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"
20 #include "run-command.h"
23 #include "split-index.h"
25 #include "fsmonitor.h"
26 #include "thread-utils.h"
28 #include "sparse-index.h"
29 #include "csum-file.h"
30 #include "promisor-remote.h"
33 /* Mask for the name length in ce_flags in the on-disk index */
35 #define CE_NAMEMASK (0x0fff)
39 * The first letter should be 'A'..'Z' for extensions that are not
40 * necessary for a correct operation (i.e. optimization data).
41 * When new extensions are added that _needs_ to be understood in
42 * order to correctly interpret the index file, pick character that
43 * is outside the range, to cause the reader to abort.
46 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
47 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
48 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
49 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
50 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
51 #define CACHE_EXT_FSMONITOR 0x46534D4E /* "FSMN" */
52 #define CACHE_EXT_ENDOFINDEXENTRIES 0x454F4945 /* "EOIE" */
53 #define CACHE_EXT_INDEXENTRYOFFSETTABLE 0x49454F54 /* "IEOT" */
54 #define CACHE_EXT_SPARSE_DIRECTORIES 0x73646972 /* "sdir" */
56 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
57 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
58 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
59 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED | FSMONITOR_CHANGED)
63 * This is an estimate of the pathname length in the index. We use
64 * this for V4 index files to guess the un-deltafied size of the index
65 * in memory because of pathname deltafication. This is not required
66 * for V2/V3 index formats because their pathnames are not compressed.
67 * If the initial amount of memory set aside is not sufficient, the
68 * mem pool will allocate extra memory.
70 #define CACHE_ENTRY_PATH_LENGTH 80
72 enum index_search_mode
{
77 static inline struct cache_entry
*mem_pool__ce_alloc(struct mem_pool
*mem_pool
, size_t len
)
79 struct cache_entry
*ce
;
80 ce
= mem_pool_alloc(mem_pool
, cache_entry_size(len
));
81 ce
->mem_pool_allocated
= 1;
85 static inline struct cache_entry
*mem_pool__ce_calloc(struct mem_pool
*mem_pool
, size_t len
)
87 struct cache_entry
* ce
;
88 ce
= mem_pool_calloc(mem_pool
, 1, cache_entry_size(len
));
89 ce
->mem_pool_allocated
= 1;
93 static struct mem_pool
*find_mem_pool(struct index_state
*istate
)
95 struct mem_pool
**pool_ptr
;
97 if (istate
->split_index
&& istate
->split_index
->base
)
98 pool_ptr
= &istate
->split_index
->base
->ce_mem_pool
;
100 pool_ptr
= &istate
->ce_mem_pool
;
103 *pool_ptr
= xmalloc(sizeof(**pool_ptr
));
104 mem_pool_init(*pool_ptr
, 0);
110 static const char *alternate_index_output
;
112 static void set_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
114 if (S_ISSPARSEDIR(ce
->ce_mode
))
115 istate
->sparse_index
= INDEX_COLLAPSED
;
117 istate
->cache
[nr
] = ce
;
118 add_name_hash(istate
, ce
);
121 static void replace_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
123 struct cache_entry
*old
= istate
->cache
[nr
];
125 replace_index_entry_in_base(istate
, old
, ce
);
126 remove_name_hash(istate
, old
);
127 discard_cache_entry(old
);
128 ce
->ce_flags
&= ~CE_HASHED
;
129 set_index_entry(istate
, nr
, ce
);
130 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
131 mark_fsmonitor_invalid(istate
, ce
);
132 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
135 void rename_index_entry_at(struct index_state
*istate
, int nr
, const char *new_name
)
137 struct cache_entry
*old_entry
= istate
->cache
[nr
], *new_entry
, *refreshed
;
138 int namelen
= strlen(new_name
);
140 new_entry
= make_empty_cache_entry(istate
, namelen
);
141 copy_cache_entry(new_entry
, old_entry
);
142 new_entry
->ce_flags
&= ~CE_HASHED
;
143 new_entry
->ce_namelen
= namelen
;
144 new_entry
->index
= 0;
145 memcpy(new_entry
->name
, new_name
, namelen
+ 1);
147 cache_tree_invalidate_path(istate
, old_entry
->name
);
148 untracked_cache_remove_from_index(istate
, old_entry
->name
);
149 remove_index_entry_at(istate
, nr
);
152 * Refresh the new index entry. Using 'refresh_cache_entry' ensures
153 * we only update stat info if the entry is otherwise up-to-date (i.e.,
154 * the contents/mode haven't changed). This ensures that we reflect the
155 * 'ctime' of the rename in the index without (incorrectly) updating
156 * the cached stat info to reflect unstaged changes on disk.
158 refreshed
= refresh_cache_entry(istate
, new_entry
, CE_MATCH_REFRESH
);
159 if (refreshed
&& refreshed
!= new_entry
) {
160 add_index_entry(istate
, refreshed
, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
161 discard_cache_entry(new_entry
);
163 add_index_entry(istate
, new_entry
, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
166 void fill_stat_data(struct stat_data
*sd
, struct stat
*st
)
168 sd
->sd_ctime
.sec
= (unsigned int)st
->st_ctime
;
169 sd
->sd_mtime
.sec
= (unsigned int)st
->st_mtime
;
170 sd
->sd_ctime
.nsec
= ST_CTIME_NSEC(*st
);
171 sd
->sd_mtime
.nsec
= ST_MTIME_NSEC(*st
);
172 sd
->sd_dev
= st
->st_dev
;
173 sd
->sd_ino
= st
->st_ino
;
174 sd
->sd_uid
= st
->st_uid
;
175 sd
->sd_gid
= st
->st_gid
;
176 sd
->sd_size
= st
->st_size
;
179 int match_stat_data(const struct stat_data
*sd
, struct stat
*st
)
183 if (sd
->sd_mtime
.sec
!= (unsigned int)st
->st_mtime
)
184 changed
|= MTIME_CHANGED
;
185 if (trust_ctime
&& check_stat
&&
186 sd
->sd_ctime
.sec
!= (unsigned int)st
->st_ctime
)
187 changed
|= CTIME_CHANGED
;
190 if (check_stat
&& sd
->sd_mtime
.nsec
!= ST_MTIME_NSEC(*st
))
191 changed
|= MTIME_CHANGED
;
192 if (trust_ctime
&& check_stat
&&
193 sd
->sd_ctime
.nsec
!= ST_CTIME_NSEC(*st
))
194 changed
|= CTIME_CHANGED
;
198 if (sd
->sd_uid
!= (unsigned int) st
->st_uid
||
199 sd
->sd_gid
!= (unsigned int) st
->st_gid
)
200 changed
|= OWNER_CHANGED
;
201 if (sd
->sd_ino
!= (unsigned int) st
->st_ino
)
202 changed
|= INODE_CHANGED
;
207 * st_dev breaks on network filesystems where different
208 * clients will have different views of what "device"
209 * the filesystem is on
211 if (check_stat
&& sd
->sd_dev
!= (unsigned int) st
->st_dev
)
212 changed
|= INODE_CHANGED
;
215 if (sd
->sd_size
!= (unsigned int) st
->st_size
)
216 changed
|= DATA_CHANGED
;
222 * This only updates the "non-critical" parts of the directory
223 * cache, ie the parts that aren't tracked by GIT, and only used
224 * to validate the cache.
226 void fill_stat_cache_info(struct index_state
*istate
, struct cache_entry
*ce
, struct stat
*st
)
228 fill_stat_data(&ce
->ce_stat_data
, st
);
230 if (assume_unchanged
)
231 ce
->ce_flags
|= CE_VALID
;
233 if (S_ISREG(st
->st_mode
)) {
234 ce_mark_uptodate(ce
);
235 mark_fsmonitor_valid(istate
, ce
);
239 static int ce_compare_data(struct index_state
*istate
,
240 const struct cache_entry
*ce
,
244 int fd
= git_open_cloexec(ce
->name
, O_RDONLY
);
247 struct object_id oid
;
248 if (!index_fd(istate
, &oid
, fd
, st
, OBJ_BLOB
, ce
->name
, 0))
249 match
= !oideq(&oid
, &ce
->oid
);
250 /* index_fd() closed the file descriptor already */
255 static int ce_compare_link(const struct cache_entry
*ce
, size_t expected_size
)
260 enum object_type type
;
261 struct strbuf sb
= STRBUF_INIT
;
263 if (strbuf_readlink(&sb
, ce
->name
, expected_size
))
266 buffer
= read_object_file(&ce
->oid
, &type
, &size
);
269 match
= memcmp(buffer
, sb
.buf
, size
);
276 static int ce_compare_gitlink(const struct cache_entry
*ce
)
278 struct object_id oid
;
281 * We don't actually require that the .git directory
282 * under GITLINK directory be a valid git directory. It
283 * might even be missing (in case nobody populated that
286 * If so, we consider it always to match.
288 if (resolve_gitlink_ref(ce
->name
, "HEAD", &oid
) < 0)
290 return !oideq(&oid
, &ce
->oid
);
293 static int ce_modified_check_fs(struct index_state
*istate
,
294 const struct cache_entry
*ce
,
297 switch (st
->st_mode
& S_IFMT
) {
299 if (ce_compare_data(istate
, ce
, st
))
303 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
307 if (S_ISGITLINK(ce
->ce_mode
))
308 return ce_compare_gitlink(ce
) ? DATA_CHANGED
: 0;
309 /* else fallthrough */
316 static int ce_match_stat_basic(const struct cache_entry
*ce
, struct stat
*st
)
318 unsigned int changed
= 0;
320 if (ce
->ce_flags
& CE_REMOVE
)
321 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
323 switch (ce
->ce_mode
& S_IFMT
) {
325 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
326 /* We consider only the owner x bit to be relevant for
329 if (trust_executable_bit
&&
330 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
331 changed
|= MODE_CHANGED
;
334 if (!S_ISLNK(st
->st_mode
) &&
335 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
336 changed
|= TYPE_CHANGED
;
339 /* We ignore most of the st_xxx fields for gitlinks */
340 if (!S_ISDIR(st
->st_mode
))
341 changed
|= TYPE_CHANGED
;
342 else if (ce_compare_gitlink(ce
))
343 changed
|= DATA_CHANGED
;
346 BUG("unsupported ce_mode: %o", ce
->ce_mode
);
349 changed
|= match_stat_data(&ce
->ce_stat_data
, st
);
351 /* Racily smudged entry? */
352 if (!ce
->ce_stat_data
.sd_size
) {
353 if (!is_empty_blob_sha1(ce
->oid
.hash
))
354 changed
|= DATA_CHANGED
;
360 static int is_racy_stat(const struct index_state
*istate
,
361 const struct stat_data
*sd
)
363 return (istate
->timestamp
.sec
&&
365 /* nanosecond timestamped files can also be racy! */
366 (istate
->timestamp
.sec
< sd
->sd_mtime
.sec
||
367 (istate
->timestamp
.sec
== sd
->sd_mtime
.sec
&&
368 istate
->timestamp
.nsec
<= sd
->sd_mtime
.nsec
))
370 istate
->timestamp
.sec
<= sd
->sd_mtime
.sec
375 int is_racy_timestamp(const struct index_state
*istate
,
376 const struct cache_entry
*ce
)
378 return (!S_ISGITLINK(ce
->ce_mode
) &&
379 is_racy_stat(istate
, &ce
->ce_stat_data
));
382 int match_stat_data_racy(const struct index_state
*istate
,
383 const struct stat_data
*sd
, struct stat
*st
)
385 if (is_racy_stat(istate
, sd
))
386 return MTIME_CHANGED
;
387 return match_stat_data(sd
, st
);
390 int ie_match_stat(struct index_state
*istate
,
391 const struct cache_entry
*ce
, struct stat
*st
,
392 unsigned int options
)
394 unsigned int changed
;
395 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
396 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
397 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
398 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
400 if (!ignore_fsmonitor
)
401 refresh_fsmonitor(istate
);
403 * If it's marked as always valid in the index, it's
404 * valid whatever the checked-out copy says.
406 * skip-worktree has the same effect with higher precedence
408 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
))
410 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
412 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
))
416 * Intent-to-add entries have not been added, so the index entry
417 * by definition never matches what is in the work tree until it
418 * actually gets added.
420 if (ce_intent_to_add(ce
))
421 return DATA_CHANGED
| TYPE_CHANGED
| MODE_CHANGED
;
423 changed
= ce_match_stat_basic(ce
, st
);
426 * Within 1 second of this sequence:
427 * echo xyzzy >file && git-update-index --add file
428 * running this command:
430 * would give a falsely clean cache entry. The mtime and
431 * length match the cache, and other stat fields do not change.
433 * We could detect this at update-index time (the cache entry
434 * being registered/updated records the same time as "now")
435 * and delay the return from git-update-index, but that would
436 * effectively mean we can make at most one commit per second,
437 * which is not acceptable. Instead, we check cache entries
438 * whose mtime are the same as the index file timestamp more
439 * carefully than others.
441 if (!changed
&& is_racy_timestamp(istate
, ce
)) {
442 if (assume_racy_is_modified
)
443 changed
|= DATA_CHANGED
;
445 changed
|= ce_modified_check_fs(istate
, ce
, st
);
451 int ie_modified(struct index_state
*istate
,
452 const struct cache_entry
*ce
,
453 struct stat
*st
, unsigned int options
)
455 int changed
, changed_fs
;
457 changed
= ie_match_stat(istate
, ce
, st
, options
);
461 * If the mode or type has changed, there's no point in trying
462 * to refresh the entry - it's not going to match
464 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
468 * Immediately after read-tree or update-index --cacheinfo,
469 * the length field is zero, as we have never even read the
470 * lstat(2) information once, and we cannot trust DATA_CHANGED
471 * returned by ie_match_stat() which in turn was returned by
472 * ce_match_stat_basic() to signal that the filesize of the
473 * blob changed. We have to actually go to the filesystem to
474 * see if the contents match, and if so, should answer "unchanged".
476 * The logic does not apply to gitlinks, as ce_match_stat_basic()
477 * already has checked the actual HEAD from the filesystem in the
478 * subproject. If ie_match_stat() already said it is different,
479 * then we know it is.
481 if ((changed
& DATA_CHANGED
) &&
482 (S_ISGITLINK(ce
->ce_mode
) || ce
->ce_stat_data
.sd_size
!= 0))
485 changed_fs
= ce_modified_check_fs(istate
, ce
, st
);
487 return changed
| changed_fs
;
491 int base_name_compare(const char *name1
, size_t len1
, int mode1
,
492 const char *name2
, size_t len2
, int mode2
)
494 unsigned char c1
, c2
;
495 size_t len
= len1
< len2
? len1
: len2
;
498 cmp
= memcmp(name1
, name2
, len
);
503 if (!c1
&& S_ISDIR(mode1
))
505 if (!c2
&& S_ISDIR(mode2
))
507 return (c1
< c2
) ? -1 : (c1
> c2
) ? 1 : 0;
511 * df_name_compare() is identical to base_name_compare(), except it
512 * compares conflicting directory/file entries as equal. Note that
513 * while a directory name compares as equal to a regular file, they
514 * then individually compare _differently_ to a filename that has
515 * a dot after the basename (because '\0' < '.' < '/').
517 * This is used by routines that want to traverse the git namespace
518 * but then handle conflicting entries together when possible.
520 int df_name_compare(const char *name1
, size_t len1
, int mode1
,
521 const char *name2
, size_t len2
, int mode2
)
523 unsigned char c1
, c2
;
524 size_t len
= len1
< len2
? len1
: len2
;
527 cmp
= memcmp(name1
, name2
, len
);
530 /* Directories and files compare equal (same length, same name) */
534 if (!c1
&& S_ISDIR(mode1
))
537 if (!c2
&& S_ISDIR(mode2
))
539 if (c1
== '/' && !c2
)
541 if (c2
== '/' && !c1
)
546 int name_compare(const char *name1
, size_t len1
, const char *name2
, size_t len2
)
548 size_t min_len
= (len1
< len2
) ? len1
: len2
;
549 int cmp
= memcmp(name1
, name2
, min_len
);
559 int cache_name_stage_compare(const char *name1
, int len1
, int stage1
, const char *name2
, int len2
, int stage2
)
563 cmp
= name_compare(name1
, len1
, name2
, len2
);
574 static int index_name_stage_pos(struct index_state
*istate
,
575 const char *name
, int namelen
,
577 enum index_search_mode search_mode
)
582 last
= istate
->cache_nr
;
583 while (last
> first
) {
584 int next
= first
+ ((last
- first
) >> 1);
585 struct cache_entry
*ce
= istate
->cache
[next
];
586 int cmp
= cache_name_stage_compare(name
, namelen
, stage
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
596 if (search_mode
== EXPAND_SPARSE
&& istate
->sparse_index
&&
598 /* Note: first <= istate->cache_nr */
599 struct cache_entry
*ce
= istate
->cache
[first
- 1];
602 * If we are in a sparse-index _and_ the entry before the
603 * insertion position is a sparse-directory entry that is
604 * an ancestor of 'name', then we need to expand the index
605 * and search again. This will only trigger once, because
606 * thereafter the index is fully expanded.
608 if (S_ISSPARSEDIR(ce
->ce_mode
) &&
609 ce_namelen(ce
) < namelen
&&
610 !strncmp(name
, ce
->name
, ce_namelen(ce
))) {
611 ensure_full_index(istate
);
612 return index_name_stage_pos(istate
, name
, namelen
, stage
, search_mode
);
619 int index_name_pos(struct index_state
*istate
, const char *name
, int namelen
)
621 return index_name_stage_pos(istate
, name
, namelen
, 0, EXPAND_SPARSE
);
624 int index_name_pos_sparse(struct index_state
*istate
, const char *name
, int namelen
)
626 return index_name_stage_pos(istate
, name
, namelen
, 0, NO_EXPAND_SPARSE
);
629 int index_entry_exists(struct index_state
*istate
, const char *name
, int namelen
)
631 return index_name_stage_pos(istate
, name
, namelen
, 0, NO_EXPAND_SPARSE
) >= 0;
634 int remove_index_entry_at(struct index_state
*istate
, int pos
)
636 struct cache_entry
*ce
= istate
->cache
[pos
];
638 record_resolve_undo(istate
, ce
);
639 remove_name_hash(istate
, ce
);
640 save_or_free_index_entry(istate
, ce
);
641 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
643 if (pos
>= istate
->cache_nr
)
645 MOVE_ARRAY(istate
->cache
+ pos
, istate
->cache
+ pos
+ 1,
646 istate
->cache_nr
- pos
);
651 * Remove all cache entries marked for removal, that is where
652 * CE_REMOVE is set in ce_flags. This is much more effective than
653 * calling remove_index_entry_at() for each entry to be removed.
655 void remove_marked_cache_entries(struct index_state
*istate
, int invalidate
)
657 struct cache_entry
**ce_array
= istate
->cache
;
660 for (i
= j
= 0; i
< istate
->cache_nr
; i
++) {
661 if (ce_array
[i
]->ce_flags
& CE_REMOVE
) {
663 cache_tree_invalidate_path(istate
,
665 untracked_cache_remove_from_index(istate
,
668 remove_name_hash(istate
, ce_array
[i
]);
669 save_or_free_index_entry(istate
, ce_array
[i
]);
672 ce_array
[j
++] = ce_array
[i
];
674 if (j
== istate
->cache_nr
)
676 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
677 istate
->cache_nr
= j
;
680 int remove_file_from_index(struct index_state
*istate
, const char *path
)
682 int pos
= index_name_pos(istate
, path
, strlen(path
));
685 cache_tree_invalidate_path(istate
, path
);
686 untracked_cache_remove_from_index(istate
, path
);
687 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
688 remove_index_entry_at(istate
, pos
);
692 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
694 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
697 static int index_name_pos_also_unmerged(struct index_state
*istate
,
698 const char *path
, int namelen
)
700 int pos
= index_name_pos(istate
, path
, namelen
);
701 struct cache_entry
*ce
;
706 /* maybe unmerged? */
708 if (pos
>= istate
->cache_nr
||
709 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
712 /* order of preference: stage 2, 1, 3 */
713 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
714 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
715 !compare_name(ce
, path
, namelen
))
720 static int different_name(struct cache_entry
*ce
, struct cache_entry
*alias
)
722 int len
= ce_namelen(ce
);
723 return ce_namelen(alias
) != len
|| memcmp(ce
->name
, alias
->name
, len
);
727 * If we add a filename that aliases in the cache, we will use the
728 * name that we already have - but we don't want to update the same
729 * alias twice, because that implies that there were actually two
730 * different files with aliasing names!
732 * So we use the CE_ADDED flag to verify that the alias was an old
733 * one before we accept it as
735 static struct cache_entry
*create_alias_ce(struct index_state
*istate
,
736 struct cache_entry
*ce
,
737 struct cache_entry
*alias
)
740 struct cache_entry
*new_entry
;
742 if (alias
->ce_flags
& CE_ADDED
)
743 die(_("will not add file alias '%s' ('%s' already exists in index)"),
744 ce
->name
, alias
->name
);
746 /* Ok, create the new entry using the name of the existing alias */
747 len
= ce_namelen(alias
);
748 new_entry
= make_empty_cache_entry(istate
, len
);
749 memcpy(new_entry
->name
, alias
->name
, len
);
750 copy_cache_entry(new_entry
, ce
);
751 save_or_free_index_entry(istate
, ce
);
755 void set_object_name_for_intent_to_add_entry(struct cache_entry
*ce
)
757 struct object_id oid
;
758 if (write_object_file("", 0, OBJ_BLOB
, &oid
))
759 die(_("cannot create an empty blob in the object database"));
760 oidcpy(&ce
->oid
, &oid
);
763 int add_to_index(struct index_state
*istate
, const char *path
, struct stat
*st
, int flags
)
765 int namelen
, was_same
;
766 mode_t st_mode
= st
->st_mode
;
767 struct cache_entry
*ce
, *alias
= NULL
;
768 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_IGNORE_SKIP_WORKTREE
|CE_MATCH_RACY_IS_DIRTY
;
769 int verbose
= flags
& (ADD_CACHE_VERBOSE
| ADD_CACHE_PRETEND
);
770 int pretend
= flags
& ADD_CACHE_PRETEND
;
771 int intent_only
= flags
& ADD_CACHE_INTENT
;
772 int add_option
= (ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
|
773 (intent_only
? ADD_CACHE_NEW_ONLY
: 0));
774 unsigned hash_flags
= pretend
? 0 : HASH_WRITE_OBJECT
;
775 struct object_id oid
;
777 if (flags
& ADD_CACHE_RENORMALIZE
)
778 hash_flags
|= HASH_RENORMALIZE
;
780 if (!S_ISREG(st_mode
) && !S_ISLNK(st_mode
) && !S_ISDIR(st_mode
))
781 return error(_("%s: can only add regular files, symbolic links or git-directories"), path
);
783 namelen
= strlen(path
);
784 if (S_ISDIR(st_mode
)) {
785 if (resolve_gitlink_ref(path
, "HEAD", &oid
) < 0)
786 return error(_("'%s' does not have a commit checked out"), path
);
787 while (namelen
&& path
[namelen
-1] == '/')
790 ce
= make_empty_cache_entry(istate
, namelen
);
791 memcpy(ce
->name
, path
, namelen
);
792 ce
->ce_namelen
= namelen
;
794 fill_stat_cache_info(istate
, ce
, st
);
796 ce
->ce_flags
|= CE_INTENT_TO_ADD
;
799 if (trust_executable_bit
&& has_symlinks
) {
800 ce
->ce_mode
= create_ce_mode(st_mode
);
802 /* If there is an existing entry, pick the mode bits and type
803 * from it, otherwise assume unexecutable regular file.
805 struct cache_entry
*ent
;
806 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
808 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
809 ce
->ce_mode
= ce_mode_from_stat(ent
, st_mode
);
812 /* When core.ignorecase=true, determine if a directory of the same name but differing
813 * case already exists within the Git repository. If it does, ensure the directory
814 * case of the file being added to the repository matches (is folded into) the existing
815 * entry's directory case.
818 adjust_dirname_case(istate
, ce
->name
);
820 if (!(flags
& ADD_CACHE_RENORMALIZE
)) {
821 alias
= index_file_exists(istate
, ce
->name
,
822 ce_namelen(ce
), ignore_case
);
825 !ie_match_stat(istate
, alias
, st
, ce_option
)) {
826 /* Nothing changed, really */
827 if (!S_ISGITLINK(alias
->ce_mode
))
828 ce_mark_uptodate(alias
);
829 alias
->ce_flags
|= CE_ADDED
;
831 discard_cache_entry(ce
);
836 if (index_path(istate
, &ce
->oid
, path
, st
, hash_flags
)) {
837 discard_cache_entry(ce
);
838 return error(_("unable to index file '%s'"), path
);
841 set_object_name_for_intent_to_add_entry(ce
);
843 if (ignore_case
&& alias
&& different_name(ce
, alias
))
844 ce
= create_alias_ce(istate
, ce
, alias
);
845 ce
->ce_flags
|= CE_ADDED
;
847 /* It was suspected to be racily clean, but it turns out to be Ok */
850 oideq(&alias
->oid
, &ce
->oid
) &&
851 ce
->ce_mode
== alias
->ce_mode
);
854 discard_cache_entry(ce
);
855 else if (add_index_entry(istate
, ce
, add_option
)) {
856 discard_cache_entry(ce
);
857 return error(_("unable to add '%s' to index"), path
);
859 if (verbose
&& !was_same
)
860 printf("add '%s'\n", path
);
864 int add_file_to_index(struct index_state
*istate
, const char *path
, int flags
)
867 if (lstat(path
, &st
))
868 die_errno(_("unable to stat '%s'"), path
);
869 return add_to_index(istate
, path
, &st
, flags
);
872 struct cache_entry
*make_empty_cache_entry(struct index_state
*istate
, size_t len
)
874 return mem_pool__ce_calloc(find_mem_pool(istate
), len
);
877 struct cache_entry
*make_empty_transient_cache_entry(size_t len
,
878 struct mem_pool
*ce_mem_pool
)
881 return mem_pool__ce_calloc(ce_mem_pool
, len
);
882 return xcalloc(1, cache_entry_size(len
));
885 enum verify_path_result
{
891 static enum verify_path_result
verify_path_internal(const char *, unsigned);
893 int verify_path(const char *path
, unsigned mode
)
895 return verify_path_internal(path
, mode
) == PATH_OK
;
898 struct cache_entry
*make_cache_entry(struct index_state
*istate
,
900 const struct object_id
*oid
,
903 unsigned int refresh_options
)
905 struct cache_entry
*ce
, *ret
;
908 if (verify_path_internal(path
, mode
) == PATH_INVALID
) {
909 error(_("invalid path '%s'"), path
);
914 ce
= make_empty_cache_entry(istate
, len
);
916 oidcpy(&ce
->oid
, oid
);
917 memcpy(ce
->name
, path
, len
);
918 ce
->ce_flags
= create_ce_flags(stage
);
919 ce
->ce_namelen
= len
;
920 ce
->ce_mode
= create_ce_mode(mode
);
922 ret
= refresh_cache_entry(istate
, ce
, refresh_options
);
924 discard_cache_entry(ce
);
928 struct cache_entry
*make_transient_cache_entry(unsigned int mode
,
929 const struct object_id
*oid
,
932 struct mem_pool
*ce_mem_pool
)
934 struct cache_entry
*ce
;
937 if (!verify_path(path
, mode
)) {
938 error(_("invalid path '%s'"), path
);
943 ce
= make_empty_transient_cache_entry(len
, ce_mem_pool
);
945 oidcpy(&ce
->oid
, oid
);
946 memcpy(ce
->name
, path
, len
);
947 ce
->ce_flags
= create_ce_flags(stage
);
948 ce
->ce_namelen
= len
;
949 ce
->ce_mode
= create_ce_mode(mode
);
955 * Chmod an index entry with either +x or -x.
957 * Returns -1 if the chmod for the particular cache entry failed (if it's
958 * not a regular file), -2 if an invalid flip argument is passed in, 0
961 int chmod_index_entry(struct index_state
*istate
, struct cache_entry
*ce
,
964 if (!S_ISREG(ce
->ce_mode
))
971 ce
->ce_mode
&= ~0111;
976 cache_tree_invalidate_path(istate
, ce
->name
);
977 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
978 mark_fsmonitor_invalid(istate
, ce
);
979 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
984 int ce_same_name(const struct cache_entry
*a
, const struct cache_entry
*b
)
986 int len
= ce_namelen(a
);
987 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
991 * We fundamentally don't like some paths: we don't want
992 * dot or dot-dot anywhere, and for obvious reasons don't
993 * want to recurse into ".git" either.
995 * Also, we don't want double slashes or slashes at the
996 * end that can make pathnames ambiguous.
998 static int verify_dotfile(const char *rest
, unsigned mode
)
1001 * The first character was '.', but that
1002 * has already been discarded, we now test
1006 /* "." is not allowed */
1007 if (*rest
== '\0' || is_dir_sep(*rest
))
1012 * ".git" followed by NUL or slash is bad. Note that we match
1013 * case-insensitively here, even if ignore_case is not set.
1014 * This outlaws ".GIT" everywhere out of an abundance of caution,
1015 * since there's really no good reason to allow it.
1017 * Once we've seen ".git", we can also find ".gitmodules", etc (also
1018 * case-insensitively).
1022 if (rest
[1] != 'i' && rest
[1] != 'I')
1024 if (rest
[2] != 't' && rest
[2] != 'T')
1026 if (rest
[3] == '\0' || is_dir_sep(rest
[3]))
1028 if (S_ISLNK(mode
)) {
1030 if (skip_iprefix(rest
, "modules", &rest
) &&
1031 (*rest
== '\0' || is_dir_sep(*rest
)))
1036 if (rest
[1] == '\0' || is_dir_sep(rest
[1]))
1042 static enum verify_path_result
verify_path_internal(const char *path
,
1047 if (has_dos_drive_prefix(path
))
1048 return PATH_INVALID
;
1050 if (!is_valid_path(path
))
1051 return PATH_INVALID
;
1057 if (is_dir_sep(c
)) {
1061 if (is_hfs_dotgit(path
))
1062 return PATH_INVALID
;
1063 if (S_ISLNK(mode
)) {
1064 if (is_hfs_dotgitmodules(path
))
1065 return PATH_INVALID
;
1069 #if defined GIT_WINDOWS_NATIVE || defined __CYGWIN__
1071 return PATH_INVALID
;
1073 if (is_ntfs_dotgit(path
))
1074 return PATH_INVALID
;
1075 if (S_ISLNK(mode
)) {
1076 if (is_ntfs_dotgitmodules(path
))
1077 return PATH_INVALID
;
1082 if ((c
== '.' && !verify_dotfile(path
, mode
)) ||
1084 return PATH_INVALID
;
1086 * allow terminating directory separators for
1087 * sparse directory entries.
1090 return S_ISDIR(mode
) ? PATH_DIR_WITH_SEP
:
1092 } else if (c
== '\\' && protect_ntfs
) {
1093 if (is_ntfs_dotgit(path
))
1094 return PATH_INVALID
;
1095 if (S_ISLNK(mode
)) {
1096 if (is_ntfs_dotgitmodules(path
))
1097 return PATH_INVALID
;
1106 * Do we have another file that has the beginning components being a
1107 * proper superset of the name we're trying to add?
1109 static int has_file_name(struct index_state
*istate
,
1110 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
1113 int len
= ce_namelen(ce
);
1114 int stage
= ce_stage(ce
);
1115 const char *name
= ce
->name
;
1117 while (pos
< istate
->cache_nr
) {
1118 struct cache_entry
*p
= istate
->cache
[pos
++];
1120 if (len
>= ce_namelen(p
))
1122 if (memcmp(name
, p
->name
, len
))
1124 if (ce_stage(p
) != stage
)
1126 if (p
->name
[len
] != '/')
1128 if (p
->ce_flags
& CE_REMOVE
)
1133 remove_index_entry_at(istate
, --pos
);
1140 * Like strcmp(), but also return the offset of the first change.
1141 * If strings are equal, return the length.
1143 int strcmp_offset(const char *s1
, const char *s2
, size_t *first_change
)
1148 return strcmp(s1
, s2
);
1150 for (k
= 0; s1
[k
] == s2
[k
]; k
++)
1155 return (unsigned char)s1
[k
] - (unsigned char)s2
[k
];
1159 * Do we have another file with a pathname that is a proper
1160 * subset of the name we're trying to add?
1162 * That is, is there another file in the index with a path
1163 * that matches a sub-directory in the given entry?
1165 static int has_dir_name(struct index_state
*istate
,
1166 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
1169 int stage
= ce_stage(ce
);
1170 const char *name
= ce
->name
;
1171 const char *slash
= name
+ ce_namelen(ce
);
1176 * We are frequently called during an iteration on a sorted
1177 * list of pathnames and while building a new index. Therefore,
1178 * there is a high probability that this entry will eventually
1179 * be appended to the index, rather than inserted in the middle.
1180 * If we can confirm that, we can avoid binary searches on the
1181 * components of the pathname.
1183 * Compare the entry's full path with the last path in the index.
1185 if (istate
->cache_nr
> 0) {
1186 cmp_last
= strcmp_offset(name
,
1187 istate
->cache
[istate
->cache_nr
- 1]->name
,
1190 if (name
[len_eq_last
] != '/') {
1192 * The entry sorts AFTER the last one in the
1195 * If there were a conflict with "file", then our
1196 * name would start with "file/" and the last index
1197 * entry would start with "file" but not "file/".
1199 * The next character after common prefix is
1200 * not '/', so there can be no conflict.
1205 * The entry sorts AFTER the last one in the
1206 * index, and the next character after common
1209 * Either the last index entry is a file in
1210 * conflict with this entry, or it has a name
1211 * which sorts between this entry and the
1212 * potential conflicting file.
1214 * In both cases, we fall through to the loop
1215 * below and let the regular search code handle it.
1218 } else if (cmp_last
== 0) {
1220 * The entry exactly matches the last one in the
1221 * index, but because of multiple stage and CE_REMOVE
1222 * items, we fall through and let the regular search
1232 if (*--slash
== '/')
1234 if (slash
<= ce
->name
)
1239 pos
= index_name_stage_pos(istate
, name
, len
, stage
, EXPAND_SPARSE
);
1242 * Found one, but not so fast. This could
1243 * be a marker that says "I was here, but
1244 * I am being removed". Such an entry is
1245 * not a part of the resulting tree, and
1246 * it is Ok to have a directory at the same
1249 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
1253 remove_index_entry_at(istate
, pos
);
1261 * Trivial optimization: if we find an entry that
1262 * already matches the sub-directory, then we know
1263 * we're ok, and we can exit.
1265 while (pos
< istate
->cache_nr
) {
1266 struct cache_entry
*p
= istate
->cache
[pos
];
1267 if ((ce_namelen(p
) <= len
) ||
1268 (p
->name
[len
] != '/') ||
1269 memcmp(p
->name
, name
, len
))
1270 break; /* not our subdirectory */
1271 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
1273 * p is at the same stage as our entry, and
1274 * is a subdirectory of what we are looking
1275 * at, so we cannot have conflicts at our
1276 * level or anything shorter.
1285 /* We may be in a situation where we already have path/file and path
1286 * is being added, or we already have path and path/file is being
1287 * added. Either one would result in a nonsense tree that has path
1288 * twice when git-write-tree tries to write it out. Prevent it.
1290 * If ok-to-replace is specified, we remove the conflicting entries
1291 * from the cache so the caller should recompute the insert position.
1292 * When this happens, we return non-zero.
1294 static int check_file_directory_conflict(struct index_state
*istate
,
1295 const struct cache_entry
*ce
,
1296 int pos
, int ok_to_replace
)
1301 * When ce is an "I am going away" entry, we allow it to be added
1303 if (ce
->ce_flags
& CE_REMOVE
)
1307 * We check if the path is a sub-path of a subsequent pathname
1308 * first, since removing those will not change the position
1311 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
1314 * Then check if the path might have a clashing sub-directory
1317 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
1320 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1323 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
1324 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
1325 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
1326 int new_only
= option
& ADD_CACHE_NEW_ONLY
;
1329 * If this entry's path sorts after the last entry in the index,
1330 * we can avoid searching for it.
1332 if (istate
->cache_nr
> 0 &&
1333 strcmp(ce
->name
, istate
->cache
[istate
->cache_nr
- 1]->name
) > 0)
1334 pos
= index_pos_to_insert_pos(istate
->cache_nr
);
1336 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
), EXPAND_SPARSE
);
1339 * Cache tree path should be invalidated only after index_name_stage_pos,
1340 * in case it expands a sparse index.
1342 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1343 cache_tree_invalidate_path(istate
, ce
->name
);
1345 /* existing match? Just replace it. */
1348 replace_index_entry(istate
, pos
, ce
);
1353 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1354 untracked_cache_add_to_index(istate
, ce
->name
);
1357 * Inserting a merged entry ("stage 0") into the index
1358 * will always replace all non-merged entries..
1360 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
1361 while (ce_same_name(istate
->cache
[pos
], ce
)) {
1363 if (!remove_index_entry_at(istate
, pos
))
1370 if (verify_path_internal(ce
->name
, ce
->ce_mode
) == PATH_INVALID
)
1371 return error(_("invalid path '%s'"), ce
->name
);
1373 if (!skip_df_check
&&
1374 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
1376 return error(_("'%s' appears as both a file and as a directory"),
1378 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
), EXPAND_SPARSE
);
1384 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1388 if (option
& ADD_CACHE_JUST_APPEND
)
1389 pos
= istate
->cache_nr
;
1392 ret
= add_index_entry_with_check(istate
, ce
, option
);
1398 /* Make sure the array is big enough .. */
1399 ALLOC_GROW(istate
->cache
, istate
->cache_nr
+ 1, istate
->cache_alloc
);
1403 if (istate
->cache_nr
> pos
+ 1)
1404 MOVE_ARRAY(istate
->cache
+ pos
+ 1, istate
->cache
+ pos
,
1405 istate
->cache_nr
- pos
- 1);
1406 set_index_entry(istate
, pos
, ce
);
1407 istate
->cache_changed
|= CE_ENTRY_ADDED
;
1412 * "refresh" does not calculate a new sha1 file or bring the
1413 * cache up-to-date for mode/content changes. But what it
1414 * _does_ do is to "re-match" the stat information of a file
1415 * with the cache, so that you can refresh the cache for a
1416 * file that hasn't been changed but where the stat entry is
1419 * For example, you'd want to do this after doing a "git-read-tree",
1420 * to link up the stat cache details with the proper files.
1422 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
1423 struct cache_entry
*ce
,
1424 unsigned int options
, int *err
,
1430 struct cache_entry
*updated
;
1432 int refresh
= options
& CE_MATCH_REFRESH
;
1433 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
1434 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
1435 int ignore_missing
= options
& CE_MATCH_IGNORE_MISSING
;
1436 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
1438 if (!refresh
|| ce_uptodate(ce
))
1441 if (!ignore_fsmonitor
)
1442 refresh_fsmonitor(istate
);
1444 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1445 * that the change to the work tree does not matter and told
1448 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
)) {
1449 ce_mark_uptodate(ce
);
1452 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
)) {
1453 ce_mark_uptodate(ce
);
1456 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
)) {
1457 ce_mark_uptodate(ce
);
1461 if (has_symlink_leading_path(ce
->name
, ce_namelen(ce
))) {
1471 if (lstat(ce
->name
, &st
) < 0) {
1472 if (ignore_missing
&& errno
== ENOENT
)
1479 changed
= ie_match_stat(istate
, ce
, &st
, options
);
1481 *changed_ret
= changed
;
1484 * The path is unchanged. If we were told to ignore
1485 * valid bit, then we did the actual stat check and
1486 * found that the entry is unmodified. If the entry
1487 * is not marked VALID, this is the place to mark it
1488 * valid again, under "assume unchanged" mode.
1490 if (ignore_valid
&& assume_unchanged
&&
1491 !(ce
->ce_flags
& CE_VALID
))
1492 ; /* mark this one VALID again */
1495 * We do not mark the index itself "modified"
1496 * because CE_UPTODATE flag is in-core only;
1497 * we are not going to write this change out.
1499 if (!S_ISGITLINK(ce
->ce_mode
)) {
1500 ce_mark_uptodate(ce
);
1501 mark_fsmonitor_valid(istate
, ce
);
1509 if (ie_modified(istate
, ce
, &st
, options
)) {
1515 updated
= make_empty_cache_entry(istate
, ce_namelen(ce
));
1516 copy_cache_entry(updated
, ce
);
1517 memcpy(updated
->name
, ce
->name
, ce
->ce_namelen
+ 1);
1518 fill_stat_cache_info(istate
, updated
, &st
);
1520 * If ignore_valid is not set, we should leave CE_VALID bit
1521 * alone. Otherwise, paths marked with --no-assume-unchanged
1522 * (i.e. things to be edited) will reacquire CE_VALID bit
1523 * automatically, which is not really what we want.
1525 if (!ignore_valid
&& assume_unchanged
&&
1526 !(ce
->ce_flags
& CE_VALID
))
1527 updated
->ce_flags
&= ~CE_VALID
;
1529 /* istate->cache_changed is updated in the caller */
1533 static void show_file(const char * fmt
, const char * name
, int in_porcelain
,
1534 int * first
, const char *header_msg
)
1536 if (in_porcelain
&& *first
&& header_msg
) {
1537 printf("%s\n", header_msg
);
1543 int repo_refresh_and_write_index(struct repository
*repo
,
1544 unsigned int refresh_flags
,
1545 unsigned int write_flags
,
1547 const struct pathspec
*pathspec
,
1548 char *seen
, const char *header_msg
)
1550 struct lock_file lock_file
= LOCK_INIT
;
1553 fd
= repo_hold_locked_index(repo
, &lock_file
, 0);
1554 if (!gentle
&& fd
< 0)
1556 if (refresh_index(repo
->index
, refresh_flags
, pathspec
, seen
, header_msg
))
1558 if (0 <= fd
&& write_locked_index(repo
->index
, &lock_file
, COMMIT_LOCK
| write_flags
))
1564 int refresh_index(struct index_state
*istate
, unsigned int flags
,
1565 const struct pathspec
*pathspec
,
1566 char *seen
, const char *header_msg
)
1570 int really
= (flags
& REFRESH_REALLY
) != 0;
1571 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
1572 int quiet
= (flags
& REFRESH_QUIET
) != 0;
1573 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
1574 int ignore_submodules
= (flags
& REFRESH_IGNORE_SUBMODULES
) != 0;
1575 int ignore_skip_worktree
= (flags
& REFRESH_IGNORE_SKIP_WORKTREE
) != 0;
1577 int in_porcelain
= (flags
& REFRESH_IN_PORCELAIN
);
1578 unsigned int options
= (CE_MATCH_REFRESH
|
1579 (really
? CE_MATCH_IGNORE_VALID
: 0) |
1580 (not_new
? CE_MATCH_IGNORE_MISSING
: 0));
1581 const char *modified_fmt
;
1582 const char *deleted_fmt
;
1583 const char *typechange_fmt
;
1584 const char *added_fmt
;
1585 const char *unmerged_fmt
;
1586 struct progress
*progress
= NULL
;
1587 int t2_sum_lstat
= 0;
1588 int t2_sum_scan
= 0;
1590 if (flags
& REFRESH_PROGRESS
&& isatty(2))
1591 progress
= start_delayed_progress(_("Refresh index"),
1594 trace_performance_enter();
1595 modified_fmt
= in_porcelain
? "M\t%s\n" : "%s: needs update\n";
1596 deleted_fmt
= in_porcelain
? "D\t%s\n" : "%s: needs update\n";
1597 typechange_fmt
= in_porcelain
? "T\t%s\n" : "%s: needs update\n";
1598 added_fmt
= in_porcelain
? "A\t%s\n" : "%s: needs update\n";
1599 unmerged_fmt
= in_porcelain
? "U\t%s\n" : "%s: needs merge\n";
1601 * Use the multi-threaded preload_index() to refresh most of the
1602 * cache entries quickly then in the single threaded loop below,
1603 * we only have to do the special cases that are left.
1605 preload_index(istate
, pathspec
, 0);
1606 trace2_region_enter("index", "refresh", NULL
);
1608 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1609 struct cache_entry
*ce
, *new_entry
;
1610 int cache_errno
= 0;
1613 int t2_did_lstat
= 0;
1614 int t2_did_scan
= 0;
1616 ce
= istate
->cache
[i
];
1617 if (ignore_submodules
&& S_ISGITLINK(ce
->ce_mode
))
1619 if (ignore_skip_worktree
&& ce_skip_worktree(ce
))
1623 * If this entry is a sparse directory, then there isn't
1624 * any stat() information to update. Ignore the entry.
1626 if (S_ISSPARSEDIR(ce
->ce_mode
))
1629 if (pathspec
&& !ce_path_match(istate
, ce
, pathspec
, seen
))
1633 while ((i
< istate
->cache_nr
) &&
1634 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
1640 show_file(unmerged_fmt
, ce
->name
, in_porcelain
,
1641 &first
, header_msg
);
1649 new_entry
= refresh_cache_ent(istate
, ce
, options
,
1650 &cache_errno
, &changed
,
1651 &t2_did_lstat
, &t2_did_scan
);
1652 t2_sum_lstat
+= t2_did_lstat
;
1653 t2_sum_scan
+= t2_did_scan
;
1654 if (new_entry
== ce
)
1656 display_progress(progress
, i
);
1660 if (really
&& cache_errno
== EINVAL
) {
1661 /* If we are doing --really-refresh that
1662 * means the index is not valid anymore.
1664 ce
->ce_flags
&= ~CE_VALID
;
1665 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
1666 mark_fsmonitor_invalid(istate
, ce
);
1667 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
1672 if (cache_errno
== ENOENT
)
1674 else if (ce_intent_to_add(ce
))
1675 fmt
= added_fmt
; /* must be before other checks */
1676 else if (changed
& TYPE_CHANGED
)
1677 fmt
= typechange_fmt
;
1681 ce
->name
, in_porcelain
, &first
, header_msg
);
1686 replace_index_entry(istate
, i
, new_entry
);
1688 trace2_data_intmax("index", NULL
, "refresh/sum_lstat", t2_sum_lstat
);
1689 trace2_data_intmax("index", NULL
, "refresh/sum_scan", t2_sum_scan
);
1690 trace2_region_leave("index", "refresh", NULL
);
1691 display_progress(progress
, istate
->cache_nr
);
1692 stop_progress(&progress
);
1693 trace_performance_leave("refresh index");
1697 struct cache_entry
*refresh_cache_entry(struct index_state
*istate
,
1698 struct cache_entry
*ce
,
1699 unsigned int options
)
1701 return refresh_cache_ent(istate
, ce
, options
, NULL
, NULL
, NULL
, NULL
);
1705 /*****************************************************************
1707 *****************************************************************/
1709 #define INDEX_FORMAT_DEFAULT 3
1711 static unsigned int get_index_format_default(struct repository
*r
)
1713 char *envversion
= getenv("GIT_INDEX_VERSION");
1715 unsigned int version
= INDEX_FORMAT_DEFAULT
;
1718 prepare_repo_settings(r
);
1720 if (r
->settings
.index_version
>= 0)
1721 version
= r
->settings
.index_version
;
1722 if (version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1723 warning(_("index.version set, but the value is invalid.\n"
1724 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1725 return INDEX_FORMAT_DEFAULT
;
1730 version
= strtoul(envversion
, &endp
, 10);
1732 version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1733 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1734 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1735 version
= INDEX_FORMAT_DEFAULT
;
1741 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1742 * Again - this is just a (very strong in practice) heuristic that
1743 * the inode hasn't changed.
1745 * We save the fields in big-endian order to allow using the
1746 * index file over NFS transparently.
1748 struct ondisk_cache_entry
{
1749 struct cache_time ctime
;
1750 struct cache_time mtime
;
1758 * unsigned char hash[hashsz];
1760 * if (flags & CE_EXTENDED)
1763 unsigned char data
[GIT_MAX_RAWSZ
+ 2 * sizeof(uint16_t)];
1764 char name
[FLEX_ARRAY
];
1767 /* These are only used for v3 or lower */
1768 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1769 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,data) + (len) + 8) & ~7)
1770 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1771 #define ondisk_data_size(flags, len) (the_hash_algo->rawsz + \
1772 ((flags & CE_EXTENDED) ? 2 : 1) * sizeof(uint16_t) + len)
1773 #define ondisk_data_size_max(len) (ondisk_data_size(CE_EXTENDED, len))
1774 #define ondisk_ce_size(ce) (ondisk_cache_entry_size(ondisk_data_size((ce)->ce_flags, ce_namelen(ce))))
1776 /* Allow fsck to force verification of the index checksum. */
1777 int verify_index_checksum
;
1779 /* Allow fsck to force verification of the cache entry order. */
1780 int verify_ce_order
;
1782 static int verify_hdr(const struct cache_header
*hdr
, unsigned long size
)
1785 unsigned char hash
[GIT_MAX_RAWSZ
];
1787 unsigned char *start
, *end
;
1788 struct object_id oid
;
1790 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
1791 return error(_("bad signature 0x%08x"), hdr
->hdr_signature
);
1792 hdr_version
= ntohl(hdr
->hdr_version
);
1793 if (hdr_version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< hdr_version
)
1794 return error(_("bad index version %d"), hdr_version
);
1796 if (!verify_index_checksum
)
1799 end
= (unsigned char *)hdr
+ size
;
1800 start
= end
- the_hash_algo
->rawsz
;
1801 oidread(&oid
, start
);
1802 if (oideq(&oid
, null_oid()))
1805 the_hash_algo
->init_fn(&c
);
1806 the_hash_algo
->update_fn(&c
, hdr
, size
- the_hash_algo
->rawsz
);
1807 the_hash_algo
->final_fn(hash
, &c
);
1808 if (!hasheq(hash
, start
))
1809 return error(_("bad index file sha1 signature"));
1813 static int read_index_extension(struct index_state
*istate
,
1814 const char *ext
, const char *data
, unsigned long sz
)
1816 switch (CACHE_EXT(ext
)) {
1817 case CACHE_EXT_TREE
:
1818 istate
->cache_tree
= cache_tree_read(data
, sz
);
1820 case CACHE_EXT_RESOLVE_UNDO
:
1821 istate
->resolve_undo
= resolve_undo_read(data
, sz
);
1823 case CACHE_EXT_LINK
:
1824 if (read_link_extension(istate
, data
, sz
))
1827 case CACHE_EXT_UNTRACKED
:
1828 istate
->untracked
= read_untracked_extension(data
, sz
);
1830 case CACHE_EXT_FSMONITOR
:
1831 read_fsmonitor_extension(istate
, data
, sz
);
1833 case CACHE_EXT_ENDOFINDEXENTRIES
:
1834 case CACHE_EXT_INDEXENTRYOFFSETTABLE
:
1835 /* already handled in do_read_index() */
1837 case CACHE_EXT_SPARSE_DIRECTORIES
:
1838 /* no content, only an indicator */
1839 istate
->sparse_index
= INDEX_COLLAPSED
;
1842 if (*ext
< 'A' || 'Z' < *ext
)
1843 return error(_("index uses %.4s extension, which we do not understand"),
1845 fprintf_ln(stderr
, _("ignoring %.4s extension"), ext
);
1852 * Parses the contents of the cache entry contained within the 'ondisk' buffer
1853 * into a new incore 'cache_entry'.
1855 * Note that 'char *ondisk' may not be aligned to a 4-byte address interval in
1856 * index v4, so we cannot cast it to 'struct ondisk_cache_entry *' and access
1857 * its members. Instead, we use the byte offsets of members within the struct to
1858 * identify where 'get_be16()', 'get_be32()', and 'oidread()' (which can all
1859 * read from an unaligned memory buffer) should read from the 'ondisk' buffer
1860 * into the corresponding incore 'cache_entry' members.
1862 static struct cache_entry
*create_from_disk(struct mem_pool
*ce_mem_pool
,
1863 unsigned int version
,
1865 unsigned long *ent_size
,
1866 const struct cache_entry
*previous_ce
)
1868 struct cache_entry
*ce
;
1871 const unsigned hashsz
= the_hash_algo
->rawsz
;
1872 const char *flagsp
= ondisk
+ offsetof(struct ondisk_cache_entry
, data
) + hashsz
;
1874 size_t copy_len
= 0;
1876 * Adjacent cache entries tend to share the leading paths, so it makes
1877 * sense to only store the differences in later entries. In the v4
1878 * on-disk format of the index, each on-disk cache entry stores the
1879 * number of bytes to be stripped from the end of the previous name,
1880 * and the bytes to append to the result, to come up with its name.
1882 int expand_name_field
= version
== 4;
1884 /* On-disk flags are just 16 bits */
1885 flags
= get_be16(flagsp
);
1886 len
= flags
& CE_NAMEMASK
;
1888 if (flags
& CE_EXTENDED
) {
1890 extended_flags
= get_be16(flagsp
+ sizeof(uint16_t)) << 16;
1891 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1892 if (extended_flags
& ~CE_EXTENDED_FLAGS
)
1893 die(_("unknown index entry format 0x%08x"), extended_flags
);
1894 flags
|= extended_flags
;
1895 name
= (const char *)(flagsp
+ 2 * sizeof(uint16_t));
1898 name
= (const char *)(flagsp
+ sizeof(uint16_t));
1900 if (expand_name_field
) {
1901 const unsigned char *cp
= (const unsigned char *)name
;
1902 size_t strip_len
, previous_len
;
1904 /* If we're at the beginning of a block, ignore the previous name */
1905 strip_len
= decode_varint(&cp
);
1907 previous_len
= previous_ce
->ce_namelen
;
1908 if (previous_len
< strip_len
)
1909 die(_("malformed name field in the index, near path '%s'"),
1911 copy_len
= previous_len
- strip_len
;
1913 name
= (const char *)cp
;
1916 if (len
== CE_NAMEMASK
) {
1918 if (expand_name_field
)
1922 ce
= mem_pool__ce_alloc(ce_mem_pool
, len
);
1925 * NEEDSWORK: using 'offsetof()' is cumbersome and should be replaced
1926 * with something more akin to 'load_bitmap_entries_v1()'s use of
1927 * 'read_be16'/'read_be32'. For consistency with the corresponding
1928 * ondisk entry write function ('copy_cache_entry_to_ondisk()'), this
1929 * should be done at the same time as removing references to
1930 * 'ondisk_cache_entry' there.
1932 ce
->ce_stat_data
.sd_ctime
.sec
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, ctime
)
1933 + offsetof(struct cache_time
, sec
));
1934 ce
->ce_stat_data
.sd_mtime
.sec
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, mtime
)
1935 + offsetof(struct cache_time
, sec
));
1936 ce
->ce_stat_data
.sd_ctime
.nsec
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, ctime
)
1937 + offsetof(struct cache_time
, nsec
));
1938 ce
->ce_stat_data
.sd_mtime
.nsec
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, mtime
)
1939 + offsetof(struct cache_time
, nsec
));
1940 ce
->ce_stat_data
.sd_dev
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, dev
));
1941 ce
->ce_stat_data
.sd_ino
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, ino
));
1942 ce
->ce_mode
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, mode
));
1943 ce
->ce_stat_data
.sd_uid
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, uid
));
1944 ce
->ce_stat_data
.sd_gid
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, gid
));
1945 ce
->ce_stat_data
.sd_size
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, size
));
1946 ce
->ce_flags
= flags
& ~CE_NAMEMASK
;
1947 ce
->ce_namelen
= len
;
1949 oidread(&ce
->oid
, (const unsigned char *)ondisk
+ offsetof(struct ondisk_cache_entry
, data
));
1951 if (expand_name_field
) {
1953 memcpy(ce
->name
, previous_ce
->name
, copy_len
);
1954 memcpy(ce
->name
+ copy_len
, name
, len
+ 1 - copy_len
);
1955 *ent_size
= (name
- ((char *)ondisk
)) + len
+ 1 - copy_len
;
1957 memcpy(ce
->name
, name
, len
+ 1);
1958 *ent_size
= ondisk_ce_size(ce
);
1963 static void check_ce_order(struct index_state
*istate
)
1967 if (!verify_ce_order
)
1970 for (i
= 1; i
< istate
->cache_nr
; i
++) {
1971 struct cache_entry
*ce
= istate
->cache
[i
- 1];
1972 struct cache_entry
*next_ce
= istate
->cache
[i
];
1973 int name_compare
= strcmp(ce
->name
, next_ce
->name
);
1975 if (0 < name_compare
)
1976 die(_("unordered stage entries in index"));
1977 if (!name_compare
) {
1979 die(_("multiple stage entries for merged file '%s'"),
1981 if (ce_stage(ce
) > ce_stage(next_ce
))
1982 die(_("unordered stage entries for '%s'"),
1988 static void tweak_untracked_cache(struct index_state
*istate
)
1990 struct repository
*r
= the_repository
;
1992 prepare_repo_settings(r
);
1994 switch (r
->settings
.core_untracked_cache
) {
1995 case UNTRACKED_CACHE_REMOVE
:
1996 remove_untracked_cache(istate
);
1998 case UNTRACKED_CACHE_WRITE
:
1999 add_untracked_cache(istate
);
2001 case UNTRACKED_CACHE_KEEP
:
2003 * Either an explicit "core.untrackedCache=keep", the
2004 * default if "core.untrackedCache" isn't configured,
2005 * or a fallback on an unknown "core.untrackedCache"
2012 static void tweak_split_index(struct index_state
*istate
)
2014 switch (git_config_get_split_index()) {
2015 case -1: /* unset: do nothing */
2018 remove_split_index(istate
);
2021 add_split_index(istate
);
2023 default: /* unknown value: do nothing */
2028 static void post_read_index_from(struct index_state
*istate
)
2030 check_ce_order(istate
);
2031 tweak_untracked_cache(istate
);
2032 tweak_split_index(istate
);
2033 tweak_fsmonitor(istate
);
2036 static size_t estimate_cache_size_from_compressed(unsigned int entries
)
2038 return entries
* (sizeof(struct cache_entry
) + CACHE_ENTRY_PATH_LENGTH
);
2041 static size_t estimate_cache_size(size_t ondisk_size
, unsigned int entries
)
2043 long per_entry
= sizeof(struct cache_entry
) - sizeof(struct ondisk_cache_entry
);
2046 * Account for potential alignment differences.
2048 per_entry
+= align_padding_size(per_entry
, 0);
2049 return ondisk_size
+ entries
* per_entry
;
2052 struct index_entry_offset
2054 /* starting byte offset into index file, count of index entries in this block */
2058 struct index_entry_offset_table
2061 struct index_entry_offset entries
[FLEX_ARRAY
];
2064 static struct index_entry_offset_table
*read_ieot_extension(const char *mmap
, size_t mmap_size
, size_t offset
);
2065 static void write_ieot_extension(struct strbuf
*sb
, struct index_entry_offset_table
*ieot
);
2067 static size_t read_eoie_extension(const char *mmap
, size_t mmap_size
);
2068 static void write_eoie_extension(struct strbuf
*sb
, git_hash_ctx
*eoie_context
, size_t offset
);
2070 struct load_index_extensions
2073 struct index_state
*istate
;
2076 unsigned long src_offset
;
2079 static void *load_index_extensions(void *_data
)
2081 struct load_index_extensions
*p
= _data
;
2082 unsigned long src_offset
= p
->src_offset
;
2084 while (src_offset
<= p
->mmap_size
- the_hash_algo
->rawsz
- 8) {
2085 /* After an array of active_nr index entries,
2086 * there can be arbitrary number of extended
2087 * sections, each of which is prefixed with
2088 * extension name (4-byte) and section length
2089 * in 4-byte network byte order.
2091 uint32_t extsize
= get_be32(p
->mmap
+ src_offset
+ 4);
2092 if (read_index_extension(p
->istate
,
2093 p
->mmap
+ src_offset
,
2094 p
->mmap
+ src_offset
+ 8,
2096 munmap((void *)p
->mmap
, p
->mmap_size
);
2097 die(_("index file corrupt"));
2100 src_offset
+= extsize
;
2107 * A helper function that will load the specified range of cache entries
2108 * from the memory mapped file and add them to the given index.
2110 static unsigned long load_cache_entry_block(struct index_state
*istate
,
2111 struct mem_pool
*ce_mem_pool
, int offset
, int nr
, const char *mmap
,
2112 unsigned long start_offset
, const struct cache_entry
*previous_ce
)
2115 unsigned long src_offset
= start_offset
;
2117 for (i
= offset
; i
< offset
+ nr
; i
++) {
2118 struct cache_entry
*ce
;
2119 unsigned long consumed
;
2121 ce
= create_from_disk(ce_mem_pool
, istate
->version
,
2123 &consumed
, previous_ce
);
2124 set_index_entry(istate
, i
, ce
);
2126 src_offset
+= consumed
;
2129 return src_offset
- start_offset
;
2132 static unsigned long load_all_cache_entries(struct index_state
*istate
,
2133 const char *mmap
, size_t mmap_size
, unsigned long src_offset
)
2135 unsigned long consumed
;
2137 istate
->ce_mem_pool
= xmalloc(sizeof(*istate
->ce_mem_pool
));
2138 if (istate
->version
== 4) {
2139 mem_pool_init(istate
->ce_mem_pool
,
2140 estimate_cache_size_from_compressed(istate
->cache_nr
));
2142 mem_pool_init(istate
->ce_mem_pool
,
2143 estimate_cache_size(mmap_size
, istate
->cache_nr
));
2146 consumed
= load_cache_entry_block(istate
, istate
->ce_mem_pool
,
2147 0, istate
->cache_nr
, mmap
, src_offset
, NULL
);
2152 * Mostly randomly chosen maximum thread counts: we
2153 * cap the parallelism to online_cpus() threads, and we want
2154 * to have at least 10000 cache entries per thread for it to
2155 * be worth starting a thread.
2158 #define THREAD_COST (10000)
2160 struct load_cache_entries_thread_data
2163 struct index_state
*istate
;
2164 struct mem_pool
*ce_mem_pool
;
2167 struct index_entry_offset_table
*ieot
;
2168 int ieot_start
; /* starting index into the ieot array */
2169 int ieot_blocks
; /* count of ieot entries to process */
2170 unsigned long consumed
; /* return # of bytes in index file processed */
2174 * A thread proc to run the load_cache_entries() computation
2175 * across multiple background threads.
2177 static void *load_cache_entries_thread(void *_data
)
2179 struct load_cache_entries_thread_data
*p
= _data
;
2182 /* iterate across all ieot blocks assigned to this thread */
2183 for (i
= p
->ieot_start
; i
< p
->ieot_start
+ p
->ieot_blocks
; i
++) {
2184 p
->consumed
+= load_cache_entry_block(p
->istate
, p
->ce_mem_pool
,
2185 p
->offset
, p
->ieot
->entries
[i
].nr
, p
->mmap
, p
->ieot
->entries
[i
].offset
, NULL
);
2186 p
->offset
+= p
->ieot
->entries
[i
].nr
;
2191 static unsigned long load_cache_entries_threaded(struct index_state
*istate
, const char *mmap
, size_t mmap_size
,
2192 int nr_threads
, struct index_entry_offset_table
*ieot
)
2194 int i
, offset
, ieot_blocks
, ieot_start
, err
;
2195 struct load_cache_entries_thread_data
*data
;
2196 unsigned long consumed
= 0;
2198 /* a little sanity checking */
2199 if (istate
->name_hash_initialized
)
2200 BUG("the name hash isn't thread safe");
2202 istate
->ce_mem_pool
= xmalloc(sizeof(*istate
->ce_mem_pool
));
2203 mem_pool_init(istate
->ce_mem_pool
, 0);
2205 /* ensure we have no more threads than we have blocks to process */
2206 if (nr_threads
> ieot
->nr
)
2207 nr_threads
= ieot
->nr
;
2208 CALLOC_ARRAY(data
, nr_threads
);
2210 offset
= ieot_start
= 0;
2211 ieot_blocks
= DIV_ROUND_UP(ieot
->nr
, nr_threads
);
2212 for (i
= 0; i
< nr_threads
; i
++) {
2213 struct load_cache_entries_thread_data
*p
= &data
[i
];
2216 if (ieot_start
+ ieot_blocks
> ieot
->nr
)
2217 ieot_blocks
= ieot
->nr
- ieot_start
;
2223 p
->ieot_start
= ieot_start
;
2224 p
->ieot_blocks
= ieot_blocks
;
2226 /* create a mem_pool for each thread */
2228 for (j
= p
->ieot_start
; j
< p
->ieot_start
+ p
->ieot_blocks
; j
++)
2229 nr
+= p
->ieot
->entries
[j
].nr
;
2230 p
->ce_mem_pool
= xmalloc(sizeof(*istate
->ce_mem_pool
));
2231 if (istate
->version
== 4) {
2232 mem_pool_init(p
->ce_mem_pool
,
2233 estimate_cache_size_from_compressed(nr
));
2235 mem_pool_init(p
->ce_mem_pool
,
2236 estimate_cache_size(mmap_size
, nr
));
2239 err
= pthread_create(&p
->pthread
, NULL
, load_cache_entries_thread
, p
);
2241 die(_("unable to create load_cache_entries thread: %s"), strerror(err
));
2243 /* increment by the number of cache entries in the ieot block being processed */
2244 for (j
= 0; j
< ieot_blocks
; j
++)
2245 offset
+= ieot
->entries
[ieot_start
+ j
].nr
;
2246 ieot_start
+= ieot_blocks
;
2249 for (i
= 0; i
< nr_threads
; i
++) {
2250 struct load_cache_entries_thread_data
*p
= &data
[i
];
2252 err
= pthread_join(p
->pthread
, NULL
);
2254 die(_("unable to join load_cache_entries thread: %s"), strerror(err
));
2255 mem_pool_combine(istate
->ce_mem_pool
, p
->ce_mem_pool
);
2256 consumed
+= p
->consumed
;
2264 static void set_new_index_sparsity(struct index_state
*istate
)
2267 * If the index's repo exists, mark it sparse according to
2270 prepare_repo_settings(istate
->repo
);
2271 if (!istate
->repo
->settings
.command_requires_full_index
&&
2272 is_sparse_index_allowed(istate
, 0))
2273 istate
->sparse_index
= 1;
2276 /* remember to discard_cache() before reading a different cache! */
2277 int do_read_index(struct index_state
*istate
, const char *path
, int must_exist
)
2281 unsigned long src_offset
;
2282 const struct cache_header
*hdr
;
2285 struct load_index_extensions p
;
2286 size_t extension_offset
= 0;
2287 int nr_threads
, cpus
;
2288 struct index_entry_offset_table
*ieot
= NULL
;
2290 if (istate
->initialized
)
2291 return istate
->cache_nr
;
2293 istate
->timestamp
.sec
= 0;
2294 istate
->timestamp
.nsec
= 0;
2295 fd
= open(path
, O_RDONLY
);
2297 if (!must_exist
&& errno
== ENOENT
) {
2298 set_new_index_sparsity(istate
);
2301 die_errno(_("%s: index file open failed"), path
);
2305 die_errno(_("%s: cannot stat the open index"), path
);
2307 mmap_size
= xsize_t(st
.st_size
);
2308 if (mmap_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2309 die(_("%s: index file smaller than expected"), path
);
2311 mmap
= xmmap_gently(NULL
, mmap_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2312 if (mmap
== MAP_FAILED
)
2313 die_errno(_("%s: unable to map index file%s"), path
,
2317 hdr
= (const struct cache_header
*)mmap
;
2318 if (verify_hdr(hdr
, mmap_size
) < 0)
2321 oidread(&istate
->oid
, (const unsigned char *)hdr
+ mmap_size
- the_hash_algo
->rawsz
);
2322 istate
->version
= ntohl(hdr
->hdr_version
);
2323 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
2324 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
2325 CALLOC_ARRAY(istate
->cache
, istate
->cache_alloc
);
2326 istate
->initialized
= 1;
2330 p
.mmap_size
= mmap_size
;
2332 src_offset
= sizeof(*hdr
);
2334 if (git_config_get_index_threads(&nr_threads
))
2337 /* TODO: does creating more threads than cores help? */
2339 nr_threads
= istate
->cache_nr
/ THREAD_COST
;
2340 cpus
= online_cpus();
2341 if (nr_threads
> cpus
)
2348 if (nr_threads
> 1) {
2349 extension_offset
= read_eoie_extension(mmap
, mmap_size
);
2350 if (extension_offset
) {
2353 p
.src_offset
= extension_offset
;
2354 err
= pthread_create(&p
.pthread
, NULL
, load_index_extensions
, &p
);
2356 die(_("unable to create load_index_extensions thread: %s"), strerror(err
));
2363 * Locate and read the index entry offset table so that we can use it
2364 * to multi-thread the reading of the cache entries.
2366 if (extension_offset
&& nr_threads
> 1)
2367 ieot
= read_ieot_extension(mmap
, mmap_size
, extension_offset
);
2370 src_offset
+= load_cache_entries_threaded(istate
, mmap
, mmap_size
, nr_threads
, ieot
);
2373 src_offset
+= load_all_cache_entries(istate
, mmap
, mmap_size
, src_offset
);
2376 istate
->timestamp
.sec
= st
.st_mtime
;
2377 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
2379 /* if we created a thread, join it otherwise load the extensions on the primary thread */
2380 if (extension_offset
) {
2381 int ret
= pthread_join(p
.pthread
, NULL
);
2383 die(_("unable to join load_index_extensions thread: %s"), strerror(ret
));
2385 p
.src_offset
= src_offset
;
2386 load_index_extensions(&p
);
2388 munmap((void *)mmap
, mmap_size
);
2391 * TODO trace2: replace "the_repository" with the actual repo instance
2392 * that is associated with the given "istate".
2394 trace2_data_intmax("index", the_repository
, "read/version",
2396 trace2_data_intmax("index", the_repository
, "read/cache_nr",
2400 * If the command explicitly requires a full index, force it
2401 * to be full. Otherwise, correct the sparsity based on repository
2402 * settings and other properties of the index (if necessary).
2404 prepare_repo_settings(istate
->repo
);
2405 if (istate
->repo
->settings
.command_requires_full_index
)
2406 ensure_full_index(istate
);
2408 ensure_correct_sparsity(istate
);
2410 return istate
->cache_nr
;
2413 munmap((void *)mmap
, mmap_size
);
2414 die(_("index file corrupt"));
2418 * Signal that the shared index is used by updating its mtime.
2420 * This way, shared index can be removed if they have not been used
2423 static void freshen_shared_index(const char *shared_index
, int warn
)
2425 if (!check_and_freshen_file(shared_index
, 1) && warn
)
2426 warning(_("could not freshen shared index '%s'"), shared_index
);
2429 int read_index_from(struct index_state
*istate
, const char *path
,
2432 struct split_index
*split_index
;
2437 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2438 if (istate
->initialized
)
2439 return istate
->cache_nr
;
2442 * TODO trace2: replace "the_repository" with the actual repo instance
2443 * that is associated with the given "istate".
2445 trace2_region_enter_printf("index", "do_read_index", the_repository
,
2447 trace_performance_enter();
2448 ret
= do_read_index(istate
, path
, 0);
2449 trace_performance_leave("read cache %s", path
);
2450 trace2_region_leave_printf("index", "do_read_index", the_repository
,
2453 split_index
= istate
->split_index
;
2454 if (!split_index
|| is_null_oid(&split_index
->base_oid
)) {
2455 post_read_index_from(istate
);
2459 trace_performance_enter();
2460 if (split_index
->base
)
2461 release_index(split_index
->base
);
2463 ALLOC_ARRAY(split_index
->base
, 1);
2464 index_state_init(split_index
->base
, istate
->repo
);
2466 base_oid_hex
= oid_to_hex(&split_index
->base_oid
);
2467 base_path
= xstrfmt("%s/sharedindex.%s", gitdir
, base_oid_hex
);
2468 trace2_region_enter_printf("index", "shared/do_read_index",
2469 the_repository
, "%s", base_path
);
2470 ret
= do_read_index(split_index
->base
, base_path
, 0);
2471 trace2_region_leave_printf("index", "shared/do_read_index",
2472 the_repository
, "%s", base_path
);
2474 char *path_copy
= xstrdup(path
);
2475 char *base_path2
= xstrfmt("%s/sharedindex.%s",
2476 dirname(path_copy
), base_oid_hex
);
2478 trace2_region_enter_printf("index", "shared/do_read_index",
2479 the_repository
, "%s", base_path2
);
2480 ret
= do_read_index(split_index
->base
, base_path2
, 1);
2481 trace2_region_leave_printf("index", "shared/do_read_index",
2482 the_repository
, "%s", base_path2
);
2485 if (!oideq(&split_index
->base_oid
, &split_index
->base
->oid
))
2486 die(_("broken index, expect %s in %s, got %s"),
2487 base_oid_hex
, base_path
,
2488 oid_to_hex(&split_index
->base
->oid
));
2490 freshen_shared_index(base_path
, 0);
2491 merge_base_index(istate
);
2492 post_read_index_from(istate
);
2493 trace_performance_leave("read cache %s", base_path
);
2498 int is_index_unborn(struct index_state
*istate
)
2500 return (!istate
->cache_nr
&& !istate
->timestamp
.sec
);
2503 void index_state_init(struct index_state
*istate
, struct repository
*r
)
2505 struct index_state blank
= INDEX_STATE_INIT(r
);
2506 memcpy(istate
, &blank
, sizeof(*istate
));
2509 void release_index(struct index_state
*istate
)
2512 * Cache entries in istate->cache[] should have been allocated
2513 * from the memory pool associated with this index, or from an
2514 * associated split_index. There is no need to free individual
2515 * cache entries. validate_cache_entries can detect when this
2516 * assertion does not hold.
2518 validate_cache_entries(istate
);
2520 resolve_undo_clear_index(istate
);
2521 free_name_hash(istate
);
2522 cache_tree_free(&(istate
->cache_tree
));
2523 free(istate
->fsmonitor_last_update
);
2524 free(istate
->cache
);
2525 discard_split_index(istate
);
2526 free_untracked_cache(istate
->untracked
);
2528 if (istate
->sparse_checkout_patterns
) {
2529 clear_pattern_list(istate
->sparse_checkout_patterns
);
2530 FREE_AND_NULL(istate
->sparse_checkout_patterns
);
2533 if (istate
->ce_mem_pool
) {
2534 mem_pool_discard(istate
->ce_mem_pool
, should_validate_cache_entries());
2535 FREE_AND_NULL(istate
->ce_mem_pool
);
2539 void discard_index(struct index_state
*istate
)
2541 release_index(istate
);
2542 index_state_init(istate
, istate
->repo
);
2546 * Validate the cache entries of this index.
2547 * All cache entries associated with this index
2548 * should have been allocated by the memory pool
2549 * associated with this index, or by a referenced
2552 void validate_cache_entries(const struct index_state
*istate
)
2556 if (!should_validate_cache_entries() ||!istate
|| !istate
->initialized
)
2559 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2561 BUG("cache entry is not allocated from expected memory pool");
2562 } else if (!istate
->ce_mem_pool
||
2563 !mem_pool_contains(istate
->ce_mem_pool
, istate
->cache
[i
])) {
2564 if (!istate
->split_index
||
2565 !istate
->split_index
->base
||
2566 !istate
->split_index
->base
->ce_mem_pool
||
2567 !mem_pool_contains(istate
->split_index
->base
->ce_mem_pool
, istate
->cache
[i
])) {
2568 BUG("cache entry is not allocated from expected memory pool");
2573 if (istate
->split_index
)
2574 validate_cache_entries(istate
->split_index
->base
);
2577 int unmerged_index(const struct index_state
*istate
)
2580 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2581 if (ce_stage(istate
->cache
[i
]))
2587 int repo_index_has_changes(struct repository
*repo
,
2591 struct index_state
*istate
= repo
->index
;
2592 struct object_id cmp
;
2596 cmp
= tree
->object
.oid
;
2597 if (tree
|| !get_oid_tree("HEAD", &cmp
)) {
2598 struct diff_options opt
;
2600 repo_diff_setup(repo
, &opt
);
2601 opt
.flags
.exit_with_status
= 1;
2603 opt
.flags
.quick
= 1;
2604 diff_setup_done(&opt
);
2605 do_diff_cache(&cmp
, &opt
);
2607 for (i
= 0; sb
&& i
< diff_queued_diff
.nr
; i
++) {
2609 strbuf_addch(sb
, ' ');
2610 strbuf_addstr(sb
, diff_queued_diff
.queue
[i
]->two
->path
);
2613 return opt
.flags
.has_changes
!= 0;
2615 /* TODO: audit for interaction with sparse-index. */
2616 ensure_full_index(istate
);
2617 for (i
= 0; sb
&& i
< istate
->cache_nr
; i
++) {
2619 strbuf_addch(sb
, ' ');
2620 strbuf_addstr(sb
, istate
->cache
[i
]->name
);
2622 return !!istate
->cache_nr
;
2626 static int write_index_ext_header(struct hashfile
*f
,
2627 git_hash_ctx
*eoie_f
,
2631 hashwrite_be32(f
, ext
);
2632 hashwrite_be32(f
, sz
);
2637 the_hash_algo
->update_fn(eoie_f
, &ext
, sizeof(ext
));
2638 the_hash_algo
->update_fn(eoie_f
, &sz
, sizeof(sz
));
2643 static void ce_smudge_racily_clean_entry(struct index_state
*istate
,
2644 struct cache_entry
*ce
)
2647 * The only thing we care about in this function is to smudge the
2648 * falsely clean entry due to touch-update-touch race, so we leave
2649 * everything else as they are. We are called for entries whose
2650 * ce_stat_data.sd_mtime match the index file mtime.
2652 * Note that this actually does not do much for gitlinks, for
2653 * which ce_match_stat_basic() always goes to the actual
2654 * contents. The caller checks with is_racy_timestamp() which
2655 * always says "no" for gitlinks, so we are not called for them ;-)
2659 if (lstat(ce
->name
, &st
) < 0)
2661 if (ce_match_stat_basic(ce
, &st
))
2663 if (ce_modified_check_fs(istate
, ce
, &st
)) {
2664 /* This is "racily clean"; smudge it. Note that this
2665 * is a tricky code. At first glance, it may appear
2666 * that it can break with this sequence:
2668 * $ echo xyzzy >frotz
2669 * $ git-update-index --add frotz
2672 * $ echo filfre >nitfol
2673 * $ git-update-index --add nitfol
2675 * but it does not. When the second update-index runs,
2676 * it notices that the entry "frotz" has the same timestamp
2677 * as index, and if we were to smudge it by resetting its
2678 * size to zero here, then the object name recorded
2679 * in index is the 6-byte file but the cached stat information
2680 * becomes zero --- which would then match what we would
2681 * obtain from the filesystem next time we stat("frotz").
2683 * However, the second update-index, before calling
2684 * this function, notices that the cached size is 6
2685 * bytes and what is on the filesystem is an empty
2686 * file, and never calls us, so the cached size information
2687 * for "frotz" stays 6 which does not match the filesystem.
2689 ce
->ce_stat_data
.sd_size
= 0;
2693 /* Copy miscellaneous fields but not the name */
2694 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry
*ondisk
,
2695 struct cache_entry
*ce
)
2698 const unsigned hashsz
= the_hash_algo
->rawsz
;
2699 uint16_t *flagsp
= (uint16_t *)(ondisk
->data
+ hashsz
);
2701 ondisk
->ctime
.sec
= htonl(ce
->ce_stat_data
.sd_ctime
.sec
);
2702 ondisk
->mtime
.sec
= htonl(ce
->ce_stat_data
.sd_mtime
.sec
);
2703 ondisk
->ctime
.nsec
= htonl(ce
->ce_stat_data
.sd_ctime
.nsec
);
2704 ondisk
->mtime
.nsec
= htonl(ce
->ce_stat_data
.sd_mtime
.nsec
);
2705 ondisk
->dev
= htonl(ce
->ce_stat_data
.sd_dev
);
2706 ondisk
->ino
= htonl(ce
->ce_stat_data
.sd_ino
);
2707 ondisk
->mode
= htonl(ce
->ce_mode
);
2708 ondisk
->uid
= htonl(ce
->ce_stat_data
.sd_uid
);
2709 ondisk
->gid
= htonl(ce
->ce_stat_data
.sd_gid
);
2710 ondisk
->size
= htonl(ce
->ce_stat_data
.sd_size
);
2711 hashcpy(ondisk
->data
, ce
->oid
.hash
);
2713 flags
= ce
->ce_flags
& ~CE_NAMEMASK
;
2714 flags
|= (ce_namelen(ce
) >= CE_NAMEMASK
? CE_NAMEMASK
: ce_namelen(ce
));
2715 flagsp
[0] = htons(flags
);
2716 if (ce
->ce_flags
& CE_EXTENDED
) {
2717 flagsp
[1] = htons((ce
->ce_flags
& CE_EXTENDED_FLAGS
) >> 16);
2721 static int ce_write_entry(struct hashfile
*f
, struct cache_entry
*ce
,
2722 struct strbuf
*previous_name
, struct ondisk_cache_entry
*ondisk
)
2725 unsigned int saved_namelen
;
2726 int stripped_name
= 0;
2727 static unsigned char padding
[8] = { 0x00 };
2729 if (ce
->ce_flags
& CE_STRIP_NAME
) {
2730 saved_namelen
= ce_namelen(ce
);
2735 size
= offsetof(struct ondisk_cache_entry
,data
) + ondisk_data_size(ce
->ce_flags
, 0);
2737 if (!previous_name
) {
2738 int len
= ce_namelen(ce
);
2739 copy_cache_entry_to_ondisk(ondisk
, ce
);
2740 hashwrite(f
, ondisk
, size
);
2741 hashwrite(f
, ce
->name
, len
);
2742 hashwrite(f
, padding
, align_padding_size(size
, len
));
2744 int common
, to_remove
, prefix_size
;
2745 unsigned char to_remove_vi
[16];
2747 (ce
->name
[common
] &&
2748 common
< previous_name
->len
&&
2749 ce
->name
[common
] == previous_name
->buf
[common
]);
2751 ; /* still matching */
2752 to_remove
= previous_name
->len
- common
;
2753 prefix_size
= encode_varint(to_remove
, to_remove_vi
);
2755 copy_cache_entry_to_ondisk(ondisk
, ce
);
2756 hashwrite(f
, ondisk
, size
);
2757 hashwrite(f
, to_remove_vi
, prefix_size
);
2758 hashwrite(f
, ce
->name
+ common
, ce_namelen(ce
) - common
);
2759 hashwrite(f
, padding
, 1);
2761 strbuf_splice(previous_name
, common
, to_remove
,
2762 ce
->name
+ common
, ce_namelen(ce
) - common
);
2764 if (stripped_name
) {
2765 ce
->ce_namelen
= saved_namelen
;
2766 ce
->ce_flags
&= ~CE_STRIP_NAME
;
2773 * This function verifies if index_state has the correct sha1 of the
2774 * index file. Don't die if we have any other failure, just return 0.
2776 static int verify_index_from(const struct index_state
*istate
, const char *path
)
2781 unsigned char hash
[GIT_MAX_RAWSZ
];
2783 if (!istate
->initialized
)
2786 fd
= open(path
, O_RDONLY
);
2793 if (st
.st_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2796 n
= pread_in_full(fd
, hash
, the_hash_algo
->rawsz
, st
.st_size
- the_hash_algo
->rawsz
);
2797 if (n
!= the_hash_algo
->rawsz
)
2800 if (!hasheq(istate
->oid
.hash
, hash
))
2811 static int repo_verify_index(struct repository
*repo
)
2813 return verify_index_from(repo
->index
, repo
->index_file
);
2816 int has_racy_timestamp(struct index_state
*istate
)
2818 int entries
= istate
->cache_nr
;
2821 for (i
= 0; i
< entries
; i
++) {
2822 struct cache_entry
*ce
= istate
->cache
[i
];
2823 if (is_racy_timestamp(istate
, ce
))
2829 void repo_update_index_if_able(struct repository
*repo
,
2830 struct lock_file
*lockfile
)
2832 if ((repo
->index
->cache_changed
||
2833 has_racy_timestamp(repo
->index
)) &&
2834 repo_verify_index(repo
))
2835 write_locked_index(repo
->index
, lockfile
, COMMIT_LOCK
);
2837 rollback_lock_file(lockfile
);
2840 static int record_eoie(void)
2844 if (!git_config_get_bool("index.recordendofindexentries", &val
))
2848 * As a convenience, the end of index entries extension
2849 * used for threading is written by default if the user
2850 * explicitly requested threaded index reads.
2852 return !git_config_get_index_threads(&val
) && val
!= 1;
2855 static int record_ieot(void)
2859 if (!git_config_get_bool("index.recordoffsettable", &val
))
2863 * As a convenience, the offset table used for threading is
2864 * written by default if the user explicitly requested
2865 * threaded index reads.
2867 return !git_config_get_index_threads(&val
) && val
!= 1;
2871 * On success, `tempfile` is closed. If it is the temporary file
2872 * of a `struct lock_file`, we will therefore effectively perform
2873 * a 'close_lock_file_gently()`. Since that is an implementation
2874 * detail of lockfiles, callers of `do_write_index()` should not
2877 static int do_write_index(struct index_state
*istate
, struct tempfile
*tempfile
,
2878 int strip_extensions
, unsigned flags
)
2880 uint64_t start
= getnanotime();
2882 git_hash_ctx
*eoie_c
= NULL
;
2883 struct cache_header hdr
;
2884 int i
, err
= 0, removed
, extended
, hdr_version
;
2885 struct cache_entry
**cache
= istate
->cache
;
2886 int entries
= istate
->cache_nr
;
2888 struct ondisk_cache_entry ondisk
;
2889 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
2890 int drop_cache_tree
= istate
->drop_cache_tree
;
2892 int csum_fsync_flag
;
2893 int ieot_entries
= 1;
2894 struct index_entry_offset_table
*ieot
= NULL
;
2896 struct repository
*r
= istate
->repo
;
2898 f
= hashfd(tempfile
->fd
, tempfile
->filename
.buf
);
2900 prepare_repo_settings(r
);
2901 f
->skip_hash
= r
->settings
.index_skip_hash
;
2903 for (i
= removed
= extended
= 0; i
< entries
; i
++) {
2904 if (cache
[i
]->ce_flags
& CE_REMOVE
)
2907 /* reduce extended entries if possible */
2908 cache
[i
]->ce_flags
&= ~CE_EXTENDED
;
2909 if (cache
[i
]->ce_flags
& CE_EXTENDED_FLAGS
) {
2911 cache
[i
]->ce_flags
|= CE_EXTENDED
;
2915 if (!istate
->version
)
2916 istate
->version
= get_index_format_default(the_repository
);
2918 /* demote version 3 to version 2 when the latter suffices */
2919 if (istate
->version
== 3 || istate
->version
== 2)
2920 istate
->version
= extended
? 3 : 2;
2922 hdr_version
= istate
->version
;
2924 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
2925 hdr
.hdr_version
= htonl(hdr_version
);
2926 hdr
.hdr_entries
= htonl(entries
- removed
);
2928 hashwrite(f
, &hdr
, sizeof(hdr
));
2930 if (!HAVE_THREADS
|| git_config_get_index_threads(&nr_threads
))
2933 if (nr_threads
!= 1 && record_ieot()) {
2934 int ieot_blocks
, cpus
;
2937 * ensure default number of ieot blocks maps evenly to the
2938 * default number of threads that will process them leaving
2939 * room for the thread to load the index extensions.
2942 ieot_blocks
= istate
->cache_nr
/ THREAD_COST
;
2943 cpus
= online_cpus();
2944 if (ieot_blocks
> cpus
- 1)
2945 ieot_blocks
= cpus
- 1;
2947 ieot_blocks
= nr_threads
;
2948 if (ieot_blocks
> istate
->cache_nr
)
2949 ieot_blocks
= istate
->cache_nr
;
2953 * no reason to write out the IEOT extension if we don't
2954 * have enough blocks to utilize multi-threading
2956 if (ieot_blocks
> 1) {
2957 ieot
= xcalloc(1, sizeof(struct index_entry_offset_table
)
2958 + (ieot_blocks
* sizeof(struct index_entry_offset
)));
2959 ieot_entries
= DIV_ROUND_UP(entries
, ieot_blocks
);
2963 offset
= hashfile_total(f
);
2966 previous_name
= (hdr_version
== 4) ? &previous_name_buf
: NULL
;
2968 for (i
= 0; i
< entries
; i
++) {
2969 struct cache_entry
*ce
= cache
[i
];
2970 if (ce
->ce_flags
& CE_REMOVE
)
2972 if (!ce_uptodate(ce
) && is_racy_timestamp(istate
, ce
))
2973 ce_smudge_racily_clean_entry(istate
, ce
);
2974 if (is_null_oid(&ce
->oid
)) {
2975 static const char msg
[] = "cache entry has null sha1: %s";
2976 static int allow
= -1;
2979 allow
= git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2981 warning(msg
, ce
->name
);
2983 err
= error(msg
, ce
->name
);
2985 drop_cache_tree
= 1;
2987 if (ieot
&& i
&& (i
% ieot_entries
== 0)) {
2988 ieot
->entries
[ieot
->nr
].nr
= nr
;
2989 ieot
->entries
[ieot
->nr
].offset
= offset
;
2992 * If we have a V4 index, set the first byte to an invalid
2993 * character to ensure there is nothing common with the previous
2997 previous_name
->buf
[0] = 0;
3000 offset
= hashfile_total(f
);
3002 if (ce_write_entry(f
, ce
, previous_name
, (struct ondisk_cache_entry
*)&ondisk
) < 0)
3010 ieot
->entries
[ieot
->nr
].nr
= nr
;
3011 ieot
->entries
[ieot
->nr
].offset
= offset
;
3014 strbuf_release(&previous_name_buf
);
3021 offset
= hashfile_total(f
);
3024 * The extension headers must be hashed on their own for the
3025 * EOIE extension. Create a hashfile here to compute that hash.
3027 if (offset
&& record_eoie()) {
3028 CALLOC_ARRAY(eoie_c
, 1);
3029 the_hash_algo
->init_fn(eoie_c
);
3033 * Lets write out CACHE_EXT_INDEXENTRYOFFSETTABLE first so that we
3034 * can minimize the number of extensions we have to scan through to
3035 * find it during load. Write it out regardless of the
3036 * strip_extensions parameter as we need it when loading the shared
3040 struct strbuf sb
= STRBUF_INIT
;
3042 write_ieot_extension(&sb
, ieot
);
3043 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_INDEXENTRYOFFSETTABLE
, sb
.len
) < 0;
3044 hashwrite(f
, sb
.buf
, sb
.len
);
3045 strbuf_release(&sb
);
3051 if (!strip_extensions
&& istate
->split_index
&&
3052 !is_null_oid(&istate
->split_index
->base_oid
)) {
3053 struct strbuf sb
= STRBUF_INIT
;
3055 if (istate
->sparse_index
)
3056 die(_("cannot write split index for a sparse index"));
3058 err
= write_link_extension(&sb
, istate
) < 0 ||
3059 write_index_ext_header(f
, eoie_c
, CACHE_EXT_LINK
,
3061 hashwrite(f
, sb
.buf
, sb
.len
);
3062 strbuf_release(&sb
);
3066 if (!strip_extensions
&& !drop_cache_tree
&& istate
->cache_tree
) {
3067 struct strbuf sb
= STRBUF_INIT
;
3069 cache_tree_write(&sb
, istate
->cache_tree
);
3070 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_TREE
, sb
.len
) < 0;
3071 hashwrite(f
, sb
.buf
, sb
.len
);
3072 strbuf_release(&sb
);
3076 if (!strip_extensions
&& istate
->resolve_undo
) {
3077 struct strbuf sb
= STRBUF_INIT
;
3079 resolve_undo_write(&sb
, istate
->resolve_undo
);
3080 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_RESOLVE_UNDO
,
3082 hashwrite(f
, sb
.buf
, sb
.len
);
3083 strbuf_release(&sb
);
3087 if (!strip_extensions
&& istate
->untracked
) {
3088 struct strbuf sb
= STRBUF_INIT
;
3090 write_untracked_extension(&sb
, istate
->untracked
);
3091 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_UNTRACKED
,
3093 hashwrite(f
, sb
.buf
, sb
.len
);
3094 strbuf_release(&sb
);
3098 if (!strip_extensions
&& istate
->fsmonitor_last_update
) {
3099 struct strbuf sb
= STRBUF_INIT
;
3101 write_fsmonitor_extension(&sb
, istate
);
3102 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_FSMONITOR
, sb
.len
) < 0;
3103 hashwrite(f
, sb
.buf
, sb
.len
);
3104 strbuf_release(&sb
);
3108 if (istate
->sparse_index
) {
3109 if (write_index_ext_header(f
, eoie_c
, CACHE_EXT_SPARSE_DIRECTORIES
, 0) < 0)
3114 * CACHE_EXT_ENDOFINDEXENTRIES must be written as the last entry before the SHA1
3115 * so that it can be found and processed before all the index entries are
3116 * read. Write it out regardless of the strip_extensions parameter as we need it
3117 * when loading the shared index.
3120 struct strbuf sb
= STRBUF_INIT
;
3122 write_eoie_extension(&sb
, eoie_c
, offset
);
3123 err
= write_index_ext_header(f
, NULL
, CACHE_EXT_ENDOFINDEXENTRIES
, sb
.len
) < 0;
3124 hashwrite(f
, sb
.buf
, sb
.len
);
3125 strbuf_release(&sb
);
3130 csum_fsync_flag
= 0;
3131 if (!alternate_index_output
&& (flags
& COMMIT_LOCK
))
3132 csum_fsync_flag
= CSUM_FSYNC
;
3134 finalize_hashfile(f
, istate
->oid
.hash
, FSYNC_COMPONENT_INDEX
,
3135 CSUM_HASH_IN_STREAM
| csum_fsync_flag
);
3137 if (close_tempfile_gently(tempfile
)) {
3138 error(_("could not close '%s'"), get_tempfile_path(tempfile
));
3141 if (stat(get_tempfile_path(tempfile
), &st
))
3143 istate
->timestamp
.sec
= (unsigned int)st
.st_mtime
;
3144 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
3145 trace_performance_since(start
, "write index, changed mask = %x", istate
->cache_changed
);
3148 * TODO trace2: replace "the_repository" with the actual repo instance
3149 * that is associated with the given "istate".
3151 trace2_data_intmax("index", the_repository
, "write/version",
3153 trace2_data_intmax("index", the_repository
, "write/cache_nr",
3159 void set_alternate_index_output(const char *name
)
3161 alternate_index_output
= name
;
3164 static int commit_locked_index(struct lock_file
*lk
)
3166 if (alternate_index_output
)
3167 return commit_lock_file_to(lk
, alternate_index_output
);
3169 return commit_lock_file(lk
);
3172 static int do_write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
3176 int was_full
= istate
->sparse_index
== INDEX_EXPANDED
;
3178 ret
= convert_to_sparse(istate
, 0);
3181 warning(_("failed to convert to a sparse-index"));
3186 * TODO trace2: replace "the_repository" with the actual repo instance
3187 * that is associated with the given "istate".
3189 trace2_region_enter_printf("index", "do_write_index", the_repository
,
3190 "%s", get_lock_file_path(lock
));
3191 ret
= do_write_index(istate
, lock
->tempfile
, 0, flags
);
3192 trace2_region_leave_printf("index", "do_write_index", the_repository
,
3193 "%s", get_lock_file_path(lock
));
3196 ensure_full_index(istate
);
3200 if (flags
& COMMIT_LOCK
)
3201 ret
= commit_locked_index(lock
);
3203 ret
= close_lock_file_gently(lock
);
3205 run_hooks_l("post-index-change",
3206 istate
->updated_workdir
? "1" : "0",
3207 istate
->updated_skipworktree
? "1" : "0", NULL
);
3208 istate
->updated_workdir
= 0;
3209 istate
->updated_skipworktree
= 0;
3214 static int write_split_index(struct index_state
*istate
,
3215 struct lock_file
*lock
,
3219 prepare_to_write_split_index(istate
);
3220 ret
= do_write_locked_index(istate
, lock
, flags
);
3221 finish_writing_split_index(istate
);
3225 static const char *shared_index_expire
= "2.weeks.ago";
3227 static unsigned long get_shared_index_expire_date(void)
3229 static unsigned long shared_index_expire_date
;
3230 static int shared_index_expire_date_prepared
;
3232 if (!shared_index_expire_date_prepared
) {
3233 git_config_get_expiry("splitindex.sharedindexexpire",
3234 &shared_index_expire
);
3235 shared_index_expire_date
= approxidate(shared_index_expire
);
3236 shared_index_expire_date_prepared
= 1;
3239 return shared_index_expire_date
;
3242 static int should_delete_shared_index(const char *shared_index_path
)
3245 unsigned long expiration
;
3247 /* Check timestamp */
3248 expiration
= get_shared_index_expire_date();
3251 if (stat(shared_index_path
, &st
))
3252 return error_errno(_("could not stat '%s'"), shared_index_path
);
3253 if (st
.st_mtime
> expiration
)
3259 static int clean_shared_index_files(const char *current_hex
)
3262 DIR *dir
= opendir(get_git_dir());
3265 return error_errno(_("unable to open git dir: %s"), get_git_dir());
3267 while ((de
= readdir(dir
)) != NULL
) {
3268 const char *sha1_hex
;
3269 const char *shared_index_path
;
3270 if (!skip_prefix(de
->d_name
, "sharedindex.", &sha1_hex
))
3272 if (!strcmp(sha1_hex
, current_hex
))
3274 shared_index_path
= git_path("%s", de
->d_name
);
3275 if (should_delete_shared_index(shared_index_path
) > 0 &&
3276 unlink(shared_index_path
))
3277 warning_errno(_("unable to unlink: %s"), shared_index_path
);
3284 static int write_shared_index(struct index_state
*istate
,
3285 struct tempfile
**temp
, unsigned flags
)
3287 struct split_index
*si
= istate
->split_index
;
3288 int ret
, was_full
= !istate
->sparse_index
;
3290 move_cache_to_base_index(istate
);
3291 convert_to_sparse(istate
, 0);
3293 trace2_region_enter_printf("index", "shared/do_write_index",
3294 the_repository
, "%s", get_tempfile_path(*temp
));
3295 ret
= do_write_index(si
->base
, *temp
, 1, flags
);
3296 trace2_region_leave_printf("index", "shared/do_write_index",
3297 the_repository
, "%s", get_tempfile_path(*temp
));
3300 ensure_full_index(istate
);
3304 ret
= adjust_shared_perm(get_tempfile_path(*temp
));
3306 error(_("cannot fix permission bits on '%s'"), get_tempfile_path(*temp
));
3309 ret
= rename_tempfile(temp
,
3310 git_path("sharedindex.%s", oid_to_hex(&si
->base
->oid
)));
3312 oidcpy(&si
->base_oid
, &si
->base
->oid
);
3313 clean_shared_index_files(oid_to_hex(&si
->base
->oid
));
3319 static const int default_max_percent_split_change
= 20;
3321 static int too_many_not_shared_entries(struct index_state
*istate
)
3323 int i
, not_shared
= 0;
3324 int max_split
= git_config_get_max_percent_split_change();
3326 switch (max_split
) {
3328 /* not or badly configured: use the default value */
3329 max_split
= default_max_percent_split_change
;
3332 return 1; /* 0% means always write a new shared index */
3334 return 0; /* 100% means never write a new shared index */
3336 break; /* just use the configured value */
3339 /* Count not shared entries */
3340 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3341 struct cache_entry
*ce
= istate
->cache
[i
];
3346 return (int64_t)istate
->cache_nr
* max_split
< (int64_t)not_shared
* 100;
3349 int write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
3352 int new_shared_index
, ret
, test_split_index_env
;
3353 struct split_index
*si
= istate
->split_index
;
3355 if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0))
3356 cache_tree_verify(the_repository
, istate
);
3358 if ((flags
& SKIP_IF_UNCHANGED
) && !istate
->cache_changed
) {
3359 if (flags
& COMMIT_LOCK
)
3360 rollback_lock_file(lock
);
3364 if (istate
->fsmonitor_last_update
)
3365 fill_fsmonitor_bitmap(istate
);
3367 test_split_index_env
= git_env_bool("GIT_TEST_SPLIT_INDEX", 0);
3369 if ((!si
&& !test_split_index_env
) ||
3370 alternate_index_output
||
3371 (istate
->cache_changed
& ~EXTMASK
)) {
3373 oidclr(&si
->base_oid
);
3374 ret
= do_write_locked_index(istate
, lock
, flags
);
3378 if (test_split_index_env
) {
3380 si
= init_split_index(istate
);
3381 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3383 int v
= si
->base_oid
.hash
[0];
3385 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3388 if (too_many_not_shared_entries(istate
))
3389 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3391 new_shared_index
= istate
->cache_changed
& SPLIT_INDEX_ORDERED
;
3393 if (new_shared_index
) {
3394 struct tempfile
*temp
;
3397 /* Same initial permissions as the main .git/index file */
3398 temp
= mks_tempfile_sm(git_path("sharedindex_XXXXXX"), 0, 0666);
3400 oidclr(&si
->base_oid
);
3401 ret
= do_write_locked_index(istate
, lock
, flags
);
3404 ret
= write_shared_index(istate
, &temp
, flags
);
3406 saved_errno
= errno
;
3407 if (is_tempfile_active(temp
))
3408 delete_tempfile(&temp
);
3409 errno
= saved_errno
;
3415 ret
= write_split_index(istate
, lock
, flags
);
3417 /* Freshen the shared index only if the split-index was written */
3418 if (!ret
&& !new_shared_index
&& !is_null_oid(&si
->base_oid
)) {
3419 const char *shared_index
= git_path("sharedindex.%s",
3420 oid_to_hex(&si
->base_oid
));
3421 freshen_shared_index(shared_index
, 1);
3425 if (flags
& COMMIT_LOCK
)
3426 rollback_lock_file(lock
);
3431 * Read the index file that is potentially unmerged into given
3432 * index_state, dropping any unmerged entries to stage #0 (potentially
3433 * resulting in a path appearing as both a file and a directory in the
3434 * index; the caller is responsible to clear out the extra entries
3435 * before writing the index to a tree). Returns true if the index is
3436 * unmerged. Callers who want to refuse to work from an unmerged
3437 * state can call this and check its return value, instead of calling
3440 int repo_read_index_unmerged(struct repository
*repo
)
3442 struct index_state
*istate
;
3446 repo_read_index(repo
);
3447 istate
= repo
->index
;
3448 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3449 struct cache_entry
*ce
= istate
->cache
[i
];
3450 struct cache_entry
*new_ce
;
3456 len
= ce_namelen(ce
);
3457 new_ce
= make_empty_cache_entry(istate
, len
);
3458 memcpy(new_ce
->name
, ce
->name
, len
);
3459 new_ce
->ce_flags
= create_ce_flags(0) | CE_CONFLICTED
;
3460 new_ce
->ce_namelen
= len
;
3461 new_ce
->ce_mode
= ce
->ce_mode
;
3462 if (add_index_entry(istate
, new_ce
, ADD_CACHE_SKIP_DFCHECK
))
3463 return error(_("%s: cannot drop to stage #0"),
3470 * Returns 1 if the path is an "other" path with respect to
3471 * the index; that is, the path is not mentioned in the index at all,
3472 * either as a file, a directory with some files in the index,
3473 * or as an unmerged entry.
3475 * We helpfully remove a trailing "/" from directories so that
3476 * the output of read_directory can be used as-is.
3478 int index_name_is_other(struct index_state
*istate
, const char *name
,
3482 if (namelen
&& name
[namelen
- 1] == '/')
3484 pos
= index_name_pos(istate
, name
, namelen
);
3486 return 0; /* exact match */
3488 if (pos
< istate
->cache_nr
) {
3489 struct cache_entry
*ce
= istate
->cache
[pos
];
3490 if (ce_namelen(ce
) == namelen
&&
3491 !memcmp(ce
->name
, name
, namelen
))
3492 return 0; /* Yup, this one exists unmerged */
3497 void *read_blob_data_from_index(struct index_state
*istate
,
3498 const char *path
, unsigned long *size
)
3502 enum object_type type
;
3506 pos
= index_name_pos(istate
, path
, len
);
3509 * We might be in the middle of a merge, in which
3510 * case we would read stage #2 (ours).
3514 (pos
< 0 && i
< istate
->cache_nr
&&
3515 !strcmp(istate
->cache
[i
]->name
, path
));
3517 if (ce_stage(istate
->cache
[i
]) == 2)
3522 data
= read_object_file(&istate
->cache
[pos
]->oid
, &type
, &sz
);
3523 if (!data
|| type
!= OBJ_BLOB
) {
3532 void stat_validity_clear(struct stat_validity
*sv
)
3534 FREE_AND_NULL(sv
->sd
);
3537 int stat_validity_check(struct stat_validity
*sv
, const char *path
)
3541 if (stat(path
, &st
) < 0)
3542 return sv
->sd
== NULL
;
3545 return S_ISREG(st
.st_mode
) && !match_stat_data(sv
->sd
, &st
);
3548 void stat_validity_update(struct stat_validity
*sv
, int fd
)
3552 if (fstat(fd
, &st
) < 0 || !S_ISREG(st
.st_mode
))
3553 stat_validity_clear(sv
);
3556 CALLOC_ARRAY(sv
->sd
, 1);
3557 fill_stat_data(sv
->sd
, &st
);
3561 void move_index_extensions(struct index_state
*dst
, struct index_state
*src
)
3563 dst
->untracked
= src
->untracked
;
3564 src
->untracked
= NULL
;
3565 dst
->cache_tree
= src
->cache_tree
;
3566 src
->cache_tree
= NULL
;
3569 struct cache_entry
*dup_cache_entry(const struct cache_entry
*ce
,
3570 struct index_state
*istate
)
3572 unsigned int size
= ce_size(ce
);
3573 int mem_pool_allocated
;
3574 struct cache_entry
*new_entry
= make_empty_cache_entry(istate
, ce_namelen(ce
));
3575 mem_pool_allocated
= new_entry
->mem_pool_allocated
;
3577 memcpy(new_entry
, ce
, size
);
3578 new_entry
->mem_pool_allocated
= mem_pool_allocated
;
3582 void discard_cache_entry(struct cache_entry
*ce
)
3584 if (ce
&& should_validate_cache_entries())
3585 memset(ce
, 0xCD, cache_entry_size(ce
->ce_namelen
));
3587 if (ce
&& ce
->mem_pool_allocated
)
3593 int should_validate_cache_entries(void)
3595 static int validate_index_cache_entries
= -1;
3597 if (validate_index_cache_entries
< 0) {
3598 if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))
3599 validate_index_cache_entries
= 1;
3601 validate_index_cache_entries
= 0;
3604 return validate_index_cache_entries
;
3607 #define EOIE_SIZE (4 + GIT_SHA1_RAWSZ) /* <4-byte offset> + <20-byte hash> */
3608 #define EOIE_SIZE_WITH_HEADER (4 + 4 + EOIE_SIZE) /* <4-byte signature> + <4-byte length> + EOIE_SIZE */
3610 static size_t read_eoie_extension(const char *mmap
, size_t mmap_size
)
3613 * The end of index entries (EOIE) extension is guaranteed to be last
3614 * so that it can be found by scanning backwards from the EOF.
3621 const char *index
, *eoie
;
3623 size_t offset
, src_offset
;
3624 unsigned char hash
[GIT_MAX_RAWSZ
];
3627 /* ensure we have an index big enough to contain an EOIE extension */
3628 if (mmap_size
< sizeof(struct cache_header
) + EOIE_SIZE_WITH_HEADER
+ the_hash_algo
->rawsz
)
3631 /* validate the extension signature */
3632 index
= eoie
= mmap
+ mmap_size
- EOIE_SIZE_WITH_HEADER
- the_hash_algo
->rawsz
;
3633 if (CACHE_EXT(index
) != CACHE_EXT_ENDOFINDEXENTRIES
)
3635 index
+= sizeof(uint32_t);
3637 /* validate the extension size */
3638 extsize
= get_be32(index
);
3639 if (extsize
!= EOIE_SIZE
)
3641 index
+= sizeof(uint32_t);
3644 * Validate the offset we're going to look for the first extension
3645 * signature is after the index header and before the eoie extension.
3647 offset
= get_be32(index
);
3648 if (mmap
+ offset
< mmap
+ sizeof(struct cache_header
))
3650 if (mmap
+ offset
>= eoie
)
3652 index
+= sizeof(uint32_t);
3655 * The hash is computed over extension types and their sizes (but not
3656 * their contents). E.g. if we have "TREE" extension that is N-bytes
3657 * long, "REUC" extension that is M-bytes long, followed by "EOIE",
3658 * then the hash would be:
3660 * SHA-1("TREE" + <binary representation of N> +
3661 * "REUC" + <binary representation of M>)
3663 src_offset
= offset
;
3664 the_hash_algo
->init_fn(&c
);
3665 while (src_offset
< mmap_size
- the_hash_algo
->rawsz
- EOIE_SIZE_WITH_HEADER
) {
3666 /* After an array of active_nr index entries,
3667 * there can be arbitrary number of extended
3668 * sections, each of which is prefixed with
3669 * extension name (4-byte) and section length
3670 * in 4-byte network byte order.
3673 memcpy(&extsize
, mmap
+ src_offset
+ 4, 4);
3674 extsize
= ntohl(extsize
);
3676 /* verify the extension size isn't so large it will wrap around */
3677 if (src_offset
+ 8 + extsize
< src_offset
)
3680 the_hash_algo
->update_fn(&c
, mmap
+ src_offset
, 8);
3683 src_offset
+= extsize
;
3685 the_hash_algo
->final_fn(hash
, &c
);
3686 if (!hasheq(hash
, (const unsigned char *)index
))
3689 /* Validate that the extension offsets returned us back to the eoie extension. */
3690 if (src_offset
!= mmap_size
- the_hash_algo
->rawsz
- EOIE_SIZE_WITH_HEADER
)
3696 static void write_eoie_extension(struct strbuf
*sb
, git_hash_ctx
*eoie_context
, size_t offset
)
3699 unsigned char hash
[GIT_MAX_RAWSZ
];
3702 put_be32(&buffer
, offset
);
3703 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3706 the_hash_algo
->final_fn(hash
, eoie_context
);
3707 strbuf_add(sb
, hash
, the_hash_algo
->rawsz
);
3710 #define IEOT_VERSION (1)
3712 static struct index_entry_offset_table
*read_ieot_extension(const char *mmap
, size_t mmap_size
, size_t offset
)
3714 const char *index
= NULL
;
3715 uint32_t extsize
, ext_version
;
3716 struct index_entry_offset_table
*ieot
;
3719 /* find the IEOT extension */
3722 while (offset
<= mmap_size
- the_hash_algo
->rawsz
- 8) {
3723 extsize
= get_be32(mmap
+ offset
+ 4);
3724 if (CACHE_EXT((mmap
+ offset
)) == CACHE_EXT_INDEXENTRYOFFSETTABLE
) {
3725 index
= mmap
+ offset
+ 4 + 4;
3734 /* validate the version is IEOT_VERSION */
3735 ext_version
= get_be32(index
);
3736 if (ext_version
!= IEOT_VERSION
) {
3737 error("invalid IEOT version %d", ext_version
);
3740 index
+= sizeof(uint32_t);
3742 /* extension size - version bytes / bytes per entry */
3743 nr
= (extsize
- sizeof(uint32_t)) / (sizeof(uint32_t) + sizeof(uint32_t));
3745 error("invalid number of IEOT entries %d", nr
);
3748 ieot
= xmalloc(sizeof(struct index_entry_offset_table
)
3749 + (nr
* sizeof(struct index_entry_offset
)));
3751 for (i
= 0; i
< nr
; i
++) {
3752 ieot
->entries
[i
].offset
= get_be32(index
);
3753 index
+= sizeof(uint32_t);
3754 ieot
->entries
[i
].nr
= get_be32(index
);
3755 index
+= sizeof(uint32_t);
3761 static void write_ieot_extension(struct strbuf
*sb
, struct index_entry_offset_table
*ieot
)
3767 put_be32(&buffer
, IEOT_VERSION
);
3768 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3771 for (i
= 0; i
< ieot
->nr
; i
++) {
3774 put_be32(&buffer
, ieot
->entries
[i
].offset
);
3775 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3778 put_be32(&buffer
, ieot
->entries
[i
].nr
);
3779 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3783 void prefetch_cache_entries(const struct index_state
*istate
,
3784 must_prefetch_predicate must_prefetch
)
3787 struct oid_array to_fetch
= OID_ARRAY_INIT
;
3789 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3790 struct cache_entry
*ce
= istate
->cache
[i
];
3792 if (S_ISGITLINK(ce
->ce_mode
) || !must_prefetch(ce
))
3794 if (!oid_object_info_extended(the_repository
, &ce
->oid
,
3796 OBJECT_INFO_FOR_PREFETCH
))
3798 oid_array_append(&to_fetch
, &ce
->oid
);
3800 promisor_remote_get_direct(the_repository
,
3801 to_fetch
.oid
, to_fetch
.nr
);
3802 oid_array_clear(&to_fetch
);