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 unsigned hash_flags
= pretend
? 0 : 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 enum verify_path_result
{
871 static enum verify_path_result
verify_path_internal(const char *, unsigned);
873 int verify_path(const char *path
, unsigned mode
)
875 return verify_path_internal(path
, mode
) == PATH_OK
;
878 struct cache_entry
*make_cache_entry(struct index_state
*istate
,
880 const struct object_id
*oid
,
883 unsigned int refresh_options
)
885 struct cache_entry
*ce
, *ret
;
888 if (verify_path_internal(path
, mode
) == PATH_INVALID
) {
889 error(_("invalid path '%s'"), path
);
894 ce
= make_empty_cache_entry(istate
, len
);
896 oidcpy(&ce
->oid
, oid
);
897 memcpy(ce
->name
, path
, len
);
898 ce
->ce_flags
= create_ce_flags(stage
);
899 ce
->ce_namelen
= len
;
900 ce
->ce_mode
= create_ce_mode(mode
);
902 ret
= refresh_cache_entry(istate
, ce
, refresh_options
);
904 discard_cache_entry(ce
);
908 struct cache_entry
*make_transient_cache_entry(unsigned int mode
,
909 const struct object_id
*oid
,
912 struct mem_pool
*ce_mem_pool
)
914 struct cache_entry
*ce
;
917 if (!verify_path(path
, mode
)) {
918 error(_("invalid path '%s'"), path
);
923 ce
= make_empty_transient_cache_entry(len
, ce_mem_pool
);
925 oidcpy(&ce
->oid
, oid
);
926 memcpy(ce
->name
, path
, len
);
927 ce
->ce_flags
= create_ce_flags(stage
);
928 ce
->ce_namelen
= len
;
929 ce
->ce_mode
= create_ce_mode(mode
);
935 * Chmod an index entry with either +x or -x.
937 * Returns -1 if the chmod for the particular cache entry failed (if it's
938 * not a regular file), -2 if an invalid flip argument is passed in, 0
941 int chmod_index_entry(struct index_state
*istate
, struct cache_entry
*ce
,
944 if (!S_ISREG(ce
->ce_mode
))
951 ce
->ce_mode
&= ~0111;
956 cache_tree_invalidate_path(istate
, ce
->name
);
957 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
958 mark_fsmonitor_invalid(istate
, ce
);
959 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
964 int ce_same_name(const struct cache_entry
*a
, const struct cache_entry
*b
)
966 int len
= ce_namelen(a
);
967 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
971 * We fundamentally don't like some paths: we don't want
972 * dot or dot-dot anywhere, and for obvious reasons don't
973 * want to recurse into ".git" either.
975 * Also, we don't want double slashes or slashes at the
976 * end that can make pathnames ambiguous.
978 static int verify_dotfile(const char *rest
, unsigned mode
)
981 * The first character was '.', but that
982 * has already been discarded, we now test
986 /* "." is not allowed */
987 if (*rest
== '\0' || is_dir_sep(*rest
))
992 * ".git" followed by NUL or slash is bad. Note that we match
993 * case-insensitively here, even if ignore_case is not set.
994 * This outlaws ".GIT" everywhere out of an abundance of caution,
995 * since there's really no good reason to allow it.
997 * Once we've seen ".git", we can also find ".gitmodules", etc (also
998 * case-insensitively).
1002 if (rest
[1] != 'i' && rest
[1] != 'I')
1004 if (rest
[2] != 't' && rest
[2] != 'T')
1006 if (rest
[3] == '\0' || is_dir_sep(rest
[3]))
1008 if (S_ISLNK(mode
)) {
1010 if (skip_iprefix(rest
, "modules", &rest
) &&
1011 (*rest
== '\0' || is_dir_sep(*rest
)))
1016 if (rest
[1] == '\0' || is_dir_sep(rest
[1]))
1022 static enum verify_path_result
verify_path_internal(const char *path
,
1027 if (has_dos_drive_prefix(path
))
1028 return PATH_INVALID
;
1030 if (!is_valid_path(path
))
1031 return PATH_INVALID
;
1037 if (is_dir_sep(c
)) {
1041 if (is_hfs_dotgit(path
))
1042 return PATH_INVALID
;
1043 if (S_ISLNK(mode
)) {
1044 if (is_hfs_dotgitmodules(path
))
1045 return PATH_INVALID
;
1049 #if defined GIT_WINDOWS_NATIVE || defined __CYGWIN__
1051 return PATH_INVALID
;
1053 if (is_ntfs_dotgit(path
))
1054 return PATH_INVALID
;
1055 if (S_ISLNK(mode
)) {
1056 if (is_ntfs_dotgitmodules(path
))
1057 return PATH_INVALID
;
1062 if ((c
== '.' && !verify_dotfile(path
, mode
)) ||
1064 return PATH_INVALID
;
1066 * allow terminating directory separators for
1067 * sparse directory entries.
1070 return S_ISDIR(mode
) ? PATH_DIR_WITH_SEP
:
1072 } else if (c
== '\\' && protect_ntfs
) {
1073 if (is_ntfs_dotgit(path
))
1074 return PATH_INVALID
;
1075 if (S_ISLNK(mode
)) {
1076 if (is_ntfs_dotgitmodules(path
))
1077 return PATH_INVALID
;
1086 * Do we have another file that has the beginning components being a
1087 * proper superset of the name we're trying to add?
1089 static int has_file_name(struct index_state
*istate
,
1090 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
1093 int len
= ce_namelen(ce
);
1094 int stage
= ce_stage(ce
);
1095 const char *name
= ce
->name
;
1097 while (pos
< istate
->cache_nr
) {
1098 struct cache_entry
*p
= istate
->cache
[pos
++];
1100 if (len
>= ce_namelen(p
))
1102 if (memcmp(name
, p
->name
, len
))
1104 if (ce_stage(p
) != stage
)
1106 if (p
->name
[len
] != '/')
1108 if (p
->ce_flags
& CE_REMOVE
)
1113 remove_index_entry_at(istate
, --pos
);
1120 * Like strcmp(), but also return the offset of the first change.
1121 * If strings are equal, return the length.
1123 int strcmp_offset(const char *s1
, const char *s2
, size_t *first_change
)
1128 return strcmp(s1
, s2
);
1130 for (k
= 0; s1
[k
] == s2
[k
]; k
++)
1135 return (unsigned char)s1
[k
] - (unsigned char)s2
[k
];
1139 * Do we have another file with a pathname that is a proper
1140 * subset of the name we're trying to add?
1142 * That is, is there another file in the index with a path
1143 * that matches a sub-directory in the given entry?
1145 static int has_dir_name(struct index_state
*istate
,
1146 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
1149 int stage
= ce_stage(ce
);
1150 const char *name
= ce
->name
;
1151 const char *slash
= name
+ ce_namelen(ce
);
1156 * We are frequently called during an iteration on a sorted
1157 * list of pathnames and while building a new index. Therefore,
1158 * there is a high probability that this entry will eventually
1159 * be appended to the index, rather than inserted in the middle.
1160 * If we can confirm that, we can avoid binary searches on the
1161 * components of the pathname.
1163 * Compare the entry's full path with the last path in the index.
1165 if (istate
->cache_nr
> 0) {
1166 cmp_last
= strcmp_offset(name
,
1167 istate
->cache
[istate
->cache_nr
- 1]->name
,
1170 if (len_eq_last
== 0) {
1172 * The entry sorts AFTER the last one in the
1173 * index and their paths have no common prefix,
1174 * so there cannot be a F/D conflict.
1179 * The entry sorts AFTER the last one in the
1180 * index, but has a common prefix. Fall through
1181 * to the loop below to disect the entry's path
1182 * and see where the difference is.
1185 } else if (cmp_last
== 0) {
1187 * The entry exactly matches the last one in the
1188 * index, but because of multiple stage and CE_REMOVE
1189 * items, we fall through and let the regular search
1199 if (*--slash
== '/')
1201 if (slash
<= ce
->name
)
1208 * (len + 1) is a directory boundary (including
1209 * the trailing slash). And since the loop is
1210 * decrementing "slash", the first iteration is
1211 * the longest directory prefix; subsequent
1212 * iterations consider parent directories.
1215 if (len
+ 1 <= len_eq_last
) {
1217 * The directory prefix (including the trailing
1218 * slash) also appears as a prefix in the last
1219 * entry, so the remainder cannot collide (because
1220 * strcmp said the whole path was greater).
1225 * LT: last: xxx/file_A
1231 if (len
> len_eq_last
) {
1233 * This part of the directory prefix (excluding
1234 * the trailing slash) is longer than the known
1235 * equal portions, so this sub-directory cannot
1236 * collide with a file.
1245 * This is a possible collision. Fall through and
1246 * let the regular search code handle it.
1253 pos
= index_name_stage_pos(istate
, name
, len
, stage
, EXPAND_SPARSE
);
1256 * Found one, but not so fast. This could
1257 * be a marker that says "I was here, but
1258 * I am being removed". Such an entry is
1259 * not a part of the resulting tree, and
1260 * it is Ok to have a directory at the same
1263 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
1267 remove_index_entry_at(istate
, pos
);
1275 * Trivial optimization: if we find an entry that
1276 * already matches the sub-directory, then we know
1277 * we're ok, and we can exit.
1279 while (pos
< istate
->cache_nr
) {
1280 struct cache_entry
*p
= istate
->cache
[pos
];
1281 if ((ce_namelen(p
) <= len
) ||
1282 (p
->name
[len
] != '/') ||
1283 memcmp(p
->name
, name
, len
))
1284 break; /* not our subdirectory */
1285 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
1287 * p is at the same stage as our entry, and
1288 * is a subdirectory of what we are looking
1289 * at, so we cannot have conflicts at our
1290 * level or anything shorter.
1299 /* We may be in a situation where we already have path/file and path
1300 * is being added, or we already have path and path/file is being
1301 * added. Either one would result in a nonsense tree that has path
1302 * twice when git-write-tree tries to write it out. Prevent it.
1304 * If ok-to-replace is specified, we remove the conflicting entries
1305 * from the cache so the caller should recompute the insert position.
1306 * When this happens, we return non-zero.
1308 static int check_file_directory_conflict(struct index_state
*istate
,
1309 const struct cache_entry
*ce
,
1310 int pos
, int ok_to_replace
)
1315 * When ce is an "I am going away" entry, we allow it to be added
1317 if (ce
->ce_flags
& CE_REMOVE
)
1321 * We check if the path is a sub-path of a subsequent pathname
1322 * first, since removing those will not change the position
1325 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
1328 * Then check if the path might have a clashing sub-directory
1331 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
1334 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1337 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
1338 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
1339 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
1340 int new_only
= option
& ADD_CACHE_NEW_ONLY
;
1343 * If this entry's path sorts after the last entry in the index,
1344 * we can avoid searching for it.
1346 if (istate
->cache_nr
> 0 &&
1347 strcmp(ce
->name
, istate
->cache
[istate
->cache_nr
- 1]->name
) > 0)
1348 pos
= index_pos_to_insert_pos(istate
->cache_nr
);
1350 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
), EXPAND_SPARSE
);
1353 * Cache tree path should be invalidated only after index_name_stage_pos,
1354 * in case it expands a sparse index.
1356 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1357 cache_tree_invalidate_path(istate
, ce
->name
);
1359 /* existing match? Just replace it. */
1362 replace_index_entry(istate
, pos
, ce
);
1367 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1368 untracked_cache_add_to_index(istate
, ce
->name
);
1371 * Inserting a merged entry ("stage 0") into the index
1372 * will always replace all non-merged entries..
1374 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
1375 while (ce_same_name(istate
->cache
[pos
], ce
)) {
1377 if (!remove_index_entry_at(istate
, pos
))
1384 if (verify_path_internal(ce
->name
, ce
->ce_mode
) == PATH_INVALID
)
1385 return error(_("invalid path '%s'"), ce
->name
);
1387 if (!skip_df_check
&&
1388 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
1390 return error(_("'%s' appears as both a file and as a directory"),
1392 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
), EXPAND_SPARSE
);
1398 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1402 if (option
& ADD_CACHE_JUST_APPEND
)
1403 pos
= istate
->cache_nr
;
1406 ret
= add_index_entry_with_check(istate
, ce
, option
);
1412 /* Make sure the array is big enough .. */
1413 ALLOC_GROW(istate
->cache
, istate
->cache_nr
+ 1, istate
->cache_alloc
);
1417 if (istate
->cache_nr
> pos
+ 1)
1418 MOVE_ARRAY(istate
->cache
+ pos
+ 1, istate
->cache
+ pos
,
1419 istate
->cache_nr
- pos
- 1);
1420 set_index_entry(istate
, pos
, ce
);
1421 istate
->cache_changed
|= CE_ENTRY_ADDED
;
1426 * "refresh" does not calculate a new sha1 file or bring the
1427 * cache up-to-date for mode/content changes. But what it
1428 * _does_ do is to "re-match" the stat information of a file
1429 * with the cache, so that you can refresh the cache for a
1430 * file that hasn't been changed but where the stat entry is
1433 * For example, you'd want to do this after doing a "git-read-tree",
1434 * to link up the stat cache details with the proper files.
1436 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
1437 struct cache_entry
*ce
,
1438 unsigned int options
, int *err
,
1444 struct cache_entry
*updated
;
1446 int refresh
= options
& CE_MATCH_REFRESH
;
1447 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
1448 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
1449 int ignore_missing
= options
& CE_MATCH_IGNORE_MISSING
;
1450 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
1452 if (!refresh
|| ce_uptodate(ce
))
1455 if (!ignore_fsmonitor
)
1456 refresh_fsmonitor(istate
);
1458 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1459 * that the change to the work tree does not matter and told
1462 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
)) {
1463 ce_mark_uptodate(ce
);
1466 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
)) {
1467 ce_mark_uptodate(ce
);
1470 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
)) {
1471 ce_mark_uptodate(ce
);
1475 if (has_symlink_leading_path(ce
->name
, ce_namelen(ce
))) {
1485 if (lstat(ce
->name
, &st
) < 0) {
1486 if (ignore_missing
&& errno
== ENOENT
)
1493 changed
= ie_match_stat(istate
, ce
, &st
, options
);
1495 *changed_ret
= changed
;
1498 * The path is unchanged. If we were told to ignore
1499 * valid bit, then we did the actual stat check and
1500 * found that the entry is unmodified. If the entry
1501 * is not marked VALID, this is the place to mark it
1502 * valid again, under "assume unchanged" mode.
1504 if (ignore_valid
&& assume_unchanged
&&
1505 !(ce
->ce_flags
& CE_VALID
))
1506 ; /* mark this one VALID again */
1509 * We do not mark the index itself "modified"
1510 * because CE_UPTODATE flag is in-core only;
1511 * we are not going to write this change out.
1513 if (!S_ISGITLINK(ce
->ce_mode
)) {
1514 ce_mark_uptodate(ce
);
1515 mark_fsmonitor_valid(istate
, ce
);
1523 if (ie_modified(istate
, ce
, &st
, options
)) {
1529 updated
= make_empty_cache_entry(istate
, ce_namelen(ce
));
1530 copy_cache_entry(updated
, ce
);
1531 memcpy(updated
->name
, ce
->name
, ce
->ce_namelen
+ 1);
1532 fill_stat_cache_info(istate
, updated
, &st
);
1534 * If ignore_valid is not set, we should leave CE_VALID bit
1535 * alone. Otherwise, paths marked with --no-assume-unchanged
1536 * (i.e. things to be edited) will reacquire CE_VALID bit
1537 * automatically, which is not really what we want.
1539 if (!ignore_valid
&& assume_unchanged
&&
1540 !(ce
->ce_flags
& CE_VALID
))
1541 updated
->ce_flags
&= ~CE_VALID
;
1543 /* istate->cache_changed is updated in the caller */
1547 static void show_file(const char * fmt
, const char * name
, int in_porcelain
,
1548 int * first
, const char *header_msg
)
1550 if (in_porcelain
&& *first
&& header_msg
) {
1551 printf("%s\n", header_msg
);
1557 int repo_refresh_and_write_index(struct repository
*repo
,
1558 unsigned int refresh_flags
,
1559 unsigned int write_flags
,
1561 const struct pathspec
*pathspec
,
1562 char *seen
, const char *header_msg
)
1564 struct lock_file lock_file
= LOCK_INIT
;
1567 fd
= repo_hold_locked_index(repo
, &lock_file
, 0);
1568 if (!gentle
&& fd
< 0)
1570 if (refresh_index(repo
->index
, refresh_flags
, pathspec
, seen
, header_msg
))
1572 if (0 <= fd
&& write_locked_index(repo
->index
, &lock_file
, COMMIT_LOCK
| write_flags
))
1578 int refresh_index(struct index_state
*istate
, unsigned int flags
,
1579 const struct pathspec
*pathspec
,
1580 char *seen
, const char *header_msg
)
1584 int really
= (flags
& REFRESH_REALLY
) != 0;
1585 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
1586 int quiet
= (flags
& REFRESH_QUIET
) != 0;
1587 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
1588 int ignore_submodules
= (flags
& REFRESH_IGNORE_SUBMODULES
) != 0;
1589 int ignore_skip_worktree
= (flags
& REFRESH_IGNORE_SKIP_WORKTREE
) != 0;
1591 int in_porcelain
= (flags
& REFRESH_IN_PORCELAIN
);
1592 unsigned int options
= (CE_MATCH_REFRESH
|
1593 (really
? CE_MATCH_IGNORE_VALID
: 0) |
1594 (not_new
? CE_MATCH_IGNORE_MISSING
: 0));
1595 const char *modified_fmt
;
1596 const char *deleted_fmt
;
1597 const char *typechange_fmt
;
1598 const char *added_fmt
;
1599 const char *unmerged_fmt
;
1600 struct progress
*progress
= NULL
;
1601 int t2_sum_lstat
= 0;
1602 int t2_sum_scan
= 0;
1604 if (flags
& REFRESH_PROGRESS
&& isatty(2))
1605 progress
= start_delayed_progress(_("Refresh index"),
1608 trace_performance_enter();
1609 modified_fmt
= in_porcelain
? "M\t%s\n" : "%s: needs update\n";
1610 deleted_fmt
= in_porcelain
? "D\t%s\n" : "%s: needs update\n";
1611 typechange_fmt
= in_porcelain
? "T\t%s\n" : "%s: needs update\n";
1612 added_fmt
= in_porcelain
? "A\t%s\n" : "%s: needs update\n";
1613 unmerged_fmt
= in_porcelain
? "U\t%s\n" : "%s: needs merge\n";
1615 * Use the multi-threaded preload_index() to refresh most of the
1616 * cache entries quickly then in the single threaded loop below,
1617 * we only have to do the special cases that are left.
1619 preload_index(istate
, pathspec
, 0);
1620 trace2_region_enter("index", "refresh", NULL
);
1622 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1623 struct cache_entry
*ce
, *new_entry
;
1624 int cache_errno
= 0;
1627 int t2_did_lstat
= 0;
1628 int t2_did_scan
= 0;
1630 ce
= istate
->cache
[i
];
1631 if (ignore_submodules
&& S_ISGITLINK(ce
->ce_mode
))
1633 if (ignore_skip_worktree
&& ce_skip_worktree(ce
))
1637 * If this entry is a sparse directory, then there isn't
1638 * any stat() information to update. Ignore the entry.
1640 if (S_ISSPARSEDIR(ce
->ce_mode
))
1643 if (pathspec
&& !ce_path_match(istate
, ce
, pathspec
, seen
))
1647 while ((i
< istate
->cache_nr
) &&
1648 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
1654 show_file(unmerged_fmt
, ce
->name
, in_porcelain
,
1655 &first
, header_msg
);
1663 new_entry
= refresh_cache_ent(istate
, ce
, options
,
1664 &cache_errno
, &changed
,
1665 &t2_did_lstat
, &t2_did_scan
);
1666 t2_sum_lstat
+= t2_did_lstat
;
1667 t2_sum_scan
+= t2_did_scan
;
1668 if (new_entry
== ce
)
1670 display_progress(progress
, i
);
1674 if (really
&& cache_errno
== EINVAL
) {
1675 /* If we are doing --really-refresh that
1676 * means the index is not valid anymore.
1678 ce
->ce_flags
&= ~CE_VALID
;
1679 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
1680 mark_fsmonitor_invalid(istate
, ce
);
1681 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
1686 if (cache_errno
== ENOENT
)
1688 else if (ce_intent_to_add(ce
))
1689 fmt
= added_fmt
; /* must be before other checks */
1690 else if (changed
& TYPE_CHANGED
)
1691 fmt
= typechange_fmt
;
1695 ce
->name
, in_porcelain
, &first
, header_msg
);
1700 replace_index_entry(istate
, i
, new_entry
);
1702 trace2_data_intmax("index", NULL
, "refresh/sum_lstat", t2_sum_lstat
);
1703 trace2_data_intmax("index", NULL
, "refresh/sum_scan", t2_sum_scan
);
1704 trace2_region_leave("index", "refresh", NULL
);
1705 display_progress(progress
, istate
->cache_nr
);
1706 stop_progress(&progress
);
1707 trace_performance_leave("refresh index");
1711 struct cache_entry
*refresh_cache_entry(struct index_state
*istate
,
1712 struct cache_entry
*ce
,
1713 unsigned int options
)
1715 return refresh_cache_ent(istate
, ce
, options
, NULL
, NULL
, NULL
, NULL
);
1719 /*****************************************************************
1721 *****************************************************************/
1723 #define INDEX_FORMAT_DEFAULT 3
1725 static unsigned int get_index_format_default(struct repository
*r
)
1727 char *envversion
= getenv("GIT_INDEX_VERSION");
1729 unsigned int version
= INDEX_FORMAT_DEFAULT
;
1732 prepare_repo_settings(r
);
1734 if (r
->settings
.index_version
>= 0)
1735 version
= r
->settings
.index_version
;
1736 if (version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1737 warning(_("index.version set, but the value is invalid.\n"
1738 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1739 return INDEX_FORMAT_DEFAULT
;
1744 version
= strtoul(envversion
, &endp
, 10);
1746 version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1747 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1748 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1749 version
= INDEX_FORMAT_DEFAULT
;
1755 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1756 * Again - this is just a (very strong in practice) heuristic that
1757 * the inode hasn't changed.
1759 * We save the fields in big-endian order to allow using the
1760 * index file over NFS transparently.
1762 struct ondisk_cache_entry
{
1763 struct cache_time ctime
;
1764 struct cache_time mtime
;
1772 * unsigned char hash[hashsz];
1774 * if (flags & CE_EXTENDED)
1777 unsigned char data
[GIT_MAX_RAWSZ
+ 2 * sizeof(uint16_t)];
1778 char name
[FLEX_ARRAY
];
1781 /* These are only used for v3 or lower */
1782 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1783 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,data) + (len) + 8) & ~7)
1784 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1785 #define ondisk_data_size(flags, len) (the_hash_algo->rawsz + \
1786 ((flags & CE_EXTENDED) ? 2 : 1) * sizeof(uint16_t) + len)
1787 #define ondisk_data_size_max(len) (ondisk_data_size(CE_EXTENDED, len))
1788 #define ondisk_ce_size(ce) (ondisk_cache_entry_size(ondisk_data_size((ce)->ce_flags, ce_namelen(ce))))
1790 /* Allow fsck to force verification of the index checksum. */
1791 int verify_index_checksum
;
1793 /* Allow fsck to force verification of the cache entry order. */
1794 int verify_ce_order
;
1796 static int verify_hdr(const struct cache_header
*hdr
, unsigned long size
)
1799 unsigned char hash
[GIT_MAX_RAWSZ
];
1802 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
1803 return error(_("bad signature 0x%08x"), hdr
->hdr_signature
);
1804 hdr_version
= ntohl(hdr
->hdr_version
);
1805 if (hdr_version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< hdr_version
)
1806 return error(_("bad index version %d"), hdr_version
);
1808 if (!verify_index_checksum
)
1811 the_hash_algo
->init_fn(&c
);
1812 the_hash_algo
->update_fn(&c
, hdr
, size
- the_hash_algo
->rawsz
);
1813 the_hash_algo
->final_fn(hash
, &c
);
1814 if (!hasheq(hash
, (unsigned char *)hdr
+ size
- the_hash_algo
->rawsz
))
1815 return error(_("bad index file sha1 signature"));
1819 static int read_index_extension(struct index_state
*istate
,
1820 const char *ext
, const char *data
, unsigned long sz
)
1822 switch (CACHE_EXT(ext
)) {
1823 case CACHE_EXT_TREE
:
1824 istate
->cache_tree
= cache_tree_read(data
, sz
);
1826 case CACHE_EXT_RESOLVE_UNDO
:
1827 istate
->resolve_undo
= resolve_undo_read(data
, sz
);
1829 case CACHE_EXT_LINK
:
1830 if (read_link_extension(istate
, data
, sz
))
1833 case CACHE_EXT_UNTRACKED
:
1834 istate
->untracked
= read_untracked_extension(data
, sz
);
1836 case CACHE_EXT_FSMONITOR
:
1837 read_fsmonitor_extension(istate
, data
, sz
);
1839 case CACHE_EXT_ENDOFINDEXENTRIES
:
1840 case CACHE_EXT_INDEXENTRYOFFSETTABLE
:
1841 /* already handled in do_read_index() */
1843 case CACHE_EXT_SPARSE_DIRECTORIES
:
1844 /* no content, only an indicator */
1845 istate
->sparse_index
= 1;
1848 if (*ext
< 'A' || 'Z' < *ext
)
1849 return error(_("index uses %.4s extension, which we do not understand"),
1851 fprintf_ln(stderr
, _("ignoring %.4s extension"), ext
);
1857 static struct cache_entry
*create_from_disk(struct mem_pool
*ce_mem_pool
,
1858 unsigned int version
,
1859 struct ondisk_cache_entry
*ondisk
,
1860 unsigned long *ent_size
,
1861 const struct cache_entry
*previous_ce
)
1863 struct cache_entry
*ce
;
1866 const unsigned hashsz
= the_hash_algo
->rawsz
;
1867 const uint16_t *flagsp
= (const uint16_t *)(ondisk
->data
+ hashsz
);
1869 size_t copy_len
= 0;
1871 * Adjacent cache entries tend to share the leading paths, so it makes
1872 * sense to only store the differences in later entries. In the v4
1873 * on-disk format of the index, each on-disk cache entry stores the
1874 * number of bytes to be stripped from the end of the previous name,
1875 * and the bytes to append to the result, to come up with its name.
1877 int expand_name_field
= version
== 4;
1879 /* On-disk flags are just 16 bits */
1880 flags
= get_be16(flagsp
);
1881 len
= flags
& CE_NAMEMASK
;
1883 if (flags
& CE_EXTENDED
) {
1885 extended_flags
= get_be16(flagsp
+ 1) << 16;
1886 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1887 if (extended_flags
& ~CE_EXTENDED_FLAGS
)
1888 die(_("unknown index entry format 0x%08x"), extended_flags
);
1889 flags
|= extended_flags
;
1890 name
= (const char *)(flagsp
+ 2);
1893 name
= (const char *)(flagsp
+ 1);
1895 if (expand_name_field
) {
1896 const unsigned char *cp
= (const unsigned char *)name
;
1897 size_t strip_len
, previous_len
;
1899 /* If we're at the beginning of a block, ignore the previous name */
1900 strip_len
= decode_varint(&cp
);
1902 previous_len
= previous_ce
->ce_namelen
;
1903 if (previous_len
< strip_len
)
1904 die(_("malformed name field in the index, near path '%s'"),
1906 copy_len
= previous_len
- strip_len
;
1908 name
= (const char *)cp
;
1911 if (len
== CE_NAMEMASK
) {
1913 if (expand_name_field
)
1917 ce
= mem_pool__ce_alloc(ce_mem_pool
, len
);
1919 ce
->ce_stat_data
.sd_ctime
.sec
= get_be32(&ondisk
->ctime
.sec
);
1920 ce
->ce_stat_data
.sd_mtime
.sec
= get_be32(&ondisk
->mtime
.sec
);
1921 ce
->ce_stat_data
.sd_ctime
.nsec
= get_be32(&ondisk
->ctime
.nsec
);
1922 ce
->ce_stat_data
.sd_mtime
.nsec
= get_be32(&ondisk
->mtime
.nsec
);
1923 ce
->ce_stat_data
.sd_dev
= get_be32(&ondisk
->dev
);
1924 ce
->ce_stat_data
.sd_ino
= get_be32(&ondisk
->ino
);
1925 ce
->ce_mode
= get_be32(&ondisk
->mode
);
1926 ce
->ce_stat_data
.sd_uid
= get_be32(&ondisk
->uid
);
1927 ce
->ce_stat_data
.sd_gid
= get_be32(&ondisk
->gid
);
1928 ce
->ce_stat_data
.sd_size
= get_be32(&ondisk
->size
);
1929 ce
->ce_flags
= flags
& ~CE_NAMEMASK
;
1930 ce
->ce_namelen
= len
;
1932 oidread(&ce
->oid
, ondisk
->data
);
1933 memcpy(ce
->name
, name
, len
);
1934 ce
->name
[len
] = '\0';
1936 if (expand_name_field
) {
1938 memcpy(ce
->name
, previous_ce
->name
, copy_len
);
1939 memcpy(ce
->name
+ copy_len
, name
, len
+ 1 - copy_len
);
1940 *ent_size
= (name
- ((char *)ondisk
)) + len
+ 1 - copy_len
;
1942 memcpy(ce
->name
, name
, len
+ 1);
1943 *ent_size
= ondisk_ce_size(ce
);
1948 static void check_ce_order(struct index_state
*istate
)
1952 if (!verify_ce_order
)
1955 for (i
= 1; i
< istate
->cache_nr
; i
++) {
1956 struct cache_entry
*ce
= istate
->cache
[i
- 1];
1957 struct cache_entry
*next_ce
= istate
->cache
[i
];
1958 int name_compare
= strcmp(ce
->name
, next_ce
->name
);
1960 if (0 < name_compare
)
1961 die(_("unordered stage entries in index"));
1962 if (!name_compare
) {
1964 die(_("multiple stage entries for merged file '%s'"),
1966 if (ce_stage(ce
) > ce_stage(next_ce
))
1967 die(_("unordered stage entries for '%s'"),
1973 static void tweak_untracked_cache(struct index_state
*istate
)
1975 struct repository
*r
= the_repository
;
1977 prepare_repo_settings(r
);
1979 switch (r
->settings
.core_untracked_cache
) {
1980 case UNTRACKED_CACHE_REMOVE
:
1981 remove_untracked_cache(istate
);
1983 case UNTRACKED_CACHE_WRITE
:
1984 add_untracked_cache(istate
);
1986 case UNTRACKED_CACHE_KEEP
:
1988 * Either an explicit "core.untrackedCache=keep", the
1989 * default if "core.untrackedCache" isn't configured,
1990 * or a fallback on an unknown "core.untrackedCache"
1997 static void tweak_split_index(struct index_state
*istate
)
1999 switch (git_config_get_split_index()) {
2000 case -1: /* unset: do nothing */
2003 remove_split_index(istate
);
2006 add_split_index(istate
);
2008 default: /* unknown value: do nothing */
2013 static void post_read_index_from(struct index_state
*istate
)
2015 check_ce_order(istate
);
2016 tweak_untracked_cache(istate
);
2017 tweak_split_index(istate
);
2018 tweak_fsmonitor(istate
);
2021 static size_t estimate_cache_size_from_compressed(unsigned int entries
)
2023 return entries
* (sizeof(struct cache_entry
) + CACHE_ENTRY_PATH_LENGTH
);
2026 static size_t estimate_cache_size(size_t ondisk_size
, unsigned int entries
)
2028 long per_entry
= sizeof(struct cache_entry
) - sizeof(struct ondisk_cache_entry
);
2031 * Account for potential alignment differences.
2033 per_entry
+= align_padding_size(per_entry
, 0);
2034 return ondisk_size
+ entries
* per_entry
;
2037 struct index_entry_offset
2039 /* starting byte offset into index file, count of index entries in this block */
2043 struct index_entry_offset_table
2046 struct index_entry_offset entries
[FLEX_ARRAY
];
2049 static struct index_entry_offset_table
*read_ieot_extension(const char *mmap
, size_t mmap_size
, size_t offset
);
2050 static void write_ieot_extension(struct strbuf
*sb
, struct index_entry_offset_table
*ieot
);
2052 static size_t read_eoie_extension(const char *mmap
, size_t mmap_size
);
2053 static void write_eoie_extension(struct strbuf
*sb
, git_hash_ctx
*eoie_context
, size_t offset
);
2055 struct load_index_extensions
2058 struct index_state
*istate
;
2061 unsigned long src_offset
;
2064 static void *load_index_extensions(void *_data
)
2066 struct load_index_extensions
*p
= _data
;
2067 unsigned long src_offset
= p
->src_offset
;
2069 while (src_offset
<= p
->mmap_size
- the_hash_algo
->rawsz
- 8) {
2070 /* After an array of active_nr index entries,
2071 * there can be arbitrary number of extended
2072 * sections, each of which is prefixed with
2073 * extension name (4-byte) and section length
2074 * in 4-byte network byte order.
2076 uint32_t extsize
= get_be32(p
->mmap
+ src_offset
+ 4);
2077 if (read_index_extension(p
->istate
,
2078 p
->mmap
+ src_offset
,
2079 p
->mmap
+ src_offset
+ 8,
2081 munmap((void *)p
->mmap
, p
->mmap_size
);
2082 die(_("index file corrupt"));
2085 src_offset
+= extsize
;
2092 * A helper function that will load the specified range of cache entries
2093 * from the memory mapped file and add them to the given index.
2095 static unsigned long load_cache_entry_block(struct index_state
*istate
,
2096 struct mem_pool
*ce_mem_pool
, int offset
, int nr
, const char *mmap
,
2097 unsigned long start_offset
, const struct cache_entry
*previous_ce
)
2100 unsigned long src_offset
= start_offset
;
2102 for (i
= offset
; i
< offset
+ nr
; i
++) {
2103 struct ondisk_cache_entry
*disk_ce
;
2104 struct cache_entry
*ce
;
2105 unsigned long consumed
;
2107 disk_ce
= (struct ondisk_cache_entry
*)(mmap
+ src_offset
);
2108 ce
= create_from_disk(ce_mem_pool
, istate
->version
, disk_ce
, &consumed
, previous_ce
);
2109 set_index_entry(istate
, i
, ce
);
2111 src_offset
+= consumed
;
2114 return src_offset
- start_offset
;
2117 static unsigned long load_all_cache_entries(struct index_state
*istate
,
2118 const char *mmap
, size_t mmap_size
, unsigned long src_offset
)
2120 unsigned long consumed
;
2122 istate
->ce_mem_pool
= xmalloc(sizeof(*istate
->ce_mem_pool
));
2123 if (istate
->version
== 4) {
2124 mem_pool_init(istate
->ce_mem_pool
,
2125 estimate_cache_size_from_compressed(istate
->cache_nr
));
2127 mem_pool_init(istate
->ce_mem_pool
,
2128 estimate_cache_size(mmap_size
, istate
->cache_nr
));
2131 consumed
= load_cache_entry_block(istate
, istate
->ce_mem_pool
,
2132 0, istate
->cache_nr
, mmap
, src_offset
, NULL
);
2137 * Mostly randomly chosen maximum thread counts: we
2138 * cap the parallelism to online_cpus() threads, and we want
2139 * to have at least 10000 cache entries per thread for it to
2140 * be worth starting a thread.
2143 #define THREAD_COST (10000)
2145 struct load_cache_entries_thread_data
2148 struct index_state
*istate
;
2149 struct mem_pool
*ce_mem_pool
;
2152 struct index_entry_offset_table
*ieot
;
2153 int ieot_start
; /* starting index into the ieot array */
2154 int ieot_blocks
; /* count of ieot entries to process */
2155 unsigned long consumed
; /* return # of bytes in index file processed */
2159 * A thread proc to run the load_cache_entries() computation
2160 * across multiple background threads.
2162 static void *load_cache_entries_thread(void *_data
)
2164 struct load_cache_entries_thread_data
*p
= _data
;
2167 /* iterate across all ieot blocks assigned to this thread */
2168 for (i
= p
->ieot_start
; i
< p
->ieot_start
+ p
->ieot_blocks
; i
++) {
2169 p
->consumed
+= load_cache_entry_block(p
->istate
, p
->ce_mem_pool
,
2170 p
->offset
, p
->ieot
->entries
[i
].nr
, p
->mmap
, p
->ieot
->entries
[i
].offset
, NULL
);
2171 p
->offset
+= p
->ieot
->entries
[i
].nr
;
2176 static unsigned long load_cache_entries_threaded(struct index_state
*istate
, const char *mmap
, size_t mmap_size
,
2177 int nr_threads
, struct index_entry_offset_table
*ieot
)
2179 int i
, offset
, ieot_blocks
, ieot_start
, err
;
2180 struct load_cache_entries_thread_data
*data
;
2181 unsigned long consumed
= 0;
2183 /* a little sanity checking */
2184 if (istate
->name_hash_initialized
)
2185 BUG("the name hash isn't thread safe");
2187 istate
->ce_mem_pool
= xmalloc(sizeof(*istate
->ce_mem_pool
));
2188 mem_pool_init(istate
->ce_mem_pool
, 0);
2190 /* ensure we have no more threads than we have blocks to process */
2191 if (nr_threads
> ieot
->nr
)
2192 nr_threads
= ieot
->nr
;
2193 CALLOC_ARRAY(data
, nr_threads
);
2195 offset
= ieot_start
= 0;
2196 ieot_blocks
= DIV_ROUND_UP(ieot
->nr
, nr_threads
);
2197 for (i
= 0; i
< nr_threads
; i
++) {
2198 struct load_cache_entries_thread_data
*p
= &data
[i
];
2201 if (ieot_start
+ ieot_blocks
> ieot
->nr
)
2202 ieot_blocks
= ieot
->nr
- ieot_start
;
2208 p
->ieot_start
= ieot_start
;
2209 p
->ieot_blocks
= ieot_blocks
;
2211 /* create a mem_pool for each thread */
2213 for (j
= p
->ieot_start
; j
< p
->ieot_start
+ p
->ieot_blocks
; j
++)
2214 nr
+= p
->ieot
->entries
[j
].nr
;
2215 p
->ce_mem_pool
= xmalloc(sizeof(*istate
->ce_mem_pool
));
2216 if (istate
->version
== 4) {
2217 mem_pool_init(p
->ce_mem_pool
,
2218 estimate_cache_size_from_compressed(nr
));
2220 mem_pool_init(p
->ce_mem_pool
,
2221 estimate_cache_size(mmap_size
, nr
));
2224 err
= pthread_create(&p
->pthread
, NULL
, load_cache_entries_thread
, p
);
2226 die(_("unable to create load_cache_entries thread: %s"), strerror(err
));
2228 /* increment by the number of cache entries in the ieot block being processed */
2229 for (j
= 0; j
< ieot_blocks
; j
++)
2230 offset
+= ieot
->entries
[ieot_start
+ j
].nr
;
2231 ieot_start
+= ieot_blocks
;
2234 for (i
= 0; i
< nr_threads
; i
++) {
2235 struct load_cache_entries_thread_data
*p
= &data
[i
];
2237 err
= pthread_join(p
->pthread
, NULL
);
2239 die(_("unable to join load_cache_entries thread: %s"), strerror(err
));
2240 mem_pool_combine(istate
->ce_mem_pool
, p
->ce_mem_pool
);
2241 consumed
+= p
->consumed
;
2249 /* remember to discard_cache() before reading a different cache! */
2250 int do_read_index(struct index_state
*istate
, const char *path
, int must_exist
)
2254 unsigned long src_offset
;
2255 const struct cache_header
*hdr
;
2258 struct load_index_extensions p
;
2259 size_t extension_offset
= 0;
2260 int nr_threads
, cpus
;
2261 struct index_entry_offset_table
*ieot
= NULL
;
2263 if (istate
->initialized
)
2264 return istate
->cache_nr
;
2266 istate
->timestamp
.sec
= 0;
2267 istate
->timestamp
.nsec
= 0;
2268 fd
= open(path
, O_RDONLY
);
2270 if (!must_exist
&& errno
== ENOENT
)
2272 die_errno(_("%s: index file open failed"), path
);
2276 die_errno(_("%s: cannot stat the open index"), path
);
2278 mmap_size
= xsize_t(st
.st_size
);
2279 if (mmap_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2280 die(_("%s: index file smaller than expected"), path
);
2282 mmap
= xmmap_gently(NULL
, mmap_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2283 if (mmap
== MAP_FAILED
)
2284 die_errno(_("%s: unable to map index file%s"), path
,
2288 hdr
= (const struct cache_header
*)mmap
;
2289 if (verify_hdr(hdr
, mmap_size
) < 0)
2292 oidread(&istate
->oid
, (const unsigned char *)hdr
+ mmap_size
- the_hash_algo
->rawsz
);
2293 istate
->version
= ntohl(hdr
->hdr_version
);
2294 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
2295 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
2296 CALLOC_ARRAY(istate
->cache
, istate
->cache_alloc
);
2297 istate
->initialized
= 1;
2301 p
.mmap_size
= mmap_size
;
2303 src_offset
= sizeof(*hdr
);
2305 if (git_config_get_index_threads(&nr_threads
))
2308 /* TODO: does creating more threads than cores help? */
2310 nr_threads
= istate
->cache_nr
/ THREAD_COST
;
2311 cpus
= online_cpus();
2312 if (nr_threads
> cpus
)
2319 if (nr_threads
> 1) {
2320 extension_offset
= read_eoie_extension(mmap
, mmap_size
);
2321 if (extension_offset
) {
2324 p
.src_offset
= extension_offset
;
2325 err
= pthread_create(&p
.pthread
, NULL
, load_index_extensions
, &p
);
2327 die(_("unable to create load_index_extensions thread: %s"), strerror(err
));
2334 * Locate and read the index entry offset table so that we can use it
2335 * to multi-thread the reading of the cache entries.
2337 if (extension_offset
&& nr_threads
> 1)
2338 ieot
= read_ieot_extension(mmap
, mmap_size
, extension_offset
);
2341 src_offset
+= load_cache_entries_threaded(istate
, mmap
, mmap_size
, nr_threads
, ieot
);
2344 src_offset
+= load_all_cache_entries(istate
, mmap
, mmap_size
, src_offset
);
2347 istate
->timestamp
.sec
= st
.st_mtime
;
2348 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
2350 /* if we created a thread, join it otherwise load the extensions on the primary thread */
2351 if (extension_offset
) {
2352 int ret
= pthread_join(p
.pthread
, NULL
);
2354 die(_("unable to join load_index_extensions thread: %s"), strerror(ret
));
2356 p
.src_offset
= src_offset
;
2357 load_index_extensions(&p
);
2359 munmap((void *)mmap
, mmap_size
);
2362 * TODO trace2: replace "the_repository" with the actual repo instance
2363 * that is associated with the given "istate".
2365 trace2_data_intmax("index", the_repository
, "read/version",
2367 trace2_data_intmax("index", the_repository
, "read/cache_nr",
2371 istate
->repo
= the_repository
;
2374 * If the command explicitly requires a full index, force it
2375 * to be full. Otherwise, correct the sparsity based on repository
2376 * settings and other properties of the index (if necessary).
2378 prepare_repo_settings(istate
->repo
);
2379 if (istate
->repo
->settings
.command_requires_full_index
)
2380 ensure_full_index(istate
);
2382 ensure_correct_sparsity(istate
);
2384 return istate
->cache_nr
;
2387 munmap((void *)mmap
, mmap_size
);
2388 die(_("index file corrupt"));
2392 * Signal that the shared index is used by updating its mtime.
2394 * This way, shared index can be removed if they have not been used
2397 static void freshen_shared_index(const char *shared_index
, int warn
)
2399 if (!check_and_freshen_file(shared_index
, 1) && warn
)
2400 warning(_("could not freshen shared index '%s'"), shared_index
);
2403 int read_index_from(struct index_state
*istate
, const char *path
,
2406 struct split_index
*split_index
;
2411 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2412 if (istate
->initialized
)
2413 return istate
->cache_nr
;
2416 * TODO trace2: replace "the_repository" with the actual repo instance
2417 * that is associated with the given "istate".
2419 trace2_region_enter_printf("index", "do_read_index", the_repository
,
2421 trace_performance_enter();
2422 ret
= do_read_index(istate
, path
, 0);
2423 trace_performance_leave("read cache %s", path
);
2424 trace2_region_leave_printf("index", "do_read_index", the_repository
,
2427 split_index
= istate
->split_index
;
2428 if (!split_index
|| is_null_oid(&split_index
->base_oid
)) {
2429 post_read_index_from(istate
);
2433 trace_performance_enter();
2434 if (split_index
->base
)
2435 discard_index(split_index
->base
);
2437 CALLOC_ARRAY(split_index
->base
, 1);
2439 base_oid_hex
= oid_to_hex(&split_index
->base_oid
);
2440 base_path
= xstrfmt("%s/sharedindex.%s", gitdir
, base_oid_hex
);
2441 trace2_region_enter_printf("index", "shared/do_read_index",
2442 the_repository
, "%s", base_path
);
2443 ret
= do_read_index(split_index
->base
, base_path
, 0);
2444 trace2_region_leave_printf("index", "shared/do_read_index",
2445 the_repository
, "%s", base_path
);
2447 char *path_copy
= xstrdup(path
);
2448 const char *base_path2
= xstrfmt("%s/sharedindex.%s",
2452 trace2_region_enter_printf("index", "shared/do_read_index",
2453 the_repository
, "%s", base_path2
);
2454 ret
= do_read_index(split_index
->base
, base_path2
, 1);
2455 trace2_region_leave_printf("index", "shared/do_read_index",
2456 the_repository
, "%s", base_path2
);
2458 if (!oideq(&split_index
->base_oid
, &split_index
->base
->oid
))
2459 die(_("broken index, expect %s in %s, got %s"),
2460 base_oid_hex
, base_path
,
2461 oid_to_hex(&split_index
->base
->oid
));
2463 freshen_shared_index(base_path
, 0);
2464 merge_base_index(istate
);
2465 post_read_index_from(istate
);
2466 trace_performance_leave("read cache %s", base_path
);
2471 int is_index_unborn(struct index_state
*istate
)
2473 return (!istate
->cache_nr
&& !istate
->timestamp
.sec
);
2476 int discard_index(struct index_state
*istate
)
2479 * Cache entries in istate->cache[] should have been allocated
2480 * from the memory pool associated with this index, or from an
2481 * associated split_index. There is no need to free individual
2482 * cache entries. validate_cache_entries can detect when this
2483 * assertion does not hold.
2485 validate_cache_entries(istate
);
2487 resolve_undo_clear_index(istate
);
2488 istate
->cache_nr
= 0;
2489 istate
->cache_changed
= 0;
2490 istate
->timestamp
.sec
= 0;
2491 istate
->timestamp
.nsec
= 0;
2492 free_name_hash(istate
);
2493 cache_tree_free(&(istate
->cache_tree
));
2494 istate
->initialized
= 0;
2495 istate
->fsmonitor_has_run_once
= 0;
2496 FREE_AND_NULL(istate
->fsmonitor_last_update
);
2497 FREE_AND_NULL(istate
->cache
);
2498 istate
->cache_alloc
= 0;
2499 discard_split_index(istate
);
2500 free_untracked_cache(istate
->untracked
);
2501 istate
->untracked
= NULL
;
2503 if (istate
->ce_mem_pool
) {
2504 mem_pool_discard(istate
->ce_mem_pool
, should_validate_cache_entries());
2505 FREE_AND_NULL(istate
->ce_mem_pool
);
2512 * Validate the cache entries of this index.
2513 * All cache entries associated with this index
2514 * should have been allocated by the memory pool
2515 * associated with this index, or by a referenced
2518 void validate_cache_entries(const struct index_state
*istate
)
2522 if (!should_validate_cache_entries() ||!istate
|| !istate
->initialized
)
2525 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2527 BUG("cache entry is not allocated from expected memory pool");
2528 } else if (!istate
->ce_mem_pool
||
2529 !mem_pool_contains(istate
->ce_mem_pool
, istate
->cache
[i
])) {
2530 if (!istate
->split_index
||
2531 !istate
->split_index
->base
||
2532 !istate
->split_index
->base
->ce_mem_pool
||
2533 !mem_pool_contains(istate
->split_index
->base
->ce_mem_pool
, istate
->cache
[i
])) {
2534 BUG("cache entry is not allocated from expected memory pool");
2539 if (istate
->split_index
)
2540 validate_cache_entries(istate
->split_index
->base
);
2543 int unmerged_index(const struct index_state
*istate
)
2546 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2547 if (ce_stage(istate
->cache
[i
]))
2553 int repo_index_has_changes(struct repository
*repo
,
2557 struct index_state
*istate
= repo
->index
;
2558 struct object_id cmp
;
2562 cmp
= tree
->object
.oid
;
2563 if (tree
|| !get_oid_tree("HEAD", &cmp
)) {
2564 struct diff_options opt
;
2566 repo_diff_setup(repo
, &opt
);
2567 opt
.flags
.exit_with_status
= 1;
2569 opt
.flags
.quick
= 1;
2570 diff_setup_done(&opt
);
2571 do_diff_cache(&cmp
, &opt
);
2573 for (i
= 0; sb
&& i
< diff_queued_diff
.nr
; i
++) {
2575 strbuf_addch(sb
, ' ');
2576 strbuf_addstr(sb
, diff_queued_diff
.queue
[i
]->two
->path
);
2579 return opt
.flags
.has_changes
!= 0;
2581 /* TODO: audit for interaction with sparse-index. */
2582 ensure_full_index(istate
);
2583 for (i
= 0; sb
&& i
< istate
->cache_nr
; i
++) {
2585 strbuf_addch(sb
, ' ');
2586 strbuf_addstr(sb
, istate
->cache
[i
]->name
);
2588 return !!istate
->cache_nr
;
2592 static int write_index_ext_header(struct hashfile
*f
,
2593 git_hash_ctx
*eoie_f
,
2597 hashwrite_be32(f
, ext
);
2598 hashwrite_be32(f
, sz
);
2603 the_hash_algo
->update_fn(eoie_f
, &ext
, sizeof(ext
));
2604 the_hash_algo
->update_fn(eoie_f
, &sz
, sizeof(sz
));
2609 static void ce_smudge_racily_clean_entry(struct index_state
*istate
,
2610 struct cache_entry
*ce
)
2613 * The only thing we care about in this function is to smudge the
2614 * falsely clean entry due to touch-update-touch race, so we leave
2615 * everything else as they are. We are called for entries whose
2616 * ce_stat_data.sd_mtime match the index file mtime.
2618 * Note that this actually does not do much for gitlinks, for
2619 * which ce_match_stat_basic() always goes to the actual
2620 * contents. The caller checks with is_racy_timestamp() which
2621 * always says "no" for gitlinks, so we are not called for them ;-)
2625 if (lstat(ce
->name
, &st
) < 0)
2627 if (ce_match_stat_basic(ce
, &st
))
2629 if (ce_modified_check_fs(istate
, ce
, &st
)) {
2630 /* This is "racily clean"; smudge it. Note that this
2631 * is a tricky code. At first glance, it may appear
2632 * that it can break with this sequence:
2634 * $ echo xyzzy >frotz
2635 * $ git-update-index --add frotz
2638 * $ echo filfre >nitfol
2639 * $ git-update-index --add nitfol
2641 * but it does not. When the second update-index runs,
2642 * it notices that the entry "frotz" has the same timestamp
2643 * as index, and if we were to smudge it by resetting its
2644 * size to zero here, then the object name recorded
2645 * in index is the 6-byte file but the cached stat information
2646 * becomes zero --- which would then match what we would
2647 * obtain from the filesystem next time we stat("frotz").
2649 * However, the second update-index, before calling
2650 * this function, notices that the cached size is 6
2651 * bytes and what is on the filesystem is an empty
2652 * file, and never calls us, so the cached size information
2653 * for "frotz" stays 6 which does not match the filesystem.
2655 ce
->ce_stat_data
.sd_size
= 0;
2659 /* Copy miscellaneous fields but not the name */
2660 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry
*ondisk
,
2661 struct cache_entry
*ce
)
2664 const unsigned hashsz
= the_hash_algo
->rawsz
;
2665 uint16_t *flagsp
= (uint16_t *)(ondisk
->data
+ hashsz
);
2667 ondisk
->ctime
.sec
= htonl(ce
->ce_stat_data
.sd_ctime
.sec
);
2668 ondisk
->mtime
.sec
= htonl(ce
->ce_stat_data
.sd_mtime
.sec
);
2669 ondisk
->ctime
.nsec
= htonl(ce
->ce_stat_data
.sd_ctime
.nsec
);
2670 ondisk
->mtime
.nsec
= htonl(ce
->ce_stat_data
.sd_mtime
.nsec
);
2671 ondisk
->dev
= htonl(ce
->ce_stat_data
.sd_dev
);
2672 ondisk
->ino
= htonl(ce
->ce_stat_data
.sd_ino
);
2673 ondisk
->mode
= htonl(ce
->ce_mode
);
2674 ondisk
->uid
= htonl(ce
->ce_stat_data
.sd_uid
);
2675 ondisk
->gid
= htonl(ce
->ce_stat_data
.sd_gid
);
2676 ondisk
->size
= htonl(ce
->ce_stat_data
.sd_size
);
2677 hashcpy(ondisk
->data
, ce
->oid
.hash
);
2679 flags
= ce
->ce_flags
& ~CE_NAMEMASK
;
2680 flags
|= (ce_namelen(ce
) >= CE_NAMEMASK
? CE_NAMEMASK
: ce_namelen(ce
));
2681 flagsp
[0] = htons(flags
);
2682 if (ce
->ce_flags
& CE_EXTENDED
) {
2683 flagsp
[1] = htons((ce
->ce_flags
& CE_EXTENDED_FLAGS
) >> 16);
2687 static int ce_write_entry(struct hashfile
*f
, struct cache_entry
*ce
,
2688 struct strbuf
*previous_name
, struct ondisk_cache_entry
*ondisk
)
2691 unsigned int saved_namelen
;
2692 int stripped_name
= 0;
2693 static unsigned char padding
[8] = { 0x00 };
2695 if (ce
->ce_flags
& CE_STRIP_NAME
) {
2696 saved_namelen
= ce_namelen(ce
);
2701 size
= offsetof(struct ondisk_cache_entry
,data
) + ondisk_data_size(ce
->ce_flags
, 0);
2703 if (!previous_name
) {
2704 int len
= ce_namelen(ce
);
2705 copy_cache_entry_to_ondisk(ondisk
, ce
);
2706 hashwrite(f
, ondisk
, size
);
2707 hashwrite(f
, ce
->name
, len
);
2708 hashwrite(f
, padding
, align_padding_size(size
, len
));
2710 int common
, to_remove
, prefix_size
;
2711 unsigned char to_remove_vi
[16];
2713 (ce
->name
[common
] &&
2714 common
< previous_name
->len
&&
2715 ce
->name
[common
] == previous_name
->buf
[common
]);
2717 ; /* still matching */
2718 to_remove
= previous_name
->len
- common
;
2719 prefix_size
= encode_varint(to_remove
, to_remove_vi
);
2721 copy_cache_entry_to_ondisk(ondisk
, ce
);
2722 hashwrite(f
, ondisk
, size
);
2723 hashwrite(f
, to_remove_vi
, prefix_size
);
2724 hashwrite(f
, ce
->name
+ common
, ce_namelen(ce
) - common
);
2725 hashwrite(f
, padding
, 1);
2727 strbuf_splice(previous_name
, common
, to_remove
,
2728 ce
->name
+ common
, ce_namelen(ce
) - common
);
2730 if (stripped_name
) {
2731 ce
->ce_namelen
= saved_namelen
;
2732 ce
->ce_flags
&= ~CE_STRIP_NAME
;
2739 * This function verifies if index_state has the correct sha1 of the
2740 * index file. Don't die if we have any other failure, just return 0.
2742 static int verify_index_from(const struct index_state
*istate
, const char *path
)
2747 unsigned char hash
[GIT_MAX_RAWSZ
];
2749 if (!istate
->initialized
)
2752 fd
= open(path
, O_RDONLY
);
2759 if (st
.st_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2762 n
= pread_in_full(fd
, hash
, the_hash_algo
->rawsz
, st
.st_size
- the_hash_algo
->rawsz
);
2763 if (n
!= the_hash_algo
->rawsz
)
2766 if (!hasheq(istate
->oid
.hash
, hash
))
2777 static int repo_verify_index(struct repository
*repo
)
2779 return verify_index_from(repo
->index
, repo
->index_file
);
2782 static int has_racy_timestamp(struct index_state
*istate
)
2784 int entries
= istate
->cache_nr
;
2787 for (i
= 0; i
< entries
; i
++) {
2788 struct cache_entry
*ce
= istate
->cache
[i
];
2789 if (is_racy_timestamp(istate
, ce
))
2795 void repo_update_index_if_able(struct repository
*repo
,
2796 struct lock_file
*lockfile
)
2798 if ((repo
->index
->cache_changed
||
2799 has_racy_timestamp(repo
->index
)) &&
2800 repo_verify_index(repo
))
2801 write_locked_index(repo
->index
, lockfile
, COMMIT_LOCK
);
2803 rollback_lock_file(lockfile
);
2806 static int record_eoie(void)
2810 if (!git_config_get_bool("index.recordendofindexentries", &val
))
2814 * As a convenience, the end of index entries extension
2815 * used for threading is written by default if the user
2816 * explicitly requested threaded index reads.
2818 return !git_config_get_index_threads(&val
) && val
!= 1;
2821 static int record_ieot(void)
2825 if (!git_config_get_bool("index.recordoffsettable", &val
))
2829 * As a convenience, the offset table used for threading is
2830 * written by default if the user explicitly requested
2831 * threaded index reads.
2833 return !git_config_get_index_threads(&val
) && val
!= 1;
2837 * On success, `tempfile` is closed. If it is the temporary file
2838 * of a `struct lock_file`, we will therefore effectively perform
2839 * a 'close_lock_file_gently()`. Since that is an implementation
2840 * detail of lockfiles, callers of `do_write_index()` should not
2843 static int do_write_index(struct index_state
*istate
, struct tempfile
*tempfile
,
2844 int strip_extensions
)
2846 uint64_t start
= getnanotime();
2848 git_hash_ctx
*eoie_c
= NULL
;
2849 struct cache_header hdr
;
2850 int i
, err
= 0, removed
, extended
, hdr_version
;
2851 struct cache_entry
**cache
= istate
->cache
;
2852 int entries
= istate
->cache_nr
;
2854 struct ondisk_cache_entry ondisk
;
2855 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
2856 int drop_cache_tree
= istate
->drop_cache_tree
;
2858 int ieot_entries
= 1;
2859 struct index_entry_offset_table
*ieot
= NULL
;
2862 f
= hashfd(tempfile
->fd
, tempfile
->filename
.buf
);
2864 for (i
= removed
= extended
= 0; i
< entries
; i
++) {
2865 if (cache
[i
]->ce_flags
& CE_REMOVE
)
2868 /* reduce extended entries if possible */
2869 cache
[i
]->ce_flags
&= ~CE_EXTENDED
;
2870 if (cache
[i
]->ce_flags
& CE_EXTENDED_FLAGS
) {
2872 cache
[i
]->ce_flags
|= CE_EXTENDED
;
2876 if (!istate
->version
)
2877 istate
->version
= get_index_format_default(the_repository
);
2879 /* demote version 3 to version 2 when the latter suffices */
2880 if (istate
->version
== 3 || istate
->version
== 2)
2881 istate
->version
= extended
? 3 : 2;
2883 hdr_version
= istate
->version
;
2885 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
2886 hdr
.hdr_version
= htonl(hdr_version
);
2887 hdr
.hdr_entries
= htonl(entries
- removed
);
2889 hashwrite(f
, &hdr
, sizeof(hdr
));
2891 if (!HAVE_THREADS
|| git_config_get_index_threads(&nr_threads
))
2894 if (nr_threads
!= 1 && record_ieot()) {
2895 int ieot_blocks
, cpus
;
2898 * ensure default number of ieot blocks maps evenly to the
2899 * default number of threads that will process them leaving
2900 * room for the thread to load the index extensions.
2903 ieot_blocks
= istate
->cache_nr
/ THREAD_COST
;
2904 cpus
= online_cpus();
2905 if (ieot_blocks
> cpus
- 1)
2906 ieot_blocks
= cpus
- 1;
2908 ieot_blocks
= nr_threads
;
2909 if (ieot_blocks
> istate
->cache_nr
)
2910 ieot_blocks
= istate
->cache_nr
;
2914 * no reason to write out the IEOT extension if we don't
2915 * have enough blocks to utilize multi-threading
2917 if (ieot_blocks
> 1) {
2918 ieot
= xcalloc(1, sizeof(struct index_entry_offset_table
)
2919 + (ieot_blocks
* sizeof(struct index_entry_offset
)));
2920 ieot_entries
= DIV_ROUND_UP(entries
, ieot_blocks
);
2924 offset
= hashfile_total(f
);
2927 previous_name
= (hdr_version
== 4) ? &previous_name_buf
: NULL
;
2929 for (i
= 0; i
< entries
; i
++) {
2930 struct cache_entry
*ce
= cache
[i
];
2931 if (ce
->ce_flags
& CE_REMOVE
)
2933 if (!ce_uptodate(ce
) && is_racy_timestamp(istate
, ce
))
2934 ce_smudge_racily_clean_entry(istate
, ce
);
2935 if (is_null_oid(&ce
->oid
)) {
2936 static const char msg
[] = "cache entry has null sha1: %s";
2937 static int allow
= -1;
2940 allow
= git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2942 warning(msg
, ce
->name
);
2944 err
= error(msg
, ce
->name
);
2946 drop_cache_tree
= 1;
2948 if (ieot
&& i
&& (i
% ieot_entries
== 0)) {
2949 ieot
->entries
[ieot
->nr
].nr
= nr
;
2950 ieot
->entries
[ieot
->nr
].offset
= offset
;
2953 * If we have a V4 index, set the first byte to an invalid
2954 * character to ensure there is nothing common with the previous
2958 previous_name
->buf
[0] = 0;
2961 offset
= hashfile_total(f
);
2963 if (ce_write_entry(f
, ce
, previous_name
, (struct ondisk_cache_entry
*)&ondisk
) < 0)
2971 ieot
->entries
[ieot
->nr
].nr
= nr
;
2972 ieot
->entries
[ieot
->nr
].offset
= offset
;
2975 strbuf_release(&previous_name_buf
);
2982 offset
= hashfile_total(f
);
2985 * The extension headers must be hashed on their own for the
2986 * EOIE extension. Create a hashfile here to compute that hash.
2988 if (offset
&& record_eoie()) {
2989 CALLOC_ARRAY(eoie_c
, 1);
2990 the_hash_algo
->init_fn(eoie_c
);
2994 * Lets write out CACHE_EXT_INDEXENTRYOFFSETTABLE first so that we
2995 * can minimize the number of extensions we have to scan through to
2996 * find it during load. Write it out regardless of the
2997 * strip_extensions parameter as we need it when loading the shared
3001 struct strbuf sb
= STRBUF_INIT
;
3003 write_ieot_extension(&sb
, ieot
);
3004 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_INDEXENTRYOFFSETTABLE
, sb
.len
) < 0;
3005 hashwrite(f
, sb
.buf
, sb
.len
);
3006 strbuf_release(&sb
);
3012 if (!strip_extensions
&& istate
->split_index
&&
3013 !is_null_oid(&istate
->split_index
->base_oid
)) {
3014 struct strbuf sb
= STRBUF_INIT
;
3016 err
= write_link_extension(&sb
, istate
) < 0 ||
3017 write_index_ext_header(f
, eoie_c
, CACHE_EXT_LINK
,
3019 hashwrite(f
, sb
.buf
, sb
.len
);
3020 strbuf_release(&sb
);
3024 if (!strip_extensions
&& !drop_cache_tree
&& istate
->cache_tree
) {
3025 struct strbuf sb
= STRBUF_INIT
;
3027 cache_tree_write(&sb
, istate
->cache_tree
);
3028 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_TREE
, sb
.len
) < 0;
3029 hashwrite(f
, sb
.buf
, sb
.len
);
3030 strbuf_release(&sb
);
3034 if (!strip_extensions
&& istate
->resolve_undo
) {
3035 struct strbuf sb
= STRBUF_INIT
;
3037 resolve_undo_write(&sb
, istate
->resolve_undo
);
3038 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_RESOLVE_UNDO
,
3040 hashwrite(f
, sb
.buf
, sb
.len
);
3041 strbuf_release(&sb
);
3045 if (!strip_extensions
&& istate
->untracked
) {
3046 struct strbuf sb
= STRBUF_INIT
;
3048 write_untracked_extension(&sb
, istate
->untracked
);
3049 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_UNTRACKED
,
3051 hashwrite(f
, sb
.buf
, sb
.len
);
3052 strbuf_release(&sb
);
3056 if (!strip_extensions
&& istate
->fsmonitor_last_update
) {
3057 struct strbuf sb
= STRBUF_INIT
;
3059 write_fsmonitor_extension(&sb
, istate
);
3060 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_FSMONITOR
, sb
.len
) < 0;
3061 hashwrite(f
, sb
.buf
, sb
.len
);
3062 strbuf_release(&sb
);
3066 if (istate
->sparse_index
) {
3067 if (write_index_ext_header(f
, eoie_c
, CACHE_EXT_SPARSE_DIRECTORIES
, 0) < 0)
3072 * CACHE_EXT_ENDOFINDEXENTRIES must be written as the last entry before the SHA1
3073 * so that it can be found and processed before all the index entries are
3074 * read. Write it out regardless of the strip_extensions parameter as we need it
3075 * when loading the shared index.
3078 struct strbuf sb
= STRBUF_INIT
;
3080 write_eoie_extension(&sb
, eoie_c
, offset
);
3081 err
= write_index_ext_header(f
, NULL
, CACHE_EXT_ENDOFINDEXENTRIES
, sb
.len
) < 0;
3082 hashwrite(f
, sb
.buf
, sb
.len
);
3083 strbuf_release(&sb
);
3088 finalize_hashfile(f
, istate
->oid
.hash
, CSUM_HASH_IN_STREAM
);
3089 if (close_tempfile_gently(tempfile
)) {
3090 error(_("could not close '%s'"), get_tempfile_path(tempfile
));
3093 if (stat(get_tempfile_path(tempfile
), &st
))
3095 istate
->timestamp
.sec
= (unsigned int)st
.st_mtime
;
3096 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
3097 trace_performance_since(start
, "write index, changed mask = %x", istate
->cache_changed
);
3100 * TODO trace2: replace "the_repository" with the actual repo instance
3101 * that is associated with the given "istate".
3103 trace2_data_intmax("index", the_repository
, "write/version",
3105 trace2_data_intmax("index", the_repository
, "write/cache_nr",
3111 void set_alternate_index_output(const char *name
)
3113 alternate_index_output
= name
;
3116 static int commit_locked_index(struct lock_file
*lk
)
3118 if (alternate_index_output
)
3119 return commit_lock_file_to(lk
, alternate_index_output
);
3121 return commit_lock_file(lk
);
3124 static int do_write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
3128 int was_full
= !istate
->sparse_index
;
3130 ret
= convert_to_sparse(istate
, 0);
3133 warning(_("failed to convert to a sparse-index"));
3138 * TODO trace2: replace "the_repository" with the actual repo instance
3139 * that is associated with the given "istate".
3141 trace2_region_enter_printf("index", "do_write_index", the_repository
,
3142 "%s", get_lock_file_path(lock
));
3143 ret
= do_write_index(istate
, lock
->tempfile
, 0);
3144 trace2_region_leave_printf("index", "do_write_index", the_repository
,
3145 "%s", get_lock_file_path(lock
));
3148 ensure_full_index(istate
);
3152 if (flags
& COMMIT_LOCK
)
3153 ret
= commit_locked_index(lock
);
3155 ret
= close_lock_file_gently(lock
);
3157 run_hook_le(NULL
, "post-index-change",
3158 istate
->updated_workdir
? "1" : "0",
3159 istate
->updated_skipworktree
? "1" : "0", NULL
);
3160 istate
->updated_workdir
= 0;
3161 istate
->updated_skipworktree
= 0;
3166 static int write_split_index(struct index_state
*istate
,
3167 struct lock_file
*lock
,
3171 prepare_to_write_split_index(istate
);
3172 ret
= do_write_locked_index(istate
, lock
, flags
);
3173 finish_writing_split_index(istate
);
3177 static const char *shared_index_expire
= "2.weeks.ago";
3179 static unsigned long get_shared_index_expire_date(void)
3181 static unsigned long shared_index_expire_date
;
3182 static int shared_index_expire_date_prepared
;
3184 if (!shared_index_expire_date_prepared
) {
3185 git_config_get_expiry("splitindex.sharedindexexpire",
3186 &shared_index_expire
);
3187 shared_index_expire_date
= approxidate(shared_index_expire
);
3188 shared_index_expire_date_prepared
= 1;
3191 return shared_index_expire_date
;
3194 static int should_delete_shared_index(const char *shared_index_path
)
3197 unsigned long expiration
;
3199 /* Check timestamp */
3200 expiration
= get_shared_index_expire_date();
3203 if (stat(shared_index_path
, &st
))
3204 return error_errno(_("could not stat '%s'"), shared_index_path
);
3205 if (st
.st_mtime
> expiration
)
3211 static int clean_shared_index_files(const char *current_hex
)
3214 DIR *dir
= opendir(get_git_dir());
3217 return error_errno(_("unable to open git dir: %s"), get_git_dir());
3219 while ((de
= readdir(dir
)) != NULL
) {
3220 const char *sha1_hex
;
3221 const char *shared_index_path
;
3222 if (!skip_prefix(de
->d_name
, "sharedindex.", &sha1_hex
))
3224 if (!strcmp(sha1_hex
, current_hex
))
3226 shared_index_path
= git_path("%s", de
->d_name
);
3227 if (should_delete_shared_index(shared_index_path
) > 0 &&
3228 unlink(shared_index_path
))
3229 warning_errno(_("unable to unlink: %s"), shared_index_path
);
3236 static int write_shared_index(struct index_state
*istate
,
3237 struct tempfile
**temp
)
3239 struct split_index
*si
= istate
->split_index
;
3240 int ret
, was_full
= !istate
->sparse_index
;
3242 move_cache_to_base_index(istate
);
3243 convert_to_sparse(istate
, 0);
3245 trace2_region_enter_printf("index", "shared/do_write_index",
3246 the_repository
, "%s", get_tempfile_path(*temp
));
3247 ret
= do_write_index(si
->base
, *temp
, 1);
3248 trace2_region_leave_printf("index", "shared/do_write_index",
3249 the_repository
, "%s", get_tempfile_path(*temp
));
3252 ensure_full_index(istate
);
3256 ret
= adjust_shared_perm(get_tempfile_path(*temp
));
3258 error(_("cannot fix permission bits on '%s'"), get_tempfile_path(*temp
));
3261 ret
= rename_tempfile(temp
,
3262 git_path("sharedindex.%s", oid_to_hex(&si
->base
->oid
)));
3264 oidcpy(&si
->base_oid
, &si
->base
->oid
);
3265 clean_shared_index_files(oid_to_hex(&si
->base
->oid
));
3271 static const int default_max_percent_split_change
= 20;
3273 static int too_many_not_shared_entries(struct index_state
*istate
)
3275 int i
, not_shared
= 0;
3276 int max_split
= git_config_get_max_percent_split_change();
3278 switch (max_split
) {
3280 /* not or badly configured: use the default value */
3281 max_split
= default_max_percent_split_change
;
3284 return 1; /* 0% means always write a new shared index */
3286 return 0; /* 100% means never write a new shared index */
3288 break; /* just use the configured value */
3291 /* Count not shared entries */
3292 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3293 struct cache_entry
*ce
= istate
->cache
[i
];
3298 return (int64_t)istate
->cache_nr
* max_split
< (int64_t)not_shared
* 100;
3301 int write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
3304 int new_shared_index
, ret
, test_split_index_env
;
3305 struct split_index
*si
= istate
->split_index
;
3307 if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0))
3308 cache_tree_verify(the_repository
, istate
);
3310 if ((flags
& SKIP_IF_UNCHANGED
) && !istate
->cache_changed
) {
3311 if (flags
& COMMIT_LOCK
)
3312 rollback_lock_file(lock
);
3316 if (istate
->fsmonitor_last_update
)
3317 fill_fsmonitor_bitmap(istate
);
3319 test_split_index_env
= git_env_bool("GIT_TEST_SPLIT_INDEX", 0);
3321 if ((!si
&& !test_split_index_env
) ||
3322 alternate_index_output
||
3323 (istate
->cache_changed
& ~EXTMASK
)) {
3325 oidclr(&si
->base_oid
);
3326 ret
= do_write_locked_index(istate
, lock
, flags
);
3330 if (test_split_index_env
) {
3332 si
= init_split_index(istate
);
3333 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3335 int v
= si
->base_oid
.hash
[0];
3337 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3340 if (too_many_not_shared_entries(istate
))
3341 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3343 new_shared_index
= istate
->cache_changed
& SPLIT_INDEX_ORDERED
;
3345 if (new_shared_index
) {
3346 struct tempfile
*temp
;
3349 /* Same initial permissions as the main .git/index file */
3350 temp
= mks_tempfile_sm(git_path("sharedindex_XXXXXX"), 0, 0666);
3352 oidclr(&si
->base_oid
);
3353 ret
= do_write_locked_index(istate
, lock
, flags
);
3356 ret
= write_shared_index(istate
, &temp
);
3358 saved_errno
= errno
;
3359 if (is_tempfile_active(temp
))
3360 delete_tempfile(&temp
);
3361 errno
= saved_errno
;
3367 ret
= write_split_index(istate
, lock
, flags
);
3369 /* Freshen the shared index only if the split-index was written */
3370 if (!ret
&& !new_shared_index
&& !is_null_oid(&si
->base_oid
)) {
3371 const char *shared_index
= git_path("sharedindex.%s",
3372 oid_to_hex(&si
->base_oid
));
3373 freshen_shared_index(shared_index
, 1);
3377 if (flags
& COMMIT_LOCK
)
3378 rollback_lock_file(lock
);
3383 * Read the index file that is potentially unmerged into given
3384 * index_state, dropping any unmerged entries to stage #0 (potentially
3385 * resulting in a path appearing as both a file and a directory in the
3386 * index; the caller is responsible to clear out the extra entries
3387 * before writing the index to a tree). Returns true if the index is
3388 * unmerged. Callers who want to refuse to work from an unmerged
3389 * state can call this and check its return value, instead of calling
3392 int repo_read_index_unmerged(struct repository
*repo
)
3394 struct index_state
*istate
;
3398 repo_read_index(repo
);
3399 istate
= repo
->index
;
3400 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3401 struct cache_entry
*ce
= istate
->cache
[i
];
3402 struct cache_entry
*new_ce
;
3408 len
= ce_namelen(ce
);
3409 new_ce
= make_empty_cache_entry(istate
, len
);
3410 memcpy(new_ce
->name
, ce
->name
, len
);
3411 new_ce
->ce_flags
= create_ce_flags(0) | CE_CONFLICTED
;
3412 new_ce
->ce_namelen
= len
;
3413 new_ce
->ce_mode
= ce
->ce_mode
;
3414 if (add_index_entry(istate
, new_ce
, ADD_CACHE_SKIP_DFCHECK
))
3415 return error(_("%s: cannot drop to stage #0"),
3422 * Returns 1 if the path is an "other" path with respect to
3423 * the index; that is, the path is not mentioned in the index at all,
3424 * either as a file, a directory with some files in the index,
3425 * or as an unmerged entry.
3427 * We helpfully remove a trailing "/" from directories so that
3428 * the output of read_directory can be used as-is.
3430 int index_name_is_other(struct index_state
*istate
, const char *name
,
3434 if (namelen
&& name
[namelen
- 1] == '/')
3436 pos
= index_name_pos(istate
, name
, namelen
);
3438 return 0; /* exact match */
3440 if (pos
< istate
->cache_nr
) {
3441 struct cache_entry
*ce
= istate
->cache
[pos
];
3442 if (ce_namelen(ce
) == namelen
&&
3443 !memcmp(ce
->name
, name
, namelen
))
3444 return 0; /* Yup, this one exists unmerged */
3449 void *read_blob_data_from_index(struct index_state
*istate
,
3450 const char *path
, unsigned long *size
)
3454 enum object_type type
;
3458 pos
= index_name_pos(istate
, path
, len
);
3461 * We might be in the middle of a merge, in which
3462 * case we would read stage #2 (ours).
3466 (pos
< 0 && i
< istate
->cache_nr
&&
3467 !strcmp(istate
->cache
[i
]->name
, path
));
3469 if (ce_stage(istate
->cache
[i
]) == 2)
3474 data
= read_object_file(&istate
->cache
[pos
]->oid
, &type
, &sz
);
3475 if (!data
|| type
!= OBJ_BLOB
) {
3484 void stat_validity_clear(struct stat_validity
*sv
)
3486 FREE_AND_NULL(sv
->sd
);
3489 int stat_validity_check(struct stat_validity
*sv
, const char *path
)
3493 if (stat(path
, &st
) < 0)
3494 return sv
->sd
== NULL
;
3497 return S_ISREG(st
.st_mode
) && !match_stat_data(sv
->sd
, &st
);
3500 void stat_validity_update(struct stat_validity
*sv
, int fd
)
3504 if (fstat(fd
, &st
) < 0 || !S_ISREG(st
.st_mode
))
3505 stat_validity_clear(sv
);
3508 CALLOC_ARRAY(sv
->sd
, 1);
3509 fill_stat_data(sv
->sd
, &st
);
3513 void move_index_extensions(struct index_state
*dst
, struct index_state
*src
)
3515 dst
->untracked
= src
->untracked
;
3516 src
->untracked
= NULL
;
3517 dst
->cache_tree
= src
->cache_tree
;
3518 src
->cache_tree
= NULL
;
3521 struct cache_entry
*dup_cache_entry(const struct cache_entry
*ce
,
3522 struct index_state
*istate
)
3524 unsigned int size
= ce_size(ce
);
3525 int mem_pool_allocated
;
3526 struct cache_entry
*new_entry
= make_empty_cache_entry(istate
, ce_namelen(ce
));
3527 mem_pool_allocated
= new_entry
->mem_pool_allocated
;
3529 memcpy(new_entry
, ce
, size
);
3530 new_entry
->mem_pool_allocated
= mem_pool_allocated
;
3534 void discard_cache_entry(struct cache_entry
*ce
)
3536 if (ce
&& should_validate_cache_entries())
3537 memset(ce
, 0xCD, cache_entry_size(ce
->ce_namelen
));
3539 if (ce
&& ce
->mem_pool_allocated
)
3545 int should_validate_cache_entries(void)
3547 static int validate_index_cache_entries
= -1;
3549 if (validate_index_cache_entries
< 0) {
3550 if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))
3551 validate_index_cache_entries
= 1;
3553 validate_index_cache_entries
= 0;
3556 return validate_index_cache_entries
;
3559 #define EOIE_SIZE (4 + GIT_SHA1_RAWSZ) /* <4-byte offset> + <20-byte hash> */
3560 #define EOIE_SIZE_WITH_HEADER (4 + 4 + EOIE_SIZE) /* <4-byte signature> + <4-byte length> + EOIE_SIZE */
3562 static size_t read_eoie_extension(const char *mmap
, size_t mmap_size
)
3565 * The end of index entries (EOIE) extension is guaranteed to be last
3566 * so that it can be found by scanning backwards from the EOF.
3573 const char *index
, *eoie
;
3575 size_t offset
, src_offset
;
3576 unsigned char hash
[GIT_MAX_RAWSZ
];
3579 /* ensure we have an index big enough to contain an EOIE extension */
3580 if (mmap_size
< sizeof(struct cache_header
) + EOIE_SIZE_WITH_HEADER
+ the_hash_algo
->rawsz
)
3583 /* validate the extension signature */
3584 index
= eoie
= mmap
+ mmap_size
- EOIE_SIZE_WITH_HEADER
- the_hash_algo
->rawsz
;
3585 if (CACHE_EXT(index
) != CACHE_EXT_ENDOFINDEXENTRIES
)
3587 index
+= sizeof(uint32_t);
3589 /* validate the extension size */
3590 extsize
= get_be32(index
);
3591 if (extsize
!= EOIE_SIZE
)
3593 index
+= sizeof(uint32_t);
3596 * Validate the offset we're going to look for the first extension
3597 * signature is after the index header and before the eoie extension.
3599 offset
= get_be32(index
);
3600 if (mmap
+ offset
< mmap
+ sizeof(struct cache_header
))
3602 if (mmap
+ offset
>= eoie
)
3604 index
+= sizeof(uint32_t);
3607 * The hash is computed over extension types and their sizes (but not
3608 * their contents). E.g. if we have "TREE" extension that is N-bytes
3609 * long, "REUC" extension that is M-bytes long, followed by "EOIE",
3610 * then the hash would be:
3612 * SHA-1("TREE" + <binary representation of N> +
3613 * "REUC" + <binary representation of M>)
3615 src_offset
= offset
;
3616 the_hash_algo
->init_fn(&c
);
3617 while (src_offset
< mmap_size
- the_hash_algo
->rawsz
- EOIE_SIZE_WITH_HEADER
) {
3618 /* After an array of active_nr index entries,
3619 * there can be arbitrary number of extended
3620 * sections, each of which is prefixed with
3621 * extension name (4-byte) and section length
3622 * in 4-byte network byte order.
3625 memcpy(&extsize
, mmap
+ src_offset
+ 4, 4);
3626 extsize
= ntohl(extsize
);
3628 /* verify the extension size isn't so large it will wrap around */
3629 if (src_offset
+ 8 + extsize
< src_offset
)
3632 the_hash_algo
->update_fn(&c
, mmap
+ src_offset
, 8);
3635 src_offset
+= extsize
;
3637 the_hash_algo
->final_fn(hash
, &c
);
3638 if (!hasheq(hash
, (const unsigned char *)index
))
3641 /* Validate that the extension offsets returned us back to the eoie extension. */
3642 if (src_offset
!= mmap_size
- the_hash_algo
->rawsz
- EOIE_SIZE_WITH_HEADER
)
3648 static void write_eoie_extension(struct strbuf
*sb
, git_hash_ctx
*eoie_context
, size_t offset
)
3651 unsigned char hash
[GIT_MAX_RAWSZ
];
3654 put_be32(&buffer
, offset
);
3655 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3658 the_hash_algo
->final_fn(hash
, eoie_context
);
3659 strbuf_add(sb
, hash
, the_hash_algo
->rawsz
);
3662 #define IEOT_VERSION (1)
3664 static struct index_entry_offset_table
*read_ieot_extension(const char *mmap
, size_t mmap_size
, size_t offset
)
3666 const char *index
= NULL
;
3667 uint32_t extsize
, ext_version
;
3668 struct index_entry_offset_table
*ieot
;
3671 /* find the IEOT extension */
3674 while (offset
<= mmap_size
- the_hash_algo
->rawsz
- 8) {
3675 extsize
= get_be32(mmap
+ offset
+ 4);
3676 if (CACHE_EXT((mmap
+ offset
)) == CACHE_EXT_INDEXENTRYOFFSETTABLE
) {
3677 index
= mmap
+ offset
+ 4 + 4;
3686 /* validate the version is IEOT_VERSION */
3687 ext_version
= get_be32(index
);
3688 if (ext_version
!= IEOT_VERSION
) {
3689 error("invalid IEOT version %d", ext_version
);
3692 index
+= sizeof(uint32_t);
3694 /* extension size - version bytes / bytes per entry */
3695 nr
= (extsize
- sizeof(uint32_t)) / (sizeof(uint32_t) + sizeof(uint32_t));
3697 error("invalid number of IEOT entries %d", nr
);
3700 ieot
= xmalloc(sizeof(struct index_entry_offset_table
)
3701 + (nr
* sizeof(struct index_entry_offset
)));
3703 for (i
= 0; i
< nr
; i
++) {
3704 ieot
->entries
[i
].offset
= get_be32(index
);
3705 index
+= sizeof(uint32_t);
3706 ieot
->entries
[i
].nr
= get_be32(index
);
3707 index
+= sizeof(uint32_t);
3713 static void write_ieot_extension(struct strbuf
*sb
, struct index_entry_offset_table
*ieot
)
3719 put_be32(&buffer
, IEOT_VERSION
);
3720 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3723 for (i
= 0; i
< ieot
->nr
; i
++) {
3726 put_be32(&buffer
, ieot
->entries
[i
].offset
);
3727 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3730 put_be32(&buffer
, ieot
->entries
[i
].nr
);
3731 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3735 void prefetch_cache_entries(const struct index_state
*istate
,
3736 must_prefetch_predicate must_prefetch
)
3739 struct oid_array to_fetch
= OID_ARRAY_INIT
;
3741 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3742 struct cache_entry
*ce
= istate
->cache
[i
];
3744 if (S_ISGITLINK(ce
->ce_mode
) || !must_prefetch(ce
))
3746 if (!oid_object_info_extended(the_repository
, &ce
->oid
,
3748 OBJECT_INFO_FOR_PREFETCH
))
3750 oid_array_append(&to_fetch
, &ce
->oid
);
3752 promisor_remote_get_direct(the_repository
,
3753 to_fetch
.oid
, to_fetch
.nr
);
3754 oid_array_clear(&to_fetch
);