2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
12 #include "cache-tree.h"
15 #include "object-store.h"
19 #include "resolve-undo.h"
20 #include "run-command.h"
23 #include "split-index.h"
25 #include "fsmonitor.h"
26 #include "thread-utils.h"
28 #include "sparse-index.h"
29 #include "csum-file.h"
30 #include "promisor-remote.h"
32 /* Mask for the name length in ce_flags in the on-disk index */
34 #define CE_NAMEMASK (0x0fff)
38 * The first letter should be 'A'..'Z' for extensions that are not
39 * necessary for a correct operation (i.e. optimization data).
40 * When new extensions are added that _needs_ to be understood in
41 * order to correctly interpret the index file, pick character that
42 * is outside the range, to cause the reader to abort.
45 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
46 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
47 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
48 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
49 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
50 #define CACHE_EXT_FSMONITOR 0x46534D4E /* "FSMN" */
51 #define CACHE_EXT_ENDOFINDEXENTRIES 0x454F4945 /* "EOIE" */
52 #define CACHE_EXT_INDEXENTRYOFFSETTABLE 0x49454F54 /* "IEOT" */
53 #define CACHE_EXT_SPARSE_DIRECTORIES 0x73646972 /* "sdir" */
55 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
56 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
57 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
58 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED | FSMONITOR_CHANGED)
62 * This is an estimate of the pathname length in the index. We use
63 * this for V4 index files to guess the un-deltafied size of the index
64 * in memory because of pathname deltafication. This is not required
65 * for V2/V3 index formats because their pathnames are not compressed.
66 * If the initial amount of memory set aside is not sufficient, the
67 * mem pool will allocate extra memory.
69 #define CACHE_ENTRY_PATH_LENGTH 80
71 static inline struct cache_entry
*mem_pool__ce_alloc(struct mem_pool
*mem_pool
, size_t len
)
73 struct cache_entry
*ce
;
74 ce
= mem_pool_alloc(mem_pool
, cache_entry_size(len
));
75 ce
->mem_pool_allocated
= 1;
79 static inline struct cache_entry
*mem_pool__ce_calloc(struct mem_pool
*mem_pool
, size_t len
)
81 struct cache_entry
* ce
;
82 ce
= mem_pool_calloc(mem_pool
, 1, cache_entry_size(len
));
83 ce
->mem_pool_allocated
= 1;
87 static struct mem_pool
*find_mem_pool(struct index_state
*istate
)
89 struct mem_pool
**pool_ptr
;
91 if (istate
->split_index
&& istate
->split_index
->base
)
92 pool_ptr
= &istate
->split_index
->base
->ce_mem_pool
;
94 pool_ptr
= &istate
->ce_mem_pool
;
97 *pool_ptr
= xmalloc(sizeof(**pool_ptr
));
98 mem_pool_init(*pool_ptr
, 0);
104 static const char *alternate_index_output
;
106 static void set_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
108 if (S_ISSPARSEDIR(ce
->ce_mode
))
109 istate
->sparse_index
= 1;
111 istate
->cache
[nr
] = ce
;
112 add_name_hash(istate
, ce
);
115 static void replace_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
117 struct cache_entry
*old
= istate
->cache
[nr
];
119 replace_index_entry_in_base(istate
, old
, ce
);
120 remove_name_hash(istate
, old
);
121 discard_cache_entry(old
);
122 ce
->ce_flags
&= ~CE_HASHED
;
123 set_index_entry(istate
, nr
, ce
);
124 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
125 mark_fsmonitor_invalid(istate
, ce
);
126 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
129 void rename_index_entry_at(struct index_state
*istate
, int nr
, const char *new_name
)
131 struct cache_entry
*old_entry
= istate
->cache
[nr
], *new_entry
;
132 int namelen
= strlen(new_name
);
134 new_entry
= make_empty_cache_entry(istate
, namelen
);
135 copy_cache_entry(new_entry
, old_entry
);
136 new_entry
->ce_flags
&= ~CE_HASHED
;
137 new_entry
->ce_namelen
= namelen
;
138 new_entry
->index
= 0;
139 memcpy(new_entry
->name
, new_name
, namelen
+ 1);
141 cache_tree_invalidate_path(istate
, old_entry
->name
);
142 untracked_cache_remove_from_index(istate
, old_entry
->name
);
143 remove_index_entry_at(istate
, nr
);
144 add_index_entry(istate
, new_entry
, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
147 void fill_stat_data(struct stat_data
*sd
, struct stat
*st
)
149 sd
->sd_ctime
.sec
= (unsigned int)st
->st_ctime
;
150 sd
->sd_mtime
.sec
= (unsigned int)st
->st_mtime
;
151 sd
->sd_ctime
.nsec
= ST_CTIME_NSEC(*st
);
152 sd
->sd_mtime
.nsec
= ST_MTIME_NSEC(*st
);
153 sd
->sd_dev
= st
->st_dev
;
154 sd
->sd_ino
= st
->st_ino
;
155 sd
->sd_uid
= st
->st_uid
;
156 sd
->sd_gid
= st
->st_gid
;
157 sd
->sd_size
= st
->st_size
;
160 int match_stat_data(const struct stat_data
*sd
, struct stat
*st
)
164 if (sd
->sd_mtime
.sec
!= (unsigned int)st
->st_mtime
)
165 changed
|= MTIME_CHANGED
;
166 if (trust_ctime
&& check_stat
&&
167 sd
->sd_ctime
.sec
!= (unsigned int)st
->st_ctime
)
168 changed
|= CTIME_CHANGED
;
171 if (check_stat
&& sd
->sd_mtime
.nsec
!= ST_MTIME_NSEC(*st
))
172 changed
|= MTIME_CHANGED
;
173 if (trust_ctime
&& check_stat
&&
174 sd
->sd_ctime
.nsec
!= ST_CTIME_NSEC(*st
))
175 changed
|= CTIME_CHANGED
;
179 if (sd
->sd_uid
!= (unsigned int) st
->st_uid
||
180 sd
->sd_gid
!= (unsigned int) st
->st_gid
)
181 changed
|= OWNER_CHANGED
;
182 if (sd
->sd_ino
!= (unsigned int) st
->st_ino
)
183 changed
|= INODE_CHANGED
;
188 * st_dev breaks on network filesystems where different
189 * clients will have different views of what "device"
190 * the filesystem is on
192 if (check_stat
&& sd
->sd_dev
!= (unsigned int) st
->st_dev
)
193 changed
|= INODE_CHANGED
;
196 if (sd
->sd_size
!= (unsigned int) st
->st_size
)
197 changed
|= DATA_CHANGED
;
203 * This only updates the "non-critical" parts of the directory
204 * cache, ie the parts that aren't tracked by GIT, and only used
205 * to validate the cache.
207 void fill_stat_cache_info(struct index_state
*istate
, struct cache_entry
*ce
, struct stat
*st
)
209 fill_stat_data(&ce
->ce_stat_data
, st
);
211 if (assume_unchanged
)
212 ce
->ce_flags
|= CE_VALID
;
214 if (S_ISREG(st
->st_mode
)) {
215 ce_mark_uptodate(ce
);
216 mark_fsmonitor_valid(istate
, ce
);
220 static int ce_compare_data(struct index_state
*istate
,
221 const struct cache_entry
*ce
,
225 int fd
= git_open_cloexec(ce
->name
, O_RDONLY
);
228 struct object_id oid
;
229 if (!index_fd(istate
, &oid
, fd
, st
, OBJ_BLOB
, ce
->name
, 0))
230 match
= !oideq(&oid
, &ce
->oid
);
231 /* index_fd() closed the file descriptor already */
236 static int ce_compare_link(const struct cache_entry
*ce
, size_t expected_size
)
241 enum object_type type
;
242 struct strbuf sb
= STRBUF_INIT
;
244 if (strbuf_readlink(&sb
, ce
->name
, expected_size
))
247 buffer
= read_object_file(&ce
->oid
, &type
, &size
);
250 match
= memcmp(buffer
, sb
.buf
, size
);
257 static int ce_compare_gitlink(const struct cache_entry
*ce
)
259 struct object_id oid
;
262 * We don't actually require that the .git directory
263 * under GITLINK directory be a valid git directory. It
264 * might even be missing (in case nobody populated that
267 * If so, we consider it always to match.
269 if (resolve_gitlink_ref(ce
->name
, "HEAD", &oid
) < 0)
271 return !oideq(&oid
, &ce
->oid
);
274 static int ce_modified_check_fs(struct index_state
*istate
,
275 const struct cache_entry
*ce
,
278 switch (st
->st_mode
& S_IFMT
) {
280 if (ce_compare_data(istate
, ce
, st
))
284 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
288 if (S_ISGITLINK(ce
->ce_mode
))
289 return ce_compare_gitlink(ce
) ? DATA_CHANGED
: 0;
290 /* else fallthrough */
297 static int ce_match_stat_basic(const struct cache_entry
*ce
, struct stat
*st
)
299 unsigned int changed
= 0;
301 if (ce
->ce_flags
& CE_REMOVE
)
302 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
304 switch (ce
->ce_mode
& S_IFMT
) {
306 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
307 /* We consider only the owner x bit to be relevant for
310 if (trust_executable_bit
&&
311 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
312 changed
|= MODE_CHANGED
;
315 if (!S_ISLNK(st
->st_mode
) &&
316 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
317 changed
|= TYPE_CHANGED
;
320 /* We ignore most of the st_xxx fields for gitlinks */
321 if (!S_ISDIR(st
->st_mode
))
322 changed
|= TYPE_CHANGED
;
323 else if (ce_compare_gitlink(ce
))
324 changed
|= DATA_CHANGED
;
327 BUG("unsupported ce_mode: %o", ce
->ce_mode
);
330 changed
|= match_stat_data(&ce
->ce_stat_data
, st
);
332 /* Racily smudged entry? */
333 if (!ce
->ce_stat_data
.sd_size
) {
334 if (!is_empty_blob_sha1(ce
->oid
.hash
))
335 changed
|= DATA_CHANGED
;
341 static int is_racy_stat(const struct index_state
*istate
,
342 const struct stat_data
*sd
)
344 return (istate
->timestamp
.sec
&&
346 /* nanosecond timestamped files can also be racy! */
347 (istate
->timestamp
.sec
< sd
->sd_mtime
.sec
||
348 (istate
->timestamp
.sec
== sd
->sd_mtime
.sec
&&
349 istate
->timestamp
.nsec
<= sd
->sd_mtime
.nsec
))
351 istate
->timestamp
.sec
<= sd
->sd_mtime
.sec
356 int is_racy_timestamp(const struct index_state
*istate
,
357 const struct cache_entry
*ce
)
359 return (!S_ISGITLINK(ce
->ce_mode
) &&
360 is_racy_stat(istate
, &ce
->ce_stat_data
));
363 int match_stat_data_racy(const struct index_state
*istate
,
364 const struct stat_data
*sd
, struct stat
*st
)
366 if (is_racy_stat(istate
, sd
))
367 return MTIME_CHANGED
;
368 return match_stat_data(sd
, st
);
371 int ie_match_stat(struct index_state
*istate
,
372 const struct cache_entry
*ce
, struct stat
*st
,
373 unsigned int options
)
375 unsigned int changed
;
376 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
377 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
378 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
379 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
381 if (!ignore_fsmonitor
)
382 refresh_fsmonitor(istate
);
384 * If it's marked as always valid in the index, it's
385 * valid whatever the checked-out copy says.
387 * skip-worktree has the same effect with higher precedence
389 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
))
391 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
393 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
))
397 * Intent-to-add entries have not been added, so the index entry
398 * by definition never matches what is in the work tree until it
399 * actually gets added.
401 if (ce_intent_to_add(ce
))
402 return DATA_CHANGED
| TYPE_CHANGED
| MODE_CHANGED
;
404 changed
= ce_match_stat_basic(ce
, st
);
407 * Within 1 second of this sequence:
408 * echo xyzzy >file && git-update-index --add file
409 * running this command:
411 * would give a falsely clean cache entry. The mtime and
412 * length match the cache, and other stat fields do not change.
414 * We could detect this at update-index time (the cache entry
415 * being registered/updated records the same time as "now")
416 * and delay the return from git-update-index, but that would
417 * effectively mean we can make at most one commit per second,
418 * which is not acceptable. Instead, we check cache entries
419 * whose mtime are the same as the index file timestamp more
420 * carefully than others.
422 if (!changed
&& is_racy_timestamp(istate
, ce
)) {
423 if (assume_racy_is_modified
)
424 changed
|= DATA_CHANGED
;
426 changed
|= ce_modified_check_fs(istate
, ce
, st
);
432 int ie_modified(struct index_state
*istate
,
433 const struct cache_entry
*ce
,
434 struct stat
*st
, unsigned int options
)
436 int changed
, changed_fs
;
438 changed
= ie_match_stat(istate
, ce
, st
, options
);
442 * If the mode or type has changed, there's no point in trying
443 * to refresh the entry - it's not going to match
445 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
449 * Immediately after read-tree or update-index --cacheinfo,
450 * the length field is zero, as we have never even read the
451 * lstat(2) information once, and we cannot trust DATA_CHANGED
452 * returned by ie_match_stat() which in turn was returned by
453 * ce_match_stat_basic() to signal that the filesize of the
454 * blob changed. We have to actually go to the filesystem to
455 * see if the contents match, and if so, should answer "unchanged".
457 * The logic does not apply to gitlinks, as ce_match_stat_basic()
458 * already has checked the actual HEAD from the filesystem in the
459 * subproject. If ie_match_stat() already said it is different,
460 * then we know it is.
462 if ((changed
& DATA_CHANGED
) &&
463 (S_ISGITLINK(ce
->ce_mode
) || ce
->ce_stat_data
.sd_size
!= 0))
466 changed_fs
= ce_modified_check_fs(istate
, ce
, st
);
468 return changed
| changed_fs
;
472 int base_name_compare(const char *name1
, int len1
, int mode1
,
473 const char *name2
, int len2
, int mode2
)
475 unsigned char c1
, c2
;
476 int len
= len1
< len2
? len1
: len2
;
479 cmp
= memcmp(name1
, name2
, len
);
484 if (!c1
&& S_ISDIR(mode1
))
486 if (!c2
&& S_ISDIR(mode2
))
488 return (c1
< c2
) ? -1 : (c1
> c2
) ? 1 : 0;
492 * df_name_compare() is identical to base_name_compare(), except it
493 * compares conflicting directory/file entries as equal. Note that
494 * while a directory name compares as equal to a regular file, they
495 * then individually compare _differently_ to a filename that has
496 * a dot after the basename (because '\0' < '.' < '/').
498 * This is used by routines that want to traverse the git namespace
499 * but then handle conflicting entries together when possible.
501 int df_name_compare(const char *name1
, int len1
, int mode1
,
502 const char *name2
, int len2
, int mode2
)
504 int len
= len1
< len2
? len1
: len2
, cmp
;
505 unsigned char c1
, c2
;
507 cmp
= memcmp(name1
, name2
, len
);
510 /* Directories and files compare equal (same length, same name) */
514 if (!c1
&& S_ISDIR(mode1
))
517 if (!c2
&& S_ISDIR(mode2
))
519 if (c1
== '/' && !c2
)
521 if (c2
== '/' && !c1
)
526 int name_compare(const char *name1
, size_t len1
, const char *name2
, size_t len2
)
528 size_t min_len
= (len1
< len2
) ? len1
: len2
;
529 int cmp
= memcmp(name1
, name2
, min_len
);
539 int cache_name_stage_compare(const char *name1
, int len1
, int stage1
, const char *name2
, int len2
, int stage2
)
543 cmp
= name_compare(name1
, len1
, name2
, len2
);
554 static int index_name_stage_pos(struct index_state
*istate
, const char *name
, int namelen
, int stage
)
559 last
= istate
->cache_nr
;
560 while (last
> first
) {
561 int next
= first
+ ((last
- first
) >> 1);
562 struct cache_entry
*ce
= istate
->cache
[next
];
563 int cmp
= cache_name_stage_compare(name
, namelen
, stage
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
573 if (istate
->sparse_index
&&
575 /* Note: first <= istate->cache_nr */
576 struct cache_entry
*ce
= istate
->cache
[first
- 1];
579 * If we are in a sparse-index _and_ the entry before the
580 * insertion position is a sparse-directory entry that is
581 * an ancestor of 'name', then we need to expand the index
582 * and search again. This will only trigger once, because
583 * thereafter the index is fully expanded.
585 if (S_ISSPARSEDIR(ce
->ce_mode
) &&
586 ce_namelen(ce
) < namelen
&&
587 !strncmp(name
, ce
->name
, ce_namelen(ce
))) {
588 ensure_full_index(istate
);
589 return index_name_stage_pos(istate
, name
, namelen
, stage
);
596 int index_name_pos(struct index_state
*istate
, const char *name
, int namelen
)
598 return index_name_stage_pos(istate
, name
, namelen
, 0);
601 int remove_index_entry_at(struct index_state
*istate
, int pos
)
603 struct cache_entry
*ce
= istate
->cache
[pos
];
605 record_resolve_undo(istate
, ce
);
606 remove_name_hash(istate
, ce
);
607 save_or_free_index_entry(istate
, ce
);
608 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
610 if (pos
>= istate
->cache_nr
)
612 MOVE_ARRAY(istate
->cache
+ pos
, istate
->cache
+ pos
+ 1,
613 istate
->cache_nr
- pos
);
618 * Remove all cache entries marked for removal, that is where
619 * CE_REMOVE is set in ce_flags. This is much more effective than
620 * calling remove_index_entry_at() for each entry to be removed.
622 void remove_marked_cache_entries(struct index_state
*istate
, int invalidate
)
624 struct cache_entry
**ce_array
= istate
->cache
;
627 for (i
= j
= 0; i
< istate
->cache_nr
; i
++) {
628 if (ce_array
[i
]->ce_flags
& CE_REMOVE
) {
630 cache_tree_invalidate_path(istate
,
632 untracked_cache_remove_from_index(istate
,
635 remove_name_hash(istate
, ce_array
[i
]);
636 save_or_free_index_entry(istate
, ce_array
[i
]);
639 ce_array
[j
++] = ce_array
[i
];
641 if (j
== istate
->cache_nr
)
643 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
644 istate
->cache_nr
= j
;
647 int remove_file_from_index(struct index_state
*istate
, const char *path
)
649 int pos
= index_name_pos(istate
, path
, strlen(path
));
652 cache_tree_invalidate_path(istate
, path
);
653 untracked_cache_remove_from_index(istate
, path
);
654 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
655 remove_index_entry_at(istate
, pos
);
659 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
661 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
664 static int index_name_pos_also_unmerged(struct index_state
*istate
,
665 const char *path
, int namelen
)
667 int pos
= index_name_pos(istate
, path
, namelen
);
668 struct cache_entry
*ce
;
673 /* maybe unmerged? */
675 if (pos
>= istate
->cache_nr
||
676 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
679 /* order of preference: stage 2, 1, 3 */
680 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
681 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
682 !compare_name(ce
, path
, namelen
))
687 static int different_name(struct cache_entry
*ce
, struct cache_entry
*alias
)
689 int len
= ce_namelen(ce
);
690 return ce_namelen(alias
) != len
|| memcmp(ce
->name
, alias
->name
, len
);
694 * If we add a filename that aliases in the cache, we will use the
695 * name that we already have - but we don't want to update the same
696 * alias twice, because that implies that there were actually two
697 * different files with aliasing names!
699 * So we use the CE_ADDED flag to verify that the alias was an old
700 * one before we accept it as
702 static struct cache_entry
*create_alias_ce(struct index_state
*istate
,
703 struct cache_entry
*ce
,
704 struct cache_entry
*alias
)
707 struct cache_entry
*new_entry
;
709 if (alias
->ce_flags
& CE_ADDED
)
710 die(_("will not add file alias '%s' ('%s' already exists in index)"),
711 ce
->name
, alias
->name
);
713 /* Ok, create the new entry using the name of the existing alias */
714 len
= ce_namelen(alias
);
715 new_entry
= make_empty_cache_entry(istate
, len
);
716 memcpy(new_entry
->name
, alias
->name
, len
);
717 copy_cache_entry(new_entry
, ce
);
718 save_or_free_index_entry(istate
, ce
);
722 void set_object_name_for_intent_to_add_entry(struct cache_entry
*ce
)
724 struct object_id oid
;
725 if (write_object_file("", 0, blob_type
, &oid
))
726 die(_("cannot create an empty blob in the object database"));
727 oidcpy(&ce
->oid
, &oid
);
730 int add_to_index(struct index_state
*istate
, const char *path
, struct stat
*st
, int flags
)
732 int namelen
, was_same
;
733 mode_t st_mode
= st
->st_mode
;
734 struct cache_entry
*ce
, *alias
= NULL
;
735 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_IGNORE_SKIP_WORKTREE
|CE_MATCH_RACY_IS_DIRTY
;
736 int verbose
= flags
& (ADD_CACHE_VERBOSE
| ADD_CACHE_PRETEND
);
737 int pretend
= flags
& ADD_CACHE_PRETEND
;
738 int intent_only
= flags
& ADD_CACHE_INTENT
;
739 int add_option
= (ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
|
740 (intent_only
? ADD_CACHE_NEW_ONLY
: 0));
741 int hash_flags
= HASH_WRITE_OBJECT
;
742 struct object_id oid
;
744 if (flags
& ADD_CACHE_RENORMALIZE
)
745 hash_flags
|= HASH_RENORMALIZE
;
747 if (!S_ISREG(st_mode
) && !S_ISLNK(st_mode
) && !S_ISDIR(st_mode
))
748 return error(_("%s: can only add regular files, symbolic links or git-directories"), path
);
750 namelen
= strlen(path
);
751 if (S_ISDIR(st_mode
)) {
752 if (resolve_gitlink_ref(path
, "HEAD", &oid
) < 0)
753 return error(_("'%s' does not have a commit checked out"), path
);
754 while (namelen
&& path
[namelen
-1] == '/')
757 ce
= make_empty_cache_entry(istate
, namelen
);
758 memcpy(ce
->name
, path
, namelen
);
759 ce
->ce_namelen
= namelen
;
761 fill_stat_cache_info(istate
, ce
, st
);
763 ce
->ce_flags
|= CE_INTENT_TO_ADD
;
766 if (trust_executable_bit
&& has_symlinks
) {
767 ce
->ce_mode
= create_ce_mode(st_mode
);
769 /* If there is an existing entry, pick the mode bits and type
770 * from it, otherwise assume unexecutable regular file.
772 struct cache_entry
*ent
;
773 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
775 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
776 ce
->ce_mode
= ce_mode_from_stat(ent
, st_mode
);
779 /* When core.ignorecase=true, determine if a directory of the same name but differing
780 * case already exists within the Git repository. If it does, ensure the directory
781 * case of the file being added to the repository matches (is folded into) the existing
782 * entry's directory case.
785 adjust_dirname_case(istate
, ce
->name
);
787 if (!(flags
& ADD_CACHE_RENORMALIZE
)) {
788 alias
= index_file_exists(istate
, ce
->name
,
789 ce_namelen(ce
), ignore_case
);
792 !ie_match_stat(istate
, alias
, st
, ce_option
)) {
793 /* Nothing changed, really */
794 if (!S_ISGITLINK(alias
->ce_mode
))
795 ce_mark_uptodate(alias
);
796 alias
->ce_flags
|= CE_ADDED
;
798 discard_cache_entry(ce
);
803 if (index_path(istate
, &ce
->oid
, path
, st
, hash_flags
)) {
804 discard_cache_entry(ce
);
805 return error(_("unable to index file '%s'"), path
);
808 set_object_name_for_intent_to_add_entry(ce
);
810 if (ignore_case
&& alias
&& different_name(ce
, alias
))
811 ce
= create_alias_ce(istate
, ce
, alias
);
812 ce
->ce_flags
|= CE_ADDED
;
814 /* It was suspected to be racily clean, but it turns out to be Ok */
817 oideq(&alias
->oid
, &ce
->oid
) &&
818 ce
->ce_mode
== alias
->ce_mode
);
821 discard_cache_entry(ce
);
822 else if (add_index_entry(istate
, ce
, add_option
)) {
823 discard_cache_entry(ce
);
824 return error(_("unable to add '%s' to index"), path
);
826 if (verbose
&& !was_same
)
827 printf("add '%s'\n", path
);
831 int add_file_to_index(struct index_state
*istate
, const char *path
, int flags
)
834 if (lstat(path
, &st
))
835 die_errno(_("unable to stat '%s'"), path
);
836 return add_to_index(istate
, path
, &st
, flags
);
839 struct cache_entry
*make_empty_cache_entry(struct index_state
*istate
, size_t len
)
841 return mem_pool__ce_calloc(find_mem_pool(istate
), len
);
844 struct cache_entry
*make_empty_transient_cache_entry(size_t len
,
845 struct mem_pool
*ce_mem_pool
)
848 return mem_pool__ce_calloc(ce_mem_pool
, len
);
849 return xcalloc(1, cache_entry_size(len
));
852 struct cache_entry
*make_cache_entry(struct index_state
*istate
,
854 const struct object_id
*oid
,
857 unsigned int refresh_options
)
859 struct cache_entry
*ce
, *ret
;
862 if (!verify_path(path
, mode
)) {
863 error(_("invalid path '%s'"), path
);
868 ce
= make_empty_cache_entry(istate
, len
);
870 oidcpy(&ce
->oid
, oid
);
871 memcpy(ce
->name
, path
, len
);
872 ce
->ce_flags
= create_ce_flags(stage
);
873 ce
->ce_namelen
= len
;
874 ce
->ce_mode
= create_ce_mode(mode
);
876 ret
= refresh_cache_entry(istate
, ce
, refresh_options
);
878 discard_cache_entry(ce
);
882 struct cache_entry
*make_transient_cache_entry(unsigned int mode
,
883 const struct object_id
*oid
,
886 struct mem_pool
*ce_mem_pool
)
888 struct cache_entry
*ce
;
891 if (!verify_path(path
, mode
)) {
892 error(_("invalid path '%s'"), path
);
897 ce
= make_empty_transient_cache_entry(len
, ce_mem_pool
);
899 oidcpy(&ce
->oid
, oid
);
900 memcpy(ce
->name
, path
, len
);
901 ce
->ce_flags
= create_ce_flags(stage
);
902 ce
->ce_namelen
= len
;
903 ce
->ce_mode
= create_ce_mode(mode
);
909 * Chmod an index entry with either +x or -x.
911 * Returns -1 if the chmod for the particular cache entry failed (if it's
912 * not a regular file), -2 if an invalid flip argument is passed in, 0
915 int chmod_index_entry(struct index_state
*istate
, struct cache_entry
*ce
,
918 if (!S_ISREG(ce
->ce_mode
))
925 ce
->ce_mode
&= ~0111;
930 cache_tree_invalidate_path(istate
, ce
->name
);
931 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
932 mark_fsmonitor_invalid(istate
, ce
);
933 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
938 int ce_same_name(const struct cache_entry
*a
, const struct cache_entry
*b
)
940 int len
= ce_namelen(a
);
941 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
945 * We fundamentally don't like some paths: we don't want
946 * dot or dot-dot anywhere, and for obvious reasons don't
947 * want to recurse into ".git" either.
949 * Also, we don't want double slashes or slashes at the
950 * end that can make pathnames ambiguous.
952 static int verify_dotfile(const char *rest
, unsigned mode
)
955 * The first character was '.', but that
956 * has already been discarded, we now test
960 /* "." is not allowed */
961 if (*rest
== '\0' || is_dir_sep(*rest
))
966 * ".git" followed by NUL or slash is bad. Note that we match
967 * case-insensitively here, even if ignore_case is not set.
968 * This outlaws ".GIT" everywhere out of an abundance of caution,
969 * since there's really no good reason to allow it.
971 * Once we've seen ".git", we can also find ".gitmodules", etc (also
972 * case-insensitively).
976 if (rest
[1] != 'i' && rest
[1] != 'I')
978 if (rest
[2] != 't' && rest
[2] != 'T')
980 if (rest
[3] == '\0' || is_dir_sep(rest
[3]))
984 if (skip_iprefix(rest
, "modules", &rest
) &&
985 (*rest
== '\0' || is_dir_sep(*rest
)))
990 if (rest
[1] == '\0' || is_dir_sep(rest
[1]))
996 int verify_path(const char *path
, unsigned mode
)
1000 if (has_dos_drive_prefix(path
))
1003 if (!is_valid_path(path
))
1010 if (is_dir_sep(c
)) {
1014 if (is_hfs_dotgit(path
))
1016 if (S_ISLNK(mode
)) {
1017 if (is_hfs_dotgitmodules(path
))
1022 #if defined GIT_WINDOWS_NATIVE || defined __CYGWIN__
1026 if (is_ntfs_dotgit(path
))
1028 if (S_ISLNK(mode
)) {
1029 if (is_ntfs_dotgitmodules(path
))
1035 if ((c
== '.' && !verify_dotfile(path
, mode
)) ||
1039 * allow terminating directory separators for
1040 * sparse directory entries.
1043 return S_ISDIR(mode
);
1044 } else if (c
== '\\' && protect_ntfs
) {
1045 if (is_ntfs_dotgit(path
))
1047 if (S_ISLNK(mode
)) {
1048 if (is_ntfs_dotgitmodules(path
))
1058 * Do we have another file that has the beginning components being a
1059 * proper superset of the name we're trying to add?
1061 static int has_file_name(struct index_state
*istate
,
1062 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
1065 int len
= ce_namelen(ce
);
1066 int stage
= ce_stage(ce
);
1067 const char *name
= ce
->name
;
1069 while (pos
< istate
->cache_nr
) {
1070 struct cache_entry
*p
= istate
->cache
[pos
++];
1072 if (len
>= ce_namelen(p
))
1074 if (memcmp(name
, p
->name
, len
))
1076 if (ce_stage(p
) != stage
)
1078 if (p
->name
[len
] != '/')
1080 if (p
->ce_flags
& CE_REMOVE
)
1085 remove_index_entry_at(istate
, --pos
);
1092 * Like strcmp(), but also return the offset of the first change.
1093 * If strings are equal, return the length.
1095 int strcmp_offset(const char *s1
, const char *s2
, size_t *first_change
)
1100 return strcmp(s1
, s2
);
1102 for (k
= 0; s1
[k
] == s2
[k
]; k
++)
1107 return (unsigned char)s1
[k
] - (unsigned char)s2
[k
];
1111 * Do we have another file with a pathname that is a proper
1112 * subset of the name we're trying to add?
1114 * That is, is there another file in the index with a path
1115 * that matches a sub-directory in the given entry?
1117 static int has_dir_name(struct index_state
*istate
,
1118 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
1121 int stage
= ce_stage(ce
);
1122 const char *name
= ce
->name
;
1123 const char *slash
= name
+ ce_namelen(ce
);
1128 * We are frequently called during an iteration on a sorted
1129 * list of pathnames and while building a new index. Therefore,
1130 * there is a high probability that this entry will eventually
1131 * be appended to the index, rather than inserted in the middle.
1132 * If we can confirm that, we can avoid binary searches on the
1133 * components of the pathname.
1135 * Compare the entry's full path with the last path in the index.
1137 if (istate
->cache_nr
> 0) {
1138 cmp_last
= strcmp_offset(name
,
1139 istate
->cache
[istate
->cache_nr
- 1]->name
,
1142 if (len_eq_last
== 0) {
1144 * The entry sorts AFTER the last one in the
1145 * index and their paths have no common prefix,
1146 * so there cannot be a F/D conflict.
1151 * The entry sorts AFTER the last one in the
1152 * index, but has a common prefix. Fall through
1153 * to the loop below to disect the entry's path
1154 * and see where the difference is.
1157 } else if (cmp_last
== 0) {
1159 * The entry exactly matches the last one in the
1160 * index, but because of multiple stage and CE_REMOVE
1161 * items, we fall through and let the regular search
1171 if (*--slash
== '/')
1173 if (slash
<= ce
->name
)
1180 * (len + 1) is a directory boundary (including
1181 * the trailing slash). And since the loop is
1182 * decrementing "slash", the first iteration is
1183 * the longest directory prefix; subsequent
1184 * iterations consider parent directories.
1187 if (len
+ 1 <= len_eq_last
) {
1189 * The directory prefix (including the trailing
1190 * slash) also appears as a prefix in the last
1191 * entry, so the remainder cannot collide (because
1192 * strcmp said the whole path was greater).
1197 * LT: last: xxx/file_A
1203 if (len
> len_eq_last
) {
1205 * This part of the directory prefix (excluding
1206 * the trailing slash) is longer than the known
1207 * equal portions, so this sub-directory cannot
1208 * collide with a file.
1217 * This is a possible collision. Fall through and
1218 * let the regular search code handle it.
1225 pos
= index_name_stage_pos(istate
, name
, len
, stage
);
1228 * Found one, but not so fast. This could
1229 * be a marker that says "I was here, but
1230 * I am being removed". Such an entry is
1231 * not a part of the resulting tree, and
1232 * it is Ok to have a directory at the same
1235 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
1239 remove_index_entry_at(istate
, pos
);
1247 * Trivial optimization: if we find an entry that
1248 * already matches the sub-directory, then we know
1249 * we're ok, and we can exit.
1251 while (pos
< istate
->cache_nr
) {
1252 struct cache_entry
*p
= istate
->cache
[pos
];
1253 if ((ce_namelen(p
) <= len
) ||
1254 (p
->name
[len
] != '/') ||
1255 memcmp(p
->name
, name
, len
))
1256 break; /* not our subdirectory */
1257 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
1259 * p is at the same stage as our entry, and
1260 * is a subdirectory of what we are looking
1261 * at, so we cannot have conflicts at our
1262 * level or anything shorter.
1271 /* We may be in a situation where we already have path/file and path
1272 * is being added, or we already have path and path/file is being
1273 * added. Either one would result in a nonsense tree that has path
1274 * twice when git-write-tree tries to write it out. Prevent it.
1276 * If ok-to-replace is specified, we remove the conflicting entries
1277 * from the cache so the caller should recompute the insert position.
1278 * When this happens, we return non-zero.
1280 static int check_file_directory_conflict(struct index_state
*istate
,
1281 const struct cache_entry
*ce
,
1282 int pos
, int ok_to_replace
)
1287 * When ce is an "I am going away" entry, we allow it to be added
1289 if (ce
->ce_flags
& CE_REMOVE
)
1293 * We check if the path is a sub-path of a subsequent pathname
1294 * first, since removing those will not change the position
1297 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
1300 * Then check if the path might have a clashing sub-directory
1303 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
1306 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1309 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
1310 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
1311 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
1312 int new_only
= option
& ADD_CACHE_NEW_ONLY
;
1314 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1315 cache_tree_invalidate_path(istate
, ce
->name
);
1318 * If this entry's path sorts after the last entry in the index,
1319 * we can avoid searching for it.
1321 if (istate
->cache_nr
> 0 &&
1322 strcmp(ce
->name
, istate
->cache
[istate
->cache_nr
- 1]->name
) > 0)
1323 pos
= index_pos_to_insert_pos(istate
->cache_nr
);
1325 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1327 /* existing match? Just replace it. */
1330 replace_index_entry(istate
, pos
, ce
);
1335 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1336 untracked_cache_add_to_index(istate
, ce
->name
);
1339 * Inserting a merged entry ("stage 0") into the index
1340 * will always replace all non-merged entries..
1342 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
1343 while (ce_same_name(istate
->cache
[pos
], ce
)) {
1345 if (!remove_index_entry_at(istate
, pos
))
1352 if (!verify_path(ce
->name
, ce
->ce_mode
))
1353 return error(_("invalid path '%s'"), ce
->name
);
1355 if (!skip_df_check
&&
1356 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
1358 return error(_("'%s' appears as both a file and as a directory"),
1360 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1366 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1370 if (option
& ADD_CACHE_JUST_APPEND
)
1371 pos
= istate
->cache_nr
;
1374 ret
= add_index_entry_with_check(istate
, ce
, option
);
1380 /* Make sure the array is big enough .. */
1381 ALLOC_GROW(istate
->cache
, istate
->cache_nr
+ 1, istate
->cache_alloc
);
1385 if (istate
->cache_nr
> pos
+ 1)
1386 MOVE_ARRAY(istate
->cache
+ pos
+ 1, istate
->cache
+ pos
,
1387 istate
->cache_nr
- pos
- 1);
1388 set_index_entry(istate
, pos
, ce
);
1389 istate
->cache_changed
|= CE_ENTRY_ADDED
;
1394 * "refresh" does not calculate a new sha1 file or bring the
1395 * cache up-to-date for mode/content changes. But what it
1396 * _does_ do is to "re-match" the stat information of a file
1397 * with the cache, so that you can refresh the cache for a
1398 * file that hasn't been changed but where the stat entry is
1401 * For example, you'd want to do this after doing a "git-read-tree",
1402 * to link up the stat cache details with the proper files.
1404 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
1405 struct cache_entry
*ce
,
1406 unsigned int options
, int *err
,
1412 struct cache_entry
*updated
;
1414 int refresh
= options
& CE_MATCH_REFRESH
;
1415 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
1416 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
1417 int ignore_missing
= options
& CE_MATCH_IGNORE_MISSING
;
1418 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
1420 if (!refresh
|| ce_uptodate(ce
))
1423 if (!ignore_fsmonitor
)
1424 refresh_fsmonitor(istate
);
1426 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1427 * that the change to the work tree does not matter and told
1430 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
)) {
1431 ce_mark_uptodate(ce
);
1434 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
)) {
1435 ce_mark_uptodate(ce
);
1438 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
)) {
1439 ce_mark_uptodate(ce
);
1443 if (has_symlink_leading_path(ce
->name
, ce_namelen(ce
))) {
1453 if (lstat(ce
->name
, &st
) < 0) {
1454 if (ignore_missing
&& errno
== ENOENT
)
1461 changed
= ie_match_stat(istate
, ce
, &st
, options
);
1463 *changed_ret
= changed
;
1466 * The path is unchanged. If we were told to ignore
1467 * valid bit, then we did the actual stat check and
1468 * found that the entry is unmodified. If the entry
1469 * is not marked VALID, this is the place to mark it
1470 * valid again, under "assume unchanged" mode.
1472 if (ignore_valid
&& assume_unchanged
&&
1473 !(ce
->ce_flags
& CE_VALID
))
1474 ; /* mark this one VALID again */
1477 * We do not mark the index itself "modified"
1478 * because CE_UPTODATE flag is in-core only;
1479 * we are not going to write this change out.
1481 if (!S_ISGITLINK(ce
->ce_mode
)) {
1482 ce_mark_uptodate(ce
);
1483 mark_fsmonitor_valid(istate
, ce
);
1491 if (ie_modified(istate
, ce
, &st
, options
)) {
1497 updated
= make_empty_cache_entry(istate
, ce_namelen(ce
));
1498 copy_cache_entry(updated
, ce
);
1499 memcpy(updated
->name
, ce
->name
, ce
->ce_namelen
+ 1);
1500 fill_stat_cache_info(istate
, updated
, &st
);
1502 * If ignore_valid is not set, we should leave CE_VALID bit
1503 * alone. Otherwise, paths marked with --no-assume-unchanged
1504 * (i.e. things to be edited) will reacquire CE_VALID bit
1505 * automatically, which is not really what we want.
1507 if (!ignore_valid
&& assume_unchanged
&&
1508 !(ce
->ce_flags
& CE_VALID
))
1509 updated
->ce_flags
&= ~CE_VALID
;
1511 /* istate->cache_changed is updated in the caller */
1515 static void show_file(const char * fmt
, const char * name
, int in_porcelain
,
1516 int * first
, const char *header_msg
)
1518 if (in_porcelain
&& *first
&& header_msg
) {
1519 printf("%s\n", header_msg
);
1525 int repo_refresh_and_write_index(struct repository
*repo
,
1526 unsigned int refresh_flags
,
1527 unsigned int write_flags
,
1529 const struct pathspec
*pathspec
,
1530 char *seen
, const char *header_msg
)
1532 struct lock_file lock_file
= LOCK_INIT
;
1535 fd
= repo_hold_locked_index(repo
, &lock_file
, 0);
1536 if (!gentle
&& fd
< 0)
1538 if (refresh_index(repo
->index
, refresh_flags
, pathspec
, seen
, header_msg
))
1540 if (0 <= fd
&& write_locked_index(repo
->index
, &lock_file
, COMMIT_LOCK
| write_flags
))
1546 int refresh_index(struct index_state
*istate
, unsigned int flags
,
1547 const struct pathspec
*pathspec
,
1548 char *seen
, const char *header_msg
)
1552 int really
= (flags
& REFRESH_REALLY
) != 0;
1553 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
1554 int quiet
= (flags
& REFRESH_QUIET
) != 0;
1555 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
1556 int ignore_submodules
= (flags
& REFRESH_IGNORE_SUBMODULES
) != 0;
1557 int ignore_skip_worktree
= (flags
& REFRESH_IGNORE_SKIP_WORKTREE
) != 0;
1559 int in_porcelain
= (flags
& REFRESH_IN_PORCELAIN
);
1560 unsigned int options
= (CE_MATCH_REFRESH
|
1561 (really
? CE_MATCH_IGNORE_VALID
: 0) |
1562 (not_new
? CE_MATCH_IGNORE_MISSING
: 0));
1563 const char *modified_fmt
;
1564 const char *deleted_fmt
;
1565 const char *typechange_fmt
;
1566 const char *added_fmt
;
1567 const char *unmerged_fmt
;
1568 struct progress
*progress
= NULL
;
1569 int t2_sum_lstat
= 0;
1570 int t2_sum_scan
= 0;
1572 if (flags
& REFRESH_PROGRESS
&& isatty(2))
1573 progress
= start_delayed_progress(_("Refresh index"),
1576 trace_performance_enter();
1577 modified_fmt
= in_porcelain
? "M\t%s\n" : "%s: needs update\n";
1578 deleted_fmt
= in_porcelain
? "D\t%s\n" : "%s: needs update\n";
1579 typechange_fmt
= in_porcelain
? "T\t%s\n" : "%s: needs update\n";
1580 added_fmt
= in_porcelain
? "A\t%s\n" : "%s: needs update\n";
1581 unmerged_fmt
= in_porcelain
? "U\t%s\n" : "%s: needs merge\n";
1583 * Use the multi-threaded preload_index() to refresh most of the
1584 * cache entries quickly then in the single threaded loop below,
1585 * we only have to do the special cases that are left.
1587 preload_index(istate
, pathspec
, 0);
1588 trace2_region_enter("index", "refresh", NULL
);
1590 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1591 struct cache_entry
*ce
, *new_entry
;
1592 int cache_errno
= 0;
1595 int t2_did_lstat
= 0;
1596 int t2_did_scan
= 0;
1598 ce
= istate
->cache
[i
];
1599 if (ignore_submodules
&& S_ISGITLINK(ce
->ce_mode
))
1601 if (ignore_skip_worktree
&& ce_skip_worktree(ce
))
1605 * If this entry is a sparse directory, then there isn't
1606 * any stat() information to update. Ignore the entry.
1608 if (S_ISSPARSEDIR(ce
->ce_mode
))
1611 if (pathspec
&& !ce_path_match(istate
, ce
, pathspec
, seen
))
1615 while ((i
< istate
->cache_nr
) &&
1616 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
1622 show_file(unmerged_fmt
, ce
->name
, in_porcelain
,
1623 &first
, header_msg
);
1631 new_entry
= refresh_cache_ent(istate
, ce
, options
,
1632 &cache_errno
, &changed
,
1633 &t2_did_lstat
, &t2_did_scan
);
1634 t2_sum_lstat
+= t2_did_lstat
;
1635 t2_sum_scan
+= t2_did_scan
;
1636 if (new_entry
== ce
)
1638 display_progress(progress
, i
);
1642 if (really
&& cache_errno
== EINVAL
) {
1643 /* If we are doing --really-refresh that
1644 * means the index is not valid anymore.
1646 ce
->ce_flags
&= ~CE_VALID
;
1647 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
1648 mark_fsmonitor_invalid(istate
, ce
);
1649 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
1654 if (cache_errno
== ENOENT
)
1656 else if (ce_intent_to_add(ce
))
1657 fmt
= added_fmt
; /* must be before other checks */
1658 else if (changed
& TYPE_CHANGED
)
1659 fmt
= typechange_fmt
;
1663 ce
->name
, in_porcelain
, &first
, header_msg
);
1668 replace_index_entry(istate
, i
, new_entry
);
1670 trace2_data_intmax("index", NULL
, "refresh/sum_lstat", t2_sum_lstat
);
1671 trace2_data_intmax("index", NULL
, "refresh/sum_scan", t2_sum_scan
);
1672 trace2_region_leave("index", "refresh", NULL
);
1673 display_progress(progress
, istate
->cache_nr
);
1674 stop_progress(&progress
);
1675 trace_performance_leave("refresh index");
1679 struct cache_entry
*refresh_cache_entry(struct index_state
*istate
,
1680 struct cache_entry
*ce
,
1681 unsigned int options
)
1683 return refresh_cache_ent(istate
, ce
, options
, NULL
, NULL
, NULL
, NULL
);
1687 /*****************************************************************
1689 *****************************************************************/
1691 #define INDEX_FORMAT_DEFAULT 3
1693 static unsigned int get_index_format_default(struct repository
*r
)
1695 char *envversion
= getenv("GIT_INDEX_VERSION");
1697 unsigned int version
= INDEX_FORMAT_DEFAULT
;
1700 prepare_repo_settings(r
);
1702 if (r
->settings
.index_version
>= 0)
1703 version
= r
->settings
.index_version
;
1704 if (version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1705 warning(_("index.version set, but the value is invalid.\n"
1706 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1707 return INDEX_FORMAT_DEFAULT
;
1712 version
= strtoul(envversion
, &endp
, 10);
1714 version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1715 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1716 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1717 version
= INDEX_FORMAT_DEFAULT
;
1723 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1724 * Again - this is just a (very strong in practice) heuristic that
1725 * the inode hasn't changed.
1727 * We save the fields in big-endian order to allow using the
1728 * index file over NFS transparently.
1730 struct ondisk_cache_entry
{
1731 struct cache_time ctime
;
1732 struct cache_time mtime
;
1740 * unsigned char hash[hashsz];
1742 * if (flags & CE_EXTENDED)
1745 unsigned char data
[GIT_MAX_RAWSZ
+ 2 * sizeof(uint16_t)];
1746 char name
[FLEX_ARRAY
];
1749 /* These are only used for v3 or lower */
1750 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1751 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,data) + (len) + 8) & ~7)
1752 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1753 #define ondisk_data_size(flags, len) (the_hash_algo->rawsz + \
1754 ((flags & CE_EXTENDED) ? 2 : 1) * sizeof(uint16_t) + len)
1755 #define ondisk_data_size_max(len) (ondisk_data_size(CE_EXTENDED, len))
1756 #define ondisk_ce_size(ce) (ondisk_cache_entry_size(ondisk_data_size((ce)->ce_flags, ce_namelen(ce))))
1758 /* Allow fsck to force verification of the index checksum. */
1759 int verify_index_checksum
;
1761 /* Allow fsck to force verification of the cache entry order. */
1762 int verify_ce_order
;
1764 static int verify_hdr(const struct cache_header
*hdr
, unsigned long size
)
1767 unsigned char hash
[GIT_MAX_RAWSZ
];
1770 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
1771 return error(_("bad signature 0x%08x"), hdr
->hdr_signature
);
1772 hdr_version
= ntohl(hdr
->hdr_version
);
1773 if (hdr_version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< hdr_version
)
1774 return error(_("bad index version %d"), hdr_version
);
1776 if (!verify_index_checksum
)
1779 the_hash_algo
->init_fn(&c
);
1780 the_hash_algo
->update_fn(&c
, hdr
, size
- the_hash_algo
->rawsz
);
1781 the_hash_algo
->final_fn(hash
, &c
);
1782 if (!hasheq(hash
, (unsigned char *)hdr
+ size
- the_hash_algo
->rawsz
))
1783 return error(_("bad index file sha1 signature"));
1787 static int read_index_extension(struct index_state
*istate
,
1788 const char *ext
, const char *data
, unsigned long sz
)
1790 switch (CACHE_EXT(ext
)) {
1791 case CACHE_EXT_TREE
:
1792 istate
->cache_tree
= cache_tree_read(data
, sz
);
1794 case CACHE_EXT_RESOLVE_UNDO
:
1795 istate
->resolve_undo
= resolve_undo_read(data
, sz
);
1797 case CACHE_EXT_LINK
:
1798 if (read_link_extension(istate
, data
, sz
))
1801 case CACHE_EXT_UNTRACKED
:
1802 istate
->untracked
= read_untracked_extension(data
, sz
);
1804 case CACHE_EXT_FSMONITOR
:
1805 read_fsmonitor_extension(istate
, data
, sz
);
1807 case CACHE_EXT_ENDOFINDEXENTRIES
:
1808 case CACHE_EXT_INDEXENTRYOFFSETTABLE
:
1809 /* already handled in do_read_index() */
1811 case CACHE_EXT_SPARSE_DIRECTORIES
:
1812 /* no content, only an indicator */
1813 istate
->sparse_index
= 1;
1816 if (*ext
< 'A' || 'Z' < *ext
)
1817 return error(_("index uses %.4s extension, which we do not understand"),
1819 fprintf_ln(stderr
, _("ignoring %.4s extension"), ext
);
1825 static struct cache_entry
*create_from_disk(struct mem_pool
*ce_mem_pool
,
1826 unsigned int version
,
1827 struct ondisk_cache_entry
*ondisk
,
1828 unsigned long *ent_size
,
1829 const struct cache_entry
*previous_ce
)
1831 struct cache_entry
*ce
;
1834 const unsigned hashsz
= the_hash_algo
->rawsz
;
1835 const uint16_t *flagsp
= (const uint16_t *)(ondisk
->data
+ hashsz
);
1837 size_t copy_len
= 0;
1839 * Adjacent cache entries tend to share the leading paths, so it makes
1840 * sense to only store the differences in later entries. In the v4
1841 * on-disk format of the index, each on-disk cache entry stores the
1842 * number of bytes to be stripped from the end of the previous name,
1843 * and the bytes to append to the result, to come up with its name.
1845 int expand_name_field
= version
== 4;
1847 /* On-disk flags are just 16 bits */
1848 flags
= get_be16(flagsp
);
1849 len
= flags
& CE_NAMEMASK
;
1851 if (flags
& CE_EXTENDED
) {
1853 extended_flags
= get_be16(flagsp
+ 1) << 16;
1854 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1855 if (extended_flags
& ~CE_EXTENDED_FLAGS
)
1856 die(_("unknown index entry format 0x%08x"), extended_flags
);
1857 flags
|= extended_flags
;
1858 name
= (const char *)(flagsp
+ 2);
1861 name
= (const char *)(flagsp
+ 1);
1863 if (expand_name_field
) {
1864 const unsigned char *cp
= (const unsigned char *)name
;
1865 size_t strip_len
, previous_len
;
1867 /* If we're at the beginning of a block, ignore the previous name */
1868 strip_len
= decode_varint(&cp
);
1870 previous_len
= previous_ce
->ce_namelen
;
1871 if (previous_len
< strip_len
)
1872 die(_("malformed name field in the index, near path '%s'"),
1874 copy_len
= previous_len
- strip_len
;
1876 name
= (const char *)cp
;
1879 if (len
== CE_NAMEMASK
) {
1881 if (expand_name_field
)
1885 ce
= mem_pool__ce_alloc(ce_mem_pool
, len
);
1887 ce
->ce_stat_data
.sd_ctime
.sec
= get_be32(&ondisk
->ctime
.sec
);
1888 ce
->ce_stat_data
.sd_mtime
.sec
= get_be32(&ondisk
->mtime
.sec
);
1889 ce
->ce_stat_data
.sd_ctime
.nsec
= get_be32(&ondisk
->ctime
.nsec
);
1890 ce
->ce_stat_data
.sd_mtime
.nsec
= get_be32(&ondisk
->mtime
.nsec
);
1891 ce
->ce_stat_data
.sd_dev
= get_be32(&ondisk
->dev
);
1892 ce
->ce_stat_data
.sd_ino
= get_be32(&ondisk
->ino
);
1893 ce
->ce_mode
= get_be32(&ondisk
->mode
);
1894 ce
->ce_stat_data
.sd_uid
= get_be32(&ondisk
->uid
);
1895 ce
->ce_stat_data
.sd_gid
= get_be32(&ondisk
->gid
);
1896 ce
->ce_stat_data
.sd_size
= get_be32(&ondisk
->size
);
1897 ce
->ce_flags
= flags
& ~CE_NAMEMASK
;
1898 ce
->ce_namelen
= len
;
1900 oidread(&ce
->oid
, ondisk
->data
);
1901 memcpy(ce
->name
, name
, len
);
1902 ce
->name
[len
] = '\0';
1904 if (expand_name_field
) {
1906 memcpy(ce
->name
, previous_ce
->name
, copy_len
);
1907 memcpy(ce
->name
+ copy_len
, name
, len
+ 1 - copy_len
);
1908 *ent_size
= (name
- ((char *)ondisk
)) + len
+ 1 - copy_len
;
1910 memcpy(ce
->name
, name
, len
+ 1);
1911 *ent_size
= ondisk_ce_size(ce
);
1916 static void check_ce_order(struct index_state
*istate
)
1920 if (!verify_ce_order
)
1923 for (i
= 1; i
< istate
->cache_nr
; i
++) {
1924 struct cache_entry
*ce
= istate
->cache
[i
- 1];
1925 struct cache_entry
*next_ce
= istate
->cache
[i
];
1926 int name_compare
= strcmp(ce
->name
, next_ce
->name
);
1928 if (0 < name_compare
)
1929 die(_("unordered stage entries in index"));
1930 if (!name_compare
) {
1932 die(_("multiple stage entries for merged file '%s'"),
1934 if (ce_stage(ce
) > ce_stage(next_ce
))
1935 die(_("unordered stage entries for '%s'"),
1941 static void tweak_untracked_cache(struct index_state
*istate
)
1943 struct repository
*r
= the_repository
;
1945 prepare_repo_settings(r
);
1947 switch (r
->settings
.core_untracked_cache
) {
1948 case UNTRACKED_CACHE_REMOVE
:
1949 remove_untracked_cache(istate
);
1951 case UNTRACKED_CACHE_WRITE
:
1952 add_untracked_cache(istate
);
1954 case UNTRACKED_CACHE_KEEP
:
1956 * Either an explicit "core.untrackedCache=keep", the
1957 * default if "core.untrackedCache" isn't configured,
1958 * or a fallback on an unknown "core.untrackedCache"
1965 static void tweak_split_index(struct index_state
*istate
)
1967 switch (git_config_get_split_index()) {
1968 case -1: /* unset: do nothing */
1971 remove_split_index(istate
);
1974 add_split_index(istate
);
1976 default: /* unknown value: do nothing */
1981 static void post_read_index_from(struct index_state
*istate
)
1983 check_ce_order(istate
);
1984 tweak_untracked_cache(istate
);
1985 tweak_split_index(istate
);
1986 tweak_fsmonitor(istate
);
1989 static size_t estimate_cache_size_from_compressed(unsigned int entries
)
1991 return entries
* (sizeof(struct cache_entry
) + CACHE_ENTRY_PATH_LENGTH
);
1994 static size_t estimate_cache_size(size_t ondisk_size
, unsigned int entries
)
1996 long per_entry
= sizeof(struct cache_entry
) - sizeof(struct ondisk_cache_entry
);
1999 * Account for potential alignment differences.
2001 per_entry
+= align_padding_size(per_entry
, 0);
2002 return ondisk_size
+ entries
* per_entry
;
2005 struct index_entry_offset
2007 /* starting byte offset into index file, count of index entries in this block */
2011 struct index_entry_offset_table
2014 struct index_entry_offset entries
[FLEX_ARRAY
];
2017 static struct index_entry_offset_table
*read_ieot_extension(const char *mmap
, size_t mmap_size
, size_t offset
);
2018 static void write_ieot_extension(struct strbuf
*sb
, struct index_entry_offset_table
*ieot
);
2020 static size_t read_eoie_extension(const char *mmap
, size_t mmap_size
);
2021 static void write_eoie_extension(struct strbuf
*sb
, git_hash_ctx
*eoie_context
, size_t offset
);
2023 struct load_index_extensions
2026 struct index_state
*istate
;
2029 unsigned long src_offset
;
2032 static void *load_index_extensions(void *_data
)
2034 struct load_index_extensions
*p
= _data
;
2035 unsigned long src_offset
= p
->src_offset
;
2037 while (src_offset
<= p
->mmap_size
- the_hash_algo
->rawsz
- 8) {
2038 /* After an array of active_nr index entries,
2039 * there can be arbitrary number of extended
2040 * sections, each of which is prefixed with
2041 * extension name (4-byte) and section length
2042 * in 4-byte network byte order.
2044 uint32_t extsize
= get_be32(p
->mmap
+ src_offset
+ 4);
2045 if (read_index_extension(p
->istate
,
2046 p
->mmap
+ src_offset
,
2047 p
->mmap
+ src_offset
+ 8,
2049 munmap((void *)p
->mmap
, p
->mmap_size
);
2050 die(_("index file corrupt"));
2053 src_offset
+= extsize
;
2060 * A helper function that will load the specified range of cache entries
2061 * from the memory mapped file and add them to the given index.
2063 static unsigned long load_cache_entry_block(struct index_state
*istate
,
2064 struct mem_pool
*ce_mem_pool
, int offset
, int nr
, const char *mmap
,
2065 unsigned long start_offset
, const struct cache_entry
*previous_ce
)
2068 unsigned long src_offset
= start_offset
;
2070 for (i
= offset
; i
< offset
+ nr
; i
++) {
2071 struct ondisk_cache_entry
*disk_ce
;
2072 struct cache_entry
*ce
;
2073 unsigned long consumed
;
2075 disk_ce
= (struct ondisk_cache_entry
*)(mmap
+ src_offset
);
2076 ce
= create_from_disk(ce_mem_pool
, istate
->version
, disk_ce
, &consumed
, previous_ce
);
2077 set_index_entry(istate
, i
, ce
);
2079 src_offset
+= consumed
;
2082 return src_offset
- start_offset
;
2085 static unsigned long load_all_cache_entries(struct index_state
*istate
,
2086 const char *mmap
, size_t mmap_size
, unsigned long src_offset
)
2088 unsigned long consumed
;
2090 istate
->ce_mem_pool
= xmalloc(sizeof(*istate
->ce_mem_pool
));
2091 if (istate
->version
== 4) {
2092 mem_pool_init(istate
->ce_mem_pool
,
2093 estimate_cache_size_from_compressed(istate
->cache_nr
));
2095 mem_pool_init(istate
->ce_mem_pool
,
2096 estimate_cache_size(mmap_size
, istate
->cache_nr
));
2099 consumed
= load_cache_entry_block(istate
, istate
->ce_mem_pool
,
2100 0, istate
->cache_nr
, mmap
, src_offset
, NULL
);
2105 * Mostly randomly chosen maximum thread counts: we
2106 * cap the parallelism to online_cpus() threads, and we want
2107 * to have at least 10000 cache entries per thread for it to
2108 * be worth starting a thread.
2111 #define THREAD_COST (10000)
2113 struct load_cache_entries_thread_data
2116 struct index_state
*istate
;
2117 struct mem_pool
*ce_mem_pool
;
2120 struct index_entry_offset_table
*ieot
;
2121 int ieot_start
; /* starting index into the ieot array */
2122 int ieot_blocks
; /* count of ieot entries to process */
2123 unsigned long consumed
; /* return # of bytes in index file processed */
2127 * A thread proc to run the load_cache_entries() computation
2128 * across multiple background threads.
2130 static void *load_cache_entries_thread(void *_data
)
2132 struct load_cache_entries_thread_data
*p
= _data
;
2135 /* iterate across all ieot blocks assigned to this thread */
2136 for (i
= p
->ieot_start
; i
< p
->ieot_start
+ p
->ieot_blocks
; i
++) {
2137 p
->consumed
+= load_cache_entry_block(p
->istate
, p
->ce_mem_pool
,
2138 p
->offset
, p
->ieot
->entries
[i
].nr
, p
->mmap
, p
->ieot
->entries
[i
].offset
, NULL
);
2139 p
->offset
+= p
->ieot
->entries
[i
].nr
;
2144 static unsigned long load_cache_entries_threaded(struct index_state
*istate
, const char *mmap
, size_t mmap_size
,
2145 int nr_threads
, struct index_entry_offset_table
*ieot
)
2147 int i
, offset
, ieot_blocks
, ieot_start
, err
;
2148 struct load_cache_entries_thread_data
*data
;
2149 unsigned long consumed
= 0;
2151 /* a little sanity checking */
2152 if (istate
->name_hash_initialized
)
2153 BUG("the name hash isn't thread safe");
2155 istate
->ce_mem_pool
= xmalloc(sizeof(*istate
->ce_mem_pool
));
2156 mem_pool_init(istate
->ce_mem_pool
, 0);
2158 /* ensure we have no more threads than we have blocks to process */
2159 if (nr_threads
> ieot
->nr
)
2160 nr_threads
= ieot
->nr
;
2161 CALLOC_ARRAY(data
, nr_threads
);
2163 offset
= ieot_start
= 0;
2164 ieot_blocks
= DIV_ROUND_UP(ieot
->nr
, nr_threads
);
2165 for (i
= 0; i
< nr_threads
; i
++) {
2166 struct load_cache_entries_thread_data
*p
= &data
[i
];
2169 if (ieot_start
+ ieot_blocks
> ieot
->nr
)
2170 ieot_blocks
= ieot
->nr
- ieot_start
;
2176 p
->ieot_start
= ieot_start
;
2177 p
->ieot_blocks
= ieot_blocks
;
2179 /* create a mem_pool for each thread */
2181 for (j
= p
->ieot_start
; j
< p
->ieot_start
+ p
->ieot_blocks
; j
++)
2182 nr
+= p
->ieot
->entries
[j
].nr
;
2183 p
->ce_mem_pool
= xmalloc(sizeof(*istate
->ce_mem_pool
));
2184 if (istate
->version
== 4) {
2185 mem_pool_init(p
->ce_mem_pool
,
2186 estimate_cache_size_from_compressed(nr
));
2188 mem_pool_init(p
->ce_mem_pool
,
2189 estimate_cache_size(mmap_size
, nr
));
2192 err
= pthread_create(&p
->pthread
, NULL
, load_cache_entries_thread
, p
);
2194 die(_("unable to create load_cache_entries thread: %s"), strerror(err
));
2196 /* increment by the number of cache entries in the ieot block being processed */
2197 for (j
= 0; j
< ieot_blocks
; j
++)
2198 offset
+= ieot
->entries
[ieot_start
+ j
].nr
;
2199 ieot_start
+= ieot_blocks
;
2202 for (i
= 0; i
< nr_threads
; i
++) {
2203 struct load_cache_entries_thread_data
*p
= &data
[i
];
2205 err
= pthread_join(p
->pthread
, NULL
);
2207 die(_("unable to join load_cache_entries thread: %s"), strerror(err
));
2208 mem_pool_combine(istate
->ce_mem_pool
, p
->ce_mem_pool
);
2209 consumed
+= p
->consumed
;
2217 /* remember to discard_cache() before reading a different cache! */
2218 int do_read_index(struct index_state
*istate
, const char *path
, int must_exist
)
2222 unsigned long src_offset
;
2223 const struct cache_header
*hdr
;
2226 struct load_index_extensions p
;
2227 size_t extension_offset
= 0;
2228 int nr_threads
, cpus
;
2229 struct index_entry_offset_table
*ieot
= NULL
;
2231 if (istate
->initialized
)
2232 return istate
->cache_nr
;
2234 istate
->timestamp
.sec
= 0;
2235 istate
->timestamp
.nsec
= 0;
2236 fd
= open(path
, O_RDONLY
);
2238 if (!must_exist
&& errno
== ENOENT
)
2240 die_errno(_("%s: index file open failed"), path
);
2244 die_errno(_("%s: cannot stat the open index"), path
);
2246 mmap_size
= xsize_t(st
.st_size
);
2247 if (mmap_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2248 die(_("%s: index file smaller than expected"), path
);
2250 mmap
= xmmap_gently(NULL
, mmap_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2251 if (mmap
== MAP_FAILED
)
2252 die_errno(_("%s: unable to map index file%s"), path
,
2256 hdr
= (const struct cache_header
*)mmap
;
2257 if (verify_hdr(hdr
, mmap_size
) < 0)
2260 oidread(&istate
->oid
, (const unsigned char *)hdr
+ mmap_size
- the_hash_algo
->rawsz
);
2261 istate
->version
= ntohl(hdr
->hdr_version
);
2262 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
2263 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
2264 CALLOC_ARRAY(istate
->cache
, istate
->cache_alloc
);
2265 istate
->initialized
= 1;
2269 p
.mmap_size
= mmap_size
;
2271 src_offset
= sizeof(*hdr
);
2273 if (git_config_get_index_threads(&nr_threads
))
2276 /* TODO: does creating more threads than cores help? */
2278 nr_threads
= istate
->cache_nr
/ THREAD_COST
;
2279 cpus
= online_cpus();
2280 if (nr_threads
> cpus
)
2287 if (nr_threads
> 1) {
2288 extension_offset
= read_eoie_extension(mmap
, mmap_size
);
2289 if (extension_offset
) {
2292 p
.src_offset
= extension_offset
;
2293 err
= pthread_create(&p
.pthread
, NULL
, load_index_extensions
, &p
);
2295 die(_("unable to create load_index_extensions thread: %s"), strerror(err
));
2302 * Locate and read the index entry offset table so that we can use it
2303 * to multi-thread the reading of the cache entries.
2305 if (extension_offset
&& nr_threads
> 1)
2306 ieot
= read_ieot_extension(mmap
, mmap_size
, extension_offset
);
2309 src_offset
+= load_cache_entries_threaded(istate
, mmap
, mmap_size
, nr_threads
, ieot
);
2312 src_offset
+= load_all_cache_entries(istate
, mmap
, mmap_size
, src_offset
);
2315 istate
->timestamp
.sec
= st
.st_mtime
;
2316 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
2318 /* if we created a thread, join it otherwise load the extensions on the primary thread */
2319 if (extension_offset
) {
2320 int ret
= pthread_join(p
.pthread
, NULL
);
2322 die(_("unable to join load_index_extensions thread: %s"), strerror(ret
));
2324 p
.src_offset
= src_offset
;
2325 load_index_extensions(&p
);
2327 munmap((void *)mmap
, mmap_size
);
2330 * TODO trace2: replace "the_repository" with the actual repo instance
2331 * that is associated with the given "istate".
2333 trace2_data_intmax("index", the_repository
, "read/version",
2335 trace2_data_intmax("index", the_repository
, "read/cache_nr",
2339 istate
->repo
= the_repository
;
2340 prepare_repo_settings(istate
->repo
);
2341 if (istate
->repo
->settings
.command_requires_full_index
)
2342 ensure_full_index(istate
);
2344 return istate
->cache_nr
;
2347 munmap((void *)mmap
, mmap_size
);
2348 die(_("index file corrupt"));
2352 * Signal that the shared index is used by updating its mtime.
2354 * This way, shared index can be removed if they have not been used
2357 static void freshen_shared_index(const char *shared_index
, int warn
)
2359 if (!check_and_freshen_file(shared_index
, 1) && warn
)
2360 warning(_("could not freshen shared index '%s'"), shared_index
);
2363 int read_index_from(struct index_state
*istate
, const char *path
,
2366 struct split_index
*split_index
;
2371 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2372 if (istate
->initialized
)
2373 return istate
->cache_nr
;
2376 * TODO trace2: replace "the_repository" with the actual repo instance
2377 * that is associated with the given "istate".
2379 trace2_region_enter_printf("index", "do_read_index", the_repository
,
2381 trace_performance_enter();
2382 ret
= do_read_index(istate
, path
, 0);
2383 trace_performance_leave("read cache %s", path
);
2384 trace2_region_leave_printf("index", "do_read_index", the_repository
,
2387 split_index
= istate
->split_index
;
2388 if (!split_index
|| is_null_oid(&split_index
->base_oid
)) {
2389 post_read_index_from(istate
);
2393 trace_performance_enter();
2394 if (split_index
->base
)
2395 discard_index(split_index
->base
);
2397 CALLOC_ARRAY(split_index
->base
, 1);
2399 base_oid_hex
= oid_to_hex(&split_index
->base_oid
);
2400 base_path
= xstrfmt("%s/sharedindex.%s", gitdir
, base_oid_hex
);
2401 trace2_region_enter_printf("index", "shared/do_read_index",
2402 the_repository
, "%s", base_path
);
2403 ret
= do_read_index(split_index
->base
, base_path
, 0);
2404 trace2_region_leave_printf("index", "shared/do_read_index",
2405 the_repository
, "%s", base_path
);
2407 char *path_copy
= xstrdup(path
);
2408 const char *base_path2
= xstrfmt("%s/sharedindex.%s",
2412 trace2_region_enter_printf("index", "shared/do_read_index",
2413 the_repository
, "%s", base_path2
);
2414 ret
= do_read_index(split_index
->base
, base_path2
, 1);
2415 trace2_region_leave_printf("index", "shared/do_read_index",
2416 the_repository
, "%s", base_path2
);
2418 if (!oideq(&split_index
->base_oid
, &split_index
->base
->oid
))
2419 die(_("broken index, expect %s in %s, got %s"),
2420 base_oid_hex
, base_path
,
2421 oid_to_hex(&split_index
->base
->oid
));
2423 freshen_shared_index(base_path
, 0);
2424 merge_base_index(istate
);
2425 post_read_index_from(istate
);
2426 trace_performance_leave("read cache %s", base_path
);
2431 int is_index_unborn(struct index_state
*istate
)
2433 return (!istate
->cache_nr
&& !istate
->timestamp
.sec
);
2436 int discard_index(struct index_state
*istate
)
2439 * Cache entries in istate->cache[] should have been allocated
2440 * from the memory pool associated with this index, or from an
2441 * associated split_index. There is no need to free individual
2442 * cache entries. validate_cache_entries can detect when this
2443 * assertion does not hold.
2445 validate_cache_entries(istate
);
2447 resolve_undo_clear_index(istate
);
2448 istate
->cache_nr
= 0;
2449 istate
->cache_changed
= 0;
2450 istate
->timestamp
.sec
= 0;
2451 istate
->timestamp
.nsec
= 0;
2452 free_name_hash(istate
);
2453 cache_tree_free(&(istate
->cache_tree
));
2454 istate
->initialized
= 0;
2455 istate
->fsmonitor_has_run_once
= 0;
2456 FREE_AND_NULL(istate
->fsmonitor_last_update
);
2457 FREE_AND_NULL(istate
->cache
);
2458 istate
->cache_alloc
= 0;
2459 discard_split_index(istate
);
2460 free_untracked_cache(istate
->untracked
);
2461 istate
->untracked
= NULL
;
2463 if (istate
->ce_mem_pool
) {
2464 mem_pool_discard(istate
->ce_mem_pool
, should_validate_cache_entries());
2465 FREE_AND_NULL(istate
->ce_mem_pool
);
2472 * Validate the cache entries of this index.
2473 * All cache entries associated with this index
2474 * should have been allocated by the memory pool
2475 * associated with this index, or by a referenced
2478 void validate_cache_entries(const struct index_state
*istate
)
2482 if (!should_validate_cache_entries() ||!istate
|| !istate
->initialized
)
2485 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2487 BUG("cache entry is not allocated from expected memory pool");
2488 } else if (!istate
->ce_mem_pool
||
2489 !mem_pool_contains(istate
->ce_mem_pool
, istate
->cache
[i
])) {
2490 if (!istate
->split_index
||
2491 !istate
->split_index
->base
||
2492 !istate
->split_index
->base
->ce_mem_pool
||
2493 !mem_pool_contains(istate
->split_index
->base
->ce_mem_pool
, istate
->cache
[i
])) {
2494 BUG("cache entry is not allocated from expected memory pool");
2499 if (istate
->split_index
)
2500 validate_cache_entries(istate
->split_index
->base
);
2503 int unmerged_index(const struct index_state
*istate
)
2506 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2507 if (ce_stage(istate
->cache
[i
]))
2513 int repo_index_has_changes(struct repository
*repo
,
2517 struct index_state
*istate
= repo
->index
;
2518 struct object_id cmp
;
2522 cmp
= tree
->object
.oid
;
2523 if (tree
|| !get_oid_tree("HEAD", &cmp
)) {
2524 struct diff_options opt
;
2526 repo_diff_setup(repo
, &opt
);
2527 opt
.flags
.exit_with_status
= 1;
2529 opt
.flags
.quick
= 1;
2530 diff_setup_done(&opt
);
2531 do_diff_cache(&cmp
, &opt
);
2533 for (i
= 0; sb
&& i
< diff_queued_diff
.nr
; i
++) {
2535 strbuf_addch(sb
, ' ');
2536 strbuf_addstr(sb
, diff_queued_diff
.queue
[i
]->two
->path
);
2539 return opt
.flags
.has_changes
!= 0;
2541 /* TODO: audit for interaction with sparse-index. */
2542 ensure_full_index(istate
);
2543 for (i
= 0; sb
&& i
< istate
->cache_nr
; i
++) {
2545 strbuf_addch(sb
, ' ');
2546 strbuf_addstr(sb
, istate
->cache
[i
]->name
);
2548 return !!istate
->cache_nr
;
2552 static int write_index_ext_header(struct hashfile
*f
,
2553 git_hash_ctx
*eoie_f
,
2557 hashwrite_be32(f
, ext
);
2558 hashwrite_be32(f
, sz
);
2563 the_hash_algo
->update_fn(eoie_f
, &ext
, sizeof(ext
));
2564 the_hash_algo
->update_fn(eoie_f
, &sz
, sizeof(sz
));
2569 static void ce_smudge_racily_clean_entry(struct index_state
*istate
,
2570 struct cache_entry
*ce
)
2573 * The only thing we care about in this function is to smudge the
2574 * falsely clean entry due to touch-update-touch race, so we leave
2575 * everything else as they are. We are called for entries whose
2576 * ce_stat_data.sd_mtime match the index file mtime.
2578 * Note that this actually does not do much for gitlinks, for
2579 * which ce_match_stat_basic() always goes to the actual
2580 * contents. The caller checks with is_racy_timestamp() which
2581 * always says "no" for gitlinks, so we are not called for them ;-)
2585 if (lstat(ce
->name
, &st
) < 0)
2587 if (ce_match_stat_basic(ce
, &st
))
2589 if (ce_modified_check_fs(istate
, ce
, &st
)) {
2590 /* This is "racily clean"; smudge it. Note that this
2591 * is a tricky code. At first glance, it may appear
2592 * that it can break with this sequence:
2594 * $ echo xyzzy >frotz
2595 * $ git-update-index --add frotz
2598 * $ echo filfre >nitfol
2599 * $ git-update-index --add nitfol
2601 * but it does not. When the second update-index runs,
2602 * it notices that the entry "frotz" has the same timestamp
2603 * as index, and if we were to smudge it by resetting its
2604 * size to zero here, then the object name recorded
2605 * in index is the 6-byte file but the cached stat information
2606 * becomes zero --- which would then match what we would
2607 * obtain from the filesystem next time we stat("frotz").
2609 * However, the second update-index, before calling
2610 * this function, notices that the cached size is 6
2611 * bytes and what is on the filesystem is an empty
2612 * file, and never calls us, so the cached size information
2613 * for "frotz" stays 6 which does not match the filesystem.
2615 ce
->ce_stat_data
.sd_size
= 0;
2619 /* Copy miscellaneous fields but not the name */
2620 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry
*ondisk
,
2621 struct cache_entry
*ce
)
2624 const unsigned hashsz
= the_hash_algo
->rawsz
;
2625 uint16_t *flagsp
= (uint16_t *)(ondisk
->data
+ hashsz
);
2627 ondisk
->ctime
.sec
= htonl(ce
->ce_stat_data
.sd_ctime
.sec
);
2628 ondisk
->mtime
.sec
= htonl(ce
->ce_stat_data
.sd_mtime
.sec
);
2629 ondisk
->ctime
.nsec
= htonl(ce
->ce_stat_data
.sd_ctime
.nsec
);
2630 ondisk
->mtime
.nsec
= htonl(ce
->ce_stat_data
.sd_mtime
.nsec
);
2631 ondisk
->dev
= htonl(ce
->ce_stat_data
.sd_dev
);
2632 ondisk
->ino
= htonl(ce
->ce_stat_data
.sd_ino
);
2633 ondisk
->mode
= htonl(ce
->ce_mode
);
2634 ondisk
->uid
= htonl(ce
->ce_stat_data
.sd_uid
);
2635 ondisk
->gid
= htonl(ce
->ce_stat_data
.sd_gid
);
2636 ondisk
->size
= htonl(ce
->ce_stat_data
.sd_size
);
2637 hashcpy(ondisk
->data
, ce
->oid
.hash
);
2639 flags
= ce
->ce_flags
& ~CE_NAMEMASK
;
2640 flags
|= (ce_namelen(ce
) >= CE_NAMEMASK
? CE_NAMEMASK
: ce_namelen(ce
));
2641 flagsp
[0] = htons(flags
);
2642 if (ce
->ce_flags
& CE_EXTENDED
) {
2643 flagsp
[1] = htons((ce
->ce_flags
& CE_EXTENDED_FLAGS
) >> 16);
2647 static int ce_write_entry(struct hashfile
*f
, struct cache_entry
*ce
,
2648 struct strbuf
*previous_name
, struct ondisk_cache_entry
*ondisk
)
2651 unsigned int saved_namelen
;
2652 int stripped_name
= 0;
2653 static unsigned char padding
[8] = { 0x00 };
2655 if (ce
->ce_flags
& CE_STRIP_NAME
) {
2656 saved_namelen
= ce_namelen(ce
);
2661 size
= offsetof(struct ondisk_cache_entry
,data
) + ondisk_data_size(ce
->ce_flags
, 0);
2663 if (!previous_name
) {
2664 int len
= ce_namelen(ce
);
2665 copy_cache_entry_to_ondisk(ondisk
, ce
);
2666 hashwrite(f
, ondisk
, size
);
2667 hashwrite(f
, ce
->name
, len
);
2668 hashwrite(f
, padding
, align_padding_size(size
, len
));
2670 int common
, to_remove
, prefix_size
;
2671 unsigned char to_remove_vi
[16];
2673 (ce
->name
[common
] &&
2674 common
< previous_name
->len
&&
2675 ce
->name
[common
] == previous_name
->buf
[common
]);
2677 ; /* still matching */
2678 to_remove
= previous_name
->len
- common
;
2679 prefix_size
= encode_varint(to_remove
, to_remove_vi
);
2681 copy_cache_entry_to_ondisk(ondisk
, ce
);
2682 hashwrite(f
, ondisk
, size
);
2683 hashwrite(f
, to_remove_vi
, prefix_size
);
2684 hashwrite(f
, ce
->name
+ common
, ce_namelen(ce
) - common
);
2685 hashwrite(f
, padding
, 1);
2687 strbuf_splice(previous_name
, common
, to_remove
,
2688 ce
->name
+ common
, ce_namelen(ce
) - common
);
2690 if (stripped_name
) {
2691 ce
->ce_namelen
= saved_namelen
;
2692 ce
->ce_flags
&= ~CE_STRIP_NAME
;
2699 * This function verifies if index_state has the correct sha1 of the
2700 * index file. Don't die if we have any other failure, just return 0.
2702 static int verify_index_from(const struct index_state
*istate
, const char *path
)
2707 unsigned char hash
[GIT_MAX_RAWSZ
];
2709 if (!istate
->initialized
)
2712 fd
= open(path
, O_RDONLY
);
2719 if (st
.st_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2722 n
= pread_in_full(fd
, hash
, the_hash_algo
->rawsz
, st
.st_size
- the_hash_algo
->rawsz
);
2723 if (n
!= the_hash_algo
->rawsz
)
2726 if (!hasheq(istate
->oid
.hash
, hash
))
2737 static int repo_verify_index(struct repository
*repo
)
2739 return verify_index_from(repo
->index
, repo
->index_file
);
2742 static int has_racy_timestamp(struct index_state
*istate
)
2744 int entries
= istate
->cache_nr
;
2747 for (i
= 0; i
< entries
; i
++) {
2748 struct cache_entry
*ce
= istate
->cache
[i
];
2749 if (is_racy_timestamp(istate
, ce
))
2755 void repo_update_index_if_able(struct repository
*repo
,
2756 struct lock_file
*lockfile
)
2758 if ((repo
->index
->cache_changed
||
2759 has_racy_timestamp(repo
->index
)) &&
2760 repo_verify_index(repo
))
2761 write_locked_index(repo
->index
, lockfile
, COMMIT_LOCK
);
2763 rollback_lock_file(lockfile
);
2766 static int record_eoie(void)
2770 if (!git_config_get_bool("index.recordendofindexentries", &val
))
2774 * As a convenience, the end of index entries extension
2775 * used for threading is written by default if the user
2776 * explicitly requested threaded index reads.
2778 return !git_config_get_index_threads(&val
) && val
!= 1;
2781 static int record_ieot(void)
2785 if (!git_config_get_bool("index.recordoffsettable", &val
))
2789 * As a convenience, the offset table used for threading is
2790 * written by default if the user explicitly requested
2791 * threaded index reads.
2793 return !git_config_get_index_threads(&val
) && val
!= 1;
2797 * On success, `tempfile` is closed. If it is the temporary file
2798 * of a `struct lock_file`, we will therefore effectively perform
2799 * a 'close_lock_file_gently()`. Since that is an implementation
2800 * detail of lockfiles, callers of `do_write_index()` should not
2803 static int do_write_index(struct index_state
*istate
, struct tempfile
*tempfile
,
2804 int strip_extensions
)
2806 uint64_t start
= getnanotime();
2808 git_hash_ctx
*eoie_c
= NULL
;
2809 struct cache_header hdr
;
2810 int i
, err
= 0, removed
, extended
, hdr_version
;
2811 struct cache_entry
**cache
= istate
->cache
;
2812 int entries
= istate
->cache_nr
;
2814 struct ondisk_cache_entry ondisk
;
2815 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
2816 int drop_cache_tree
= istate
->drop_cache_tree
;
2818 int ieot_entries
= 1;
2819 struct index_entry_offset_table
*ieot
= NULL
;
2822 f
= hashfd(tempfile
->fd
, tempfile
->filename
.buf
);
2824 for (i
= removed
= extended
= 0; i
< entries
; i
++) {
2825 if (cache
[i
]->ce_flags
& CE_REMOVE
)
2828 /* reduce extended entries if possible */
2829 cache
[i
]->ce_flags
&= ~CE_EXTENDED
;
2830 if (cache
[i
]->ce_flags
& CE_EXTENDED_FLAGS
) {
2832 cache
[i
]->ce_flags
|= CE_EXTENDED
;
2836 if (!istate
->version
)
2837 istate
->version
= get_index_format_default(the_repository
);
2839 /* demote version 3 to version 2 when the latter suffices */
2840 if (istate
->version
== 3 || istate
->version
== 2)
2841 istate
->version
= extended
? 3 : 2;
2843 hdr_version
= istate
->version
;
2845 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
2846 hdr
.hdr_version
= htonl(hdr_version
);
2847 hdr
.hdr_entries
= htonl(entries
- removed
);
2849 hashwrite(f
, &hdr
, sizeof(hdr
));
2851 if (!HAVE_THREADS
|| git_config_get_index_threads(&nr_threads
))
2854 if (nr_threads
!= 1 && record_ieot()) {
2855 int ieot_blocks
, cpus
;
2858 * ensure default number of ieot blocks maps evenly to the
2859 * default number of threads that will process them leaving
2860 * room for the thread to load the index extensions.
2863 ieot_blocks
= istate
->cache_nr
/ THREAD_COST
;
2864 cpus
= online_cpus();
2865 if (ieot_blocks
> cpus
- 1)
2866 ieot_blocks
= cpus
- 1;
2868 ieot_blocks
= nr_threads
;
2869 if (ieot_blocks
> istate
->cache_nr
)
2870 ieot_blocks
= istate
->cache_nr
;
2874 * no reason to write out the IEOT extension if we don't
2875 * have enough blocks to utilize multi-threading
2877 if (ieot_blocks
> 1) {
2878 ieot
= xcalloc(1, sizeof(struct index_entry_offset_table
)
2879 + (ieot_blocks
* sizeof(struct index_entry_offset
)));
2880 ieot_entries
= DIV_ROUND_UP(entries
, ieot_blocks
);
2884 offset
= hashfile_total(f
);
2887 previous_name
= (hdr_version
== 4) ? &previous_name_buf
: NULL
;
2889 for (i
= 0; i
< entries
; i
++) {
2890 struct cache_entry
*ce
= cache
[i
];
2891 if (ce
->ce_flags
& CE_REMOVE
)
2893 if (!ce_uptodate(ce
) && is_racy_timestamp(istate
, ce
))
2894 ce_smudge_racily_clean_entry(istate
, ce
);
2895 if (is_null_oid(&ce
->oid
)) {
2896 static const char msg
[] = "cache entry has null sha1: %s";
2897 static int allow
= -1;
2900 allow
= git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2902 warning(msg
, ce
->name
);
2904 err
= error(msg
, ce
->name
);
2906 drop_cache_tree
= 1;
2908 if (ieot
&& i
&& (i
% ieot_entries
== 0)) {
2909 ieot
->entries
[ieot
->nr
].nr
= nr
;
2910 ieot
->entries
[ieot
->nr
].offset
= offset
;
2913 * If we have a V4 index, set the first byte to an invalid
2914 * character to ensure there is nothing common with the previous
2918 previous_name
->buf
[0] = 0;
2921 offset
= hashfile_total(f
);
2923 if (ce_write_entry(f
, ce
, previous_name
, (struct ondisk_cache_entry
*)&ondisk
) < 0)
2931 ieot
->entries
[ieot
->nr
].nr
= nr
;
2932 ieot
->entries
[ieot
->nr
].offset
= offset
;
2935 strbuf_release(&previous_name_buf
);
2942 offset
= hashfile_total(f
);
2945 * The extension headers must be hashed on their own for the
2946 * EOIE extension. Create a hashfile here to compute that hash.
2948 if (offset
&& record_eoie()) {
2949 CALLOC_ARRAY(eoie_c
, 1);
2950 the_hash_algo
->init_fn(eoie_c
);
2954 * Lets write out CACHE_EXT_INDEXENTRYOFFSETTABLE first so that we
2955 * can minimize the number of extensions we have to scan through to
2956 * find it during load. Write it out regardless of the
2957 * strip_extensions parameter as we need it when loading the shared
2961 struct strbuf sb
= STRBUF_INIT
;
2963 write_ieot_extension(&sb
, ieot
);
2964 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_INDEXENTRYOFFSETTABLE
, sb
.len
) < 0;
2965 hashwrite(f
, sb
.buf
, sb
.len
);
2966 strbuf_release(&sb
);
2972 if (!strip_extensions
&& istate
->split_index
&&
2973 !is_null_oid(&istate
->split_index
->base_oid
)) {
2974 struct strbuf sb
= STRBUF_INIT
;
2976 err
= write_link_extension(&sb
, istate
) < 0 ||
2977 write_index_ext_header(f
, eoie_c
, CACHE_EXT_LINK
,
2979 hashwrite(f
, sb
.buf
, sb
.len
);
2980 strbuf_release(&sb
);
2984 if (!strip_extensions
&& !drop_cache_tree
&& istate
->cache_tree
) {
2985 struct strbuf sb
= STRBUF_INIT
;
2987 cache_tree_write(&sb
, istate
->cache_tree
);
2988 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_TREE
, sb
.len
) < 0;
2989 hashwrite(f
, sb
.buf
, sb
.len
);
2990 strbuf_release(&sb
);
2994 if (!strip_extensions
&& istate
->resolve_undo
) {
2995 struct strbuf sb
= STRBUF_INIT
;
2997 resolve_undo_write(&sb
, istate
->resolve_undo
);
2998 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_RESOLVE_UNDO
,
3000 hashwrite(f
, sb
.buf
, sb
.len
);
3001 strbuf_release(&sb
);
3005 if (!strip_extensions
&& istate
->untracked
) {
3006 struct strbuf sb
= STRBUF_INIT
;
3008 write_untracked_extension(&sb
, istate
->untracked
);
3009 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_UNTRACKED
,
3011 hashwrite(f
, sb
.buf
, sb
.len
);
3012 strbuf_release(&sb
);
3016 if (!strip_extensions
&& istate
->fsmonitor_last_update
) {
3017 struct strbuf sb
= STRBUF_INIT
;
3019 write_fsmonitor_extension(&sb
, istate
);
3020 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_FSMONITOR
, sb
.len
) < 0;
3021 hashwrite(f
, sb
.buf
, sb
.len
);
3022 strbuf_release(&sb
);
3026 if (istate
->sparse_index
) {
3027 if (write_index_ext_header(f
, eoie_c
, CACHE_EXT_SPARSE_DIRECTORIES
, 0) < 0)
3032 * CACHE_EXT_ENDOFINDEXENTRIES must be written as the last entry before the SHA1
3033 * so that it can be found and processed before all the index entries are
3034 * read. Write it out regardless of the strip_extensions parameter as we need it
3035 * when loading the shared index.
3038 struct strbuf sb
= STRBUF_INIT
;
3040 write_eoie_extension(&sb
, eoie_c
, offset
);
3041 err
= write_index_ext_header(f
, NULL
, CACHE_EXT_ENDOFINDEXENTRIES
, sb
.len
) < 0;
3042 hashwrite(f
, sb
.buf
, sb
.len
);
3043 strbuf_release(&sb
);
3048 finalize_hashfile(f
, istate
->oid
.hash
, CSUM_HASH_IN_STREAM
);
3049 if (close_tempfile_gently(tempfile
)) {
3050 error(_("could not close '%s'"), get_tempfile_path(tempfile
));
3053 if (stat(get_tempfile_path(tempfile
), &st
))
3055 istate
->timestamp
.sec
= (unsigned int)st
.st_mtime
;
3056 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
3057 trace_performance_since(start
, "write index, changed mask = %x", istate
->cache_changed
);
3060 * TODO trace2: replace "the_repository" with the actual repo instance
3061 * that is associated with the given "istate".
3063 trace2_data_intmax("index", the_repository
, "write/version",
3065 trace2_data_intmax("index", the_repository
, "write/cache_nr",
3071 void set_alternate_index_output(const char *name
)
3073 alternate_index_output
= name
;
3076 static int commit_locked_index(struct lock_file
*lk
)
3078 if (alternate_index_output
)
3079 return commit_lock_file_to(lk
, alternate_index_output
);
3081 return commit_lock_file(lk
);
3084 static int do_write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
3088 int was_full
= !istate
->sparse_index
;
3090 ret
= convert_to_sparse(istate
, 0);
3093 warning(_("failed to convert to a sparse-index"));
3098 * TODO trace2: replace "the_repository" with the actual repo instance
3099 * that is associated with the given "istate".
3101 trace2_region_enter_printf("index", "do_write_index", the_repository
,
3102 "%s", get_lock_file_path(lock
));
3103 ret
= do_write_index(istate
, lock
->tempfile
, 0);
3104 trace2_region_leave_printf("index", "do_write_index", the_repository
,
3105 "%s", get_lock_file_path(lock
));
3108 ensure_full_index(istate
);
3112 if (flags
& COMMIT_LOCK
)
3113 ret
= commit_locked_index(lock
);
3115 ret
= close_lock_file_gently(lock
);
3117 run_hook_le(NULL
, "post-index-change",
3118 istate
->updated_workdir
? "1" : "0",
3119 istate
->updated_skipworktree
? "1" : "0", NULL
);
3120 istate
->updated_workdir
= 0;
3121 istate
->updated_skipworktree
= 0;
3126 static int write_split_index(struct index_state
*istate
,
3127 struct lock_file
*lock
,
3131 prepare_to_write_split_index(istate
);
3132 ret
= do_write_locked_index(istate
, lock
, flags
);
3133 finish_writing_split_index(istate
);
3137 static const char *shared_index_expire
= "2.weeks.ago";
3139 static unsigned long get_shared_index_expire_date(void)
3141 static unsigned long shared_index_expire_date
;
3142 static int shared_index_expire_date_prepared
;
3144 if (!shared_index_expire_date_prepared
) {
3145 git_config_get_expiry("splitindex.sharedindexexpire",
3146 &shared_index_expire
);
3147 shared_index_expire_date
= approxidate(shared_index_expire
);
3148 shared_index_expire_date_prepared
= 1;
3151 return shared_index_expire_date
;
3154 static int should_delete_shared_index(const char *shared_index_path
)
3157 unsigned long expiration
;
3159 /* Check timestamp */
3160 expiration
= get_shared_index_expire_date();
3163 if (stat(shared_index_path
, &st
))
3164 return error_errno(_("could not stat '%s'"), shared_index_path
);
3165 if (st
.st_mtime
> expiration
)
3171 static int clean_shared_index_files(const char *current_hex
)
3174 DIR *dir
= opendir(get_git_dir());
3177 return error_errno(_("unable to open git dir: %s"), get_git_dir());
3179 while ((de
= readdir(dir
)) != NULL
) {
3180 const char *sha1_hex
;
3181 const char *shared_index_path
;
3182 if (!skip_prefix(de
->d_name
, "sharedindex.", &sha1_hex
))
3184 if (!strcmp(sha1_hex
, current_hex
))
3186 shared_index_path
= git_path("%s", de
->d_name
);
3187 if (should_delete_shared_index(shared_index_path
) > 0 &&
3188 unlink(shared_index_path
))
3189 warning_errno(_("unable to unlink: %s"), shared_index_path
);
3196 static int write_shared_index(struct index_state
*istate
,
3197 struct tempfile
**temp
)
3199 struct split_index
*si
= istate
->split_index
;
3200 int ret
, was_full
= !istate
->sparse_index
;
3202 move_cache_to_base_index(istate
);
3203 convert_to_sparse(istate
, 0);
3205 trace2_region_enter_printf("index", "shared/do_write_index",
3206 the_repository
, "%s", get_tempfile_path(*temp
));
3207 ret
= do_write_index(si
->base
, *temp
, 1);
3208 trace2_region_leave_printf("index", "shared/do_write_index",
3209 the_repository
, "%s", get_tempfile_path(*temp
));
3212 ensure_full_index(istate
);
3216 ret
= adjust_shared_perm(get_tempfile_path(*temp
));
3218 error(_("cannot fix permission bits on '%s'"), get_tempfile_path(*temp
));
3221 ret
= rename_tempfile(temp
,
3222 git_path("sharedindex.%s", oid_to_hex(&si
->base
->oid
)));
3224 oidcpy(&si
->base_oid
, &si
->base
->oid
);
3225 clean_shared_index_files(oid_to_hex(&si
->base
->oid
));
3231 static const int default_max_percent_split_change
= 20;
3233 static int too_many_not_shared_entries(struct index_state
*istate
)
3235 int i
, not_shared
= 0;
3236 int max_split
= git_config_get_max_percent_split_change();
3238 switch (max_split
) {
3240 /* not or badly configured: use the default value */
3241 max_split
= default_max_percent_split_change
;
3244 return 1; /* 0% means always write a new shared index */
3246 return 0; /* 100% means never write a new shared index */
3248 break; /* just use the configured value */
3251 /* Count not shared entries */
3252 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3253 struct cache_entry
*ce
= istate
->cache
[i
];
3258 return (int64_t)istate
->cache_nr
* max_split
< (int64_t)not_shared
* 100;
3261 int write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
3264 int new_shared_index
, ret
, test_split_index_env
;
3265 struct split_index
*si
= istate
->split_index
;
3267 if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0))
3268 cache_tree_verify(the_repository
, istate
);
3270 if ((flags
& SKIP_IF_UNCHANGED
) && !istate
->cache_changed
) {
3271 if (flags
& COMMIT_LOCK
)
3272 rollback_lock_file(lock
);
3276 if (istate
->fsmonitor_last_update
)
3277 fill_fsmonitor_bitmap(istate
);
3279 test_split_index_env
= git_env_bool("GIT_TEST_SPLIT_INDEX", 0);
3281 if ((!si
&& !test_split_index_env
) ||
3282 alternate_index_output
||
3283 (istate
->cache_changed
& ~EXTMASK
)) {
3285 oidclr(&si
->base_oid
);
3286 ret
= do_write_locked_index(istate
, lock
, flags
);
3290 if (test_split_index_env
) {
3292 si
= init_split_index(istate
);
3293 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3295 int v
= si
->base_oid
.hash
[0];
3297 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3300 if (too_many_not_shared_entries(istate
))
3301 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3303 new_shared_index
= istate
->cache_changed
& SPLIT_INDEX_ORDERED
;
3305 if (new_shared_index
) {
3306 struct tempfile
*temp
;
3309 /* Same initial permissions as the main .git/index file */
3310 temp
= mks_tempfile_sm(git_path("sharedindex_XXXXXX"), 0, 0666);
3312 oidclr(&si
->base_oid
);
3313 ret
= do_write_locked_index(istate
, lock
, flags
);
3316 ret
= write_shared_index(istate
, &temp
);
3318 saved_errno
= errno
;
3319 if (is_tempfile_active(temp
))
3320 delete_tempfile(&temp
);
3321 errno
= saved_errno
;
3327 ret
= write_split_index(istate
, lock
, flags
);
3329 /* Freshen the shared index only if the split-index was written */
3330 if (!ret
&& !new_shared_index
&& !is_null_oid(&si
->base_oid
)) {
3331 const char *shared_index
= git_path("sharedindex.%s",
3332 oid_to_hex(&si
->base_oid
));
3333 freshen_shared_index(shared_index
, 1);
3337 if (flags
& COMMIT_LOCK
)
3338 rollback_lock_file(lock
);
3343 * Read the index file that is potentially unmerged into given
3344 * index_state, dropping any unmerged entries to stage #0 (potentially
3345 * resulting in a path appearing as both a file and a directory in the
3346 * index; the caller is responsible to clear out the extra entries
3347 * before writing the index to a tree). Returns true if the index is
3348 * unmerged. Callers who want to refuse to work from an unmerged
3349 * state can call this and check its return value, instead of calling
3352 int repo_read_index_unmerged(struct repository
*repo
)
3354 struct index_state
*istate
;
3358 repo_read_index(repo
);
3359 istate
= repo
->index
;
3360 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3361 struct cache_entry
*ce
= istate
->cache
[i
];
3362 struct cache_entry
*new_ce
;
3368 len
= ce_namelen(ce
);
3369 new_ce
= make_empty_cache_entry(istate
, len
);
3370 memcpy(new_ce
->name
, ce
->name
, len
);
3371 new_ce
->ce_flags
= create_ce_flags(0) | CE_CONFLICTED
;
3372 new_ce
->ce_namelen
= len
;
3373 new_ce
->ce_mode
= ce
->ce_mode
;
3374 if (add_index_entry(istate
, new_ce
, ADD_CACHE_SKIP_DFCHECK
))
3375 return error(_("%s: cannot drop to stage #0"),
3382 * Returns 1 if the path is an "other" path with respect to
3383 * the index; that is, the path is not mentioned in the index at all,
3384 * either as a file, a directory with some files in the index,
3385 * or as an unmerged entry.
3387 * We helpfully remove a trailing "/" from directories so that
3388 * the output of read_directory can be used as-is.
3390 int index_name_is_other(struct index_state
*istate
, const char *name
,
3394 if (namelen
&& name
[namelen
- 1] == '/')
3396 pos
= index_name_pos(istate
, name
, namelen
);
3398 return 0; /* exact match */
3400 if (pos
< istate
->cache_nr
) {
3401 struct cache_entry
*ce
= istate
->cache
[pos
];
3402 if (ce_namelen(ce
) == namelen
&&
3403 !memcmp(ce
->name
, name
, namelen
))
3404 return 0; /* Yup, this one exists unmerged */
3409 void *read_blob_data_from_index(struct index_state
*istate
,
3410 const char *path
, unsigned long *size
)
3414 enum object_type type
;
3418 pos
= index_name_pos(istate
, path
, len
);
3421 * We might be in the middle of a merge, in which
3422 * case we would read stage #2 (ours).
3426 (pos
< 0 && i
< istate
->cache_nr
&&
3427 !strcmp(istate
->cache
[i
]->name
, path
));
3429 if (ce_stage(istate
->cache
[i
]) == 2)
3434 data
= read_object_file(&istate
->cache
[pos
]->oid
, &type
, &sz
);
3435 if (!data
|| type
!= OBJ_BLOB
) {
3444 void stat_validity_clear(struct stat_validity
*sv
)
3446 FREE_AND_NULL(sv
->sd
);
3449 int stat_validity_check(struct stat_validity
*sv
, const char *path
)
3453 if (stat(path
, &st
) < 0)
3454 return sv
->sd
== NULL
;
3457 return S_ISREG(st
.st_mode
) && !match_stat_data(sv
->sd
, &st
);
3460 void stat_validity_update(struct stat_validity
*sv
, int fd
)
3464 if (fstat(fd
, &st
) < 0 || !S_ISREG(st
.st_mode
))
3465 stat_validity_clear(sv
);
3468 CALLOC_ARRAY(sv
->sd
, 1);
3469 fill_stat_data(sv
->sd
, &st
);
3473 void move_index_extensions(struct index_state
*dst
, struct index_state
*src
)
3475 dst
->untracked
= src
->untracked
;
3476 src
->untracked
= NULL
;
3477 dst
->cache_tree
= src
->cache_tree
;
3478 src
->cache_tree
= NULL
;
3481 struct cache_entry
*dup_cache_entry(const struct cache_entry
*ce
,
3482 struct index_state
*istate
)
3484 unsigned int size
= ce_size(ce
);
3485 int mem_pool_allocated
;
3486 struct cache_entry
*new_entry
= make_empty_cache_entry(istate
, ce_namelen(ce
));
3487 mem_pool_allocated
= new_entry
->mem_pool_allocated
;
3489 memcpy(new_entry
, ce
, size
);
3490 new_entry
->mem_pool_allocated
= mem_pool_allocated
;
3494 void discard_cache_entry(struct cache_entry
*ce
)
3496 if (ce
&& should_validate_cache_entries())
3497 memset(ce
, 0xCD, cache_entry_size(ce
->ce_namelen
));
3499 if (ce
&& ce
->mem_pool_allocated
)
3505 int should_validate_cache_entries(void)
3507 static int validate_index_cache_entries
= -1;
3509 if (validate_index_cache_entries
< 0) {
3510 if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))
3511 validate_index_cache_entries
= 1;
3513 validate_index_cache_entries
= 0;
3516 return validate_index_cache_entries
;
3519 #define EOIE_SIZE (4 + GIT_SHA1_RAWSZ) /* <4-byte offset> + <20-byte hash> */
3520 #define EOIE_SIZE_WITH_HEADER (4 + 4 + EOIE_SIZE) /* <4-byte signature> + <4-byte length> + EOIE_SIZE */
3522 static size_t read_eoie_extension(const char *mmap
, size_t mmap_size
)
3525 * The end of index entries (EOIE) extension is guaranteed to be last
3526 * so that it can be found by scanning backwards from the EOF.
3533 const char *index
, *eoie
;
3535 size_t offset
, src_offset
;
3536 unsigned char hash
[GIT_MAX_RAWSZ
];
3539 /* ensure we have an index big enough to contain an EOIE extension */
3540 if (mmap_size
< sizeof(struct cache_header
) + EOIE_SIZE_WITH_HEADER
+ the_hash_algo
->rawsz
)
3543 /* validate the extension signature */
3544 index
= eoie
= mmap
+ mmap_size
- EOIE_SIZE_WITH_HEADER
- the_hash_algo
->rawsz
;
3545 if (CACHE_EXT(index
) != CACHE_EXT_ENDOFINDEXENTRIES
)
3547 index
+= sizeof(uint32_t);
3549 /* validate the extension size */
3550 extsize
= get_be32(index
);
3551 if (extsize
!= EOIE_SIZE
)
3553 index
+= sizeof(uint32_t);
3556 * Validate the offset we're going to look for the first extension
3557 * signature is after the index header and before the eoie extension.
3559 offset
= get_be32(index
);
3560 if (mmap
+ offset
< mmap
+ sizeof(struct cache_header
))
3562 if (mmap
+ offset
>= eoie
)
3564 index
+= sizeof(uint32_t);
3567 * The hash is computed over extension types and their sizes (but not
3568 * their contents). E.g. if we have "TREE" extension that is N-bytes
3569 * long, "REUC" extension that is M-bytes long, followed by "EOIE",
3570 * then the hash would be:
3572 * SHA-1("TREE" + <binary representation of N> +
3573 * "REUC" + <binary representation of M>)
3575 src_offset
= offset
;
3576 the_hash_algo
->init_fn(&c
);
3577 while (src_offset
< mmap_size
- the_hash_algo
->rawsz
- EOIE_SIZE_WITH_HEADER
) {
3578 /* After an array of active_nr index entries,
3579 * there can be arbitrary number of extended
3580 * sections, each of which is prefixed with
3581 * extension name (4-byte) and section length
3582 * in 4-byte network byte order.
3585 memcpy(&extsize
, mmap
+ src_offset
+ 4, 4);
3586 extsize
= ntohl(extsize
);
3588 /* verify the extension size isn't so large it will wrap around */
3589 if (src_offset
+ 8 + extsize
< src_offset
)
3592 the_hash_algo
->update_fn(&c
, mmap
+ src_offset
, 8);
3595 src_offset
+= extsize
;
3597 the_hash_algo
->final_fn(hash
, &c
);
3598 if (!hasheq(hash
, (const unsigned char *)index
))
3601 /* Validate that the extension offsets returned us back to the eoie extension. */
3602 if (src_offset
!= mmap_size
- the_hash_algo
->rawsz
- EOIE_SIZE_WITH_HEADER
)
3608 static void write_eoie_extension(struct strbuf
*sb
, git_hash_ctx
*eoie_context
, size_t offset
)
3611 unsigned char hash
[GIT_MAX_RAWSZ
];
3614 put_be32(&buffer
, offset
);
3615 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3618 the_hash_algo
->final_fn(hash
, eoie_context
);
3619 strbuf_add(sb
, hash
, the_hash_algo
->rawsz
);
3622 #define IEOT_VERSION (1)
3624 static struct index_entry_offset_table
*read_ieot_extension(const char *mmap
, size_t mmap_size
, size_t offset
)
3626 const char *index
= NULL
;
3627 uint32_t extsize
, ext_version
;
3628 struct index_entry_offset_table
*ieot
;
3631 /* find the IEOT extension */
3634 while (offset
<= mmap_size
- the_hash_algo
->rawsz
- 8) {
3635 extsize
= get_be32(mmap
+ offset
+ 4);
3636 if (CACHE_EXT((mmap
+ offset
)) == CACHE_EXT_INDEXENTRYOFFSETTABLE
) {
3637 index
= mmap
+ offset
+ 4 + 4;
3646 /* validate the version is IEOT_VERSION */
3647 ext_version
= get_be32(index
);
3648 if (ext_version
!= IEOT_VERSION
) {
3649 error("invalid IEOT version %d", ext_version
);
3652 index
+= sizeof(uint32_t);
3654 /* extension size - version bytes / bytes per entry */
3655 nr
= (extsize
- sizeof(uint32_t)) / (sizeof(uint32_t) + sizeof(uint32_t));
3657 error("invalid number of IEOT entries %d", nr
);
3660 ieot
= xmalloc(sizeof(struct index_entry_offset_table
)
3661 + (nr
* sizeof(struct index_entry_offset
)));
3663 for (i
= 0; i
< nr
; i
++) {
3664 ieot
->entries
[i
].offset
= get_be32(index
);
3665 index
+= sizeof(uint32_t);
3666 ieot
->entries
[i
].nr
= get_be32(index
);
3667 index
+= sizeof(uint32_t);
3673 static void write_ieot_extension(struct strbuf
*sb
, struct index_entry_offset_table
*ieot
)
3679 put_be32(&buffer
, IEOT_VERSION
);
3680 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3683 for (i
= 0; i
< ieot
->nr
; i
++) {
3686 put_be32(&buffer
, ieot
->entries
[i
].offset
);
3687 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3690 put_be32(&buffer
, ieot
->entries
[i
].nr
);
3691 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3695 void prefetch_cache_entries(const struct index_state
*istate
,
3696 must_prefetch_predicate must_prefetch
)
3699 struct oid_array to_fetch
= OID_ARRAY_INIT
;
3701 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3702 struct cache_entry
*ce
= istate
->cache
[i
];
3704 if (S_ISGITLINK(ce
->ce_mode
) || !must_prefetch(ce
))
3706 if (!oid_object_info_extended(the_repository
, &ce
->oid
,
3708 OBJECT_INFO_FOR_PREFETCH
))
3710 oid_array_append(&to_fetch
, &ce
->oid
);
3712 promisor_remote_get_direct(the_repository
,
3713 to_fetch
.oid
, to_fetch
.nr
);
3714 oid_array_clear(&to_fetch
);