2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
12 #include "cache-tree.h"
15 #include "object-store.h"
19 #include "resolve-undo.h"
20 #include "run-command.h"
23 #include "split-index.h"
25 #include "fsmonitor.h"
26 #include "thread-utils.h"
28 #include "sparse-index.h"
29 #include "csum-file.h"
30 #include "promisor-remote.h"
33 /* Mask for the name length in ce_flags in the on-disk index */
35 #define CE_NAMEMASK (0x0fff)
39 * The first letter should be 'A'..'Z' for extensions that are not
40 * necessary for a correct operation (i.e. optimization data).
41 * When new extensions are added that _needs_ to be understood in
42 * order to correctly interpret the index file, pick character that
43 * is outside the range, to cause the reader to abort.
46 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
47 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
48 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
49 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
50 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
51 #define CACHE_EXT_FSMONITOR 0x46534D4E /* "FSMN" */
52 #define CACHE_EXT_ENDOFINDEXENTRIES 0x454F4945 /* "EOIE" */
53 #define CACHE_EXT_INDEXENTRYOFFSETTABLE 0x49454F54 /* "IEOT" */
54 #define CACHE_EXT_SPARSE_DIRECTORIES 0x73646972 /* "sdir" */
56 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
57 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
58 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
59 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED | FSMONITOR_CHANGED)
63 * This is an estimate of the pathname length in the index. We use
64 * this for V4 index files to guess the un-deltafied size of the index
65 * in memory because of pathname deltafication. This is not required
66 * for V2/V3 index formats because their pathnames are not compressed.
67 * If the initial amount of memory set aside is not sufficient, the
68 * mem pool will allocate extra memory.
70 #define CACHE_ENTRY_PATH_LENGTH 80
72 enum index_search_mode
{
77 static inline struct cache_entry
*mem_pool__ce_alloc(struct mem_pool
*mem_pool
, size_t len
)
79 struct cache_entry
*ce
;
80 ce
= mem_pool_alloc(mem_pool
, cache_entry_size(len
));
81 ce
->mem_pool_allocated
= 1;
85 static inline struct cache_entry
*mem_pool__ce_calloc(struct mem_pool
*mem_pool
, size_t len
)
87 struct cache_entry
* ce
;
88 ce
= mem_pool_calloc(mem_pool
, 1, cache_entry_size(len
));
89 ce
->mem_pool_allocated
= 1;
93 static struct mem_pool
*find_mem_pool(struct index_state
*istate
)
95 struct mem_pool
**pool_ptr
;
97 if (istate
->split_index
&& istate
->split_index
->base
)
98 pool_ptr
= &istate
->split_index
->base
->ce_mem_pool
;
100 pool_ptr
= &istate
->ce_mem_pool
;
103 *pool_ptr
= xmalloc(sizeof(**pool_ptr
));
104 mem_pool_init(*pool_ptr
, 0);
110 static const char *alternate_index_output
;
112 static void set_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
114 if (S_ISSPARSEDIR(ce
->ce_mode
))
115 istate
->sparse_index
= INDEX_COLLAPSED
;
117 istate
->cache
[nr
] = ce
;
118 add_name_hash(istate
, ce
);
121 static void replace_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
123 struct cache_entry
*old
= istate
->cache
[nr
];
125 replace_index_entry_in_base(istate
, old
, ce
);
126 remove_name_hash(istate
, old
);
127 discard_cache_entry(old
);
128 ce
->ce_flags
&= ~CE_HASHED
;
129 set_index_entry(istate
, nr
, ce
);
130 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
131 mark_fsmonitor_invalid(istate
, ce
);
132 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
135 void rename_index_entry_at(struct index_state
*istate
, int nr
, const char *new_name
)
137 struct cache_entry
*old_entry
= istate
->cache
[nr
], *new_entry
, *refreshed
;
138 int namelen
= strlen(new_name
);
140 new_entry
= make_empty_cache_entry(istate
, namelen
);
141 copy_cache_entry(new_entry
, old_entry
);
142 new_entry
->ce_flags
&= ~CE_HASHED
;
143 new_entry
->ce_namelen
= namelen
;
144 new_entry
->index
= 0;
145 memcpy(new_entry
->name
, new_name
, namelen
+ 1);
147 cache_tree_invalidate_path(istate
, old_entry
->name
);
148 untracked_cache_remove_from_index(istate
, old_entry
->name
);
149 remove_index_entry_at(istate
, nr
);
152 * Refresh the new index entry. Using 'refresh_cache_entry' ensures
153 * we only update stat info if the entry is otherwise up-to-date (i.e.,
154 * the contents/mode haven't changed). This ensures that we reflect the
155 * 'ctime' of the rename in the index without (incorrectly) updating
156 * the cached stat info to reflect unstaged changes on disk.
158 refreshed
= refresh_cache_entry(istate
, new_entry
, CE_MATCH_REFRESH
);
159 if (refreshed
&& refreshed
!= new_entry
) {
160 add_index_entry(istate
, refreshed
, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
161 discard_cache_entry(new_entry
);
163 add_index_entry(istate
, new_entry
, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
166 void fill_stat_data(struct stat_data
*sd
, struct stat
*st
)
168 sd
->sd_ctime
.sec
= (unsigned int)st
->st_ctime
;
169 sd
->sd_mtime
.sec
= (unsigned int)st
->st_mtime
;
170 sd
->sd_ctime
.nsec
= ST_CTIME_NSEC(*st
);
171 sd
->sd_mtime
.nsec
= ST_MTIME_NSEC(*st
);
172 sd
->sd_dev
= st
->st_dev
;
173 sd
->sd_ino
= st
->st_ino
;
174 sd
->sd_uid
= st
->st_uid
;
175 sd
->sd_gid
= st
->st_gid
;
176 sd
->sd_size
= st
->st_size
;
179 int match_stat_data(const struct stat_data
*sd
, struct stat
*st
)
183 if (sd
->sd_mtime
.sec
!= (unsigned int)st
->st_mtime
)
184 changed
|= MTIME_CHANGED
;
185 if (trust_ctime
&& check_stat
&&
186 sd
->sd_ctime
.sec
!= (unsigned int)st
->st_ctime
)
187 changed
|= CTIME_CHANGED
;
190 if (check_stat
&& sd
->sd_mtime
.nsec
!= ST_MTIME_NSEC(*st
))
191 changed
|= MTIME_CHANGED
;
192 if (trust_ctime
&& check_stat
&&
193 sd
->sd_ctime
.nsec
!= ST_CTIME_NSEC(*st
))
194 changed
|= CTIME_CHANGED
;
198 if (sd
->sd_uid
!= (unsigned int) st
->st_uid
||
199 sd
->sd_gid
!= (unsigned int) st
->st_gid
)
200 changed
|= OWNER_CHANGED
;
201 if (sd
->sd_ino
!= (unsigned int) st
->st_ino
)
202 changed
|= INODE_CHANGED
;
207 * st_dev breaks on network filesystems where different
208 * clients will have different views of what "device"
209 * the filesystem is on
211 if (check_stat
&& sd
->sd_dev
!= (unsigned int) st
->st_dev
)
212 changed
|= INODE_CHANGED
;
215 if (sd
->sd_size
!= (unsigned int) st
->st_size
)
216 changed
|= DATA_CHANGED
;
222 * This only updates the "non-critical" parts of the directory
223 * cache, ie the parts that aren't tracked by GIT, and only used
224 * to validate the cache.
226 void fill_stat_cache_info(struct index_state
*istate
, struct cache_entry
*ce
, struct stat
*st
)
228 fill_stat_data(&ce
->ce_stat_data
, st
);
230 if (assume_unchanged
)
231 ce
->ce_flags
|= CE_VALID
;
233 if (S_ISREG(st
->st_mode
)) {
234 ce_mark_uptodate(ce
);
235 mark_fsmonitor_valid(istate
, ce
);
239 static int ce_compare_data(struct index_state
*istate
,
240 const struct cache_entry
*ce
,
244 int fd
= git_open_cloexec(ce
->name
, O_RDONLY
);
247 struct object_id oid
;
248 if (!index_fd(istate
, &oid
, fd
, st
, OBJ_BLOB
, ce
->name
, 0))
249 match
= !oideq(&oid
, &ce
->oid
);
250 /* index_fd() closed the file descriptor already */
255 static int ce_compare_link(const struct cache_entry
*ce
, size_t expected_size
)
260 enum object_type type
;
261 struct strbuf sb
= STRBUF_INIT
;
263 if (strbuf_readlink(&sb
, ce
->name
, expected_size
))
266 buffer
= read_object_file(&ce
->oid
, &type
, &size
);
269 match
= memcmp(buffer
, sb
.buf
, size
);
276 static int ce_compare_gitlink(const struct cache_entry
*ce
)
278 struct object_id oid
;
281 * We don't actually require that the .git directory
282 * under GITLINK directory be a valid git directory. It
283 * might even be missing (in case nobody populated that
286 * If so, we consider it always to match.
288 if (resolve_gitlink_ref(ce
->name
, "HEAD", &oid
) < 0)
290 return !oideq(&oid
, &ce
->oid
);
293 static int ce_modified_check_fs(struct index_state
*istate
,
294 const struct cache_entry
*ce
,
297 switch (st
->st_mode
& S_IFMT
) {
299 if (ce_compare_data(istate
, ce
, st
))
303 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
307 if (S_ISGITLINK(ce
->ce_mode
))
308 return ce_compare_gitlink(ce
) ? DATA_CHANGED
: 0;
309 /* else fallthrough */
316 static int ce_match_stat_basic(const struct cache_entry
*ce
, struct stat
*st
)
318 unsigned int changed
= 0;
320 if (ce
->ce_flags
& CE_REMOVE
)
321 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
323 switch (ce
->ce_mode
& S_IFMT
) {
325 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
326 /* We consider only the owner x bit to be relevant for
329 if (trust_executable_bit
&&
330 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
331 changed
|= MODE_CHANGED
;
334 if (!S_ISLNK(st
->st_mode
) &&
335 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
336 changed
|= TYPE_CHANGED
;
339 /* We ignore most of the st_xxx fields for gitlinks */
340 if (!S_ISDIR(st
->st_mode
))
341 changed
|= TYPE_CHANGED
;
342 else if (ce_compare_gitlink(ce
))
343 changed
|= DATA_CHANGED
;
346 BUG("unsupported ce_mode: %o", ce
->ce_mode
);
349 changed
|= match_stat_data(&ce
->ce_stat_data
, st
);
351 /* Racily smudged entry? */
352 if (!ce
->ce_stat_data
.sd_size
) {
353 if (!is_empty_blob_sha1(ce
->oid
.hash
))
354 changed
|= DATA_CHANGED
;
360 static int is_racy_stat(const struct index_state
*istate
,
361 const struct stat_data
*sd
)
363 return (istate
->timestamp
.sec
&&
365 /* nanosecond timestamped files can also be racy! */
366 (istate
->timestamp
.sec
< sd
->sd_mtime
.sec
||
367 (istate
->timestamp
.sec
== sd
->sd_mtime
.sec
&&
368 istate
->timestamp
.nsec
<= sd
->sd_mtime
.nsec
))
370 istate
->timestamp
.sec
<= sd
->sd_mtime
.sec
375 int is_racy_timestamp(const struct index_state
*istate
,
376 const struct cache_entry
*ce
)
378 return (!S_ISGITLINK(ce
->ce_mode
) &&
379 is_racy_stat(istate
, &ce
->ce_stat_data
));
382 int match_stat_data_racy(const struct index_state
*istate
,
383 const struct stat_data
*sd
, struct stat
*st
)
385 if (is_racy_stat(istate
, sd
))
386 return MTIME_CHANGED
;
387 return match_stat_data(sd
, st
);
390 int ie_match_stat(struct index_state
*istate
,
391 const struct cache_entry
*ce
, struct stat
*st
,
392 unsigned int options
)
394 unsigned int changed
;
395 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
396 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
397 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
398 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
400 if (!ignore_fsmonitor
)
401 refresh_fsmonitor(istate
);
403 * If it's marked as always valid in the index, it's
404 * valid whatever the checked-out copy says.
406 * skip-worktree has the same effect with higher precedence
408 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
))
410 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
412 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
))
416 * Intent-to-add entries have not been added, so the index entry
417 * by definition never matches what is in the work tree until it
418 * actually gets added.
420 if (ce_intent_to_add(ce
))
421 return DATA_CHANGED
| TYPE_CHANGED
| MODE_CHANGED
;
423 changed
= ce_match_stat_basic(ce
, st
);
426 * Within 1 second of this sequence:
427 * echo xyzzy >file && git-update-index --add file
428 * running this command:
430 * would give a falsely clean cache entry. The mtime and
431 * length match the cache, and other stat fields do not change.
433 * We could detect this at update-index time (the cache entry
434 * being registered/updated records the same time as "now")
435 * and delay the return from git-update-index, but that would
436 * effectively mean we can make at most one commit per second,
437 * which is not acceptable. Instead, we check cache entries
438 * whose mtime are the same as the index file timestamp more
439 * carefully than others.
441 if (!changed
&& is_racy_timestamp(istate
, ce
)) {
442 if (assume_racy_is_modified
)
443 changed
|= DATA_CHANGED
;
445 changed
|= ce_modified_check_fs(istate
, ce
, st
);
451 int ie_modified(struct index_state
*istate
,
452 const struct cache_entry
*ce
,
453 struct stat
*st
, unsigned int options
)
455 int changed
, changed_fs
;
457 changed
= ie_match_stat(istate
, ce
, st
, options
);
461 * If the mode or type has changed, there's no point in trying
462 * to refresh the entry - it's not going to match
464 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
468 * Immediately after read-tree or update-index --cacheinfo,
469 * the length field is zero, as we have never even read the
470 * lstat(2) information once, and we cannot trust DATA_CHANGED
471 * returned by ie_match_stat() which in turn was returned by
472 * ce_match_stat_basic() to signal that the filesize of the
473 * blob changed. We have to actually go to the filesystem to
474 * see if the contents match, and if so, should answer "unchanged".
476 * The logic does not apply to gitlinks, as ce_match_stat_basic()
477 * already has checked the actual HEAD from the filesystem in the
478 * subproject. If ie_match_stat() already said it is different,
479 * then we know it is.
481 if ((changed
& DATA_CHANGED
) &&
482 (S_ISGITLINK(ce
->ce_mode
) || ce
->ce_stat_data
.sd_size
!= 0))
485 changed_fs
= ce_modified_check_fs(istate
, ce
, st
);
487 return changed
| changed_fs
;
491 int base_name_compare(const char *name1
, int len1
, int mode1
,
492 const char *name2
, int len2
, int mode2
)
494 unsigned char c1
, c2
;
495 int len
= len1
< len2
? len1
: len2
;
498 cmp
= memcmp(name1
, name2
, len
);
503 if (!c1
&& S_ISDIR(mode1
))
505 if (!c2
&& S_ISDIR(mode2
))
507 return (c1
< c2
) ? -1 : (c1
> c2
) ? 1 : 0;
511 * df_name_compare() is identical to base_name_compare(), except it
512 * compares conflicting directory/file entries as equal. Note that
513 * while a directory name compares as equal to a regular file, they
514 * then individually compare _differently_ to a filename that has
515 * a dot after the basename (because '\0' < '.' < '/').
517 * This is used by routines that want to traverse the git namespace
518 * but then handle conflicting entries together when possible.
520 int df_name_compare(const char *name1
, int len1
, int mode1
,
521 const char *name2
, int len2
, int mode2
)
523 int len
= len1
< len2
? len1
: len2
, cmp
;
524 unsigned char c1
, c2
;
526 cmp
= memcmp(name1
, name2
, len
);
529 /* Directories and files compare equal (same length, same name) */
533 if (!c1
&& S_ISDIR(mode1
))
536 if (!c2
&& S_ISDIR(mode2
))
538 if (c1
== '/' && !c2
)
540 if (c2
== '/' && !c1
)
545 int name_compare(const char *name1
, size_t len1
, const char *name2
, size_t len2
)
547 size_t min_len
= (len1
< len2
) ? len1
: len2
;
548 int cmp
= memcmp(name1
, name2
, min_len
);
558 int cache_name_stage_compare(const char *name1
, int len1
, int stage1
, const char *name2
, int len2
, int stage2
)
562 cmp
= name_compare(name1
, len1
, name2
, len2
);
573 static int index_name_stage_pos(struct index_state
*istate
,
574 const char *name
, int namelen
,
576 enum index_search_mode search_mode
)
581 last
= istate
->cache_nr
;
582 while (last
> first
) {
583 int next
= first
+ ((last
- first
) >> 1);
584 struct cache_entry
*ce
= istate
->cache
[next
];
585 int cmp
= cache_name_stage_compare(name
, namelen
, stage
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
595 if (search_mode
== EXPAND_SPARSE
&& istate
->sparse_index
&&
597 /* Note: first <= istate->cache_nr */
598 struct cache_entry
*ce
= istate
->cache
[first
- 1];
601 * If we are in a sparse-index _and_ the entry before the
602 * insertion position is a sparse-directory entry that is
603 * an ancestor of 'name', then we need to expand the index
604 * and search again. This will only trigger once, because
605 * thereafter the index is fully expanded.
607 if (S_ISSPARSEDIR(ce
->ce_mode
) &&
608 ce_namelen(ce
) < namelen
&&
609 !strncmp(name
, ce
->name
, ce_namelen(ce
))) {
610 ensure_full_index(istate
);
611 return index_name_stage_pos(istate
, name
, namelen
, stage
, search_mode
);
618 int index_name_pos(struct index_state
*istate
, const char *name
, int namelen
)
620 return index_name_stage_pos(istate
, name
, namelen
, 0, EXPAND_SPARSE
);
623 int index_entry_exists(struct index_state
*istate
, const char *name
, int namelen
)
625 return index_name_stage_pos(istate
, name
, namelen
, 0, NO_EXPAND_SPARSE
) >= 0;
628 int remove_index_entry_at(struct index_state
*istate
, int pos
)
630 struct cache_entry
*ce
= istate
->cache
[pos
];
632 record_resolve_undo(istate
, ce
);
633 remove_name_hash(istate
, ce
);
634 save_or_free_index_entry(istate
, ce
);
635 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
637 if (pos
>= istate
->cache_nr
)
639 MOVE_ARRAY(istate
->cache
+ pos
, istate
->cache
+ pos
+ 1,
640 istate
->cache_nr
- pos
);
645 * Remove all cache entries marked for removal, that is where
646 * CE_REMOVE is set in ce_flags. This is much more effective than
647 * calling remove_index_entry_at() for each entry to be removed.
649 void remove_marked_cache_entries(struct index_state
*istate
, int invalidate
)
651 struct cache_entry
**ce_array
= istate
->cache
;
654 for (i
= j
= 0; i
< istate
->cache_nr
; i
++) {
655 if (ce_array
[i
]->ce_flags
& CE_REMOVE
) {
657 cache_tree_invalidate_path(istate
,
659 untracked_cache_remove_from_index(istate
,
662 remove_name_hash(istate
, ce_array
[i
]);
663 save_or_free_index_entry(istate
, ce_array
[i
]);
666 ce_array
[j
++] = ce_array
[i
];
668 if (j
== istate
->cache_nr
)
670 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
671 istate
->cache_nr
= j
;
674 int remove_file_from_index(struct index_state
*istate
, const char *path
)
676 int pos
= index_name_pos(istate
, path
, strlen(path
));
679 cache_tree_invalidate_path(istate
, path
);
680 untracked_cache_remove_from_index(istate
, path
);
681 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
682 remove_index_entry_at(istate
, pos
);
686 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
688 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
691 static int index_name_pos_also_unmerged(struct index_state
*istate
,
692 const char *path
, int namelen
)
694 int pos
= index_name_pos(istate
, path
, namelen
);
695 struct cache_entry
*ce
;
700 /* maybe unmerged? */
702 if (pos
>= istate
->cache_nr
||
703 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
706 /* order of preference: stage 2, 1, 3 */
707 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
708 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
709 !compare_name(ce
, path
, namelen
))
714 static int different_name(struct cache_entry
*ce
, struct cache_entry
*alias
)
716 int len
= ce_namelen(ce
);
717 return ce_namelen(alias
) != len
|| memcmp(ce
->name
, alias
->name
, len
);
721 * If we add a filename that aliases in the cache, we will use the
722 * name that we already have - but we don't want to update the same
723 * alias twice, because that implies that there were actually two
724 * different files with aliasing names!
726 * So we use the CE_ADDED flag to verify that the alias was an old
727 * one before we accept it as
729 static struct cache_entry
*create_alias_ce(struct index_state
*istate
,
730 struct cache_entry
*ce
,
731 struct cache_entry
*alias
)
734 struct cache_entry
*new_entry
;
736 if (alias
->ce_flags
& CE_ADDED
)
737 die(_("will not add file alias '%s' ('%s' already exists in index)"),
738 ce
->name
, alias
->name
);
740 /* Ok, create the new entry using the name of the existing alias */
741 len
= ce_namelen(alias
);
742 new_entry
= make_empty_cache_entry(istate
, len
);
743 memcpy(new_entry
->name
, alias
->name
, len
);
744 copy_cache_entry(new_entry
, ce
);
745 save_or_free_index_entry(istate
, ce
);
749 void set_object_name_for_intent_to_add_entry(struct cache_entry
*ce
)
751 struct object_id oid
;
752 if (write_object_file("", 0, OBJ_BLOB
, &oid
))
753 die(_("cannot create an empty blob in the object database"));
754 oidcpy(&ce
->oid
, &oid
);
757 int add_to_index(struct index_state
*istate
, const char *path
, struct stat
*st
, int flags
)
759 int namelen
, was_same
;
760 mode_t st_mode
= st
->st_mode
;
761 struct cache_entry
*ce
, *alias
= NULL
;
762 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_IGNORE_SKIP_WORKTREE
|CE_MATCH_RACY_IS_DIRTY
;
763 int verbose
= flags
& (ADD_CACHE_VERBOSE
| ADD_CACHE_PRETEND
);
764 int pretend
= flags
& ADD_CACHE_PRETEND
;
765 int intent_only
= flags
& ADD_CACHE_INTENT
;
766 int add_option
= (ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
|
767 (intent_only
? ADD_CACHE_NEW_ONLY
: 0));
768 unsigned hash_flags
= pretend
? 0 : HASH_WRITE_OBJECT
;
769 struct object_id oid
;
771 if (flags
& ADD_CACHE_RENORMALIZE
)
772 hash_flags
|= HASH_RENORMALIZE
;
774 if (!S_ISREG(st_mode
) && !S_ISLNK(st_mode
) && !S_ISDIR(st_mode
))
775 return error(_("%s: can only add regular files, symbolic links or git-directories"), path
);
777 namelen
= strlen(path
);
778 if (S_ISDIR(st_mode
)) {
779 if (resolve_gitlink_ref(path
, "HEAD", &oid
) < 0)
780 return error(_("'%s' does not have a commit checked out"), path
);
781 while (namelen
&& path
[namelen
-1] == '/')
784 ce
= make_empty_cache_entry(istate
, namelen
);
785 memcpy(ce
->name
, path
, namelen
);
786 ce
->ce_namelen
= namelen
;
788 fill_stat_cache_info(istate
, ce
, st
);
790 ce
->ce_flags
|= CE_INTENT_TO_ADD
;
793 if (trust_executable_bit
&& has_symlinks
) {
794 ce
->ce_mode
= create_ce_mode(st_mode
);
796 /* If there is an existing entry, pick the mode bits and type
797 * from it, otherwise assume unexecutable regular file.
799 struct cache_entry
*ent
;
800 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
802 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
803 ce
->ce_mode
= ce_mode_from_stat(ent
, st_mode
);
806 /* When core.ignorecase=true, determine if a directory of the same name but differing
807 * case already exists within the Git repository. If it does, ensure the directory
808 * case of the file being added to the repository matches (is folded into) the existing
809 * entry's directory case.
812 adjust_dirname_case(istate
, ce
->name
);
814 if (!(flags
& ADD_CACHE_RENORMALIZE
)) {
815 alias
= index_file_exists(istate
, ce
->name
,
816 ce_namelen(ce
), ignore_case
);
819 !ie_match_stat(istate
, alias
, st
, ce_option
)) {
820 /* Nothing changed, really */
821 if (!S_ISGITLINK(alias
->ce_mode
))
822 ce_mark_uptodate(alias
);
823 alias
->ce_flags
|= CE_ADDED
;
825 discard_cache_entry(ce
);
830 if (index_path(istate
, &ce
->oid
, path
, st
, hash_flags
)) {
831 discard_cache_entry(ce
);
832 return error(_("unable to index file '%s'"), path
);
835 set_object_name_for_intent_to_add_entry(ce
);
837 if (ignore_case
&& alias
&& different_name(ce
, alias
))
838 ce
= create_alias_ce(istate
, ce
, alias
);
839 ce
->ce_flags
|= CE_ADDED
;
841 /* It was suspected to be racily clean, but it turns out to be Ok */
844 oideq(&alias
->oid
, &ce
->oid
) &&
845 ce
->ce_mode
== alias
->ce_mode
);
848 discard_cache_entry(ce
);
849 else if (add_index_entry(istate
, ce
, add_option
)) {
850 discard_cache_entry(ce
);
851 return error(_("unable to add '%s' to index"), path
);
853 if (verbose
&& !was_same
)
854 printf("add '%s'\n", path
);
858 int add_file_to_index(struct index_state
*istate
, const char *path
, int flags
)
861 if (lstat(path
, &st
))
862 die_errno(_("unable to stat '%s'"), path
);
863 return add_to_index(istate
, path
, &st
, flags
);
866 struct cache_entry
*make_empty_cache_entry(struct index_state
*istate
, size_t len
)
868 return mem_pool__ce_calloc(find_mem_pool(istate
), len
);
871 struct cache_entry
*make_empty_transient_cache_entry(size_t len
,
872 struct mem_pool
*ce_mem_pool
)
875 return mem_pool__ce_calloc(ce_mem_pool
, len
);
876 return xcalloc(1, cache_entry_size(len
));
879 enum verify_path_result
{
885 static enum verify_path_result
verify_path_internal(const char *, unsigned);
887 int verify_path(const char *path
, unsigned mode
)
889 return verify_path_internal(path
, mode
) == PATH_OK
;
892 struct cache_entry
*make_cache_entry(struct index_state
*istate
,
894 const struct object_id
*oid
,
897 unsigned int refresh_options
)
899 struct cache_entry
*ce
, *ret
;
902 if (verify_path_internal(path
, mode
) == PATH_INVALID
) {
903 error(_("invalid path '%s'"), path
);
908 ce
= make_empty_cache_entry(istate
, len
);
910 oidcpy(&ce
->oid
, oid
);
911 memcpy(ce
->name
, path
, len
);
912 ce
->ce_flags
= create_ce_flags(stage
);
913 ce
->ce_namelen
= len
;
914 ce
->ce_mode
= create_ce_mode(mode
);
916 ret
= refresh_cache_entry(istate
, ce
, refresh_options
);
918 discard_cache_entry(ce
);
922 struct cache_entry
*make_transient_cache_entry(unsigned int mode
,
923 const struct object_id
*oid
,
926 struct mem_pool
*ce_mem_pool
)
928 struct cache_entry
*ce
;
931 if (!verify_path(path
, mode
)) {
932 error(_("invalid path '%s'"), path
);
937 ce
= make_empty_transient_cache_entry(len
, ce_mem_pool
);
939 oidcpy(&ce
->oid
, oid
);
940 memcpy(ce
->name
, path
, len
);
941 ce
->ce_flags
= create_ce_flags(stage
);
942 ce
->ce_namelen
= len
;
943 ce
->ce_mode
= create_ce_mode(mode
);
949 * Chmod an index entry with either +x or -x.
951 * Returns -1 if the chmod for the particular cache entry failed (if it's
952 * not a regular file), -2 if an invalid flip argument is passed in, 0
955 int chmod_index_entry(struct index_state
*istate
, struct cache_entry
*ce
,
958 if (!S_ISREG(ce
->ce_mode
))
965 ce
->ce_mode
&= ~0111;
970 cache_tree_invalidate_path(istate
, ce
->name
);
971 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
972 mark_fsmonitor_invalid(istate
, ce
);
973 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
978 int ce_same_name(const struct cache_entry
*a
, const struct cache_entry
*b
)
980 int len
= ce_namelen(a
);
981 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
985 * We fundamentally don't like some paths: we don't want
986 * dot or dot-dot anywhere, and for obvious reasons don't
987 * want to recurse into ".git" either.
989 * Also, we don't want double slashes or slashes at the
990 * end that can make pathnames ambiguous.
992 static int verify_dotfile(const char *rest
, unsigned mode
)
995 * The first character was '.', but that
996 * has already been discarded, we now test
1000 /* "." is not allowed */
1001 if (*rest
== '\0' || is_dir_sep(*rest
))
1006 * ".git" followed by NUL or slash is bad. Note that we match
1007 * case-insensitively here, even if ignore_case is not set.
1008 * This outlaws ".GIT" everywhere out of an abundance of caution,
1009 * since there's really no good reason to allow it.
1011 * Once we've seen ".git", we can also find ".gitmodules", etc (also
1012 * case-insensitively).
1016 if (rest
[1] != 'i' && rest
[1] != 'I')
1018 if (rest
[2] != 't' && rest
[2] != 'T')
1020 if (rest
[3] == '\0' || is_dir_sep(rest
[3]))
1022 if (S_ISLNK(mode
)) {
1024 if (skip_iprefix(rest
, "modules", &rest
) &&
1025 (*rest
== '\0' || is_dir_sep(*rest
)))
1030 if (rest
[1] == '\0' || is_dir_sep(rest
[1]))
1036 static enum verify_path_result
verify_path_internal(const char *path
,
1041 if (has_dos_drive_prefix(path
))
1042 return PATH_INVALID
;
1044 if (!is_valid_path(path
))
1045 return PATH_INVALID
;
1051 if (is_dir_sep(c
)) {
1055 if (is_hfs_dotgit(path
))
1056 return PATH_INVALID
;
1057 if (S_ISLNK(mode
)) {
1058 if (is_hfs_dotgitmodules(path
))
1059 return PATH_INVALID
;
1063 #if defined GIT_WINDOWS_NATIVE || defined __CYGWIN__
1065 return PATH_INVALID
;
1067 if (is_ntfs_dotgit(path
))
1068 return PATH_INVALID
;
1069 if (S_ISLNK(mode
)) {
1070 if (is_ntfs_dotgitmodules(path
))
1071 return PATH_INVALID
;
1076 if ((c
== '.' && !verify_dotfile(path
, mode
)) ||
1078 return PATH_INVALID
;
1080 * allow terminating directory separators for
1081 * sparse directory entries.
1084 return S_ISDIR(mode
) ? PATH_DIR_WITH_SEP
:
1086 } else if (c
== '\\' && protect_ntfs
) {
1087 if (is_ntfs_dotgit(path
))
1088 return PATH_INVALID
;
1089 if (S_ISLNK(mode
)) {
1090 if (is_ntfs_dotgitmodules(path
))
1091 return PATH_INVALID
;
1100 * Do we have another file that has the beginning components being a
1101 * proper superset of the name we're trying to add?
1103 static int has_file_name(struct index_state
*istate
,
1104 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
1107 int len
= ce_namelen(ce
);
1108 int stage
= ce_stage(ce
);
1109 const char *name
= ce
->name
;
1111 while (pos
< istate
->cache_nr
) {
1112 struct cache_entry
*p
= istate
->cache
[pos
++];
1114 if (len
>= ce_namelen(p
))
1116 if (memcmp(name
, p
->name
, len
))
1118 if (ce_stage(p
) != stage
)
1120 if (p
->name
[len
] != '/')
1122 if (p
->ce_flags
& CE_REMOVE
)
1127 remove_index_entry_at(istate
, --pos
);
1134 * Like strcmp(), but also return the offset of the first change.
1135 * If strings are equal, return the length.
1137 int strcmp_offset(const char *s1
, const char *s2
, size_t *first_change
)
1142 return strcmp(s1
, s2
);
1144 for (k
= 0; s1
[k
] == s2
[k
]; k
++)
1149 return (unsigned char)s1
[k
] - (unsigned char)s2
[k
];
1153 * Do we have another file with a pathname that is a proper
1154 * subset of the name we're trying to add?
1156 * That is, is there another file in the index with a path
1157 * that matches a sub-directory in the given entry?
1159 static int has_dir_name(struct index_state
*istate
,
1160 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
1163 int stage
= ce_stage(ce
);
1164 const char *name
= ce
->name
;
1165 const char *slash
= name
+ ce_namelen(ce
);
1170 * We are frequently called during an iteration on a sorted
1171 * list of pathnames and while building a new index. Therefore,
1172 * there is a high probability that this entry will eventually
1173 * be appended to the index, rather than inserted in the middle.
1174 * If we can confirm that, we can avoid binary searches on the
1175 * components of the pathname.
1177 * Compare the entry's full path with the last path in the index.
1179 if (istate
->cache_nr
> 0) {
1180 cmp_last
= strcmp_offset(name
,
1181 istate
->cache
[istate
->cache_nr
- 1]->name
,
1184 if (len_eq_last
== 0) {
1186 * The entry sorts AFTER the last one in the
1187 * index and their paths have no common prefix,
1188 * so there cannot be a F/D conflict.
1193 * The entry sorts AFTER the last one in the
1194 * index, but has a common prefix. Fall through
1195 * to the loop below to disect the entry's path
1196 * and see where the difference is.
1199 } else if (cmp_last
== 0) {
1201 * The entry exactly matches the last one in the
1202 * index, but because of multiple stage and CE_REMOVE
1203 * items, we fall through and let the regular search
1213 if (*--slash
== '/')
1215 if (slash
<= ce
->name
)
1222 * (len + 1) is a directory boundary (including
1223 * the trailing slash). And since the loop is
1224 * decrementing "slash", the first iteration is
1225 * the longest directory prefix; subsequent
1226 * iterations consider parent directories.
1229 if (len
+ 1 <= len_eq_last
) {
1231 * The directory prefix (including the trailing
1232 * slash) also appears as a prefix in the last
1233 * entry, so the remainder cannot collide (because
1234 * strcmp said the whole path was greater).
1239 * LT: last: xxx/file_A
1245 if (len
> len_eq_last
) {
1247 * This part of the directory prefix (excluding
1248 * the trailing slash) is longer than the known
1249 * equal portions, so this sub-directory cannot
1250 * collide with a file.
1259 * This is a possible collision. Fall through and
1260 * let the regular search code handle it.
1267 pos
= index_name_stage_pos(istate
, name
, len
, stage
, EXPAND_SPARSE
);
1270 * Found one, but not so fast. This could
1271 * be a marker that says "I was here, but
1272 * I am being removed". Such an entry is
1273 * not a part of the resulting tree, and
1274 * it is Ok to have a directory at the same
1277 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
1281 remove_index_entry_at(istate
, pos
);
1289 * Trivial optimization: if we find an entry that
1290 * already matches the sub-directory, then we know
1291 * we're ok, and we can exit.
1293 while (pos
< istate
->cache_nr
) {
1294 struct cache_entry
*p
= istate
->cache
[pos
];
1295 if ((ce_namelen(p
) <= len
) ||
1296 (p
->name
[len
] != '/') ||
1297 memcmp(p
->name
, name
, len
))
1298 break; /* not our subdirectory */
1299 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
1301 * p is at the same stage as our entry, and
1302 * is a subdirectory of what we are looking
1303 * at, so we cannot have conflicts at our
1304 * level or anything shorter.
1313 /* We may be in a situation where we already have path/file and path
1314 * is being added, or we already have path and path/file is being
1315 * added. Either one would result in a nonsense tree that has path
1316 * twice when git-write-tree tries to write it out. Prevent it.
1318 * If ok-to-replace is specified, we remove the conflicting entries
1319 * from the cache so the caller should recompute the insert position.
1320 * When this happens, we return non-zero.
1322 static int check_file_directory_conflict(struct index_state
*istate
,
1323 const struct cache_entry
*ce
,
1324 int pos
, int ok_to_replace
)
1329 * When ce is an "I am going away" entry, we allow it to be added
1331 if (ce
->ce_flags
& CE_REMOVE
)
1335 * We check if the path is a sub-path of a subsequent pathname
1336 * first, since removing those will not change the position
1339 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
1342 * Then check if the path might have a clashing sub-directory
1345 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
1348 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1351 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
1352 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
1353 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
1354 int new_only
= option
& ADD_CACHE_NEW_ONLY
;
1357 * If this entry's path sorts after the last entry in the index,
1358 * we can avoid searching for it.
1360 if (istate
->cache_nr
> 0 &&
1361 strcmp(ce
->name
, istate
->cache
[istate
->cache_nr
- 1]->name
) > 0)
1362 pos
= index_pos_to_insert_pos(istate
->cache_nr
);
1364 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
), EXPAND_SPARSE
);
1367 * Cache tree path should be invalidated only after index_name_stage_pos,
1368 * in case it expands a sparse index.
1370 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1371 cache_tree_invalidate_path(istate
, ce
->name
);
1373 /* existing match? Just replace it. */
1376 replace_index_entry(istate
, pos
, ce
);
1381 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1382 untracked_cache_add_to_index(istate
, ce
->name
);
1385 * Inserting a merged entry ("stage 0") into the index
1386 * will always replace all non-merged entries..
1388 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
1389 while (ce_same_name(istate
->cache
[pos
], ce
)) {
1391 if (!remove_index_entry_at(istate
, pos
))
1398 if (verify_path_internal(ce
->name
, ce
->ce_mode
) == PATH_INVALID
)
1399 return error(_("invalid path '%s'"), ce
->name
);
1401 if (!skip_df_check
&&
1402 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
1404 return error(_("'%s' appears as both a file and as a directory"),
1406 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
), EXPAND_SPARSE
);
1412 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1416 if (option
& ADD_CACHE_JUST_APPEND
)
1417 pos
= istate
->cache_nr
;
1420 ret
= add_index_entry_with_check(istate
, ce
, option
);
1426 /* Make sure the array is big enough .. */
1427 ALLOC_GROW(istate
->cache
, istate
->cache_nr
+ 1, istate
->cache_alloc
);
1431 if (istate
->cache_nr
> pos
+ 1)
1432 MOVE_ARRAY(istate
->cache
+ pos
+ 1, istate
->cache
+ pos
,
1433 istate
->cache_nr
- pos
- 1);
1434 set_index_entry(istate
, pos
, ce
);
1435 istate
->cache_changed
|= CE_ENTRY_ADDED
;
1440 * "refresh" does not calculate a new sha1 file or bring the
1441 * cache up-to-date for mode/content changes. But what it
1442 * _does_ do is to "re-match" the stat information of a file
1443 * with the cache, so that you can refresh the cache for a
1444 * file that hasn't been changed but where the stat entry is
1447 * For example, you'd want to do this after doing a "git-read-tree",
1448 * to link up the stat cache details with the proper files.
1450 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
1451 struct cache_entry
*ce
,
1452 unsigned int options
, int *err
,
1458 struct cache_entry
*updated
;
1460 int refresh
= options
& CE_MATCH_REFRESH
;
1461 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
1462 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
1463 int ignore_missing
= options
& CE_MATCH_IGNORE_MISSING
;
1464 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
1466 if (!refresh
|| ce_uptodate(ce
))
1469 if (!ignore_fsmonitor
)
1470 refresh_fsmonitor(istate
);
1472 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1473 * that the change to the work tree does not matter and told
1476 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
)) {
1477 ce_mark_uptodate(ce
);
1480 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
)) {
1481 ce_mark_uptodate(ce
);
1484 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
)) {
1485 ce_mark_uptodate(ce
);
1489 if (has_symlink_leading_path(ce
->name
, ce_namelen(ce
))) {
1499 if (lstat(ce
->name
, &st
) < 0) {
1500 if (ignore_missing
&& errno
== ENOENT
)
1507 changed
= ie_match_stat(istate
, ce
, &st
, options
);
1509 *changed_ret
= changed
;
1512 * The path is unchanged. If we were told to ignore
1513 * valid bit, then we did the actual stat check and
1514 * found that the entry is unmodified. If the entry
1515 * is not marked VALID, this is the place to mark it
1516 * valid again, under "assume unchanged" mode.
1518 if (ignore_valid
&& assume_unchanged
&&
1519 !(ce
->ce_flags
& CE_VALID
))
1520 ; /* mark this one VALID again */
1523 * We do not mark the index itself "modified"
1524 * because CE_UPTODATE flag is in-core only;
1525 * we are not going to write this change out.
1527 if (!S_ISGITLINK(ce
->ce_mode
)) {
1528 ce_mark_uptodate(ce
);
1529 mark_fsmonitor_valid(istate
, ce
);
1537 if (ie_modified(istate
, ce
, &st
, options
)) {
1543 updated
= make_empty_cache_entry(istate
, ce_namelen(ce
));
1544 copy_cache_entry(updated
, ce
);
1545 memcpy(updated
->name
, ce
->name
, ce
->ce_namelen
+ 1);
1546 fill_stat_cache_info(istate
, updated
, &st
);
1548 * If ignore_valid is not set, we should leave CE_VALID bit
1549 * alone. Otherwise, paths marked with --no-assume-unchanged
1550 * (i.e. things to be edited) will reacquire CE_VALID bit
1551 * automatically, which is not really what we want.
1553 if (!ignore_valid
&& assume_unchanged
&&
1554 !(ce
->ce_flags
& CE_VALID
))
1555 updated
->ce_flags
&= ~CE_VALID
;
1557 /* istate->cache_changed is updated in the caller */
1561 static void show_file(const char * fmt
, const char * name
, int in_porcelain
,
1562 int * first
, const char *header_msg
)
1564 if (in_porcelain
&& *first
&& header_msg
) {
1565 printf("%s\n", header_msg
);
1571 int repo_refresh_and_write_index(struct repository
*repo
,
1572 unsigned int refresh_flags
,
1573 unsigned int write_flags
,
1575 const struct pathspec
*pathspec
,
1576 char *seen
, const char *header_msg
)
1578 struct lock_file lock_file
= LOCK_INIT
;
1581 fd
= repo_hold_locked_index(repo
, &lock_file
, 0);
1582 if (!gentle
&& fd
< 0)
1584 if (refresh_index(repo
->index
, refresh_flags
, pathspec
, seen
, header_msg
))
1586 if (0 <= fd
&& write_locked_index(repo
->index
, &lock_file
, COMMIT_LOCK
| write_flags
))
1592 int refresh_index(struct index_state
*istate
, unsigned int flags
,
1593 const struct pathspec
*pathspec
,
1594 char *seen
, const char *header_msg
)
1598 int really
= (flags
& REFRESH_REALLY
) != 0;
1599 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
1600 int quiet
= (flags
& REFRESH_QUIET
) != 0;
1601 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
1602 int ignore_submodules
= (flags
& REFRESH_IGNORE_SUBMODULES
) != 0;
1603 int ignore_skip_worktree
= (flags
& REFRESH_IGNORE_SKIP_WORKTREE
) != 0;
1605 int in_porcelain
= (flags
& REFRESH_IN_PORCELAIN
);
1606 unsigned int options
= (CE_MATCH_REFRESH
|
1607 (really
? CE_MATCH_IGNORE_VALID
: 0) |
1608 (not_new
? CE_MATCH_IGNORE_MISSING
: 0));
1609 const char *modified_fmt
;
1610 const char *deleted_fmt
;
1611 const char *typechange_fmt
;
1612 const char *added_fmt
;
1613 const char *unmerged_fmt
;
1614 struct progress
*progress
= NULL
;
1615 int t2_sum_lstat
= 0;
1616 int t2_sum_scan
= 0;
1618 if (flags
& REFRESH_PROGRESS
&& isatty(2))
1619 progress
= start_delayed_progress(_("Refresh index"),
1622 trace_performance_enter();
1623 modified_fmt
= in_porcelain
? "M\t%s\n" : "%s: needs update\n";
1624 deleted_fmt
= in_porcelain
? "D\t%s\n" : "%s: needs update\n";
1625 typechange_fmt
= in_porcelain
? "T\t%s\n" : "%s: needs update\n";
1626 added_fmt
= in_porcelain
? "A\t%s\n" : "%s: needs update\n";
1627 unmerged_fmt
= in_porcelain
? "U\t%s\n" : "%s: needs merge\n";
1629 * Use the multi-threaded preload_index() to refresh most of the
1630 * cache entries quickly then in the single threaded loop below,
1631 * we only have to do the special cases that are left.
1633 preload_index(istate
, pathspec
, 0);
1634 trace2_region_enter("index", "refresh", NULL
);
1636 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1637 struct cache_entry
*ce
, *new_entry
;
1638 int cache_errno
= 0;
1641 int t2_did_lstat
= 0;
1642 int t2_did_scan
= 0;
1644 ce
= istate
->cache
[i
];
1645 if (ignore_submodules
&& S_ISGITLINK(ce
->ce_mode
))
1647 if (ignore_skip_worktree
&& ce_skip_worktree(ce
))
1651 * If this entry is a sparse directory, then there isn't
1652 * any stat() information to update. Ignore the entry.
1654 if (S_ISSPARSEDIR(ce
->ce_mode
))
1657 if (pathspec
&& !ce_path_match(istate
, ce
, pathspec
, seen
))
1661 while ((i
< istate
->cache_nr
) &&
1662 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
1668 show_file(unmerged_fmt
, ce
->name
, in_porcelain
,
1669 &first
, header_msg
);
1677 new_entry
= refresh_cache_ent(istate
, ce
, options
,
1678 &cache_errno
, &changed
,
1679 &t2_did_lstat
, &t2_did_scan
);
1680 t2_sum_lstat
+= t2_did_lstat
;
1681 t2_sum_scan
+= t2_did_scan
;
1682 if (new_entry
== ce
)
1684 display_progress(progress
, i
);
1688 if (really
&& cache_errno
== EINVAL
) {
1689 /* If we are doing --really-refresh that
1690 * means the index is not valid anymore.
1692 ce
->ce_flags
&= ~CE_VALID
;
1693 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
1694 mark_fsmonitor_invalid(istate
, ce
);
1695 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
1700 if (cache_errno
== ENOENT
)
1702 else if (ce_intent_to_add(ce
))
1703 fmt
= added_fmt
; /* must be before other checks */
1704 else if (changed
& TYPE_CHANGED
)
1705 fmt
= typechange_fmt
;
1709 ce
->name
, in_porcelain
, &first
, header_msg
);
1714 replace_index_entry(istate
, i
, new_entry
);
1716 trace2_data_intmax("index", NULL
, "refresh/sum_lstat", t2_sum_lstat
);
1717 trace2_data_intmax("index", NULL
, "refresh/sum_scan", t2_sum_scan
);
1718 trace2_region_leave("index", "refresh", NULL
);
1719 display_progress(progress
, istate
->cache_nr
);
1720 stop_progress(&progress
);
1721 trace_performance_leave("refresh index");
1725 struct cache_entry
*refresh_cache_entry(struct index_state
*istate
,
1726 struct cache_entry
*ce
,
1727 unsigned int options
)
1729 return refresh_cache_ent(istate
, ce
, options
, NULL
, NULL
, NULL
, NULL
);
1733 /*****************************************************************
1735 *****************************************************************/
1737 #define INDEX_FORMAT_DEFAULT 3
1739 static unsigned int get_index_format_default(struct repository
*r
)
1741 char *envversion
= getenv("GIT_INDEX_VERSION");
1743 unsigned int version
= INDEX_FORMAT_DEFAULT
;
1746 prepare_repo_settings(r
);
1748 if (r
->settings
.index_version
>= 0)
1749 version
= r
->settings
.index_version
;
1750 if (version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1751 warning(_("index.version set, but the value is invalid.\n"
1752 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1753 return INDEX_FORMAT_DEFAULT
;
1758 version
= strtoul(envversion
, &endp
, 10);
1760 version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1761 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1762 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1763 version
= INDEX_FORMAT_DEFAULT
;
1769 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1770 * Again - this is just a (very strong in practice) heuristic that
1771 * the inode hasn't changed.
1773 * We save the fields in big-endian order to allow using the
1774 * index file over NFS transparently.
1776 struct ondisk_cache_entry
{
1777 struct cache_time ctime
;
1778 struct cache_time mtime
;
1786 * unsigned char hash[hashsz];
1788 * if (flags & CE_EXTENDED)
1791 unsigned char data
[GIT_MAX_RAWSZ
+ 2 * sizeof(uint16_t)];
1792 char name
[FLEX_ARRAY
];
1795 /* These are only used for v3 or lower */
1796 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1797 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,data) + (len) + 8) & ~7)
1798 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1799 #define ondisk_data_size(flags, len) (the_hash_algo->rawsz + \
1800 ((flags & CE_EXTENDED) ? 2 : 1) * sizeof(uint16_t) + len)
1801 #define ondisk_data_size_max(len) (ondisk_data_size(CE_EXTENDED, len))
1802 #define ondisk_ce_size(ce) (ondisk_cache_entry_size(ondisk_data_size((ce)->ce_flags, ce_namelen(ce))))
1804 /* Allow fsck to force verification of the index checksum. */
1805 int verify_index_checksum
;
1807 /* Allow fsck to force verification of the cache entry order. */
1808 int verify_ce_order
;
1810 static int verify_hdr(const struct cache_header
*hdr
, unsigned long size
)
1813 unsigned char hash
[GIT_MAX_RAWSZ
];
1816 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
1817 return error(_("bad signature 0x%08x"), hdr
->hdr_signature
);
1818 hdr_version
= ntohl(hdr
->hdr_version
);
1819 if (hdr_version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< hdr_version
)
1820 return error(_("bad index version %d"), hdr_version
);
1822 if (!verify_index_checksum
)
1825 the_hash_algo
->init_fn(&c
);
1826 the_hash_algo
->update_fn(&c
, hdr
, size
- the_hash_algo
->rawsz
);
1827 the_hash_algo
->final_fn(hash
, &c
);
1828 if (!hasheq(hash
, (unsigned char *)hdr
+ size
- the_hash_algo
->rawsz
))
1829 return error(_("bad index file sha1 signature"));
1833 static int read_index_extension(struct index_state
*istate
,
1834 const char *ext
, const char *data
, unsigned long sz
)
1836 switch (CACHE_EXT(ext
)) {
1837 case CACHE_EXT_TREE
:
1838 istate
->cache_tree
= cache_tree_read(data
, sz
);
1840 case CACHE_EXT_RESOLVE_UNDO
:
1841 istate
->resolve_undo
= resolve_undo_read(data
, sz
);
1843 case CACHE_EXT_LINK
:
1844 if (read_link_extension(istate
, data
, sz
))
1847 case CACHE_EXT_UNTRACKED
:
1848 istate
->untracked
= read_untracked_extension(data
, sz
);
1850 case CACHE_EXT_FSMONITOR
:
1851 read_fsmonitor_extension(istate
, data
, sz
);
1853 case CACHE_EXT_ENDOFINDEXENTRIES
:
1854 case CACHE_EXT_INDEXENTRYOFFSETTABLE
:
1855 /* already handled in do_read_index() */
1857 case CACHE_EXT_SPARSE_DIRECTORIES
:
1858 /* no content, only an indicator */
1859 istate
->sparse_index
= INDEX_COLLAPSED
;
1862 if (*ext
< 'A' || 'Z' < *ext
)
1863 return error(_("index uses %.4s extension, which we do not understand"),
1865 fprintf_ln(stderr
, _("ignoring %.4s extension"), ext
);
1871 static struct cache_entry
*create_from_disk(struct mem_pool
*ce_mem_pool
,
1872 unsigned int version
,
1873 struct ondisk_cache_entry
*ondisk
,
1874 unsigned long *ent_size
,
1875 const struct cache_entry
*previous_ce
)
1877 struct cache_entry
*ce
;
1880 const unsigned hashsz
= the_hash_algo
->rawsz
;
1881 const uint16_t *flagsp
= (const uint16_t *)(ondisk
->data
+ hashsz
);
1883 size_t copy_len
= 0;
1885 * Adjacent cache entries tend to share the leading paths, so it makes
1886 * sense to only store the differences in later entries. In the v4
1887 * on-disk format of the index, each on-disk cache entry stores the
1888 * number of bytes to be stripped from the end of the previous name,
1889 * and the bytes to append to the result, to come up with its name.
1891 int expand_name_field
= version
== 4;
1893 /* On-disk flags are just 16 bits */
1894 flags
= get_be16(flagsp
);
1895 len
= flags
& CE_NAMEMASK
;
1897 if (flags
& CE_EXTENDED
) {
1899 extended_flags
= get_be16(flagsp
+ 1) << 16;
1900 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1901 if (extended_flags
& ~CE_EXTENDED_FLAGS
)
1902 die(_("unknown index entry format 0x%08x"), extended_flags
);
1903 flags
|= extended_flags
;
1904 name
= (const char *)(flagsp
+ 2);
1907 name
= (const char *)(flagsp
+ 1);
1909 if (expand_name_field
) {
1910 const unsigned char *cp
= (const unsigned char *)name
;
1911 size_t strip_len
, previous_len
;
1913 /* If we're at the beginning of a block, ignore the previous name */
1914 strip_len
= decode_varint(&cp
);
1916 previous_len
= previous_ce
->ce_namelen
;
1917 if (previous_len
< strip_len
)
1918 die(_("malformed name field in the index, near path '%s'"),
1920 copy_len
= previous_len
- strip_len
;
1922 name
= (const char *)cp
;
1925 if (len
== CE_NAMEMASK
) {
1927 if (expand_name_field
)
1931 ce
= mem_pool__ce_alloc(ce_mem_pool
, len
);
1933 ce
->ce_stat_data
.sd_ctime
.sec
= get_be32(&ondisk
->ctime
.sec
);
1934 ce
->ce_stat_data
.sd_mtime
.sec
= get_be32(&ondisk
->mtime
.sec
);
1935 ce
->ce_stat_data
.sd_ctime
.nsec
= get_be32(&ondisk
->ctime
.nsec
);
1936 ce
->ce_stat_data
.sd_mtime
.nsec
= get_be32(&ondisk
->mtime
.nsec
);
1937 ce
->ce_stat_data
.sd_dev
= get_be32(&ondisk
->dev
);
1938 ce
->ce_stat_data
.sd_ino
= get_be32(&ondisk
->ino
);
1939 ce
->ce_mode
= get_be32(&ondisk
->mode
);
1940 ce
->ce_stat_data
.sd_uid
= get_be32(&ondisk
->uid
);
1941 ce
->ce_stat_data
.sd_gid
= get_be32(&ondisk
->gid
);
1942 ce
->ce_stat_data
.sd_size
= get_be32(&ondisk
->size
);
1943 ce
->ce_flags
= flags
& ~CE_NAMEMASK
;
1944 ce
->ce_namelen
= len
;
1946 oidread(&ce
->oid
, ondisk
->data
);
1948 if (expand_name_field
) {
1950 memcpy(ce
->name
, previous_ce
->name
, copy_len
);
1951 memcpy(ce
->name
+ copy_len
, name
, len
+ 1 - copy_len
);
1952 *ent_size
= (name
- ((char *)ondisk
)) + len
+ 1 - copy_len
;
1954 memcpy(ce
->name
, name
, len
+ 1);
1955 *ent_size
= ondisk_ce_size(ce
);
1960 static void check_ce_order(struct index_state
*istate
)
1964 if (!verify_ce_order
)
1967 for (i
= 1; i
< istate
->cache_nr
; i
++) {
1968 struct cache_entry
*ce
= istate
->cache
[i
- 1];
1969 struct cache_entry
*next_ce
= istate
->cache
[i
];
1970 int name_compare
= strcmp(ce
->name
, next_ce
->name
);
1972 if (0 < name_compare
)
1973 die(_("unordered stage entries in index"));
1974 if (!name_compare
) {
1976 die(_("multiple stage entries for merged file '%s'"),
1978 if (ce_stage(ce
) > ce_stage(next_ce
))
1979 die(_("unordered stage entries for '%s'"),
1985 static void tweak_untracked_cache(struct index_state
*istate
)
1987 struct repository
*r
= the_repository
;
1989 prepare_repo_settings(r
);
1991 switch (r
->settings
.core_untracked_cache
) {
1992 case UNTRACKED_CACHE_REMOVE
:
1993 remove_untracked_cache(istate
);
1995 case UNTRACKED_CACHE_WRITE
:
1996 add_untracked_cache(istate
);
1998 case UNTRACKED_CACHE_KEEP
:
2000 * Either an explicit "core.untrackedCache=keep", the
2001 * default if "core.untrackedCache" isn't configured,
2002 * or a fallback on an unknown "core.untrackedCache"
2009 static void tweak_split_index(struct index_state
*istate
)
2011 switch (git_config_get_split_index()) {
2012 case -1: /* unset: do nothing */
2015 remove_split_index(istate
);
2018 add_split_index(istate
);
2020 default: /* unknown value: do nothing */
2025 static void post_read_index_from(struct index_state
*istate
)
2027 check_ce_order(istate
);
2028 tweak_untracked_cache(istate
);
2029 tweak_split_index(istate
);
2030 tweak_fsmonitor(istate
);
2033 static size_t estimate_cache_size_from_compressed(unsigned int entries
)
2035 return entries
* (sizeof(struct cache_entry
) + CACHE_ENTRY_PATH_LENGTH
);
2038 static size_t estimate_cache_size(size_t ondisk_size
, unsigned int entries
)
2040 long per_entry
= sizeof(struct cache_entry
) - sizeof(struct ondisk_cache_entry
);
2043 * Account for potential alignment differences.
2045 per_entry
+= align_padding_size(per_entry
, 0);
2046 return ondisk_size
+ entries
* per_entry
;
2049 struct index_entry_offset
2051 /* starting byte offset into index file, count of index entries in this block */
2055 struct index_entry_offset_table
2058 struct index_entry_offset entries
[FLEX_ARRAY
];
2061 static struct index_entry_offset_table
*read_ieot_extension(const char *mmap
, size_t mmap_size
, size_t offset
);
2062 static void write_ieot_extension(struct strbuf
*sb
, struct index_entry_offset_table
*ieot
);
2064 static size_t read_eoie_extension(const char *mmap
, size_t mmap_size
);
2065 static void write_eoie_extension(struct strbuf
*sb
, git_hash_ctx
*eoie_context
, size_t offset
);
2067 struct load_index_extensions
2070 struct index_state
*istate
;
2073 unsigned long src_offset
;
2076 static void *load_index_extensions(void *_data
)
2078 struct load_index_extensions
*p
= _data
;
2079 unsigned long src_offset
= p
->src_offset
;
2081 while (src_offset
<= p
->mmap_size
- the_hash_algo
->rawsz
- 8) {
2082 /* After an array of active_nr index entries,
2083 * there can be arbitrary number of extended
2084 * sections, each of which is prefixed with
2085 * extension name (4-byte) and section length
2086 * in 4-byte network byte order.
2088 uint32_t extsize
= get_be32(p
->mmap
+ src_offset
+ 4);
2089 if (read_index_extension(p
->istate
,
2090 p
->mmap
+ src_offset
,
2091 p
->mmap
+ src_offset
+ 8,
2093 munmap((void *)p
->mmap
, p
->mmap_size
);
2094 die(_("index file corrupt"));
2097 src_offset
+= extsize
;
2104 * A helper function that will load the specified range of cache entries
2105 * from the memory mapped file and add them to the given index.
2107 static unsigned long load_cache_entry_block(struct index_state
*istate
,
2108 struct mem_pool
*ce_mem_pool
, int offset
, int nr
, const char *mmap
,
2109 unsigned long start_offset
, const struct cache_entry
*previous_ce
)
2112 unsigned long src_offset
= start_offset
;
2114 for (i
= offset
; i
< offset
+ nr
; i
++) {
2115 struct ondisk_cache_entry
*disk_ce
;
2116 struct cache_entry
*ce
;
2117 unsigned long consumed
;
2119 disk_ce
= (struct ondisk_cache_entry
*)(mmap
+ src_offset
);
2120 ce
= create_from_disk(ce_mem_pool
, istate
->version
, disk_ce
, &consumed
, previous_ce
);
2121 set_index_entry(istate
, i
, ce
);
2123 src_offset
+= consumed
;
2126 return src_offset
- start_offset
;
2129 static unsigned long load_all_cache_entries(struct index_state
*istate
,
2130 const char *mmap
, size_t mmap_size
, unsigned long src_offset
)
2132 unsigned long consumed
;
2134 istate
->ce_mem_pool
= xmalloc(sizeof(*istate
->ce_mem_pool
));
2135 if (istate
->version
== 4) {
2136 mem_pool_init(istate
->ce_mem_pool
,
2137 estimate_cache_size_from_compressed(istate
->cache_nr
));
2139 mem_pool_init(istate
->ce_mem_pool
,
2140 estimate_cache_size(mmap_size
, istate
->cache_nr
));
2143 consumed
= load_cache_entry_block(istate
, istate
->ce_mem_pool
,
2144 0, istate
->cache_nr
, mmap
, src_offset
, NULL
);
2149 * Mostly randomly chosen maximum thread counts: we
2150 * cap the parallelism to online_cpus() threads, and we want
2151 * to have at least 10000 cache entries per thread for it to
2152 * be worth starting a thread.
2155 #define THREAD_COST (10000)
2157 struct load_cache_entries_thread_data
2160 struct index_state
*istate
;
2161 struct mem_pool
*ce_mem_pool
;
2164 struct index_entry_offset_table
*ieot
;
2165 int ieot_start
; /* starting index into the ieot array */
2166 int ieot_blocks
; /* count of ieot entries to process */
2167 unsigned long consumed
; /* return # of bytes in index file processed */
2171 * A thread proc to run the load_cache_entries() computation
2172 * across multiple background threads.
2174 static void *load_cache_entries_thread(void *_data
)
2176 struct load_cache_entries_thread_data
*p
= _data
;
2179 /* iterate across all ieot blocks assigned to this thread */
2180 for (i
= p
->ieot_start
; i
< p
->ieot_start
+ p
->ieot_blocks
; i
++) {
2181 p
->consumed
+= load_cache_entry_block(p
->istate
, p
->ce_mem_pool
,
2182 p
->offset
, p
->ieot
->entries
[i
].nr
, p
->mmap
, p
->ieot
->entries
[i
].offset
, NULL
);
2183 p
->offset
+= p
->ieot
->entries
[i
].nr
;
2188 static unsigned long load_cache_entries_threaded(struct index_state
*istate
, const char *mmap
, size_t mmap_size
,
2189 int nr_threads
, struct index_entry_offset_table
*ieot
)
2191 int i
, offset
, ieot_blocks
, ieot_start
, err
;
2192 struct load_cache_entries_thread_data
*data
;
2193 unsigned long consumed
= 0;
2195 /* a little sanity checking */
2196 if (istate
->name_hash_initialized
)
2197 BUG("the name hash isn't thread safe");
2199 istate
->ce_mem_pool
= xmalloc(sizeof(*istate
->ce_mem_pool
));
2200 mem_pool_init(istate
->ce_mem_pool
, 0);
2202 /* ensure we have no more threads than we have blocks to process */
2203 if (nr_threads
> ieot
->nr
)
2204 nr_threads
= ieot
->nr
;
2205 CALLOC_ARRAY(data
, nr_threads
);
2207 offset
= ieot_start
= 0;
2208 ieot_blocks
= DIV_ROUND_UP(ieot
->nr
, nr_threads
);
2209 for (i
= 0; i
< nr_threads
; i
++) {
2210 struct load_cache_entries_thread_data
*p
= &data
[i
];
2213 if (ieot_start
+ ieot_blocks
> ieot
->nr
)
2214 ieot_blocks
= ieot
->nr
- ieot_start
;
2220 p
->ieot_start
= ieot_start
;
2221 p
->ieot_blocks
= ieot_blocks
;
2223 /* create a mem_pool for each thread */
2225 for (j
= p
->ieot_start
; j
< p
->ieot_start
+ p
->ieot_blocks
; j
++)
2226 nr
+= p
->ieot
->entries
[j
].nr
;
2227 p
->ce_mem_pool
= xmalloc(sizeof(*istate
->ce_mem_pool
));
2228 if (istate
->version
== 4) {
2229 mem_pool_init(p
->ce_mem_pool
,
2230 estimate_cache_size_from_compressed(nr
));
2232 mem_pool_init(p
->ce_mem_pool
,
2233 estimate_cache_size(mmap_size
, nr
));
2236 err
= pthread_create(&p
->pthread
, NULL
, load_cache_entries_thread
, p
);
2238 die(_("unable to create load_cache_entries thread: %s"), strerror(err
));
2240 /* increment by the number of cache entries in the ieot block being processed */
2241 for (j
= 0; j
< ieot_blocks
; j
++)
2242 offset
+= ieot
->entries
[ieot_start
+ j
].nr
;
2243 ieot_start
+= ieot_blocks
;
2246 for (i
= 0; i
< nr_threads
; i
++) {
2247 struct load_cache_entries_thread_data
*p
= &data
[i
];
2249 err
= pthread_join(p
->pthread
, NULL
);
2251 die(_("unable to join load_cache_entries thread: %s"), strerror(err
));
2252 mem_pool_combine(istate
->ce_mem_pool
, p
->ce_mem_pool
);
2253 consumed
+= p
->consumed
;
2261 static void set_new_index_sparsity(struct index_state
*istate
)
2264 * If the index's repo exists, mark it sparse according to
2268 prepare_repo_settings(istate
->repo
);
2269 if (!istate
->repo
->settings
.command_requires_full_index
&&
2270 is_sparse_index_allowed(istate
, 0))
2271 istate
->sparse_index
= 1;
2275 /* remember to discard_cache() before reading a different cache! */
2276 int do_read_index(struct index_state
*istate
, const char *path
, int must_exist
)
2280 unsigned long src_offset
;
2281 const struct cache_header
*hdr
;
2284 struct load_index_extensions p
;
2285 size_t extension_offset
= 0;
2286 int nr_threads
, cpus
;
2287 struct index_entry_offset_table
*ieot
= NULL
;
2289 if (istate
->initialized
)
2290 return istate
->cache_nr
;
2292 istate
->timestamp
.sec
= 0;
2293 istate
->timestamp
.nsec
= 0;
2294 fd
= open(path
, O_RDONLY
);
2296 if (!must_exist
&& errno
== ENOENT
) {
2297 set_new_index_sparsity(istate
);
2300 die_errno(_("%s: index file open failed"), path
);
2304 die_errno(_("%s: cannot stat the open index"), path
);
2306 mmap_size
= xsize_t(st
.st_size
);
2307 if (mmap_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2308 die(_("%s: index file smaller than expected"), path
);
2310 mmap
= xmmap_gently(NULL
, mmap_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2311 if (mmap
== MAP_FAILED
)
2312 die_errno(_("%s: unable to map index file%s"), path
,
2316 hdr
= (const struct cache_header
*)mmap
;
2317 if (verify_hdr(hdr
, mmap_size
) < 0)
2320 oidread(&istate
->oid
, (const unsigned char *)hdr
+ mmap_size
- the_hash_algo
->rawsz
);
2321 istate
->version
= ntohl(hdr
->hdr_version
);
2322 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
2323 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
2324 CALLOC_ARRAY(istate
->cache
, istate
->cache_alloc
);
2325 istate
->initialized
= 1;
2329 p
.mmap_size
= mmap_size
;
2331 src_offset
= sizeof(*hdr
);
2333 if (git_config_get_index_threads(&nr_threads
))
2336 /* TODO: does creating more threads than cores help? */
2338 nr_threads
= istate
->cache_nr
/ THREAD_COST
;
2339 cpus
= online_cpus();
2340 if (nr_threads
> cpus
)
2347 if (nr_threads
> 1) {
2348 extension_offset
= read_eoie_extension(mmap
, mmap_size
);
2349 if (extension_offset
) {
2352 p
.src_offset
= extension_offset
;
2353 err
= pthread_create(&p
.pthread
, NULL
, load_index_extensions
, &p
);
2355 die(_("unable to create load_index_extensions thread: %s"), strerror(err
));
2362 * Locate and read the index entry offset table so that we can use it
2363 * to multi-thread the reading of the cache entries.
2365 if (extension_offset
&& nr_threads
> 1)
2366 ieot
= read_ieot_extension(mmap
, mmap_size
, extension_offset
);
2369 src_offset
+= load_cache_entries_threaded(istate
, mmap
, mmap_size
, nr_threads
, ieot
);
2372 src_offset
+= load_all_cache_entries(istate
, mmap
, mmap_size
, src_offset
);
2375 istate
->timestamp
.sec
= st
.st_mtime
;
2376 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
2378 /* if we created a thread, join it otherwise load the extensions on the primary thread */
2379 if (extension_offset
) {
2380 int ret
= pthread_join(p
.pthread
, NULL
);
2382 die(_("unable to join load_index_extensions thread: %s"), strerror(ret
));
2384 p
.src_offset
= src_offset
;
2385 load_index_extensions(&p
);
2387 munmap((void *)mmap
, mmap_size
);
2390 * TODO trace2: replace "the_repository" with the actual repo instance
2391 * that is associated with the given "istate".
2393 trace2_data_intmax("index", the_repository
, "read/version",
2395 trace2_data_intmax("index", the_repository
, "read/cache_nr",
2399 istate
->repo
= the_repository
;
2402 * If the command explicitly requires a full index, force it
2403 * to be full. Otherwise, correct the sparsity based on repository
2404 * settings and other properties of the index (if necessary).
2406 prepare_repo_settings(istate
->repo
);
2407 if (istate
->repo
->settings
.command_requires_full_index
)
2408 ensure_full_index(istate
);
2410 ensure_correct_sparsity(istate
);
2412 return istate
->cache_nr
;
2415 munmap((void *)mmap
, mmap_size
);
2416 die(_("index file corrupt"));
2420 * Signal that the shared index is used by updating its mtime.
2422 * This way, shared index can be removed if they have not been used
2425 static void freshen_shared_index(const char *shared_index
, int warn
)
2427 if (!check_and_freshen_file(shared_index
, 1) && warn
)
2428 warning(_("could not freshen shared index '%s'"), shared_index
);
2431 int read_index_from(struct index_state
*istate
, const char *path
,
2434 struct split_index
*split_index
;
2439 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2440 if (istate
->initialized
)
2441 return istate
->cache_nr
;
2444 * TODO trace2: replace "the_repository" with the actual repo instance
2445 * that is associated with the given "istate".
2447 trace2_region_enter_printf("index", "do_read_index", the_repository
,
2449 trace_performance_enter();
2450 ret
= do_read_index(istate
, path
, 0);
2451 trace_performance_leave("read cache %s", path
);
2452 trace2_region_leave_printf("index", "do_read_index", the_repository
,
2455 split_index
= istate
->split_index
;
2456 if (!split_index
|| is_null_oid(&split_index
->base_oid
)) {
2457 post_read_index_from(istate
);
2461 trace_performance_enter();
2462 if (split_index
->base
)
2463 discard_index(split_index
->base
);
2465 CALLOC_ARRAY(split_index
->base
, 1);
2467 base_oid_hex
= oid_to_hex(&split_index
->base_oid
);
2468 base_path
= xstrfmt("%s/sharedindex.%s", gitdir
, base_oid_hex
);
2469 trace2_region_enter_printf("index", "shared/do_read_index",
2470 the_repository
, "%s", base_path
);
2471 ret
= do_read_index(split_index
->base
, base_path
, 0);
2472 trace2_region_leave_printf("index", "shared/do_read_index",
2473 the_repository
, "%s", base_path
);
2475 char *path_copy
= xstrdup(path
);
2476 char *base_path2
= xstrfmt("%s/sharedindex.%s",
2477 dirname(path_copy
), base_oid_hex
);
2479 trace2_region_enter_printf("index", "shared/do_read_index",
2480 the_repository
, "%s", base_path2
);
2481 ret
= do_read_index(split_index
->base
, base_path2
, 1);
2482 trace2_region_leave_printf("index", "shared/do_read_index",
2483 the_repository
, "%s", base_path2
);
2486 if (!oideq(&split_index
->base_oid
, &split_index
->base
->oid
))
2487 die(_("broken index, expect %s in %s, got %s"),
2488 base_oid_hex
, base_path
,
2489 oid_to_hex(&split_index
->base
->oid
));
2491 freshen_shared_index(base_path
, 0);
2492 merge_base_index(istate
);
2493 post_read_index_from(istate
);
2494 trace_performance_leave("read cache %s", base_path
);
2499 int is_index_unborn(struct index_state
*istate
)
2501 return (!istate
->cache_nr
&& !istate
->timestamp
.sec
);
2504 int discard_index(struct index_state
*istate
)
2507 * Cache entries in istate->cache[] should have been allocated
2508 * from the memory pool associated with this index, or from an
2509 * associated split_index. There is no need to free individual
2510 * cache entries. validate_cache_entries can detect when this
2511 * assertion does not hold.
2513 validate_cache_entries(istate
);
2515 resolve_undo_clear_index(istate
);
2516 istate
->cache_nr
= 0;
2517 istate
->cache_changed
= 0;
2518 istate
->timestamp
.sec
= 0;
2519 istate
->timestamp
.nsec
= 0;
2520 free_name_hash(istate
);
2521 cache_tree_free(&(istate
->cache_tree
));
2522 istate
->initialized
= 0;
2523 istate
->fsmonitor_has_run_once
= 0;
2524 FREE_AND_NULL(istate
->fsmonitor_last_update
);
2525 FREE_AND_NULL(istate
->cache
);
2526 istate
->cache_alloc
= 0;
2527 discard_split_index(istate
);
2528 free_untracked_cache(istate
->untracked
);
2529 istate
->untracked
= NULL
;
2531 if (istate
->ce_mem_pool
) {
2532 mem_pool_discard(istate
->ce_mem_pool
, should_validate_cache_entries());
2533 FREE_AND_NULL(istate
->ce_mem_pool
);
2540 * Validate the cache entries of this index.
2541 * All cache entries associated with this index
2542 * should have been allocated by the memory pool
2543 * associated with this index, or by a referenced
2546 void validate_cache_entries(const struct index_state
*istate
)
2550 if (!should_validate_cache_entries() ||!istate
|| !istate
->initialized
)
2553 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2555 BUG("cache entry is not allocated from expected memory pool");
2556 } else if (!istate
->ce_mem_pool
||
2557 !mem_pool_contains(istate
->ce_mem_pool
, istate
->cache
[i
])) {
2558 if (!istate
->split_index
||
2559 !istate
->split_index
->base
||
2560 !istate
->split_index
->base
->ce_mem_pool
||
2561 !mem_pool_contains(istate
->split_index
->base
->ce_mem_pool
, istate
->cache
[i
])) {
2562 BUG("cache entry is not allocated from expected memory pool");
2567 if (istate
->split_index
)
2568 validate_cache_entries(istate
->split_index
->base
);
2571 int unmerged_index(const struct index_state
*istate
)
2574 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2575 if (ce_stage(istate
->cache
[i
]))
2581 int repo_index_has_changes(struct repository
*repo
,
2585 struct index_state
*istate
= repo
->index
;
2586 struct object_id cmp
;
2590 cmp
= tree
->object
.oid
;
2591 if (tree
|| !get_oid_tree("HEAD", &cmp
)) {
2592 struct diff_options opt
;
2594 repo_diff_setup(repo
, &opt
);
2595 opt
.flags
.exit_with_status
= 1;
2597 opt
.flags
.quick
= 1;
2598 diff_setup_done(&opt
);
2599 do_diff_cache(&cmp
, &opt
);
2601 for (i
= 0; sb
&& i
< diff_queued_diff
.nr
; i
++) {
2603 strbuf_addch(sb
, ' ');
2604 strbuf_addstr(sb
, diff_queued_diff
.queue
[i
]->two
->path
);
2607 return opt
.flags
.has_changes
!= 0;
2609 /* TODO: audit for interaction with sparse-index. */
2610 ensure_full_index(istate
);
2611 for (i
= 0; sb
&& i
< istate
->cache_nr
; i
++) {
2613 strbuf_addch(sb
, ' ');
2614 strbuf_addstr(sb
, istate
->cache
[i
]->name
);
2616 return !!istate
->cache_nr
;
2620 static int write_index_ext_header(struct hashfile
*f
,
2621 git_hash_ctx
*eoie_f
,
2625 hashwrite_be32(f
, ext
);
2626 hashwrite_be32(f
, sz
);
2631 the_hash_algo
->update_fn(eoie_f
, &ext
, sizeof(ext
));
2632 the_hash_algo
->update_fn(eoie_f
, &sz
, sizeof(sz
));
2637 static void ce_smudge_racily_clean_entry(struct index_state
*istate
,
2638 struct cache_entry
*ce
)
2641 * The only thing we care about in this function is to smudge the
2642 * falsely clean entry due to touch-update-touch race, so we leave
2643 * everything else as they are. We are called for entries whose
2644 * ce_stat_data.sd_mtime match the index file mtime.
2646 * Note that this actually does not do much for gitlinks, for
2647 * which ce_match_stat_basic() always goes to the actual
2648 * contents. The caller checks with is_racy_timestamp() which
2649 * always says "no" for gitlinks, so we are not called for them ;-)
2653 if (lstat(ce
->name
, &st
) < 0)
2655 if (ce_match_stat_basic(ce
, &st
))
2657 if (ce_modified_check_fs(istate
, ce
, &st
)) {
2658 /* This is "racily clean"; smudge it. Note that this
2659 * is a tricky code. At first glance, it may appear
2660 * that it can break with this sequence:
2662 * $ echo xyzzy >frotz
2663 * $ git-update-index --add frotz
2666 * $ echo filfre >nitfol
2667 * $ git-update-index --add nitfol
2669 * but it does not. When the second update-index runs,
2670 * it notices that the entry "frotz" has the same timestamp
2671 * as index, and if we were to smudge it by resetting its
2672 * size to zero here, then the object name recorded
2673 * in index is the 6-byte file but the cached stat information
2674 * becomes zero --- which would then match what we would
2675 * obtain from the filesystem next time we stat("frotz").
2677 * However, the second update-index, before calling
2678 * this function, notices that the cached size is 6
2679 * bytes and what is on the filesystem is an empty
2680 * file, and never calls us, so the cached size information
2681 * for "frotz" stays 6 which does not match the filesystem.
2683 ce
->ce_stat_data
.sd_size
= 0;
2687 /* Copy miscellaneous fields but not the name */
2688 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry
*ondisk
,
2689 struct cache_entry
*ce
)
2692 const unsigned hashsz
= the_hash_algo
->rawsz
;
2693 uint16_t *flagsp
= (uint16_t *)(ondisk
->data
+ hashsz
);
2695 ondisk
->ctime
.sec
= htonl(ce
->ce_stat_data
.sd_ctime
.sec
);
2696 ondisk
->mtime
.sec
= htonl(ce
->ce_stat_data
.sd_mtime
.sec
);
2697 ondisk
->ctime
.nsec
= htonl(ce
->ce_stat_data
.sd_ctime
.nsec
);
2698 ondisk
->mtime
.nsec
= htonl(ce
->ce_stat_data
.sd_mtime
.nsec
);
2699 ondisk
->dev
= htonl(ce
->ce_stat_data
.sd_dev
);
2700 ondisk
->ino
= htonl(ce
->ce_stat_data
.sd_ino
);
2701 ondisk
->mode
= htonl(ce
->ce_mode
);
2702 ondisk
->uid
= htonl(ce
->ce_stat_data
.sd_uid
);
2703 ondisk
->gid
= htonl(ce
->ce_stat_data
.sd_gid
);
2704 ondisk
->size
= htonl(ce
->ce_stat_data
.sd_size
);
2705 hashcpy(ondisk
->data
, ce
->oid
.hash
);
2707 flags
= ce
->ce_flags
& ~CE_NAMEMASK
;
2708 flags
|= (ce_namelen(ce
) >= CE_NAMEMASK
? CE_NAMEMASK
: ce_namelen(ce
));
2709 flagsp
[0] = htons(flags
);
2710 if (ce
->ce_flags
& CE_EXTENDED
) {
2711 flagsp
[1] = htons((ce
->ce_flags
& CE_EXTENDED_FLAGS
) >> 16);
2715 static int ce_write_entry(struct hashfile
*f
, struct cache_entry
*ce
,
2716 struct strbuf
*previous_name
, struct ondisk_cache_entry
*ondisk
)
2719 unsigned int saved_namelen
;
2720 int stripped_name
= 0;
2721 static unsigned char padding
[8] = { 0x00 };
2723 if (ce
->ce_flags
& CE_STRIP_NAME
) {
2724 saved_namelen
= ce_namelen(ce
);
2729 size
= offsetof(struct ondisk_cache_entry
,data
) + ondisk_data_size(ce
->ce_flags
, 0);
2731 if (!previous_name
) {
2732 int len
= ce_namelen(ce
);
2733 copy_cache_entry_to_ondisk(ondisk
, ce
);
2734 hashwrite(f
, ondisk
, size
);
2735 hashwrite(f
, ce
->name
, len
);
2736 hashwrite(f
, padding
, align_padding_size(size
, len
));
2738 int common
, to_remove
, prefix_size
;
2739 unsigned char to_remove_vi
[16];
2741 (ce
->name
[common
] &&
2742 common
< previous_name
->len
&&
2743 ce
->name
[common
] == previous_name
->buf
[common
]);
2745 ; /* still matching */
2746 to_remove
= previous_name
->len
- common
;
2747 prefix_size
= encode_varint(to_remove
, to_remove_vi
);
2749 copy_cache_entry_to_ondisk(ondisk
, ce
);
2750 hashwrite(f
, ondisk
, size
);
2751 hashwrite(f
, to_remove_vi
, prefix_size
);
2752 hashwrite(f
, ce
->name
+ common
, ce_namelen(ce
) - common
);
2753 hashwrite(f
, padding
, 1);
2755 strbuf_splice(previous_name
, common
, to_remove
,
2756 ce
->name
+ common
, ce_namelen(ce
) - common
);
2758 if (stripped_name
) {
2759 ce
->ce_namelen
= saved_namelen
;
2760 ce
->ce_flags
&= ~CE_STRIP_NAME
;
2767 * This function verifies if index_state has the correct sha1 of the
2768 * index file. Don't die if we have any other failure, just return 0.
2770 static int verify_index_from(const struct index_state
*istate
, const char *path
)
2775 unsigned char hash
[GIT_MAX_RAWSZ
];
2777 if (!istate
->initialized
)
2780 fd
= open(path
, O_RDONLY
);
2787 if (st
.st_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2790 n
= pread_in_full(fd
, hash
, the_hash_algo
->rawsz
, st
.st_size
- the_hash_algo
->rawsz
);
2791 if (n
!= the_hash_algo
->rawsz
)
2794 if (!hasheq(istate
->oid
.hash
, hash
))
2805 static int repo_verify_index(struct repository
*repo
)
2807 return verify_index_from(repo
->index
, repo
->index_file
);
2810 int has_racy_timestamp(struct index_state
*istate
)
2812 int entries
= istate
->cache_nr
;
2815 for (i
= 0; i
< entries
; i
++) {
2816 struct cache_entry
*ce
= istate
->cache
[i
];
2817 if (is_racy_timestamp(istate
, ce
))
2823 void repo_update_index_if_able(struct repository
*repo
,
2824 struct lock_file
*lockfile
)
2826 if ((repo
->index
->cache_changed
||
2827 has_racy_timestamp(repo
->index
)) &&
2828 repo_verify_index(repo
))
2829 write_locked_index(repo
->index
, lockfile
, COMMIT_LOCK
);
2831 rollback_lock_file(lockfile
);
2834 static int record_eoie(void)
2838 if (!git_config_get_bool("index.recordendofindexentries", &val
))
2842 * As a convenience, the end of index entries extension
2843 * used for threading is written by default if the user
2844 * explicitly requested threaded index reads.
2846 return !git_config_get_index_threads(&val
) && val
!= 1;
2849 static int record_ieot(void)
2853 if (!git_config_get_bool("index.recordoffsettable", &val
))
2857 * As a convenience, the offset table used for threading is
2858 * written by default if the user explicitly requested
2859 * threaded index reads.
2861 return !git_config_get_index_threads(&val
) && val
!= 1;
2865 * On success, `tempfile` is closed. If it is the temporary file
2866 * of a `struct lock_file`, we will therefore effectively perform
2867 * a 'close_lock_file_gently()`. Since that is an implementation
2868 * detail of lockfiles, callers of `do_write_index()` should not
2871 static int do_write_index(struct index_state
*istate
, struct tempfile
*tempfile
,
2872 int strip_extensions
, unsigned flags
)
2874 uint64_t start
= getnanotime();
2876 git_hash_ctx
*eoie_c
= NULL
;
2877 struct cache_header hdr
;
2878 int i
, err
= 0, removed
, extended
, hdr_version
;
2879 struct cache_entry
**cache
= istate
->cache
;
2880 int entries
= istate
->cache_nr
;
2882 struct ondisk_cache_entry ondisk
;
2883 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
2884 int drop_cache_tree
= istate
->drop_cache_tree
;
2886 int csum_fsync_flag
;
2887 int ieot_entries
= 1;
2888 struct index_entry_offset_table
*ieot
= NULL
;
2891 f
= hashfd(tempfile
->fd
, tempfile
->filename
.buf
);
2893 for (i
= removed
= extended
= 0; i
< entries
; i
++) {
2894 if (cache
[i
]->ce_flags
& CE_REMOVE
)
2897 /* reduce extended entries if possible */
2898 cache
[i
]->ce_flags
&= ~CE_EXTENDED
;
2899 if (cache
[i
]->ce_flags
& CE_EXTENDED_FLAGS
) {
2901 cache
[i
]->ce_flags
|= CE_EXTENDED
;
2905 if (!istate
->version
)
2906 istate
->version
= get_index_format_default(the_repository
);
2908 /* demote version 3 to version 2 when the latter suffices */
2909 if (istate
->version
== 3 || istate
->version
== 2)
2910 istate
->version
= extended
? 3 : 2;
2912 hdr_version
= istate
->version
;
2914 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
2915 hdr
.hdr_version
= htonl(hdr_version
);
2916 hdr
.hdr_entries
= htonl(entries
- removed
);
2918 hashwrite(f
, &hdr
, sizeof(hdr
));
2920 if (!HAVE_THREADS
|| git_config_get_index_threads(&nr_threads
))
2923 if (nr_threads
!= 1 && record_ieot()) {
2924 int ieot_blocks
, cpus
;
2927 * ensure default number of ieot blocks maps evenly to the
2928 * default number of threads that will process them leaving
2929 * room for the thread to load the index extensions.
2932 ieot_blocks
= istate
->cache_nr
/ THREAD_COST
;
2933 cpus
= online_cpus();
2934 if (ieot_blocks
> cpus
- 1)
2935 ieot_blocks
= cpus
- 1;
2937 ieot_blocks
= nr_threads
;
2938 if (ieot_blocks
> istate
->cache_nr
)
2939 ieot_blocks
= istate
->cache_nr
;
2943 * no reason to write out the IEOT extension if we don't
2944 * have enough blocks to utilize multi-threading
2946 if (ieot_blocks
> 1) {
2947 ieot
= xcalloc(1, sizeof(struct index_entry_offset_table
)
2948 + (ieot_blocks
* sizeof(struct index_entry_offset
)));
2949 ieot_entries
= DIV_ROUND_UP(entries
, ieot_blocks
);
2953 offset
= hashfile_total(f
);
2956 previous_name
= (hdr_version
== 4) ? &previous_name_buf
: NULL
;
2958 for (i
= 0; i
< entries
; i
++) {
2959 struct cache_entry
*ce
= cache
[i
];
2960 if (ce
->ce_flags
& CE_REMOVE
)
2962 if (!ce_uptodate(ce
) && is_racy_timestamp(istate
, ce
))
2963 ce_smudge_racily_clean_entry(istate
, ce
);
2964 if (is_null_oid(&ce
->oid
)) {
2965 static const char msg
[] = "cache entry has null sha1: %s";
2966 static int allow
= -1;
2969 allow
= git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2971 warning(msg
, ce
->name
);
2973 err
= error(msg
, ce
->name
);
2975 drop_cache_tree
= 1;
2977 if (ieot
&& i
&& (i
% ieot_entries
== 0)) {
2978 ieot
->entries
[ieot
->nr
].nr
= nr
;
2979 ieot
->entries
[ieot
->nr
].offset
= offset
;
2982 * If we have a V4 index, set the first byte to an invalid
2983 * character to ensure there is nothing common with the previous
2987 previous_name
->buf
[0] = 0;
2990 offset
= hashfile_total(f
);
2992 if (ce_write_entry(f
, ce
, previous_name
, (struct ondisk_cache_entry
*)&ondisk
) < 0)
3000 ieot
->entries
[ieot
->nr
].nr
= nr
;
3001 ieot
->entries
[ieot
->nr
].offset
= offset
;
3004 strbuf_release(&previous_name_buf
);
3011 offset
= hashfile_total(f
);
3014 * The extension headers must be hashed on their own for the
3015 * EOIE extension. Create a hashfile here to compute that hash.
3017 if (offset
&& record_eoie()) {
3018 CALLOC_ARRAY(eoie_c
, 1);
3019 the_hash_algo
->init_fn(eoie_c
);
3023 * Lets write out CACHE_EXT_INDEXENTRYOFFSETTABLE first so that we
3024 * can minimize the number of extensions we have to scan through to
3025 * find it during load. Write it out regardless of the
3026 * strip_extensions parameter as we need it when loading the shared
3030 struct strbuf sb
= STRBUF_INIT
;
3032 write_ieot_extension(&sb
, ieot
);
3033 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_INDEXENTRYOFFSETTABLE
, sb
.len
) < 0;
3034 hashwrite(f
, sb
.buf
, sb
.len
);
3035 strbuf_release(&sb
);
3041 if (!strip_extensions
&& istate
->split_index
&&
3042 !is_null_oid(&istate
->split_index
->base_oid
)) {
3043 struct strbuf sb
= STRBUF_INIT
;
3045 if (istate
->sparse_index
)
3046 die(_("cannot write split index for a sparse index"));
3048 err
= write_link_extension(&sb
, istate
) < 0 ||
3049 write_index_ext_header(f
, eoie_c
, CACHE_EXT_LINK
,
3051 hashwrite(f
, sb
.buf
, sb
.len
);
3052 strbuf_release(&sb
);
3056 if (!strip_extensions
&& !drop_cache_tree
&& istate
->cache_tree
) {
3057 struct strbuf sb
= STRBUF_INIT
;
3059 cache_tree_write(&sb
, istate
->cache_tree
);
3060 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_TREE
, sb
.len
) < 0;
3061 hashwrite(f
, sb
.buf
, sb
.len
);
3062 strbuf_release(&sb
);
3066 if (!strip_extensions
&& istate
->resolve_undo
) {
3067 struct strbuf sb
= STRBUF_INIT
;
3069 resolve_undo_write(&sb
, istate
->resolve_undo
);
3070 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_RESOLVE_UNDO
,
3072 hashwrite(f
, sb
.buf
, sb
.len
);
3073 strbuf_release(&sb
);
3077 if (!strip_extensions
&& istate
->untracked
) {
3078 struct strbuf sb
= STRBUF_INIT
;
3080 write_untracked_extension(&sb
, istate
->untracked
);
3081 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_UNTRACKED
,
3083 hashwrite(f
, sb
.buf
, sb
.len
);
3084 strbuf_release(&sb
);
3088 if (!strip_extensions
&& istate
->fsmonitor_last_update
) {
3089 struct strbuf sb
= STRBUF_INIT
;
3091 write_fsmonitor_extension(&sb
, istate
);
3092 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_FSMONITOR
, sb
.len
) < 0;
3093 hashwrite(f
, sb
.buf
, sb
.len
);
3094 strbuf_release(&sb
);
3098 if (istate
->sparse_index
) {
3099 if (write_index_ext_header(f
, eoie_c
, CACHE_EXT_SPARSE_DIRECTORIES
, 0) < 0)
3104 * CACHE_EXT_ENDOFINDEXENTRIES must be written as the last entry before the SHA1
3105 * so that it can be found and processed before all the index entries are
3106 * read. Write it out regardless of the strip_extensions parameter as we need it
3107 * when loading the shared index.
3110 struct strbuf sb
= STRBUF_INIT
;
3112 write_eoie_extension(&sb
, eoie_c
, offset
);
3113 err
= write_index_ext_header(f
, NULL
, CACHE_EXT_ENDOFINDEXENTRIES
, sb
.len
) < 0;
3114 hashwrite(f
, sb
.buf
, sb
.len
);
3115 strbuf_release(&sb
);
3120 csum_fsync_flag
= 0;
3121 if (!alternate_index_output
&& (flags
& COMMIT_LOCK
))
3122 csum_fsync_flag
= CSUM_FSYNC
;
3124 finalize_hashfile(f
, istate
->oid
.hash
, FSYNC_COMPONENT_INDEX
,
3125 CSUM_HASH_IN_STREAM
| csum_fsync_flag
);
3127 if (close_tempfile_gently(tempfile
)) {
3128 error(_("could not close '%s'"), get_tempfile_path(tempfile
));
3131 if (stat(get_tempfile_path(tempfile
), &st
))
3133 istate
->timestamp
.sec
= (unsigned int)st
.st_mtime
;
3134 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
3135 trace_performance_since(start
, "write index, changed mask = %x", istate
->cache_changed
);
3138 * TODO trace2: replace "the_repository" with the actual repo instance
3139 * that is associated with the given "istate".
3141 trace2_data_intmax("index", the_repository
, "write/version",
3143 trace2_data_intmax("index", the_repository
, "write/cache_nr",
3149 void set_alternate_index_output(const char *name
)
3151 alternate_index_output
= name
;
3154 static int commit_locked_index(struct lock_file
*lk
)
3156 if (alternate_index_output
)
3157 return commit_lock_file_to(lk
, alternate_index_output
);
3159 return commit_lock_file(lk
);
3162 static int do_write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
3166 int was_full
= istate
->sparse_index
== INDEX_EXPANDED
;
3168 ret
= convert_to_sparse(istate
, 0);
3171 warning(_("failed to convert to a sparse-index"));
3176 * TODO trace2: replace "the_repository" with the actual repo instance
3177 * that is associated with the given "istate".
3179 trace2_region_enter_printf("index", "do_write_index", the_repository
,
3180 "%s", get_lock_file_path(lock
));
3181 ret
= do_write_index(istate
, lock
->tempfile
, 0, flags
);
3182 trace2_region_leave_printf("index", "do_write_index", the_repository
,
3183 "%s", get_lock_file_path(lock
));
3186 ensure_full_index(istate
);
3190 if (flags
& COMMIT_LOCK
)
3191 ret
= commit_locked_index(lock
);
3193 ret
= close_lock_file_gently(lock
);
3195 run_hooks_l("post-index-change",
3196 istate
->updated_workdir
? "1" : "0",
3197 istate
->updated_skipworktree
? "1" : "0", NULL
);
3198 istate
->updated_workdir
= 0;
3199 istate
->updated_skipworktree
= 0;
3204 static int write_split_index(struct index_state
*istate
,
3205 struct lock_file
*lock
,
3209 prepare_to_write_split_index(istate
);
3210 ret
= do_write_locked_index(istate
, lock
, flags
);
3211 finish_writing_split_index(istate
);
3215 static const char *shared_index_expire
= "2.weeks.ago";
3217 static unsigned long get_shared_index_expire_date(void)
3219 static unsigned long shared_index_expire_date
;
3220 static int shared_index_expire_date_prepared
;
3222 if (!shared_index_expire_date_prepared
) {
3223 git_config_get_expiry("splitindex.sharedindexexpire",
3224 &shared_index_expire
);
3225 shared_index_expire_date
= approxidate(shared_index_expire
);
3226 shared_index_expire_date_prepared
= 1;
3229 return shared_index_expire_date
;
3232 static int should_delete_shared_index(const char *shared_index_path
)
3235 unsigned long expiration
;
3237 /* Check timestamp */
3238 expiration
= get_shared_index_expire_date();
3241 if (stat(shared_index_path
, &st
))
3242 return error_errno(_("could not stat '%s'"), shared_index_path
);
3243 if (st
.st_mtime
> expiration
)
3249 static int clean_shared_index_files(const char *current_hex
)
3252 DIR *dir
= opendir(get_git_dir());
3255 return error_errno(_("unable to open git dir: %s"), get_git_dir());
3257 while ((de
= readdir(dir
)) != NULL
) {
3258 const char *sha1_hex
;
3259 const char *shared_index_path
;
3260 if (!skip_prefix(de
->d_name
, "sharedindex.", &sha1_hex
))
3262 if (!strcmp(sha1_hex
, current_hex
))
3264 shared_index_path
= git_path("%s", de
->d_name
);
3265 if (should_delete_shared_index(shared_index_path
) > 0 &&
3266 unlink(shared_index_path
))
3267 warning_errno(_("unable to unlink: %s"), shared_index_path
);
3274 static int write_shared_index(struct index_state
*istate
,
3275 struct tempfile
**temp
, unsigned flags
)
3277 struct split_index
*si
= istate
->split_index
;
3278 int ret
, was_full
= !istate
->sparse_index
;
3280 move_cache_to_base_index(istate
);
3281 convert_to_sparse(istate
, 0);
3283 trace2_region_enter_printf("index", "shared/do_write_index",
3284 the_repository
, "%s", get_tempfile_path(*temp
));
3285 ret
= do_write_index(si
->base
, *temp
, 1, flags
);
3286 trace2_region_leave_printf("index", "shared/do_write_index",
3287 the_repository
, "%s", get_tempfile_path(*temp
));
3290 ensure_full_index(istate
);
3294 ret
= adjust_shared_perm(get_tempfile_path(*temp
));
3296 error(_("cannot fix permission bits on '%s'"), get_tempfile_path(*temp
));
3299 ret
= rename_tempfile(temp
,
3300 git_path("sharedindex.%s", oid_to_hex(&si
->base
->oid
)));
3302 oidcpy(&si
->base_oid
, &si
->base
->oid
);
3303 clean_shared_index_files(oid_to_hex(&si
->base
->oid
));
3309 static const int default_max_percent_split_change
= 20;
3311 static int too_many_not_shared_entries(struct index_state
*istate
)
3313 int i
, not_shared
= 0;
3314 int max_split
= git_config_get_max_percent_split_change();
3316 switch (max_split
) {
3318 /* not or badly configured: use the default value */
3319 max_split
= default_max_percent_split_change
;
3322 return 1; /* 0% means always write a new shared index */
3324 return 0; /* 100% means never write a new shared index */
3326 break; /* just use the configured value */
3329 /* Count not shared entries */
3330 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3331 struct cache_entry
*ce
= istate
->cache
[i
];
3336 return (int64_t)istate
->cache_nr
* max_split
< (int64_t)not_shared
* 100;
3339 int write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
3342 int new_shared_index
, ret
, test_split_index_env
;
3343 struct split_index
*si
= istate
->split_index
;
3345 if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0))
3346 cache_tree_verify(the_repository
, istate
);
3348 if ((flags
& SKIP_IF_UNCHANGED
) && !istate
->cache_changed
) {
3349 if (flags
& COMMIT_LOCK
)
3350 rollback_lock_file(lock
);
3354 if (istate
->fsmonitor_last_update
)
3355 fill_fsmonitor_bitmap(istate
);
3357 test_split_index_env
= git_env_bool("GIT_TEST_SPLIT_INDEX", 0);
3359 if ((!si
&& !test_split_index_env
) ||
3360 alternate_index_output
||
3361 (istate
->cache_changed
& ~EXTMASK
)) {
3363 oidclr(&si
->base_oid
);
3364 ret
= do_write_locked_index(istate
, lock
, flags
);
3368 if (test_split_index_env
) {
3370 si
= init_split_index(istate
);
3371 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3373 int v
= si
->base_oid
.hash
[0];
3375 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3378 if (too_many_not_shared_entries(istate
))
3379 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3381 new_shared_index
= istate
->cache_changed
& SPLIT_INDEX_ORDERED
;
3383 if (new_shared_index
) {
3384 struct tempfile
*temp
;
3387 /* Same initial permissions as the main .git/index file */
3388 temp
= mks_tempfile_sm(git_path("sharedindex_XXXXXX"), 0, 0666);
3390 oidclr(&si
->base_oid
);
3391 ret
= do_write_locked_index(istate
, lock
, flags
);
3394 ret
= write_shared_index(istate
, &temp
, flags
);
3396 saved_errno
= errno
;
3397 if (is_tempfile_active(temp
))
3398 delete_tempfile(&temp
);
3399 errno
= saved_errno
;
3405 ret
= write_split_index(istate
, lock
, flags
);
3407 /* Freshen the shared index only if the split-index was written */
3408 if (!ret
&& !new_shared_index
&& !is_null_oid(&si
->base_oid
)) {
3409 const char *shared_index
= git_path("sharedindex.%s",
3410 oid_to_hex(&si
->base_oid
));
3411 freshen_shared_index(shared_index
, 1);
3415 if (flags
& COMMIT_LOCK
)
3416 rollback_lock_file(lock
);
3421 * Read the index file that is potentially unmerged into given
3422 * index_state, dropping any unmerged entries to stage #0 (potentially
3423 * resulting in a path appearing as both a file and a directory in the
3424 * index; the caller is responsible to clear out the extra entries
3425 * before writing the index to a tree). Returns true if the index is
3426 * unmerged. Callers who want to refuse to work from an unmerged
3427 * state can call this and check its return value, instead of calling
3430 int repo_read_index_unmerged(struct repository
*repo
)
3432 struct index_state
*istate
;
3436 repo_read_index(repo
);
3437 istate
= repo
->index
;
3438 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3439 struct cache_entry
*ce
= istate
->cache
[i
];
3440 struct cache_entry
*new_ce
;
3446 len
= ce_namelen(ce
);
3447 new_ce
= make_empty_cache_entry(istate
, len
);
3448 memcpy(new_ce
->name
, ce
->name
, len
);
3449 new_ce
->ce_flags
= create_ce_flags(0) | CE_CONFLICTED
;
3450 new_ce
->ce_namelen
= len
;
3451 new_ce
->ce_mode
= ce
->ce_mode
;
3452 if (add_index_entry(istate
, new_ce
, ADD_CACHE_SKIP_DFCHECK
))
3453 return error(_("%s: cannot drop to stage #0"),
3460 * Returns 1 if the path is an "other" path with respect to
3461 * the index; that is, the path is not mentioned in the index at all,
3462 * either as a file, a directory with some files in the index,
3463 * or as an unmerged entry.
3465 * We helpfully remove a trailing "/" from directories so that
3466 * the output of read_directory can be used as-is.
3468 int index_name_is_other(struct index_state
*istate
, const char *name
,
3472 if (namelen
&& name
[namelen
- 1] == '/')
3474 pos
= index_name_pos(istate
, name
, namelen
);
3476 return 0; /* exact match */
3478 if (pos
< istate
->cache_nr
) {
3479 struct cache_entry
*ce
= istate
->cache
[pos
];
3480 if (ce_namelen(ce
) == namelen
&&
3481 !memcmp(ce
->name
, name
, namelen
))
3482 return 0; /* Yup, this one exists unmerged */
3487 void *read_blob_data_from_index(struct index_state
*istate
,
3488 const char *path
, unsigned long *size
)
3492 enum object_type type
;
3496 pos
= index_name_pos(istate
, path
, len
);
3499 * We might be in the middle of a merge, in which
3500 * case we would read stage #2 (ours).
3504 (pos
< 0 && i
< istate
->cache_nr
&&
3505 !strcmp(istate
->cache
[i
]->name
, path
));
3507 if (ce_stage(istate
->cache
[i
]) == 2)
3512 data
= read_object_file(&istate
->cache
[pos
]->oid
, &type
, &sz
);
3513 if (!data
|| type
!= OBJ_BLOB
) {
3522 void stat_validity_clear(struct stat_validity
*sv
)
3524 FREE_AND_NULL(sv
->sd
);
3527 int stat_validity_check(struct stat_validity
*sv
, const char *path
)
3531 if (stat(path
, &st
) < 0)
3532 return sv
->sd
== NULL
;
3535 return S_ISREG(st
.st_mode
) && !match_stat_data(sv
->sd
, &st
);
3538 void stat_validity_update(struct stat_validity
*sv
, int fd
)
3542 if (fstat(fd
, &st
) < 0 || !S_ISREG(st
.st_mode
))
3543 stat_validity_clear(sv
);
3546 CALLOC_ARRAY(sv
->sd
, 1);
3547 fill_stat_data(sv
->sd
, &st
);
3551 void move_index_extensions(struct index_state
*dst
, struct index_state
*src
)
3553 dst
->untracked
= src
->untracked
;
3554 src
->untracked
= NULL
;
3555 dst
->cache_tree
= src
->cache_tree
;
3556 src
->cache_tree
= NULL
;
3559 struct cache_entry
*dup_cache_entry(const struct cache_entry
*ce
,
3560 struct index_state
*istate
)
3562 unsigned int size
= ce_size(ce
);
3563 int mem_pool_allocated
;
3564 struct cache_entry
*new_entry
= make_empty_cache_entry(istate
, ce_namelen(ce
));
3565 mem_pool_allocated
= new_entry
->mem_pool_allocated
;
3567 memcpy(new_entry
, ce
, size
);
3568 new_entry
->mem_pool_allocated
= mem_pool_allocated
;
3572 void discard_cache_entry(struct cache_entry
*ce
)
3574 if (ce
&& should_validate_cache_entries())
3575 memset(ce
, 0xCD, cache_entry_size(ce
->ce_namelen
));
3577 if (ce
&& ce
->mem_pool_allocated
)
3583 int should_validate_cache_entries(void)
3585 static int validate_index_cache_entries
= -1;
3587 if (validate_index_cache_entries
< 0) {
3588 if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))
3589 validate_index_cache_entries
= 1;
3591 validate_index_cache_entries
= 0;
3594 return validate_index_cache_entries
;
3597 #define EOIE_SIZE (4 + GIT_SHA1_RAWSZ) /* <4-byte offset> + <20-byte hash> */
3598 #define EOIE_SIZE_WITH_HEADER (4 + 4 + EOIE_SIZE) /* <4-byte signature> + <4-byte length> + EOIE_SIZE */
3600 static size_t read_eoie_extension(const char *mmap
, size_t mmap_size
)
3603 * The end of index entries (EOIE) extension is guaranteed to be last
3604 * so that it can be found by scanning backwards from the EOF.
3611 const char *index
, *eoie
;
3613 size_t offset
, src_offset
;
3614 unsigned char hash
[GIT_MAX_RAWSZ
];
3617 /* ensure we have an index big enough to contain an EOIE extension */
3618 if (mmap_size
< sizeof(struct cache_header
) + EOIE_SIZE_WITH_HEADER
+ the_hash_algo
->rawsz
)
3621 /* validate the extension signature */
3622 index
= eoie
= mmap
+ mmap_size
- EOIE_SIZE_WITH_HEADER
- the_hash_algo
->rawsz
;
3623 if (CACHE_EXT(index
) != CACHE_EXT_ENDOFINDEXENTRIES
)
3625 index
+= sizeof(uint32_t);
3627 /* validate the extension size */
3628 extsize
= get_be32(index
);
3629 if (extsize
!= EOIE_SIZE
)
3631 index
+= sizeof(uint32_t);
3634 * Validate the offset we're going to look for the first extension
3635 * signature is after the index header and before the eoie extension.
3637 offset
= get_be32(index
);
3638 if (mmap
+ offset
< mmap
+ sizeof(struct cache_header
))
3640 if (mmap
+ offset
>= eoie
)
3642 index
+= sizeof(uint32_t);
3645 * The hash is computed over extension types and their sizes (but not
3646 * their contents). E.g. if we have "TREE" extension that is N-bytes
3647 * long, "REUC" extension that is M-bytes long, followed by "EOIE",
3648 * then the hash would be:
3650 * SHA-1("TREE" + <binary representation of N> +
3651 * "REUC" + <binary representation of M>)
3653 src_offset
= offset
;
3654 the_hash_algo
->init_fn(&c
);
3655 while (src_offset
< mmap_size
- the_hash_algo
->rawsz
- EOIE_SIZE_WITH_HEADER
) {
3656 /* After an array of active_nr index entries,
3657 * there can be arbitrary number of extended
3658 * sections, each of which is prefixed with
3659 * extension name (4-byte) and section length
3660 * in 4-byte network byte order.
3663 memcpy(&extsize
, mmap
+ src_offset
+ 4, 4);
3664 extsize
= ntohl(extsize
);
3666 /* verify the extension size isn't so large it will wrap around */
3667 if (src_offset
+ 8 + extsize
< src_offset
)
3670 the_hash_algo
->update_fn(&c
, mmap
+ src_offset
, 8);
3673 src_offset
+= extsize
;
3675 the_hash_algo
->final_fn(hash
, &c
);
3676 if (!hasheq(hash
, (const unsigned char *)index
))
3679 /* Validate that the extension offsets returned us back to the eoie extension. */
3680 if (src_offset
!= mmap_size
- the_hash_algo
->rawsz
- EOIE_SIZE_WITH_HEADER
)
3686 static void write_eoie_extension(struct strbuf
*sb
, git_hash_ctx
*eoie_context
, size_t offset
)
3689 unsigned char hash
[GIT_MAX_RAWSZ
];
3692 put_be32(&buffer
, offset
);
3693 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3696 the_hash_algo
->final_fn(hash
, eoie_context
);
3697 strbuf_add(sb
, hash
, the_hash_algo
->rawsz
);
3700 #define IEOT_VERSION (1)
3702 static struct index_entry_offset_table
*read_ieot_extension(const char *mmap
, size_t mmap_size
, size_t offset
)
3704 const char *index
= NULL
;
3705 uint32_t extsize
, ext_version
;
3706 struct index_entry_offset_table
*ieot
;
3709 /* find the IEOT extension */
3712 while (offset
<= mmap_size
- the_hash_algo
->rawsz
- 8) {
3713 extsize
= get_be32(mmap
+ offset
+ 4);
3714 if (CACHE_EXT((mmap
+ offset
)) == CACHE_EXT_INDEXENTRYOFFSETTABLE
) {
3715 index
= mmap
+ offset
+ 4 + 4;
3724 /* validate the version is IEOT_VERSION */
3725 ext_version
= get_be32(index
);
3726 if (ext_version
!= IEOT_VERSION
) {
3727 error("invalid IEOT version %d", ext_version
);
3730 index
+= sizeof(uint32_t);
3732 /* extension size - version bytes / bytes per entry */
3733 nr
= (extsize
- sizeof(uint32_t)) / (sizeof(uint32_t) + sizeof(uint32_t));
3735 error("invalid number of IEOT entries %d", nr
);
3738 ieot
= xmalloc(sizeof(struct index_entry_offset_table
)
3739 + (nr
* sizeof(struct index_entry_offset
)));
3741 for (i
= 0; i
< nr
; i
++) {
3742 ieot
->entries
[i
].offset
= get_be32(index
);
3743 index
+= sizeof(uint32_t);
3744 ieot
->entries
[i
].nr
= get_be32(index
);
3745 index
+= sizeof(uint32_t);
3751 static void write_ieot_extension(struct strbuf
*sb
, struct index_entry_offset_table
*ieot
)
3757 put_be32(&buffer
, IEOT_VERSION
);
3758 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3761 for (i
= 0; i
< ieot
->nr
; i
++) {
3764 put_be32(&buffer
, ieot
->entries
[i
].offset
);
3765 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3768 put_be32(&buffer
, ieot
->entries
[i
].nr
);
3769 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3773 void prefetch_cache_entries(const struct index_state
*istate
,
3774 must_prefetch_predicate must_prefetch
)
3777 struct oid_array to_fetch
= OID_ARRAY_INIT
;
3779 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3780 struct cache_entry
*ce
= istate
->cache
[i
];
3782 if (S_ISGITLINK(ce
->ce_mode
) || !must_prefetch(ce
))
3784 if (!oid_object_info_extended(the_repository
, &ce
->oid
,
3786 OBJECT_INFO_FOR_PREFETCH
))
3788 oid_array_append(&to_fetch
, &ce
->oid
);
3790 promisor_remote_get_direct(the_repository
,
3791 to_fetch
.oid
, to_fetch
.nr
);
3792 oid_array_clear(&to_fetch
);