2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
12 #include "cache-tree.h"
15 #include "object-store.h"
19 #include "resolve-undo.h"
22 #include "split-index.h"
24 #include "fsmonitor.h"
25 #include "thread-utils.h"
28 /* Mask for the name length in ce_flags in the on-disk index */
30 #define CE_NAMEMASK (0x0fff)
34 * The first letter should be 'A'..'Z' for extensions that are not
35 * necessary for a correct operation (i.e. optimization data).
36 * When new extensions are added that _needs_ to be understood in
37 * order to correctly interpret the index file, pick character that
38 * is outside the range, to cause the reader to abort.
41 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
42 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
43 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
44 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
45 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
46 #define CACHE_EXT_FSMONITOR 0x46534D4E /* "FSMN" */
47 #define CACHE_EXT_ENDOFINDEXENTRIES 0x454F4945 /* "EOIE" */
48 #define CACHE_EXT_INDEXENTRYOFFSETTABLE 0x49454F54 /* "IEOT" */
50 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
51 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
52 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
53 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED | FSMONITOR_CHANGED)
57 * This is an estimate of the pathname length in the index. We use
58 * this for V4 index files to guess the un-deltafied size of the index
59 * in memory because of pathname deltafication. This is not required
60 * for V2/V3 index formats because their pathnames are not compressed.
61 * If the initial amount of memory set aside is not sufficient, the
62 * mem pool will allocate extra memory.
64 #define CACHE_ENTRY_PATH_LENGTH 80
66 static inline struct cache_entry
*mem_pool__ce_alloc(struct mem_pool
*mem_pool
, size_t len
)
68 struct cache_entry
*ce
;
69 ce
= mem_pool_alloc(mem_pool
, cache_entry_size(len
));
70 ce
->mem_pool_allocated
= 1;
74 static inline struct cache_entry
*mem_pool__ce_calloc(struct mem_pool
*mem_pool
, size_t len
)
76 struct cache_entry
* ce
;
77 ce
= mem_pool_calloc(mem_pool
, 1, cache_entry_size(len
));
78 ce
->mem_pool_allocated
= 1;
82 static struct mem_pool
*find_mem_pool(struct index_state
*istate
)
84 struct mem_pool
**pool_ptr
;
86 if (istate
->split_index
&& istate
->split_index
->base
)
87 pool_ptr
= &istate
->split_index
->base
->ce_mem_pool
;
89 pool_ptr
= &istate
->ce_mem_pool
;
92 mem_pool_init(pool_ptr
, 0);
97 static const char *alternate_index_output
;
99 static void set_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
101 istate
->cache
[nr
] = ce
;
102 add_name_hash(istate
, ce
);
105 static void replace_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
107 struct cache_entry
*old
= istate
->cache
[nr
];
109 replace_index_entry_in_base(istate
, old
, ce
);
110 remove_name_hash(istate
, old
);
111 discard_cache_entry(old
);
112 ce
->ce_flags
&= ~CE_HASHED
;
113 set_index_entry(istate
, nr
, ce
);
114 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
115 mark_fsmonitor_invalid(istate
, ce
);
116 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
119 void rename_index_entry_at(struct index_state
*istate
, int nr
, const char *new_name
)
121 struct cache_entry
*old_entry
= istate
->cache
[nr
], *new_entry
;
122 int namelen
= strlen(new_name
);
124 new_entry
= make_empty_cache_entry(istate
, namelen
);
125 copy_cache_entry(new_entry
, old_entry
);
126 new_entry
->ce_flags
&= ~CE_HASHED
;
127 new_entry
->ce_namelen
= namelen
;
128 new_entry
->index
= 0;
129 memcpy(new_entry
->name
, new_name
, namelen
+ 1);
131 cache_tree_invalidate_path(istate
, old_entry
->name
);
132 untracked_cache_remove_from_index(istate
, old_entry
->name
);
133 remove_index_entry_at(istate
, nr
);
134 add_index_entry(istate
, new_entry
, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
137 void fill_stat_data(struct stat_data
*sd
, struct stat
*st
)
139 sd
->sd_ctime
.sec
= (unsigned int)st
->st_ctime
;
140 sd
->sd_mtime
.sec
= (unsigned int)st
->st_mtime
;
141 sd
->sd_ctime
.nsec
= ST_CTIME_NSEC(*st
);
142 sd
->sd_mtime
.nsec
= ST_MTIME_NSEC(*st
);
143 sd
->sd_dev
= st
->st_dev
;
144 sd
->sd_ino
= st
->st_ino
;
145 sd
->sd_uid
= st
->st_uid
;
146 sd
->sd_gid
= st
->st_gid
;
147 sd
->sd_size
= st
->st_size
;
150 int match_stat_data(const struct stat_data
*sd
, struct stat
*st
)
154 if (sd
->sd_mtime
.sec
!= (unsigned int)st
->st_mtime
)
155 changed
|= MTIME_CHANGED
;
156 if (trust_ctime
&& check_stat
&&
157 sd
->sd_ctime
.sec
!= (unsigned int)st
->st_ctime
)
158 changed
|= CTIME_CHANGED
;
161 if (check_stat
&& sd
->sd_mtime
.nsec
!= ST_MTIME_NSEC(*st
))
162 changed
|= MTIME_CHANGED
;
163 if (trust_ctime
&& check_stat
&&
164 sd
->sd_ctime
.nsec
!= ST_CTIME_NSEC(*st
))
165 changed
|= CTIME_CHANGED
;
169 if (sd
->sd_uid
!= (unsigned int) st
->st_uid
||
170 sd
->sd_gid
!= (unsigned int) st
->st_gid
)
171 changed
|= OWNER_CHANGED
;
172 if (sd
->sd_ino
!= (unsigned int) st
->st_ino
)
173 changed
|= INODE_CHANGED
;
178 * st_dev breaks on network filesystems where different
179 * clients will have different views of what "device"
180 * the filesystem is on
182 if (check_stat
&& sd
->sd_dev
!= (unsigned int) st
->st_dev
)
183 changed
|= INODE_CHANGED
;
186 if (sd
->sd_size
!= (unsigned int) st
->st_size
)
187 changed
|= DATA_CHANGED
;
193 * This only updates the "non-critical" parts of the directory
194 * cache, ie the parts that aren't tracked by GIT, and only used
195 * to validate the cache.
197 void fill_stat_cache_info(struct cache_entry
*ce
, struct stat
*st
)
199 fill_stat_data(&ce
->ce_stat_data
, st
);
201 if (assume_unchanged
)
202 ce
->ce_flags
|= CE_VALID
;
204 if (S_ISREG(st
->st_mode
)) {
205 ce_mark_uptodate(ce
);
206 mark_fsmonitor_valid(ce
);
210 static int ce_compare_data(struct index_state
*istate
,
211 const struct cache_entry
*ce
,
215 int fd
= git_open_cloexec(ce
->name
, O_RDONLY
);
218 struct object_id oid
;
219 if (!index_fd(istate
, &oid
, fd
, st
, OBJ_BLOB
, ce
->name
, 0))
220 match
= !oideq(&oid
, &ce
->oid
);
221 /* index_fd() closed the file descriptor already */
226 static int ce_compare_link(const struct cache_entry
*ce
, size_t expected_size
)
231 enum object_type type
;
232 struct strbuf sb
= STRBUF_INIT
;
234 if (strbuf_readlink(&sb
, ce
->name
, expected_size
))
237 buffer
= read_object_file(&ce
->oid
, &type
, &size
);
240 match
= memcmp(buffer
, sb
.buf
, size
);
247 static int ce_compare_gitlink(const struct cache_entry
*ce
)
249 struct object_id oid
;
252 * We don't actually require that the .git directory
253 * under GITLINK directory be a valid git directory. It
254 * might even be missing (in case nobody populated that
257 * If so, we consider it always to match.
259 if (resolve_gitlink_ref(ce
->name
, "HEAD", &oid
) < 0)
261 return !oideq(&oid
, &ce
->oid
);
264 static int ce_modified_check_fs(struct index_state
*istate
,
265 const struct cache_entry
*ce
,
268 switch (st
->st_mode
& S_IFMT
) {
270 if (ce_compare_data(istate
, ce
, st
))
274 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
278 if (S_ISGITLINK(ce
->ce_mode
))
279 return ce_compare_gitlink(ce
) ? DATA_CHANGED
: 0;
280 /* else fallthrough */
287 static int ce_match_stat_basic(const struct cache_entry
*ce
, struct stat
*st
)
289 unsigned int changed
= 0;
291 if (ce
->ce_flags
& CE_REMOVE
)
292 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
294 switch (ce
->ce_mode
& S_IFMT
) {
296 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
297 /* We consider only the owner x bit to be relevant for
300 if (trust_executable_bit
&&
301 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
302 changed
|= MODE_CHANGED
;
305 if (!S_ISLNK(st
->st_mode
) &&
306 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
307 changed
|= TYPE_CHANGED
;
310 /* We ignore most of the st_xxx fields for gitlinks */
311 if (!S_ISDIR(st
->st_mode
))
312 changed
|= TYPE_CHANGED
;
313 else if (ce_compare_gitlink(ce
))
314 changed
|= DATA_CHANGED
;
317 BUG("unsupported ce_mode: %o", ce
->ce_mode
);
320 changed
|= match_stat_data(&ce
->ce_stat_data
, st
);
322 /* Racily smudged entry? */
323 if (!ce
->ce_stat_data
.sd_size
) {
324 if (!is_empty_blob_sha1(ce
->oid
.hash
))
325 changed
|= DATA_CHANGED
;
331 static int is_racy_stat(const struct index_state
*istate
,
332 const struct stat_data
*sd
)
334 return (istate
->timestamp
.sec
&&
336 /* nanosecond timestamped files can also be racy! */
337 (istate
->timestamp
.sec
< sd
->sd_mtime
.sec
||
338 (istate
->timestamp
.sec
== sd
->sd_mtime
.sec
&&
339 istate
->timestamp
.nsec
<= sd
->sd_mtime
.nsec
))
341 istate
->timestamp
.sec
<= sd
->sd_mtime
.sec
346 int is_racy_timestamp(const struct index_state
*istate
,
347 const struct cache_entry
*ce
)
349 return (!S_ISGITLINK(ce
->ce_mode
) &&
350 is_racy_stat(istate
, &ce
->ce_stat_data
));
353 int match_stat_data_racy(const struct index_state
*istate
,
354 const struct stat_data
*sd
, struct stat
*st
)
356 if (is_racy_stat(istate
, sd
))
357 return MTIME_CHANGED
;
358 return match_stat_data(sd
, st
);
361 int ie_match_stat(struct index_state
*istate
,
362 const struct cache_entry
*ce
, struct stat
*st
,
363 unsigned int options
)
365 unsigned int changed
;
366 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
367 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
368 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
369 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
371 if (!ignore_fsmonitor
)
372 refresh_fsmonitor(istate
);
374 * If it's marked as always valid in the index, it's
375 * valid whatever the checked-out copy says.
377 * skip-worktree has the same effect with higher precedence
379 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
))
381 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
383 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
))
387 * Intent-to-add entries have not been added, so the index entry
388 * by definition never matches what is in the work tree until it
389 * actually gets added.
391 if (ce_intent_to_add(ce
))
392 return DATA_CHANGED
| TYPE_CHANGED
| MODE_CHANGED
;
394 changed
= ce_match_stat_basic(ce
, st
);
397 * Within 1 second of this sequence:
398 * echo xyzzy >file && git-update-index --add file
399 * running this command:
401 * would give a falsely clean cache entry. The mtime and
402 * length match the cache, and other stat fields do not change.
404 * We could detect this at update-index time (the cache entry
405 * being registered/updated records the same time as "now")
406 * and delay the return from git-update-index, but that would
407 * effectively mean we can make at most one commit per second,
408 * which is not acceptable. Instead, we check cache entries
409 * whose mtime are the same as the index file timestamp more
410 * carefully than others.
412 if (!changed
&& is_racy_timestamp(istate
, ce
)) {
413 if (assume_racy_is_modified
)
414 changed
|= DATA_CHANGED
;
416 changed
|= ce_modified_check_fs(istate
, ce
, st
);
422 int ie_modified(struct index_state
*istate
,
423 const struct cache_entry
*ce
,
424 struct stat
*st
, unsigned int options
)
426 int changed
, changed_fs
;
428 changed
= ie_match_stat(istate
, ce
, st
, options
);
432 * If the mode or type has changed, there's no point in trying
433 * to refresh the entry - it's not going to match
435 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
439 * Immediately after read-tree or update-index --cacheinfo,
440 * the length field is zero, as we have never even read the
441 * lstat(2) information once, and we cannot trust DATA_CHANGED
442 * returned by ie_match_stat() which in turn was returned by
443 * ce_match_stat_basic() to signal that the filesize of the
444 * blob changed. We have to actually go to the filesystem to
445 * see if the contents match, and if so, should answer "unchanged".
447 * The logic does not apply to gitlinks, as ce_match_stat_basic()
448 * already has checked the actual HEAD from the filesystem in the
449 * subproject. If ie_match_stat() already said it is different,
450 * then we know it is.
452 if ((changed
& DATA_CHANGED
) &&
453 (S_ISGITLINK(ce
->ce_mode
) || ce
->ce_stat_data
.sd_size
!= 0))
456 changed_fs
= ce_modified_check_fs(istate
, ce
, st
);
458 return changed
| changed_fs
;
462 int base_name_compare(const char *name1
, int len1
, int mode1
,
463 const char *name2
, int len2
, int mode2
)
465 unsigned char c1
, c2
;
466 int len
= len1
< len2
? len1
: len2
;
469 cmp
= memcmp(name1
, name2
, len
);
474 if (!c1
&& S_ISDIR(mode1
))
476 if (!c2
&& S_ISDIR(mode2
))
478 return (c1
< c2
) ? -1 : (c1
> c2
) ? 1 : 0;
482 * df_name_compare() is identical to base_name_compare(), except it
483 * compares conflicting directory/file entries as equal. Note that
484 * while a directory name compares as equal to a regular file, they
485 * then individually compare _differently_ to a filename that has
486 * a dot after the basename (because '\0' < '.' < '/').
488 * This is used by routines that want to traverse the git namespace
489 * but then handle conflicting entries together when possible.
491 int df_name_compare(const char *name1
, int len1
, int mode1
,
492 const char *name2
, int len2
, int mode2
)
494 int len
= len1
< len2
? len1
: len2
, cmp
;
495 unsigned char c1
, c2
;
497 cmp
= memcmp(name1
, name2
, len
);
500 /* Directories and files compare equal (same length, same name) */
504 if (!c1
&& S_ISDIR(mode1
))
507 if (!c2
&& S_ISDIR(mode2
))
509 if (c1
== '/' && !c2
)
511 if (c2
== '/' && !c1
)
516 int name_compare(const char *name1
, size_t len1
, const char *name2
, size_t len2
)
518 size_t min_len
= (len1
< len2
) ? len1
: len2
;
519 int cmp
= memcmp(name1
, name2
, min_len
);
529 int cache_name_stage_compare(const char *name1
, int len1
, int stage1
, const char *name2
, int len2
, int stage2
)
533 cmp
= name_compare(name1
, len1
, name2
, len2
);
544 static int index_name_stage_pos(const struct index_state
*istate
, const char *name
, int namelen
, int stage
)
549 last
= istate
->cache_nr
;
550 while (last
> first
) {
551 int next
= (last
+ first
) >> 1;
552 struct cache_entry
*ce
= istate
->cache
[next
];
553 int cmp
= cache_name_stage_compare(name
, namelen
, stage
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
565 int index_name_pos(const struct index_state
*istate
, const char *name
, int namelen
)
567 return index_name_stage_pos(istate
, name
, namelen
, 0);
570 int remove_index_entry_at(struct index_state
*istate
, int pos
)
572 struct cache_entry
*ce
= istate
->cache
[pos
];
574 record_resolve_undo(istate
, ce
);
575 remove_name_hash(istate
, ce
);
576 save_or_free_index_entry(istate
, ce
);
577 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
579 if (pos
>= istate
->cache_nr
)
581 MOVE_ARRAY(istate
->cache
+ pos
, istate
->cache
+ pos
+ 1,
582 istate
->cache_nr
- pos
);
587 * Remove all cache entries marked for removal, that is where
588 * CE_REMOVE is set in ce_flags. This is much more effective than
589 * calling remove_index_entry_at() for each entry to be removed.
591 void remove_marked_cache_entries(struct index_state
*istate
)
593 struct cache_entry
**ce_array
= istate
->cache
;
596 for (i
= j
= 0; i
< istate
->cache_nr
; i
++) {
597 if (ce_array
[i
]->ce_flags
& CE_REMOVE
) {
598 remove_name_hash(istate
, ce_array
[i
]);
599 save_or_free_index_entry(istate
, ce_array
[i
]);
602 ce_array
[j
++] = ce_array
[i
];
604 if (j
== istate
->cache_nr
)
606 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
607 istate
->cache_nr
= j
;
610 int remove_file_from_index(struct index_state
*istate
, const char *path
)
612 int pos
= index_name_pos(istate
, path
, strlen(path
));
615 cache_tree_invalidate_path(istate
, path
);
616 untracked_cache_remove_from_index(istate
, path
);
617 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
618 remove_index_entry_at(istate
, pos
);
622 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
624 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
627 static int index_name_pos_also_unmerged(struct index_state
*istate
,
628 const char *path
, int namelen
)
630 int pos
= index_name_pos(istate
, path
, namelen
);
631 struct cache_entry
*ce
;
636 /* maybe unmerged? */
638 if (pos
>= istate
->cache_nr
||
639 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
642 /* order of preference: stage 2, 1, 3 */
643 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
644 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
645 !compare_name(ce
, path
, namelen
))
650 static int different_name(struct cache_entry
*ce
, struct cache_entry
*alias
)
652 int len
= ce_namelen(ce
);
653 return ce_namelen(alias
) != len
|| memcmp(ce
->name
, alias
->name
, len
);
657 * If we add a filename that aliases in the cache, we will use the
658 * name that we already have - but we don't want to update the same
659 * alias twice, because that implies that there were actually two
660 * different files with aliasing names!
662 * So we use the CE_ADDED flag to verify that the alias was an old
663 * one before we accept it as
665 static struct cache_entry
*create_alias_ce(struct index_state
*istate
,
666 struct cache_entry
*ce
,
667 struct cache_entry
*alias
)
670 struct cache_entry
*new_entry
;
672 if (alias
->ce_flags
& CE_ADDED
)
673 die(_("will not add file alias '%s' ('%s' already exists in index)"),
674 ce
->name
, alias
->name
);
676 /* Ok, create the new entry using the name of the existing alias */
677 len
= ce_namelen(alias
);
678 new_entry
= make_empty_cache_entry(istate
, len
);
679 memcpy(new_entry
->name
, alias
->name
, len
);
680 copy_cache_entry(new_entry
, ce
);
681 save_or_free_index_entry(istate
, ce
);
685 void set_object_name_for_intent_to_add_entry(struct cache_entry
*ce
)
687 struct object_id oid
;
688 if (write_object_file("", 0, blob_type
, &oid
))
689 die(_("cannot create an empty blob in the object database"));
690 oidcpy(&ce
->oid
, &oid
);
693 int add_to_index(struct index_state
*istate
, const char *path
, struct stat
*st
, int flags
)
695 int namelen
, was_same
;
696 mode_t st_mode
= st
->st_mode
;
697 struct cache_entry
*ce
, *alias
= NULL
;
698 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_IGNORE_SKIP_WORKTREE
|CE_MATCH_RACY_IS_DIRTY
;
699 int verbose
= flags
& (ADD_CACHE_VERBOSE
| ADD_CACHE_PRETEND
);
700 int pretend
= flags
& ADD_CACHE_PRETEND
;
701 int intent_only
= flags
& ADD_CACHE_INTENT
;
702 int add_option
= (ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
|
703 (intent_only
? ADD_CACHE_NEW_ONLY
: 0));
704 int hash_flags
= HASH_WRITE_OBJECT
;
706 if (flags
& ADD_CACHE_RENORMALIZE
)
707 hash_flags
|= HASH_RENORMALIZE
;
709 if (!S_ISREG(st_mode
) && !S_ISLNK(st_mode
) && !S_ISDIR(st_mode
))
710 return error(_("%s: can only add regular files, symbolic links or git-directories"), path
);
712 namelen
= strlen(path
);
713 if (S_ISDIR(st_mode
)) {
714 while (namelen
&& path
[namelen
-1] == '/')
717 ce
= make_empty_cache_entry(istate
, namelen
);
718 memcpy(ce
->name
, path
, namelen
);
719 ce
->ce_namelen
= namelen
;
721 fill_stat_cache_info(ce
, st
);
723 ce
->ce_flags
|= CE_INTENT_TO_ADD
;
726 if (trust_executable_bit
&& has_symlinks
) {
727 ce
->ce_mode
= create_ce_mode(st_mode
);
729 /* If there is an existing entry, pick the mode bits and type
730 * from it, otherwise assume unexecutable regular file.
732 struct cache_entry
*ent
;
733 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
735 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
736 ce
->ce_mode
= ce_mode_from_stat(ent
, st_mode
);
739 /* When core.ignorecase=true, determine if a directory of the same name but differing
740 * case already exists within the Git repository. If it does, ensure the directory
741 * case of the file being added to the repository matches (is folded into) the existing
742 * entry's directory case.
745 adjust_dirname_case(istate
, ce
->name
);
747 if (!(flags
& ADD_CACHE_RENORMALIZE
)) {
748 alias
= index_file_exists(istate
, ce
->name
,
749 ce_namelen(ce
), ignore_case
);
752 !ie_match_stat(istate
, alias
, st
, ce_option
)) {
753 /* Nothing changed, really */
754 if (!S_ISGITLINK(alias
->ce_mode
))
755 ce_mark_uptodate(alias
);
756 alias
->ce_flags
|= CE_ADDED
;
758 discard_cache_entry(ce
);
763 if (index_path(istate
, &ce
->oid
, path
, st
, hash_flags
)) {
764 discard_cache_entry(ce
);
765 return error(_("unable to index file '%s'"), path
);
768 set_object_name_for_intent_to_add_entry(ce
);
770 if (ignore_case
&& alias
&& different_name(ce
, alias
))
771 ce
= create_alias_ce(istate
, ce
, alias
);
772 ce
->ce_flags
|= CE_ADDED
;
774 /* It was suspected to be racily clean, but it turns out to be Ok */
777 oideq(&alias
->oid
, &ce
->oid
) &&
778 ce
->ce_mode
== alias
->ce_mode
);
781 discard_cache_entry(ce
);
782 else if (add_index_entry(istate
, ce
, add_option
)) {
783 discard_cache_entry(ce
);
784 return error(_("unable to add '%s' to index"), path
);
786 if (verbose
&& !was_same
)
787 printf("add '%s'\n", path
);
791 int add_file_to_index(struct index_state
*istate
, const char *path
, int flags
)
794 if (lstat(path
, &st
))
795 die_errno(_("unable to stat '%s'"), path
);
796 return add_to_index(istate
, path
, &st
, flags
);
799 struct cache_entry
*make_empty_cache_entry(struct index_state
*istate
, size_t len
)
801 return mem_pool__ce_calloc(find_mem_pool(istate
), len
);
804 struct cache_entry
*make_empty_transient_cache_entry(size_t len
)
806 return xcalloc(1, cache_entry_size(len
));
809 struct cache_entry
*make_cache_entry(struct index_state
*istate
,
811 const struct object_id
*oid
,
814 unsigned int refresh_options
)
816 struct cache_entry
*ce
, *ret
;
819 if (!verify_path(path
, mode
)) {
820 error(_("invalid path '%s'"), path
);
825 ce
= make_empty_cache_entry(istate
, len
);
827 oidcpy(&ce
->oid
, oid
);
828 memcpy(ce
->name
, path
, len
);
829 ce
->ce_flags
= create_ce_flags(stage
);
830 ce
->ce_namelen
= len
;
831 ce
->ce_mode
= create_ce_mode(mode
);
833 ret
= refresh_cache_entry(istate
, ce
, refresh_options
);
835 discard_cache_entry(ce
);
839 struct cache_entry
*make_transient_cache_entry(unsigned int mode
, const struct object_id
*oid
,
840 const char *path
, int stage
)
842 struct cache_entry
*ce
;
845 if (!verify_path(path
, mode
)) {
846 error(_("invalid path '%s'"), path
);
851 ce
= make_empty_transient_cache_entry(len
);
853 oidcpy(&ce
->oid
, oid
);
854 memcpy(ce
->name
, path
, len
);
855 ce
->ce_flags
= create_ce_flags(stage
);
856 ce
->ce_namelen
= len
;
857 ce
->ce_mode
= create_ce_mode(mode
);
863 * Chmod an index entry with either +x or -x.
865 * Returns -1 if the chmod for the particular cache entry failed (if it's
866 * not a regular file), -2 if an invalid flip argument is passed in, 0
869 int chmod_index_entry(struct index_state
*istate
, struct cache_entry
*ce
,
872 if (!S_ISREG(ce
->ce_mode
))
879 ce
->ce_mode
&= ~0111;
884 cache_tree_invalidate_path(istate
, ce
->name
);
885 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
886 mark_fsmonitor_invalid(istate
, ce
);
887 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
892 int ce_same_name(const struct cache_entry
*a
, const struct cache_entry
*b
)
894 int len
= ce_namelen(a
);
895 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
899 * We fundamentally don't like some paths: we don't want
900 * dot or dot-dot anywhere, and for obvious reasons don't
901 * want to recurse into ".git" either.
903 * Also, we don't want double slashes or slashes at the
904 * end that can make pathnames ambiguous.
906 static int verify_dotfile(const char *rest
, unsigned mode
)
909 * The first character was '.', but that
910 * has already been discarded, we now test
914 /* "." is not allowed */
915 if (*rest
== '\0' || is_dir_sep(*rest
))
920 * ".git" followed by NUL or slash is bad. Note that we match
921 * case-insensitively here, even if ignore_case is not set.
922 * This outlaws ".GIT" everywhere out of an abundance of caution,
923 * since there's really no good reason to allow it.
925 * Once we've seen ".git", we can also find ".gitmodules", etc (also
926 * case-insensitively).
930 if (rest
[1] != 'i' && rest
[1] != 'I')
932 if (rest
[2] != 't' && rest
[2] != 'T')
934 if (rest
[3] == '\0' || is_dir_sep(rest
[3]))
938 if (skip_iprefix(rest
, "modules", &rest
) &&
939 (*rest
== '\0' || is_dir_sep(*rest
)))
944 if (rest
[1] == '\0' || is_dir_sep(rest
[1]))
950 int verify_path(const char *path
, unsigned mode
)
954 if (has_dos_drive_prefix(path
))
964 if (is_hfs_dotgit(path
))
967 if (is_hfs_dotgitmodules(path
))
972 if (is_ntfs_dotgit(path
))
975 if (is_ntfs_dotgitmodules(path
))
981 if ((c
== '.' && !verify_dotfile(path
, mode
)) ||
982 is_dir_sep(c
) || c
== '\0')
990 * Do we have another file that has the beginning components being a
991 * proper superset of the name we're trying to add?
993 static int has_file_name(struct index_state
*istate
,
994 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
997 int len
= ce_namelen(ce
);
998 int stage
= ce_stage(ce
);
999 const char *name
= ce
->name
;
1001 while (pos
< istate
->cache_nr
) {
1002 struct cache_entry
*p
= istate
->cache
[pos
++];
1004 if (len
>= ce_namelen(p
))
1006 if (memcmp(name
, p
->name
, len
))
1008 if (ce_stage(p
) != stage
)
1010 if (p
->name
[len
] != '/')
1012 if (p
->ce_flags
& CE_REMOVE
)
1017 remove_index_entry_at(istate
, --pos
);
1024 * Like strcmp(), but also return the offset of the first change.
1025 * If strings are equal, return the length.
1027 int strcmp_offset(const char *s1
, const char *s2
, size_t *first_change
)
1032 return strcmp(s1
, s2
);
1034 for (k
= 0; s1
[k
] == s2
[k
]; k
++)
1039 return (unsigned char)s1
[k
] - (unsigned char)s2
[k
];
1043 * Do we have another file with a pathname that is a proper
1044 * subset of the name we're trying to add?
1046 * That is, is there another file in the index with a path
1047 * that matches a sub-directory in the given entry?
1049 static int has_dir_name(struct index_state
*istate
,
1050 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
1053 int stage
= ce_stage(ce
);
1054 const char *name
= ce
->name
;
1055 const char *slash
= name
+ ce_namelen(ce
);
1060 * We are frequently called during an iteration on a sorted
1061 * list of pathnames and while building a new index. Therefore,
1062 * there is a high probability that this entry will eventually
1063 * be appended to the index, rather than inserted in the middle.
1064 * If we can confirm that, we can avoid binary searches on the
1065 * components of the pathname.
1067 * Compare the entry's full path with the last path in the index.
1069 if (istate
->cache_nr
> 0) {
1070 cmp_last
= strcmp_offset(name
,
1071 istate
->cache
[istate
->cache_nr
- 1]->name
,
1074 if (len_eq_last
== 0) {
1076 * The entry sorts AFTER the last one in the
1077 * index and their paths have no common prefix,
1078 * so there cannot be a F/D conflict.
1083 * The entry sorts AFTER the last one in the
1084 * index, but has a common prefix. Fall through
1085 * to the loop below to disect the entry's path
1086 * and see where the difference is.
1089 } else if (cmp_last
== 0) {
1091 * The entry exactly matches the last one in the
1092 * index, but because of multiple stage and CE_REMOVE
1093 * items, we fall through and let the regular search
1103 if (*--slash
== '/')
1105 if (slash
<= ce
->name
)
1112 * (len + 1) is a directory boundary (including
1113 * the trailing slash). And since the loop is
1114 * decrementing "slash", the first iteration is
1115 * the longest directory prefix; subsequent
1116 * iterations consider parent directories.
1119 if (len
+ 1 <= len_eq_last
) {
1121 * The directory prefix (including the trailing
1122 * slash) also appears as a prefix in the last
1123 * entry, so the remainder cannot collide (because
1124 * strcmp said the whole path was greater).
1129 * LT: last: xxx/file_A
1135 if (len
> len_eq_last
) {
1137 * This part of the directory prefix (excluding
1138 * the trailing slash) is longer than the known
1139 * equal portions, so this sub-directory cannot
1140 * collide with a file.
1148 if (istate
->cache_nr
> 0 &&
1149 ce_namelen(istate
->cache
[istate
->cache_nr
- 1]) > len
) {
1151 * The directory prefix lines up with part of
1152 * a longer file or directory name, but sorts
1153 * after it, so this sub-directory cannot
1154 * collide with a file.
1156 * last: xxx/yy-file (because '-' sorts before '/')
1163 * This is a possible collision. Fall through and
1164 * let the regular search code handle it.
1171 pos
= index_name_stage_pos(istate
, name
, len
, stage
);
1174 * Found one, but not so fast. This could
1175 * be a marker that says "I was here, but
1176 * I am being removed". Such an entry is
1177 * not a part of the resulting tree, and
1178 * it is Ok to have a directory at the same
1181 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
1185 remove_index_entry_at(istate
, pos
);
1193 * Trivial optimization: if we find an entry that
1194 * already matches the sub-directory, then we know
1195 * we're ok, and we can exit.
1197 while (pos
< istate
->cache_nr
) {
1198 struct cache_entry
*p
= istate
->cache
[pos
];
1199 if ((ce_namelen(p
) <= len
) ||
1200 (p
->name
[len
] != '/') ||
1201 memcmp(p
->name
, name
, len
))
1202 break; /* not our subdirectory */
1203 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
1205 * p is at the same stage as our entry, and
1206 * is a subdirectory of what we are looking
1207 * at, so we cannot have conflicts at our
1208 * level or anything shorter.
1217 /* We may be in a situation where we already have path/file and path
1218 * is being added, or we already have path and path/file is being
1219 * added. Either one would result in a nonsense tree that has path
1220 * twice when git-write-tree tries to write it out. Prevent it.
1222 * If ok-to-replace is specified, we remove the conflicting entries
1223 * from the cache so the caller should recompute the insert position.
1224 * When this happens, we return non-zero.
1226 static int check_file_directory_conflict(struct index_state
*istate
,
1227 const struct cache_entry
*ce
,
1228 int pos
, int ok_to_replace
)
1233 * When ce is an "I am going away" entry, we allow it to be added
1235 if (ce
->ce_flags
& CE_REMOVE
)
1239 * We check if the path is a sub-path of a subsequent pathname
1240 * first, since removing those will not change the position
1243 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
1246 * Then check if the path might have a clashing sub-directory
1249 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
1252 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1255 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
1256 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
1257 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
1258 int new_only
= option
& ADD_CACHE_NEW_ONLY
;
1260 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1261 cache_tree_invalidate_path(istate
, ce
->name
);
1264 * If this entry's path sorts after the last entry in the index,
1265 * we can avoid searching for it.
1267 if (istate
->cache_nr
> 0 &&
1268 strcmp(ce
->name
, istate
->cache
[istate
->cache_nr
- 1]->name
) > 0)
1269 pos
= -istate
->cache_nr
- 1;
1271 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1273 /* existing match? Just replace it. */
1276 replace_index_entry(istate
, pos
, ce
);
1281 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1282 untracked_cache_add_to_index(istate
, ce
->name
);
1285 * Inserting a merged entry ("stage 0") into the index
1286 * will always replace all non-merged entries..
1288 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
1289 while (ce_same_name(istate
->cache
[pos
], ce
)) {
1291 if (!remove_index_entry_at(istate
, pos
))
1298 if (!verify_path(ce
->name
, ce
->ce_mode
))
1299 return error(_("invalid path '%s'"), ce
->name
);
1301 if (!skip_df_check
&&
1302 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
1304 return error(_("'%s' appears as both a file and as a directory"),
1306 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1312 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1316 if (option
& ADD_CACHE_JUST_APPEND
)
1317 pos
= istate
->cache_nr
;
1320 ret
= add_index_entry_with_check(istate
, ce
, option
);
1326 /* Make sure the array is big enough .. */
1327 ALLOC_GROW(istate
->cache
, istate
->cache_nr
+ 1, istate
->cache_alloc
);
1331 if (istate
->cache_nr
> pos
+ 1)
1332 MOVE_ARRAY(istate
->cache
+ pos
+ 1, istate
->cache
+ pos
,
1333 istate
->cache_nr
- pos
- 1);
1334 set_index_entry(istate
, pos
, ce
);
1335 istate
->cache_changed
|= CE_ENTRY_ADDED
;
1340 * "refresh" does not calculate a new sha1 file or bring the
1341 * cache up-to-date for mode/content changes. But what it
1342 * _does_ do is to "re-match" the stat information of a file
1343 * with the cache, so that you can refresh the cache for a
1344 * file that hasn't been changed but where the stat entry is
1347 * For example, you'd want to do this after doing a "git-read-tree",
1348 * to link up the stat cache details with the proper files.
1350 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
1351 struct cache_entry
*ce
,
1352 unsigned int options
, int *err
,
1356 struct cache_entry
*updated
;
1358 int refresh
= options
& CE_MATCH_REFRESH
;
1359 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
1360 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
1361 int ignore_missing
= options
& CE_MATCH_IGNORE_MISSING
;
1362 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
1364 if (!refresh
|| ce_uptodate(ce
))
1367 if (!ignore_fsmonitor
)
1368 refresh_fsmonitor(istate
);
1370 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1371 * that the change to the work tree does not matter and told
1374 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
)) {
1375 ce_mark_uptodate(ce
);
1378 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
)) {
1379 ce_mark_uptodate(ce
);
1382 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
)) {
1383 ce_mark_uptodate(ce
);
1387 if (has_symlink_leading_path(ce
->name
, ce_namelen(ce
))) {
1395 if (lstat(ce
->name
, &st
) < 0) {
1396 if (ignore_missing
&& errno
== ENOENT
)
1403 changed
= ie_match_stat(istate
, ce
, &st
, options
);
1405 *changed_ret
= changed
;
1408 * The path is unchanged. If we were told to ignore
1409 * valid bit, then we did the actual stat check and
1410 * found that the entry is unmodified. If the entry
1411 * is not marked VALID, this is the place to mark it
1412 * valid again, under "assume unchanged" mode.
1414 if (ignore_valid
&& assume_unchanged
&&
1415 !(ce
->ce_flags
& CE_VALID
))
1416 ; /* mark this one VALID again */
1419 * We do not mark the index itself "modified"
1420 * because CE_UPTODATE flag is in-core only;
1421 * we are not going to write this change out.
1423 if (!S_ISGITLINK(ce
->ce_mode
)) {
1424 ce_mark_uptodate(ce
);
1425 mark_fsmonitor_valid(ce
);
1431 if (ie_modified(istate
, ce
, &st
, options
)) {
1437 updated
= make_empty_cache_entry(istate
, ce_namelen(ce
));
1438 copy_cache_entry(updated
, ce
);
1439 memcpy(updated
->name
, ce
->name
, ce
->ce_namelen
+ 1);
1440 fill_stat_cache_info(updated
, &st
);
1442 * If ignore_valid is not set, we should leave CE_VALID bit
1443 * alone. Otherwise, paths marked with --no-assume-unchanged
1444 * (i.e. things to be edited) will reacquire CE_VALID bit
1445 * automatically, which is not really what we want.
1447 if (!ignore_valid
&& assume_unchanged
&&
1448 !(ce
->ce_flags
& CE_VALID
))
1449 updated
->ce_flags
&= ~CE_VALID
;
1451 /* istate->cache_changed is updated in the caller */
1455 static void show_file(const char * fmt
, const char * name
, int in_porcelain
,
1456 int * first
, const char *header_msg
)
1458 if (in_porcelain
&& *first
&& header_msg
) {
1459 printf("%s\n", header_msg
);
1465 int refresh_index(struct index_state
*istate
, unsigned int flags
,
1466 const struct pathspec
*pathspec
,
1467 char *seen
, const char *header_msg
)
1471 int really
= (flags
& REFRESH_REALLY
) != 0;
1472 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
1473 int quiet
= (flags
& REFRESH_QUIET
) != 0;
1474 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
1475 int ignore_submodules
= (flags
& REFRESH_IGNORE_SUBMODULES
) != 0;
1477 int in_porcelain
= (flags
& REFRESH_IN_PORCELAIN
);
1478 unsigned int options
= (CE_MATCH_REFRESH
|
1479 (really
? CE_MATCH_IGNORE_VALID
: 0) |
1480 (not_new
? CE_MATCH_IGNORE_MISSING
: 0));
1481 const char *modified_fmt
;
1482 const char *deleted_fmt
;
1483 const char *typechange_fmt
;
1484 const char *added_fmt
;
1485 const char *unmerged_fmt
;
1486 struct progress
*progress
= NULL
;
1488 if (flags
& REFRESH_PROGRESS
&& isatty(2))
1489 progress
= start_delayed_progress(_("Refresh index"),
1492 trace_performance_enter();
1493 modified_fmt
= in_porcelain
? "M\t%s\n" : "%s: needs update\n";
1494 deleted_fmt
= in_porcelain
? "D\t%s\n" : "%s: needs update\n";
1495 typechange_fmt
= in_porcelain
? "T\t%s\n" : "%s: needs update\n";
1496 added_fmt
= in_porcelain
? "A\t%s\n" : "%s: needs update\n";
1497 unmerged_fmt
= in_porcelain
? "U\t%s\n" : "%s: needs merge\n";
1499 * Use the multi-threaded preload_index() to refresh most of the
1500 * cache entries quickly then in the single threaded loop below,
1501 * we only have to do the special cases that are left.
1503 preload_index(istate
, pathspec
, 0);
1504 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1505 struct cache_entry
*ce
, *new_entry
;
1506 int cache_errno
= 0;
1510 ce
= istate
->cache
[i
];
1511 if (ignore_submodules
&& S_ISGITLINK(ce
->ce_mode
))
1514 if (pathspec
&& !ce_path_match(istate
, ce
, pathspec
, seen
))
1518 while ((i
< istate
->cache_nr
) &&
1519 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
1525 show_file(unmerged_fmt
, ce
->name
, in_porcelain
,
1526 &first
, header_msg
);
1534 new_entry
= refresh_cache_ent(istate
, ce
, options
, &cache_errno
, &changed
);
1535 if (new_entry
== ce
)
1538 display_progress(progress
, i
);
1542 if (really
&& cache_errno
== EINVAL
) {
1543 /* If we are doing --really-refresh that
1544 * means the index is not valid anymore.
1546 ce
->ce_flags
&= ~CE_VALID
;
1547 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
1548 mark_fsmonitor_invalid(istate
, ce
);
1549 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
1554 if (cache_errno
== ENOENT
)
1556 else if (ce_intent_to_add(ce
))
1557 fmt
= added_fmt
; /* must be before other checks */
1558 else if (changed
& TYPE_CHANGED
)
1559 fmt
= typechange_fmt
;
1563 ce
->name
, in_porcelain
, &first
, header_msg
);
1568 replace_index_entry(istate
, i
, new_entry
);
1571 display_progress(progress
, istate
->cache_nr
);
1572 stop_progress(&progress
);
1574 trace_performance_leave("refresh index");
1578 struct cache_entry
*refresh_cache_entry(struct index_state
*istate
,
1579 struct cache_entry
*ce
,
1580 unsigned int options
)
1582 return refresh_cache_ent(istate
, ce
, options
, NULL
, NULL
);
1586 /*****************************************************************
1588 *****************************************************************/
1590 #define INDEX_FORMAT_DEFAULT 3
1592 static unsigned int get_index_format_default(void)
1594 char *envversion
= getenv("GIT_INDEX_VERSION");
1597 unsigned int version
= INDEX_FORMAT_DEFAULT
;
1600 if (!git_config_get_int("index.version", &value
))
1602 if (version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1603 warning(_("index.version set, but the value is invalid.\n"
1604 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1605 return INDEX_FORMAT_DEFAULT
;
1610 version
= strtoul(envversion
, &endp
, 10);
1612 version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1613 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1614 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1615 version
= INDEX_FORMAT_DEFAULT
;
1621 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1622 * Again - this is just a (very strong in practice) heuristic that
1623 * the inode hasn't changed.
1625 * We save the fields in big-endian order to allow using the
1626 * index file over NFS transparently.
1628 struct ondisk_cache_entry
{
1629 struct cache_time ctime
;
1630 struct cache_time mtime
;
1637 unsigned char sha1
[20];
1639 char name
[FLEX_ARRAY
]; /* more */
1643 * This struct is used when CE_EXTENDED bit is 1
1644 * The struct must match ondisk_cache_entry exactly from
1647 struct ondisk_cache_entry_extended
{
1648 struct cache_time ctime
;
1649 struct cache_time mtime
;
1656 unsigned char sha1
[20];
1659 char name
[FLEX_ARRAY
]; /* more */
1662 /* These are only used for v3 or lower */
1663 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1664 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1665 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1666 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1667 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1668 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1669 ondisk_cache_entry_size(ce_namelen(ce)))
1671 /* Allow fsck to force verification of the index checksum. */
1672 int verify_index_checksum
;
1674 /* Allow fsck to force verification of the cache entry order. */
1675 int verify_ce_order
;
1677 static int verify_hdr(const struct cache_header
*hdr
, unsigned long size
)
1680 unsigned char hash
[GIT_MAX_RAWSZ
];
1683 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
1684 return error(_("bad signature 0x%08x"), hdr
->hdr_signature
);
1685 hdr_version
= ntohl(hdr
->hdr_version
);
1686 if (hdr_version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< hdr_version
)
1687 return error(_("bad index version %d"), hdr_version
);
1689 if (!verify_index_checksum
)
1692 the_hash_algo
->init_fn(&c
);
1693 the_hash_algo
->update_fn(&c
, hdr
, size
- the_hash_algo
->rawsz
);
1694 the_hash_algo
->final_fn(hash
, &c
);
1695 if (!hasheq(hash
, (unsigned char *)hdr
+ size
- the_hash_algo
->rawsz
))
1696 return error(_("bad index file sha1 signature"));
1700 static int read_index_extension(struct index_state
*istate
,
1701 const char *ext
, const char *data
, unsigned long sz
)
1703 switch (CACHE_EXT(ext
)) {
1704 case CACHE_EXT_TREE
:
1705 istate
->cache_tree
= cache_tree_read(data
, sz
);
1707 case CACHE_EXT_RESOLVE_UNDO
:
1708 istate
->resolve_undo
= resolve_undo_read(data
, sz
);
1710 case CACHE_EXT_LINK
:
1711 if (read_link_extension(istate
, data
, sz
))
1714 case CACHE_EXT_UNTRACKED
:
1715 istate
->untracked
= read_untracked_extension(data
, sz
);
1717 case CACHE_EXT_FSMONITOR
:
1718 read_fsmonitor_extension(istate
, data
, sz
);
1720 case CACHE_EXT_ENDOFINDEXENTRIES
:
1721 case CACHE_EXT_INDEXENTRYOFFSETTABLE
:
1722 /* already handled in do_read_index() */
1725 if (*ext
< 'A' || 'Z' < *ext
)
1726 return error(_("index uses %.4s extension, which we do not understand"),
1728 fprintf_ln(stderr
, _("ignoring %.4s extension"), ext
);
1734 static struct cache_entry
*create_from_disk(struct mem_pool
*ce_mem_pool
,
1735 unsigned int version
,
1736 struct ondisk_cache_entry
*ondisk
,
1737 unsigned long *ent_size
,
1738 const struct cache_entry
*previous_ce
)
1740 struct cache_entry
*ce
;
1744 size_t copy_len
= 0;
1746 * Adjacent cache entries tend to share the leading paths, so it makes
1747 * sense to only store the differences in later entries. In the v4
1748 * on-disk format of the index, each on-disk cache entry stores the
1749 * number of bytes to be stripped from the end of the previous name,
1750 * and the bytes to append to the result, to come up with its name.
1752 int expand_name_field
= version
== 4;
1754 /* On-disk flags are just 16 bits */
1755 flags
= get_be16(&ondisk
->flags
);
1756 len
= flags
& CE_NAMEMASK
;
1758 if (flags
& CE_EXTENDED
) {
1759 struct ondisk_cache_entry_extended
*ondisk2
;
1761 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
1762 extended_flags
= get_be16(&ondisk2
->flags2
) << 16;
1763 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1764 if (extended_flags
& ~CE_EXTENDED_FLAGS
)
1765 die(_("unknown index entry format 0x%08x"), extended_flags
);
1766 flags
|= extended_flags
;
1767 name
= ondisk2
->name
;
1770 name
= ondisk
->name
;
1772 if (expand_name_field
) {
1773 const unsigned char *cp
= (const unsigned char *)name
;
1774 size_t strip_len
, previous_len
;
1776 /* If we're at the begining of a block, ignore the previous name */
1777 strip_len
= decode_varint(&cp
);
1779 previous_len
= previous_ce
->ce_namelen
;
1780 if (previous_len
< strip_len
)
1781 die(_("malformed name field in the index, near path '%s'"),
1783 copy_len
= previous_len
- strip_len
;
1785 name
= (const char *)cp
;
1788 if (len
== CE_NAMEMASK
) {
1790 if (expand_name_field
)
1794 ce
= mem_pool__ce_alloc(ce_mem_pool
, len
);
1796 ce
->ce_stat_data
.sd_ctime
.sec
= get_be32(&ondisk
->ctime
.sec
);
1797 ce
->ce_stat_data
.sd_mtime
.sec
= get_be32(&ondisk
->mtime
.sec
);
1798 ce
->ce_stat_data
.sd_ctime
.nsec
= get_be32(&ondisk
->ctime
.nsec
);
1799 ce
->ce_stat_data
.sd_mtime
.nsec
= get_be32(&ondisk
->mtime
.nsec
);
1800 ce
->ce_stat_data
.sd_dev
= get_be32(&ondisk
->dev
);
1801 ce
->ce_stat_data
.sd_ino
= get_be32(&ondisk
->ino
);
1802 ce
->ce_mode
= get_be32(&ondisk
->mode
);
1803 ce
->ce_stat_data
.sd_uid
= get_be32(&ondisk
->uid
);
1804 ce
->ce_stat_data
.sd_gid
= get_be32(&ondisk
->gid
);
1805 ce
->ce_stat_data
.sd_size
= get_be32(&ondisk
->size
);
1806 ce
->ce_flags
= flags
& ~CE_NAMEMASK
;
1807 ce
->ce_namelen
= len
;
1809 hashcpy(ce
->oid
.hash
, ondisk
->sha1
);
1811 if (expand_name_field
) {
1813 memcpy(ce
->name
, previous_ce
->name
, copy_len
);
1814 memcpy(ce
->name
+ copy_len
, name
, len
+ 1 - copy_len
);
1815 *ent_size
= (name
- ((char *)ondisk
)) + len
+ 1 - copy_len
;
1817 memcpy(ce
->name
, name
, len
+ 1);
1818 *ent_size
= ondisk_ce_size(ce
);
1823 static void check_ce_order(struct index_state
*istate
)
1827 if (!verify_ce_order
)
1830 for (i
= 1; i
< istate
->cache_nr
; i
++) {
1831 struct cache_entry
*ce
= istate
->cache
[i
- 1];
1832 struct cache_entry
*next_ce
= istate
->cache
[i
];
1833 int name_compare
= strcmp(ce
->name
, next_ce
->name
);
1835 if (0 < name_compare
)
1836 die(_("unordered stage entries in index"));
1837 if (!name_compare
) {
1839 die(_("multiple stage entries for merged file '%s'"),
1841 if (ce_stage(ce
) > ce_stage(next_ce
))
1842 die(_("unordered stage entries for '%s'"),
1848 static void tweak_untracked_cache(struct index_state
*istate
)
1850 switch (git_config_get_untracked_cache()) {
1851 case -1: /* keep: do nothing */
1854 remove_untracked_cache(istate
);
1857 add_untracked_cache(istate
);
1859 default: /* unknown value: do nothing */
1864 static void tweak_split_index(struct index_state
*istate
)
1866 switch (git_config_get_split_index()) {
1867 case -1: /* unset: do nothing */
1870 remove_split_index(istate
);
1873 add_split_index(istate
);
1875 default: /* unknown value: do nothing */
1880 static void post_read_index_from(struct index_state
*istate
)
1882 check_ce_order(istate
);
1883 tweak_untracked_cache(istate
);
1884 tweak_split_index(istate
);
1885 tweak_fsmonitor(istate
);
1888 static size_t estimate_cache_size_from_compressed(unsigned int entries
)
1890 return entries
* (sizeof(struct cache_entry
) + CACHE_ENTRY_PATH_LENGTH
);
1893 static size_t estimate_cache_size(size_t ondisk_size
, unsigned int entries
)
1895 long per_entry
= sizeof(struct cache_entry
) - sizeof(struct ondisk_cache_entry
);
1898 * Account for potential alignment differences.
1900 per_entry
+= align_padding_size(sizeof(struct cache_entry
), -sizeof(struct ondisk_cache_entry
));
1901 return ondisk_size
+ entries
* per_entry
;
1904 struct index_entry_offset
1906 /* starting byte offset into index file, count of index entries in this block */
1910 struct index_entry_offset_table
1913 struct index_entry_offset entries
[FLEX_ARRAY
];
1916 static struct index_entry_offset_table
*read_ieot_extension(const char *mmap
, size_t mmap_size
, size_t offset
);
1917 static void write_ieot_extension(struct strbuf
*sb
, struct index_entry_offset_table
*ieot
);
1919 static size_t read_eoie_extension(const char *mmap
, size_t mmap_size
);
1920 static void write_eoie_extension(struct strbuf
*sb
, git_hash_ctx
*eoie_context
, size_t offset
);
1922 struct load_index_extensions
1925 struct index_state
*istate
;
1928 unsigned long src_offset
;
1931 static void *load_index_extensions(void *_data
)
1933 struct load_index_extensions
*p
= _data
;
1934 unsigned long src_offset
= p
->src_offset
;
1936 while (src_offset
<= p
->mmap_size
- the_hash_algo
->rawsz
- 8) {
1937 /* After an array of active_nr index entries,
1938 * there can be arbitrary number of extended
1939 * sections, each of which is prefixed with
1940 * extension name (4-byte) and section length
1941 * in 4-byte network byte order.
1943 uint32_t extsize
= get_be32(p
->mmap
+ src_offset
+ 4);
1944 if (read_index_extension(p
->istate
,
1945 p
->mmap
+ src_offset
,
1946 p
->mmap
+ src_offset
+ 8,
1948 munmap((void *)p
->mmap
, p
->mmap_size
);
1949 die(_("index file corrupt"));
1952 src_offset
+= extsize
;
1959 * A helper function that will load the specified range of cache entries
1960 * from the memory mapped file and add them to the given index.
1962 static unsigned long load_cache_entry_block(struct index_state
*istate
,
1963 struct mem_pool
*ce_mem_pool
, int offset
, int nr
, const char *mmap
,
1964 unsigned long start_offset
, const struct cache_entry
*previous_ce
)
1967 unsigned long src_offset
= start_offset
;
1969 for (i
= offset
; i
< offset
+ nr
; i
++) {
1970 struct ondisk_cache_entry
*disk_ce
;
1971 struct cache_entry
*ce
;
1972 unsigned long consumed
;
1974 disk_ce
= (struct ondisk_cache_entry
*)(mmap
+ src_offset
);
1975 ce
= create_from_disk(ce_mem_pool
, istate
->version
, disk_ce
, &consumed
, previous_ce
);
1976 set_index_entry(istate
, i
, ce
);
1978 src_offset
+= consumed
;
1981 return src_offset
- start_offset
;
1984 static unsigned long load_all_cache_entries(struct index_state
*istate
,
1985 const char *mmap
, size_t mmap_size
, unsigned long src_offset
)
1987 unsigned long consumed
;
1989 if (istate
->version
== 4) {
1990 mem_pool_init(&istate
->ce_mem_pool
,
1991 estimate_cache_size_from_compressed(istate
->cache_nr
));
1993 mem_pool_init(&istate
->ce_mem_pool
,
1994 estimate_cache_size(mmap_size
, istate
->cache_nr
));
1997 consumed
= load_cache_entry_block(istate
, istate
->ce_mem_pool
,
1998 0, istate
->cache_nr
, mmap
, src_offset
, NULL
);
2003 * Mostly randomly chosen maximum thread counts: we
2004 * cap the parallelism to online_cpus() threads, and we want
2005 * to have at least 10000 cache entries per thread for it to
2006 * be worth starting a thread.
2009 #define THREAD_COST (10000)
2011 struct load_cache_entries_thread_data
2014 struct index_state
*istate
;
2015 struct mem_pool
*ce_mem_pool
;
2018 struct index_entry_offset_table
*ieot
;
2019 int ieot_start
; /* starting index into the ieot array */
2020 int ieot_blocks
; /* count of ieot entries to process */
2021 unsigned long consumed
; /* return # of bytes in index file processed */
2025 * A thread proc to run the load_cache_entries() computation
2026 * across multiple background threads.
2028 static void *load_cache_entries_thread(void *_data
)
2030 struct load_cache_entries_thread_data
*p
= _data
;
2033 /* iterate across all ieot blocks assigned to this thread */
2034 for (i
= p
->ieot_start
; i
< p
->ieot_start
+ p
->ieot_blocks
; i
++) {
2035 p
->consumed
+= load_cache_entry_block(p
->istate
, p
->ce_mem_pool
,
2036 p
->offset
, p
->ieot
->entries
[i
].nr
, p
->mmap
, p
->ieot
->entries
[i
].offset
, NULL
);
2037 p
->offset
+= p
->ieot
->entries
[i
].nr
;
2042 static unsigned long load_cache_entries_threaded(struct index_state
*istate
, const char *mmap
, size_t mmap_size
,
2043 unsigned long src_offset
, int nr_threads
, struct index_entry_offset_table
*ieot
)
2045 int i
, offset
, ieot_blocks
, ieot_start
, err
;
2046 struct load_cache_entries_thread_data
*data
;
2047 unsigned long consumed
= 0;
2049 /* a little sanity checking */
2050 if (istate
->name_hash_initialized
)
2051 BUG("the name hash isn't thread safe");
2053 mem_pool_init(&istate
->ce_mem_pool
, 0);
2055 /* ensure we have no more threads than we have blocks to process */
2056 if (nr_threads
> ieot
->nr
)
2057 nr_threads
= ieot
->nr
;
2058 data
= xcalloc(nr_threads
, sizeof(*data
));
2060 offset
= ieot_start
= 0;
2061 ieot_blocks
= DIV_ROUND_UP(ieot
->nr
, nr_threads
);
2062 for (i
= 0; i
< nr_threads
; i
++) {
2063 struct load_cache_entries_thread_data
*p
= &data
[i
];
2066 if (ieot_start
+ ieot_blocks
> ieot
->nr
)
2067 ieot_blocks
= ieot
->nr
- ieot_start
;
2073 p
->ieot_start
= ieot_start
;
2074 p
->ieot_blocks
= ieot_blocks
;
2076 /* create a mem_pool for each thread */
2078 for (j
= p
->ieot_start
; j
< p
->ieot_start
+ p
->ieot_blocks
; j
++)
2079 nr
+= p
->ieot
->entries
[j
].nr
;
2080 if (istate
->version
== 4) {
2081 mem_pool_init(&p
->ce_mem_pool
,
2082 estimate_cache_size_from_compressed(nr
));
2084 mem_pool_init(&p
->ce_mem_pool
,
2085 estimate_cache_size(mmap_size
, nr
));
2088 err
= pthread_create(&p
->pthread
, NULL
, load_cache_entries_thread
, p
);
2090 die(_("unable to create load_cache_entries thread: %s"), strerror(err
));
2092 /* increment by the number of cache entries in the ieot block being processed */
2093 for (j
= 0; j
< ieot_blocks
; j
++)
2094 offset
+= ieot
->entries
[ieot_start
+ j
].nr
;
2095 ieot_start
+= ieot_blocks
;
2098 for (i
= 0; i
< nr_threads
; i
++) {
2099 struct load_cache_entries_thread_data
*p
= &data
[i
];
2101 err
= pthread_join(p
->pthread
, NULL
);
2103 die(_("unable to join load_cache_entries thread: %s"), strerror(err
));
2104 mem_pool_combine(istate
->ce_mem_pool
, p
->ce_mem_pool
);
2105 consumed
+= p
->consumed
;
2113 /* remember to discard_cache() before reading a different cache! */
2114 int do_read_index(struct index_state
*istate
, const char *path
, int must_exist
)
2118 unsigned long src_offset
;
2119 const struct cache_header
*hdr
;
2122 struct load_index_extensions p
;
2123 size_t extension_offset
= 0;
2124 int nr_threads
, cpus
;
2125 struct index_entry_offset_table
*ieot
= NULL
;
2127 if (istate
->initialized
)
2128 return istate
->cache_nr
;
2130 istate
->timestamp
.sec
= 0;
2131 istate
->timestamp
.nsec
= 0;
2132 fd
= open(path
, O_RDONLY
);
2134 if (!must_exist
&& errno
== ENOENT
)
2136 die_errno(_("%s: index file open failed"), path
);
2140 die_errno(_("%s: cannot stat the open index"), path
);
2142 mmap_size
= xsize_t(st
.st_size
);
2143 if (mmap_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2144 die(_("%s: index file smaller than expected"), path
);
2146 mmap
= xmmap(NULL
, mmap_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2147 if (mmap
== MAP_FAILED
)
2148 die_errno(_("%s: unable to map index file"), path
);
2151 hdr
= (const struct cache_header
*)mmap
;
2152 if (verify_hdr(hdr
, mmap_size
) < 0)
2155 hashcpy(istate
->oid
.hash
, (const unsigned char *)hdr
+ mmap_size
- the_hash_algo
->rawsz
);
2156 istate
->version
= ntohl(hdr
->hdr_version
);
2157 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
2158 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
2159 istate
->cache
= xcalloc(istate
->cache_alloc
, sizeof(*istate
->cache
));
2160 istate
->initialized
= 1;
2164 p
.mmap_size
= mmap_size
;
2166 src_offset
= sizeof(*hdr
);
2168 if (git_config_get_index_threads(&nr_threads
))
2171 /* TODO: does creating more threads than cores help? */
2173 nr_threads
= istate
->cache_nr
/ THREAD_COST
;
2174 cpus
= online_cpus();
2175 if (nr_threads
> cpus
)
2182 if (nr_threads
> 1) {
2183 extension_offset
= read_eoie_extension(mmap
, mmap_size
);
2184 if (extension_offset
) {
2187 p
.src_offset
= extension_offset
;
2188 err
= pthread_create(&p
.pthread
, NULL
, load_index_extensions
, &p
);
2190 die(_("unable to create load_index_extensions thread: %s"), strerror(err
));
2197 * Locate and read the index entry offset table so that we can use it
2198 * to multi-thread the reading of the cache entries.
2200 if (extension_offset
&& nr_threads
> 1)
2201 ieot
= read_ieot_extension(mmap
, mmap_size
, extension_offset
);
2204 src_offset
+= load_cache_entries_threaded(istate
, mmap
, mmap_size
, src_offset
, nr_threads
, ieot
);
2207 src_offset
+= load_all_cache_entries(istate
, mmap
, mmap_size
, src_offset
);
2210 istate
->timestamp
.sec
= st
.st_mtime
;
2211 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
2213 /* if we created a thread, join it otherwise load the extensions on the primary thread */
2214 if (extension_offset
) {
2215 int ret
= pthread_join(p
.pthread
, NULL
);
2217 die(_("unable to join load_index_extensions thread: %s"), strerror(ret
));
2219 p
.src_offset
= src_offset
;
2220 load_index_extensions(&p
);
2222 munmap((void *)mmap
, mmap_size
);
2225 * TODO trace2: replace "the_repository" with the actual repo instance
2226 * that is associated with the given "istate".
2228 trace2_data_intmax("index", the_repository
, "read/version",
2230 trace2_data_intmax("index", the_repository
, "read/cache_nr",
2233 return istate
->cache_nr
;
2236 munmap((void *)mmap
, mmap_size
);
2237 die(_("index file corrupt"));
2241 * Signal that the shared index is used by updating its mtime.
2243 * This way, shared index can be removed if they have not been used
2246 static void freshen_shared_index(const char *shared_index
, int warn
)
2248 if (!check_and_freshen_file(shared_index
, 1) && warn
)
2249 warning(_("could not freshen shared index '%s'"), shared_index
);
2252 int read_index_from(struct index_state
*istate
, const char *path
,
2255 struct split_index
*split_index
;
2260 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2261 if (istate
->initialized
)
2262 return istate
->cache_nr
;
2265 * TODO trace2: replace "the_repository" with the actual repo instance
2266 * that is associated with the given "istate".
2268 trace2_region_enter_printf("index", "do_read_index", the_repository
,
2270 trace_performance_enter();
2271 ret
= do_read_index(istate
, path
, 0);
2272 trace_performance_leave("read cache %s", path
);
2273 trace2_region_leave_printf("index", "do_read_index", the_repository
,
2276 split_index
= istate
->split_index
;
2277 if (!split_index
|| is_null_oid(&split_index
->base_oid
)) {
2278 post_read_index_from(istate
);
2282 trace_performance_enter();
2283 if (split_index
->base
)
2284 discard_index(split_index
->base
);
2286 split_index
->base
= xcalloc(1, sizeof(*split_index
->base
));
2288 base_oid_hex
= oid_to_hex(&split_index
->base_oid
);
2289 base_path
= xstrfmt("%s/sharedindex.%s", gitdir
, base_oid_hex
);
2290 trace2_region_enter_printf("index", "shared/do_read_index",
2291 the_repository
, "%s", base_path
);
2292 ret
= do_read_index(split_index
->base
, base_path
, 1);
2293 trace2_region_leave_printf("index", "shared/do_read_index",
2294 the_repository
, "%s", base_path
);
2295 if (!oideq(&split_index
->base_oid
, &split_index
->base
->oid
))
2296 die(_("broken index, expect %s in %s, got %s"),
2297 base_oid_hex
, base_path
,
2298 oid_to_hex(&split_index
->base
->oid
));
2300 freshen_shared_index(base_path
, 0);
2301 merge_base_index(istate
);
2302 post_read_index_from(istate
);
2303 trace_performance_leave("read cache %s", base_path
);
2308 int is_index_unborn(struct index_state
*istate
)
2310 return (!istate
->cache_nr
&& !istate
->timestamp
.sec
);
2313 int discard_index(struct index_state
*istate
)
2316 * Cache entries in istate->cache[] should have been allocated
2317 * from the memory pool associated with this index, or from an
2318 * associated split_index. There is no need to free individual
2319 * cache entries. validate_cache_entries can detect when this
2320 * assertion does not hold.
2322 validate_cache_entries(istate
);
2324 resolve_undo_clear_index(istate
);
2325 istate
->cache_nr
= 0;
2326 istate
->cache_changed
= 0;
2327 istate
->timestamp
.sec
= 0;
2328 istate
->timestamp
.nsec
= 0;
2329 free_name_hash(istate
);
2330 cache_tree_free(&(istate
->cache_tree
));
2331 istate
->initialized
= 0;
2332 FREE_AND_NULL(istate
->cache
);
2333 istate
->cache_alloc
= 0;
2334 discard_split_index(istate
);
2335 free_untracked_cache(istate
->untracked
);
2336 istate
->untracked
= NULL
;
2338 if (istate
->ce_mem_pool
) {
2339 mem_pool_discard(istate
->ce_mem_pool
, should_validate_cache_entries());
2340 istate
->ce_mem_pool
= NULL
;
2347 * Validate the cache entries of this index.
2348 * All cache entries associated with this index
2349 * should have been allocated by the memory pool
2350 * associated with this index, or by a referenced
2353 void validate_cache_entries(const struct index_state
*istate
)
2357 if (!should_validate_cache_entries() ||!istate
|| !istate
->initialized
)
2360 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2362 BUG("cache entry is not allocated from expected memory pool");
2363 } else if (!istate
->ce_mem_pool
||
2364 !mem_pool_contains(istate
->ce_mem_pool
, istate
->cache
[i
])) {
2365 if (!istate
->split_index
||
2366 !istate
->split_index
->base
||
2367 !istate
->split_index
->base
->ce_mem_pool
||
2368 !mem_pool_contains(istate
->split_index
->base
->ce_mem_pool
, istate
->cache
[i
])) {
2369 BUG("cache entry is not allocated from expected memory pool");
2374 if (istate
->split_index
)
2375 validate_cache_entries(istate
->split_index
->base
);
2378 int unmerged_index(const struct index_state
*istate
)
2381 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2382 if (ce_stage(istate
->cache
[i
]))
2388 int repo_index_has_changes(struct repository
*repo
,
2392 struct index_state
*istate
= repo
->index
;
2393 struct object_id cmp
;
2397 cmp
= tree
->object
.oid
;
2398 if (tree
|| !get_oid_tree("HEAD", &cmp
)) {
2399 struct diff_options opt
;
2401 repo_diff_setup(repo
, &opt
);
2402 opt
.flags
.exit_with_status
= 1;
2404 opt
.flags
.quick
= 1;
2405 do_diff_cache(&cmp
, &opt
);
2407 for (i
= 0; sb
&& i
< diff_queued_diff
.nr
; i
++) {
2409 strbuf_addch(sb
, ' ');
2410 strbuf_addstr(sb
, diff_queued_diff
.queue
[i
]->two
->path
);
2413 return opt
.flags
.has_changes
!= 0;
2415 for (i
= 0; sb
&& i
< istate
->cache_nr
; i
++) {
2417 strbuf_addch(sb
, ' ');
2418 strbuf_addstr(sb
, istate
->cache
[i
]->name
);
2420 return !!istate
->cache_nr
;
2424 #define WRITE_BUFFER_SIZE 8192
2425 static unsigned char write_buffer
[WRITE_BUFFER_SIZE
];
2426 static unsigned long write_buffer_len
;
2428 static int ce_write_flush(git_hash_ctx
*context
, int fd
)
2430 unsigned int buffered
= write_buffer_len
;
2432 the_hash_algo
->update_fn(context
, write_buffer
, buffered
);
2433 if (write_in_full(fd
, write_buffer
, buffered
) < 0)
2435 write_buffer_len
= 0;
2440 static int ce_write(git_hash_ctx
*context
, int fd
, void *data
, unsigned int len
)
2443 unsigned int buffered
= write_buffer_len
;
2444 unsigned int partial
= WRITE_BUFFER_SIZE
- buffered
;
2447 memcpy(write_buffer
+ buffered
, data
, partial
);
2448 buffered
+= partial
;
2449 if (buffered
== WRITE_BUFFER_SIZE
) {
2450 write_buffer_len
= buffered
;
2451 if (ce_write_flush(context
, fd
))
2455 write_buffer_len
= buffered
;
2457 data
= (char *) data
+ partial
;
2462 static int write_index_ext_header(git_hash_ctx
*context
, git_hash_ctx
*eoie_context
,
2463 int fd
, unsigned int ext
, unsigned int sz
)
2468 the_hash_algo
->update_fn(eoie_context
, &ext
, 4);
2469 the_hash_algo
->update_fn(eoie_context
, &sz
, 4);
2471 return ((ce_write(context
, fd
, &ext
, 4) < 0) ||
2472 (ce_write(context
, fd
, &sz
, 4) < 0)) ? -1 : 0;
2475 static int ce_flush(git_hash_ctx
*context
, int fd
, unsigned char *hash
)
2477 unsigned int left
= write_buffer_len
;
2480 write_buffer_len
= 0;
2481 the_hash_algo
->update_fn(context
, write_buffer
, left
);
2484 /* Flush first if not enough space for hash signature */
2485 if (left
+ the_hash_algo
->rawsz
> WRITE_BUFFER_SIZE
) {
2486 if (write_in_full(fd
, write_buffer
, left
) < 0)
2491 /* Append the hash signature at the end */
2492 the_hash_algo
->final_fn(write_buffer
+ left
, context
);
2493 hashcpy(hash
, write_buffer
+ left
);
2494 left
+= the_hash_algo
->rawsz
;
2495 return (write_in_full(fd
, write_buffer
, left
) < 0) ? -1 : 0;
2498 static void ce_smudge_racily_clean_entry(struct index_state
*istate
,
2499 struct cache_entry
*ce
)
2502 * The only thing we care about in this function is to smudge the
2503 * falsely clean entry due to touch-update-touch race, so we leave
2504 * everything else as they are. We are called for entries whose
2505 * ce_stat_data.sd_mtime match the index file mtime.
2507 * Note that this actually does not do much for gitlinks, for
2508 * which ce_match_stat_basic() always goes to the actual
2509 * contents. The caller checks with is_racy_timestamp() which
2510 * always says "no" for gitlinks, so we are not called for them ;-)
2514 if (lstat(ce
->name
, &st
) < 0)
2516 if (ce_match_stat_basic(ce
, &st
))
2518 if (ce_modified_check_fs(istate
, ce
, &st
)) {
2519 /* This is "racily clean"; smudge it. Note that this
2520 * is a tricky code. At first glance, it may appear
2521 * that it can break with this sequence:
2523 * $ echo xyzzy >frotz
2524 * $ git-update-index --add frotz
2527 * $ echo filfre >nitfol
2528 * $ git-update-index --add nitfol
2530 * but it does not. When the second update-index runs,
2531 * it notices that the entry "frotz" has the same timestamp
2532 * as index, and if we were to smudge it by resetting its
2533 * size to zero here, then the object name recorded
2534 * in index is the 6-byte file but the cached stat information
2535 * becomes zero --- which would then match what we would
2536 * obtain from the filesystem next time we stat("frotz").
2538 * However, the second update-index, before calling
2539 * this function, notices that the cached size is 6
2540 * bytes and what is on the filesystem is an empty
2541 * file, and never calls us, so the cached size information
2542 * for "frotz" stays 6 which does not match the filesystem.
2544 ce
->ce_stat_data
.sd_size
= 0;
2548 /* Copy miscellaneous fields but not the name */
2549 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry
*ondisk
,
2550 struct cache_entry
*ce
)
2554 ondisk
->ctime
.sec
= htonl(ce
->ce_stat_data
.sd_ctime
.sec
);
2555 ondisk
->mtime
.sec
= htonl(ce
->ce_stat_data
.sd_mtime
.sec
);
2556 ondisk
->ctime
.nsec
= htonl(ce
->ce_stat_data
.sd_ctime
.nsec
);
2557 ondisk
->mtime
.nsec
= htonl(ce
->ce_stat_data
.sd_mtime
.nsec
);
2558 ondisk
->dev
= htonl(ce
->ce_stat_data
.sd_dev
);
2559 ondisk
->ino
= htonl(ce
->ce_stat_data
.sd_ino
);
2560 ondisk
->mode
= htonl(ce
->ce_mode
);
2561 ondisk
->uid
= htonl(ce
->ce_stat_data
.sd_uid
);
2562 ondisk
->gid
= htonl(ce
->ce_stat_data
.sd_gid
);
2563 ondisk
->size
= htonl(ce
->ce_stat_data
.sd_size
);
2564 hashcpy(ondisk
->sha1
, ce
->oid
.hash
);
2566 flags
= ce
->ce_flags
& ~CE_NAMEMASK
;
2567 flags
|= (ce_namelen(ce
) >= CE_NAMEMASK
? CE_NAMEMASK
: ce_namelen(ce
));
2568 ondisk
->flags
= htons(flags
);
2569 if (ce
->ce_flags
& CE_EXTENDED
) {
2570 struct ondisk_cache_entry_extended
*ondisk2
;
2571 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
2572 ondisk2
->flags2
= htons((ce
->ce_flags
& CE_EXTENDED_FLAGS
) >> 16);
2576 static int ce_write_entry(git_hash_ctx
*c
, int fd
, struct cache_entry
*ce
,
2577 struct strbuf
*previous_name
, struct ondisk_cache_entry
*ondisk
)
2581 unsigned int saved_namelen
;
2582 int stripped_name
= 0;
2583 static unsigned char padding
[8] = { 0x00 };
2585 if (ce
->ce_flags
& CE_STRIP_NAME
) {
2586 saved_namelen
= ce_namelen(ce
);
2591 if (ce
->ce_flags
& CE_EXTENDED
)
2592 size
= offsetof(struct ondisk_cache_entry_extended
, name
);
2594 size
= offsetof(struct ondisk_cache_entry
, name
);
2596 if (!previous_name
) {
2597 int len
= ce_namelen(ce
);
2598 copy_cache_entry_to_ondisk(ondisk
, ce
);
2599 result
= ce_write(c
, fd
, ondisk
, size
);
2601 result
= ce_write(c
, fd
, ce
->name
, len
);
2603 result
= ce_write(c
, fd
, padding
, align_padding_size(size
, len
));
2605 int common
, to_remove
, prefix_size
;
2606 unsigned char to_remove_vi
[16];
2608 (ce
->name
[common
] &&
2609 common
< previous_name
->len
&&
2610 ce
->name
[common
] == previous_name
->buf
[common
]);
2612 ; /* still matching */
2613 to_remove
= previous_name
->len
- common
;
2614 prefix_size
= encode_varint(to_remove
, to_remove_vi
);
2616 copy_cache_entry_to_ondisk(ondisk
, ce
);
2617 result
= ce_write(c
, fd
, ondisk
, size
);
2619 result
= ce_write(c
, fd
, to_remove_vi
, prefix_size
);
2621 result
= ce_write(c
, fd
, ce
->name
+ common
, ce_namelen(ce
) - common
);
2623 result
= ce_write(c
, fd
, padding
, 1);
2625 strbuf_splice(previous_name
, common
, to_remove
,
2626 ce
->name
+ common
, ce_namelen(ce
) - common
);
2628 if (stripped_name
) {
2629 ce
->ce_namelen
= saved_namelen
;
2630 ce
->ce_flags
&= ~CE_STRIP_NAME
;
2637 * This function verifies if index_state has the correct sha1 of the
2638 * index file. Don't die if we have any other failure, just return 0.
2640 static int verify_index_from(const struct index_state
*istate
, const char *path
)
2645 unsigned char hash
[GIT_MAX_RAWSZ
];
2647 if (!istate
->initialized
)
2650 fd
= open(path
, O_RDONLY
);
2657 if (st
.st_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2660 n
= pread_in_full(fd
, hash
, the_hash_algo
->rawsz
, st
.st_size
- the_hash_algo
->rawsz
);
2661 if (n
!= the_hash_algo
->rawsz
)
2664 if (!hasheq(istate
->oid
.hash
, hash
))
2675 static int repo_verify_index(struct repository
*repo
)
2677 return verify_index_from(repo
->index
, repo
->index_file
);
2680 static int has_racy_timestamp(struct index_state
*istate
)
2682 int entries
= istate
->cache_nr
;
2685 for (i
= 0; i
< entries
; i
++) {
2686 struct cache_entry
*ce
= istate
->cache
[i
];
2687 if (is_racy_timestamp(istate
, ce
))
2693 void repo_update_index_if_able(struct repository
*repo
,
2694 struct lock_file
*lockfile
)
2696 if ((repo
->index
->cache_changed
||
2697 has_racy_timestamp(repo
->index
)) &&
2698 repo_verify_index(repo
))
2699 write_locked_index(repo
->index
, lockfile
, COMMIT_LOCK
);
2701 rollback_lock_file(lockfile
);
2704 static int record_eoie(void)
2708 if (!git_config_get_bool("index.recordendofindexentries", &val
))
2712 * As a convenience, the end of index entries extension
2713 * used for threading is written by default if the user
2714 * explicitly requested threaded index reads.
2716 return !git_config_get_index_threads(&val
) && val
!= 1;
2719 static int record_ieot(void)
2723 if (!git_config_get_bool("index.recordoffsettable", &val
))
2727 * As a convenience, the offset table used for threading is
2728 * written by default if the user explicitly requested
2729 * threaded index reads.
2731 return !git_config_get_index_threads(&val
) && val
!= 1;
2735 * On success, `tempfile` is closed. If it is the temporary file
2736 * of a `struct lock_file`, we will therefore effectively perform
2737 * a 'close_lock_file_gently()`. Since that is an implementation
2738 * detail of lockfiles, callers of `do_write_index()` should not
2741 static int do_write_index(struct index_state
*istate
, struct tempfile
*tempfile
,
2742 int strip_extensions
)
2744 uint64_t start
= getnanotime();
2745 int newfd
= tempfile
->fd
;
2746 git_hash_ctx c
, eoie_c
;
2747 struct cache_header hdr
;
2748 int i
, err
= 0, removed
, extended
, hdr_version
;
2749 struct cache_entry
**cache
= istate
->cache
;
2750 int entries
= istate
->cache_nr
;
2752 struct ondisk_cache_entry_extended ondisk
;
2753 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
2754 int drop_cache_tree
= istate
->drop_cache_tree
;
2756 int ieot_entries
= 1;
2757 struct index_entry_offset_table
*ieot
= NULL
;
2760 for (i
= removed
= extended
= 0; i
< entries
; i
++) {
2761 if (cache
[i
]->ce_flags
& CE_REMOVE
)
2764 /* reduce extended entries if possible */
2765 cache
[i
]->ce_flags
&= ~CE_EXTENDED
;
2766 if (cache
[i
]->ce_flags
& CE_EXTENDED_FLAGS
) {
2768 cache
[i
]->ce_flags
|= CE_EXTENDED
;
2772 if (!istate
->version
) {
2773 istate
->version
= get_index_format_default();
2774 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0))
2775 init_split_index(istate
);
2778 /* demote version 3 to version 2 when the latter suffices */
2779 if (istate
->version
== 3 || istate
->version
== 2)
2780 istate
->version
= extended
? 3 : 2;
2782 hdr_version
= istate
->version
;
2784 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
2785 hdr
.hdr_version
= htonl(hdr_version
);
2786 hdr
.hdr_entries
= htonl(entries
- removed
);
2788 the_hash_algo
->init_fn(&c
);
2789 if (ce_write(&c
, newfd
, &hdr
, sizeof(hdr
)) < 0)
2792 if (!HAVE_THREADS
|| git_config_get_index_threads(&nr_threads
))
2795 if (nr_threads
!= 1 && record_ieot()) {
2796 int ieot_blocks
, cpus
;
2799 * ensure default number of ieot blocks maps evenly to the
2800 * default number of threads that will process them leaving
2801 * room for the thread to load the index extensions.
2804 ieot_blocks
= istate
->cache_nr
/ THREAD_COST
;
2805 cpus
= online_cpus();
2806 if (ieot_blocks
> cpus
- 1)
2807 ieot_blocks
= cpus
- 1;
2809 ieot_blocks
= nr_threads
;
2810 if (ieot_blocks
> istate
->cache_nr
)
2811 ieot_blocks
= istate
->cache_nr
;
2815 * no reason to write out the IEOT extension if we don't
2816 * have enough blocks to utilize multi-threading
2818 if (ieot_blocks
> 1) {
2819 ieot
= xcalloc(1, sizeof(struct index_entry_offset_table
)
2820 + (ieot_blocks
* sizeof(struct index_entry_offset
)));
2821 ieot_entries
= DIV_ROUND_UP(entries
, ieot_blocks
);
2825 offset
= lseek(newfd
, 0, SEEK_CUR
);
2830 offset
+= write_buffer_len
;
2832 previous_name
= (hdr_version
== 4) ? &previous_name_buf
: NULL
;
2834 for (i
= 0; i
< entries
; i
++) {
2835 struct cache_entry
*ce
= cache
[i
];
2836 if (ce
->ce_flags
& CE_REMOVE
)
2838 if (!ce_uptodate(ce
) && is_racy_timestamp(istate
, ce
))
2839 ce_smudge_racily_clean_entry(istate
, ce
);
2840 if (is_null_oid(&ce
->oid
)) {
2841 static const char msg
[] = "cache entry has null sha1: %s";
2842 static int allow
= -1;
2845 allow
= git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2847 warning(msg
, ce
->name
);
2849 err
= error(msg
, ce
->name
);
2851 drop_cache_tree
= 1;
2853 if (ieot
&& i
&& (i
% ieot_entries
== 0)) {
2854 ieot
->entries
[ieot
->nr
].nr
= nr
;
2855 ieot
->entries
[ieot
->nr
].offset
= offset
;
2858 * If we have a V4 index, set the first byte to an invalid
2859 * character to ensure there is nothing common with the previous
2863 previous_name
->buf
[0] = 0;
2865 offset
= lseek(newfd
, 0, SEEK_CUR
);
2870 offset
+= write_buffer_len
;
2872 if (ce_write_entry(&c
, newfd
, ce
, previous_name
, (struct ondisk_cache_entry
*)&ondisk
) < 0)
2880 ieot
->entries
[ieot
->nr
].nr
= nr
;
2881 ieot
->entries
[ieot
->nr
].offset
= offset
;
2884 strbuf_release(&previous_name_buf
);
2891 /* Write extension data here */
2892 offset
= lseek(newfd
, 0, SEEK_CUR
);
2897 offset
+= write_buffer_len
;
2898 the_hash_algo
->init_fn(&eoie_c
);
2901 * Lets write out CACHE_EXT_INDEXENTRYOFFSETTABLE first so that we
2902 * can minimize the number of extensions we have to scan through to
2903 * find it during load. Write it out regardless of the
2904 * strip_extensions parameter as we need it when loading the shared
2908 struct strbuf sb
= STRBUF_INIT
;
2910 write_ieot_extension(&sb
, ieot
);
2911 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_INDEXENTRYOFFSETTABLE
, sb
.len
) < 0
2912 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2913 strbuf_release(&sb
);
2919 if (!strip_extensions
&& istate
->split_index
) {
2920 struct strbuf sb
= STRBUF_INIT
;
2922 err
= write_link_extension(&sb
, istate
) < 0 ||
2923 write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_LINK
,
2925 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2926 strbuf_release(&sb
);
2930 if (!strip_extensions
&& !drop_cache_tree
&& istate
->cache_tree
) {
2931 struct strbuf sb
= STRBUF_INIT
;
2933 cache_tree_write(&sb
, istate
->cache_tree
);
2934 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_TREE
, sb
.len
) < 0
2935 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2936 strbuf_release(&sb
);
2940 if (!strip_extensions
&& istate
->resolve_undo
) {
2941 struct strbuf sb
= STRBUF_INIT
;
2943 resolve_undo_write(&sb
, istate
->resolve_undo
);
2944 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_RESOLVE_UNDO
,
2946 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2947 strbuf_release(&sb
);
2951 if (!strip_extensions
&& istate
->untracked
) {
2952 struct strbuf sb
= STRBUF_INIT
;
2954 write_untracked_extension(&sb
, istate
->untracked
);
2955 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_UNTRACKED
,
2957 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2958 strbuf_release(&sb
);
2962 if (!strip_extensions
&& istate
->fsmonitor_last_update
) {
2963 struct strbuf sb
= STRBUF_INIT
;
2965 write_fsmonitor_extension(&sb
, istate
);
2966 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_FSMONITOR
, sb
.len
) < 0
2967 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2968 strbuf_release(&sb
);
2974 * CACHE_EXT_ENDOFINDEXENTRIES must be written as the last entry before the SHA1
2975 * so that it can be found and processed before all the index entries are
2976 * read. Write it out regardless of the strip_extensions parameter as we need it
2977 * when loading the shared index.
2979 if (offset
&& record_eoie()) {
2980 struct strbuf sb
= STRBUF_INIT
;
2982 write_eoie_extension(&sb
, &eoie_c
, offset
);
2983 err
= write_index_ext_header(&c
, NULL
, newfd
, CACHE_EXT_ENDOFINDEXENTRIES
, sb
.len
) < 0
2984 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2985 strbuf_release(&sb
);
2990 if (ce_flush(&c
, newfd
, istate
->oid
.hash
))
2992 if (close_tempfile_gently(tempfile
)) {
2993 error(_("could not close '%s'"), tempfile
->filename
.buf
);
2996 if (stat(tempfile
->filename
.buf
, &st
))
2998 istate
->timestamp
.sec
= (unsigned int)st
.st_mtime
;
2999 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
3000 trace_performance_since(start
, "write index, changed mask = %x", istate
->cache_changed
);
3003 * TODO trace2: replace "the_repository" with the actual repo instance
3004 * that is associated with the given "istate".
3006 trace2_data_intmax("index", the_repository
, "write/version",
3008 trace2_data_intmax("index", the_repository
, "write/cache_nr",
3014 void set_alternate_index_output(const char *name
)
3016 alternate_index_output
= name
;
3019 static int commit_locked_index(struct lock_file
*lk
)
3021 if (alternate_index_output
)
3022 return commit_lock_file_to(lk
, alternate_index_output
);
3024 return commit_lock_file(lk
);
3027 static int do_write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
3033 * TODO trace2: replace "the_repository" with the actual repo instance
3034 * that is associated with the given "istate".
3036 trace2_region_enter_printf("index", "do_write_index", the_repository
,
3037 "%s", lock
->tempfile
->filename
.buf
);
3038 ret
= do_write_index(istate
, lock
->tempfile
, 0);
3039 trace2_region_leave_printf("index", "do_write_index", the_repository
,
3040 "%s", lock
->tempfile
->filename
.buf
);
3044 if (flags
& COMMIT_LOCK
)
3045 return commit_locked_index(lock
);
3046 return close_lock_file_gently(lock
);
3049 static int write_split_index(struct index_state
*istate
,
3050 struct lock_file
*lock
,
3054 prepare_to_write_split_index(istate
);
3055 ret
= do_write_locked_index(istate
, lock
, flags
);
3056 finish_writing_split_index(istate
);
3060 static const char *shared_index_expire
= "2.weeks.ago";
3062 static unsigned long get_shared_index_expire_date(void)
3064 static unsigned long shared_index_expire_date
;
3065 static int shared_index_expire_date_prepared
;
3067 if (!shared_index_expire_date_prepared
) {
3068 git_config_get_expiry("splitindex.sharedindexexpire",
3069 &shared_index_expire
);
3070 shared_index_expire_date
= approxidate(shared_index_expire
);
3071 shared_index_expire_date_prepared
= 1;
3074 return shared_index_expire_date
;
3077 static int should_delete_shared_index(const char *shared_index_path
)
3080 unsigned long expiration
;
3082 /* Check timestamp */
3083 expiration
= get_shared_index_expire_date();
3086 if (stat(shared_index_path
, &st
))
3087 return error_errno(_("could not stat '%s'"), shared_index_path
);
3088 if (st
.st_mtime
> expiration
)
3094 static int clean_shared_index_files(const char *current_hex
)
3097 DIR *dir
= opendir(get_git_dir());
3100 return error_errno(_("unable to open git dir: %s"), get_git_dir());
3102 while ((de
= readdir(dir
)) != NULL
) {
3103 const char *sha1_hex
;
3104 const char *shared_index_path
;
3105 if (!skip_prefix(de
->d_name
, "sharedindex.", &sha1_hex
))
3107 if (!strcmp(sha1_hex
, current_hex
))
3109 shared_index_path
= git_path("%s", de
->d_name
);
3110 if (should_delete_shared_index(shared_index_path
) > 0 &&
3111 unlink(shared_index_path
))
3112 warning_errno(_("unable to unlink: %s"), shared_index_path
);
3119 static int write_shared_index(struct index_state
*istate
,
3120 struct tempfile
**temp
)
3122 struct split_index
*si
= istate
->split_index
;
3125 move_cache_to_base_index(istate
);
3127 trace2_region_enter_printf("index", "shared/do_write_index",
3128 the_repository
, "%s", (*temp
)->filename
.buf
);
3129 ret
= do_write_index(si
->base
, *temp
, 1);
3130 trace2_region_enter_printf("index", "shared/do_write_index",
3131 the_repository
, "%s", (*temp
)->filename
.buf
);
3135 ret
= adjust_shared_perm(get_tempfile_path(*temp
));
3137 error(_("cannot fix permission bits on '%s'"), get_tempfile_path(*temp
));
3140 ret
= rename_tempfile(temp
,
3141 git_path("sharedindex.%s", oid_to_hex(&si
->base
->oid
)));
3143 oidcpy(&si
->base_oid
, &si
->base
->oid
);
3144 clean_shared_index_files(oid_to_hex(&si
->base
->oid
));
3150 static const int default_max_percent_split_change
= 20;
3152 static int too_many_not_shared_entries(struct index_state
*istate
)
3154 int i
, not_shared
= 0;
3155 int max_split
= git_config_get_max_percent_split_change();
3157 switch (max_split
) {
3159 /* not or badly configured: use the default value */
3160 max_split
= default_max_percent_split_change
;
3163 return 1; /* 0% means always write a new shared index */
3165 return 0; /* 100% means never write a new shared index */
3167 break; /* just use the configured value */
3170 /* Count not shared entries */
3171 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3172 struct cache_entry
*ce
= istate
->cache
[i
];
3177 return (int64_t)istate
->cache_nr
* max_split
< (int64_t)not_shared
* 100;
3180 int write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
3183 int new_shared_index
, ret
;
3184 struct split_index
*si
= istate
->split_index
;
3186 if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0))
3187 cache_tree_verify(the_repository
, istate
);
3189 if ((flags
& SKIP_IF_UNCHANGED
) && !istate
->cache_changed
) {
3190 if (flags
& COMMIT_LOCK
)
3191 rollback_lock_file(lock
);
3195 if (istate
->fsmonitor_last_update
)
3196 fill_fsmonitor_bitmap(istate
);
3198 if (!si
|| alternate_index_output
||
3199 (istate
->cache_changed
& ~EXTMASK
)) {
3201 oidclr(&si
->base_oid
);
3202 ret
= do_write_locked_index(istate
, lock
, flags
);
3206 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0)) {
3207 int v
= si
->base_oid
.hash
[0];
3209 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3211 if (too_many_not_shared_entries(istate
))
3212 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3214 new_shared_index
= istate
->cache_changed
& SPLIT_INDEX_ORDERED
;
3216 if (new_shared_index
) {
3217 struct tempfile
*temp
;
3220 /* Same initial permissions as the main .git/index file */
3221 temp
= mks_tempfile_sm(git_path("sharedindex_XXXXXX"), 0, 0666);
3223 oidclr(&si
->base_oid
);
3224 ret
= do_write_locked_index(istate
, lock
, flags
);
3227 ret
= write_shared_index(istate
, &temp
);
3229 saved_errno
= errno
;
3230 if (is_tempfile_active(temp
))
3231 delete_tempfile(&temp
);
3232 errno
= saved_errno
;
3238 ret
= write_split_index(istate
, lock
, flags
);
3240 /* Freshen the shared index only if the split-index was written */
3241 if (!ret
&& !new_shared_index
) {
3242 const char *shared_index
= git_path("sharedindex.%s",
3243 oid_to_hex(&si
->base_oid
));
3244 freshen_shared_index(shared_index
, 1);
3248 if (flags
& COMMIT_LOCK
)
3249 rollback_lock_file(lock
);
3254 * Read the index file that is potentially unmerged into given
3255 * index_state, dropping any unmerged entries to stage #0 (potentially
3256 * resulting in a path appearing as both a file and a directory in the
3257 * index; the caller is responsible to clear out the extra entries
3258 * before writing the index to a tree). Returns true if the index is
3259 * unmerged. Callers who want to refuse to work from an unmerged
3260 * state can call this and check its return value, instead of calling
3263 int repo_read_index_unmerged(struct repository
*repo
)
3265 struct index_state
*istate
;
3269 repo_read_index(repo
);
3270 istate
= repo
->index
;
3271 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3272 struct cache_entry
*ce
= istate
->cache
[i
];
3273 struct cache_entry
*new_ce
;
3279 len
= ce_namelen(ce
);
3280 new_ce
= make_empty_cache_entry(istate
, len
);
3281 memcpy(new_ce
->name
, ce
->name
, len
);
3282 new_ce
->ce_flags
= create_ce_flags(0) | CE_CONFLICTED
;
3283 new_ce
->ce_namelen
= len
;
3284 new_ce
->ce_mode
= ce
->ce_mode
;
3285 if (add_index_entry(istate
, new_ce
, ADD_CACHE_SKIP_DFCHECK
))
3286 return error(_("%s: cannot drop to stage #0"),
3293 * Returns 1 if the path is an "other" path with respect to
3294 * the index; that is, the path is not mentioned in the index at all,
3295 * either as a file, a directory with some files in the index,
3296 * or as an unmerged entry.
3298 * We helpfully remove a trailing "/" from directories so that
3299 * the output of read_directory can be used as-is.
3301 int index_name_is_other(const struct index_state
*istate
, const char *name
,
3305 if (namelen
&& name
[namelen
- 1] == '/')
3307 pos
= index_name_pos(istate
, name
, namelen
);
3309 return 0; /* exact match */
3311 if (pos
< istate
->cache_nr
) {
3312 struct cache_entry
*ce
= istate
->cache
[pos
];
3313 if (ce_namelen(ce
) == namelen
&&
3314 !memcmp(ce
->name
, name
, namelen
))
3315 return 0; /* Yup, this one exists unmerged */
3320 void *read_blob_data_from_index(const struct index_state
*istate
,
3321 const char *path
, unsigned long *size
)
3325 enum object_type type
;
3329 pos
= index_name_pos(istate
, path
, len
);
3332 * We might be in the middle of a merge, in which
3333 * case we would read stage #2 (ours).
3337 (pos
< 0 && i
< istate
->cache_nr
&&
3338 !strcmp(istate
->cache
[i
]->name
, path
));
3340 if (ce_stage(istate
->cache
[i
]) == 2)
3345 data
= read_object_file(&istate
->cache
[pos
]->oid
, &type
, &sz
);
3346 if (!data
|| type
!= OBJ_BLOB
) {
3355 void stat_validity_clear(struct stat_validity
*sv
)
3357 FREE_AND_NULL(sv
->sd
);
3360 int stat_validity_check(struct stat_validity
*sv
, const char *path
)
3364 if (stat(path
, &st
) < 0)
3365 return sv
->sd
== NULL
;
3368 return S_ISREG(st
.st_mode
) && !match_stat_data(sv
->sd
, &st
);
3371 void stat_validity_update(struct stat_validity
*sv
, int fd
)
3375 if (fstat(fd
, &st
) < 0 || !S_ISREG(st
.st_mode
))
3376 stat_validity_clear(sv
);
3379 sv
->sd
= xcalloc(1, sizeof(struct stat_data
));
3380 fill_stat_data(sv
->sd
, &st
);
3384 void move_index_extensions(struct index_state
*dst
, struct index_state
*src
)
3386 dst
->untracked
= src
->untracked
;
3387 src
->untracked
= NULL
;
3388 dst
->cache_tree
= src
->cache_tree
;
3389 src
->cache_tree
= NULL
;
3392 struct cache_entry
*dup_cache_entry(const struct cache_entry
*ce
,
3393 struct index_state
*istate
)
3395 unsigned int size
= ce_size(ce
);
3396 int mem_pool_allocated
;
3397 struct cache_entry
*new_entry
= make_empty_cache_entry(istate
, ce_namelen(ce
));
3398 mem_pool_allocated
= new_entry
->mem_pool_allocated
;
3400 memcpy(new_entry
, ce
, size
);
3401 new_entry
->mem_pool_allocated
= mem_pool_allocated
;
3405 void discard_cache_entry(struct cache_entry
*ce
)
3407 if (ce
&& should_validate_cache_entries())
3408 memset(ce
, 0xCD, cache_entry_size(ce
->ce_namelen
));
3410 if (ce
&& ce
->mem_pool_allocated
)
3416 int should_validate_cache_entries(void)
3418 static int validate_index_cache_entries
= -1;
3420 if (validate_index_cache_entries
< 0) {
3421 if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))
3422 validate_index_cache_entries
= 1;
3424 validate_index_cache_entries
= 0;
3427 return validate_index_cache_entries
;
3430 #define EOIE_SIZE (4 + GIT_SHA1_RAWSZ) /* <4-byte offset> + <20-byte hash> */
3431 #define EOIE_SIZE_WITH_HEADER (4 + 4 + EOIE_SIZE) /* <4-byte signature> + <4-byte length> + EOIE_SIZE */
3433 static size_t read_eoie_extension(const char *mmap
, size_t mmap_size
)
3436 * The end of index entries (EOIE) extension is guaranteed to be last
3437 * so that it can be found by scanning backwards from the EOF.
3444 const char *index
, *eoie
;
3446 size_t offset
, src_offset
;
3447 unsigned char hash
[GIT_MAX_RAWSZ
];
3450 /* ensure we have an index big enough to contain an EOIE extension */
3451 if (mmap_size
< sizeof(struct cache_header
) + EOIE_SIZE_WITH_HEADER
+ the_hash_algo
->rawsz
)
3454 /* validate the extension signature */
3455 index
= eoie
= mmap
+ mmap_size
- EOIE_SIZE_WITH_HEADER
- the_hash_algo
->rawsz
;
3456 if (CACHE_EXT(index
) != CACHE_EXT_ENDOFINDEXENTRIES
)
3458 index
+= sizeof(uint32_t);
3460 /* validate the extension size */
3461 extsize
= get_be32(index
);
3462 if (extsize
!= EOIE_SIZE
)
3464 index
+= sizeof(uint32_t);
3467 * Validate the offset we're going to look for the first extension
3468 * signature is after the index header and before the eoie extension.
3470 offset
= get_be32(index
);
3471 if (mmap
+ offset
< mmap
+ sizeof(struct cache_header
))
3473 if (mmap
+ offset
>= eoie
)
3475 index
+= sizeof(uint32_t);
3478 * The hash is computed over extension types and their sizes (but not
3479 * their contents). E.g. if we have "TREE" extension that is N-bytes
3480 * long, "REUC" extension that is M-bytes long, followed by "EOIE",
3481 * then the hash would be:
3483 * SHA-1("TREE" + <binary representation of N> +
3484 * "REUC" + <binary representation of M>)
3486 src_offset
= offset
;
3487 the_hash_algo
->init_fn(&c
);
3488 while (src_offset
< mmap_size
- the_hash_algo
->rawsz
- EOIE_SIZE_WITH_HEADER
) {
3489 /* After an array of active_nr index entries,
3490 * there can be arbitrary number of extended
3491 * sections, each of which is prefixed with
3492 * extension name (4-byte) and section length
3493 * in 4-byte network byte order.
3496 memcpy(&extsize
, mmap
+ src_offset
+ 4, 4);
3497 extsize
= ntohl(extsize
);
3499 /* verify the extension size isn't so large it will wrap around */
3500 if (src_offset
+ 8 + extsize
< src_offset
)
3503 the_hash_algo
->update_fn(&c
, mmap
+ src_offset
, 8);
3506 src_offset
+= extsize
;
3508 the_hash_algo
->final_fn(hash
, &c
);
3509 if (!hasheq(hash
, (const unsigned char *)index
))
3512 /* Validate that the extension offsets returned us back to the eoie extension. */
3513 if (src_offset
!= mmap_size
- the_hash_algo
->rawsz
- EOIE_SIZE_WITH_HEADER
)
3519 static void write_eoie_extension(struct strbuf
*sb
, git_hash_ctx
*eoie_context
, size_t offset
)
3522 unsigned char hash
[GIT_MAX_RAWSZ
];
3525 put_be32(&buffer
, offset
);
3526 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3529 the_hash_algo
->final_fn(hash
, eoie_context
);
3530 strbuf_add(sb
, hash
, the_hash_algo
->rawsz
);
3533 #define IEOT_VERSION (1)
3535 static struct index_entry_offset_table
*read_ieot_extension(const char *mmap
, size_t mmap_size
, size_t offset
)
3537 const char *index
= NULL
;
3538 uint32_t extsize
, ext_version
;
3539 struct index_entry_offset_table
*ieot
;
3542 /* find the IEOT extension */
3545 while (offset
<= mmap_size
- the_hash_algo
->rawsz
- 8) {
3546 extsize
= get_be32(mmap
+ offset
+ 4);
3547 if (CACHE_EXT((mmap
+ offset
)) == CACHE_EXT_INDEXENTRYOFFSETTABLE
) {
3548 index
= mmap
+ offset
+ 4 + 4;
3557 /* validate the version is IEOT_VERSION */
3558 ext_version
= get_be32(index
);
3559 if (ext_version
!= IEOT_VERSION
) {
3560 error("invalid IEOT version %d", ext_version
);
3563 index
+= sizeof(uint32_t);
3565 /* extension size - version bytes / bytes per entry */
3566 nr
= (extsize
- sizeof(uint32_t)) / (sizeof(uint32_t) + sizeof(uint32_t));
3568 error("invalid number of IEOT entries %d", nr
);
3571 ieot
= xmalloc(sizeof(struct index_entry_offset_table
)
3572 + (nr
* sizeof(struct index_entry_offset
)));
3574 for (i
= 0; i
< nr
; i
++) {
3575 ieot
->entries
[i
].offset
= get_be32(index
);
3576 index
+= sizeof(uint32_t);
3577 ieot
->entries
[i
].nr
= get_be32(index
);
3578 index
+= sizeof(uint32_t);
3584 static void write_ieot_extension(struct strbuf
*sb
, struct index_entry_offset_table
*ieot
)
3590 put_be32(&buffer
, IEOT_VERSION
);
3591 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3594 for (i
= 0; i
< ieot
->nr
; i
++) {
3597 put_be32(&buffer
, ieot
->entries
[i
].offset
);
3598 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3601 put_be32(&buffer
, ieot
->entries
[i
].nr
);
3602 strbuf_add(sb
, &buffer
, sizeof(uint32_t));