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"
32 /* Mask for the name length in ce_flags in the on-disk index */
34 #define CE_NAMEMASK (0x0fff)
38 * The first letter should be 'A'..'Z' for extensions that are not
39 * necessary for a correct operation (i.e. optimization data).
40 * When new extensions are added that _needs_ to be understood in
41 * order to correctly interpret the index file, pick character that
42 * is outside the range, to cause the reader to abort.
45 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
46 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
47 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
48 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
49 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
50 #define CACHE_EXT_FSMONITOR 0x46534D4E /* "FSMN" */
51 #define CACHE_EXT_ENDOFINDEXENTRIES 0x454F4945 /* "EOIE" */
52 #define CACHE_EXT_INDEXENTRYOFFSETTABLE 0x49454F54 /* "IEOT" */
53 #define CACHE_EXT_SPARSE_DIRECTORIES 0x73646972 /* "sdir" */
55 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
56 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
57 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
58 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED | FSMONITOR_CHANGED)
62 * This is an estimate of the pathname length in the index. We use
63 * this for V4 index files to guess the un-deltafied size of the index
64 * in memory because of pathname deltafication. This is not required
65 * for V2/V3 index formats because their pathnames are not compressed.
66 * If the initial amount of memory set aside is not sufficient, the
67 * mem pool will allocate extra memory.
69 #define CACHE_ENTRY_PATH_LENGTH 80
71 enum index_search_mode
{
76 static inline struct cache_entry
*mem_pool__ce_alloc(struct mem_pool
*mem_pool
, size_t len
)
78 struct cache_entry
*ce
;
79 ce
= mem_pool_alloc(mem_pool
, cache_entry_size(len
));
80 ce
->mem_pool_allocated
= 1;
84 static inline struct cache_entry
*mem_pool__ce_calloc(struct mem_pool
*mem_pool
, size_t len
)
86 struct cache_entry
* ce
;
87 ce
= mem_pool_calloc(mem_pool
, 1, cache_entry_size(len
));
88 ce
->mem_pool_allocated
= 1;
92 static struct mem_pool
*find_mem_pool(struct index_state
*istate
)
94 struct mem_pool
**pool_ptr
;
96 if (istate
->split_index
&& istate
->split_index
->base
)
97 pool_ptr
= &istate
->split_index
->base
->ce_mem_pool
;
99 pool_ptr
= &istate
->ce_mem_pool
;
102 *pool_ptr
= xmalloc(sizeof(**pool_ptr
));
103 mem_pool_init(*pool_ptr
, 0);
109 static const char *alternate_index_output
;
111 static void set_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
113 if (S_ISSPARSEDIR(ce
->ce_mode
))
114 istate
->sparse_index
= 1;
116 istate
->cache
[nr
] = ce
;
117 add_name_hash(istate
, ce
);
120 static void replace_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
122 struct cache_entry
*old
= istate
->cache
[nr
];
124 replace_index_entry_in_base(istate
, old
, ce
);
125 remove_name_hash(istate
, old
);
126 discard_cache_entry(old
);
127 ce
->ce_flags
&= ~CE_HASHED
;
128 set_index_entry(istate
, nr
, ce
);
129 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
130 mark_fsmonitor_invalid(istate
, ce
);
131 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
134 void rename_index_entry_at(struct index_state
*istate
, int nr
, const char *new_name
)
136 struct cache_entry
*old_entry
= istate
->cache
[nr
], *new_entry
;
137 int namelen
= strlen(new_name
);
139 new_entry
= make_empty_cache_entry(istate
, namelen
);
140 copy_cache_entry(new_entry
, old_entry
);
141 new_entry
->ce_flags
&= ~CE_HASHED
;
142 new_entry
->ce_namelen
= namelen
;
143 new_entry
->index
= 0;
144 memcpy(new_entry
->name
, new_name
, namelen
+ 1);
146 cache_tree_invalidate_path(istate
, old_entry
->name
);
147 untracked_cache_remove_from_index(istate
, old_entry
->name
);
148 remove_index_entry_at(istate
, nr
);
149 add_index_entry(istate
, new_entry
, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
152 void fill_stat_data(struct stat_data
*sd
, struct stat
*st
)
154 sd
->sd_ctime
.sec
= (unsigned int)st
->st_ctime
;
155 sd
->sd_mtime
.sec
= (unsigned int)st
->st_mtime
;
156 sd
->sd_ctime
.nsec
= ST_CTIME_NSEC(*st
);
157 sd
->sd_mtime
.nsec
= ST_MTIME_NSEC(*st
);
158 sd
->sd_dev
= st
->st_dev
;
159 sd
->sd_ino
= st
->st_ino
;
160 sd
->sd_uid
= st
->st_uid
;
161 sd
->sd_gid
= st
->st_gid
;
162 sd
->sd_size
= st
->st_size
;
165 int match_stat_data(const struct stat_data
*sd
, struct stat
*st
)
169 if (sd
->sd_mtime
.sec
!= (unsigned int)st
->st_mtime
)
170 changed
|= MTIME_CHANGED
;
171 if (trust_ctime
&& check_stat
&&
172 sd
->sd_ctime
.sec
!= (unsigned int)st
->st_ctime
)
173 changed
|= CTIME_CHANGED
;
176 if (check_stat
&& sd
->sd_mtime
.nsec
!= ST_MTIME_NSEC(*st
))
177 changed
|= MTIME_CHANGED
;
178 if (trust_ctime
&& check_stat
&&
179 sd
->sd_ctime
.nsec
!= ST_CTIME_NSEC(*st
))
180 changed
|= CTIME_CHANGED
;
184 if (sd
->sd_uid
!= (unsigned int) st
->st_uid
||
185 sd
->sd_gid
!= (unsigned int) st
->st_gid
)
186 changed
|= OWNER_CHANGED
;
187 if (sd
->sd_ino
!= (unsigned int) st
->st_ino
)
188 changed
|= INODE_CHANGED
;
193 * st_dev breaks on network filesystems where different
194 * clients will have different views of what "device"
195 * the filesystem is on
197 if (check_stat
&& sd
->sd_dev
!= (unsigned int) st
->st_dev
)
198 changed
|= INODE_CHANGED
;
201 if (sd
->sd_size
!= (unsigned int) st
->st_size
)
202 changed
|= DATA_CHANGED
;
208 * This only updates the "non-critical" parts of the directory
209 * cache, ie the parts that aren't tracked by GIT, and only used
210 * to validate the cache.
212 void fill_stat_cache_info(struct index_state
*istate
, struct cache_entry
*ce
, struct stat
*st
)
214 fill_stat_data(&ce
->ce_stat_data
, st
);
216 if (assume_unchanged
)
217 ce
->ce_flags
|= CE_VALID
;
219 if (S_ISREG(st
->st_mode
)) {
220 ce_mark_uptodate(ce
);
221 mark_fsmonitor_valid(istate
, ce
);
225 static int ce_compare_data(struct index_state
*istate
,
226 const struct cache_entry
*ce
,
230 int fd
= git_open_cloexec(ce
->name
, O_RDONLY
);
233 struct object_id oid
;
234 if (!index_fd(istate
, &oid
, fd
, st
, OBJ_BLOB
, ce
->name
, 0))
235 match
= !oideq(&oid
, &ce
->oid
);
236 /* index_fd() closed the file descriptor already */
241 static int ce_compare_link(const struct cache_entry
*ce
, size_t expected_size
)
246 enum object_type type
;
247 struct strbuf sb
= STRBUF_INIT
;
249 if (strbuf_readlink(&sb
, ce
->name
, expected_size
))
252 buffer
= read_object_file(&ce
->oid
, &type
, &size
);
255 match
= memcmp(buffer
, sb
.buf
, size
);
262 static int ce_compare_gitlink(const struct cache_entry
*ce
)
264 struct object_id oid
;
267 * We don't actually require that the .git directory
268 * under GITLINK directory be a valid git directory. It
269 * might even be missing (in case nobody populated that
272 * If so, we consider it always to match.
274 if (resolve_gitlink_ref(ce
->name
, "HEAD", &oid
) < 0)
276 return !oideq(&oid
, &ce
->oid
);
279 static int ce_modified_check_fs(struct index_state
*istate
,
280 const struct cache_entry
*ce
,
283 switch (st
->st_mode
& S_IFMT
) {
285 if (ce_compare_data(istate
, ce
, st
))
289 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
293 if (S_ISGITLINK(ce
->ce_mode
))
294 return ce_compare_gitlink(ce
) ? DATA_CHANGED
: 0;
295 /* else fallthrough */
302 static int ce_match_stat_basic(const struct cache_entry
*ce
, struct stat
*st
)
304 unsigned int changed
= 0;
306 if (ce
->ce_flags
& CE_REMOVE
)
307 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
309 switch (ce
->ce_mode
& S_IFMT
) {
311 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
312 /* We consider only the owner x bit to be relevant for
315 if (trust_executable_bit
&&
316 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
317 changed
|= MODE_CHANGED
;
320 if (!S_ISLNK(st
->st_mode
) &&
321 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
322 changed
|= TYPE_CHANGED
;
325 /* We ignore most of the st_xxx fields for gitlinks */
326 if (!S_ISDIR(st
->st_mode
))
327 changed
|= TYPE_CHANGED
;
328 else if (ce_compare_gitlink(ce
))
329 changed
|= DATA_CHANGED
;
332 BUG("unsupported ce_mode: %o", ce
->ce_mode
);
335 changed
|= match_stat_data(&ce
->ce_stat_data
, st
);
337 /* Racily smudged entry? */
338 if (!ce
->ce_stat_data
.sd_size
) {
339 if (!is_empty_blob_sha1(ce
->oid
.hash
))
340 changed
|= DATA_CHANGED
;
346 static int is_racy_stat(const struct index_state
*istate
,
347 const struct stat_data
*sd
)
349 return (istate
->timestamp
.sec
&&
351 /* nanosecond timestamped files can also be racy! */
352 (istate
->timestamp
.sec
< sd
->sd_mtime
.sec
||
353 (istate
->timestamp
.sec
== sd
->sd_mtime
.sec
&&
354 istate
->timestamp
.nsec
<= sd
->sd_mtime
.nsec
))
356 istate
->timestamp
.sec
<= sd
->sd_mtime
.sec
361 int is_racy_timestamp(const struct index_state
*istate
,
362 const struct cache_entry
*ce
)
364 return (!S_ISGITLINK(ce
->ce_mode
) &&
365 is_racy_stat(istate
, &ce
->ce_stat_data
));
368 int match_stat_data_racy(const struct index_state
*istate
,
369 const struct stat_data
*sd
, struct stat
*st
)
371 if (is_racy_stat(istate
, sd
))
372 return MTIME_CHANGED
;
373 return match_stat_data(sd
, st
);
376 int ie_match_stat(struct index_state
*istate
,
377 const struct cache_entry
*ce
, struct stat
*st
,
378 unsigned int options
)
380 unsigned int changed
;
381 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
382 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
383 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
384 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
386 if (!ignore_fsmonitor
)
387 refresh_fsmonitor(istate
);
389 * If it's marked as always valid in the index, it's
390 * valid whatever the checked-out copy says.
392 * skip-worktree has the same effect with higher precedence
394 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
))
396 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
398 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
))
402 * Intent-to-add entries have not been added, so the index entry
403 * by definition never matches what is in the work tree until it
404 * actually gets added.
406 if (ce_intent_to_add(ce
))
407 return DATA_CHANGED
| TYPE_CHANGED
| MODE_CHANGED
;
409 changed
= ce_match_stat_basic(ce
, st
);
412 * Within 1 second of this sequence:
413 * echo xyzzy >file && git-update-index --add file
414 * running this command:
416 * would give a falsely clean cache entry. The mtime and
417 * length match the cache, and other stat fields do not change.
419 * We could detect this at update-index time (the cache entry
420 * being registered/updated records the same time as "now")
421 * and delay the return from git-update-index, but that would
422 * effectively mean we can make at most one commit per second,
423 * which is not acceptable. Instead, we check cache entries
424 * whose mtime are the same as the index file timestamp more
425 * carefully than others.
427 if (!changed
&& is_racy_timestamp(istate
, ce
)) {
428 if (assume_racy_is_modified
)
429 changed
|= DATA_CHANGED
;
431 changed
|= ce_modified_check_fs(istate
, ce
, st
);
437 int ie_modified(struct index_state
*istate
,
438 const struct cache_entry
*ce
,
439 struct stat
*st
, unsigned int options
)
441 int changed
, changed_fs
;
443 changed
= ie_match_stat(istate
, ce
, st
, options
);
447 * If the mode or type has changed, there's no point in trying
448 * to refresh the entry - it's not going to match
450 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
454 * Immediately after read-tree or update-index --cacheinfo,
455 * the length field is zero, as we have never even read the
456 * lstat(2) information once, and we cannot trust DATA_CHANGED
457 * returned by ie_match_stat() which in turn was returned by
458 * ce_match_stat_basic() to signal that the filesize of the
459 * blob changed. We have to actually go to the filesystem to
460 * see if the contents match, and if so, should answer "unchanged".
462 * The logic does not apply to gitlinks, as ce_match_stat_basic()
463 * already has checked the actual HEAD from the filesystem in the
464 * subproject. If ie_match_stat() already said it is different,
465 * then we know it is.
467 if ((changed
& DATA_CHANGED
) &&
468 (S_ISGITLINK(ce
->ce_mode
) || ce
->ce_stat_data
.sd_size
!= 0))
471 changed_fs
= ce_modified_check_fs(istate
, ce
, st
);
473 return changed
| changed_fs
;
477 int base_name_compare(const char *name1
, int len1
, int mode1
,
478 const char *name2
, int len2
, int mode2
)
480 unsigned char c1
, c2
;
481 int len
= len1
< len2
? len1
: len2
;
484 cmp
= memcmp(name1
, name2
, len
);
489 if (!c1
&& S_ISDIR(mode1
))
491 if (!c2
&& S_ISDIR(mode2
))
493 return (c1
< c2
) ? -1 : (c1
> c2
) ? 1 : 0;
497 * df_name_compare() is identical to base_name_compare(), except it
498 * compares conflicting directory/file entries as equal. Note that
499 * while a directory name compares as equal to a regular file, they
500 * then individually compare _differently_ to a filename that has
501 * a dot after the basename (because '\0' < '.' < '/').
503 * This is used by routines that want to traverse the git namespace
504 * but then handle conflicting entries together when possible.
506 int df_name_compare(const char *name1
, int len1
, int mode1
,
507 const char *name2
, int len2
, int mode2
)
509 int len
= len1
< len2
? len1
: len2
, cmp
;
510 unsigned char c1
, c2
;
512 cmp
= memcmp(name1
, name2
, len
);
515 /* Directories and files compare equal (same length, same name) */
519 if (!c1
&& S_ISDIR(mode1
))
522 if (!c2
&& S_ISDIR(mode2
))
524 if (c1
== '/' && !c2
)
526 if (c2
== '/' && !c1
)
531 int name_compare(const char *name1
, size_t len1
, const char *name2
, size_t len2
)
533 size_t min_len
= (len1
< len2
) ? len1
: len2
;
534 int cmp
= memcmp(name1
, name2
, min_len
);
544 int cache_name_stage_compare(const char *name1
, int len1
, int stage1
, const char *name2
, int len2
, int stage2
)
548 cmp
= name_compare(name1
, len1
, name2
, len2
);
559 static int index_name_stage_pos(struct index_state
*istate
,
560 const char *name
, int namelen
,
562 enum index_search_mode search_mode
)
567 last
= istate
->cache_nr
;
568 while (last
> first
) {
569 int next
= first
+ ((last
- first
) >> 1);
570 struct cache_entry
*ce
= istate
->cache
[next
];
571 int cmp
= cache_name_stage_compare(name
, namelen
, stage
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
581 if (search_mode
== EXPAND_SPARSE
&& istate
->sparse_index
&&
583 /* Note: first <= istate->cache_nr */
584 struct cache_entry
*ce
= istate
->cache
[first
- 1];
587 * If we are in a sparse-index _and_ the entry before the
588 * insertion position is a sparse-directory entry that is
589 * an ancestor of 'name', then we need to expand the index
590 * and search again. This will only trigger once, because
591 * thereafter the index is fully expanded.
593 if (S_ISSPARSEDIR(ce
->ce_mode
) &&
594 ce_namelen(ce
) < namelen
&&
595 !strncmp(name
, ce
->name
, ce_namelen(ce
))) {
596 ensure_full_index(istate
);
597 return index_name_stage_pos(istate
, name
, namelen
, stage
, search_mode
);
604 int index_name_pos(struct index_state
*istate
, const char *name
, int namelen
)
606 return index_name_stage_pos(istate
, name
, namelen
, 0, EXPAND_SPARSE
);
609 int index_entry_exists(struct index_state
*istate
, const char *name
, int namelen
)
611 return index_name_stage_pos(istate
, name
, namelen
, 0, NO_EXPAND_SPARSE
) >= 0;
614 int remove_index_entry_at(struct index_state
*istate
, int pos
)
616 struct cache_entry
*ce
= istate
->cache
[pos
];
618 record_resolve_undo(istate
, ce
);
619 remove_name_hash(istate
, ce
);
620 save_or_free_index_entry(istate
, ce
);
621 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
623 if (pos
>= istate
->cache_nr
)
625 MOVE_ARRAY(istate
->cache
+ pos
, istate
->cache
+ pos
+ 1,
626 istate
->cache_nr
- pos
);
631 * Remove all cache entries marked for removal, that is where
632 * CE_REMOVE is set in ce_flags. This is much more effective than
633 * calling remove_index_entry_at() for each entry to be removed.
635 void remove_marked_cache_entries(struct index_state
*istate
, int invalidate
)
637 struct cache_entry
**ce_array
= istate
->cache
;
640 for (i
= j
= 0; i
< istate
->cache_nr
; i
++) {
641 if (ce_array
[i
]->ce_flags
& CE_REMOVE
) {
643 cache_tree_invalidate_path(istate
,
645 untracked_cache_remove_from_index(istate
,
648 remove_name_hash(istate
, ce_array
[i
]);
649 save_or_free_index_entry(istate
, ce_array
[i
]);
652 ce_array
[j
++] = ce_array
[i
];
654 if (j
== istate
->cache_nr
)
656 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
657 istate
->cache_nr
= j
;
660 int remove_file_from_index(struct index_state
*istate
, const char *path
)
662 int pos
= index_name_pos(istate
, path
, strlen(path
));
665 cache_tree_invalidate_path(istate
, path
);
666 untracked_cache_remove_from_index(istate
, path
);
667 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
668 remove_index_entry_at(istate
, pos
);
672 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
674 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
677 static int index_name_pos_also_unmerged(struct index_state
*istate
,
678 const char *path
, int namelen
)
680 int pos
= index_name_pos(istate
, path
, namelen
);
681 struct cache_entry
*ce
;
686 /* maybe unmerged? */
688 if (pos
>= istate
->cache_nr
||
689 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
692 /* order of preference: stage 2, 1, 3 */
693 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
694 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
695 !compare_name(ce
, path
, namelen
))
700 static int different_name(struct cache_entry
*ce
, struct cache_entry
*alias
)
702 int len
= ce_namelen(ce
);
703 return ce_namelen(alias
) != len
|| memcmp(ce
->name
, alias
->name
, len
);
707 * If we add a filename that aliases in the cache, we will use the
708 * name that we already have - but we don't want to update the same
709 * alias twice, because that implies that there were actually two
710 * different files with aliasing names!
712 * So we use the CE_ADDED flag to verify that the alias was an old
713 * one before we accept it as
715 static struct cache_entry
*create_alias_ce(struct index_state
*istate
,
716 struct cache_entry
*ce
,
717 struct cache_entry
*alias
)
720 struct cache_entry
*new_entry
;
722 if (alias
->ce_flags
& CE_ADDED
)
723 die(_("will not add file alias '%s' ('%s' already exists in index)"),
724 ce
->name
, alias
->name
);
726 /* Ok, create the new entry using the name of the existing alias */
727 len
= ce_namelen(alias
);
728 new_entry
= make_empty_cache_entry(istate
, len
);
729 memcpy(new_entry
->name
, alias
->name
, len
);
730 copy_cache_entry(new_entry
, ce
);
731 save_or_free_index_entry(istate
, ce
);
735 void set_object_name_for_intent_to_add_entry(struct cache_entry
*ce
)
737 struct object_id oid
;
738 if (write_object_file("", 0, blob_type
, &oid
))
739 die(_("cannot create an empty blob in the object database"));
740 oidcpy(&ce
->oid
, &oid
);
743 int add_to_index(struct index_state
*istate
, const char *path
, struct stat
*st
, int flags
)
745 int namelen
, was_same
;
746 mode_t st_mode
= st
->st_mode
;
747 struct cache_entry
*ce
, *alias
= NULL
;
748 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_IGNORE_SKIP_WORKTREE
|CE_MATCH_RACY_IS_DIRTY
;
749 int verbose
= flags
& (ADD_CACHE_VERBOSE
| ADD_CACHE_PRETEND
);
750 int pretend
= flags
& ADD_CACHE_PRETEND
;
751 int intent_only
= flags
& ADD_CACHE_INTENT
;
752 int add_option
= (ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
|
753 (intent_only
? ADD_CACHE_NEW_ONLY
: 0));
754 int hash_flags
= HASH_WRITE_OBJECT
;
755 struct object_id oid
;
757 if (flags
& ADD_CACHE_RENORMALIZE
)
758 hash_flags
|= HASH_RENORMALIZE
;
760 if (!S_ISREG(st_mode
) && !S_ISLNK(st_mode
) && !S_ISDIR(st_mode
))
761 return error(_("%s: can only add regular files, symbolic links or git-directories"), path
);
763 namelen
= strlen(path
);
764 if (S_ISDIR(st_mode
)) {
765 if (resolve_gitlink_ref(path
, "HEAD", &oid
) < 0)
766 return error(_("'%s' does not have a commit checked out"), path
);
767 while (namelen
&& path
[namelen
-1] == '/')
770 ce
= make_empty_cache_entry(istate
, namelen
);
771 memcpy(ce
->name
, path
, namelen
);
772 ce
->ce_namelen
= namelen
;
774 fill_stat_cache_info(istate
, ce
, st
);
776 ce
->ce_flags
|= CE_INTENT_TO_ADD
;
779 if (trust_executable_bit
&& has_symlinks
) {
780 ce
->ce_mode
= create_ce_mode(st_mode
);
782 /* If there is an existing entry, pick the mode bits and type
783 * from it, otherwise assume unexecutable regular file.
785 struct cache_entry
*ent
;
786 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
788 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
789 ce
->ce_mode
= ce_mode_from_stat(ent
, st_mode
);
792 /* When core.ignorecase=true, determine if a directory of the same name but differing
793 * case already exists within the Git repository. If it does, ensure the directory
794 * case of the file being added to the repository matches (is folded into) the existing
795 * entry's directory case.
798 adjust_dirname_case(istate
, ce
->name
);
800 if (!(flags
& ADD_CACHE_RENORMALIZE
)) {
801 alias
= index_file_exists(istate
, ce
->name
,
802 ce_namelen(ce
), ignore_case
);
805 !ie_match_stat(istate
, alias
, st
, ce_option
)) {
806 /* Nothing changed, really */
807 if (!S_ISGITLINK(alias
->ce_mode
))
808 ce_mark_uptodate(alias
);
809 alias
->ce_flags
|= CE_ADDED
;
811 discard_cache_entry(ce
);
816 if (index_path(istate
, &ce
->oid
, path
, st
, hash_flags
)) {
817 discard_cache_entry(ce
);
818 return error(_("unable to index file '%s'"), path
);
821 set_object_name_for_intent_to_add_entry(ce
);
823 if (ignore_case
&& alias
&& different_name(ce
, alias
))
824 ce
= create_alias_ce(istate
, ce
, alias
);
825 ce
->ce_flags
|= CE_ADDED
;
827 /* It was suspected to be racily clean, but it turns out to be Ok */
830 oideq(&alias
->oid
, &ce
->oid
) &&
831 ce
->ce_mode
== alias
->ce_mode
);
834 discard_cache_entry(ce
);
835 else if (add_index_entry(istate
, ce
, add_option
)) {
836 discard_cache_entry(ce
);
837 return error(_("unable to add '%s' to index"), path
);
839 if (verbose
&& !was_same
)
840 printf("add '%s'\n", path
);
844 int add_file_to_index(struct index_state
*istate
, const char *path
, int flags
)
847 if (lstat(path
, &st
))
848 die_errno(_("unable to stat '%s'"), path
);
849 return add_to_index(istate
, path
, &st
, flags
);
852 struct cache_entry
*make_empty_cache_entry(struct index_state
*istate
, size_t len
)
854 return mem_pool__ce_calloc(find_mem_pool(istate
), len
);
857 struct cache_entry
*make_empty_transient_cache_entry(size_t len
,
858 struct mem_pool
*ce_mem_pool
)
861 return mem_pool__ce_calloc(ce_mem_pool
, len
);
862 return xcalloc(1, cache_entry_size(len
));
865 struct cache_entry
*make_cache_entry(struct index_state
*istate
,
867 const struct object_id
*oid
,
870 unsigned int refresh_options
)
872 struct cache_entry
*ce
, *ret
;
875 if (!verify_path(path
, mode
)) {
876 error(_("invalid path '%s'"), path
);
881 ce
= make_empty_cache_entry(istate
, len
);
883 oidcpy(&ce
->oid
, oid
);
884 memcpy(ce
->name
, path
, len
);
885 ce
->ce_flags
= create_ce_flags(stage
);
886 ce
->ce_namelen
= len
;
887 ce
->ce_mode
= create_ce_mode(mode
);
889 ret
= refresh_cache_entry(istate
, ce
, refresh_options
);
891 discard_cache_entry(ce
);
895 struct cache_entry
*make_transient_cache_entry(unsigned int mode
,
896 const struct object_id
*oid
,
899 struct mem_pool
*ce_mem_pool
)
901 struct cache_entry
*ce
;
904 if (!verify_path(path
, mode
)) {
905 error(_("invalid path '%s'"), path
);
910 ce
= make_empty_transient_cache_entry(len
, ce_mem_pool
);
912 oidcpy(&ce
->oid
, oid
);
913 memcpy(ce
->name
, path
, len
);
914 ce
->ce_flags
= create_ce_flags(stage
);
915 ce
->ce_namelen
= len
;
916 ce
->ce_mode
= create_ce_mode(mode
);
922 * Chmod an index entry with either +x or -x.
924 * Returns -1 if the chmod for the particular cache entry failed (if it's
925 * not a regular file), -2 if an invalid flip argument is passed in, 0
928 int chmod_index_entry(struct index_state
*istate
, struct cache_entry
*ce
,
931 if (!S_ISREG(ce
->ce_mode
))
938 ce
->ce_mode
&= ~0111;
943 cache_tree_invalidate_path(istate
, ce
->name
);
944 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
945 mark_fsmonitor_invalid(istate
, ce
);
946 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
951 int ce_same_name(const struct cache_entry
*a
, const struct cache_entry
*b
)
953 int len
= ce_namelen(a
);
954 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
958 * We fundamentally don't like some paths: we don't want
959 * dot or dot-dot anywhere, and for obvious reasons don't
960 * want to recurse into ".git" either.
962 * Also, we don't want double slashes or slashes at the
963 * end that can make pathnames ambiguous.
965 static int verify_dotfile(const char *rest
, unsigned mode
)
968 * The first character was '.', but that
969 * has already been discarded, we now test
973 /* "." is not allowed */
974 if (*rest
== '\0' || is_dir_sep(*rest
))
979 * ".git" followed by NUL or slash is bad. Note that we match
980 * case-insensitively here, even if ignore_case is not set.
981 * This outlaws ".GIT" everywhere out of an abundance of caution,
982 * since there's really no good reason to allow it.
984 * Once we've seen ".git", we can also find ".gitmodules", etc (also
985 * case-insensitively).
989 if (rest
[1] != 'i' && rest
[1] != 'I')
991 if (rest
[2] != 't' && rest
[2] != 'T')
993 if (rest
[3] == '\0' || is_dir_sep(rest
[3]))
997 if (skip_iprefix(rest
, "modules", &rest
) &&
998 (*rest
== '\0' || is_dir_sep(*rest
)))
1003 if (rest
[1] == '\0' || is_dir_sep(rest
[1]))
1009 int verify_path(const char *path
, unsigned mode
)
1013 if (has_dos_drive_prefix(path
))
1016 if (!is_valid_path(path
))
1023 if (is_dir_sep(c
)) {
1027 if (is_hfs_dotgit(path
))
1029 if (S_ISLNK(mode
)) {
1030 if (is_hfs_dotgitmodules(path
))
1035 #if defined GIT_WINDOWS_NATIVE || defined __CYGWIN__
1039 if (is_ntfs_dotgit(path
))
1041 if (S_ISLNK(mode
)) {
1042 if (is_ntfs_dotgitmodules(path
))
1048 if ((c
== '.' && !verify_dotfile(path
, mode
)) ||
1052 * allow terminating directory separators for
1053 * sparse directory entries.
1056 return S_ISDIR(mode
);
1057 } else if (c
== '\\' && protect_ntfs
) {
1058 if (is_ntfs_dotgit(path
))
1060 if (S_ISLNK(mode
)) {
1061 if (is_ntfs_dotgitmodules(path
))
1071 * Do we have another file that has the beginning components being a
1072 * proper superset of the name we're trying to add?
1074 static int has_file_name(struct index_state
*istate
,
1075 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
1078 int len
= ce_namelen(ce
);
1079 int stage
= ce_stage(ce
);
1080 const char *name
= ce
->name
;
1082 while (pos
< istate
->cache_nr
) {
1083 struct cache_entry
*p
= istate
->cache
[pos
++];
1085 if (len
>= ce_namelen(p
))
1087 if (memcmp(name
, p
->name
, len
))
1089 if (ce_stage(p
) != stage
)
1091 if (p
->name
[len
] != '/')
1093 if (p
->ce_flags
& CE_REMOVE
)
1098 remove_index_entry_at(istate
, --pos
);
1105 * Like strcmp(), but also return the offset of the first change.
1106 * If strings are equal, return the length.
1108 int strcmp_offset(const char *s1
, const char *s2
, size_t *first_change
)
1113 return strcmp(s1
, s2
);
1115 for (k
= 0; s1
[k
] == s2
[k
]; k
++)
1120 return (unsigned char)s1
[k
] - (unsigned char)s2
[k
];
1124 * Do we have another file with a pathname that is a proper
1125 * subset of the name we're trying to add?
1127 * That is, is there another file in the index with a path
1128 * that matches a sub-directory in the given entry?
1130 static int has_dir_name(struct index_state
*istate
,
1131 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
1134 int stage
= ce_stage(ce
);
1135 const char *name
= ce
->name
;
1136 const char *slash
= name
+ ce_namelen(ce
);
1141 * We are frequently called during an iteration on a sorted
1142 * list of pathnames and while building a new index. Therefore,
1143 * there is a high probability that this entry will eventually
1144 * be appended to the index, rather than inserted in the middle.
1145 * If we can confirm that, we can avoid binary searches on the
1146 * components of the pathname.
1148 * Compare the entry's full path with the last path in the index.
1150 if (istate
->cache_nr
> 0) {
1151 cmp_last
= strcmp_offset(name
,
1152 istate
->cache
[istate
->cache_nr
- 1]->name
,
1155 if (len_eq_last
== 0) {
1157 * The entry sorts AFTER the last one in the
1158 * index and their paths have no common prefix,
1159 * so there cannot be a F/D conflict.
1164 * The entry sorts AFTER the last one in the
1165 * index, but has a common prefix. Fall through
1166 * to the loop below to disect the entry's path
1167 * and see where the difference is.
1170 } else if (cmp_last
== 0) {
1172 * The entry exactly matches the last one in the
1173 * index, but because of multiple stage and CE_REMOVE
1174 * items, we fall through and let the regular search
1184 if (*--slash
== '/')
1186 if (slash
<= ce
->name
)
1193 * (len + 1) is a directory boundary (including
1194 * the trailing slash). And since the loop is
1195 * decrementing "slash", the first iteration is
1196 * the longest directory prefix; subsequent
1197 * iterations consider parent directories.
1200 if (len
+ 1 <= len_eq_last
) {
1202 * The directory prefix (including the trailing
1203 * slash) also appears as a prefix in the last
1204 * entry, so the remainder cannot collide (because
1205 * strcmp said the whole path was greater).
1210 * LT: last: xxx/file_A
1216 if (len
> len_eq_last
) {
1218 * This part of the directory prefix (excluding
1219 * the trailing slash) is longer than the known
1220 * equal portions, so this sub-directory cannot
1221 * collide with a file.
1230 * This is a possible collision. Fall through and
1231 * let the regular search code handle it.
1238 pos
= index_name_stage_pos(istate
, name
, len
, stage
, EXPAND_SPARSE
);
1241 * Found one, but not so fast. This could
1242 * be a marker that says "I was here, but
1243 * I am being removed". Such an entry is
1244 * not a part of the resulting tree, and
1245 * it is Ok to have a directory at the same
1248 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
1252 remove_index_entry_at(istate
, pos
);
1260 * Trivial optimization: if we find an entry that
1261 * already matches the sub-directory, then we know
1262 * we're ok, and we can exit.
1264 while (pos
< istate
->cache_nr
) {
1265 struct cache_entry
*p
= istate
->cache
[pos
];
1266 if ((ce_namelen(p
) <= len
) ||
1267 (p
->name
[len
] != '/') ||
1268 memcmp(p
->name
, name
, len
))
1269 break; /* not our subdirectory */
1270 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
1272 * p is at the same stage as our entry, and
1273 * is a subdirectory of what we are looking
1274 * at, so we cannot have conflicts at our
1275 * level or anything shorter.
1284 /* We may be in a situation where we already have path/file and path
1285 * is being added, or we already have path and path/file is being
1286 * added. Either one would result in a nonsense tree that has path
1287 * twice when git-write-tree tries to write it out. Prevent it.
1289 * If ok-to-replace is specified, we remove the conflicting entries
1290 * from the cache so the caller should recompute the insert position.
1291 * When this happens, we return non-zero.
1293 static int check_file_directory_conflict(struct index_state
*istate
,
1294 const struct cache_entry
*ce
,
1295 int pos
, int ok_to_replace
)
1300 * When ce is an "I am going away" entry, we allow it to be added
1302 if (ce
->ce_flags
& CE_REMOVE
)
1306 * We check if the path is a sub-path of a subsequent pathname
1307 * first, since removing those will not change the position
1310 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
1313 * Then check if the path might have a clashing sub-directory
1316 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
1319 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1322 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
1323 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
1324 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
1325 int new_only
= option
& ADD_CACHE_NEW_ONLY
;
1327 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1328 cache_tree_invalidate_path(istate
, ce
->name
);
1331 * If this entry's path sorts after the last entry in the index,
1332 * we can avoid searching for it.
1334 if (istate
->cache_nr
> 0 &&
1335 strcmp(ce
->name
, istate
->cache
[istate
->cache_nr
- 1]->name
) > 0)
1336 pos
= index_pos_to_insert_pos(istate
->cache_nr
);
1338 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
), EXPAND_SPARSE
);
1340 /* existing match? Just replace it. */
1343 replace_index_entry(istate
, pos
, ce
);
1348 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1349 untracked_cache_add_to_index(istate
, ce
->name
);
1352 * Inserting a merged entry ("stage 0") into the index
1353 * will always replace all non-merged entries..
1355 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
1356 while (ce_same_name(istate
->cache
[pos
], ce
)) {
1358 if (!remove_index_entry_at(istate
, pos
))
1365 if (!verify_path(ce
->name
, ce
->ce_mode
))
1366 return error(_("invalid path '%s'"), ce
->name
);
1368 if (!skip_df_check
&&
1369 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
1371 return error(_("'%s' appears as both a file and as a directory"),
1373 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
), EXPAND_SPARSE
);
1379 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1383 if (option
& ADD_CACHE_JUST_APPEND
)
1384 pos
= istate
->cache_nr
;
1387 ret
= add_index_entry_with_check(istate
, ce
, option
);
1393 /* Make sure the array is big enough .. */
1394 ALLOC_GROW(istate
->cache
, istate
->cache_nr
+ 1, istate
->cache_alloc
);
1398 if (istate
->cache_nr
> pos
+ 1)
1399 MOVE_ARRAY(istate
->cache
+ pos
+ 1, istate
->cache
+ pos
,
1400 istate
->cache_nr
- pos
- 1);
1401 set_index_entry(istate
, pos
, ce
);
1402 istate
->cache_changed
|= CE_ENTRY_ADDED
;
1407 * "refresh" does not calculate a new sha1 file or bring the
1408 * cache up-to-date for mode/content changes. But what it
1409 * _does_ do is to "re-match" the stat information of a file
1410 * with the cache, so that you can refresh the cache for a
1411 * file that hasn't been changed but where the stat entry is
1414 * For example, you'd want to do this after doing a "git-read-tree",
1415 * to link up the stat cache details with the proper files.
1417 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
1418 struct cache_entry
*ce
,
1419 unsigned int options
, int *err
,
1425 struct cache_entry
*updated
;
1427 int refresh
= options
& CE_MATCH_REFRESH
;
1428 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
1429 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
1430 int ignore_missing
= options
& CE_MATCH_IGNORE_MISSING
;
1431 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
1433 if (!refresh
|| ce_uptodate(ce
))
1436 if (!ignore_fsmonitor
)
1437 refresh_fsmonitor(istate
);
1439 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1440 * that the change to the work tree does not matter and told
1443 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
)) {
1444 ce_mark_uptodate(ce
);
1447 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
)) {
1448 ce_mark_uptodate(ce
);
1451 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
)) {
1452 ce_mark_uptodate(ce
);
1456 if (has_symlink_leading_path(ce
->name
, ce_namelen(ce
))) {
1466 if (lstat(ce
->name
, &st
) < 0) {
1467 if (ignore_missing
&& errno
== ENOENT
)
1474 changed
= ie_match_stat(istate
, ce
, &st
, options
);
1476 *changed_ret
= changed
;
1479 * The path is unchanged. If we were told to ignore
1480 * valid bit, then we did the actual stat check and
1481 * found that the entry is unmodified. If the entry
1482 * is not marked VALID, this is the place to mark it
1483 * valid again, under "assume unchanged" mode.
1485 if (ignore_valid
&& assume_unchanged
&&
1486 !(ce
->ce_flags
& CE_VALID
))
1487 ; /* mark this one VALID again */
1490 * We do not mark the index itself "modified"
1491 * because CE_UPTODATE flag is in-core only;
1492 * we are not going to write this change out.
1494 if (!S_ISGITLINK(ce
->ce_mode
)) {
1495 ce_mark_uptodate(ce
);
1496 mark_fsmonitor_valid(istate
, ce
);
1504 if (ie_modified(istate
, ce
, &st
, options
)) {
1510 updated
= make_empty_cache_entry(istate
, ce_namelen(ce
));
1511 copy_cache_entry(updated
, ce
);
1512 memcpy(updated
->name
, ce
->name
, ce
->ce_namelen
+ 1);
1513 fill_stat_cache_info(istate
, updated
, &st
);
1515 * If ignore_valid is not set, we should leave CE_VALID bit
1516 * alone. Otherwise, paths marked with --no-assume-unchanged
1517 * (i.e. things to be edited) will reacquire CE_VALID bit
1518 * automatically, which is not really what we want.
1520 if (!ignore_valid
&& assume_unchanged
&&
1521 !(ce
->ce_flags
& CE_VALID
))
1522 updated
->ce_flags
&= ~CE_VALID
;
1524 /* istate->cache_changed is updated in the caller */
1528 static void show_file(const char * fmt
, const char * name
, int in_porcelain
,
1529 int * first
, const char *header_msg
)
1531 if (in_porcelain
&& *first
&& header_msg
) {
1532 printf("%s\n", header_msg
);
1538 int repo_refresh_and_write_index(struct repository
*repo
,
1539 unsigned int refresh_flags
,
1540 unsigned int write_flags
,
1542 const struct pathspec
*pathspec
,
1543 char *seen
, const char *header_msg
)
1545 struct lock_file lock_file
= LOCK_INIT
;
1548 fd
= repo_hold_locked_index(repo
, &lock_file
, 0);
1549 if (!gentle
&& fd
< 0)
1551 if (refresh_index(repo
->index
, refresh_flags
, pathspec
, seen
, header_msg
))
1553 if (0 <= fd
&& write_locked_index(repo
->index
, &lock_file
, COMMIT_LOCK
| write_flags
))
1559 int refresh_index(struct index_state
*istate
, unsigned int flags
,
1560 const struct pathspec
*pathspec
,
1561 char *seen
, const char *header_msg
)
1565 int really
= (flags
& REFRESH_REALLY
) != 0;
1566 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
1567 int quiet
= (flags
& REFRESH_QUIET
) != 0;
1568 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
1569 int ignore_submodules
= (flags
& REFRESH_IGNORE_SUBMODULES
) != 0;
1570 int ignore_skip_worktree
= (flags
& REFRESH_IGNORE_SKIP_WORKTREE
) != 0;
1572 int in_porcelain
= (flags
& REFRESH_IN_PORCELAIN
);
1573 unsigned int options
= (CE_MATCH_REFRESH
|
1574 (really
? CE_MATCH_IGNORE_VALID
: 0) |
1575 (not_new
? CE_MATCH_IGNORE_MISSING
: 0));
1576 const char *modified_fmt
;
1577 const char *deleted_fmt
;
1578 const char *typechange_fmt
;
1579 const char *added_fmt
;
1580 const char *unmerged_fmt
;
1581 struct progress
*progress
= NULL
;
1582 int t2_sum_lstat
= 0;
1583 int t2_sum_scan
= 0;
1585 if (flags
& REFRESH_PROGRESS
&& isatty(2))
1586 progress
= start_delayed_progress(_("Refresh index"),
1589 trace_performance_enter();
1590 modified_fmt
= in_porcelain
? "M\t%s\n" : "%s: needs update\n";
1591 deleted_fmt
= in_porcelain
? "D\t%s\n" : "%s: needs update\n";
1592 typechange_fmt
= in_porcelain
? "T\t%s\n" : "%s: needs update\n";
1593 added_fmt
= in_porcelain
? "A\t%s\n" : "%s: needs update\n";
1594 unmerged_fmt
= in_porcelain
? "U\t%s\n" : "%s: needs merge\n";
1596 * Use the multi-threaded preload_index() to refresh most of the
1597 * cache entries quickly then in the single threaded loop below,
1598 * we only have to do the special cases that are left.
1600 preload_index(istate
, pathspec
, 0);
1601 trace2_region_enter("index", "refresh", NULL
);
1603 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1604 struct cache_entry
*ce
, *new_entry
;
1605 int cache_errno
= 0;
1608 int t2_did_lstat
= 0;
1609 int t2_did_scan
= 0;
1611 ce
= istate
->cache
[i
];
1612 if (ignore_submodules
&& S_ISGITLINK(ce
->ce_mode
))
1614 if (ignore_skip_worktree
&& ce_skip_worktree(ce
))
1618 * If this entry is a sparse directory, then there isn't
1619 * any stat() information to update. Ignore the entry.
1621 if (S_ISSPARSEDIR(ce
->ce_mode
))
1624 if (pathspec
&& !ce_path_match(istate
, ce
, pathspec
, seen
))
1628 while ((i
< istate
->cache_nr
) &&
1629 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
1635 show_file(unmerged_fmt
, ce
->name
, in_porcelain
,
1636 &first
, header_msg
);
1644 new_entry
= refresh_cache_ent(istate
, ce
, options
,
1645 &cache_errno
, &changed
,
1646 &t2_did_lstat
, &t2_did_scan
);
1647 t2_sum_lstat
+= t2_did_lstat
;
1648 t2_sum_scan
+= t2_did_scan
;
1649 if (new_entry
== ce
)
1651 display_progress(progress
, i
);
1655 if (really
&& cache_errno
== EINVAL
) {
1656 /* If we are doing --really-refresh that
1657 * means the index is not valid anymore.
1659 ce
->ce_flags
&= ~CE_VALID
;
1660 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
1661 mark_fsmonitor_invalid(istate
, ce
);
1662 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
1667 if (cache_errno
== ENOENT
)
1669 else if (ce_intent_to_add(ce
))
1670 fmt
= added_fmt
; /* must be before other checks */
1671 else if (changed
& TYPE_CHANGED
)
1672 fmt
= typechange_fmt
;
1676 ce
->name
, in_porcelain
, &first
, header_msg
);
1681 replace_index_entry(istate
, i
, new_entry
);
1683 trace2_data_intmax("index", NULL
, "refresh/sum_lstat", t2_sum_lstat
);
1684 trace2_data_intmax("index", NULL
, "refresh/sum_scan", t2_sum_scan
);
1685 trace2_region_leave("index", "refresh", NULL
);
1686 display_progress(progress
, istate
->cache_nr
);
1687 stop_progress(&progress
);
1688 trace_performance_leave("refresh index");
1692 struct cache_entry
*refresh_cache_entry(struct index_state
*istate
,
1693 struct cache_entry
*ce
,
1694 unsigned int options
)
1696 return refresh_cache_ent(istate
, ce
, options
, NULL
, NULL
, NULL
, NULL
);
1700 /*****************************************************************
1702 *****************************************************************/
1704 #define INDEX_FORMAT_DEFAULT 3
1706 static unsigned int get_index_format_default(struct repository
*r
)
1708 char *envversion
= getenv("GIT_INDEX_VERSION");
1710 unsigned int version
= INDEX_FORMAT_DEFAULT
;
1713 prepare_repo_settings(r
);
1715 if (r
->settings
.index_version
>= 0)
1716 version
= r
->settings
.index_version
;
1717 if (version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1718 warning(_("index.version set, but the value is invalid.\n"
1719 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1720 return INDEX_FORMAT_DEFAULT
;
1725 version
= strtoul(envversion
, &endp
, 10);
1727 version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1728 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1729 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1730 version
= INDEX_FORMAT_DEFAULT
;
1736 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1737 * Again - this is just a (very strong in practice) heuristic that
1738 * the inode hasn't changed.
1740 * We save the fields in big-endian order to allow using the
1741 * index file over NFS transparently.
1743 struct ondisk_cache_entry
{
1744 struct cache_time ctime
;
1745 struct cache_time mtime
;
1753 * unsigned char hash[hashsz];
1755 * if (flags & CE_EXTENDED)
1758 unsigned char data
[GIT_MAX_RAWSZ
+ 2 * sizeof(uint16_t)];
1759 char name
[FLEX_ARRAY
];
1762 /* These are only used for v3 or lower */
1763 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1764 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,data) + (len) + 8) & ~7)
1765 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1766 #define ondisk_data_size(flags, len) (the_hash_algo->rawsz + \
1767 ((flags & CE_EXTENDED) ? 2 : 1) * sizeof(uint16_t) + len)
1768 #define ondisk_data_size_max(len) (ondisk_data_size(CE_EXTENDED, len))
1769 #define ondisk_ce_size(ce) (ondisk_cache_entry_size(ondisk_data_size((ce)->ce_flags, ce_namelen(ce))))
1771 /* Allow fsck to force verification of the index checksum. */
1772 int verify_index_checksum
;
1774 /* Allow fsck to force verification of the cache entry order. */
1775 int verify_ce_order
;
1777 static int verify_hdr(const struct cache_header
*hdr
, unsigned long size
)
1780 unsigned char hash
[GIT_MAX_RAWSZ
];
1783 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
1784 return error(_("bad signature 0x%08x"), hdr
->hdr_signature
);
1785 hdr_version
= ntohl(hdr
->hdr_version
);
1786 if (hdr_version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< hdr_version
)
1787 return error(_("bad index version %d"), hdr_version
);
1789 if (!verify_index_checksum
)
1792 the_hash_algo
->init_fn(&c
);
1793 the_hash_algo
->update_fn(&c
, hdr
, size
- the_hash_algo
->rawsz
);
1794 the_hash_algo
->final_fn(hash
, &c
);
1795 if (!hasheq(hash
, (unsigned char *)hdr
+ size
- the_hash_algo
->rawsz
))
1796 return error(_("bad index file sha1 signature"));
1800 static int read_index_extension(struct index_state
*istate
,
1801 const char *ext
, const char *data
, unsigned long sz
)
1803 switch (CACHE_EXT(ext
)) {
1804 case CACHE_EXT_TREE
:
1805 istate
->cache_tree
= cache_tree_read(data
, sz
);
1807 case CACHE_EXT_RESOLVE_UNDO
:
1808 istate
->resolve_undo
= resolve_undo_read(data
, sz
);
1810 case CACHE_EXT_LINK
:
1811 if (read_link_extension(istate
, data
, sz
))
1814 case CACHE_EXT_UNTRACKED
:
1815 istate
->untracked
= read_untracked_extension(data
, sz
);
1817 case CACHE_EXT_FSMONITOR
:
1818 read_fsmonitor_extension(istate
, data
, sz
);
1820 case CACHE_EXT_ENDOFINDEXENTRIES
:
1821 case CACHE_EXT_INDEXENTRYOFFSETTABLE
:
1822 /* already handled in do_read_index() */
1824 case CACHE_EXT_SPARSE_DIRECTORIES
:
1825 /* no content, only an indicator */
1826 istate
->sparse_index
= 1;
1829 if (*ext
< 'A' || 'Z' < *ext
)
1830 return error(_("index uses %.4s extension, which we do not understand"),
1832 fprintf_ln(stderr
, _("ignoring %.4s extension"), ext
);
1838 static struct cache_entry
*create_from_disk(struct mem_pool
*ce_mem_pool
,
1839 unsigned int version
,
1840 struct ondisk_cache_entry
*ondisk
,
1841 unsigned long *ent_size
,
1842 const struct cache_entry
*previous_ce
)
1844 struct cache_entry
*ce
;
1847 const unsigned hashsz
= the_hash_algo
->rawsz
;
1848 const uint16_t *flagsp
= (const uint16_t *)(ondisk
->data
+ hashsz
);
1850 size_t copy_len
= 0;
1852 * Adjacent cache entries tend to share the leading paths, so it makes
1853 * sense to only store the differences in later entries. In the v4
1854 * on-disk format of the index, each on-disk cache entry stores the
1855 * number of bytes to be stripped from the end of the previous name,
1856 * and the bytes to append to the result, to come up with its name.
1858 int expand_name_field
= version
== 4;
1860 /* On-disk flags are just 16 bits */
1861 flags
= get_be16(flagsp
);
1862 len
= flags
& CE_NAMEMASK
;
1864 if (flags
& CE_EXTENDED
) {
1866 extended_flags
= get_be16(flagsp
+ 1) << 16;
1867 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1868 if (extended_flags
& ~CE_EXTENDED_FLAGS
)
1869 die(_("unknown index entry format 0x%08x"), extended_flags
);
1870 flags
|= extended_flags
;
1871 name
= (const char *)(flagsp
+ 2);
1874 name
= (const char *)(flagsp
+ 1);
1876 if (expand_name_field
) {
1877 const unsigned char *cp
= (const unsigned char *)name
;
1878 size_t strip_len
, previous_len
;
1880 /* If we're at the beginning of a block, ignore the previous name */
1881 strip_len
= decode_varint(&cp
);
1883 previous_len
= previous_ce
->ce_namelen
;
1884 if (previous_len
< strip_len
)
1885 die(_("malformed name field in the index, near path '%s'"),
1887 copy_len
= previous_len
- strip_len
;
1889 name
= (const char *)cp
;
1892 if (len
== CE_NAMEMASK
) {
1894 if (expand_name_field
)
1898 ce
= mem_pool__ce_alloc(ce_mem_pool
, len
);
1900 ce
->ce_stat_data
.sd_ctime
.sec
= get_be32(&ondisk
->ctime
.sec
);
1901 ce
->ce_stat_data
.sd_mtime
.sec
= get_be32(&ondisk
->mtime
.sec
);
1902 ce
->ce_stat_data
.sd_ctime
.nsec
= get_be32(&ondisk
->ctime
.nsec
);
1903 ce
->ce_stat_data
.sd_mtime
.nsec
= get_be32(&ondisk
->mtime
.nsec
);
1904 ce
->ce_stat_data
.sd_dev
= get_be32(&ondisk
->dev
);
1905 ce
->ce_stat_data
.sd_ino
= get_be32(&ondisk
->ino
);
1906 ce
->ce_mode
= get_be32(&ondisk
->mode
);
1907 ce
->ce_stat_data
.sd_uid
= get_be32(&ondisk
->uid
);
1908 ce
->ce_stat_data
.sd_gid
= get_be32(&ondisk
->gid
);
1909 ce
->ce_stat_data
.sd_size
= get_be32(&ondisk
->size
);
1910 ce
->ce_flags
= flags
& ~CE_NAMEMASK
;
1911 ce
->ce_namelen
= len
;
1913 oidread(&ce
->oid
, ondisk
->data
);
1914 memcpy(ce
->name
, name
, len
);
1915 ce
->name
[len
] = '\0';
1917 if (expand_name_field
) {
1919 memcpy(ce
->name
, previous_ce
->name
, copy_len
);
1920 memcpy(ce
->name
+ copy_len
, name
, len
+ 1 - copy_len
);
1921 *ent_size
= (name
- ((char *)ondisk
)) + len
+ 1 - copy_len
;
1923 memcpy(ce
->name
, name
, len
+ 1);
1924 *ent_size
= ondisk_ce_size(ce
);
1929 static void check_ce_order(struct index_state
*istate
)
1933 if (!verify_ce_order
)
1936 for (i
= 1; i
< istate
->cache_nr
; i
++) {
1937 struct cache_entry
*ce
= istate
->cache
[i
- 1];
1938 struct cache_entry
*next_ce
= istate
->cache
[i
];
1939 int name_compare
= strcmp(ce
->name
, next_ce
->name
);
1941 if (0 < name_compare
)
1942 die(_("unordered stage entries in index"));
1943 if (!name_compare
) {
1945 die(_("multiple stage entries for merged file '%s'"),
1947 if (ce_stage(ce
) > ce_stage(next_ce
))
1948 die(_("unordered stage entries for '%s'"),
1954 static void tweak_untracked_cache(struct index_state
*istate
)
1956 struct repository
*r
= the_repository
;
1958 prepare_repo_settings(r
);
1960 switch (r
->settings
.core_untracked_cache
) {
1961 case UNTRACKED_CACHE_REMOVE
:
1962 remove_untracked_cache(istate
);
1964 case UNTRACKED_CACHE_WRITE
:
1965 add_untracked_cache(istate
);
1967 case UNTRACKED_CACHE_KEEP
:
1969 * Either an explicit "core.untrackedCache=keep", the
1970 * default if "core.untrackedCache" isn't configured,
1971 * or a fallback on an unknown "core.untrackedCache"
1978 static void tweak_split_index(struct index_state
*istate
)
1980 switch (git_config_get_split_index()) {
1981 case -1: /* unset: do nothing */
1984 remove_split_index(istate
);
1987 add_split_index(istate
);
1989 default: /* unknown value: do nothing */
1994 static void post_read_index_from(struct index_state
*istate
)
1996 check_ce_order(istate
);
1997 tweak_untracked_cache(istate
);
1998 tweak_split_index(istate
);
1999 tweak_fsmonitor(istate
);
2002 static size_t estimate_cache_size_from_compressed(unsigned int entries
)
2004 return entries
* (sizeof(struct cache_entry
) + CACHE_ENTRY_PATH_LENGTH
);
2007 static size_t estimate_cache_size(size_t ondisk_size
, unsigned int entries
)
2009 long per_entry
= sizeof(struct cache_entry
) - sizeof(struct ondisk_cache_entry
);
2012 * Account for potential alignment differences.
2014 per_entry
+= align_padding_size(per_entry
, 0);
2015 return ondisk_size
+ entries
* per_entry
;
2018 struct index_entry_offset
2020 /* starting byte offset into index file, count of index entries in this block */
2024 struct index_entry_offset_table
2027 struct index_entry_offset entries
[FLEX_ARRAY
];
2030 static struct index_entry_offset_table
*read_ieot_extension(const char *mmap
, size_t mmap_size
, size_t offset
);
2031 static void write_ieot_extension(struct strbuf
*sb
, struct index_entry_offset_table
*ieot
);
2033 static size_t read_eoie_extension(const char *mmap
, size_t mmap_size
);
2034 static void write_eoie_extension(struct strbuf
*sb
, git_hash_ctx
*eoie_context
, size_t offset
);
2036 struct load_index_extensions
2039 struct index_state
*istate
;
2042 unsigned long src_offset
;
2045 static void *load_index_extensions(void *_data
)
2047 struct load_index_extensions
*p
= _data
;
2048 unsigned long src_offset
= p
->src_offset
;
2050 while (src_offset
<= p
->mmap_size
- the_hash_algo
->rawsz
- 8) {
2051 /* After an array of active_nr index entries,
2052 * there can be arbitrary number of extended
2053 * sections, each of which is prefixed with
2054 * extension name (4-byte) and section length
2055 * in 4-byte network byte order.
2057 uint32_t extsize
= get_be32(p
->mmap
+ src_offset
+ 4);
2058 if (read_index_extension(p
->istate
,
2059 p
->mmap
+ src_offset
,
2060 p
->mmap
+ src_offset
+ 8,
2062 munmap((void *)p
->mmap
, p
->mmap_size
);
2063 die(_("index file corrupt"));
2066 src_offset
+= extsize
;
2073 * A helper function that will load the specified range of cache entries
2074 * from the memory mapped file and add them to the given index.
2076 static unsigned long load_cache_entry_block(struct index_state
*istate
,
2077 struct mem_pool
*ce_mem_pool
, int offset
, int nr
, const char *mmap
,
2078 unsigned long start_offset
, const struct cache_entry
*previous_ce
)
2081 unsigned long src_offset
= start_offset
;
2083 for (i
= offset
; i
< offset
+ nr
; i
++) {
2084 struct ondisk_cache_entry
*disk_ce
;
2085 struct cache_entry
*ce
;
2086 unsigned long consumed
;
2088 disk_ce
= (struct ondisk_cache_entry
*)(mmap
+ src_offset
);
2089 ce
= create_from_disk(ce_mem_pool
, istate
->version
, disk_ce
, &consumed
, previous_ce
);
2090 set_index_entry(istate
, i
, ce
);
2092 src_offset
+= consumed
;
2095 return src_offset
- start_offset
;
2098 static unsigned long load_all_cache_entries(struct index_state
*istate
,
2099 const char *mmap
, size_t mmap_size
, unsigned long src_offset
)
2101 unsigned long consumed
;
2103 istate
->ce_mem_pool
= xmalloc(sizeof(*istate
->ce_mem_pool
));
2104 if (istate
->version
== 4) {
2105 mem_pool_init(istate
->ce_mem_pool
,
2106 estimate_cache_size_from_compressed(istate
->cache_nr
));
2108 mem_pool_init(istate
->ce_mem_pool
,
2109 estimate_cache_size(mmap_size
, istate
->cache_nr
));
2112 consumed
= load_cache_entry_block(istate
, istate
->ce_mem_pool
,
2113 0, istate
->cache_nr
, mmap
, src_offset
, NULL
);
2118 * Mostly randomly chosen maximum thread counts: we
2119 * cap the parallelism to online_cpus() threads, and we want
2120 * to have at least 10000 cache entries per thread for it to
2121 * be worth starting a thread.
2124 #define THREAD_COST (10000)
2126 struct load_cache_entries_thread_data
2129 struct index_state
*istate
;
2130 struct mem_pool
*ce_mem_pool
;
2133 struct index_entry_offset_table
*ieot
;
2134 int ieot_start
; /* starting index into the ieot array */
2135 int ieot_blocks
; /* count of ieot entries to process */
2136 unsigned long consumed
; /* return # of bytes in index file processed */
2140 * A thread proc to run the load_cache_entries() computation
2141 * across multiple background threads.
2143 static void *load_cache_entries_thread(void *_data
)
2145 struct load_cache_entries_thread_data
*p
= _data
;
2148 /* iterate across all ieot blocks assigned to this thread */
2149 for (i
= p
->ieot_start
; i
< p
->ieot_start
+ p
->ieot_blocks
; i
++) {
2150 p
->consumed
+= load_cache_entry_block(p
->istate
, p
->ce_mem_pool
,
2151 p
->offset
, p
->ieot
->entries
[i
].nr
, p
->mmap
, p
->ieot
->entries
[i
].offset
, NULL
);
2152 p
->offset
+= p
->ieot
->entries
[i
].nr
;
2157 static unsigned long load_cache_entries_threaded(struct index_state
*istate
, const char *mmap
, size_t mmap_size
,
2158 int nr_threads
, struct index_entry_offset_table
*ieot
)
2160 int i
, offset
, ieot_blocks
, ieot_start
, err
;
2161 struct load_cache_entries_thread_data
*data
;
2162 unsigned long consumed
= 0;
2164 /* a little sanity checking */
2165 if (istate
->name_hash_initialized
)
2166 BUG("the name hash isn't thread safe");
2168 istate
->ce_mem_pool
= xmalloc(sizeof(*istate
->ce_mem_pool
));
2169 mem_pool_init(istate
->ce_mem_pool
, 0);
2171 /* ensure we have no more threads than we have blocks to process */
2172 if (nr_threads
> ieot
->nr
)
2173 nr_threads
= ieot
->nr
;
2174 CALLOC_ARRAY(data
, nr_threads
);
2176 offset
= ieot_start
= 0;
2177 ieot_blocks
= DIV_ROUND_UP(ieot
->nr
, nr_threads
);
2178 for (i
= 0; i
< nr_threads
; i
++) {
2179 struct load_cache_entries_thread_data
*p
= &data
[i
];
2182 if (ieot_start
+ ieot_blocks
> ieot
->nr
)
2183 ieot_blocks
= ieot
->nr
- ieot_start
;
2189 p
->ieot_start
= ieot_start
;
2190 p
->ieot_blocks
= ieot_blocks
;
2192 /* create a mem_pool for each thread */
2194 for (j
= p
->ieot_start
; j
< p
->ieot_start
+ p
->ieot_blocks
; j
++)
2195 nr
+= p
->ieot
->entries
[j
].nr
;
2196 p
->ce_mem_pool
= xmalloc(sizeof(*istate
->ce_mem_pool
));
2197 if (istate
->version
== 4) {
2198 mem_pool_init(p
->ce_mem_pool
,
2199 estimate_cache_size_from_compressed(nr
));
2201 mem_pool_init(p
->ce_mem_pool
,
2202 estimate_cache_size(mmap_size
, nr
));
2205 err
= pthread_create(&p
->pthread
, NULL
, load_cache_entries_thread
, p
);
2207 die(_("unable to create load_cache_entries thread: %s"), strerror(err
));
2209 /* increment by the number of cache entries in the ieot block being processed */
2210 for (j
= 0; j
< ieot_blocks
; j
++)
2211 offset
+= ieot
->entries
[ieot_start
+ j
].nr
;
2212 ieot_start
+= ieot_blocks
;
2215 for (i
= 0; i
< nr_threads
; i
++) {
2216 struct load_cache_entries_thread_data
*p
= &data
[i
];
2218 err
= pthread_join(p
->pthread
, NULL
);
2220 die(_("unable to join load_cache_entries thread: %s"), strerror(err
));
2221 mem_pool_combine(istate
->ce_mem_pool
, p
->ce_mem_pool
);
2222 consumed
+= p
->consumed
;
2230 /* remember to discard_cache() before reading a different cache! */
2231 int do_read_index(struct index_state
*istate
, const char *path
, int must_exist
)
2235 unsigned long src_offset
;
2236 const struct cache_header
*hdr
;
2239 struct load_index_extensions p
;
2240 size_t extension_offset
= 0;
2241 int nr_threads
, cpus
;
2242 struct index_entry_offset_table
*ieot
= NULL
;
2244 if (istate
->initialized
)
2245 return istate
->cache_nr
;
2247 istate
->timestamp
.sec
= 0;
2248 istate
->timestamp
.nsec
= 0;
2249 fd
= open(path
, O_RDONLY
);
2251 if (!must_exist
&& errno
== ENOENT
)
2253 die_errno(_("%s: index file open failed"), path
);
2257 die_errno(_("%s: cannot stat the open index"), path
);
2259 mmap_size
= xsize_t(st
.st_size
);
2260 if (mmap_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2261 die(_("%s: index file smaller than expected"), path
);
2263 mmap
= xmmap_gently(NULL
, mmap_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2264 if (mmap
== MAP_FAILED
)
2265 die_errno(_("%s: unable to map index file%s"), path
,
2269 hdr
= (const struct cache_header
*)mmap
;
2270 if (verify_hdr(hdr
, mmap_size
) < 0)
2273 oidread(&istate
->oid
, (const unsigned char *)hdr
+ mmap_size
- the_hash_algo
->rawsz
);
2274 istate
->version
= ntohl(hdr
->hdr_version
);
2275 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
2276 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
2277 CALLOC_ARRAY(istate
->cache
, istate
->cache_alloc
);
2278 istate
->initialized
= 1;
2282 p
.mmap_size
= mmap_size
;
2284 src_offset
= sizeof(*hdr
);
2286 if (git_config_get_index_threads(&nr_threads
))
2289 /* TODO: does creating more threads than cores help? */
2291 nr_threads
= istate
->cache_nr
/ THREAD_COST
;
2292 cpus
= online_cpus();
2293 if (nr_threads
> cpus
)
2300 if (nr_threads
> 1) {
2301 extension_offset
= read_eoie_extension(mmap
, mmap_size
);
2302 if (extension_offset
) {
2305 p
.src_offset
= extension_offset
;
2306 err
= pthread_create(&p
.pthread
, NULL
, load_index_extensions
, &p
);
2308 die(_("unable to create load_index_extensions thread: %s"), strerror(err
));
2315 * Locate and read the index entry offset table so that we can use it
2316 * to multi-thread the reading of the cache entries.
2318 if (extension_offset
&& nr_threads
> 1)
2319 ieot
= read_ieot_extension(mmap
, mmap_size
, extension_offset
);
2322 src_offset
+= load_cache_entries_threaded(istate
, mmap
, mmap_size
, nr_threads
, ieot
);
2325 src_offset
+= load_all_cache_entries(istate
, mmap
, mmap_size
, src_offset
);
2328 istate
->timestamp
.sec
= st
.st_mtime
;
2329 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
2331 /* if we created a thread, join it otherwise load the extensions on the primary thread */
2332 if (extension_offset
) {
2333 int ret
= pthread_join(p
.pthread
, NULL
);
2335 die(_("unable to join load_index_extensions thread: %s"), strerror(ret
));
2337 p
.src_offset
= src_offset
;
2338 load_index_extensions(&p
);
2340 munmap((void *)mmap
, mmap_size
);
2343 * TODO trace2: replace "the_repository" with the actual repo instance
2344 * that is associated with the given "istate".
2346 trace2_data_intmax("index", the_repository
, "read/version",
2348 trace2_data_intmax("index", the_repository
, "read/cache_nr",
2352 istate
->repo
= the_repository
;
2353 prepare_repo_settings(istate
->repo
);
2354 if (istate
->repo
->settings
.command_requires_full_index
)
2355 ensure_full_index(istate
);
2357 return istate
->cache_nr
;
2360 munmap((void *)mmap
, mmap_size
);
2361 die(_("index file corrupt"));
2365 * Signal that the shared index is used by updating its mtime.
2367 * This way, shared index can be removed if they have not been used
2370 static void freshen_shared_index(const char *shared_index
, int warn
)
2372 if (!check_and_freshen_file(shared_index
, 1) && warn
)
2373 warning(_("could not freshen shared index '%s'"), shared_index
);
2376 int read_index_from(struct index_state
*istate
, const char *path
,
2379 struct split_index
*split_index
;
2384 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2385 if (istate
->initialized
)
2386 return istate
->cache_nr
;
2389 * TODO trace2: replace "the_repository" with the actual repo instance
2390 * that is associated with the given "istate".
2392 trace2_region_enter_printf("index", "do_read_index", the_repository
,
2394 trace_performance_enter();
2395 ret
= do_read_index(istate
, path
, 0);
2396 trace_performance_leave("read cache %s", path
);
2397 trace2_region_leave_printf("index", "do_read_index", the_repository
,
2400 split_index
= istate
->split_index
;
2401 if (!split_index
|| is_null_oid(&split_index
->base_oid
)) {
2402 post_read_index_from(istate
);
2406 trace_performance_enter();
2407 if (split_index
->base
)
2408 discard_index(split_index
->base
);
2410 CALLOC_ARRAY(split_index
->base
, 1);
2412 base_oid_hex
= oid_to_hex(&split_index
->base_oid
);
2413 base_path
= xstrfmt("%s/sharedindex.%s", gitdir
, base_oid_hex
);
2414 trace2_region_enter_printf("index", "shared/do_read_index",
2415 the_repository
, "%s", base_path
);
2416 ret
= do_read_index(split_index
->base
, base_path
, 1);
2417 trace2_region_leave_printf("index", "shared/do_read_index",
2418 the_repository
, "%s", base_path
);
2419 if (!oideq(&split_index
->base_oid
, &split_index
->base
->oid
))
2420 die(_("broken index, expect %s in %s, got %s"),
2421 base_oid_hex
, base_path
,
2422 oid_to_hex(&split_index
->base
->oid
));
2424 freshen_shared_index(base_path
, 0);
2425 merge_base_index(istate
);
2426 post_read_index_from(istate
);
2427 trace_performance_leave("read cache %s", base_path
);
2432 int is_index_unborn(struct index_state
*istate
)
2434 return (!istate
->cache_nr
&& !istate
->timestamp
.sec
);
2437 int discard_index(struct index_state
*istate
)
2440 * Cache entries in istate->cache[] should have been allocated
2441 * from the memory pool associated with this index, or from an
2442 * associated split_index. There is no need to free individual
2443 * cache entries. validate_cache_entries can detect when this
2444 * assertion does not hold.
2446 validate_cache_entries(istate
);
2448 resolve_undo_clear_index(istate
);
2449 istate
->cache_nr
= 0;
2450 istate
->cache_changed
= 0;
2451 istate
->timestamp
.sec
= 0;
2452 istate
->timestamp
.nsec
= 0;
2453 free_name_hash(istate
);
2454 cache_tree_free(&(istate
->cache_tree
));
2455 istate
->initialized
= 0;
2456 istate
->fsmonitor_has_run_once
= 0;
2457 FREE_AND_NULL(istate
->fsmonitor_last_update
);
2458 FREE_AND_NULL(istate
->cache
);
2459 istate
->cache_alloc
= 0;
2460 discard_split_index(istate
);
2461 free_untracked_cache(istate
->untracked
);
2462 istate
->untracked
= NULL
;
2464 if (istate
->ce_mem_pool
) {
2465 mem_pool_discard(istate
->ce_mem_pool
, should_validate_cache_entries());
2466 FREE_AND_NULL(istate
->ce_mem_pool
);
2473 * Validate the cache entries of this index.
2474 * All cache entries associated with this index
2475 * should have been allocated by the memory pool
2476 * associated with this index, or by a referenced
2479 void validate_cache_entries(const struct index_state
*istate
)
2483 if (!should_validate_cache_entries() ||!istate
|| !istate
->initialized
)
2486 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2488 BUG("cache entry is not allocated from expected memory pool");
2489 } else if (!istate
->ce_mem_pool
||
2490 !mem_pool_contains(istate
->ce_mem_pool
, istate
->cache
[i
])) {
2491 if (!istate
->split_index
||
2492 !istate
->split_index
->base
||
2493 !istate
->split_index
->base
->ce_mem_pool
||
2494 !mem_pool_contains(istate
->split_index
->base
->ce_mem_pool
, istate
->cache
[i
])) {
2495 BUG("cache entry is not allocated from expected memory pool");
2500 if (istate
->split_index
)
2501 validate_cache_entries(istate
->split_index
->base
);
2504 int unmerged_index(const struct index_state
*istate
)
2507 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2508 if (ce_stage(istate
->cache
[i
]))
2514 int repo_index_has_changes(struct repository
*repo
,
2518 struct index_state
*istate
= repo
->index
;
2519 struct object_id cmp
;
2523 cmp
= tree
->object
.oid
;
2524 if (tree
|| !get_oid_tree("HEAD", &cmp
)) {
2525 struct diff_options opt
;
2527 repo_diff_setup(repo
, &opt
);
2528 opt
.flags
.exit_with_status
= 1;
2530 opt
.flags
.quick
= 1;
2531 diff_setup_done(&opt
);
2532 do_diff_cache(&cmp
, &opt
);
2534 for (i
= 0; sb
&& i
< diff_queued_diff
.nr
; i
++) {
2536 strbuf_addch(sb
, ' ');
2537 strbuf_addstr(sb
, diff_queued_diff
.queue
[i
]->two
->path
);
2540 return opt
.flags
.has_changes
!= 0;
2542 /* TODO: audit for interaction with sparse-index. */
2543 ensure_full_index(istate
);
2544 for (i
= 0; sb
&& i
< istate
->cache_nr
; i
++) {
2546 strbuf_addch(sb
, ' ');
2547 strbuf_addstr(sb
, istate
->cache
[i
]->name
);
2549 return !!istate
->cache_nr
;
2553 static int write_index_ext_header(struct hashfile
*f
,
2554 git_hash_ctx
*eoie_f
,
2558 hashwrite_be32(f
, ext
);
2559 hashwrite_be32(f
, sz
);
2564 the_hash_algo
->update_fn(eoie_f
, &ext
, sizeof(ext
));
2565 the_hash_algo
->update_fn(eoie_f
, &sz
, sizeof(sz
));
2570 static void ce_smudge_racily_clean_entry(struct index_state
*istate
,
2571 struct cache_entry
*ce
)
2574 * The only thing we care about in this function is to smudge the
2575 * falsely clean entry due to touch-update-touch race, so we leave
2576 * everything else as they are. We are called for entries whose
2577 * ce_stat_data.sd_mtime match the index file mtime.
2579 * Note that this actually does not do much for gitlinks, for
2580 * which ce_match_stat_basic() always goes to the actual
2581 * contents. The caller checks with is_racy_timestamp() which
2582 * always says "no" for gitlinks, so we are not called for them ;-)
2586 if (lstat(ce
->name
, &st
) < 0)
2588 if (ce_match_stat_basic(ce
, &st
))
2590 if (ce_modified_check_fs(istate
, ce
, &st
)) {
2591 /* This is "racily clean"; smudge it. Note that this
2592 * is a tricky code. At first glance, it may appear
2593 * that it can break with this sequence:
2595 * $ echo xyzzy >frotz
2596 * $ git-update-index --add frotz
2599 * $ echo filfre >nitfol
2600 * $ git-update-index --add nitfol
2602 * but it does not. When the second update-index runs,
2603 * it notices that the entry "frotz" has the same timestamp
2604 * as index, and if we were to smudge it by resetting its
2605 * size to zero here, then the object name recorded
2606 * in index is the 6-byte file but the cached stat information
2607 * becomes zero --- which would then match what we would
2608 * obtain from the filesystem next time we stat("frotz").
2610 * However, the second update-index, before calling
2611 * this function, notices that the cached size is 6
2612 * bytes and what is on the filesystem is an empty
2613 * file, and never calls us, so the cached size information
2614 * for "frotz" stays 6 which does not match the filesystem.
2616 ce
->ce_stat_data
.sd_size
= 0;
2620 /* Copy miscellaneous fields but not the name */
2621 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry
*ondisk
,
2622 struct cache_entry
*ce
)
2625 const unsigned hashsz
= the_hash_algo
->rawsz
;
2626 uint16_t *flagsp
= (uint16_t *)(ondisk
->data
+ hashsz
);
2628 ondisk
->ctime
.sec
= htonl(ce
->ce_stat_data
.sd_ctime
.sec
);
2629 ondisk
->mtime
.sec
= htonl(ce
->ce_stat_data
.sd_mtime
.sec
);
2630 ondisk
->ctime
.nsec
= htonl(ce
->ce_stat_data
.sd_ctime
.nsec
);
2631 ondisk
->mtime
.nsec
= htonl(ce
->ce_stat_data
.sd_mtime
.nsec
);
2632 ondisk
->dev
= htonl(ce
->ce_stat_data
.sd_dev
);
2633 ondisk
->ino
= htonl(ce
->ce_stat_data
.sd_ino
);
2634 ondisk
->mode
= htonl(ce
->ce_mode
);
2635 ondisk
->uid
= htonl(ce
->ce_stat_data
.sd_uid
);
2636 ondisk
->gid
= htonl(ce
->ce_stat_data
.sd_gid
);
2637 ondisk
->size
= htonl(ce
->ce_stat_data
.sd_size
);
2638 hashcpy(ondisk
->data
, ce
->oid
.hash
);
2640 flags
= ce
->ce_flags
& ~CE_NAMEMASK
;
2641 flags
|= (ce_namelen(ce
) >= CE_NAMEMASK
? CE_NAMEMASK
: ce_namelen(ce
));
2642 flagsp
[0] = htons(flags
);
2643 if (ce
->ce_flags
& CE_EXTENDED
) {
2644 flagsp
[1] = htons((ce
->ce_flags
& CE_EXTENDED_FLAGS
) >> 16);
2648 static int ce_write_entry(struct hashfile
*f
, struct cache_entry
*ce
,
2649 struct strbuf
*previous_name
, struct ondisk_cache_entry
*ondisk
)
2652 unsigned int saved_namelen
;
2653 int stripped_name
= 0;
2654 static unsigned char padding
[8] = { 0x00 };
2656 if (ce
->ce_flags
& CE_STRIP_NAME
) {
2657 saved_namelen
= ce_namelen(ce
);
2662 size
= offsetof(struct ondisk_cache_entry
,data
) + ondisk_data_size(ce
->ce_flags
, 0);
2664 if (!previous_name
) {
2665 int len
= ce_namelen(ce
);
2666 copy_cache_entry_to_ondisk(ondisk
, ce
);
2667 hashwrite(f
, ondisk
, size
);
2668 hashwrite(f
, ce
->name
, len
);
2669 hashwrite(f
, padding
, align_padding_size(size
, len
));
2671 int common
, to_remove
, prefix_size
;
2672 unsigned char to_remove_vi
[16];
2674 (ce
->name
[common
] &&
2675 common
< previous_name
->len
&&
2676 ce
->name
[common
] == previous_name
->buf
[common
]);
2678 ; /* still matching */
2679 to_remove
= previous_name
->len
- common
;
2680 prefix_size
= encode_varint(to_remove
, to_remove_vi
);
2682 copy_cache_entry_to_ondisk(ondisk
, ce
);
2683 hashwrite(f
, ondisk
, size
);
2684 hashwrite(f
, to_remove_vi
, prefix_size
);
2685 hashwrite(f
, ce
->name
+ common
, ce_namelen(ce
) - common
);
2686 hashwrite(f
, padding
, 1);
2688 strbuf_splice(previous_name
, common
, to_remove
,
2689 ce
->name
+ common
, ce_namelen(ce
) - common
);
2691 if (stripped_name
) {
2692 ce
->ce_namelen
= saved_namelen
;
2693 ce
->ce_flags
&= ~CE_STRIP_NAME
;
2700 * This function verifies if index_state has the correct sha1 of the
2701 * index file. Don't die if we have any other failure, just return 0.
2703 static int verify_index_from(const struct index_state
*istate
, const char *path
)
2708 unsigned char hash
[GIT_MAX_RAWSZ
];
2710 if (!istate
->initialized
)
2713 fd
= open(path
, O_RDONLY
);
2720 if (st
.st_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2723 n
= pread_in_full(fd
, hash
, the_hash_algo
->rawsz
, st
.st_size
- the_hash_algo
->rawsz
);
2724 if (n
!= the_hash_algo
->rawsz
)
2727 if (!hasheq(istate
->oid
.hash
, hash
))
2738 static int repo_verify_index(struct repository
*repo
)
2740 return verify_index_from(repo
->index
, repo
->index_file
);
2743 static int has_racy_timestamp(struct index_state
*istate
)
2745 int entries
= istate
->cache_nr
;
2748 for (i
= 0; i
< entries
; i
++) {
2749 struct cache_entry
*ce
= istate
->cache
[i
];
2750 if (is_racy_timestamp(istate
, ce
))
2756 void repo_update_index_if_able(struct repository
*repo
,
2757 struct lock_file
*lockfile
)
2759 if ((repo
->index
->cache_changed
||
2760 has_racy_timestamp(repo
->index
)) &&
2761 repo_verify_index(repo
))
2762 write_locked_index(repo
->index
, lockfile
, COMMIT_LOCK
);
2764 rollback_lock_file(lockfile
);
2767 static int record_eoie(void)
2771 if (!git_config_get_bool("index.recordendofindexentries", &val
))
2775 * As a convenience, the end of index entries extension
2776 * used for threading is written by default if the user
2777 * explicitly requested threaded index reads.
2779 return !git_config_get_index_threads(&val
) && val
!= 1;
2782 static int record_ieot(void)
2786 if (!git_config_get_bool("index.recordoffsettable", &val
))
2790 * As a convenience, the offset table used for threading is
2791 * written by default if the user explicitly requested
2792 * threaded index reads.
2794 return !git_config_get_index_threads(&val
) && val
!= 1;
2798 * On success, `tempfile` is closed. If it is the temporary file
2799 * of a `struct lock_file`, we will therefore effectively perform
2800 * a 'close_lock_file_gently()`. Since that is an implementation
2801 * detail of lockfiles, callers of `do_write_index()` should not
2804 static int do_write_index(struct index_state
*istate
, struct tempfile
*tempfile
,
2805 int strip_extensions
)
2807 uint64_t start
= getnanotime();
2809 git_hash_ctx
*eoie_c
= NULL
;
2810 struct cache_header hdr
;
2811 int i
, err
= 0, removed
, extended
, hdr_version
;
2812 struct cache_entry
**cache
= istate
->cache
;
2813 int entries
= istate
->cache_nr
;
2815 struct ondisk_cache_entry ondisk
;
2816 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
2817 int drop_cache_tree
= istate
->drop_cache_tree
;
2819 int ieot_entries
= 1;
2820 struct index_entry_offset_table
*ieot
= NULL
;
2823 f
= hashfd(tempfile
->fd
, tempfile
->filename
.buf
);
2825 for (i
= removed
= extended
= 0; i
< entries
; i
++) {
2826 if (cache
[i
]->ce_flags
& CE_REMOVE
)
2829 /* reduce extended entries if possible */
2830 cache
[i
]->ce_flags
&= ~CE_EXTENDED
;
2831 if (cache
[i
]->ce_flags
& CE_EXTENDED_FLAGS
) {
2833 cache
[i
]->ce_flags
|= CE_EXTENDED
;
2837 if (!istate
->version
) {
2838 istate
->version
= get_index_format_default(the_repository
);
2839 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0))
2840 init_split_index(istate
);
2843 /* demote version 3 to version 2 when the latter suffices */
2844 if (istate
->version
== 3 || istate
->version
== 2)
2845 istate
->version
= extended
? 3 : 2;
2847 hdr_version
= istate
->version
;
2849 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
2850 hdr
.hdr_version
= htonl(hdr_version
);
2851 hdr
.hdr_entries
= htonl(entries
- removed
);
2853 hashwrite(f
, &hdr
, sizeof(hdr
));
2855 if (!HAVE_THREADS
|| git_config_get_index_threads(&nr_threads
))
2858 if (nr_threads
!= 1 && record_ieot()) {
2859 int ieot_blocks
, cpus
;
2862 * ensure default number of ieot blocks maps evenly to the
2863 * default number of threads that will process them leaving
2864 * room for the thread to load the index extensions.
2867 ieot_blocks
= istate
->cache_nr
/ THREAD_COST
;
2868 cpus
= online_cpus();
2869 if (ieot_blocks
> cpus
- 1)
2870 ieot_blocks
= cpus
- 1;
2872 ieot_blocks
= nr_threads
;
2873 if (ieot_blocks
> istate
->cache_nr
)
2874 ieot_blocks
= istate
->cache_nr
;
2878 * no reason to write out the IEOT extension if we don't
2879 * have enough blocks to utilize multi-threading
2881 if (ieot_blocks
> 1) {
2882 ieot
= xcalloc(1, sizeof(struct index_entry_offset_table
)
2883 + (ieot_blocks
* sizeof(struct index_entry_offset
)));
2884 ieot_entries
= DIV_ROUND_UP(entries
, ieot_blocks
);
2888 offset
= hashfile_total(f
);
2891 previous_name
= (hdr_version
== 4) ? &previous_name_buf
: NULL
;
2893 for (i
= 0; i
< entries
; i
++) {
2894 struct cache_entry
*ce
= cache
[i
];
2895 if (ce
->ce_flags
& CE_REMOVE
)
2897 if (!ce_uptodate(ce
) && is_racy_timestamp(istate
, ce
))
2898 ce_smudge_racily_clean_entry(istate
, ce
);
2899 if (is_null_oid(&ce
->oid
)) {
2900 static const char msg
[] = "cache entry has null sha1: %s";
2901 static int allow
= -1;
2904 allow
= git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2906 warning(msg
, ce
->name
);
2908 err
= error(msg
, ce
->name
);
2910 drop_cache_tree
= 1;
2912 if (ieot
&& i
&& (i
% ieot_entries
== 0)) {
2913 ieot
->entries
[ieot
->nr
].nr
= nr
;
2914 ieot
->entries
[ieot
->nr
].offset
= offset
;
2917 * If we have a V4 index, set the first byte to an invalid
2918 * character to ensure there is nothing common with the previous
2922 previous_name
->buf
[0] = 0;
2925 offset
= hashfile_total(f
);
2927 if (ce_write_entry(f
, ce
, previous_name
, (struct ondisk_cache_entry
*)&ondisk
) < 0)
2935 ieot
->entries
[ieot
->nr
].nr
= nr
;
2936 ieot
->entries
[ieot
->nr
].offset
= offset
;
2939 strbuf_release(&previous_name_buf
);
2946 offset
= hashfile_total(f
);
2949 * The extension headers must be hashed on their own for the
2950 * EOIE extension. Create a hashfile here to compute that hash.
2952 if (offset
&& record_eoie()) {
2953 CALLOC_ARRAY(eoie_c
, 1);
2954 the_hash_algo
->init_fn(eoie_c
);
2958 * Lets write out CACHE_EXT_INDEXENTRYOFFSETTABLE first so that we
2959 * can minimize the number of extensions we have to scan through to
2960 * find it during load. Write it out regardless of the
2961 * strip_extensions parameter as we need it when loading the shared
2965 struct strbuf sb
= STRBUF_INIT
;
2967 write_ieot_extension(&sb
, ieot
);
2968 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_INDEXENTRYOFFSETTABLE
, sb
.len
) < 0;
2969 hashwrite(f
, sb
.buf
, sb
.len
);
2970 strbuf_release(&sb
);
2976 if (!strip_extensions
&& istate
->split_index
&&
2977 !is_null_oid(&istate
->split_index
->base_oid
)) {
2978 struct strbuf sb
= STRBUF_INIT
;
2980 err
= write_link_extension(&sb
, istate
) < 0 ||
2981 write_index_ext_header(f
, eoie_c
, CACHE_EXT_LINK
,
2983 hashwrite(f
, sb
.buf
, sb
.len
);
2984 strbuf_release(&sb
);
2988 if (!strip_extensions
&& !drop_cache_tree
&& istate
->cache_tree
) {
2989 struct strbuf sb
= STRBUF_INIT
;
2991 cache_tree_write(&sb
, istate
->cache_tree
);
2992 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_TREE
, sb
.len
) < 0;
2993 hashwrite(f
, sb
.buf
, sb
.len
);
2994 strbuf_release(&sb
);
2998 if (!strip_extensions
&& istate
->resolve_undo
) {
2999 struct strbuf sb
= STRBUF_INIT
;
3001 resolve_undo_write(&sb
, istate
->resolve_undo
);
3002 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_RESOLVE_UNDO
,
3004 hashwrite(f
, sb
.buf
, sb
.len
);
3005 strbuf_release(&sb
);
3009 if (!strip_extensions
&& istate
->untracked
) {
3010 struct strbuf sb
= STRBUF_INIT
;
3012 write_untracked_extension(&sb
, istate
->untracked
);
3013 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_UNTRACKED
,
3015 hashwrite(f
, sb
.buf
, sb
.len
);
3016 strbuf_release(&sb
);
3020 if (!strip_extensions
&& istate
->fsmonitor_last_update
) {
3021 struct strbuf sb
= STRBUF_INIT
;
3023 write_fsmonitor_extension(&sb
, istate
);
3024 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_FSMONITOR
, sb
.len
) < 0;
3025 hashwrite(f
, sb
.buf
, sb
.len
);
3026 strbuf_release(&sb
);
3030 if (istate
->sparse_index
) {
3031 if (write_index_ext_header(f
, eoie_c
, CACHE_EXT_SPARSE_DIRECTORIES
, 0) < 0)
3036 * CACHE_EXT_ENDOFINDEXENTRIES must be written as the last entry before the SHA1
3037 * so that it can be found and processed before all the index entries are
3038 * read. Write it out regardless of the strip_extensions parameter as we need it
3039 * when loading the shared index.
3042 struct strbuf sb
= STRBUF_INIT
;
3044 write_eoie_extension(&sb
, eoie_c
, offset
);
3045 err
= write_index_ext_header(f
, NULL
, CACHE_EXT_ENDOFINDEXENTRIES
, sb
.len
) < 0;
3046 hashwrite(f
, sb
.buf
, sb
.len
);
3047 strbuf_release(&sb
);
3052 finalize_hashfile(f
, istate
->oid
.hash
, CSUM_HASH_IN_STREAM
);
3053 if (close_tempfile_gently(tempfile
)) {
3054 error(_("could not close '%s'"), get_tempfile_path(tempfile
));
3057 if (stat(get_tempfile_path(tempfile
), &st
))
3059 istate
->timestamp
.sec
= (unsigned int)st
.st_mtime
;
3060 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
3061 trace_performance_since(start
, "write index, changed mask = %x", istate
->cache_changed
);
3064 * TODO trace2: replace "the_repository" with the actual repo instance
3065 * that is associated with the given "istate".
3067 trace2_data_intmax("index", the_repository
, "write/version",
3069 trace2_data_intmax("index", the_repository
, "write/cache_nr",
3075 void set_alternate_index_output(const char *name
)
3077 alternate_index_output
= name
;
3080 static int commit_locked_index(struct lock_file
*lk
)
3082 if (alternate_index_output
)
3083 return commit_lock_file_to(lk
, alternate_index_output
);
3085 return commit_lock_file(lk
);
3088 static int do_write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
3092 int was_full
= !istate
->sparse_index
;
3094 ret
= convert_to_sparse(istate
, 0);
3097 warning(_("failed to convert to a sparse-index"));
3102 * TODO trace2: replace "the_repository" with the actual repo instance
3103 * that is associated with the given "istate".
3105 trace2_region_enter_printf("index", "do_write_index", the_repository
,
3106 "%s", get_lock_file_path(lock
));
3107 ret
= do_write_index(istate
, lock
->tempfile
, 0);
3108 trace2_region_leave_printf("index", "do_write_index", the_repository
,
3109 "%s", get_lock_file_path(lock
));
3112 ensure_full_index(istate
);
3116 if (flags
& COMMIT_LOCK
)
3117 ret
= commit_locked_index(lock
);
3119 ret
= close_lock_file_gently(lock
);
3121 run_hook_le(NULL
, "post-index-change",
3122 istate
->updated_workdir
? "1" : "0",
3123 istate
->updated_skipworktree
? "1" : "0", NULL
);
3124 istate
->updated_workdir
= 0;
3125 istate
->updated_skipworktree
= 0;
3130 static int write_split_index(struct index_state
*istate
,
3131 struct lock_file
*lock
,
3135 prepare_to_write_split_index(istate
);
3136 ret
= do_write_locked_index(istate
, lock
, flags
);
3137 finish_writing_split_index(istate
);
3141 static const char *shared_index_expire
= "2.weeks.ago";
3143 static unsigned long get_shared_index_expire_date(void)
3145 static unsigned long shared_index_expire_date
;
3146 static int shared_index_expire_date_prepared
;
3148 if (!shared_index_expire_date_prepared
) {
3149 git_config_get_expiry("splitindex.sharedindexexpire",
3150 &shared_index_expire
);
3151 shared_index_expire_date
= approxidate(shared_index_expire
);
3152 shared_index_expire_date_prepared
= 1;
3155 return shared_index_expire_date
;
3158 static int should_delete_shared_index(const char *shared_index_path
)
3161 unsigned long expiration
;
3163 /* Check timestamp */
3164 expiration
= get_shared_index_expire_date();
3167 if (stat(shared_index_path
, &st
))
3168 return error_errno(_("could not stat '%s'"), shared_index_path
);
3169 if (st
.st_mtime
> expiration
)
3175 static int clean_shared_index_files(const char *current_hex
)
3178 DIR *dir
= opendir(get_git_dir());
3181 return error_errno(_("unable to open git dir: %s"), get_git_dir());
3183 while ((de
= readdir(dir
)) != NULL
) {
3184 const char *sha1_hex
;
3185 const char *shared_index_path
;
3186 if (!skip_prefix(de
->d_name
, "sharedindex.", &sha1_hex
))
3188 if (!strcmp(sha1_hex
, current_hex
))
3190 shared_index_path
= git_path("%s", de
->d_name
);
3191 if (should_delete_shared_index(shared_index_path
) > 0 &&
3192 unlink(shared_index_path
))
3193 warning_errno(_("unable to unlink: %s"), shared_index_path
);
3200 static int write_shared_index(struct index_state
*istate
,
3201 struct tempfile
**temp
)
3203 struct split_index
*si
= istate
->split_index
;
3204 int ret
, was_full
= !istate
->sparse_index
;
3206 move_cache_to_base_index(istate
);
3207 convert_to_sparse(istate
, 0);
3209 trace2_region_enter_printf("index", "shared/do_write_index",
3210 the_repository
, "%s", get_tempfile_path(*temp
));
3211 ret
= do_write_index(si
->base
, *temp
, 1);
3212 trace2_region_leave_printf("index", "shared/do_write_index",
3213 the_repository
, "%s", get_tempfile_path(*temp
));
3216 ensure_full_index(istate
);
3220 ret
= adjust_shared_perm(get_tempfile_path(*temp
));
3222 error(_("cannot fix permission bits on '%s'"), get_tempfile_path(*temp
));
3225 ret
= rename_tempfile(temp
,
3226 git_path("sharedindex.%s", oid_to_hex(&si
->base
->oid
)));
3228 oidcpy(&si
->base_oid
, &si
->base
->oid
);
3229 clean_shared_index_files(oid_to_hex(&si
->base
->oid
));
3235 static const int default_max_percent_split_change
= 20;
3237 static int too_many_not_shared_entries(struct index_state
*istate
)
3239 int i
, not_shared
= 0;
3240 int max_split
= git_config_get_max_percent_split_change();
3242 switch (max_split
) {
3244 /* not or badly configured: use the default value */
3245 max_split
= default_max_percent_split_change
;
3248 return 1; /* 0% means always write a new shared index */
3250 return 0; /* 100% means never write a new shared index */
3252 break; /* just use the configured value */
3255 /* Count not shared entries */
3256 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3257 struct cache_entry
*ce
= istate
->cache
[i
];
3262 return (int64_t)istate
->cache_nr
* max_split
< (int64_t)not_shared
* 100;
3265 int write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
3268 int new_shared_index
, ret
;
3269 struct split_index
*si
= istate
->split_index
;
3271 if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0))
3272 cache_tree_verify(the_repository
, istate
);
3274 if ((flags
& SKIP_IF_UNCHANGED
) && !istate
->cache_changed
) {
3275 if (flags
& COMMIT_LOCK
)
3276 rollback_lock_file(lock
);
3280 if (istate
->fsmonitor_last_update
)
3281 fill_fsmonitor_bitmap(istate
);
3283 if (!si
|| alternate_index_output
||
3284 (istate
->cache_changed
& ~EXTMASK
)) {
3286 oidclr(&si
->base_oid
);
3287 ret
= do_write_locked_index(istate
, lock
, flags
);
3291 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0)) {
3292 int v
= si
->base_oid
.hash
[0];
3294 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3296 if (too_many_not_shared_entries(istate
))
3297 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3299 new_shared_index
= istate
->cache_changed
& SPLIT_INDEX_ORDERED
;
3301 if (new_shared_index
) {
3302 struct tempfile
*temp
;
3305 /* Same initial permissions as the main .git/index file */
3306 temp
= mks_tempfile_sm(git_path("sharedindex_XXXXXX"), 0, 0666);
3308 oidclr(&si
->base_oid
);
3309 ret
= do_write_locked_index(istate
, lock
, flags
);
3312 ret
= write_shared_index(istate
, &temp
);
3314 saved_errno
= errno
;
3315 if (is_tempfile_active(temp
))
3316 delete_tempfile(&temp
);
3317 errno
= saved_errno
;
3323 ret
= write_split_index(istate
, lock
, flags
);
3325 /* Freshen the shared index only if the split-index was written */
3326 if (!ret
&& !new_shared_index
&& !is_null_oid(&si
->base_oid
)) {
3327 const char *shared_index
= git_path("sharedindex.%s",
3328 oid_to_hex(&si
->base_oid
));
3329 freshen_shared_index(shared_index
, 1);
3333 if (flags
& COMMIT_LOCK
)
3334 rollback_lock_file(lock
);
3339 * Read the index file that is potentially unmerged into given
3340 * index_state, dropping any unmerged entries to stage #0 (potentially
3341 * resulting in a path appearing as both a file and a directory in the
3342 * index; the caller is responsible to clear out the extra entries
3343 * before writing the index to a tree). Returns true if the index is
3344 * unmerged. Callers who want to refuse to work from an unmerged
3345 * state can call this and check its return value, instead of calling
3348 int repo_read_index_unmerged(struct repository
*repo
)
3350 struct index_state
*istate
;
3354 repo_read_index(repo
);
3355 istate
= repo
->index
;
3356 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3357 struct cache_entry
*ce
= istate
->cache
[i
];
3358 struct cache_entry
*new_ce
;
3364 len
= ce_namelen(ce
);
3365 new_ce
= make_empty_cache_entry(istate
, len
);
3366 memcpy(new_ce
->name
, ce
->name
, len
);
3367 new_ce
->ce_flags
= create_ce_flags(0) | CE_CONFLICTED
;
3368 new_ce
->ce_namelen
= len
;
3369 new_ce
->ce_mode
= ce
->ce_mode
;
3370 if (add_index_entry(istate
, new_ce
, ADD_CACHE_SKIP_DFCHECK
))
3371 return error(_("%s: cannot drop to stage #0"),
3378 * Returns 1 if the path is an "other" path with respect to
3379 * the index; that is, the path is not mentioned in the index at all,
3380 * either as a file, a directory with some files in the index,
3381 * or as an unmerged entry.
3383 * We helpfully remove a trailing "/" from directories so that
3384 * the output of read_directory can be used as-is.
3386 int index_name_is_other(struct index_state
*istate
, const char *name
,
3390 if (namelen
&& name
[namelen
- 1] == '/')
3392 pos
= index_name_pos(istate
, name
, namelen
);
3394 return 0; /* exact match */
3396 if (pos
< istate
->cache_nr
) {
3397 struct cache_entry
*ce
= istate
->cache
[pos
];
3398 if (ce_namelen(ce
) == namelen
&&
3399 !memcmp(ce
->name
, name
, namelen
))
3400 return 0; /* Yup, this one exists unmerged */
3405 void *read_blob_data_from_index(struct index_state
*istate
,
3406 const char *path
, unsigned long *size
)
3410 enum object_type type
;
3414 pos
= index_name_pos(istate
, path
, len
);
3417 * We might be in the middle of a merge, in which
3418 * case we would read stage #2 (ours).
3422 (pos
< 0 && i
< istate
->cache_nr
&&
3423 !strcmp(istate
->cache
[i
]->name
, path
));
3425 if (ce_stage(istate
->cache
[i
]) == 2)
3430 data
= read_object_file(&istate
->cache
[pos
]->oid
, &type
, &sz
);
3431 if (!data
|| type
!= OBJ_BLOB
) {
3440 void stat_validity_clear(struct stat_validity
*sv
)
3442 FREE_AND_NULL(sv
->sd
);
3445 int stat_validity_check(struct stat_validity
*sv
, const char *path
)
3449 if (stat(path
, &st
) < 0)
3450 return sv
->sd
== NULL
;
3453 return S_ISREG(st
.st_mode
) && !match_stat_data(sv
->sd
, &st
);
3456 void stat_validity_update(struct stat_validity
*sv
, int fd
)
3460 if (fstat(fd
, &st
) < 0 || !S_ISREG(st
.st_mode
))
3461 stat_validity_clear(sv
);
3464 CALLOC_ARRAY(sv
->sd
, 1);
3465 fill_stat_data(sv
->sd
, &st
);
3469 void move_index_extensions(struct index_state
*dst
, struct index_state
*src
)
3471 dst
->untracked
= src
->untracked
;
3472 src
->untracked
= NULL
;
3473 dst
->cache_tree
= src
->cache_tree
;
3474 src
->cache_tree
= NULL
;
3477 struct cache_entry
*dup_cache_entry(const struct cache_entry
*ce
,
3478 struct index_state
*istate
)
3480 unsigned int size
= ce_size(ce
);
3481 int mem_pool_allocated
;
3482 struct cache_entry
*new_entry
= make_empty_cache_entry(istate
, ce_namelen(ce
));
3483 mem_pool_allocated
= new_entry
->mem_pool_allocated
;
3485 memcpy(new_entry
, ce
, size
);
3486 new_entry
->mem_pool_allocated
= mem_pool_allocated
;
3490 void discard_cache_entry(struct cache_entry
*ce
)
3492 if (ce
&& should_validate_cache_entries())
3493 memset(ce
, 0xCD, cache_entry_size(ce
->ce_namelen
));
3495 if (ce
&& ce
->mem_pool_allocated
)
3501 int should_validate_cache_entries(void)
3503 static int validate_index_cache_entries
= -1;
3505 if (validate_index_cache_entries
< 0) {
3506 if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))
3507 validate_index_cache_entries
= 1;
3509 validate_index_cache_entries
= 0;
3512 return validate_index_cache_entries
;
3515 #define EOIE_SIZE (4 + GIT_SHA1_RAWSZ) /* <4-byte offset> + <20-byte hash> */
3516 #define EOIE_SIZE_WITH_HEADER (4 + 4 + EOIE_SIZE) /* <4-byte signature> + <4-byte length> + EOIE_SIZE */
3518 static size_t read_eoie_extension(const char *mmap
, size_t mmap_size
)
3521 * The end of index entries (EOIE) extension is guaranteed to be last
3522 * so that it can be found by scanning backwards from the EOF.
3529 const char *index
, *eoie
;
3531 size_t offset
, src_offset
;
3532 unsigned char hash
[GIT_MAX_RAWSZ
];
3535 /* ensure we have an index big enough to contain an EOIE extension */
3536 if (mmap_size
< sizeof(struct cache_header
) + EOIE_SIZE_WITH_HEADER
+ the_hash_algo
->rawsz
)
3539 /* validate the extension signature */
3540 index
= eoie
= mmap
+ mmap_size
- EOIE_SIZE_WITH_HEADER
- the_hash_algo
->rawsz
;
3541 if (CACHE_EXT(index
) != CACHE_EXT_ENDOFINDEXENTRIES
)
3543 index
+= sizeof(uint32_t);
3545 /* validate the extension size */
3546 extsize
= get_be32(index
);
3547 if (extsize
!= EOIE_SIZE
)
3549 index
+= sizeof(uint32_t);
3552 * Validate the offset we're going to look for the first extension
3553 * signature is after the index header and before the eoie extension.
3555 offset
= get_be32(index
);
3556 if (mmap
+ offset
< mmap
+ sizeof(struct cache_header
))
3558 if (mmap
+ offset
>= eoie
)
3560 index
+= sizeof(uint32_t);
3563 * The hash is computed over extension types and their sizes (but not
3564 * their contents). E.g. if we have "TREE" extension that is N-bytes
3565 * long, "REUC" extension that is M-bytes long, followed by "EOIE",
3566 * then the hash would be:
3568 * SHA-1("TREE" + <binary representation of N> +
3569 * "REUC" + <binary representation of M>)
3571 src_offset
= offset
;
3572 the_hash_algo
->init_fn(&c
);
3573 while (src_offset
< mmap_size
- the_hash_algo
->rawsz
- EOIE_SIZE_WITH_HEADER
) {
3574 /* After an array of active_nr index entries,
3575 * there can be arbitrary number of extended
3576 * sections, each of which is prefixed with
3577 * extension name (4-byte) and section length
3578 * in 4-byte network byte order.
3581 memcpy(&extsize
, mmap
+ src_offset
+ 4, 4);
3582 extsize
= ntohl(extsize
);
3584 /* verify the extension size isn't so large it will wrap around */
3585 if (src_offset
+ 8 + extsize
< src_offset
)
3588 the_hash_algo
->update_fn(&c
, mmap
+ src_offset
, 8);
3591 src_offset
+= extsize
;
3593 the_hash_algo
->final_fn(hash
, &c
);
3594 if (!hasheq(hash
, (const unsigned char *)index
))
3597 /* Validate that the extension offsets returned us back to the eoie extension. */
3598 if (src_offset
!= mmap_size
- the_hash_algo
->rawsz
- EOIE_SIZE_WITH_HEADER
)
3604 static void write_eoie_extension(struct strbuf
*sb
, git_hash_ctx
*eoie_context
, size_t offset
)
3607 unsigned char hash
[GIT_MAX_RAWSZ
];
3610 put_be32(&buffer
, offset
);
3611 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3614 the_hash_algo
->final_fn(hash
, eoie_context
);
3615 strbuf_add(sb
, hash
, the_hash_algo
->rawsz
);
3618 #define IEOT_VERSION (1)
3620 static struct index_entry_offset_table
*read_ieot_extension(const char *mmap
, size_t mmap_size
, size_t offset
)
3622 const char *index
= NULL
;
3623 uint32_t extsize
, ext_version
;
3624 struct index_entry_offset_table
*ieot
;
3627 /* find the IEOT extension */
3630 while (offset
<= mmap_size
- the_hash_algo
->rawsz
- 8) {
3631 extsize
= get_be32(mmap
+ offset
+ 4);
3632 if (CACHE_EXT((mmap
+ offset
)) == CACHE_EXT_INDEXENTRYOFFSETTABLE
) {
3633 index
= mmap
+ offset
+ 4 + 4;
3642 /* validate the version is IEOT_VERSION */
3643 ext_version
= get_be32(index
);
3644 if (ext_version
!= IEOT_VERSION
) {
3645 error("invalid IEOT version %d", ext_version
);
3648 index
+= sizeof(uint32_t);
3650 /* extension size - version bytes / bytes per entry */
3651 nr
= (extsize
- sizeof(uint32_t)) / (sizeof(uint32_t) + sizeof(uint32_t));
3653 error("invalid number of IEOT entries %d", nr
);
3656 ieot
= xmalloc(sizeof(struct index_entry_offset_table
)
3657 + (nr
* sizeof(struct index_entry_offset
)));
3659 for (i
= 0; i
< nr
; i
++) {
3660 ieot
->entries
[i
].offset
= get_be32(index
);
3661 index
+= sizeof(uint32_t);
3662 ieot
->entries
[i
].nr
= get_be32(index
);
3663 index
+= sizeof(uint32_t);
3669 static void write_ieot_extension(struct strbuf
*sb
, struct index_entry_offset_table
*ieot
)
3675 put_be32(&buffer
, IEOT_VERSION
);
3676 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3679 for (i
= 0; i
< ieot
->nr
; i
++) {
3682 put_be32(&buffer
, ieot
->entries
[i
].offset
);
3683 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3686 put_be32(&buffer
, ieot
->entries
[i
].nr
);
3687 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3691 void prefetch_cache_entries(const struct index_state
*istate
,
3692 must_prefetch_predicate must_prefetch
)
3695 struct oid_array to_fetch
= OID_ARRAY_INIT
;
3697 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3698 struct cache_entry
*ce
= istate
->cache
[i
];
3700 if (S_ISGITLINK(ce
->ce_mode
) || !must_prefetch(ce
))
3702 if (!oid_object_info_extended(the_repository
, &ce
->oid
,
3704 OBJECT_INFO_FOR_PREFETCH
))
3706 oid_array_append(&to_fetch
, &ce
->oid
);
3708 promisor_remote_get_direct(the_repository
,
3709 to_fetch
.oid
, to_fetch
.nr
);
3710 oid_array_clear(&to_fetch
);