct cache_entry
*ce
;
63 ce
= mem_pool_alloc(mem_pool
, cache_entry_size(len
));
64 ce
->mem_pool_allocated
= 1;
68 static inline struct cache_entry
*mem_pool__ce_calloc(struct mem_pool
*mem_pool
, size_t len
)
70 struct cache_entry
* ce
;
71 ce
= mem_pool_calloc(mem_pool
, 1, cache_entry_size(len
));
72 ce
->mem_pool_allocated
= 1;
76 static struct mem_pool
*find_mem_pool(struct index_state
*istate
)
78 struct mem_pool
**pool_ptr
;
80 if (istate
->split_index
&& istate
->split_index
->base
)
81 pool_ptr
= &istate
->split_index
->base
->ce_mem_pool
;
83 pool_ptr
= &istate
->ce_mem_pool
;
86 mem_pool_init(pool_ptr
, 0);
91 struct index_state the_index
;
92 static const char *alternate_index_output
;
94 static void set_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
96 istate
->cache
[nr
] = ce
;
97 add_name_hash(istate
, ce
);
100 static void replace_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
102 struct cache_entry
*old
= istate
->cache
[nr
];
104 replace_index_entry_in_base(istate
, old
, ce
);
105 remove_name_hash(istate
, old
);
106 discard_cache_entry(old
);
107 ce
->ce_flags
&= ~CE_HASHED
;
108 set_index_entry(istate
, nr
, ce
);
109 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
110 mark_fsmonitor_invalid(istate
, ce
);
111 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
114 void rename_index_entry_at(struct index_state
*istate
, int nr
, const char *new_name
)
116 struct cache_entry
*old_entry
= istate
->cache
[nr
], *new_entry
;
117 int namelen
= strlen(new_name
);
119 new_entry
= make_empty_cache_entry(istate
, namelen
);
120 copy_cache_entry(new_entry
, old_entry
);
121 new_entry
->ce_flags
&= ~CE_HASHED
;
122 new_entry
->ce_namelen
= namelen
;
123 new_entry
->index
= 0;
124 memcpy(new_entry
->name
, new_name
, namelen
+ 1);
126 cache_tree_invalidate_path(istate
, old_entry
->name
);
127 untracked_cache_remove_from_index(istate
, old_entry
->name
);
128 remove_index_entry_at(istate
, nr
);
129 add_index_entry(istate
, new_entry
, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
132 void fill_stat_data(struct stat_data
*sd
, struct stat
*st
)
134 sd
->sd_ctime
.sec
= (unsigned int)st
->st_ctime
;
135 sd
->sd_mtime
.sec
= (unsigned int)st
->st_mtime
;
136 sd
->sd_ctime
.nsec
= ST_CTIME_NSEC(*st
);
137 sd
->sd_mtime
.nsec
= ST_MTIME_NSEC(*st
);
138 sd
->sd_dev
= st
->st_dev
;
139 sd
->sd_ino
= st
->st_ino
;
140 sd
->sd_uid
= st
->st_uid
;
141 sd
->sd_gid
= st
->st_gid
;
142 sd
->sd_size
= st
->st_size
;
145 int match_stat_data(const struct stat_data
*sd
, struct stat
*st
)
149 if (sd
->sd_mtime
.sec
!= (unsigned int)st
->st_mtime
)
150 changed
|= MTIME_CHANGED
;
151 if (trust_ctime
&& check_stat
&&
152 sd
->sd_ctime
.sec
!= (unsigned int)st
->st_ctime
)
153 changed
|= CTIME_CHANGED
;
156 if (check_stat
&& sd
->sd_mtime
.nsec
!= ST_MTIME_NSEC(*st
))
157 changed
|= MTIME_CHANGED
;
158 if (trust_ctime
&& check_stat
&&
159 sd
->sd_ctime
.nsec
!= ST_CTIME_NSEC(*st
))
160 changed
|= CTIME_CHANGED
;
164 if (sd
->sd_uid
!= (unsigned int) st
->st_uid
||
165 sd
->sd_gid
!= (unsigned int) st
->st_gid
)
166 changed
|= OWNER_CHANGED
;
167 if (sd
->sd_ino
!= (unsigned int) st
->st_ino
)
168 changed
|= INODE_CHANGED
;
173 * st_dev breaks on network filesystems where different
174 * clients will have different views of what "device"
175 * the filesystem is on
177 if (check_stat
&& sd
->sd_dev
!= (unsigned int) st
->st_dev
)
178 changed
|= INODE_CHANGED
;
181 if (sd
->sd_size
!= (unsigned int) st
->st_size
)
182 changed
|= DATA_CHANGED
;
188 * This only updates the "non-critical" parts of the directory
189 * cache, ie the parts that aren't tracked by GIT, and only used
190 * to validate the cache.
192 void fill_stat_cache_info(struct cache_entry
*ce
, struct stat
*st
)
194 fill_stat_data(&ce
->ce_stat_data
, st
);
196 if (assume_unchanged
)
197 ce
->ce_flags
|= CE_VALID
;
199 if (S_ISREG(st
->st_mode
)) {
200 ce_mark_uptodate(ce
);
201 mark_fsmonitor_valid(ce
);
205 static int ce_compare_data(const struct cache_entry
*ce
, struct stat
*st
)
208 int fd
= git_open_cloexec(ce
->name
, O_RDONLY
);
211 struct object_id oid
;
212 if (!index_fd(&oid
, fd
, st
, OBJ_BLOB
, ce
->name
, 0))
213 match
= oidcmp(&oid
, &ce
->oid
);
214 /* index_fd() closed the file descriptor already */
219 static int ce_compare_link(const struct cache_entry
*ce
, size_t expected_size
)
224 enum object_type type
;
225 struct strbuf sb
= STRBUF_INIT
;
227 if (strbuf_readlink(&sb
, ce
->name
, expected_size
))
230 buffer
= read_object_file(&ce
->oid
, &type
, &size
);
233 match
= memcmp(buffer
, sb
.buf
, size
);
240 static int ce_compare_gitlink(const struct cache_entry
*ce
)
242 struct object_id oid
;
245 * We don't actually require that the .git directory
246 * under GITLINK directory be a valid git directory. It
247 * might even be missing (in case nobody populated that
250 * If so, we consider it always to match.
252 if (resolve_gitlink_ref(ce
->name
, "HEAD", &oid
) < 0)
254 return oidcmp(&oid
, &ce
->oid
);
257 static int ce_modified_check_fs(const struct cache_entry
*ce
, struct stat
*st
)
259 switch (st
->st_mode
& S_IFMT
) {
261 if (ce_compare_data(ce
, st
))
265 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
269 if (S_ISGITLINK(ce
->ce_mode
))
270 return ce_compare_gitlink(ce
) ? DATA_CHANGED
: 0;
271 /* else fallthrough */
278 static int ce_match_stat_basic(const struct cache_entry
*ce
, struct stat
*st
)
280 unsigned int changed
= 0;
282 if (ce
->ce_flags
& CE_REMOVE
)
283 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
285 switch (ce
->ce_mode
& S_IFMT
) {
287 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
288 /* We consider only the owner x bit to be relevant for
291 if (trust_executable_bit
&&
292 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
293 changed
|= MODE_CHANGED
;
296 if (!S_ISLNK(st
->st_mode
) &&
297 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
298 changed
|= TYPE_CHANGED
;
301 /* We ignore most of the st_xxx fields for gitlinks */
302 if (!S_ISDIR(st
->st_mode
))
303 changed
|= TYPE_CHANGED
;
304 else if (ce_compare_gitlink(ce
))
305 changed
|= DATA_CHANGED
;
308 die("internal error: ce_mode is %o", ce
->ce_mode
);
311 changed
|= match_stat_data(&ce
->ce_stat_data
, st
);
313 /* Racily smudged entry? */
314 if (!ce
->ce_stat_data
.sd_size
) {
315 if (!is_empty_blob_sha1(ce
->oid
.hash
))
316 changed
|= DATA_CHANGED
;
322 static int is_racy_stat(const struct index_state
*istate
,
323 const struct stat_data
*sd
)
325 return (istate
->timestamp
.sec
&&
327 /* nanosecond timestamped files can also be racy! */
328 (istate
->timestamp
.sec
< sd
->sd_mtime
.sec
||
329 (istate
->timestamp
.sec
== sd
->sd_mtime
.sec
&&
330 istate
->timestamp
.nsec
<= sd
->sd_mtime
.nsec
))
332 istate
->timestamp
.sec
<= sd
->sd_mtime
.sec
337 static int is_racy_timestamp(const struct index_state
*istate
,
338 const struct cache_entry
*ce
)
340 return (!S_ISGITLINK(ce
->ce_mode
) &&
341 is_racy_stat(istate
, &ce
->ce_stat_data
));
344 int match_stat_data_racy(const struct index_state
*istate
,
345 const struct stat_data
*sd
, struct stat
*st
)
347 if (is_racy_stat(istate
, sd
))
348 return MTIME_CHANGED
;
349 return match_stat_data(sd
, st
);
352 int ie_match_stat(struct index_state
*istate
,
353 const struct cache_entry
*ce
, struct stat
*st
,
354 unsigned int options
)
356 unsigned int changed
;
357 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
358 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
359 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
360 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
362 if (!ignore_fsmonitor
)
363 refresh_fsmonitor(istate
);
365 * If it's marked as always valid in the index, it's
366 * valid whatever the checked-out copy says.
368 * skip-worktree has the same effect with higher precedence
370 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
))
372 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
374 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
))
378 * Intent-to-add entries have not been added, so the index entry
379 * by definition never matches what is in the work tree until it
380 * actually gets added.
382 if (ce_intent_to_add(ce
))
383 return DATA_CHANGED
| TYPE_CHANGED
| MODE_CHANGED
;
385 changed
= ce_match_stat_basic(ce
, st
);
388 * Within 1 second of this sequence:
389 * echo xyzzy >file && git-update-index --add file
390 * running this command:
392 * would give a falsely clean cache entry. The mtime and
393 * length match the cache, and other stat fields do not change.
395 * We could detect this at update-index time (the cache entry
396 * being registered/updated records the same time as "now")
397 * and delay the return from git-update-index, but that would
398 * effectively mean we can make at most one commit per second,
399 * which is not acceptable. Instead, we check cache entries
400 * whose mtime are the same as the index file timestamp more
401 * carefully than others.
403 if (!changed
&& is_racy_timestamp(istate
, ce
)) {
404 if (assume_racy_is_modified
)
405 changed
|= DATA_CHANGED
;
407 changed
|= ce_modified_check_fs(ce
, st
);
413 int ie_modified(struct index_state
*istate
,
414 const struct cache_entry
*ce
,
415 struct stat
*st
, unsigned int options
)
417 int changed
, changed_fs
;
419 changed
= ie_match_stat(istate
, ce
, st
, options
);
423 * If the mode or type has changed, there's no point in trying
424 * to refresh the entry - it's not going to match
426 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
430 * Immediately after read-tree or update-index --cacheinfo,
431 * the length field is zero, as we have never even read the
432 * lstat(2) information once, and we cannot trust DATA_CHANGED
433 * returned by ie_match_stat() which in turn was returned by
434 * ce_match_stat_basic() to signal that the filesize of the
435 * blob changed. We have to actually go to the filesystem to
436 * see if the contents match, and if so, should answer "unchanged".
438 * The logic does not apply to gitlinks, as ce_match_stat_basic()
439 * already has checked the actual HEAD from the filesystem in the
440 * subproject. If ie_match_stat() already said it is different,
441 * then we know it is.
443 if ((changed
& DATA_CHANGED
) &&
444 (S_ISGITLINK(ce
->ce_mode
) || ce
->ce_stat_data
.sd_size
!= 0))
447 changed_fs
= ce_modified_check_fs(ce
, st
);
449 return changed
| changed_fs
;
453 int base_name_compare(const char *name1
, int len1
, int mode1
,
454 const char *name2
, int len2
, int mode2
)
456 unsigned char c1
, c2
;
457 int len
= len1
< len2
? len1
: len2
;
460 cmp
= memcmp(name1
, name2
, len
);
465 if (!c1
&& S_ISDIR(mode1
))
467 if (!c2
&& S_ISDIR(mode2
))
469 return (c1
< c2
) ? -1 : (c1
> c2
) ? 1 : 0;
473 * df_name_compare() is identical to base_name_compare(), except it
474 * compares conflicting directory/file entries as equal. Note that
475 * while a directory name compares as equal to a regular file, they
476 * then individually compare _differently_ to a filename that has
477 * a dot after the basename (because '\0' < '.' < '/').
479 * This is used by routines that want to traverse the git namespace
480 * but then handle conflicting entries together when possible.
482 int df_name_compare(const char *name1
, int len1
, int mode1
,
483 const char *name2
, int len2
, int mode2
)
485 int len
= len1
< len2
? len1
: len2
, cmp
;
486 unsigned char c1
, c2
;
488 cmp
= memcmp(name1
, name2
, len
);
491 /* Directories and files compare equal (same length, same name) */
495 if (!c1
&& S_ISDIR(mode1
))
498 if (!c2
&& S_ISDIR(mode2
))
500 if (c1
== '/' && !c2
)
502 if (c2
== '/' && !c1
)
507 int name_compare(const char *name1
, size_t len1
, const char *name2
, size_t len2
)
509 size_t min_len
= (len1
< len2
) ? len1
: len2
;
510 int cmp
= memcmp(name1
, name2
, min_len
);
520 int cache_name_stage_compare(const char *name1
, int len1
, int stage1
, const char *name2
, int len2
, int stage2
)
524 cmp
= name_compare(name1
, len1
, name2
, len2
);
535 static int index_name_stage_pos(const struct index_state
*istate
, const char *name
, int namelen
, int stage
)
540 last
= istate
->cache_nr
;
541 while (last
> first
) {
542 int next
= (last
+ first
) >> 1;
543 struct cache_entry
*ce
= istate
->cache
[next
];
544 int cmp
= cache_name_stage_compare(name
, namelen
, stage
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
556 int index_name_pos(const struct index_state
*istate
, const char *name
, int namelen
)
558 return index_name_stage_pos(istate
, name
, namelen
, 0);
561 int remove_index_entry_at(struct index_state
*istate
, int pos
)
563 struct cache_entry
*ce
= istate
->cache
[pos
];
565 record_resolve_undo(istate
, ce
);
566 remove_name_hash(istate
, ce
);
567 save_or_free_index_entry(istate
, ce
);
568 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
570 if (pos
>= istate
->cache_nr
)
572 MOVE_ARRAY(istate
->cache
+ pos
, istate
->cache
+ pos
+ 1,
573 istate
->cache_nr
- pos
);
578 * Remove all cache entries marked for removal, that is where
579 * CE_REMOVE is set in ce_flags. This is much more effective than
580 * calling remove_index_entry_at() for each entry to be removed.
582 void remove_marked_cache_entries(struct index_state
*istate
)
584 struct cache_entry
**ce_array
= istate
->cache
;
587 for (i
= j
= 0; i
< istate
->cache_nr
; i
++) {
588 if (ce_array
[i
]->ce_flags
& CE_REMOVE
) {
589 remove_name_hash(istate
, ce_array
[i
]);
590 save_or_free_index_entry(istate
, ce_array
[i
]);
593 ce_array
[j
++] = ce_array
[i
];
595 if (j
== istate
->cache_nr
)
597 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
598 istate
->cache_nr
= j
;
601 int remove_file_from_index(struct index_state
*istate
, const char *path
)
603 int pos
= index_name_pos(istate
, path
, strlen(path
));
606 cache_tree_invalidate_path(istate
, path
);
607 untracked_cache_remove_from_index(istate
, path
);
608 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
609 remove_index_entry_at(istate
, pos
);
613 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
615 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
618 static int index_name_pos_also_unmerged(struct index_state
*istate
,
619 const char *path
, int namelen
)
621 int pos
= index_name_pos(istate
, path
, namelen
);
622 struct cache_entry
*ce
;
627 /* maybe unmerged? */
629 if (pos
>= istate
->cache_nr
||
630 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
633 /* order of preference: stage 2, 1, 3 */
634 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
635 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
636 !compare_name(ce
, path
, namelen
))
641 static int different_name(struct cache_entry
*ce
, struct cache_entry
*alias
)
643 int len
= ce_namelen(ce
);
644 return ce_namelen(alias
) != len
|| memcmp(ce
->name
, alias
->name
, len
);
648 * If we add a filename that aliases in the cache, we will use the
649 * name that we already have - but we don't want to update the same
650 * alias twice, because that implies that there were actually two
651 * different files with aliasing names!
653 * So we use the CE_ADDED flag to verify that the alias was an old
654 * one before we accept it as
656 static struct cache_entry
*create_alias_ce(struct index_state
*istate
,
657 struct cache_entry
*ce
,
658 struct cache_entry
*alias
)
661 struct cache_entry
*new_entry
;
663 if (alias
->ce_flags
& CE_ADDED
)
664 die("Will not add file alias '%s' ('%s' already exists in index)", ce
->name
, alias
->name
);
666 /* Ok, create the new entry using the name of the existing alias */
667 len
= ce_namelen(alias
);
668 new_entry
= make_empty_cache_entry(istate
, len
);
669 memcpy(new_entry
->name
, alias
->name
, len
);
670 copy_cache_entry(new_entry
, ce
);
671 save_or_free_index_entry(istate
, ce
);
675 void set_object_name_for_intent_to_add_entry(struct cache_entry
*ce
)
677 struct object_id oid
;
678 if (write_object_file("", 0, blob_type
, &oid
))
679 die("cannot create an empty blob in the object database");
680 oidcpy(&ce
->oid
, &oid
);
683 int add_to_index(struct index_state
*istate
, const char *path
, struct stat
*st
, int flags
)
685 int namelen
, was_same
;
686 mode_t st_mode
= st
->st_mode
;
687 struct cache_entry
*ce
, *alias
= NULL
;
688 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_IGNORE_SKIP_WORKTREE
|CE_MATCH_RACY_IS_DIRTY
;
689 int verbose
= flags
& (ADD_CACHE_VERBOSE
| ADD_CACHE_PRETEND
);
690 int pretend
= flags
& ADD_CACHE_PRETEND
;
691 int intent_only
= flags
& ADD_CACHE_INTENT
;
692 int add_option
= (ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
|
693 (intent_only
? ADD_CACHE_NEW_ONLY
: 0));
694 int newflags
= HASH_WRITE_OBJECT
;
696 if (flags
& HASH_RENORMALIZE
)
697 newflags
|= HASH_RENORMALIZE
;
699 if (!S_ISREG(st_mode
) && !S_ISLNK(st_mode
) && !S_ISDIR(st_mode
))
700 return error("%s: can only add regular files, symbolic links or git-directories", path
);
702 namelen
= strlen(path
);
703 if (S_ISDIR(st_mode
)) {
704 while (namelen
&& path
[namelen
-1] == '/')
707 ce
= make_empty_cache_entry(istate
, namelen
);
708 memcpy(ce
->name
, path
, namelen
);
709 ce
->ce_namelen
= namelen
;
711 fill_stat_cache_info(ce
, st
);
713 ce
->ce_flags
|= CE_INTENT_TO_ADD
;
716 if (trust_executable_bit
&& has_symlinks
) {
717 ce
->ce_mode
= create_ce_mode(st_mode
);
719 /* If there is an existing entry, pick the mode bits and type
720 * from it, otherwise assume unexecutable regular file.
722 struct cache_entry
*ent
;
723 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
725 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
726 ce
->ce_mode
= ce_mode_from_stat(ent
, st_mode
);
729 /* When core.ignorecase=true, determine if a directory of the same name but differing
730 * case already exists within the Git repository. If it does, ensure the directory
731 * case of the file being added to the repository matches (is folded into) the existing
732 * entry's directory case.
735 adjust_dirname_case(istate
, ce
->name
);
737 if (!(flags
& HASH_RENORMALIZE
)) {
738 alias
= index_file_exists(istate
, ce
->name
,
739 ce_namelen(ce
), ignore_case
);
742 !ie_match_stat(istate
, alias
, st
, ce_option
)) {
743 /* Nothing changed, really */
744 if (!S_ISGITLINK(alias
->ce_mode
))
745 ce_mark_uptodate(alias
);
746 alias
->ce_flags
|= CE_ADDED
;
748 discard_cache_entry(ce
);
753 if (index_path(&ce
->oid
, path
, st
, newflags
)) {
754 discard_cache_entry(ce
);
755 return error("unable to index file %s", path
);
758 set_object_name_for_intent_to_add_entry(ce
);
760 if (ignore_case
&& alias
&& different_name(ce
, alias
))
761 ce
= create_alias_ce(istate
, ce
, alias
);
762 ce
->ce_flags
|= CE_ADDED
;
764 /* It was suspected to be racily clean, but it turns out to be Ok */
767 !oidcmp(&alias
->oid
, &ce
->oid
) &&
768 ce
->ce_mode
== alias
->ce_mode
);
771 discard_cache_entry(ce
);
772 else if (add_index_entry(istate
, ce
, add_option
)) {
773 discard_cache_entry(ce
);
774 return error("unable to add %s to index", path
);
776 if (verbose
&& !was_same
)
777 printf("add '%s'\n", path
);
781 int add_file_to_index(struct index_state
*istate
, const char *path
, int flags
)
784 if (lstat(path
, &st
))
785 die_errno("unable to stat '%s'", path
);
786 return add_to_index(istate
, path
, &st
, flags
);
789 struct cache_entry
*make_empty_cache_entry(struct index_state
*istate
, size_t len
)
791 return mem_pool__ce_calloc(find_mem_pool(istate
), len
);
794 struct cache_entry
*make_empty_transient_cache_entry(size_t len
)
796 return xcalloc(1, cache_entry_size(len
));
799 struct cache_entry
*make_cache_entry(struct index_state
*istate
,
801 const struct object_id
*oid
,
804 unsigned int refresh_options
)
806 struct cache_entry
*ce
, *ret
;
809 if (!verify_path(path
, mode
)) {
810 error("Invalid path '%s'", path
);
815 ce
= make_empty_cache_entry(istate
, len
);
817 oidcpy(&ce
->oid
, oid
);
818 memcpy(ce
->name
, path
, len
);
819 ce
->ce_flags
= create_ce_flags(stage
);
820 ce
->ce_namelen
= len
;
821 ce
->ce_mode
= create_ce_mode(mode
);
823 ret
= refresh_cache_entry(&the_index
, ce
, refresh_options
);
825 discard_cache_entry(ce
);
829 struct cache_entry
*make_transient_cache_entry(unsigned int mode
, const struct object_id
*oid
,
830 const char *path
, int stage
)
832 struct cache_entry
*ce
;
835 if (!verify_path(path
, mode
)) {
836 error("Invalid path '%s'", path
);
841 ce
= make_empty_transient_cache_entry(len
);
843 oidcpy(&ce
->oid
, oid
);
844 memcpy(ce
->name
, path
, len
);
845 ce
->ce_flags
= create_ce_flags(stage
);
846 ce
->ce_namelen
= len
;
847 ce
->ce_mode
= create_ce_mode(mode
);
853 * Chmod an index entry with either +x or -x.
855 * Returns -1 if the chmod for the particular cache entry failed (if it's
856 * not a regular file), -2 if an invalid flip argument is passed in, 0
859 int chmod_index_entry(struct index_state
*istate
, struct cache_entry
*ce
,
862 if (!S_ISREG(ce
->ce_mode
))
869 ce
->ce_mode
&= ~0111;
874 cache_tree_invalidate_path(istate
, ce
->name
);
875 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
876 mark_fsmonitor_invalid(istate
, ce
);
877 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
882 int ce_same_name(const struct cache_entry
*a
, const struct cache_entry
*b
)
884 int len
= ce_namelen(a
);
885 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
889 * We fundamentally don't like some paths: we don't want
890 * dot or dot-dot anywhere, and for obvious reasons don't
891 * want to recurse into ".git" either.
893 * Also, we don't want double slashes or slashes at the
894 * end that can make pathnames ambiguous.
896 static int verify_dotfile(const char *rest
, unsigned mode
)
899 * The first character was '.', but that
900 * has already been discarded, we now test
904 /* "." is not allowed */
905 if (*rest
== '\0' || is_dir_sep(*rest
))
910 * ".git" followed by NUL or slash is bad. Note that we match
911 * case-insensitively here, even if ignore_case is not set.
912 * This outlaws ".GIT" everywhere out of an abundance of caution,
913 * since there's really no good reason to allow it.
915 * Once we've seen ".git", we can also find ".gitmodules", etc (also
916 * case-insensitively).
920 if (rest
[1] != 'i' && rest
[1] != 'I')
922 if (rest
[2] != 't' && rest
[2] != 'T')
924 if (rest
[3] == '\0' || is_dir_sep(rest
[3]))
928 if (skip_iprefix(rest
, "modules", &rest
) &&
929 (*rest
== '\0' || is_dir_sep(*rest
)))
934 if (rest
[1] == '\0' || is_dir_sep(rest
[1]))
940 int verify_path(const char *path
, unsigned mode
)
944 if (has_dos_drive_prefix(path
))
954 if (is_hfs_dotgit(path
))
957 if (is_hfs_dotgitmodules(path
))
962 if (is_ntfs_dotgit(path
))
965 if (is_ntfs_dotgitmodules(path
))
971 if ((c
== '.' && !verify_dotfile(path
, mode
)) ||
972 is_dir_sep(c
) || c
== '\0')
980 * Do we have another file that has the beginning components being a
981 * proper superset of the name we're trying to add?
983 static int has_file_name(struct index_state
*istate
,
984 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
987 int len
= ce_namelen(ce
);
988 int stage
= ce_stage(ce
);
989 const char *name
= ce
->name
;
991 while (pos
< istate
->cache_nr
) {
992 struct cache_entry
*p
= istate
->cache
[pos
++];
994 if (len
>= ce_namelen(p
))
996 if (memcmp(name
, p
->name
, len
))
998 if (ce_stage(p
) != stage
)
1000 if (p
->name
[len
] != '/')
1002 if (p
->ce_flags
& CE_REMOVE
)
1007 remove_index_entry_at(istate
, --pos
);
1014 * Like strcmp(), but also return the offset of the first change.
1015 * If strings are equal, return the length.
1017 int strcmp_offset(const char *s1
, const char *s2
, size_t *first_change
)
1022 return strcmp(s1
, s2
);
1024 for (k
= 0; s1
[k
] == s2
[k
]; k
++)
1029 return (unsigned char)s1
[k
] - (unsigned char)s2
[k
];
1033 * Do we have another file with a pathname that is a proper
1034 * subset of the name we're trying to add?
1036 * That is, is there another file in the index with a path
1037 * that matches a sub-directory in the given entry?
1039 static int has_dir_name(struct index_state
*istate
,
1040 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
1043 int stage
= ce_stage(ce
);
1044 const char *name
= ce
->name
;
1045 const char *slash
= name
+ ce_namelen(ce
);
1050 * We are frequently called during an iteration on a sorted
1051 * list of pathnames and while building a new index. Therefore,
1052 * there is a high probability that this entry will eventually
1053 * be appended to the index, rather than inserted in the middle.
1054 * If we can confirm that, we can avoid binary searches on the
1055 * components of the pathname.
1057 * Compare the entry's full path with the last path in the index.
1059 if (istate
->cache_nr
> 0) {
1060 cmp_last
= strcmp_offset(name
,
1061 istate
->cache
[istate
->cache_nr
- 1]->name
,
1064 if (len_eq_last
== 0) {
1066 * The entry sorts AFTER the last one in the
1067 * index and their paths have no common prefix,
1068 * so there cannot be a F/D conflict.
1073 * The entry sorts AFTER the last one in the
1074 * index, but has a common prefix. Fall through
1075 * to the loop below to disect the entry's path
1076 * and see where the difference is.
1079 } else if (cmp_last
== 0) {
1081 * The entry exactly matches the last one in the
1082 * index, but because of multiple stage and CE_REMOVE
1083 * items, we fall through and let the regular search
1093 if (*--slash
== '/')
1095 if (slash
<= ce
->name
)
1102 * (len + 1) is a directory boundary (including
1103 * the trailing slash). And since the loop is
1104 * decrementing "slash", the first iteration is
1105 * the longest directory prefix; subsequent
1106 * iterations consider parent directories.
1109 if (len
+ 1 <= len_eq_last
) {
1111 * The directory prefix (including the trailing
1112 * slash) also appears as a prefix in the last
1113 * entry, so the remainder cannot collide (because
1114 * strcmp said the whole path was greater).
1119 * LT: last: xxx/file_A
1125 if (len
> len_eq_last
) {
1127 * This part of the directory prefix (excluding
1128 * the trailing slash) is longer than the known
1129 * equal portions, so this sub-directory cannot
1130 * collide with a file.
1138 if (istate
->cache_nr
> 0 &&
1139 ce_namelen(istate
->cache
[istate
->cache_nr
- 1]) > len
) {
1141 * The directory prefix lines up with part of
1142 * a longer file or directory name, but sorts
1143 * after it, so this sub-directory cannot
1144 * collide with a file.
1146 * last: xxx/yy-file (because '-' sorts before '/')
1153 * This is a possible collision. Fall through and
1154 * let the regular search code handle it.
1161 pos
= index_name_stage_pos(istate
, name
, len
, stage
);
1164 * Found one, but not so fast. This could
1165 * be a marker that says "I was here, but
1166 * I am being removed". Such an entry is
1167 * not a part of the resulting tree, and
1168 * it is Ok to have a directory at the same
1171 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
1175 remove_index_entry_at(istate
, pos
);
1183 * Trivial optimization: if we find an entry that
1184 * already matches the sub-directory, then we know
1185 * we're ok, and we can exit.
1187 while (pos
< istate
->cache_nr
) {
1188 struct cache_entry
*p
= istate
->cache
[pos
];
1189 if ((ce_namelen(p
) <= len
) ||
1190 (p
->name
[len
] != '/') ||
1191 memcmp(p
->name
, name
, len
))
1192 break; /* not our subdirectory */
1193 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
1195 * p is at the same stage as our entry, and
1196 * is a subdirectory of what we are looking
1197 * at, so we cannot have conflicts at our
1198 * level or anything shorter.
1207 /* We may be in a situation where we already have path/file and path
1208 * is being added, or we already have path and path/file is being
1209 * added. Either one would result in a nonsense tree that has path
1210 * twice when git-write-tree tries to write it out. Prevent it.
1212 * If ok-to-replace is specified, we remove the conflicting entries
1213 * from the cache so the caller should recompute the insert position.
1214 * When this happens, we return non-zero.
1216 static int check_file_directory_conflict(struct index_state
*istate
,
1217 const struct cache_entry
*ce
,
1218 int pos
, int ok_to_replace
)
1223 * When ce is an "I am going away" entry, we allow it to be added
1225 if (ce
->ce_flags
& CE_REMOVE
)
1229 * We check if the path is a sub-path of a subsequent pathname
1230 * first, since removing those will not change the position
1233 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
1236 * Then check if the path might have a clashing sub-directory
1239 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
1242 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1245 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
1246 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
1247 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
1248 int new_only
= option
& ADD_CACHE_NEW_ONLY
;
1250 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1251 cache_tree_invalidate_path(istate
, ce
->name
);
1254 * If this entry's path sorts after the last entry in the index,
1255 * we can avoid searching for it.
1257 if (istate
->cache_nr
> 0 &&
1258 strcmp(ce
->name
, istate
->cache
[istate
->cache_nr
- 1]->name
) > 0)
1259 pos
= -istate
->cache_nr
- 1;
1261 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1263 /* existing match? Just replace it. */
1266 replace_index_entry(istate
, pos
, ce
);
1271 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1272 untracked_cache_add_to_index(istate
, ce
->name
);
1275 * Inserting a merged entry ("stage 0") into the index
1276 * will always replace all non-merged entries..
1278 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
1279 while (ce_same_name(istate
->cache
[pos
], ce
)) {
1281 if (!remove_index_entry_at(istate
, pos
))
1288 if (!verify_path(ce
->name
, ce
->ce_mode
))
1289 return error("Invalid path '%s'", ce
->name
);
1291 if (!skip_df_check
&&
1292 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
1294 return error("'%s' appears as both a file and as a directory",
1296 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1302 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1306 if (option
& ADD_CACHE_JUST_APPEND
)
1307 pos
= istate
->cache_nr
;
1310 ret
= add_index_entry_with_check(istate
, ce
, option
);
1316 /* Make sure the array is big enough .. */
1317 ALLOC_GROW(istate
->cache
, istate
->cache_nr
+ 1, istate
->cache_alloc
);
1321 if (istate
->cache_nr
> pos
+ 1)
1322 MOVE_ARRAY(istate
->cache
+ pos
+ 1, istate
->cache
+ pos
,
1323 istate
->cache_nr
- pos
- 1);
1324 set_index_entry(istate
, pos
, ce
);
1325 istate
->cache_changed
|= CE_ENTRY_ADDED
;
1330 * "refresh" does not calculate a new sha1 file or bring the
1331 * cache up-to-date for mode/content changes. But what it
1332 * _does_ do is to "re-match" the stat information of a file
1333 * with the cache, so that you can refresh the cache for a
1334 * file that hasn't been changed but where the stat entry is
1337 * For example, you'd want to do this after doing a "git-read-tree",
1338 * to link up the stat cache details with the proper files.
1340 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
1341 struct cache_entry
*ce
,
1342 unsigned int options
, int *err
,
1346 struct cache_entry
*updated
;
1348 int refresh
= options
& CE_MATCH_REFRESH
;
1349 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
1350 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
1351 int ignore_missing
= options
& CE_MATCH_IGNORE_MISSING
;
1352 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
1354 if (!refresh
|| ce_uptodate(ce
))
1357 if (!ignore_fsmonitor
)
1358 refresh_fsmonitor(istate
);
1360 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1361 * that the change to the work tree does not matter and told
1364 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
)) {
1365 ce_mark_uptodate(ce
);
1368 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
)) {
1369 ce_mark_uptodate(ce
);
1372 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
)) {
1373 ce_mark_uptodate(ce
);
1377 if (has_symlink_leading_path(ce
->name
, ce_namelen(ce
))) {
1385 if (lstat(ce
->name
, &st
) < 0) {
1386 if (ignore_missing
&& errno
== ENOENT
)
1393 changed
= ie_match_stat(istate
, ce
, &st
, options
);
1395 *changed_ret
= changed
;
1398 * The path is unchanged. If we were told to ignore
1399 * valid bit, then we did the actual stat check and
1400 * found that the entry is unmodified. If the entry
1401 * is not marked VALID, this is the place to mark it
1402 * valid again, under "assume unchanged" mode.
1404 if (ignore_valid
&& assume_unchanged
&&
1405 !(ce
->ce_flags
& CE_VALID
))
1406 ; /* mark this one VALID again */
1409 * We do not mark the index itself "modified"
1410 * because CE_UPTODATE flag is in-core only;
1411 * we are not going to write this change out.
1413 if (!S_ISGITLINK(ce
->ce_mode
)) {
1414 ce_mark_uptodate(ce
);
1415 mark_fsmonitor_valid(ce
);
1421 if (ie_modified(istate
, ce
, &st
, options
)) {
1427 updated
= make_empty_cache_entry(istate
, ce_namelen(ce
));
1428 copy_cache_entry(updated
, ce
);
1429 memcpy(updated
->name
, ce
->name
, ce
->ce_namelen
+ 1);
1430 fill_stat_cache_info(updated
, &st
);
1432 * If ignore_valid is not set, we should leave CE_VALID bit
1433 * alone. Otherwise, paths marked with --no-assume-unchanged
1434 * (i.e. things to be edited) will reacquire CE_VALID bit
1435 * automatically, which is not really what we want.
1437 if (!ignore_valid
&& assume_unchanged
&&
1438 !(ce
->ce_flags
& CE_VALID
))
1439 updated
->ce_flags
&= ~CE_VALID
;
1441 /* istate->cache_changed is updated in the caller */
1445 static void show_file(const char * fmt
, const char * name
, int in_porcelain
,
1446 int * first
, const char *header_msg
)
1448 if (in_porcelain
&& *first
&& header_msg
) {
1449 printf("%s\n", header_msg
);
1455 int refresh_index(struct index_state
*istate
, unsigned int flags
,
1456 const struct pathspec
*pathspec
,
1457 char *seen
, const char *header_msg
)
1461 int really
= (flags
& REFRESH_REALLY
) != 0;
1462 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
1463 int quiet
= (flags
& REFRESH_QUIET
) != 0;
1464 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
1465 int ignore_submodules
= (flags
& REFRESH_IGNORE_SUBMODULES
) != 0;
1467 int in_porcelain
= (flags
& REFRESH_IN_PORCELAIN
);
1468 unsigned int options
= (CE_MATCH_REFRESH
|
1469 (really
? CE_MATCH_IGNORE_VALID
: 0) |
1470 (not_new
? CE_MATCH_IGNORE_MISSING
: 0));
1471 const char *modified_fmt
;
1472 const char *deleted_fmt
;
1473 const char *typechange_fmt
;
1474 const char *added_fmt
;
1475 const char *unmerged_fmt
;
1476 uint64_t start
= getnanotime();
1478 modified_fmt
= (in_porcelain
? "M\t%s\n" : "%s: needs update\n");
1479 deleted_fmt
= (in_porcelain
? "D\t%s\n" : "%s: needs update\n");
1480 typechange_fmt
= (in_porcelain
? "T\t%s\n" : "%s needs update\n");
1481 added_fmt
= (in_porcelain
? "A\t%s\n" : "%s needs update\n");
1482 unmerged_fmt
= (in_porcelain
? "U\t%s\n" : "%s: needs merge\n");
1483 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1484 struct cache_entry
*ce
, *new_entry
;
1485 int cache_errno
= 0;
1489 ce
= istate
->cache
[i
];
1490 if (ignore_submodules
&& S_ISGITLINK(ce
->ce_mode
))
1493 if (pathspec
&& !ce_path_match(ce
, pathspec
, seen
))
1497 while ((i
< istate
->cache_nr
) &&
1498 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
1504 show_file(unmerged_fmt
, ce
->name
, in_porcelain
,
1505 &first
, header_msg
);
1513 new_entry
= refresh_cache_ent(istate
, ce
, options
, &cache_errno
, &changed
);
1514 if (new_entry
== ce
)
1519 if (really
&& cache_errno
== EINVAL
) {
1520 /* If we are doing --really-refresh that
1521 * means the index is not valid anymore.
1523 ce
->ce_flags
&= ~CE_VALID
;
1524 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
1525 mark_fsmonitor_invalid(istate
, ce
);
1526 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
1531 if (cache_errno
== ENOENT
)
1533 else if (ce_intent_to_add(ce
))
1534 fmt
= added_fmt
; /* must be before other checks */
1535 else if (changed
& TYPE_CHANGED
)
1536 fmt
= typechange_fmt
;
1540 ce
->name
, in_porcelain
, &first
, header_msg
);
1545 replace_index_entry(istate
, i
, new_entry
);
1547 trace_performance_since(start
, "refresh index");
1551 struct cache_entry
*refresh_cache_entry(struct index_state
*istate
,
1552 struct cache_entry
*ce
,
1553 unsigned int options
)
1555 return refresh_cache_ent(istate
, ce
, options
, NULL
, NULL
);
1559 /*****************************************************************
1561 *****************************************************************/
1563 #define INDEX_FORMAT_DEFAULT 3
1565 static unsigned int get_index_format_default(void)
1567 char *envversion
= getenv("GIT_INDEX_VERSION");
1570 unsigned int version
= INDEX_FORMAT_DEFAULT
;
1573 if (!git_config_get_int("index.version", &value
))
1575 if (version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1576 warning(_("index.version set, but the value is invalid.\n"
1577 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1578 return INDEX_FORMAT_DEFAULT
;
1583 version
= strtoul(envversion
, &endp
, 10);
1585 version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1586 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1587 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1588 version
= INDEX_FORMAT_DEFAULT
;
1594 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1595 * Again - this is just a (very strong in practice) heuristic that
1596 * the inode hasn't changed.
1598 * We save the fields in big-endian order to allow using the
1599 * index file over NFS transparently.
1601 struct ondisk_cache_entry
{
1602 struct cache_time ctime
;
1603 struct cache_time mtime
;
1610 unsigned char sha1
[20];
1612 char name
[FLEX_ARRAY
]; /* more */
1616 * This struct is used when CE_EXTENDED bit is 1
1617 * The struct must match ondisk_cache_entry exactly from
1620 struct ondisk_cache_entry_extended
{
1621 struct cache_time ctime
;
1622 struct cache_time mtime
;
1629 unsigned char sha1
[20];
1632 char name
[FLEX_ARRAY
]; /* more */
1635 /* These are only used for v3 or lower */
1636 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1637 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1638 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1639 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1640 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1641 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1642 ondisk_cache_entry_size(ce_namelen(ce)))
1644 /* Allow fsck to force verification of the index checksum. */
1645 int verify_index_checksum
;
1647 /* Allow fsck to force verification of the cache entry order. */
1648 int verify_ce_order
;
1650 static int verify_hdr(struct cache_header
*hdr
, unsigned long size
)
1653 unsigned char hash
[GIT_MAX_RAWSZ
];
1656 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
1657 return error("bad signature");
1658 hdr_version
= ntohl(hdr
->hdr_version
);
1659 if (hdr_version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< hdr_version
)
1660 return error("bad index version %d", hdr_version
);
1662 if (!verify_index_checksum
)
1665 the_hash_algo
->init_fn(&c
);
1666 the_hash_algo
->update_fn(&c
, hdr
, size
- the_hash_algo
->rawsz
);
1667 the_hash_algo
->final_fn(hash
, &c
);
1668 if (hashcmp(hash
, (unsigned char *)hdr
+ size
- the_hash_algo
->rawsz
))
1669 return error("bad index file sha1 signature");
1673 static int read_index_extension(struct index_state
*istate
,
1674 const char *ext
, void *data
, unsigned long sz
)
1676 switch (CACHE_EXT(ext
)) {
1677 case CACHE_EXT_TREE
:
1678 istate
->cache_tree
= cache_tree_read(data
, sz
);
1680 case CACHE_EXT_RESOLVE_UNDO
:
1681 istate
->resolve_undo
= resolve_undo_read(data
, sz
);
1683 case CACHE_EXT_LINK
:
1684 if (read_link_extension(istate
, data
, sz
))
1687 case CACHE_EXT_UNTRACKED
:
1688 istate
->untracked
= read_untracked_extension(data
, sz
);
1690 case CACHE_EXT_FSMONITOR
:
1691 read_fsmonitor_extension(istate
, data
, sz
);
1694 if (*ext
< 'A' || 'Z' < *ext
)
1695 return error("index uses %.4s extension, which we do not understand",
1697 fprintf(stderr
, "ignoring %.4s extension\n", ext
);
1703 int hold_locked_index(struct lock_file
*lk
, int lock_flags
)
1705 return hold_lock_file_for_update(lk
, get_index_file(), lock_flags
);
1708 int read_index(struct index_state
*istate
)
1710 return read_index_from(istate
, get_index_file(), get_git_dir());
1713 static struct cache_entry
*cache_entry_from_ondisk(struct mem_pool
*mem_pool
,
1714 struct ondisk_cache_entry
*ondisk
,
1719 struct cache_entry
*ce
= mem_pool__ce_alloc(mem_pool
, len
);
1721 ce
->ce_stat_data
.sd_ctime
.sec
= get_be32(&ondisk
->ctime
.sec
);
1722 ce
->ce_stat_data
.sd_mtime
.sec
= get_be32(&ondisk
->mtime
.sec
);
1723 ce
->ce_stat_data
.sd_ctime
.nsec
= get_be32(&ondisk
->ctime
.nsec
);
1724 ce
->ce_stat_data
.sd_mtime
.nsec
= get_be32(&ondisk
->mtime
.nsec
);
1725 ce
->ce_stat_data
.sd_dev
= get_be32(&ondisk
->dev
);
1726 ce
->ce_stat_data
.sd_ino
= get_be32(&ondisk
->ino
);
1727 ce
->ce_mode
= get_be32(&ondisk
->mode
);
1728 ce
->ce_stat_data
.sd_uid
= get_be32(&ondisk
->uid
);
1729 ce
->ce_stat_data
.sd_gid
= get_be32(&ondisk
->gid
);
1730 ce
->ce_stat_data
.sd_size
= get_be32(&ondisk
->size
);
1731 ce
->ce_flags
= flags
& ~CE_NAMEMASK
;
1732 ce
->ce_namelen
= len
;
1734 hashcpy(ce
->oid
.hash
, ondisk
->sha1
);
1735 memcpy(ce
->name
, name
, len
);
1736 ce
->name
[len
] = '\0';
1741 * Adjacent cache entries tend to share the leading paths, so it makes
1742 * sense to only store the differences in later entries. In the v4
1743 * on-disk format of the index, each on-disk cache entry stores the
1744 * number of bytes to be stripped from the end of the previous name,
1745 * and the bytes to append to the result, to come up with its name.
1747 static unsigned long expand_name_field(struct strbuf
*name
, const char *cp_
)
1749 const unsigned char *ep
, *cp
= (const unsigned char *)cp_
;
1750 size_t len
= decode_varint(&cp
);
1752 if (name
->len
< len
)
1753 die("malformed name field in the index");
1754 strbuf_remove(name
, name
->len
- len
, len
);
1755 for (ep
= cp
; *ep
; ep
++)
1756 ; /* find the end */
1757 strbuf_add(name
, cp
, ep
- cp
);
1758 return (const char *)ep
+ 1 - cp_
;
1761 static struct cache_entry
*create_from_disk(struct mem_pool
*mem_pool
,
1762 struct ondisk_cache_entry
*ondisk
,
1763 unsigned long *ent_size
,
1764 struct strbuf
*previous_name
)
1766 struct cache_entry
*ce
;
1771 /* On-disk flags are just 16 bits */
1772 flags
= get_be16(&ondisk
->flags
);
1773 len
= flags
& CE_NAMEMASK
;
1775 if (flags
& CE_EXTENDED
) {
1776 struct ondisk_cache_entry_extended
*ondisk2
;
1778 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
1779 extended_flags
= get_be16(&ondisk2
->flags2
) << 16;
1780 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1781 if (extended_flags
& ~CE_EXTENDED_FLAGS
)
1782 die("Unknown index entry format %08x", extended_flags
);
1783 flags
|= extended_flags
;
1784 name
= ondisk2
->name
;
1787 name
= ondisk
->name
;
1789 if (!previous_name
) {
1790 /* v3 and earlier */
1791 if (len
== CE_NAMEMASK
)
1793 ce
= cache_entry_from_ondisk(mem_pool
, ondisk
, flags
, name
, len
);
1795 *ent_size
= ondisk_ce_size(ce
);
1797 unsigned long consumed
;
1798 consumed
= expand_name_field(previous_name
, name
);
1799 ce
= cache_entry_from_ondisk(mem_pool
, ondisk
, flags
,
1801 previous_name
->len
);
1803 *ent_size
= (name
- ((char *)ondisk
)) + consumed
;
1808 static void check_ce_order(struct index_state
*istate
)
1812 if (!verify_ce_order
)
1815 for (i
= 1; i
< istate
->cache_nr
; i
++) {
1816 struct cache_entry
*ce
= istate
->cache
[i
- 1];
1817 struct cache_entry
*next_ce
= istate
->cache
[i
];
1818 int name_compare
= strcmp(ce
->name
, next_ce
->name
);
1820 if (0 < name_compare
)
1821 die("unordered stage entries in index");
1822 if (!name_compare
) {
1824 die("multiple stage entries for merged file '%s'",
1826 if (ce_stage(ce
) > ce_stage(next_ce
))
1827 die("unordered stage entries for '%s'",
1833 static void tweak_untracked_cache(struct index_state
*istate
)
1835 switch (git_config_get_untracked_cache()) {
1836 case -1: /* keep: do nothing */
1839 remove_untracked_cache(istate
);
1842 add_untracked_cache(istate
);
1844 default: /* unknown value: do nothing */
1849 static void tweak_split_index(struct index_state
*istate
)
1851 switch (git_config_get_split_index()) {
1852 case -1: /* unset: do nothing */
1855 remove_split_index(istate
);
1858 add_split_index(istate
);
1860 default: /* unknown value: do nothing */
1865 static void post_read_index_from(struct index_state
*istate
)
1867 check_ce_order(istate
);
1868 tweak_untracked_cache(istate
);
1869 tweak_split_index(istate
);
1870 tweak_fsmonitor(istate
);
1873 static size_t estimate_cache_size_from_compressed(unsigned int entries
)
1875 return entries
* (sizeof(struct cache_entry
) + CACHE_ENTRY_PATH_LENGTH
);
1878 static size_t estimate_cache_size(size_t ondisk_size
, unsigned int entries
)
1880 long per_entry
= sizeof(struct cache_entry
) - sizeof(struct ondisk_cache_entry
);
1883 * Account for potential alignment differences.
1885 per_entry
+= align_padding_size(sizeof(struct cache_entry
), -sizeof(struct ondisk_cache_entry
));
1886 return ondisk_size
+ entries
* per_entry
;
1889 /* remember to discard_cache() before reading a different cache! */
1890 int do_read_index(struct index_state
*istate
, const char *path
, int must_exist
)
1894 unsigned long src_offset
;
1895 struct cache_header
*hdr
;
1898 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
1900 if (istate
->initialized
)
1901 return istate
->cache_nr
;
1903 istate
->timestamp
.sec
= 0;
1904 istate
->timestamp
.nsec
= 0;
1905 fd
= open(path
, O_RDONLY
);
1907 if (!must_exist
&& errno
== ENOENT
)
1909 die_errno("%s: index file open failed", path
);
1913 die_errno("cannot stat the open index");
1915 mmap_size
= xsize_t(st
.st_size
);
1916 if (mmap_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
1917 die("index file smaller than expected");
1919 mmap
= xmmap(NULL
, mmap_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1920 if (mmap
== MAP_FAILED
)
1921 die_errno("unable to map index file");
1925 if (verify_hdr(hdr
, mmap_size
) < 0)
1928 hashcpy(istate
->oid
.hash
, (const unsigned char *)hdr
+ mmap_size
- the_hash_algo
->rawsz
);
1929 istate
->version
= ntohl(hdr
->hdr_version
);
1930 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
1931 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
1932 istate
->cache
= xcalloc(istate
->cache_alloc
, sizeof(*istate
->cache
));
1933 istate
->initialized
= 1;
1935 if (istate
->version
== 4) {
1936 previous_name
= &previous_name_buf
;
1937 mem_pool_init(&istate
->ce_mem_pool
,
1938 estimate_cache_size_from_compressed(istate
->cache_nr
));
1940 previous_name
= NULL
;
1941 mem_pool_init(&istate
->ce_mem_pool
,
1942 estimate_cache_size(mmap_size
, istate
->cache_nr
));
1945 src_offset
= sizeof(*hdr
);
1946 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1947 struct ondisk_cache_entry
*disk_ce
;
1948 struct cache_entry
*ce
;
1949 unsigned long consumed
;
1951 disk_ce
= (struct ondisk_cache_entry
*)((char *)mmap
+ src_offset
);
1952 ce
= create_from_disk(istate
->ce_mem_pool
, disk_ce
, &consumed
, previous_name
);
1953 set_index_entry(istate
, i
, ce
);
1955 src_offset
+= consumed
;
1957 strbuf_release(&previous_name_buf
);
1958 istate
->timestamp
.sec
= st
.st_mtime
;
1959 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
1961 while (src_offset
<= mmap_size
- the_hash_algo
->rawsz
- 8) {
1962 /* After an array of active_nr index entries,
1963 * there can be arbitrary number of extended
1964 * sections, each of which is prefixed with
1965 * extension name (4-byte) and section length
1966 * in 4-byte network byte order.
1969 memcpy(&extsize
, (char *)mmap
+ src_offset
+ 4, 4);
1970 extsize
= ntohl(extsize
);
1971 if (read_index_extension(istate
,
1972 (const char *) mmap
+ src_offset
,
1973 (char *) mmap
+ src_offset
+ 8,
1977 src_offset
+= extsize
;
1979 munmap(mmap
, mmap_size
);
1980 return istate
->cache_nr
;
1983 munmap(mmap
, mmap_size
);
1984 die("index file corrupt");
1988 * Signal that the shared index is used by updating its mtime.
1990 * This way, shared index can be removed if they have not been used
1993 static void freshen_shared_index(const char *shared_index
, int warn
)
1995 if (!check_and_freshen_file(shared_index
, 1) && warn
)
1996 warning("could not freshen shared index '%s'", shared_index
);
1999 int read_index_from(struct index_state
*istate
, const char *path
,
2002 uint64_t start
= getnanotime();
2003 struct split_index
*split_index
;
2008 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2009 if (istate
->initialized
)
2010 return istate
->cache_nr
;
2012 ret
= do_read_index(istate
, path
, 0);
2013 trace_performance_since(start
, "read cache %s", path
);
2015 split_index
= istate
->split_index
;
2016 if (!split_index
|| is_null_oid(&split_index
->base_oid
)) {
2017 post_read_index_from(istate
);
2021 if (split_index
->base
)
2022 discard_index(split_index
->base
);
2024 split_index
->base
= xcalloc(1, sizeof(*split_index
->base
));
2026 base_oid_hex
= oid_to_hex(&split_index
->base_oid
);
2027 base_path
= xstrfmt("%s/sharedindex.%s", gitdir
, base_oid_hex
);
2028 ret
= do_read_index(split_index
->base
, base_path
, 1);
2029 if (oidcmp(&split_index
->base_oid
, &split_index
->base
->oid
))
2030 die("broken index, expect %s in %s, got %s",
2031 base_oid_hex
, base_path
,
2032 oid_to_hex(&split_index
->base
->oid
));
2034 freshen_shared_index(base_path
, 0);
2035 merge_base_index(istate
);
2036 post_read_index_from(istate
);
2037 trace_performance_since(start
, "read cache %s", base_path
);
2042 int is_index_unborn(struct index_state
*istate
)
2044 return (!istate
->cache_nr
&& !istate
->timestamp
.sec
);
2047 int discard_index(struct index_state
*istate
)
2050 * Cache entries in istate->cache[] should have been allocated
2051 * from the memory pool associated with this index, or from an
2052 * associated split_index. There is no need to free individual
2056 resolve_undo_clear_index(istate
);
2057 istate
->cache_nr
= 0;
2058 istate
->cache_changed
= 0;
2059 istate
->timestamp
.sec
= 0;
2060 istate
->timestamp
.nsec
= 0;
2061 free_name_hash(istate
);
2062 cache_tree_free(&(istate
->cache_tree
));
2063 istate
->initialized
= 0;
2064 FREE_AND_NULL(istate
->cache
);
2065 istate
->cache_alloc
= 0;
2066 discard_split_index(istate
);
2067 free_untracked_cache(istate
->untracked
);
2068 istate
->untracked
= NULL
;
2070 if (istate
->ce_mem_pool
) {
2071 mem_pool_discard(istate
->ce_mem_pool
);
2072 istate
->ce_mem_pool
= NULL
;
2078 int unmerged_index(const struct index_state
*istate
)
2081 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2082 if (ce_stage(istate
->cache
[i
]))
2088 #define WRITE_BUFFER_SIZE 8192
2089 static unsigned char write_buffer
[WRITE_BUFFER_SIZE
];
2090 static unsigned long write_buffer_len
;
2092 static int ce_write_flush(git_hash_ctx
*context
, int fd
)
2094 unsigned int buffered
= write_buffer_len
;
2096 the_hash_algo
->update_fn(context
, write_buffer
, buffered
);
2097 if (write_in_full(fd
, write_buffer
, buffered
) < 0)
2099 write_buffer_len
= 0;
2104 static int ce_write(git_hash_ctx
*context
, int fd
, void *data
, unsigned int len
)
2107 unsigned int buffered
= write_buffer_len
;
2108 unsigned int partial
= WRITE_BUFFER_SIZE
- buffered
;
2111 memcpy(write_buffer
+ buffered
, data
, partial
);
2112 buffered
+= partial
;
2113 if (buffered
== WRITE_BUFFER_SIZE
) {
2114 write_buffer_len
= buffered
;
2115 if (ce_write_flush(context
, fd
))
2119 write_buffer_len
= buffered
;
2121 data
= (char *) data
+ partial
;
2126 static int write_index_ext_header(git_hash_ctx
*context
, int fd
,
2127 unsigned int ext
, unsigned int sz
)
2131 return ((ce_write(context
, fd
, &ext
, 4) < 0) ||
2132 (ce_write(context
, fd
, &sz
, 4) < 0)) ? -1 : 0;
2135 static int ce_flush(git_hash_ctx
*context
, int fd
, unsigned char *hash
)
2137 unsigned int left
= write_buffer_len
;
2140 write_buffer_len
= 0;
2141 the_hash_algo
->update_fn(context
, write_buffer
, left
);
2144 /* Flush first if not enough space for hash signature */
2145 if (left
+ the_hash_algo
->rawsz
> WRITE_BUFFER_SIZE
) {
2146 if (write_in_full(fd
, write_buffer
, left
) < 0)
2151 /* Append the hash signature at the end */
2152 the_hash_algo
->final_fn(write_buffer
+ left
, context
);
2153 hashcpy(hash
, write_buffer
+ left
);
2154 left
+= the_hash_algo
->rawsz
;
2155 return (write_in_full(fd
, write_buffer
, left
) < 0) ? -1 : 0;
2158 static void ce_smudge_racily_clean_entry(struct cache_entry
*ce
)
2161 * The only thing we care about in this function is to smudge the
2162 * falsely clean entry due to touch-update-touch race, so we leave
2163 * everything else as they are. We are called for entries whose
2164 * ce_stat_data.sd_mtime match the index file mtime.
2166 * Note that this actually does not do much for gitlinks, for
2167 * which ce_match_stat_basic() always goes to the actual
2168 * contents. The caller checks with is_racy_timestamp() which
2169 * always says "no" for gitlinks, so we are not called for them ;-)
2173 if (lstat(ce
->name
, &st
) < 0)
2175 if (ce_match_stat_basic(ce
, &st
))
2177 if (ce_modified_check_fs(ce
, &st
)) {
2178 /* This is "racily clean"; smudge it. Note that this
2179 * is a tricky code. At first glance, it may appear
2180 * that it can break with this sequence:
2182 * $ echo xyzzy >frotz
2183 * $ git-update-index --add frotz
2186 * $ echo filfre >nitfol
2187 * $ git-update-index --add nitfol
2189 * but it does not. When the second update-index runs,
2190 * it notices that the entry "frotz" has the same timestamp
2191 * as index, and if we were to smudge it by resetting its
2192 * size to zero here, then the object name recorded
2193 * in index is the 6-byte file but the cached stat information
2194 * becomes zero --- which would then match what we would
2195 * obtain from the filesystem next time we stat("frotz").
2197 * However, the second update-index, before calling
2198 * this function, notices that the cached size is 6
2199 * bytes and what is on the filesystem is an empty
2200 * file, and never calls us, so the cached size information
2201 * for "frotz" stays 6 which does not match the filesystem.
2203 ce
->ce_stat_data
.sd_size
= 0;
2207 /* Copy miscellaneous fields but not the name */
2208 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry
*ondisk
,
2209 struct cache_entry
*ce
)
2213 ondisk
->ctime
.sec
= htonl(ce
->ce_stat_data
.sd_ctime
.sec
);
2214 ondisk
->mtime
.sec
= htonl(ce
->ce_stat_data
.sd_mtime
.sec
);
2215 ondisk
->ctime
.nsec
= htonl(ce
->ce_stat_data
.sd_ctime
.nsec
);
2216 ondisk
->mtime
.nsec
= htonl(ce
->ce_stat_data
.sd_mtime
.nsec
);
2217 ondisk
->dev
= htonl(ce
->ce_stat_data
.sd_dev
);
2218 ondisk
->ino
= htonl(ce
->ce_stat_data
.sd_ino
);
2219 ondisk
->mode
= htonl(ce
->ce_mode
);
2220 ondisk
->uid
= htonl(ce
->ce_stat_data
.sd_uid
);
2221 ondisk
->gid
= htonl(ce
->ce_stat_data
.sd_gid
);
2222 ondisk
->size
= htonl(ce
->ce_stat_data
.sd_size
);
2223 hashcpy(ondisk
->sha1
, ce
->oid
.hash
);
2225 flags
= ce
->ce_flags
& ~CE_NAMEMASK
;
2226 flags
|= (ce_namelen(ce
) >= CE_NAMEMASK
? CE_NAMEMASK
: ce_namelen(ce
));
2227 ondisk
->flags
= htons(flags
);
2228 if (ce
->ce_flags
& CE_EXTENDED
) {
2229 struct ondisk_cache_entry_extended
*ondisk2
;
2230 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
2231 ondisk2
->flags2
= htons((ce
->ce_flags
& CE_EXTENDED_FLAGS
) >> 16);
2235 static int ce_write_entry(git_hash_ctx
*c
, int fd
, struct cache_entry
*ce
,
2236 struct strbuf
*previous_name
, struct ondisk_cache_entry
*ondisk
)
2240 unsigned int saved_namelen
;
2241 int stripped_name
= 0;
2242 static unsigned char padding
[8] = { 0x00 };
2244 if (ce
->ce_flags
& CE_STRIP_NAME
) {
2245 saved_namelen
= ce_namelen(ce
);
2250 if (ce
->ce_flags
& CE_EXTENDED
)
2251 size
= offsetof(struct ondisk_cache_entry_extended
, name
);
2253 size
= offsetof(struct ondisk_cache_entry
, name
);
2255 if (!previous_name
) {
2256 int len
= ce_namelen(ce
);
2257 copy_cache_entry_to_ondisk(ondisk
, ce
);
2258 result
= ce_write(c
, fd
, ondisk
, size
);
2260 result
= ce_write(c
, fd
, ce
->name
, len
);
2262 result
= ce_write(c
, fd
, padding
, align_padding_size(size
, len
));
2264 int common
, to_remove
, prefix_size
;
2265 unsigned char to_remove_vi
[16];
2267 (ce
->name
[common
] &&
2268 common
< previous_name
->len
&&
2269 ce
->name
[common
] == previous_name
->buf
[common
]);
2271 ; /* still matching */
2272 to_remove
= previous_name
->len
- common
;
2273 prefix_size
= encode_varint(to_remove
, to_remove_vi
);
2275 copy_cache_entry_to_ondisk(ondisk
, ce
);
2276 result
= ce_write(c
, fd
, ondisk
, size
);
2278 result
= ce_write(c
, fd
, to_remove_vi
, prefix_size
);
2280 result
= ce_write(c
, fd
, ce
->name
+ common
, ce_namelen(ce
) - common
);
2282 result
= ce_write(c
, fd
, padding
, 1);
2284 strbuf_splice(previous_name
, common
, to_remove
,
2285 ce
->name
+ common
, ce_namelen(ce
) - common
);
2287 if (stripped_name
) {
2288 ce
->ce_namelen
= saved_namelen
;
2289 ce
->ce_flags
&= ~CE_STRIP_NAME
;
2296 * This function verifies if index_state has the correct sha1 of the
2297 * index file. Don't die if we have any other failure, just return 0.
2299 static int verify_index_from(const struct index_state
*istate
, const char *path
)
2304 unsigned char hash
[GIT_MAX_RAWSZ
];
2306 if (!istate
->initialized
)
2309 fd
= open(path
, O_RDONLY
);
2316 if (st
.st_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2319 n
= pread_in_full(fd
, hash
, the_hash_algo
->rawsz
, st
.st_size
- the_hash_algo
->rawsz
);
2320 if (n
!= the_hash_algo
->rawsz
)
2323 if (hashcmp(istate
->oid
.hash
, hash
))
2334 static int verify_index(const struct index_state
*istate
)
2336 return verify_index_from(istate
, get_index_file());
2339 static int has_racy_timestamp(struct index_state
*istate
)
2341 int entries
= istate
->cache_nr
;
2344 for (i
= 0; i
< entries
; i
++) {
2345 struct cache_entry
*ce
= istate
->cache
[i
];
2346 if (is_racy_timestamp(istate
, ce
))
2352 void update_index_if_able(struct index_state
*istate
, struct lock_file
*lockfile
)
2354 if ((istate
->cache_changed
|| has_racy_timestamp(istate
)) &&
2355 verify_index(istate
))
2356 write_locked_index(istate
, lockfile
, COMMIT_LOCK
);
2358 rollback_lock_file(lockfile
);
2362 * On success, `tempfile` is closed. If it is the temporary file
2363 * of a `struct lock_file`, we will therefore effectively perform
2364 * a 'close_lock_file_gently()`. Since that is an implementation
2365 * detail of lockfiles, callers of `do_write_index()` should not
2368 static int do_write_index(struct index_state
*istate
, struct tempfile
*tempfile
,
2369 int strip_extensions
)
2371 uint64_t start
= getnanotime();
2372 int newfd
= tempfile
->fd
;
2374 struct cache_header hdr
;
2375 int i
, err
= 0, removed
, extended
, hdr_version
;
2376 struct cache_entry
**cache
= istate
->cache
;
2377 int entries
= istate
->cache_nr
;
2379 struct ondisk_cache_entry_extended ondisk
;
2380 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
2381 int drop_cache_tree
= istate
->drop_cache_tree
;
2383 for (i
= removed
= extended
= 0; i
< entries
; i
++) {
2384 if (cache
[i
]->ce_flags
& CE_REMOVE
)
2387 /* reduce extended entries if possible */
2388 cache
[i
]->ce_flags
&= ~CE_EXTENDED
;
2389 if (cache
[i
]->ce_flags
& CE_EXTENDED_FLAGS
) {
2391 cache
[i
]->ce_flags
|= CE_EXTENDED
;
2395 if (!istate
->version
) {
2396 istate
->version
= get_index_format_default();
2397 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0))
2398 init_split_index(istate
);
2401 /* demote version 3 to version 2 when the latter suffices */
2402 if (istate
->version
== 3 || istate
->version
== 2)
2403 istate
->version
= extended
? 3 : 2;
2405 hdr_version
= istate
->version
;
2407 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
2408 hdr
.hdr_version
= htonl(hdr_version
);
2409 hdr
.hdr_entries
= htonl(entries
- removed
);
2411 the_hash_algo
->init_fn(&c
);
2412 if (ce_write(&c
, newfd
, &hdr
, sizeof(hdr
)) < 0)
2415 previous_name
= (hdr_version
== 4) ? &previous_name_buf
: NULL
;
2417 for (i
= 0; i
< entries
; i
++) {
2418 struct cache_entry
*ce
= cache
[i
];
2419 if (ce
->ce_flags
& CE_REMOVE
)
2421 if (!ce_uptodate(ce
) && is_racy_timestamp(istate
, ce
))
2422 ce_smudge_racily_clean_entry(ce
);
2423 if (is_null_oid(&ce
->oid
)) {
2424 static const char msg
[] = "cache entry has null sha1: %s";
2425 static int allow
= -1;
2428 allow
= git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2430 warning(msg
, ce
->name
);
2432 err
= error(msg
, ce
->name
);
2434 drop_cache_tree
= 1;
2436 if (ce_write_entry(&c
, newfd
, ce
, previous_name
, (struct ondisk_cache_entry
*)&ondisk
) < 0)
2442 strbuf_release(&previous_name_buf
);
2447 /* Write extension data here */
2448 if (!strip_extensions
&& istate
->split_index
) {
2449 struct strbuf sb
= STRBUF_INIT
;
2451 err
= write_link_extension(&sb
, istate
) < 0 ||
2452 write_index_ext_header(&c
, newfd
, CACHE_EXT_LINK
,
2454 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2455 strbuf_release(&sb
);
2459 if (!strip_extensions
&& !drop_cache_tree
&& istate
->cache_tree
) {
2460 struct strbuf sb
= STRBUF_INIT
;
2462 cache_tree_write(&sb
, istate
->cache_tree
);
2463 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_TREE
, sb
.len
) < 0
2464 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2465 strbuf_release(&sb
);
2469 if (!strip_extensions
&& istate
->resolve_undo
) {
2470 struct strbuf sb
= STRBUF_INIT
;
2472 resolve_undo_write(&sb
, istate
->resolve_undo
);
2473 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_RESOLVE_UNDO
,
2475 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2476 strbuf_release(&sb
);
2480 if (!strip_extensions
&& istate
->untracked
) {
2481 struct strbuf sb
= STRBUF_INIT
;
2483 write_untracked_extension(&sb
, istate
->untracked
);
2484 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_UNTRACKED
,
2486 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2487 strbuf_release(&sb
);
2491 if (!strip_extensions
&& istate
->fsmonitor_last_update
) {
2492 struct strbuf sb
= STRBUF_INIT
;
2494 write_fsmonitor_extension(&sb
, istate
);
2495 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_FSMONITOR
, sb
.len
) < 0
2496 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2497 strbuf_release(&sb
);
2502 if (ce_flush(&c
, newfd
, istate
->oid
.hash
))
2504 if (close_tempfile_gently(tempfile
)) {
2505 error(_("could not close '%s'"), tempfile
->filename
.buf
);
2508 if (stat(tempfile
->filename
.buf
, &st
))
2510 istate
->timestamp
.sec
= (unsigned int)st
.st_mtime
;
2511 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
2512 trace_performance_since(start
, "write index, changed mask = %x", istate
->cache_changed
);
2516 void set_alternate_index_output(const char *name
)
2518 alternate_index_output
= name
;
2521 static int commit_locked_index(struct lock_file
*lk
)
2523 if (alternate_index_output
)
2524 return commit_lock_file_to(lk
, alternate_index_output
);
2526 return commit_lock_file(lk
);
2529 static int do_write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
2532 int ret
= do_write_index(istate
, lock
->tempfile
, 0);
2535 if (flags
& COMMIT_LOCK
)
2536 return commit_locked_index(lock
);
2537 return close_lock_file_gently(lock
);
2540 static int write_split_index(struct index_state
*istate
,
2541 struct lock_file
*lock
,
2545 prepare_to_write_split_index(istate
);
2546 ret
= do_write_locked_index(istate
, lock
, flags
);
2547 finish_writing_split_index(istate
);
2551 static const char *shared_index_expire
= "2.weeks.ago";
2553 static unsigned long get_shared_index_expire_date(void)
2555 static unsigned long shared_index_expire_date
;
2556 static int shared_index_expire_date_prepared
;
2558 if (!shared_index_expire_date_prepared
) {
2559 git_config_get_expiry("splitindex.sharedindexexpire",
2560 &shared_index_expire
);
2561 shared_index_expire_date
= approxidate(shared_index_expire
);
2562 shared_index_expire_date_prepared
= 1;
2565 return shared_index_expire_date
;
2568 static int should_delete_shared_index(const char *shared_index_path
)
2571 unsigned long expiration
;
2573 /* Check timestamp */
2574 expiration
= get_shared_index_expire_date();
2577 if (stat(shared_index_path
, &st
))
2578 return error_errno(_("could not stat '%s'"), shared_index_path
);
2579 if (st
.st_mtime
> expiration
)
2585 static int clean_shared_index_files(const char *current_hex
)
2588 DIR *dir
= opendir(get_git_dir());
2591 return error_errno(_("unable to open git dir: %s"), get_git_dir());
2593 while ((de
= readdir(dir
)) != NULL
) {
2594 const char *sha1_hex
;
2595 const char *shared_index_path
;
2596 if (!skip_prefix(de
->d_name
, "sharedindex.", &sha1_hex
))
2598 if (!strcmp(sha1_hex
, current_hex
))
2600 shared_index_path
= git_path("%s", de
->d_name
);
2601 if (should_delete_shared_index(shared_index_path
) > 0 &&
2602 unlink(shared_index_path
))
2603 warning_errno(_("unable to unlink: %s"), shared_index_path
);
2610 static int write_shared_index(struct index_state
*istate
,
2611 struct tempfile
**temp
)
2613 struct split_index
*si
= istate
->split_index
;
2616 move_cache_to_base_index(istate
);
2617 ret
= do_write_index(si
->base
, *temp
, 1);
2620 ret
= adjust_shared_perm(get_tempfile_path(*temp
));
2622 error("cannot fix permission bits on %s", get_tempfile_path(*temp
));
2625 ret
= rename_tempfile(temp
,
2626 git_path("sharedindex.%s", oid_to_hex(&si
->base
->oid
)));
2628 oidcpy(&si
->base_oid
, &si
->base
->oid
);
2629 clean_shared_index_files(oid_to_hex(&si
->base
->oid
));
2635 static const int default_max_percent_split_change
= 20;
2637 static int too_many_not_shared_entries(struct index_state
*istate
)
2639 int i
, not_shared
= 0;
2640 int max_split
= git_config_get_max_percent_split_change();
2642 switch (max_split
) {
2644 /* not or badly configured: use the default value */
2645 max_split
= default_max_percent_split_change
;
2648 return 1; /* 0% means always write a new shared index */
2650 return 0; /* 100% means never write a new shared index */
2652 break; /* just use the configured value */
2655 /* Count not shared entries */
2656 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2657 struct cache_entry
*ce
= istate
->cache
[i
];
2662 return (int64_t)istate
->cache_nr
* max_split
< (int64_t)not_shared
* 100;
2665 int write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
2668 int new_shared_index
, ret
;
2669 struct split_index
*si
= istate
->split_index
;
2671 if ((flags
& SKIP_IF_UNCHANGED
) && !istate
->cache_changed
) {
2672 if (flags
& COMMIT_LOCK
)
2673 rollback_lock_file(lock
);
2677 if (istate
->fsmonitor_last_update
)
2678 fill_fsmonitor_bitmap(istate
);
2680 if (!si
|| alternate_index_output
||
2681 (istate
->cache_changed
& ~EXTMASK
)) {
2683 oidclr(&si
->base_oid
);
2684 ret
= do_write_locked_index(istate
, lock
, flags
);
2688 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0)) {
2689 int v
= si
->base_oid
.hash
[0];
2691 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
2693 if (too_many_not_shared_entries(istate
))
2694 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
2696 new_shared_index
= istate
->cache_changed
& SPLIT_INDEX_ORDERED
;
2698 if (new_shared_index
) {
2699 struct tempfile
*temp
;
2702 temp
= mks_tempfile(git_path("sharedindex_XXXXXX"));
2704 oidclr(&si
->base_oid
);
2705 ret
= do_write_locked_index(istate
, lock
, flags
);
2708 ret
= write_shared_index(istate
, &temp
);
2710 saved_errno
= errno
;
2711 if (is_tempfile_active(temp
))
2712 delete_tempfile(&temp
);
2713 errno
= saved_errno
;
2719 ret
= write_split_index(istate
, lock
, flags
);
2721 /* Freshen the shared index only if the split-index was written */
2722 if (!ret
&& !new_shared_index
) {
2723 const char *shared_index
= git_path("sharedindex.%s",
2724 oid_to_hex(&si
->base_oid
));
2725 freshen_shared_index(shared_index
, 1);
2729 if (flags
& COMMIT_LOCK
)
2730 rollback_lock_file(lock
);
2735 * Read the index file that is potentially unmerged into given
2736 * index_state, dropping any unmerged entries. Returns true if
2737 * the index is unmerged. Callers who want to refuse to work
2738 * from an unmerged state can call this and check its return value,
2739 * instead of calling read_cache().
2741 int read_index_unmerged(struct index_state
*istate
)
2747 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2748 struct cache_entry
*ce
= istate
->cache
[i
];
2749 struct cache_entry
*new_ce
;
2755 len
= ce_namelen(ce
);
2756 new_ce
= make_empty_cache_entry(istate
, len
);
2757 memcpy(new_ce
->name
, ce
->name
, len
);
2758 new_ce
->ce_flags
= create_ce_flags(0) | CE_CONFLICTED
;
2759 new_ce
->ce_namelen
= len
;
2760 new_ce
->ce_mode
= ce
->ce_mode
;
2761 if (add_index_entry(istate
, new_ce
, 0))
2762 return error("%s: cannot drop to stage #0",
2769 * Returns 1 if the path is an "other" path with respect to
2770 * the index; that is, the path is not mentioned in the index at all,
2771 * either as a file, a directory with some files in the index,
2772 * or as an unmerged entry.
2774 * We helpfully remove a trailing "/" from directories so that
2775 * the output of read_directory can be used as-is.
2777 int index_name_is_other(const struct index_state
*istate
, const char *name
,
2781 if (namelen
&& name
[namelen
- 1] == '/')
2783 pos
= index_name_pos(istate
, name
, namelen
);
2785 return 0; /* exact match */
2787 if (pos
< istate
->cache_nr
) {
2788 struct cache_entry
*ce
= istate
->cache
[pos
];
2789 if (ce_namelen(ce
) == namelen
&&
2790 !memcmp(ce
->name
, name
, namelen
))
2791 return 0; /* Yup, this one exists unmerged */
2796 void *read_blob_data_from_index(const struct index_state
*istate
,
2797 const char *path
, unsigned long *size
)
2801 enum object_type type
;
2805 pos
= index_name_pos(istate
, path
, len
);
2808 * We might be in the middle of a merge, in which
2809 * case we would read stage #2 (ours).
2813 (pos
< 0 && i
< istate
->cache_nr
&&
2814 !strcmp(istate
->cache
[i
]->name
, path
));
2816 if (ce_stage(istate
->cache
[i
]) == 2)
2821 data
= read_object_file(&istate
->cache
[pos
]->oid
, &type
, &sz
);
2822 if (!data
|| type
!= OBJ_BLOB
) {
2831 void stat_validity_clear(struct stat_validity
*sv
)
2833 FREE_AND_NULL(sv
->sd
);
2836 int stat_validity_check(struct stat_validity
*sv
, const char *path
)
2840 if (stat(path
, &st
) < 0)
2841 return sv
->sd
== NULL
;
2844 return S_ISREG(st
.st_mode
) && !match_stat_data(sv
->sd
, &st
);
2847 void stat_validity_update(struct stat_validity
*sv
, int fd
)
2851 if (fstat(fd
, &st
) < 0 || !S_ISREG(st
.st_mode
))
2852 stat_validity_clear(sv
);
2855 sv
->sd
= xcalloc(1, sizeof(struct stat_data
));
2856 fill_stat_data(sv
->sd
, &st
);
2860 void move_index_extensions(struct index_state
*dst
, struct index_state
*src
)
2862 dst
->untracked
= src
->untracked
;
2863 src
->untracked
= NULL
;
2866 struct cache_entry
*dup_cache_entry(const struct cache_entry
*ce
,
2867 struct index_state
*istate
)
2869 unsigned int size
= ce_size(ce
);
2870 int mem_pool_allocated
;
2871 struct cache_entry
*new_entry
= make_empty_cache_entry(istate
, ce_namelen(ce
));
2872 mem_pool_allocated
= new_entry
->mem_pool_allocated
;
2874 memcpy(new_entry
, ce
, size
);
2875 new_entry
->mem_pool_allocated
= mem_pool_allocated
;
2879 void discard_cache_entry(struct cache_entry
*ce
)
2881 if (ce
&& ce
->mem_pool_allocated
)