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"
29 /* Mask for the name length in ce_flags in the on-disk index */
31 #define CE_NAMEMASK (0x0fff)
35 * The first letter should be 'A'..'Z' for extensions that are not
36 * necessary for a correct operation (i.e. optimization data).
37 * When new extensions are added that _needs_ to be understood in
38 * order to correctly interpret the index file, pick character that
39 * is outside the range, to cause the reader to abort.
42 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
43 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
44 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
45 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
46 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
47 #define CACHE_EXT_FSMONITOR 0x46534D4E /* "FSMN" */
48 #define CACHE_EXT_ENDOFINDEXENTRIES 0x454F4945 /* "EOIE" */
49 #define CACHE_EXT_INDEXENTRYOFFSETTABLE 0x49454F54 /* "IEOT" */
51 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
52 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
53 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
54 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED | FSMONITOR_CHANGED)
58 * This is an estimate of the pathname length in the index. We use
59 * this for V4 index files to guess the un-deltafied size of the index
60 * in memory because of pathname deltafication. This is not required
61 * for V2/V3 index formats because their pathnames are not compressed.
62 * If the initial amount of memory set aside is not sufficient, the
63 * mem pool will allocate extra memory.
65 #define CACHE_ENTRY_PATH_LENGTH 80
67 static inline struct cache_entry
*mem_pool__ce_alloc(struct mem_pool
*mem_pool
, size_t len
)
69 struct cache_entry
*ce
;
70 ce
= mem_pool_alloc(mem_pool
, cache_entry_size(len
));
71 ce
->mem_pool_allocated
= 1;
75 static inline struct cache_entry
*mem_pool__ce_calloc(struct mem_pool
*mem_pool
, size_t len
)
77 struct cache_entry
* ce
;
78 ce
= mem_pool_calloc(mem_pool
, 1, cache_entry_size(len
));
79 ce
->mem_pool_allocated
= 1;
83 static struct mem_pool
*find_mem_pool(struct index_state
*istate
)
85 struct mem_pool
**pool_ptr
;
87 if (istate
->split_index
&& istate
->split_index
->base
)
88 pool_ptr
= &istate
->split_index
->base
->ce_mem_pool
;
90 pool_ptr
= &istate
->ce_mem_pool
;
93 mem_pool_init(pool_ptr
, 0);
98 static const char *alternate_index_output
;
100 static void set_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
102 istate
->cache
[nr
] = ce
;
103 add_name_hash(istate
, ce
);
106 static void replace_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
108 struct cache_entry
*old
= istate
->cache
[nr
];
110 replace_index_entry_in_base(istate
, old
, ce
);
111 remove_name_hash(istate
, old
);
112 discard_cache_entry(old
);
113 ce
->ce_flags
&= ~CE_HASHED
;
114 set_index_entry(istate
, nr
, ce
);
115 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
116 mark_fsmonitor_invalid(istate
, ce
);
117 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
120 void rename_index_entry_at(struct index_state
*istate
, int nr
, const char *new_name
)
122 struct cache_entry
*old_entry
= istate
->cache
[nr
], *new_entry
;
123 int namelen
= strlen(new_name
);
125 new_entry
= make_empty_cache_entry(istate
, namelen
);
126 copy_cache_entry(new_entry
, old_entry
);
127 new_entry
->ce_flags
&= ~CE_HASHED
;
128 new_entry
->ce_namelen
= namelen
;
129 new_entry
->index
= 0;
130 memcpy(new_entry
->name
, new_name
, namelen
+ 1);
132 cache_tree_invalidate_path(istate
, old_entry
->name
);
133 untracked_cache_remove_from_index(istate
, old_entry
->name
);
134 remove_index_entry_at(istate
, nr
);
135 add_index_entry(istate
, new_entry
, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
138 void fill_stat_data(struct stat_data
*sd
, struct stat
*st
)
140 sd
->sd_ctime
.sec
= (unsigned int)st
->st_ctime
;
141 sd
->sd_mtime
.sec
= (unsigned int)st
->st_mtime
;
142 sd
->sd_ctime
.nsec
= ST_CTIME_NSEC(*st
);
143 sd
->sd_mtime
.nsec
= ST_MTIME_NSEC(*st
);
144 sd
->sd_dev
= st
->st_dev
;
145 sd
->sd_ino
= st
->st_ino
;
146 sd
->sd_uid
= st
->st_uid
;
147 sd
->sd_gid
= st
->st_gid
;
148 sd
->sd_size
= st
->st_size
;
151 int match_stat_data(const struct stat_data
*sd
, struct stat
*st
)
155 if (sd
->sd_mtime
.sec
!= (unsigned int)st
->st_mtime
)
156 changed
|= MTIME_CHANGED
;
157 if (trust_ctime
&& check_stat
&&
158 sd
->sd_ctime
.sec
!= (unsigned int)st
->st_ctime
)
159 changed
|= CTIME_CHANGED
;
162 if (check_stat
&& sd
->sd_mtime
.nsec
!= ST_MTIME_NSEC(*st
))
163 changed
|= MTIME_CHANGED
;
164 if (trust_ctime
&& check_stat
&&
165 sd
->sd_ctime
.nsec
!= ST_CTIME_NSEC(*st
))
166 changed
|= CTIME_CHANGED
;
170 if (sd
->sd_uid
!= (unsigned int) st
->st_uid
||
171 sd
->sd_gid
!= (unsigned int) st
->st_gid
)
172 changed
|= OWNER_CHANGED
;
173 if (sd
->sd_ino
!= (unsigned int) st
->st_ino
)
174 changed
|= INODE_CHANGED
;
179 * st_dev breaks on network filesystems where different
180 * clients will have different views of what "device"
181 * the filesystem is on
183 if (check_stat
&& sd
->sd_dev
!= (unsigned int) st
->st_dev
)
184 changed
|= INODE_CHANGED
;
187 if (sd
->sd_size
!= (unsigned int) st
->st_size
)
188 changed
|= DATA_CHANGED
;
194 * This only updates the "non-critical" parts of the directory
195 * cache, ie the parts that aren't tracked by GIT, and only used
196 * to validate the cache.
198 void fill_stat_cache_info(struct cache_entry
*ce
, struct stat
*st
)
200 fill_stat_data(&ce
->ce_stat_data
, st
);
202 if (assume_unchanged
)
203 ce
->ce_flags
|= CE_VALID
;
205 if (S_ISREG(st
->st_mode
)) {
206 ce_mark_uptodate(ce
);
207 mark_fsmonitor_valid(ce
);
211 static int ce_compare_data(struct index_state
*istate
,
212 const struct cache_entry
*ce
,
216 int fd
= git_open_cloexec(ce
->name
, O_RDONLY
);
219 struct object_id oid
;
220 if (!index_fd(istate
, &oid
, fd
, st
, OBJ_BLOB
, ce
->name
, 0))
221 match
= !oideq(&oid
, &ce
->oid
);
222 /* index_fd() closed the file descriptor already */
227 static int ce_compare_link(const struct cache_entry
*ce
, size_t expected_size
)
232 enum object_type type
;
233 struct strbuf sb
= STRBUF_INIT
;
235 if (strbuf_readlink(&sb
, ce
->name
, expected_size
))
238 buffer
= read_object_file(&ce
->oid
, &type
, &size
);
241 match
= memcmp(buffer
, sb
.buf
, size
);
248 static int ce_compare_gitlink(const struct cache_entry
*ce
)
250 struct object_id oid
;
253 * We don't actually require that the .git directory
254 * under GITLINK directory be a valid git directory. It
255 * might even be missing (in case nobody populated that
258 * If so, we consider it always to match.
260 if (resolve_gitlink_ref(ce
->name
, "HEAD", &oid
) < 0)
262 return !oideq(&oid
, &ce
->oid
);
265 static int ce_modified_check_fs(struct index_state
*istate
,
266 const struct cache_entry
*ce
,
269 switch (st
->st_mode
& S_IFMT
) {
271 if (ce_compare_data(istate
, ce
, st
))
275 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
279 if (S_ISGITLINK(ce
->ce_mode
))
280 return ce_compare_gitlink(ce
) ? DATA_CHANGED
: 0;
281 /* else fallthrough */
288 static int ce_match_stat_basic(const struct cache_entry
*ce
, struct stat
*st
)
290 unsigned int changed
= 0;
292 if (ce
->ce_flags
& CE_REMOVE
)
293 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
295 switch (ce
->ce_mode
& S_IFMT
) {
297 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
298 /* We consider only the owner x bit to be relevant for
301 if (trust_executable_bit
&&
302 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
303 changed
|= MODE_CHANGED
;
306 if (!S_ISLNK(st
->st_mode
) &&
307 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
308 changed
|= TYPE_CHANGED
;
311 /* We ignore most of the st_xxx fields for gitlinks */
312 if (!S_ISDIR(st
->st_mode
))
313 changed
|= TYPE_CHANGED
;
314 else if (ce_compare_gitlink(ce
))
315 changed
|= DATA_CHANGED
;
318 BUG("unsupported ce_mode: %o", ce
->ce_mode
);
321 changed
|= match_stat_data(&ce
->ce_stat_data
, st
);
323 /* Racily smudged entry? */
324 if (!ce
->ce_stat_data
.sd_size
) {
325 if (!is_empty_blob_sha1(ce
->oid
.hash
))
326 changed
|= DATA_CHANGED
;
332 static int is_racy_stat(const struct index_state
*istate
,
333 const struct stat_data
*sd
)
335 return (istate
->timestamp
.sec
&&
337 /* nanosecond timestamped files can also be racy! */
338 (istate
->timestamp
.sec
< sd
->sd_mtime
.sec
||
339 (istate
->timestamp
.sec
== sd
->sd_mtime
.sec
&&
340 istate
->timestamp
.nsec
<= sd
->sd_mtime
.nsec
))
342 istate
->timestamp
.sec
<= sd
->sd_mtime
.sec
347 int is_racy_timestamp(const struct index_state
*istate
,
348 const struct cache_entry
*ce
)
350 return (!S_ISGITLINK(ce
->ce_mode
) &&
351 is_racy_stat(istate
, &ce
->ce_stat_data
));
354 int match_stat_data_racy(const struct index_state
*istate
,
355 const struct stat_data
*sd
, struct stat
*st
)
357 if (is_racy_stat(istate
, sd
))
358 return MTIME_CHANGED
;
359 return match_stat_data(sd
, st
);
362 int ie_match_stat(struct index_state
*istate
,
363 const struct cache_entry
*ce
, struct stat
*st
,
364 unsigned int options
)
366 unsigned int changed
;
367 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
368 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
369 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
370 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
372 if (!ignore_fsmonitor
)
373 refresh_fsmonitor(istate
);
375 * If it's marked as always valid in the index, it's
376 * valid whatever the checked-out copy says.
378 * skip-worktree has the same effect with higher precedence
380 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
))
382 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
384 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
))
388 * Intent-to-add entries have not been added, so the index entry
389 * by definition never matches what is in the work tree until it
390 * actually gets added.
392 if (ce_intent_to_add(ce
))
393 return DATA_CHANGED
| TYPE_CHANGED
| MODE_CHANGED
;
395 changed
= ce_match_stat_basic(ce
, st
);
398 * Within 1 second of this sequence:
399 * echo xyzzy >file && git-update-index --add file
400 * running this command:
402 * would give a falsely clean cache entry. The mtime and
403 * length match the cache, and other stat fields do not change.
405 * We could detect this at update-index time (the cache entry
406 * being registered/updated records the same time as "now")
407 * and delay the return from git-update-index, but that would
408 * effectively mean we can make at most one commit per second,
409 * which is not acceptable. Instead, we check cache entries
410 * whose mtime are the same as the index file timestamp more
411 * carefully than others.
413 if (!changed
&& is_racy_timestamp(istate
, ce
)) {
414 if (assume_racy_is_modified
)
415 changed
|= DATA_CHANGED
;
417 changed
|= ce_modified_check_fs(istate
, ce
, st
);
423 int ie_modified(struct index_state
*istate
,
424 const struct cache_entry
*ce
,
425 struct stat
*st
, unsigned int options
)
427 int changed
, changed_fs
;
429 changed
= ie_match_stat(istate
, ce
, st
, options
);
433 * If the mode or type has changed, there's no point in trying
434 * to refresh the entry - it's not going to match
436 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
440 * Immediately after read-tree or update-index --cacheinfo,
441 * the length field is zero, as we have never even read the
442 * lstat(2) information once, and we cannot trust DATA_CHANGED
443 * returned by ie_match_stat() which in turn was returned by
444 * ce_match_stat_basic() to signal that the filesize of the
445 * blob changed. We have to actually go to the filesystem to
446 * see if the contents match, and if so, should answer "unchanged".
448 * The logic does not apply to gitlinks, as ce_match_stat_basic()
449 * already has checked the actual HEAD from the filesystem in the
450 * subproject. If ie_match_stat() already said it is different,
451 * then we know it is.
453 if ((changed
& DATA_CHANGED
) &&
454 (S_ISGITLINK(ce
->ce_mode
) || ce
->ce_stat_data
.sd_size
!= 0))
457 changed_fs
= ce_modified_check_fs(istate
, ce
, st
);
459 return changed
| changed_fs
;
463 int base_name_compare(const char *name1
, int len1
, int mode1
,
464 const char *name2
, int len2
, int mode2
)
466 unsigned char c1
, c2
;
467 int len
= len1
< len2
? len1
: len2
;
470 cmp
= memcmp(name1
, name2
, len
);
475 if (!c1
&& S_ISDIR(mode1
))
477 if (!c2
&& S_ISDIR(mode2
))
479 return (c1
< c2
) ? -1 : (c1
> c2
) ? 1 : 0;
483 * df_name_compare() is identical to base_name_compare(), except it
484 * compares conflicting directory/file entries as equal. Note that
485 * while a directory name compares as equal to a regular file, they
486 * then individually compare _differently_ to a filename that has
487 * a dot after the basename (because '\0' < '.' < '/').
489 * This is used by routines that want to traverse the git namespace
490 * but then handle conflicting entries together when possible.
492 int df_name_compare(const char *name1
, int len1
, int mode1
,
493 const char *name2
, int len2
, int mode2
)
495 int len
= len1
< len2
? len1
: len2
, cmp
;
496 unsigned char c1
, c2
;
498 cmp
= memcmp(name1
, name2
, len
);
501 /* Directories and files compare equal (same length, same name) */
505 if (!c1
&& S_ISDIR(mode1
))
508 if (!c2
&& S_ISDIR(mode2
))
510 if (c1
== '/' && !c2
)
512 if (c2
== '/' && !c1
)
517 int name_compare(const char *name1
, size_t len1
, const char *name2
, size_t len2
)
519 size_t min_len
= (len1
< len2
) ? len1
: len2
;
520 int cmp
= memcmp(name1
, name2
, min_len
);
530 int cache_name_stage_compare(const char *name1
, int len1
, int stage1
, const char *name2
, int len2
, int stage2
)
534 cmp
= name_compare(name1
, len1
, name2
, len2
);
545 static int index_name_stage_pos(const struct index_state
*istate
, const char *name
, int namelen
, int stage
)
550 last
= istate
->cache_nr
;
551 while (last
> first
) {
552 int next
= (last
+ first
) >> 1;
553 struct cache_entry
*ce
= istate
->cache
[next
];
554 int cmp
= cache_name_stage_compare(name
, namelen
, stage
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
566 int index_name_pos(const struct index_state
*istate
, const char *name
, int namelen
)
568 return index_name_stage_pos(istate
, name
, namelen
, 0);
571 int remove_index_entry_at(struct index_state
*istate
, int pos
)
573 struct cache_entry
*ce
= istate
->cache
[pos
];
575 record_resolve_undo(istate
, ce
);
576 remove_name_hash(istate
, ce
);
577 save_or_free_index_entry(istate
, ce
);
578 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
580 if (pos
>= istate
->cache_nr
)
582 MOVE_ARRAY(istate
->cache
+ pos
, istate
->cache
+ pos
+ 1,
583 istate
->cache_nr
- pos
);
588 * Remove all cache entries marked for removal, that is where
589 * CE_REMOVE is set in ce_flags. This is much more effective than
590 * calling remove_index_entry_at() for each entry to be removed.
592 void remove_marked_cache_entries(struct index_state
*istate
, int invalidate
)
594 struct cache_entry
**ce_array
= istate
->cache
;
597 for (i
= j
= 0; i
< istate
->cache_nr
; i
++) {
598 if (ce_array
[i
]->ce_flags
& CE_REMOVE
) {
600 cache_tree_invalidate_path(istate
,
602 untracked_cache_remove_from_index(istate
,
605 remove_name_hash(istate
, ce_array
[i
]);
606 save_or_free_index_entry(istate
, ce_array
[i
]);
609 ce_array
[j
++] = ce_array
[i
];
611 if (j
== istate
->cache_nr
)
613 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
614 istate
->cache_nr
= j
;
617 int remove_file_from_index(struct index_state
*istate
, const char *path
)
619 int pos
= index_name_pos(istate
, path
, strlen(path
));
622 cache_tree_invalidate_path(istate
, path
);
623 untracked_cache_remove_from_index(istate
, path
);
624 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
625 remove_index_entry_at(istate
, pos
);
629 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
631 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
634 static int index_name_pos_also_unmerged(struct index_state
*istate
,
635 const char *path
, int namelen
)
637 int pos
= index_name_pos(istate
, path
, namelen
);
638 struct cache_entry
*ce
;
643 /* maybe unmerged? */
645 if (pos
>= istate
->cache_nr
||
646 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
649 /* order of preference: stage 2, 1, 3 */
650 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
651 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
652 !compare_name(ce
, path
, namelen
))
657 static int different_name(struct cache_entry
*ce
, struct cache_entry
*alias
)
659 int len
= ce_namelen(ce
);
660 return ce_namelen(alias
) != len
|| memcmp(ce
->name
, alias
->name
, len
);
664 * If we add a filename that aliases in the cache, we will use the
665 * name that we already have - but we don't want to update the same
666 * alias twice, because that implies that there were actually two
667 * different files with aliasing names!
669 * So we use the CE_ADDED flag to verify that the alias was an old
670 * one before we accept it as
672 static struct cache_entry
*create_alias_ce(struct index_state
*istate
,
673 struct cache_entry
*ce
,
674 struct cache_entry
*alias
)
677 struct cache_entry
*new_entry
;
679 if (alias
->ce_flags
& CE_ADDED
)
680 die(_("will not add file alias '%s' ('%s' already exists in index)"),
681 ce
->name
, alias
->name
);
683 /* Ok, create the new entry using the name of the existing alias */
684 len
= ce_namelen(alias
);
685 new_entry
= make_empty_cache_entry(istate
, len
);
686 memcpy(new_entry
->name
, alias
->name
, len
);
687 copy_cache_entry(new_entry
, ce
);
688 save_or_free_index_entry(istate
, ce
);
692 void set_object_name_for_intent_to_add_entry(struct cache_entry
*ce
)
694 struct object_id oid
;
695 if (write_object_file("", 0, blob_type
, &oid
))
696 die(_("cannot create an empty blob in the object database"));
697 oidcpy(&ce
->oid
, &oid
);
700 int add_to_index(struct index_state
*istate
, const char *path
, struct stat
*st
, int flags
)
702 int namelen
, was_same
;
703 mode_t st_mode
= st
->st_mode
;
704 struct cache_entry
*ce
, *alias
= NULL
;
705 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_IGNORE_SKIP_WORKTREE
|CE_MATCH_RACY_IS_DIRTY
;
706 int verbose
= flags
& (ADD_CACHE_VERBOSE
| ADD_CACHE_PRETEND
);
707 int pretend
= flags
& ADD_CACHE_PRETEND
;
708 int intent_only
= flags
& ADD_CACHE_INTENT
;
709 int add_option
= (ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
|
710 (intent_only
? ADD_CACHE_NEW_ONLY
: 0));
711 int hash_flags
= HASH_WRITE_OBJECT
;
713 if (flags
& ADD_CACHE_RENORMALIZE
)
714 hash_flags
|= HASH_RENORMALIZE
;
716 if (!S_ISREG(st_mode
) && !S_ISLNK(st_mode
) && !S_ISDIR(st_mode
))
717 return error(_("%s: can only add regular files, symbolic links or git-directories"), path
);
719 namelen
= strlen(path
);
720 if (S_ISDIR(st_mode
)) {
721 while (namelen
&& path
[namelen
-1] == '/')
724 ce
= make_empty_cache_entry(istate
, namelen
);
725 memcpy(ce
->name
, path
, namelen
);
726 ce
->ce_namelen
= namelen
;
728 fill_stat_cache_info(ce
, st
);
730 ce
->ce_flags
|= CE_INTENT_TO_ADD
;
733 if (trust_executable_bit
&& has_symlinks
) {
734 ce
->ce_mode
= create_ce_mode(st_mode
);
736 /* If there is an existing entry, pick the mode bits and type
737 * from it, otherwise assume unexecutable regular file.
739 struct cache_entry
*ent
;
740 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
742 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
743 ce
->ce_mode
= ce_mode_from_stat(ent
, st_mode
);
746 /* When core.ignorecase=true, determine if a directory of the same name but differing
747 * case already exists within the Git repository. If it does, ensure the directory
748 * case of the file being added to the repository matches (is folded into) the existing
749 * entry's directory case.
752 adjust_dirname_case(istate
, ce
->name
);
754 if (!(flags
& ADD_CACHE_RENORMALIZE
)) {
755 alias
= index_file_exists(istate
, ce
->name
,
756 ce_namelen(ce
), ignore_case
);
759 !ie_match_stat(istate
, alias
, st
, ce_option
)) {
760 /* Nothing changed, really */
761 if (!S_ISGITLINK(alias
->ce_mode
))
762 ce_mark_uptodate(alias
);
763 alias
->ce_flags
|= CE_ADDED
;
765 discard_cache_entry(ce
);
770 if (index_path(istate
, &ce
->oid
, path
, st
, hash_flags
)) {
771 discard_cache_entry(ce
);
772 return error(_("unable to index file '%s'"), path
);
775 set_object_name_for_intent_to_add_entry(ce
);
777 if (ignore_case
&& alias
&& different_name(ce
, alias
))
778 ce
= create_alias_ce(istate
, ce
, alias
);
779 ce
->ce_flags
|= CE_ADDED
;
781 /* It was suspected to be racily clean, but it turns out to be Ok */
784 oideq(&alias
->oid
, &ce
->oid
) &&
785 ce
->ce_mode
== alias
->ce_mode
);
788 discard_cache_entry(ce
);
789 else if (add_index_entry(istate
, ce
, add_option
)) {
790 discard_cache_entry(ce
);
791 return error(_("unable to add '%s' to index"), path
);
793 if (verbose
&& !was_same
)
794 printf("add '%s'\n", path
);
798 int add_file_to_index(struct index_state
*istate
, const char *path
, int flags
)
801 if (lstat(path
, &st
))
802 die_errno(_("unable to stat '%s'"), path
);
803 return add_to_index(istate
, path
, &st
, flags
);
806 struct cache_entry
*make_empty_cache_entry(struct index_state
*istate
, size_t len
)
808 return mem_pool__ce_calloc(find_mem_pool(istate
), len
);
811 struct cache_entry
*make_empty_transient_cache_entry(size_t len
)
813 return xcalloc(1, cache_entry_size(len
));
816 struct cache_entry
*make_cache_entry(struct index_state
*istate
,
818 const struct object_id
*oid
,
821 unsigned int refresh_options
)
823 struct cache_entry
*ce
, *ret
;
826 if (!verify_path(path
, mode
)) {
827 error(_("invalid path '%s'"), path
);
832 ce
= make_empty_cache_entry(istate
, len
);
834 oidcpy(&ce
->oid
, oid
);
835 memcpy(ce
->name
, path
, len
);
836 ce
->ce_flags
= create_ce_flags(stage
);
837 ce
->ce_namelen
= len
;
838 ce
->ce_mode
= create_ce_mode(mode
);
840 ret
= refresh_cache_entry(istate
, ce
, refresh_options
);
842 discard_cache_entry(ce
);
846 struct cache_entry
*make_transient_cache_entry(unsigned int mode
, const struct object_id
*oid
,
847 const char *path
, int stage
)
849 struct cache_entry
*ce
;
852 if (!verify_path(path
, mode
)) {
853 error(_("invalid path '%s'"), path
);
858 ce
= make_empty_transient_cache_entry(len
);
860 oidcpy(&ce
->oid
, oid
);
861 memcpy(ce
->name
, path
, len
);
862 ce
->ce_flags
= create_ce_flags(stage
);
863 ce
->ce_namelen
= len
;
864 ce
->ce_mode
= create_ce_mode(mode
);
870 * Chmod an index entry with either +x or -x.
872 * Returns -1 if the chmod for the particular cache entry failed (if it's
873 * not a regular file), -2 if an invalid flip argument is passed in, 0
876 int chmod_index_entry(struct index_state
*istate
, struct cache_entry
*ce
,
879 if (!S_ISREG(ce
->ce_mode
))
886 ce
->ce_mode
&= ~0111;
891 cache_tree_invalidate_path(istate
, ce
->name
);
892 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
893 mark_fsmonitor_invalid(istate
, ce
);
894 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
899 int ce_same_name(const struct cache_entry
*a
, const struct cache_entry
*b
)
901 int len
= ce_namelen(a
);
902 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
906 * We fundamentally don't like some paths: we don't want
907 * dot or dot-dot anywhere, and for obvious reasons don't
908 * want to recurse into ".git" either.
910 * Also, we don't want double slashes or slashes at the
911 * end that can make pathnames ambiguous.
913 static int verify_dotfile(const char *rest
, unsigned mode
)
916 * The first character was '.', but that
917 * has already been discarded, we now test
921 /* "." is not allowed */
922 if (*rest
== '\0' || is_dir_sep(*rest
))
927 * ".git" followed by NUL or slash is bad. Note that we match
928 * case-insensitively here, even if ignore_case is not set.
929 * This outlaws ".GIT" everywhere out of an abundance of caution,
930 * since there's really no good reason to allow it.
932 * Once we've seen ".git", we can also find ".gitmodules", etc (also
933 * case-insensitively).
937 if (rest
[1] != 'i' && rest
[1] != 'I')
939 if (rest
[2] != 't' && rest
[2] != 'T')
941 if (rest
[3] == '\0' || is_dir_sep(rest
[3]))
945 if (skip_iprefix(rest
, "modules", &rest
) &&
946 (*rest
== '\0' || is_dir_sep(*rest
)))
951 if (rest
[1] == '\0' || is_dir_sep(rest
[1]))
957 int verify_path(const char *path
, unsigned mode
)
961 if (has_dos_drive_prefix(path
))
971 if (is_hfs_dotgit(path
))
974 if (is_hfs_dotgitmodules(path
))
979 if (is_ntfs_dotgit(path
))
982 if (is_ntfs_dotgitmodules(path
))
988 if ((c
== '.' && !verify_dotfile(path
, mode
)) ||
989 is_dir_sep(c
) || c
== '\0')
997 * Do we have another file that has the beginning components being a
998 * proper superset of the name we're trying to add?
1000 static int has_file_name(struct index_state
*istate
,
1001 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
1004 int len
= ce_namelen(ce
);
1005 int stage
= ce_stage(ce
);
1006 const char *name
= ce
->name
;
1008 while (pos
< istate
->cache_nr
) {
1009 struct cache_entry
*p
= istate
->cache
[pos
++];
1011 if (len
>= ce_namelen(p
))
1013 if (memcmp(name
, p
->name
, len
))
1015 if (ce_stage(p
) != stage
)
1017 if (p
->name
[len
] != '/')
1019 if (p
->ce_flags
& CE_REMOVE
)
1024 remove_index_entry_at(istate
, --pos
);
1031 * Like strcmp(), but also return the offset of the first change.
1032 * If strings are equal, return the length.
1034 int strcmp_offset(const char *s1
, const char *s2
, size_t *first_change
)
1039 return strcmp(s1
, s2
);
1041 for (k
= 0; s1
[k
] == s2
[k
]; k
++)
1046 return (unsigned char)s1
[k
] - (unsigned char)s2
[k
];
1050 * Do we have another file with a pathname that is a proper
1051 * subset of the name we're trying to add?
1053 * That is, is there another file in the index with a path
1054 * that matches a sub-directory in the given entry?
1056 static int has_dir_name(struct index_state
*istate
,
1057 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
1060 int stage
= ce_stage(ce
);
1061 const char *name
= ce
->name
;
1062 const char *slash
= name
+ ce_namelen(ce
);
1067 * We are frequently called during an iteration on a sorted
1068 * list of pathnames and while building a new index. Therefore,
1069 * there is a high probability that this entry will eventually
1070 * be appended to the index, rather than inserted in the middle.
1071 * If we can confirm that, we can avoid binary searches on the
1072 * components of the pathname.
1074 * Compare the entry's full path with the last path in the index.
1076 if (istate
->cache_nr
> 0) {
1077 cmp_last
= strcmp_offset(name
,
1078 istate
->cache
[istate
->cache_nr
- 1]->name
,
1081 if (len_eq_last
== 0) {
1083 * The entry sorts AFTER the last one in the
1084 * index and their paths have no common prefix,
1085 * so there cannot be a F/D conflict.
1090 * The entry sorts AFTER the last one in the
1091 * index, but has a common prefix. Fall through
1092 * to the loop below to disect the entry's path
1093 * and see where the difference is.
1096 } else if (cmp_last
== 0) {
1098 * The entry exactly matches the last one in the
1099 * index, but because of multiple stage and CE_REMOVE
1100 * items, we fall through and let the regular search
1110 if (*--slash
== '/')
1112 if (slash
<= ce
->name
)
1119 * (len + 1) is a directory boundary (including
1120 * the trailing slash). And since the loop is
1121 * decrementing "slash", the first iteration is
1122 * the longest directory prefix; subsequent
1123 * iterations consider parent directories.
1126 if (len
+ 1 <= len_eq_last
) {
1128 * The directory prefix (including the trailing
1129 * slash) also appears as a prefix in the last
1130 * entry, so the remainder cannot collide (because
1131 * strcmp said the whole path was greater).
1136 * LT: last: xxx/file_A
1142 if (len
> len_eq_last
) {
1144 * This part of the directory prefix (excluding
1145 * the trailing slash) is longer than the known
1146 * equal portions, so this sub-directory cannot
1147 * collide with a file.
1155 if (istate
->cache_nr
> 0 &&
1156 ce_namelen(istate
->cache
[istate
->cache_nr
- 1]) > len
) {
1158 * The directory prefix lines up with part of
1159 * a longer file or directory name, but sorts
1160 * after it, so this sub-directory cannot
1161 * collide with a file.
1163 * last: xxx/yy-file (because '-' sorts before '/')
1170 * This is a possible collision. Fall through and
1171 * let the regular search code handle it.
1178 pos
= index_name_stage_pos(istate
, name
, len
, stage
);
1181 * Found one, but not so fast. This could
1182 * be a marker that says "I was here, but
1183 * I am being removed". Such an entry is
1184 * not a part of the resulting tree, and
1185 * it is Ok to have a directory at the same
1188 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
1192 remove_index_entry_at(istate
, pos
);
1200 * Trivial optimization: if we find an entry that
1201 * already matches the sub-directory, then we know
1202 * we're ok, and we can exit.
1204 while (pos
< istate
->cache_nr
) {
1205 struct cache_entry
*p
= istate
->cache
[pos
];
1206 if ((ce_namelen(p
) <= len
) ||
1207 (p
->name
[len
] != '/') ||
1208 memcmp(p
->name
, name
, len
))
1209 break; /* not our subdirectory */
1210 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
1212 * p is at the same stage as our entry, and
1213 * is a subdirectory of what we are looking
1214 * at, so we cannot have conflicts at our
1215 * level or anything shorter.
1224 /* We may be in a situation where we already have path/file and path
1225 * is being added, or we already have path and path/file is being
1226 * added. Either one would result in a nonsense tree that has path
1227 * twice when git-write-tree tries to write it out. Prevent it.
1229 * If ok-to-replace is specified, we remove the conflicting entries
1230 * from the cache so the caller should recompute the insert position.
1231 * When this happens, we return non-zero.
1233 static int check_file_directory_conflict(struct index_state
*istate
,
1234 const struct cache_entry
*ce
,
1235 int pos
, int ok_to_replace
)
1240 * When ce is an "I am going away" entry, we allow it to be added
1242 if (ce
->ce_flags
& CE_REMOVE
)
1246 * We check if the path is a sub-path of a subsequent pathname
1247 * first, since removing those will not change the position
1250 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
1253 * Then check if the path might have a clashing sub-directory
1256 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
1259 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1262 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
1263 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
1264 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
1265 int new_only
= option
& ADD_CACHE_NEW_ONLY
;
1267 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1268 cache_tree_invalidate_path(istate
, ce
->name
);
1271 * If this entry's path sorts after the last entry in the index,
1272 * we can avoid searching for it.
1274 if (istate
->cache_nr
> 0 &&
1275 strcmp(ce
->name
, istate
->cache
[istate
->cache_nr
- 1]->name
) > 0)
1276 pos
= -istate
->cache_nr
- 1;
1278 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1280 /* existing match? Just replace it. */
1283 replace_index_entry(istate
, pos
, ce
);
1288 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1289 untracked_cache_add_to_index(istate
, ce
->name
);
1292 * Inserting a merged entry ("stage 0") into the index
1293 * will always replace all non-merged entries..
1295 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
1296 while (ce_same_name(istate
->cache
[pos
], ce
)) {
1298 if (!remove_index_entry_at(istate
, pos
))
1305 if (!verify_path(ce
->name
, ce
->ce_mode
))
1306 return error(_("invalid path '%s'"), ce
->name
);
1308 if (!skip_df_check
&&
1309 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
1311 return error(_("'%s' appears as both a file and as a directory"),
1313 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1319 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1323 if (option
& ADD_CACHE_JUST_APPEND
)
1324 pos
= istate
->cache_nr
;
1327 ret
= add_index_entry_with_check(istate
, ce
, option
);
1333 /* Make sure the array is big enough .. */
1334 ALLOC_GROW(istate
->cache
, istate
->cache_nr
+ 1, istate
->cache_alloc
);
1338 if (istate
->cache_nr
> pos
+ 1)
1339 MOVE_ARRAY(istate
->cache
+ pos
+ 1, istate
->cache
+ pos
,
1340 istate
->cache_nr
- pos
- 1);
1341 set_index_entry(istate
, pos
, ce
);
1342 istate
->cache_changed
|= CE_ENTRY_ADDED
;
1347 * "refresh" does not calculate a new sha1 file or bring the
1348 * cache up-to-date for mode/content changes. But what it
1349 * _does_ do is to "re-match" the stat information of a file
1350 * with the cache, so that you can refresh the cache for a
1351 * file that hasn't been changed but where the stat entry is
1354 * For example, you'd want to do this after doing a "git-read-tree",
1355 * to link up the stat cache details with the proper files.
1357 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
1358 struct cache_entry
*ce
,
1359 unsigned int options
, int *err
,
1363 struct cache_entry
*updated
;
1365 int refresh
= options
& CE_MATCH_REFRESH
;
1366 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
1367 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
1368 int ignore_missing
= options
& CE_MATCH_IGNORE_MISSING
;
1369 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
1371 if (!refresh
|| ce_uptodate(ce
))
1374 if (!ignore_fsmonitor
)
1375 refresh_fsmonitor(istate
);
1377 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1378 * that the change to the work tree does not matter and told
1381 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
)) {
1382 ce_mark_uptodate(ce
);
1385 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
)) {
1386 ce_mark_uptodate(ce
);
1389 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
)) {
1390 ce_mark_uptodate(ce
);
1394 if (has_symlink_leading_path(ce
->name
, ce_namelen(ce
))) {
1402 if (lstat(ce
->name
, &st
) < 0) {
1403 if (ignore_missing
&& errno
== ENOENT
)
1410 changed
= ie_match_stat(istate
, ce
, &st
, options
);
1412 *changed_ret
= changed
;
1415 * The path is unchanged. If we were told to ignore
1416 * valid bit, then we did the actual stat check and
1417 * found that the entry is unmodified. If the entry
1418 * is not marked VALID, this is the place to mark it
1419 * valid again, under "assume unchanged" mode.
1421 if (ignore_valid
&& assume_unchanged
&&
1422 !(ce
->ce_flags
& CE_VALID
))
1423 ; /* mark this one VALID again */
1426 * We do not mark the index itself "modified"
1427 * because CE_UPTODATE flag is in-core only;
1428 * we are not going to write this change out.
1430 if (!S_ISGITLINK(ce
->ce_mode
)) {
1431 ce_mark_uptodate(ce
);
1432 mark_fsmonitor_valid(ce
);
1438 if (ie_modified(istate
, ce
, &st
, options
)) {
1444 updated
= make_empty_cache_entry(istate
, ce_namelen(ce
));
1445 copy_cache_entry(updated
, ce
);
1446 memcpy(updated
->name
, ce
->name
, ce
->ce_namelen
+ 1);
1447 fill_stat_cache_info(updated
, &st
);
1449 * If ignore_valid is not set, we should leave CE_VALID bit
1450 * alone. Otherwise, paths marked with --no-assume-unchanged
1451 * (i.e. things to be edited) will reacquire CE_VALID bit
1452 * automatically, which is not really what we want.
1454 if (!ignore_valid
&& assume_unchanged
&&
1455 !(ce
->ce_flags
& CE_VALID
))
1456 updated
->ce_flags
&= ~CE_VALID
;
1458 /* istate->cache_changed is updated in the caller */
1462 static void show_file(const char * fmt
, const char * name
, int in_porcelain
,
1463 int * first
, const char *header_msg
)
1465 if (in_porcelain
&& *first
&& header_msg
) {
1466 printf("%s\n", header_msg
);
1472 int refresh_index(struct index_state
*istate
, unsigned int flags
,
1473 const struct pathspec
*pathspec
,
1474 char *seen
, const char *header_msg
)
1478 int really
= (flags
& REFRESH_REALLY
) != 0;
1479 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
1480 int quiet
= (flags
& REFRESH_QUIET
) != 0;
1481 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
1482 int ignore_submodules
= (flags
& REFRESH_IGNORE_SUBMODULES
) != 0;
1484 int in_porcelain
= (flags
& REFRESH_IN_PORCELAIN
);
1485 unsigned int options
= (CE_MATCH_REFRESH
|
1486 (really
? CE_MATCH_IGNORE_VALID
: 0) |
1487 (not_new
? CE_MATCH_IGNORE_MISSING
: 0));
1488 const char *modified_fmt
;
1489 const char *deleted_fmt
;
1490 const char *typechange_fmt
;
1491 const char *added_fmt
;
1492 const char *unmerged_fmt
;
1493 struct progress
*progress
= NULL
;
1495 if (flags
& REFRESH_PROGRESS
&& isatty(2))
1496 progress
= start_delayed_progress(_("Refresh index"),
1499 trace_performance_enter();
1500 modified_fmt
= in_porcelain
? "M\t%s\n" : "%s: needs update\n";
1501 deleted_fmt
= in_porcelain
? "D\t%s\n" : "%s: needs update\n";
1502 typechange_fmt
= in_porcelain
? "T\t%s\n" : "%s: needs update\n";
1503 added_fmt
= in_porcelain
? "A\t%s\n" : "%s: needs update\n";
1504 unmerged_fmt
= in_porcelain
? "U\t%s\n" : "%s: needs merge\n";
1506 * Use the multi-threaded preload_index() to refresh most of the
1507 * cache entries quickly then in the single threaded loop below,
1508 * we only have to do the special cases that are left.
1510 preload_index(istate
, pathspec
, 0);
1511 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1512 struct cache_entry
*ce
, *new_entry
;
1513 int cache_errno
= 0;
1517 ce
= istate
->cache
[i
];
1518 if (ignore_submodules
&& S_ISGITLINK(ce
->ce_mode
))
1521 if (pathspec
&& !ce_path_match(istate
, ce
, pathspec
, seen
))
1525 while ((i
< istate
->cache_nr
) &&
1526 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
1532 show_file(unmerged_fmt
, ce
->name
, in_porcelain
,
1533 &first
, header_msg
);
1541 new_entry
= refresh_cache_ent(istate
, ce
, options
, &cache_errno
, &changed
);
1542 if (new_entry
== ce
)
1545 display_progress(progress
, i
);
1549 if (really
&& cache_errno
== EINVAL
) {
1550 /* If we are doing --really-refresh that
1551 * means the index is not valid anymore.
1553 ce
->ce_flags
&= ~CE_VALID
;
1554 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
1555 mark_fsmonitor_invalid(istate
, ce
);
1556 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
1561 if (cache_errno
== ENOENT
)
1563 else if (ce_intent_to_add(ce
))
1564 fmt
= added_fmt
; /* must be before other checks */
1565 else if (changed
& TYPE_CHANGED
)
1566 fmt
= typechange_fmt
;
1570 ce
->name
, in_porcelain
, &first
, header_msg
);
1575 replace_index_entry(istate
, i
, new_entry
);
1578 display_progress(progress
, istate
->cache_nr
);
1579 stop_progress(&progress
);
1581 trace_performance_leave("refresh index");
1585 struct cache_entry
*refresh_cache_entry(struct index_state
*istate
,
1586 struct cache_entry
*ce
,
1587 unsigned int options
)
1589 return refresh_cache_ent(istate
, ce
, options
, NULL
, NULL
);
1593 /*****************************************************************
1595 *****************************************************************/
1597 #define INDEX_FORMAT_DEFAULT 3
1599 static unsigned int get_index_format_default(void)
1601 char *envversion
= getenv("GIT_INDEX_VERSION");
1604 unsigned int version
= INDEX_FORMAT_DEFAULT
;
1607 if (!git_config_get_int("index.version", &value
))
1609 if (version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1610 warning(_("index.version set, but the value is invalid.\n"
1611 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1612 return INDEX_FORMAT_DEFAULT
;
1617 version
= strtoul(envversion
, &endp
, 10);
1619 version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1620 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1621 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1622 version
= INDEX_FORMAT_DEFAULT
;
1628 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1629 * Again - this is just a (very strong in practice) heuristic that
1630 * the inode hasn't changed.
1632 * We save the fields in big-endian order to allow using the
1633 * index file over NFS transparently.
1635 struct ondisk_cache_entry
{
1636 struct cache_time ctime
;
1637 struct cache_time mtime
;
1644 unsigned char sha1
[20];
1646 char name
[FLEX_ARRAY
]; /* more */
1650 * This struct is used when CE_EXTENDED bit is 1
1651 * The struct must match ondisk_cache_entry exactly from
1654 struct ondisk_cache_entry_extended
{
1655 struct cache_time ctime
;
1656 struct cache_time mtime
;
1663 unsigned char sha1
[20];
1666 char name
[FLEX_ARRAY
]; /* more */
1669 /* These are only used for v3 or lower */
1670 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1671 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1672 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1673 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1674 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1675 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1676 ondisk_cache_entry_size(ce_namelen(ce)))
1678 /* Allow fsck to force verification of the index checksum. */
1679 int verify_index_checksum
;
1681 /* Allow fsck to force verification of the cache entry order. */
1682 int verify_ce_order
;
1684 static int verify_hdr(const struct cache_header
*hdr
, unsigned long size
)
1687 unsigned char hash
[GIT_MAX_RAWSZ
];
1690 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
1691 return error(_("bad signature 0x%08x"), hdr
->hdr_signature
);
1692 hdr_version
= ntohl(hdr
->hdr_version
);
1693 if (hdr_version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< hdr_version
)
1694 return error(_("bad index version %d"), hdr_version
);
1696 if (!verify_index_checksum
)
1699 the_hash_algo
->init_fn(&c
);
1700 the_hash_algo
->update_fn(&c
, hdr
, size
- the_hash_algo
->rawsz
);
1701 the_hash_algo
->final_fn(hash
, &c
);
1702 if (!hasheq(hash
, (unsigned char *)hdr
+ size
- the_hash_algo
->rawsz
))
1703 return error(_("bad index file sha1 signature"));
1707 static int read_index_extension(struct index_state
*istate
,
1708 const char *ext
, const char *data
, unsigned long sz
)
1710 switch (CACHE_EXT(ext
)) {
1711 case CACHE_EXT_TREE
:
1712 istate
->cache_tree
= cache_tree_read(data
, sz
);
1714 case CACHE_EXT_RESOLVE_UNDO
:
1715 istate
->resolve_undo
= resolve_undo_read(data
, sz
);
1717 case CACHE_EXT_LINK
:
1718 if (read_link_extension(istate
, data
, sz
))
1721 case CACHE_EXT_UNTRACKED
:
1722 istate
->untracked
= read_untracked_extension(data
, sz
);
1724 case CACHE_EXT_FSMONITOR
:
1725 read_fsmonitor_extension(istate
, data
, sz
);
1727 case CACHE_EXT_ENDOFINDEXENTRIES
:
1728 case CACHE_EXT_INDEXENTRYOFFSETTABLE
:
1729 /* already handled in do_read_index() */
1732 if (*ext
< 'A' || 'Z' < *ext
)
1733 return error(_("index uses %.4s extension, which we do not understand"),
1735 fprintf_ln(stderr
, _("ignoring %.4s extension"), ext
);
1741 static struct cache_entry
*create_from_disk(struct mem_pool
*ce_mem_pool
,
1742 unsigned int version
,
1743 struct ondisk_cache_entry
*ondisk
,
1744 unsigned long *ent_size
,
1745 const struct cache_entry
*previous_ce
)
1747 struct cache_entry
*ce
;
1751 size_t copy_len
= 0;
1753 * Adjacent cache entries tend to share the leading paths, so it makes
1754 * sense to only store the differences in later entries. In the v4
1755 * on-disk format of the index, each on-disk cache entry stores the
1756 * number of bytes to be stripped from the end of the previous name,
1757 * and the bytes to append to the result, to come up with its name.
1759 int expand_name_field
= version
== 4;
1761 /* On-disk flags are just 16 bits */
1762 flags
= get_be16(&ondisk
->flags
);
1763 len
= flags
& CE_NAMEMASK
;
1765 if (flags
& CE_EXTENDED
) {
1766 struct ondisk_cache_entry_extended
*ondisk2
;
1768 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
1769 extended_flags
= get_be16(&ondisk2
->flags2
) << 16;
1770 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1771 if (extended_flags
& ~CE_EXTENDED_FLAGS
)
1772 die(_("unknown index entry format 0x%08x"), extended_flags
);
1773 flags
|= extended_flags
;
1774 name
= ondisk2
->name
;
1777 name
= ondisk
->name
;
1779 if (expand_name_field
) {
1780 const unsigned char *cp
= (const unsigned char *)name
;
1781 size_t strip_len
, previous_len
;
1783 /* If we're at the begining of a block, ignore the previous name */
1784 strip_len
= decode_varint(&cp
);
1786 previous_len
= previous_ce
->ce_namelen
;
1787 if (previous_len
< strip_len
)
1788 die(_("malformed name field in the index, near path '%s'"),
1790 copy_len
= previous_len
- strip_len
;
1792 name
= (const char *)cp
;
1795 if (len
== CE_NAMEMASK
) {
1797 if (expand_name_field
)
1801 ce
= mem_pool__ce_alloc(ce_mem_pool
, len
);
1803 ce
->ce_stat_data
.sd_ctime
.sec
= get_be32(&ondisk
->ctime
.sec
);
1804 ce
->ce_stat_data
.sd_mtime
.sec
= get_be32(&ondisk
->mtime
.sec
);
1805 ce
->ce_stat_data
.sd_ctime
.nsec
= get_be32(&ondisk
->ctime
.nsec
);
1806 ce
->ce_stat_data
.sd_mtime
.nsec
= get_be32(&ondisk
->mtime
.nsec
);
1807 ce
->ce_stat_data
.sd_dev
= get_be32(&ondisk
->dev
);
1808 ce
->ce_stat_data
.sd_ino
= get_be32(&ondisk
->ino
);
1809 ce
->ce_mode
= get_be32(&ondisk
->mode
);
1810 ce
->ce_stat_data
.sd_uid
= get_be32(&ondisk
->uid
);
1811 ce
->ce_stat_data
.sd_gid
= get_be32(&ondisk
->gid
);
1812 ce
->ce_stat_data
.sd_size
= get_be32(&ondisk
->size
);
1813 ce
->ce_flags
= flags
& ~CE_NAMEMASK
;
1814 ce
->ce_namelen
= len
;
1816 hashcpy(ce
->oid
.hash
, ondisk
->sha1
);
1818 if (expand_name_field
) {
1820 memcpy(ce
->name
, previous_ce
->name
, copy_len
);
1821 memcpy(ce
->name
+ copy_len
, name
, len
+ 1 - copy_len
);
1822 *ent_size
= (name
- ((char *)ondisk
)) + len
+ 1 - copy_len
;
1824 memcpy(ce
->name
, name
, len
+ 1);
1825 *ent_size
= ondisk_ce_size(ce
);
1830 static void check_ce_order(struct index_state
*istate
)
1834 if (!verify_ce_order
)
1837 for (i
= 1; i
< istate
->cache_nr
; i
++) {
1838 struct cache_entry
*ce
= istate
->cache
[i
- 1];
1839 struct cache_entry
*next_ce
= istate
->cache
[i
];
1840 int name_compare
= strcmp(ce
->name
, next_ce
->name
);
1842 if (0 < name_compare
)
1843 die(_("unordered stage entries in index"));
1844 if (!name_compare
) {
1846 die(_("multiple stage entries for merged file '%s'"),
1848 if (ce_stage(ce
) > ce_stage(next_ce
))
1849 die(_("unordered stage entries for '%s'"),
1855 static void tweak_untracked_cache(struct index_state
*istate
)
1857 switch (git_config_get_untracked_cache()) {
1858 case -1: /* keep: do nothing */
1861 remove_untracked_cache(istate
);
1864 add_untracked_cache(istate
);
1866 default: /* unknown value: do nothing */
1871 static void tweak_split_index(struct index_state
*istate
)
1873 switch (git_config_get_split_index()) {
1874 case -1: /* unset: do nothing */
1877 remove_split_index(istate
);
1880 add_split_index(istate
);
1882 default: /* unknown value: do nothing */
1887 static void post_read_index_from(struct index_state
*istate
)
1889 check_ce_order(istate
);
1890 tweak_untracked_cache(istate
);
1891 tweak_split_index(istate
);
1892 tweak_fsmonitor(istate
);
1895 static size_t estimate_cache_size_from_compressed(unsigned int entries
)
1897 return entries
* (sizeof(struct cache_entry
) + CACHE_ENTRY_PATH_LENGTH
);
1900 static size_t estimate_cache_size(size_t ondisk_size
, unsigned int entries
)
1902 long per_entry
= sizeof(struct cache_entry
) - sizeof(struct ondisk_cache_entry
);
1905 * Account for potential alignment differences.
1907 per_entry
+= align_padding_size(sizeof(struct cache_entry
), -sizeof(struct ondisk_cache_entry
));
1908 return ondisk_size
+ entries
* per_entry
;
1911 struct index_entry_offset
1913 /* starting byte offset into index file, count of index entries in this block */
1917 struct index_entry_offset_table
1920 struct index_entry_offset entries
[FLEX_ARRAY
];
1923 static struct index_entry_offset_table
*read_ieot_extension(const char *mmap
, size_t mmap_size
, size_t offset
);
1924 static void write_ieot_extension(struct strbuf
*sb
, struct index_entry_offset_table
*ieot
);
1926 static size_t read_eoie_extension(const char *mmap
, size_t mmap_size
);
1927 static void write_eoie_extension(struct strbuf
*sb
, git_hash_ctx
*eoie_context
, size_t offset
);
1929 struct load_index_extensions
1932 struct index_state
*istate
;
1935 unsigned long src_offset
;
1938 static void *load_index_extensions(void *_data
)
1940 struct load_index_extensions
*p
= _data
;
1941 unsigned long src_offset
= p
->src_offset
;
1943 while (src_offset
<= p
->mmap_size
- the_hash_algo
->rawsz
- 8) {
1944 /* After an array of active_nr index entries,
1945 * there can be arbitrary number of extended
1946 * sections, each of which is prefixed with
1947 * extension name (4-byte) and section length
1948 * in 4-byte network byte order.
1950 uint32_t extsize
= get_be32(p
->mmap
+ src_offset
+ 4);
1951 if (read_index_extension(p
->istate
,
1952 p
->mmap
+ src_offset
,
1953 p
->mmap
+ src_offset
+ 8,
1955 munmap((void *)p
->mmap
, p
->mmap_size
);
1956 die(_("index file corrupt"));
1959 src_offset
+= extsize
;
1966 * A helper function that will load the specified range of cache entries
1967 * from the memory mapped file and add them to the given index.
1969 static unsigned long load_cache_entry_block(struct index_state
*istate
,
1970 struct mem_pool
*ce_mem_pool
, int offset
, int nr
, const char *mmap
,
1971 unsigned long start_offset
, const struct cache_entry
*previous_ce
)
1974 unsigned long src_offset
= start_offset
;
1976 for (i
= offset
; i
< offset
+ nr
; i
++) {
1977 struct ondisk_cache_entry
*disk_ce
;
1978 struct cache_entry
*ce
;
1979 unsigned long consumed
;
1981 disk_ce
= (struct ondisk_cache_entry
*)(mmap
+ src_offset
);
1982 ce
= create_from_disk(ce_mem_pool
, istate
->version
, disk_ce
, &consumed
, previous_ce
);
1983 set_index_entry(istate
, i
, ce
);
1985 src_offset
+= consumed
;
1988 return src_offset
- start_offset
;
1991 static unsigned long load_all_cache_entries(struct index_state
*istate
,
1992 const char *mmap
, size_t mmap_size
, unsigned long src_offset
)
1994 unsigned long consumed
;
1996 if (istate
->version
== 4) {
1997 mem_pool_init(&istate
->ce_mem_pool
,
1998 estimate_cache_size_from_compressed(istate
->cache_nr
));
2000 mem_pool_init(&istate
->ce_mem_pool
,
2001 estimate_cache_size(mmap_size
, istate
->cache_nr
));
2004 consumed
= load_cache_entry_block(istate
, istate
->ce_mem_pool
,
2005 0, istate
->cache_nr
, mmap
, src_offset
, NULL
);
2010 * Mostly randomly chosen maximum thread counts: we
2011 * cap the parallelism to online_cpus() threads, and we want
2012 * to have at least 10000 cache entries per thread for it to
2013 * be worth starting a thread.
2016 #define THREAD_COST (10000)
2018 struct load_cache_entries_thread_data
2021 struct index_state
*istate
;
2022 struct mem_pool
*ce_mem_pool
;
2025 struct index_entry_offset_table
*ieot
;
2026 int ieot_start
; /* starting index into the ieot array */
2027 int ieot_blocks
; /* count of ieot entries to process */
2028 unsigned long consumed
; /* return # of bytes in index file processed */
2032 * A thread proc to run the load_cache_entries() computation
2033 * across multiple background threads.
2035 static void *load_cache_entries_thread(void *_data
)
2037 struct load_cache_entries_thread_data
*p
= _data
;
2040 /* iterate across all ieot blocks assigned to this thread */
2041 for (i
= p
->ieot_start
; i
< p
->ieot_start
+ p
->ieot_blocks
; i
++) {
2042 p
->consumed
+= load_cache_entry_block(p
->istate
, p
->ce_mem_pool
,
2043 p
->offset
, p
->ieot
->entries
[i
].nr
, p
->mmap
, p
->ieot
->entries
[i
].offset
, NULL
);
2044 p
->offset
+= p
->ieot
->entries
[i
].nr
;
2049 static unsigned long load_cache_entries_threaded(struct index_state
*istate
, const char *mmap
, size_t mmap_size
,
2050 unsigned long src_offset
, int nr_threads
, struct index_entry_offset_table
*ieot
)
2052 int i
, offset
, ieot_blocks
, ieot_start
, err
;
2053 struct load_cache_entries_thread_data
*data
;
2054 unsigned long consumed
= 0;
2056 /* a little sanity checking */
2057 if (istate
->name_hash_initialized
)
2058 BUG("the name hash isn't thread safe");
2060 mem_pool_init(&istate
->ce_mem_pool
, 0);
2062 /* ensure we have no more threads than we have blocks to process */
2063 if (nr_threads
> ieot
->nr
)
2064 nr_threads
= ieot
->nr
;
2065 data
= xcalloc(nr_threads
, sizeof(*data
));
2067 offset
= ieot_start
= 0;
2068 ieot_blocks
= DIV_ROUND_UP(ieot
->nr
, nr_threads
);
2069 for (i
= 0; i
< nr_threads
; i
++) {
2070 struct load_cache_entries_thread_data
*p
= &data
[i
];
2073 if (ieot_start
+ ieot_blocks
> ieot
->nr
)
2074 ieot_blocks
= ieot
->nr
- ieot_start
;
2080 p
->ieot_start
= ieot_start
;
2081 p
->ieot_blocks
= ieot_blocks
;
2083 /* create a mem_pool for each thread */
2085 for (j
= p
->ieot_start
; j
< p
->ieot_start
+ p
->ieot_blocks
; j
++)
2086 nr
+= p
->ieot
->entries
[j
].nr
;
2087 if (istate
->version
== 4) {
2088 mem_pool_init(&p
->ce_mem_pool
,
2089 estimate_cache_size_from_compressed(nr
));
2091 mem_pool_init(&p
->ce_mem_pool
,
2092 estimate_cache_size(mmap_size
, nr
));
2095 err
= pthread_create(&p
->pthread
, NULL
, load_cache_entries_thread
, p
);
2097 die(_("unable to create load_cache_entries thread: %s"), strerror(err
));
2099 /* increment by the number of cache entries in the ieot block being processed */
2100 for (j
= 0; j
< ieot_blocks
; j
++)
2101 offset
+= ieot
->entries
[ieot_start
+ j
].nr
;
2102 ieot_start
+= ieot_blocks
;
2105 for (i
= 0; i
< nr_threads
; i
++) {
2106 struct load_cache_entries_thread_data
*p
= &data
[i
];
2108 err
= pthread_join(p
->pthread
, NULL
);
2110 die(_("unable to join load_cache_entries thread: %s"), strerror(err
));
2111 mem_pool_combine(istate
->ce_mem_pool
, p
->ce_mem_pool
);
2112 consumed
+= p
->consumed
;
2120 /* remember to discard_cache() before reading a different cache! */
2121 int do_read_index(struct index_state
*istate
, const char *path
, int must_exist
)
2125 unsigned long src_offset
;
2126 const struct cache_header
*hdr
;
2129 struct load_index_extensions p
;
2130 size_t extension_offset
= 0;
2131 int nr_threads
, cpus
;
2132 struct index_entry_offset_table
*ieot
= NULL
;
2134 if (istate
->initialized
)
2135 return istate
->cache_nr
;
2137 istate
->timestamp
.sec
= 0;
2138 istate
->timestamp
.nsec
= 0;
2139 fd
= open(path
, O_RDONLY
);
2141 if (!must_exist
&& errno
== ENOENT
)
2143 die_errno(_("%s: index file open failed"), path
);
2147 die_errno(_("%s: cannot stat the open index"), path
);
2149 mmap_size
= xsize_t(st
.st_size
);
2150 if (mmap_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2151 die(_("%s: index file smaller than expected"), path
);
2153 mmap
= xmmap(NULL
, mmap_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2154 if (mmap
== MAP_FAILED
)
2155 die_errno(_("%s: unable to map index file"), path
);
2158 hdr
= (const struct cache_header
*)mmap
;
2159 if (verify_hdr(hdr
, mmap_size
) < 0)
2162 hashcpy(istate
->oid
.hash
, (const unsigned char *)hdr
+ mmap_size
- the_hash_algo
->rawsz
);
2163 istate
->version
= ntohl(hdr
->hdr_version
);
2164 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
2165 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
2166 istate
->cache
= xcalloc(istate
->cache_alloc
, sizeof(*istate
->cache
));
2167 istate
->initialized
= 1;
2171 p
.mmap_size
= mmap_size
;
2173 src_offset
= sizeof(*hdr
);
2175 if (git_config_get_index_threads(&nr_threads
))
2178 /* TODO: does creating more threads than cores help? */
2180 nr_threads
= istate
->cache_nr
/ THREAD_COST
;
2181 cpus
= online_cpus();
2182 if (nr_threads
> cpus
)
2189 if (nr_threads
> 1) {
2190 extension_offset
= read_eoie_extension(mmap
, mmap_size
);
2191 if (extension_offset
) {
2194 p
.src_offset
= extension_offset
;
2195 err
= pthread_create(&p
.pthread
, NULL
, load_index_extensions
, &p
);
2197 die(_("unable to create load_index_extensions thread: %s"), strerror(err
));
2204 * Locate and read the index entry offset table so that we can use it
2205 * to multi-thread the reading of the cache entries.
2207 if (extension_offset
&& nr_threads
> 1)
2208 ieot
= read_ieot_extension(mmap
, mmap_size
, extension_offset
);
2211 src_offset
+= load_cache_entries_threaded(istate
, mmap
, mmap_size
, src_offset
, nr_threads
, ieot
);
2214 src_offset
+= load_all_cache_entries(istate
, mmap
, mmap_size
, src_offset
);
2217 istate
->timestamp
.sec
= st
.st_mtime
;
2218 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
2220 /* if we created a thread, join it otherwise load the extensions on the primary thread */
2221 if (extension_offset
) {
2222 int ret
= pthread_join(p
.pthread
, NULL
);
2224 die(_("unable to join load_index_extensions thread: %s"), strerror(ret
));
2226 p
.src_offset
= src_offset
;
2227 load_index_extensions(&p
);
2229 munmap((void *)mmap
, mmap_size
);
2232 * TODO trace2: replace "the_repository" with the actual repo instance
2233 * that is associated with the given "istate".
2235 trace2_data_intmax("index", the_repository
, "read/version",
2237 trace2_data_intmax("index", the_repository
, "read/cache_nr",
2240 return istate
->cache_nr
;
2243 munmap((void *)mmap
, mmap_size
);
2244 die(_("index file corrupt"));
2248 * Signal that the shared index is used by updating its mtime.
2250 * This way, shared index can be removed if they have not been used
2253 static void freshen_shared_index(const char *shared_index
, int warn
)
2255 if (!check_and_freshen_file(shared_index
, 1) && warn
)
2256 warning(_("could not freshen shared index '%s'"), shared_index
);
2259 int read_index_from(struct index_state
*istate
, const char *path
,
2262 struct split_index
*split_index
;
2267 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2268 if (istate
->initialized
)
2269 return istate
->cache_nr
;
2272 * TODO trace2: replace "the_repository" with the actual repo instance
2273 * that is associated with the given "istate".
2275 trace2_region_enter_printf("index", "do_read_index", the_repository
,
2277 trace_performance_enter();
2278 ret
= do_read_index(istate
, path
, 0);
2279 trace_performance_leave("read cache %s", path
);
2280 trace2_region_leave_printf("index", "do_read_index", the_repository
,
2283 split_index
= istate
->split_index
;
2284 if (!split_index
|| is_null_oid(&split_index
->base_oid
)) {
2285 post_read_index_from(istate
);
2289 trace_performance_enter();
2290 if (split_index
->base
)
2291 discard_index(split_index
->base
);
2293 split_index
->base
= xcalloc(1, sizeof(*split_index
->base
));
2295 base_oid_hex
= oid_to_hex(&split_index
->base_oid
);
2296 base_path
= xstrfmt("%s/sharedindex.%s", gitdir
, base_oid_hex
);
2297 trace2_region_enter_printf("index", "shared/do_read_index",
2298 the_repository
, "%s", base_path
);
2299 ret
= do_read_index(split_index
->base
, base_path
, 1);
2300 trace2_region_leave_printf("index", "shared/do_read_index",
2301 the_repository
, "%s", base_path
);
2302 if (!oideq(&split_index
->base_oid
, &split_index
->base
->oid
))
2303 die(_("broken index, expect %s in %s, got %s"),
2304 base_oid_hex
, base_path
,
2305 oid_to_hex(&split_index
->base
->oid
));
2307 freshen_shared_index(base_path
, 0);
2308 merge_base_index(istate
);
2309 post_read_index_from(istate
);
2310 trace_performance_leave("read cache %s", base_path
);
2315 int is_index_unborn(struct index_state
*istate
)
2317 return (!istate
->cache_nr
&& !istate
->timestamp
.sec
);
2320 int discard_index(struct index_state
*istate
)
2323 * Cache entries in istate->cache[] should have been allocated
2324 * from the memory pool associated with this index, or from an
2325 * associated split_index. There is no need to free individual
2326 * cache entries. validate_cache_entries can detect when this
2327 * assertion does not hold.
2329 validate_cache_entries(istate
);
2331 resolve_undo_clear_index(istate
);
2332 istate
->cache_nr
= 0;
2333 istate
->cache_changed
= 0;
2334 istate
->timestamp
.sec
= 0;
2335 istate
->timestamp
.nsec
= 0;
2336 free_name_hash(istate
);
2337 cache_tree_free(&(istate
->cache_tree
));
2338 istate
->initialized
= 0;
2339 FREE_AND_NULL(istate
->cache
);
2340 istate
->cache_alloc
= 0;
2341 discard_split_index(istate
);
2342 free_untracked_cache(istate
->untracked
);
2343 istate
->untracked
= NULL
;
2345 if (istate
->ce_mem_pool
) {
2346 mem_pool_discard(istate
->ce_mem_pool
, should_validate_cache_entries());
2347 istate
->ce_mem_pool
= NULL
;
2354 * Validate the cache entries of this index.
2355 * All cache entries associated with this index
2356 * should have been allocated by the memory pool
2357 * associated with this index, or by a referenced
2360 void validate_cache_entries(const struct index_state
*istate
)
2364 if (!should_validate_cache_entries() ||!istate
|| !istate
->initialized
)
2367 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2369 BUG("cache entry is not allocated from expected memory pool");
2370 } else if (!istate
->ce_mem_pool
||
2371 !mem_pool_contains(istate
->ce_mem_pool
, istate
->cache
[i
])) {
2372 if (!istate
->split_index
||
2373 !istate
->split_index
->base
||
2374 !istate
->split_index
->base
->ce_mem_pool
||
2375 !mem_pool_contains(istate
->split_index
->base
->ce_mem_pool
, istate
->cache
[i
])) {
2376 BUG("cache entry is not allocated from expected memory pool");
2381 if (istate
->split_index
)
2382 validate_cache_entries(istate
->split_index
->base
);
2385 int unmerged_index(const struct index_state
*istate
)
2388 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2389 if (ce_stage(istate
->cache
[i
]))
2395 int repo_index_has_changes(struct repository
*repo
,
2399 struct index_state
*istate
= repo
->index
;
2400 struct object_id cmp
;
2404 cmp
= tree
->object
.oid
;
2405 if (tree
|| !get_oid_tree("HEAD", &cmp
)) {
2406 struct diff_options opt
;
2408 repo_diff_setup(repo
, &opt
);
2409 opt
.flags
.exit_with_status
= 1;
2411 opt
.flags
.quick
= 1;
2412 do_diff_cache(&cmp
, &opt
);
2414 for (i
= 0; sb
&& i
< diff_queued_diff
.nr
; i
++) {
2416 strbuf_addch(sb
, ' ');
2417 strbuf_addstr(sb
, diff_queued_diff
.queue
[i
]->two
->path
);
2420 return opt
.flags
.has_changes
!= 0;
2422 for (i
= 0; sb
&& i
< istate
->cache_nr
; i
++) {
2424 strbuf_addch(sb
, ' ');
2425 strbuf_addstr(sb
, istate
->cache
[i
]->name
);
2427 return !!istate
->cache_nr
;
2431 #define WRITE_BUFFER_SIZE 8192
2432 static unsigned char write_buffer
[WRITE_BUFFER_SIZE
];
2433 static unsigned long write_buffer_len
;
2435 static int ce_write_flush(git_hash_ctx
*context
, int fd
)
2437 unsigned int buffered
= write_buffer_len
;
2439 the_hash_algo
->update_fn(context
, write_buffer
, buffered
);
2440 if (write_in_full(fd
, write_buffer
, buffered
) < 0)
2442 write_buffer_len
= 0;
2447 static int ce_write(git_hash_ctx
*context
, int fd
, void *data
, unsigned int len
)
2450 unsigned int buffered
= write_buffer_len
;
2451 unsigned int partial
= WRITE_BUFFER_SIZE
- buffered
;
2454 memcpy(write_buffer
+ buffered
, data
, partial
);
2455 buffered
+= partial
;
2456 if (buffered
== WRITE_BUFFER_SIZE
) {
2457 write_buffer_len
= buffered
;
2458 if (ce_write_flush(context
, fd
))
2462 write_buffer_len
= buffered
;
2464 data
= (char *) data
+ partial
;
2469 static int write_index_ext_header(git_hash_ctx
*context
, git_hash_ctx
*eoie_context
,
2470 int fd
, unsigned int ext
, unsigned int sz
)
2475 the_hash_algo
->update_fn(eoie_context
, &ext
, 4);
2476 the_hash_algo
->update_fn(eoie_context
, &sz
, 4);
2478 return ((ce_write(context
, fd
, &ext
, 4) < 0) ||
2479 (ce_write(context
, fd
, &sz
, 4) < 0)) ? -1 : 0;
2482 static int ce_flush(git_hash_ctx
*context
, int fd
, unsigned char *hash
)
2484 unsigned int left
= write_buffer_len
;
2487 write_buffer_len
= 0;
2488 the_hash_algo
->update_fn(context
, write_buffer
, left
);
2491 /* Flush first if not enough space for hash signature */
2492 if (left
+ the_hash_algo
->rawsz
> WRITE_BUFFER_SIZE
) {
2493 if (write_in_full(fd
, write_buffer
, left
) < 0)
2498 /* Append the hash signature at the end */
2499 the_hash_algo
->final_fn(write_buffer
+ left
, context
);
2500 hashcpy(hash
, write_buffer
+ left
);
2501 left
+= the_hash_algo
->rawsz
;
2502 return (write_in_full(fd
, write_buffer
, left
) < 0) ? -1 : 0;
2505 static void ce_smudge_racily_clean_entry(struct index_state
*istate
,
2506 struct cache_entry
*ce
)
2509 * The only thing we care about in this function is to smudge the
2510 * falsely clean entry due to touch-update-touch race, so we leave
2511 * everything else as they are. We are called for entries whose
2512 * ce_stat_data.sd_mtime match the index file mtime.
2514 * Note that this actually does not do much for gitlinks, for
2515 * which ce_match_stat_basic() always goes to the actual
2516 * contents. The caller checks with is_racy_timestamp() which
2517 * always says "no" for gitlinks, so we are not called for them ;-)
2521 if (lstat(ce
->name
, &st
) < 0)
2523 if (ce_match_stat_basic(ce
, &st
))
2525 if (ce_modified_check_fs(istate
, ce
, &st
)) {
2526 /* This is "racily clean"; smudge it. Note that this
2527 * is a tricky code. At first glance, it may appear
2528 * that it can break with this sequence:
2530 * $ echo xyzzy >frotz
2531 * $ git-update-index --add frotz
2534 * $ echo filfre >nitfol
2535 * $ git-update-index --add nitfol
2537 * but it does not. When the second update-index runs,
2538 * it notices that the entry "frotz" has the same timestamp
2539 * as index, and if we were to smudge it by resetting its
2540 * size to zero here, then the object name recorded
2541 * in index is the 6-byte file but the cached stat information
2542 * becomes zero --- which would then match what we would
2543 * obtain from the filesystem next time we stat("frotz").
2545 * However, the second update-index, before calling
2546 * this function, notices that the cached size is 6
2547 * bytes and what is on the filesystem is an empty
2548 * file, and never calls us, so the cached size information
2549 * for "frotz" stays 6 which does not match the filesystem.
2551 ce
->ce_stat_data
.sd_size
= 0;
2555 /* Copy miscellaneous fields but not the name */
2556 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry
*ondisk
,
2557 struct cache_entry
*ce
)
2561 ondisk
->ctime
.sec
= htonl(ce
->ce_stat_data
.sd_ctime
.sec
);
2562 ondisk
->mtime
.sec
= htonl(ce
->ce_stat_data
.sd_mtime
.sec
);
2563 ondisk
->ctime
.nsec
= htonl(ce
->ce_stat_data
.sd_ctime
.nsec
);
2564 ondisk
->mtime
.nsec
= htonl(ce
->ce_stat_data
.sd_mtime
.nsec
);
2565 ondisk
->dev
= htonl(ce
->ce_stat_data
.sd_dev
);
2566 ondisk
->ino
= htonl(ce
->ce_stat_data
.sd_ino
);
2567 ondisk
->mode
= htonl(ce
->ce_mode
);
2568 ondisk
->uid
= htonl(ce
->ce_stat_data
.sd_uid
);
2569 ondisk
->gid
= htonl(ce
->ce_stat_data
.sd_gid
);
2570 ondisk
->size
= htonl(ce
->ce_stat_data
.sd_size
);
2571 hashcpy(ondisk
->sha1
, ce
->oid
.hash
);
2573 flags
= ce
->ce_flags
& ~CE_NAMEMASK
;
2574 flags
|= (ce_namelen(ce
) >= CE_NAMEMASK
? CE_NAMEMASK
: ce_namelen(ce
));
2575 ondisk
->flags
= htons(flags
);
2576 if (ce
->ce_flags
& CE_EXTENDED
) {
2577 struct ondisk_cache_entry_extended
*ondisk2
;
2578 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
2579 ondisk2
->flags2
= htons((ce
->ce_flags
& CE_EXTENDED_FLAGS
) >> 16);
2583 static int ce_write_entry(git_hash_ctx
*c
, int fd
, struct cache_entry
*ce
,
2584 struct strbuf
*previous_name
, struct ondisk_cache_entry
*ondisk
)
2588 unsigned int saved_namelen
;
2589 int stripped_name
= 0;
2590 static unsigned char padding
[8] = { 0x00 };
2592 if (ce
->ce_flags
& CE_STRIP_NAME
) {
2593 saved_namelen
= ce_namelen(ce
);
2598 if (ce
->ce_flags
& CE_EXTENDED
)
2599 size
= offsetof(struct ondisk_cache_entry_extended
, name
);
2601 size
= offsetof(struct ondisk_cache_entry
, name
);
2603 if (!previous_name
) {
2604 int len
= ce_namelen(ce
);
2605 copy_cache_entry_to_ondisk(ondisk
, ce
);
2606 result
= ce_write(c
, fd
, ondisk
, size
);
2608 result
= ce_write(c
, fd
, ce
->name
, len
);
2610 result
= ce_write(c
, fd
, padding
, align_padding_size(size
, len
));
2612 int common
, to_remove
, prefix_size
;
2613 unsigned char to_remove_vi
[16];
2615 (ce
->name
[common
] &&
2616 common
< previous_name
->len
&&
2617 ce
->name
[common
] == previous_name
->buf
[common
]);
2619 ; /* still matching */
2620 to_remove
= previous_name
->len
- common
;
2621 prefix_size
= encode_varint(to_remove
, to_remove_vi
);
2623 copy_cache_entry_to_ondisk(ondisk
, ce
);
2624 result
= ce_write(c
, fd
, ondisk
, size
);
2626 result
= ce_write(c
, fd
, to_remove_vi
, prefix_size
);
2628 result
= ce_write(c
, fd
, ce
->name
+ common
, ce_namelen(ce
) - common
);
2630 result
= ce_write(c
, fd
, padding
, 1);
2632 strbuf_splice(previous_name
, common
, to_remove
,
2633 ce
->name
+ common
, ce_namelen(ce
) - common
);
2635 if (stripped_name
) {
2636 ce
->ce_namelen
= saved_namelen
;
2637 ce
->ce_flags
&= ~CE_STRIP_NAME
;
2644 * This function verifies if index_state has the correct sha1 of the
2645 * index file. Don't die if we have any other failure, just return 0.
2647 static int verify_index_from(const struct index_state
*istate
, const char *path
)
2652 unsigned char hash
[GIT_MAX_RAWSZ
];
2654 if (!istate
->initialized
)
2657 fd
= open(path
, O_RDONLY
);
2664 if (st
.st_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2667 n
= pread_in_full(fd
, hash
, the_hash_algo
->rawsz
, st
.st_size
- the_hash_algo
->rawsz
);
2668 if (n
!= the_hash_algo
->rawsz
)
2671 if (!hasheq(istate
->oid
.hash
, hash
))
2682 static int repo_verify_index(struct repository
*repo
)
2684 return verify_index_from(repo
->index
, repo
->index_file
);
2687 static int has_racy_timestamp(struct index_state
*istate
)
2689 int entries
= istate
->cache_nr
;
2692 for (i
= 0; i
< entries
; i
++) {
2693 struct cache_entry
*ce
= istate
->cache
[i
];
2694 if (is_racy_timestamp(istate
, ce
))
2700 void repo_update_index_if_able(struct repository
*repo
,
2701 struct lock_file
*lockfile
)
2703 if ((repo
->index
->cache_changed
||
2704 has_racy_timestamp(repo
->index
)) &&
2705 repo_verify_index(repo
))
2706 write_locked_index(repo
->index
, lockfile
, COMMIT_LOCK
);
2708 rollback_lock_file(lockfile
);
2711 static int record_eoie(void)
2715 if (!git_config_get_bool("index.recordendofindexentries", &val
))
2719 * As a convenience, the end of index entries extension
2720 * used for threading is written by default if the user
2721 * explicitly requested threaded index reads.
2723 return !git_config_get_index_threads(&val
) && val
!= 1;
2726 static int record_ieot(void)
2730 if (!git_config_get_bool("index.recordoffsettable", &val
))
2734 * As a convenience, the offset table used for threading is
2735 * written by default if the user explicitly requested
2736 * threaded index reads.
2738 return !git_config_get_index_threads(&val
) && val
!= 1;
2742 * On success, `tempfile` is closed. If it is the temporary file
2743 * of a `struct lock_file`, we will therefore effectively perform
2744 * a 'close_lock_file_gently()`. Since that is an implementation
2745 * detail of lockfiles, callers of `do_write_index()` should not
2748 static int do_write_index(struct index_state
*istate
, struct tempfile
*tempfile
,
2749 int strip_extensions
)
2751 uint64_t start
= getnanotime();
2752 int newfd
= tempfile
->fd
;
2753 git_hash_ctx c
, eoie_c
;
2754 struct cache_header hdr
;
2755 int i
, err
= 0, removed
, extended
, hdr_version
;
2756 struct cache_entry
**cache
= istate
->cache
;
2757 int entries
= istate
->cache_nr
;
2759 struct ondisk_cache_entry_extended ondisk
;
2760 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
2761 int drop_cache_tree
= istate
->drop_cache_tree
;
2763 int ieot_entries
= 1;
2764 struct index_entry_offset_table
*ieot
= NULL
;
2767 for (i
= removed
= extended
= 0; i
< entries
; i
++) {
2768 if (cache
[i
]->ce_flags
& CE_REMOVE
)
2771 /* reduce extended entries if possible */
2772 cache
[i
]->ce_flags
&= ~CE_EXTENDED
;
2773 if (cache
[i
]->ce_flags
& CE_EXTENDED_FLAGS
) {
2775 cache
[i
]->ce_flags
|= CE_EXTENDED
;
2779 if (!istate
->version
) {
2780 istate
->version
= get_index_format_default();
2781 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0))
2782 init_split_index(istate
);
2785 /* demote version 3 to version 2 when the latter suffices */
2786 if (istate
->version
== 3 || istate
->version
== 2)
2787 istate
->version
= extended
? 3 : 2;
2789 hdr_version
= istate
->version
;
2791 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
2792 hdr
.hdr_version
= htonl(hdr_version
);
2793 hdr
.hdr_entries
= htonl(entries
- removed
);
2795 the_hash_algo
->init_fn(&c
);
2796 if (ce_write(&c
, newfd
, &hdr
, sizeof(hdr
)) < 0)
2799 if (!HAVE_THREADS
|| git_config_get_index_threads(&nr_threads
))
2802 if (nr_threads
!= 1 && record_ieot()) {
2803 int ieot_blocks
, cpus
;
2806 * ensure default number of ieot blocks maps evenly to the
2807 * default number of threads that will process them leaving
2808 * room for the thread to load the index extensions.
2811 ieot_blocks
= istate
->cache_nr
/ THREAD_COST
;
2812 cpus
= online_cpus();
2813 if (ieot_blocks
> cpus
- 1)
2814 ieot_blocks
= cpus
- 1;
2816 ieot_blocks
= nr_threads
;
2817 if (ieot_blocks
> istate
->cache_nr
)
2818 ieot_blocks
= istate
->cache_nr
;
2822 * no reason to write out the IEOT extension if we don't
2823 * have enough blocks to utilize multi-threading
2825 if (ieot_blocks
> 1) {
2826 ieot
= xcalloc(1, sizeof(struct index_entry_offset_table
)
2827 + (ieot_blocks
* sizeof(struct index_entry_offset
)));
2828 ieot_entries
= DIV_ROUND_UP(entries
, ieot_blocks
);
2832 offset
= lseek(newfd
, 0, SEEK_CUR
);
2837 offset
+= write_buffer_len
;
2839 previous_name
= (hdr_version
== 4) ? &previous_name_buf
: NULL
;
2841 for (i
= 0; i
< entries
; i
++) {
2842 struct cache_entry
*ce
= cache
[i
];
2843 if (ce
->ce_flags
& CE_REMOVE
)
2845 if (!ce_uptodate(ce
) && is_racy_timestamp(istate
, ce
))
2846 ce_smudge_racily_clean_entry(istate
, ce
);
2847 if (is_null_oid(&ce
->oid
)) {
2848 static const char msg
[] = "cache entry has null sha1: %s";
2849 static int allow
= -1;
2852 allow
= git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2854 warning(msg
, ce
->name
);
2856 err
= error(msg
, ce
->name
);
2858 drop_cache_tree
= 1;
2860 if (ieot
&& i
&& (i
% ieot_entries
== 0)) {
2861 ieot
->entries
[ieot
->nr
].nr
= nr
;
2862 ieot
->entries
[ieot
->nr
].offset
= offset
;
2865 * If we have a V4 index, set the first byte to an invalid
2866 * character to ensure there is nothing common with the previous
2870 previous_name
->buf
[0] = 0;
2872 offset
= lseek(newfd
, 0, SEEK_CUR
);
2877 offset
+= write_buffer_len
;
2879 if (ce_write_entry(&c
, newfd
, ce
, previous_name
, (struct ondisk_cache_entry
*)&ondisk
) < 0)
2887 ieot
->entries
[ieot
->nr
].nr
= nr
;
2888 ieot
->entries
[ieot
->nr
].offset
= offset
;
2891 strbuf_release(&previous_name_buf
);
2898 /* Write extension data here */
2899 offset
= lseek(newfd
, 0, SEEK_CUR
);
2904 offset
+= write_buffer_len
;
2905 the_hash_algo
->init_fn(&eoie_c
);
2908 * Lets write out CACHE_EXT_INDEXENTRYOFFSETTABLE first so that we
2909 * can minimize the number of extensions we have to scan through to
2910 * find it during load. Write it out regardless of the
2911 * strip_extensions parameter as we need it when loading the shared
2915 struct strbuf sb
= STRBUF_INIT
;
2917 write_ieot_extension(&sb
, ieot
);
2918 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_INDEXENTRYOFFSETTABLE
, sb
.len
) < 0
2919 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2920 strbuf_release(&sb
);
2926 if (!strip_extensions
&& istate
->split_index
&&
2927 !is_null_oid(&istate
->split_index
->base_oid
)) {
2928 struct strbuf sb
= STRBUF_INIT
;
2930 err
= write_link_extension(&sb
, istate
) < 0 ||
2931 write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_LINK
,
2933 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2934 strbuf_release(&sb
);
2938 if (!strip_extensions
&& !drop_cache_tree
&& istate
->cache_tree
) {
2939 struct strbuf sb
= STRBUF_INIT
;
2941 cache_tree_write(&sb
, istate
->cache_tree
);
2942 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_TREE
, sb
.len
) < 0
2943 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2944 strbuf_release(&sb
);
2948 if (!strip_extensions
&& istate
->resolve_undo
) {
2949 struct strbuf sb
= STRBUF_INIT
;
2951 resolve_undo_write(&sb
, istate
->resolve_undo
);
2952 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_RESOLVE_UNDO
,
2954 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2955 strbuf_release(&sb
);
2959 if (!strip_extensions
&& istate
->untracked
) {
2960 struct strbuf sb
= STRBUF_INIT
;
2962 write_untracked_extension(&sb
, istate
->untracked
);
2963 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_UNTRACKED
,
2965 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2966 strbuf_release(&sb
);
2970 if (!strip_extensions
&& istate
->fsmonitor_last_update
) {
2971 struct strbuf sb
= STRBUF_INIT
;
2973 write_fsmonitor_extension(&sb
, istate
);
2974 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_FSMONITOR
, sb
.len
) < 0
2975 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2976 strbuf_release(&sb
);
2982 * CACHE_EXT_ENDOFINDEXENTRIES must be written as the last entry before the SHA1
2983 * so that it can be found and processed before all the index entries are
2984 * read. Write it out regardless of the strip_extensions parameter as we need it
2985 * when loading the shared index.
2987 if (offset
&& record_eoie()) {
2988 struct strbuf sb
= STRBUF_INIT
;
2990 write_eoie_extension(&sb
, &eoie_c
, offset
);
2991 err
= write_index_ext_header(&c
, NULL
, newfd
, CACHE_EXT_ENDOFINDEXENTRIES
, sb
.len
) < 0
2992 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2993 strbuf_release(&sb
);
2998 if (ce_flush(&c
, newfd
, istate
->oid
.hash
))
3000 if (close_tempfile_gently(tempfile
)) {
3001 error(_("could not close '%s'"), tempfile
->filename
.buf
);
3004 if (stat(tempfile
->filename
.buf
, &st
))
3006 istate
->timestamp
.sec
= (unsigned int)st
.st_mtime
;
3007 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
3008 trace_performance_since(start
, "write index, changed mask = %x", istate
->cache_changed
);
3011 * TODO trace2: replace "the_repository" with the actual repo instance
3012 * that is associated with the given "istate".
3014 trace2_data_intmax("index", the_repository
, "write/version",
3016 trace2_data_intmax("index", the_repository
, "write/cache_nr",
3022 void set_alternate_index_output(const char *name
)
3024 alternate_index_output
= name
;
3027 static int commit_locked_index(struct lock_file
*lk
)
3029 if (alternate_index_output
)
3030 return commit_lock_file_to(lk
, alternate_index_output
);
3032 return commit_lock_file(lk
);
3035 static int do_write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
3041 * TODO trace2: replace "the_repository" with the actual repo instance
3042 * that is associated with the given "istate".
3044 trace2_region_enter_printf("index", "do_write_index", the_repository
,
3045 "%s", lock
->tempfile
->filename
.buf
);
3046 ret
= do_write_index(istate
, lock
->tempfile
, 0);
3047 trace2_region_leave_printf("index", "do_write_index", the_repository
,
3048 "%s", lock
->tempfile
->filename
.buf
);
3052 if (flags
& COMMIT_LOCK
)
3053 ret
= commit_locked_index(lock
);
3055 ret
= close_lock_file_gently(lock
);
3057 run_hook_le(NULL
, "post-index-change",
3058 istate
->updated_workdir
? "1" : "0",
3059 istate
->updated_skipworktree
? "1" : "0", NULL
);
3060 istate
->updated_workdir
= 0;
3061 istate
->updated_skipworktree
= 0;
3066 static int write_split_index(struct index_state
*istate
,
3067 struct lock_file
*lock
,
3071 prepare_to_write_split_index(istate
);
3072 ret
= do_write_locked_index(istate
, lock
, flags
);
3073 finish_writing_split_index(istate
);
3077 static const char *shared_index_expire
= "2.weeks.ago";
3079 static unsigned long get_shared_index_expire_date(void)
3081 static unsigned long shared_index_expire_date
;
3082 static int shared_index_expire_date_prepared
;
3084 if (!shared_index_expire_date_prepared
) {
3085 git_config_get_expiry("splitindex.sharedindexexpire",
3086 &shared_index_expire
);
3087 shared_index_expire_date
= approxidate(shared_index_expire
);
3088 shared_index_expire_date_prepared
= 1;
3091 return shared_index_expire_date
;
3094 static int should_delete_shared_index(const char *shared_index_path
)
3097 unsigned long expiration
;
3099 /* Check timestamp */
3100 expiration
= get_shared_index_expire_date();
3103 if (stat(shared_index_path
, &st
))
3104 return error_errno(_("could not stat '%s'"), shared_index_path
);
3105 if (st
.st_mtime
> expiration
)
3111 static int clean_shared_index_files(const char *current_hex
)
3114 DIR *dir
= opendir(get_git_dir());
3117 return error_errno(_("unable to open git dir: %s"), get_git_dir());
3119 while ((de
= readdir(dir
)) != NULL
) {
3120 const char *sha1_hex
;
3121 const char *shared_index_path
;
3122 if (!skip_prefix(de
->d_name
, "sharedindex.", &sha1_hex
))
3124 if (!strcmp(sha1_hex
, current_hex
))
3126 shared_index_path
= git_path("%s", de
->d_name
);
3127 if (should_delete_shared_index(shared_index_path
) > 0 &&
3128 unlink(shared_index_path
))
3129 warning_errno(_("unable to unlink: %s"), shared_index_path
);
3136 static int write_shared_index(struct index_state
*istate
,
3137 struct tempfile
**temp
)
3139 struct split_index
*si
= istate
->split_index
;
3142 move_cache_to_base_index(istate
);
3144 trace2_region_enter_printf("index", "shared/do_write_index",
3145 the_repository
, "%s", (*temp
)->filename
.buf
);
3146 ret
= do_write_index(si
->base
, *temp
, 1);
3147 trace2_region_enter_printf("index", "shared/do_write_index",
3148 the_repository
, "%s", (*temp
)->filename
.buf
);
3152 ret
= adjust_shared_perm(get_tempfile_path(*temp
));
3154 error(_("cannot fix permission bits on '%s'"), get_tempfile_path(*temp
));
3157 ret
= rename_tempfile(temp
,
3158 git_path("sharedindex.%s", oid_to_hex(&si
->base
->oid
)));
3160 oidcpy(&si
->base_oid
, &si
->base
->oid
);
3161 clean_shared_index_files(oid_to_hex(&si
->base
->oid
));
3167 static const int default_max_percent_split_change
= 20;
3169 static int too_many_not_shared_entries(struct index_state
*istate
)
3171 int i
, not_shared
= 0;
3172 int max_split
= git_config_get_max_percent_split_change();
3174 switch (max_split
) {
3176 /* not or badly configured: use the default value */
3177 max_split
= default_max_percent_split_change
;
3180 return 1; /* 0% means always write a new shared index */
3182 return 0; /* 100% means never write a new shared index */
3184 break; /* just use the configured value */
3187 /* Count not shared entries */
3188 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3189 struct cache_entry
*ce
= istate
->cache
[i
];
3194 return (int64_t)istate
->cache_nr
* max_split
< (int64_t)not_shared
* 100;
3197 int write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
3200 int new_shared_index
, ret
;
3201 struct split_index
*si
= istate
->split_index
;
3203 if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0))
3204 cache_tree_verify(the_repository
, istate
);
3206 if ((flags
& SKIP_IF_UNCHANGED
) && !istate
->cache_changed
) {
3207 if (flags
& COMMIT_LOCK
)
3208 rollback_lock_file(lock
);
3212 if (istate
->fsmonitor_last_update
)
3213 fill_fsmonitor_bitmap(istate
);
3215 if (!si
|| alternate_index_output
||
3216 (istate
->cache_changed
& ~EXTMASK
)) {
3218 oidclr(&si
->base_oid
);
3219 ret
= do_write_locked_index(istate
, lock
, flags
);
3223 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0)) {
3224 int v
= si
->base_oid
.hash
[0];
3226 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3228 if (too_many_not_shared_entries(istate
))
3229 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3231 new_shared_index
= istate
->cache_changed
& SPLIT_INDEX_ORDERED
;
3233 if (new_shared_index
) {
3234 struct tempfile
*temp
;
3237 /* Same initial permissions as the main .git/index file */
3238 temp
= mks_tempfile_sm(git_path("sharedindex_XXXXXX"), 0, 0666);
3240 oidclr(&si
->base_oid
);
3241 ret
= do_write_locked_index(istate
, lock
, flags
);
3244 ret
= write_shared_index(istate
, &temp
);
3246 saved_errno
= errno
;
3247 if (is_tempfile_active(temp
))
3248 delete_tempfile(&temp
);
3249 errno
= saved_errno
;
3255 ret
= write_split_index(istate
, lock
, flags
);
3257 /* Freshen the shared index only if the split-index was written */
3258 if (!ret
&& !new_shared_index
&& !is_null_oid(&si
->base_oid
)) {
3259 const char *shared_index
= git_path("sharedindex.%s",
3260 oid_to_hex(&si
->base_oid
));
3261 freshen_shared_index(shared_index
, 1);
3265 if (flags
& COMMIT_LOCK
)
3266 rollback_lock_file(lock
);
3271 * Read the index file that is potentially unmerged into given
3272 * index_state, dropping any unmerged entries to stage #0 (potentially
3273 * resulting in a path appearing as both a file and a directory in the
3274 * index; the caller is responsible to clear out the extra entries
3275 * before writing the index to a tree). Returns true if the index is
3276 * unmerged. Callers who want to refuse to work from an unmerged
3277 * state can call this and check its return value, instead of calling
3280 int repo_read_index_unmerged(struct repository
*repo
)
3282 struct index_state
*istate
;
3286 repo_read_index(repo
);
3287 istate
= repo
->index
;
3288 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3289 struct cache_entry
*ce
= istate
->cache
[i
];
3290 struct cache_entry
*new_ce
;
3296 len
= ce_namelen(ce
);
3297 new_ce
= make_empty_cache_entry(istate
, len
);
3298 memcpy(new_ce
->name
, ce
->name
, len
);
3299 new_ce
->ce_flags
= create_ce_flags(0) | CE_CONFLICTED
;
3300 new_ce
->ce_namelen
= len
;
3301 new_ce
->ce_mode
= ce
->ce_mode
;
3302 if (add_index_entry(istate
, new_ce
, ADD_CACHE_SKIP_DFCHECK
))
3303 return error(_("%s: cannot drop to stage #0"),
3310 * Returns 1 if the path is an "other" path with respect to
3311 * the index; that is, the path is not mentioned in the index at all,
3312 * either as a file, a directory with some files in the index,
3313 * or as an unmerged entry.
3315 * We helpfully remove a trailing "/" from directories so that
3316 * the output of read_directory can be used as-is.
3318 int index_name_is_other(const struct index_state
*istate
, const char *name
,
3322 if (namelen
&& name
[namelen
- 1] == '/')
3324 pos
= index_name_pos(istate
, name
, namelen
);
3326 return 0; /* exact match */
3328 if (pos
< istate
->cache_nr
) {
3329 struct cache_entry
*ce
= istate
->cache
[pos
];
3330 if (ce_namelen(ce
) == namelen
&&
3331 !memcmp(ce
->name
, name
, namelen
))
3332 return 0; /* Yup, this one exists unmerged */
3337 void *read_blob_data_from_index(const struct index_state
*istate
,
3338 const char *path
, unsigned long *size
)
3342 enum object_type type
;
3346 pos
= index_name_pos(istate
, path
, len
);
3349 * We might be in the middle of a merge, in which
3350 * case we would read stage #2 (ours).
3354 (pos
< 0 && i
< istate
->cache_nr
&&
3355 !strcmp(istate
->cache
[i
]->name
, path
));
3357 if (ce_stage(istate
->cache
[i
]) == 2)
3362 data
= read_object_file(&istate
->cache
[pos
]->oid
, &type
, &sz
);
3363 if (!data
|| type
!= OBJ_BLOB
) {
3372 void stat_validity_clear(struct stat_validity
*sv
)
3374 FREE_AND_NULL(sv
->sd
);
3377 int stat_validity_check(struct stat_validity
*sv
, const char *path
)
3381 if (stat(path
, &st
) < 0)
3382 return sv
->sd
== NULL
;
3385 return S_ISREG(st
.st_mode
) && !match_stat_data(sv
->sd
, &st
);
3388 void stat_validity_update(struct stat_validity
*sv
, int fd
)
3392 if (fstat(fd
, &st
) < 0 || !S_ISREG(st
.st_mode
))
3393 stat_validity_clear(sv
);
3396 sv
->sd
= xcalloc(1, sizeof(struct stat_data
));
3397 fill_stat_data(sv
->sd
, &st
);
3401 void move_index_extensions(struct index_state
*dst
, struct index_state
*src
)
3403 dst
->untracked
= src
->untracked
;
3404 src
->untracked
= NULL
;
3405 dst
->cache_tree
= src
->cache_tree
;
3406 src
->cache_tree
= NULL
;
3409 struct cache_entry
*dup_cache_entry(const struct cache_entry
*ce
,
3410 struct index_state
*istate
)
3412 unsigned int size
= ce_size(ce
);
3413 int mem_pool_allocated
;
3414 struct cache_entry
*new_entry
= make_empty_cache_entry(istate
, ce_namelen(ce
));
3415 mem_pool_allocated
= new_entry
->mem_pool_allocated
;
3417 memcpy(new_entry
, ce
, size
);
3418 new_entry
->mem_pool_allocated
= mem_pool_allocated
;
3422 void discard_cache_entry(struct cache_entry
*ce
)
3424 if (ce
&& should_validate_cache_entries())
3425 memset(ce
, 0xCD, cache_entry_size(ce
->ce_namelen
));
3427 if (ce
&& ce
->mem_pool_allocated
)
3433 int should_validate_cache_entries(void)
3435 static int validate_index_cache_entries
= -1;
3437 if (validate_index_cache_entries
< 0) {
3438 if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))
3439 validate_index_cache_entries
= 1;
3441 validate_index_cache_entries
= 0;
3444 return validate_index_cache_entries
;
3447 #define EOIE_SIZE (4 + GIT_SHA1_RAWSZ) /* <4-byte offset> + <20-byte hash> */
3448 #define EOIE_SIZE_WITH_HEADER (4 + 4 + EOIE_SIZE) /* <4-byte signature> + <4-byte length> + EOIE_SIZE */
3450 static size_t read_eoie_extension(const char *mmap
, size_t mmap_size
)
3453 * The end of index entries (EOIE) extension is guaranteed to be last
3454 * so that it can be found by scanning backwards from the EOF.
3461 const char *index
, *eoie
;
3463 size_t offset
, src_offset
;
3464 unsigned char hash
[GIT_MAX_RAWSZ
];
3467 /* ensure we have an index big enough to contain an EOIE extension */
3468 if (mmap_size
< sizeof(struct cache_header
) + EOIE_SIZE_WITH_HEADER
+ the_hash_algo
->rawsz
)
3471 /* validate the extension signature */
3472 index
= eoie
= mmap
+ mmap_size
- EOIE_SIZE_WITH_HEADER
- the_hash_algo
->rawsz
;
3473 if (CACHE_EXT(index
) != CACHE_EXT_ENDOFINDEXENTRIES
)
3475 index
+= sizeof(uint32_t);
3477 /* validate the extension size */
3478 extsize
= get_be32(index
);
3479 if (extsize
!= EOIE_SIZE
)
3481 index
+= sizeof(uint32_t);
3484 * Validate the offset we're going to look for the first extension
3485 * signature is after the index header and before the eoie extension.
3487 offset
= get_be32(index
);
3488 if (mmap
+ offset
< mmap
+ sizeof(struct cache_header
))
3490 if (mmap
+ offset
>= eoie
)
3492 index
+= sizeof(uint32_t);
3495 * The hash is computed over extension types and their sizes (but not
3496 * their contents). E.g. if we have "TREE" extension that is N-bytes
3497 * long, "REUC" extension that is M-bytes long, followed by "EOIE",
3498 * then the hash would be:
3500 * SHA-1("TREE" + <binary representation of N> +
3501 * "REUC" + <binary representation of M>)
3503 src_offset
= offset
;
3504 the_hash_algo
->init_fn(&c
);
3505 while (src_offset
< mmap_size
- the_hash_algo
->rawsz
- EOIE_SIZE_WITH_HEADER
) {
3506 /* After an array of active_nr index entries,
3507 * there can be arbitrary number of extended
3508 * sections, each of which is prefixed with
3509 * extension name (4-byte) and section length
3510 * in 4-byte network byte order.
3513 memcpy(&extsize
, mmap
+ src_offset
+ 4, 4);
3514 extsize
= ntohl(extsize
);
3516 /* verify the extension size isn't so large it will wrap around */
3517 if (src_offset
+ 8 + extsize
< src_offset
)
3520 the_hash_algo
->update_fn(&c
, mmap
+ src_offset
, 8);
3523 src_offset
+= extsize
;
3525 the_hash_algo
->final_fn(hash
, &c
);
3526 if (!hasheq(hash
, (const unsigned char *)index
))
3529 /* Validate that the extension offsets returned us back to the eoie extension. */
3530 if (src_offset
!= mmap_size
- the_hash_algo
->rawsz
- EOIE_SIZE_WITH_HEADER
)
3536 static void write_eoie_extension(struct strbuf
*sb
, git_hash_ctx
*eoie_context
, size_t offset
)
3539 unsigned char hash
[GIT_MAX_RAWSZ
];
3542 put_be32(&buffer
, offset
);
3543 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3546 the_hash_algo
->final_fn(hash
, eoie_context
);
3547 strbuf_add(sb
, hash
, the_hash_algo
->rawsz
);
3550 #define IEOT_VERSION (1)
3552 static struct index_entry_offset_table
*read_ieot_extension(const char *mmap
, size_t mmap_size
, size_t offset
)
3554 const char *index
= NULL
;
3555 uint32_t extsize
, ext_version
;
3556 struct index_entry_offset_table
*ieot
;
3559 /* find the IEOT extension */
3562 while (offset
<= mmap_size
- the_hash_algo
->rawsz
- 8) {
3563 extsize
= get_be32(mmap
+ offset
+ 4);
3564 if (CACHE_EXT((mmap
+ offset
)) == CACHE_EXT_INDEXENTRYOFFSETTABLE
) {
3565 index
= mmap
+ offset
+ 4 + 4;
3574 /* validate the version is IEOT_VERSION */
3575 ext_version
= get_be32(index
);
3576 if (ext_version
!= IEOT_VERSION
) {
3577 error("invalid IEOT version %d", ext_version
);
3580 index
+= sizeof(uint32_t);
3582 /* extension size - version bytes / bytes per entry */
3583 nr
= (extsize
- sizeof(uint32_t)) / (sizeof(uint32_t) + sizeof(uint32_t));
3585 error("invalid number of IEOT entries %d", nr
);
3588 ieot
= xmalloc(sizeof(struct index_entry_offset_table
)
3589 + (nr
* sizeof(struct index_entry_offset
)));
3591 for (i
= 0; i
< nr
; i
++) {
3592 ieot
->entries
[i
].offset
= get_be32(index
);
3593 index
+= sizeof(uint32_t);
3594 ieot
->entries
[i
].nr
= get_be32(index
);
3595 index
+= sizeof(uint32_t);
3601 static void write_ieot_extension(struct strbuf
*sb
, struct index_entry_offset_table
*ieot
)
3607 put_be32(&buffer
, IEOT_VERSION
);
3608 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3611 for (i
= 0; i
< ieot
->nr
; i
++) {
3614 put_be32(&buffer
, ieot
->entries
[i
].offset
);
3615 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3618 put_be32(&buffer
, ieot
->entries
[i
].nr
);
3619 strbuf_add(sb
, &buffer
, sizeof(uint32_t));