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 index_state
*istate
, 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(istate
, 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
= first
+ ((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
;
712 struct object_id oid
;
714 if (flags
& ADD_CACHE_RENORMALIZE
)
715 hash_flags
|= HASH_RENORMALIZE
;
717 if (!S_ISREG(st_mode
) && !S_ISLNK(st_mode
) && !S_ISDIR(st_mode
))
718 return error(_("%s: can only add regular files, symbolic links or git-directories"), path
);
720 namelen
= strlen(path
);
721 if (S_ISDIR(st_mode
)) {
722 if (resolve_gitlink_ref(path
, "HEAD", &oid
) < 0)
723 return error(_("'%s' does not have a commit checked out"), path
);
724 while (namelen
&& path
[namelen
-1] == '/')
727 ce
= make_empty_cache_entry(istate
, namelen
);
728 memcpy(ce
->name
, path
, namelen
);
729 ce
->ce_namelen
= namelen
;
731 fill_stat_cache_info(istate
, ce
, st
);
733 ce
->ce_flags
|= CE_INTENT_TO_ADD
;
736 if (trust_executable_bit
&& has_symlinks
) {
737 ce
->ce_mode
= create_ce_mode(st_mode
);
739 /* If there is an existing entry, pick the mode bits and type
740 * from it, otherwise assume unexecutable regular file.
742 struct cache_entry
*ent
;
743 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
745 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
746 ce
->ce_mode
= ce_mode_from_stat(ent
, st_mode
);
749 /* When core.ignorecase=true, determine if a directory of the same name but differing
750 * case already exists within the Git repository. If it does, ensure the directory
751 * case of the file being added to the repository matches (is folded into) the existing
752 * entry's directory case.
755 adjust_dirname_case(istate
, ce
->name
);
757 if (!(flags
& ADD_CACHE_RENORMALIZE
)) {
758 alias
= index_file_exists(istate
, ce
->name
,
759 ce_namelen(ce
), ignore_case
);
762 !ie_match_stat(istate
, alias
, st
, ce_option
)) {
763 /* Nothing changed, really */
764 if (!S_ISGITLINK(alias
->ce_mode
))
765 ce_mark_uptodate(alias
);
766 alias
->ce_flags
|= CE_ADDED
;
768 discard_cache_entry(ce
);
773 if (index_path(istate
, &ce
->oid
, path
, st
, hash_flags
)) {
774 discard_cache_entry(ce
);
775 return error(_("unable to index file '%s'"), path
);
778 set_object_name_for_intent_to_add_entry(ce
);
780 if (ignore_case
&& alias
&& different_name(ce
, alias
))
781 ce
= create_alias_ce(istate
, ce
, alias
);
782 ce
->ce_flags
|= CE_ADDED
;
784 /* It was suspected to be racily clean, but it turns out to be Ok */
787 oideq(&alias
->oid
, &ce
->oid
) &&
788 ce
->ce_mode
== alias
->ce_mode
);
791 discard_cache_entry(ce
);
792 else if (add_index_entry(istate
, ce
, add_option
)) {
793 discard_cache_entry(ce
);
794 return error(_("unable to add '%s' to index"), path
);
796 if (verbose
&& !was_same
)
797 printf("add '%s'\n", path
);
801 int add_file_to_index(struct index_state
*istate
, const char *path
, int flags
)
804 if (lstat(path
, &st
))
805 die_errno(_("unable to stat '%s'"), path
);
806 return add_to_index(istate
, path
, &st
, flags
);
809 struct cache_entry
*make_empty_cache_entry(struct index_state
*istate
, size_t len
)
811 return mem_pool__ce_calloc(find_mem_pool(istate
), len
);
814 struct cache_entry
*make_empty_transient_cache_entry(size_t len
)
816 return xcalloc(1, cache_entry_size(len
));
819 struct cache_entry
*make_cache_entry(struct index_state
*istate
,
821 const struct object_id
*oid
,
824 unsigned int refresh_options
)
826 struct cache_entry
*ce
, *ret
;
829 if (!verify_path(path
, mode
)) {
830 error(_("invalid path '%s'"), path
);
835 ce
= make_empty_cache_entry(istate
, len
);
837 oidcpy(&ce
->oid
, oid
);
838 memcpy(ce
->name
, path
, len
);
839 ce
->ce_flags
= create_ce_flags(stage
);
840 ce
->ce_namelen
= len
;
841 ce
->ce_mode
= create_ce_mode(mode
);
843 ret
= refresh_cache_entry(istate
, ce
, refresh_options
);
845 discard_cache_entry(ce
);
849 struct cache_entry
*make_transient_cache_entry(unsigned int mode
, const struct object_id
*oid
,
850 const char *path
, int stage
)
852 struct cache_entry
*ce
;
855 if (!verify_path(path
, mode
)) {
856 error(_("invalid path '%s'"), path
);
861 ce
= make_empty_transient_cache_entry(len
);
863 oidcpy(&ce
->oid
, oid
);
864 memcpy(ce
->name
, path
, len
);
865 ce
->ce_flags
= create_ce_flags(stage
);
866 ce
->ce_namelen
= len
;
867 ce
->ce_mode
= create_ce_mode(mode
);
873 * Chmod an index entry with either +x or -x.
875 * Returns -1 if the chmod for the particular cache entry failed (if it's
876 * not a regular file), -2 if an invalid flip argument is passed in, 0
879 int chmod_index_entry(struct index_state
*istate
, struct cache_entry
*ce
,
882 if (!S_ISREG(ce
->ce_mode
))
889 ce
->ce_mode
&= ~0111;
894 cache_tree_invalidate_path(istate
, ce
->name
);
895 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
896 mark_fsmonitor_invalid(istate
, ce
);
897 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
902 int ce_same_name(const struct cache_entry
*a
, const struct cache_entry
*b
)
904 int len
= ce_namelen(a
);
905 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
909 * We fundamentally don't like some paths: we don't want
910 * dot or dot-dot anywhere, and for obvious reasons don't
911 * want to recurse into ".git" either.
913 * Also, we don't want double slashes or slashes at the
914 * end that can make pathnames ambiguous.
916 static int verify_dotfile(const char *rest
, unsigned mode
)
919 * The first character was '.', but that
920 * has already been discarded, we now test
924 /* "." is not allowed */
925 if (*rest
== '\0' || is_dir_sep(*rest
))
930 * ".git" followed by NUL or slash is bad. Note that we match
931 * case-insensitively here, even if ignore_case is not set.
932 * This outlaws ".GIT" everywhere out of an abundance of caution,
933 * since there's really no good reason to allow it.
935 * Once we've seen ".git", we can also find ".gitmodules", etc (also
936 * case-insensitively).
940 if (rest
[1] != 'i' && rest
[1] != 'I')
942 if (rest
[2] != 't' && rest
[2] != 'T')
944 if (rest
[3] == '\0' || is_dir_sep(rest
[3]))
948 if (skip_iprefix(rest
, "modules", &rest
) &&
949 (*rest
== '\0' || is_dir_sep(*rest
)))
954 if (rest
[1] == '\0' || is_dir_sep(rest
[1]))
960 int verify_path(const char *path
, unsigned mode
)
964 if (has_dos_drive_prefix(path
))
974 if (is_hfs_dotgit(path
))
977 if (is_hfs_dotgitmodules(path
))
982 if (is_ntfs_dotgit(path
))
985 if (is_ntfs_dotgitmodules(path
))
991 if ((c
== '.' && !verify_dotfile(path
, mode
)) ||
992 is_dir_sep(c
) || c
== '\0')
1000 * Do we have another file that has the beginning components being a
1001 * proper superset of the name we're trying to add?
1003 static int has_file_name(struct index_state
*istate
,
1004 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
1007 int len
= ce_namelen(ce
);
1008 int stage
= ce_stage(ce
);
1009 const char *name
= ce
->name
;
1011 while (pos
< istate
->cache_nr
) {
1012 struct cache_entry
*p
= istate
->cache
[pos
++];
1014 if (len
>= ce_namelen(p
))
1016 if (memcmp(name
, p
->name
, len
))
1018 if (ce_stage(p
) != stage
)
1020 if (p
->name
[len
] != '/')
1022 if (p
->ce_flags
& CE_REMOVE
)
1027 remove_index_entry_at(istate
, --pos
);
1034 * Like strcmp(), but also return the offset of the first change.
1035 * If strings are equal, return the length.
1037 int strcmp_offset(const char *s1
, const char *s2
, size_t *first_change
)
1042 return strcmp(s1
, s2
);
1044 for (k
= 0; s1
[k
] == s2
[k
]; k
++)
1049 return (unsigned char)s1
[k
] - (unsigned char)s2
[k
];
1053 * Do we have another file with a pathname that is a proper
1054 * subset of the name we're trying to add?
1056 * That is, is there another file in the index with a path
1057 * that matches a sub-directory in the given entry?
1059 static int has_dir_name(struct index_state
*istate
,
1060 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
1063 int stage
= ce_stage(ce
);
1064 const char *name
= ce
->name
;
1065 const char *slash
= name
+ ce_namelen(ce
);
1070 * We are frequently called during an iteration on a sorted
1071 * list of pathnames and while building a new index. Therefore,
1072 * there is a high probability that this entry will eventually
1073 * be appended to the index, rather than inserted in the middle.
1074 * If we can confirm that, we can avoid binary searches on the
1075 * components of the pathname.
1077 * Compare the entry's full path with the last path in the index.
1079 if (istate
->cache_nr
> 0) {
1080 cmp_last
= strcmp_offset(name
,
1081 istate
->cache
[istate
->cache_nr
- 1]->name
,
1084 if (len_eq_last
== 0) {
1086 * The entry sorts AFTER the last one in the
1087 * index and their paths have no common prefix,
1088 * so there cannot be a F/D conflict.
1093 * The entry sorts AFTER the last one in the
1094 * index, but has a common prefix. Fall through
1095 * to the loop below to disect the entry's path
1096 * and see where the difference is.
1099 } else if (cmp_last
== 0) {
1101 * The entry exactly matches the last one in the
1102 * index, but because of multiple stage and CE_REMOVE
1103 * items, we fall through and let the regular search
1113 if (*--slash
== '/')
1115 if (slash
<= ce
->name
)
1122 * (len + 1) is a directory boundary (including
1123 * the trailing slash). And since the loop is
1124 * decrementing "slash", the first iteration is
1125 * the longest directory prefix; subsequent
1126 * iterations consider parent directories.
1129 if (len
+ 1 <= len_eq_last
) {
1131 * The directory prefix (including the trailing
1132 * slash) also appears as a prefix in the last
1133 * entry, so the remainder cannot collide (because
1134 * strcmp said the whole path was greater).
1139 * LT: last: xxx/file_A
1145 if (len
> len_eq_last
) {
1147 * This part of the directory prefix (excluding
1148 * the trailing slash) is longer than the known
1149 * equal portions, so this sub-directory cannot
1150 * collide with a file.
1158 if (istate
->cache_nr
> 0 &&
1159 ce_namelen(istate
->cache
[istate
->cache_nr
- 1]) > len
) {
1161 * The directory prefix lines up with part of
1162 * a longer file or directory name, but sorts
1163 * after it, so this sub-directory cannot
1164 * collide with a file.
1166 * last: xxx/yy-file (because '-' sorts before '/')
1173 * This is a possible collision. Fall through and
1174 * let the regular search code handle it.
1181 pos
= index_name_stage_pos(istate
, name
, len
, stage
);
1184 * Found one, but not so fast. This could
1185 * be a marker that says "I was here, but
1186 * I am being removed". Such an entry is
1187 * not a part of the resulting tree, and
1188 * it is Ok to have a directory at the same
1191 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
1195 remove_index_entry_at(istate
, pos
);
1203 * Trivial optimization: if we find an entry that
1204 * already matches the sub-directory, then we know
1205 * we're ok, and we can exit.
1207 while (pos
< istate
->cache_nr
) {
1208 struct cache_entry
*p
= istate
->cache
[pos
];
1209 if ((ce_namelen(p
) <= len
) ||
1210 (p
->name
[len
] != '/') ||
1211 memcmp(p
->name
, name
, len
))
1212 break; /* not our subdirectory */
1213 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
1215 * p is at the same stage as our entry, and
1216 * is a subdirectory of what we are looking
1217 * at, so we cannot have conflicts at our
1218 * level or anything shorter.
1227 /* We may be in a situation where we already have path/file and path
1228 * is being added, or we already have path and path/file is being
1229 * added. Either one would result in a nonsense tree that has path
1230 * twice when git-write-tree tries to write it out. Prevent it.
1232 * If ok-to-replace is specified, we remove the conflicting entries
1233 * from the cache so the caller should recompute the insert position.
1234 * When this happens, we return non-zero.
1236 static int check_file_directory_conflict(struct index_state
*istate
,
1237 const struct cache_entry
*ce
,
1238 int pos
, int ok_to_replace
)
1243 * When ce is an "I am going away" entry, we allow it to be added
1245 if (ce
->ce_flags
& CE_REMOVE
)
1249 * We check if the path is a sub-path of a subsequent pathname
1250 * first, since removing those will not change the position
1253 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
1256 * Then check if the path might have a clashing sub-directory
1259 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
1262 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1265 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
1266 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
1267 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
1268 int new_only
= option
& ADD_CACHE_NEW_ONLY
;
1270 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1271 cache_tree_invalidate_path(istate
, ce
->name
);
1274 * If this entry's path sorts after the last entry in the index,
1275 * we can avoid searching for it.
1277 if (istate
->cache_nr
> 0 &&
1278 strcmp(ce
->name
, istate
->cache
[istate
->cache_nr
- 1]->name
) > 0)
1279 pos
= -istate
->cache_nr
- 1;
1281 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1283 /* existing match? Just replace it. */
1286 replace_index_entry(istate
, pos
, ce
);
1291 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1292 untracked_cache_add_to_index(istate
, ce
->name
);
1295 * Inserting a merged entry ("stage 0") into the index
1296 * will always replace all non-merged entries..
1298 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
1299 while (ce_same_name(istate
->cache
[pos
], ce
)) {
1301 if (!remove_index_entry_at(istate
, pos
))
1308 if (!verify_path(ce
->name
, ce
->ce_mode
))
1309 return error(_("invalid path '%s'"), ce
->name
);
1311 if (!skip_df_check
&&
1312 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
1314 return error(_("'%s' appears as both a file and as a directory"),
1316 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1322 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1326 if (option
& ADD_CACHE_JUST_APPEND
)
1327 pos
= istate
->cache_nr
;
1330 ret
= add_index_entry_with_check(istate
, ce
, option
);
1336 /* Make sure the array is big enough .. */
1337 ALLOC_GROW(istate
->cache
, istate
->cache_nr
+ 1, istate
->cache_alloc
);
1341 if (istate
->cache_nr
> pos
+ 1)
1342 MOVE_ARRAY(istate
->cache
+ pos
+ 1, istate
->cache
+ pos
,
1343 istate
->cache_nr
- pos
- 1);
1344 set_index_entry(istate
, pos
, ce
);
1345 istate
->cache_changed
|= CE_ENTRY_ADDED
;
1350 * "refresh" does not calculate a new sha1 file or bring the
1351 * cache up-to-date for mode/content changes. But what it
1352 * _does_ do is to "re-match" the stat information of a file
1353 * with the cache, so that you can refresh the cache for a
1354 * file that hasn't been changed but where the stat entry is
1357 * For example, you'd want to do this after doing a "git-read-tree",
1358 * to link up the stat cache details with the proper files.
1360 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
1361 struct cache_entry
*ce
,
1362 unsigned int options
, int *err
,
1366 struct cache_entry
*updated
;
1368 int refresh
= options
& CE_MATCH_REFRESH
;
1369 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
1370 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
1371 int ignore_missing
= options
& CE_MATCH_IGNORE_MISSING
;
1372 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
1374 if (!refresh
|| ce_uptodate(ce
))
1377 if (!ignore_fsmonitor
)
1378 refresh_fsmonitor(istate
);
1380 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1381 * that the change to the work tree does not matter and told
1384 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
)) {
1385 ce_mark_uptodate(ce
);
1388 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
)) {
1389 ce_mark_uptodate(ce
);
1392 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
)) {
1393 ce_mark_uptodate(ce
);
1397 if (has_symlink_leading_path(ce
->name
, ce_namelen(ce
))) {
1405 if (lstat(ce
->name
, &st
) < 0) {
1406 if (ignore_missing
&& errno
== ENOENT
)
1413 changed
= ie_match_stat(istate
, ce
, &st
, options
);
1415 *changed_ret
= changed
;
1418 * The path is unchanged. If we were told to ignore
1419 * valid bit, then we did the actual stat check and
1420 * found that the entry is unmodified. If the entry
1421 * is not marked VALID, this is the place to mark it
1422 * valid again, under "assume unchanged" mode.
1424 if (ignore_valid
&& assume_unchanged
&&
1425 !(ce
->ce_flags
& CE_VALID
))
1426 ; /* mark this one VALID again */
1429 * We do not mark the index itself "modified"
1430 * because CE_UPTODATE flag is in-core only;
1431 * we are not going to write this change out.
1433 if (!S_ISGITLINK(ce
->ce_mode
)) {
1434 ce_mark_uptodate(ce
);
1435 mark_fsmonitor_valid(istate
, ce
);
1441 if (ie_modified(istate
, ce
, &st
, options
)) {
1447 updated
= make_empty_cache_entry(istate
, ce_namelen(ce
));
1448 copy_cache_entry(updated
, ce
);
1449 memcpy(updated
->name
, ce
->name
, ce
->ce_namelen
+ 1);
1450 fill_stat_cache_info(istate
, updated
, &st
);
1452 * If ignore_valid is not set, we should leave CE_VALID bit
1453 * alone. Otherwise, paths marked with --no-assume-unchanged
1454 * (i.e. things to be edited) will reacquire CE_VALID bit
1455 * automatically, which is not really what we want.
1457 if (!ignore_valid
&& assume_unchanged
&&
1458 !(ce
->ce_flags
& CE_VALID
))
1459 updated
->ce_flags
&= ~CE_VALID
;
1461 /* istate->cache_changed is updated in the caller */
1465 static void show_file(const char * fmt
, const char * name
, int in_porcelain
,
1466 int * first
, const char *header_msg
)
1468 if (in_porcelain
&& *first
&& header_msg
) {
1469 printf("%s\n", header_msg
);
1475 int refresh_index(struct index_state
*istate
, unsigned int flags
,
1476 const struct pathspec
*pathspec
,
1477 char *seen
, const char *header_msg
)
1481 int really
= (flags
& REFRESH_REALLY
) != 0;
1482 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
1483 int quiet
= (flags
& REFRESH_QUIET
) != 0;
1484 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
1485 int ignore_submodules
= (flags
& REFRESH_IGNORE_SUBMODULES
) != 0;
1487 int in_porcelain
= (flags
& REFRESH_IN_PORCELAIN
);
1488 unsigned int options
= (CE_MATCH_REFRESH
|
1489 (really
? CE_MATCH_IGNORE_VALID
: 0) |
1490 (not_new
? CE_MATCH_IGNORE_MISSING
: 0));
1491 const char *modified_fmt
;
1492 const char *deleted_fmt
;
1493 const char *typechange_fmt
;
1494 const char *added_fmt
;
1495 const char *unmerged_fmt
;
1496 struct progress
*progress
= NULL
;
1498 if (flags
& REFRESH_PROGRESS
&& isatty(2))
1499 progress
= start_delayed_progress(_("Refresh index"),
1502 trace_performance_enter();
1503 modified_fmt
= in_porcelain
? "M\t%s\n" : "%s: needs update\n";
1504 deleted_fmt
= in_porcelain
? "D\t%s\n" : "%s: needs update\n";
1505 typechange_fmt
= in_porcelain
? "T\t%s\n" : "%s: needs update\n";
1506 added_fmt
= in_porcelain
? "A\t%s\n" : "%s: needs update\n";
1507 unmerged_fmt
= in_porcelain
? "U\t%s\n" : "%s: needs merge\n";
1509 * Use the multi-threaded preload_index() to refresh most of the
1510 * cache entries quickly then in the single threaded loop below,
1511 * we only have to do the special cases that are left.
1513 preload_index(istate
, pathspec
, 0);
1514 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1515 struct cache_entry
*ce
, *new_entry
;
1516 int cache_errno
= 0;
1520 ce
= istate
->cache
[i
];
1521 if (ignore_submodules
&& S_ISGITLINK(ce
->ce_mode
))
1524 if (pathspec
&& !ce_path_match(istate
, ce
, pathspec
, seen
))
1528 while ((i
< istate
->cache_nr
) &&
1529 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
1535 show_file(unmerged_fmt
, ce
->name
, in_porcelain
,
1536 &first
, header_msg
);
1544 new_entry
= refresh_cache_ent(istate
, ce
, options
, &cache_errno
, &changed
);
1545 if (new_entry
== ce
)
1548 display_progress(progress
, i
);
1552 if (really
&& cache_errno
== EINVAL
) {
1553 /* If we are doing --really-refresh that
1554 * means the index is not valid anymore.
1556 ce
->ce_flags
&= ~CE_VALID
;
1557 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
1558 mark_fsmonitor_invalid(istate
, ce
);
1559 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
1564 if (cache_errno
== ENOENT
)
1566 else if (ce_intent_to_add(ce
))
1567 fmt
= added_fmt
; /* must be before other checks */
1568 else if (changed
& TYPE_CHANGED
)
1569 fmt
= typechange_fmt
;
1573 ce
->name
, in_porcelain
, &first
, header_msg
);
1578 replace_index_entry(istate
, i
, new_entry
);
1581 display_progress(progress
, istate
->cache_nr
);
1582 stop_progress(&progress
);
1584 trace_performance_leave("refresh index");
1588 struct cache_entry
*refresh_cache_entry(struct index_state
*istate
,
1589 struct cache_entry
*ce
,
1590 unsigned int options
)
1592 return refresh_cache_ent(istate
, ce
, options
, NULL
, NULL
);
1596 /*****************************************************************
1598 *****************************************************************/
1600 #define INDEX_FORMAT_DEFAULT 3
1602 static unsigned int get_index_format_default(struct repository
*r
)
1604 char *envversion
= getenv("GIT_INDEX_VERSION");
1606 unsigned int version
= INDEX_FORMAT_DEFAULT
;
1609 prepare_repo_settings(r
);
1611 if (r
->settings
.index_version
>= 0)
1612 version
= r
->settings
.index_version
;
1613 if (version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1614 warning(_("index.version set, but the value is invalid.\n"
1615 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1616 return INDEX_FORMAT_DEFAULT
;
1621 version
= strtoul(envversion
, &endp
, 10);
1623 version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1624 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1625 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1626 version
= INDEX_FORMAT_DEFAULT
;
1632 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1633 * Again - this is just a (very strong in practice) heuristic that
1634 * the inode hasn't changed.
1636 * We save the fields in big-endian order to allow using the
1637 * index file over NFS transparently.
1639 struct ondisk_cache_entry
{
1640 struct cache_time ctime
;
1641 struct cache_time mtime
;
1649 * unsigned char hash[hashsz];
1651 * if (flags & CE_EXTENDED)
1654 unsigned char data
[GIT_MAX_RAWSZ
+ 2 * sizeof(uint16_t)];
1655 char name
[FLEX_ARRAY
];
1658 /* These are only used for v3 or lower */
1659 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1660 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,data) + (len) + 8) & ~7)
1661 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1662 #define ondisk_data_size(flags, len) (the_hash_algo->rawsz + \
1663 ((flags & CE_EXTENDED) ? 2 : 1) * sizeof(uint16_t) + len)
1664 #define ondisk_data_size_max(len) (ondisk_data_size(CE_EXTENDED, len))
1665 #define ondisk_ce_size(ce) (ondisk_cache_entry_size(ondisk_data_size((ce)->ce_flags, ce_namelen(ce))))
1667 /* Allow fsck to force verification of the index checksum. */
1668 int verify_index_checksum
;
1670 /* Allow fsck to force verification of the cache entry order. */
1671 int verify_ce_order
;
1673 static int verify_hdr(const struct cache_header
*hdr
, unsigned long size
)
1676 unsigned char hash
[GIT_MAX_RAWSZ
];
1679 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
1680 return error(_("bad signature 0x%08x"), hdr
->hdr_signature
);
1681 hdr_version
= ntohl(hdr
->hdr_version
);
1682 if (hdr_version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< hdr_version
)
1683 return error(_("bad index version %d"), hdr_version
);
1685 if (!verify_index_checksum
)
1688 the_hash_algo
->init_fn(&c
);
1689 the_hash_algo
->update_fn(&c
, hdr
, size
- the_hash_algo
->rawsz
);
1690 the_hash_algo
->final_fn(hash
, &c
);
1691 if (!hasheq(hash
, (unsigned char *)hdr
+ size
- the_hash_algo
->rawsz
))
1692 return error(_("bad index file sha1 signature"));
1696 static int read_index_extension(struct index_state
*istate
,
1697 const char *ext
, const char *data
, unsigned long sz
)
1699 switch (CACHE_EXT(ext
)) {
1700 case CACHE_EXT_TREE
:
1701 istate
->cache_tree
= cache_tree_read(data
, sz
);
1703 case CACHE_EXT_RESOLVE_UNDO
:
1704 istate
->resolve_undo
= resolve_undo_read(data
, sz
);
1706 case CACHE_EXT_LINK
:
1707 if (read_link_extension(istate
, data
, sz
))
1710 case CACHE_EXT_UNTRACKED
:
1711 istate
->untracked
= read_untracked_extension(data
, sz
);
1713 case CACHE_EXT_FSMONITOR
:
1714 read_fsmonitor_extension(istate
, data
, sz
);
1716 case CACHE_EXT_ENDOFINDEXENTRIES
:
1717 case CACHE_EXT_INDEXENTRYOFFSETTABLE
:
1718 /* already handled in do_read_index() */
1721 if (*ext
< 'A' || 'Z' < *ext
)
1722 return error(_("index uses %.4s extension, which we do not understand"),
1724 fprintf_ln(stderr
, _("ignoring %.4s extension"), ext
);
1730 static struct cache_entry
*create_from_disk(struct mem_pool
*ce_mem_pool
,
1731 unsigned int version
,
1732 struct ondisk_cache_entry
*ondisk
,
1733 unsigned long *ent_size
,
1734 const struct cache_entry
*previous_ce
)
1736 struct cache_entry
*ce
;
1739 const unsigned hashsz
= the_hash_algo
->rawsz
;
1740 const uint16_t *flagsp
= (const uint16_t *)(ondisk
->data
+ hashsz
);
1742 size_t copy_len
= 0;
1744 * Adjacent cache entries tend to share the leading paths, so it makes
1745 * sense to only store the differences in later entries. In the v4
1746 * on-disk format of the index, each on-disk cache entry stores the
1747 * number of bytes to be stripped from the end of the previous name,
1748 * and the bytes to append to the result, to come up with its name.
1750 int expand_name_field
= version
== 4;
1752 /* On-disk flags are just 16 bits */
1753 flags
= get_be16(flagsp
);
1754 len
= flags
& CE_NAMEMASK
;
1756 if (flags
& CE_EXTENDED
) {
1758 extended_flags
= get_be16(flagsp
+ 1) << 16;
1759 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1760 if (extended_flags
& ~CE_EXTENDED_FLAGS
)
1761 die(_("unknown index entry format 0x%08x"), extended_flags
);
1762 flags
|= extended_flags
;
1763 name
= (const char *)(flagsp
+ 2);
1766 name
= (const char *)(flagsp
+ 1);
1768 if (expand_name_field
) {
1769 const unsigned char *cp
= (const unsigned char *)name
;
1770 size_t strip_len
, previous_len
;
1772 /* If we're at the begining of a block, ignore the previous name */
1773 strip_len
= decode_varint(&cp
);
1775 previous_len
= previous_ce
->ce_namelen
;
1776 if (previous_len
< strip_len
)
1777 die(_("malformed name field in the index, near path '%s'"),
1779 copy_len
= previous_len
- strip_len
;
1781 name
= (const char *)cp
;
1784 if (len
== CE_NAMEMASK
) {
1786 if (expand_name_field
)
1790 ce
= mem_pool__ce_alloc(ce_mem_pool
, len
);
1792 ce
->ce_stat_data
.sd_ctime
.sec
= get_be32(&ondisk
->ctime
.sec
);
1793 ce
->ce_stat_data
.sd_mtime
.sec
= get_be32(&ondisk
->mtime
.sec
);
1794 ce
->ce_stat_data
.sd_ctime
.nsec
= get_be32(&ondisk
->ctime
.nsec
);
1795 ce
->ce_stat_data
.sd_mtime
.nsec
= get_be32(&ondisk
->mtime
.nsec
);
1796 ce
->ce_stat_data
.sd_dev
= get_be32(&ondisk
->dev
);
1797 ce
->ce_stat_data
.sd_ino
= get_be32(&ondisk
->ino
);
1798 ce
->ce_mode
= get_be32(&ondisk
->mode
);
1799 ce
->ce_stat_data
.sd_uid
= get_be32(&ondisk
->uid
);
1800 ce
->ce_stat_data
.sd_gid
= get_be32(&ondisk
->gid
);
1801 ce
->ce_stat_data
.sd_size
= get_be32(&ondisk
->size
);
1802 ce
->ce_flags
= flags
& ~CE_NAMEMASK
;
1803 ce
->ce_namelen
= len
;
1805 hashcpy(ce
->oid
.hash
, ondisk
->data
);
1806 memcpy(ce
->name
, name
, len
);
1807 ce
->name
[len
] = '\0';
1809 if (expand_name_field
) {
1811 memcpy(ce
->name
, previous_ce
->name
, copy_len
);
1812 memcpy(ce
->name
+ copy_len
, name
, len
+ 1 - copy_len
);
1813 *ent_size
= (name
- ((char *)ondisk
)) + len
+ 1 - copy_len
;
1815 memcpy(ce
->name
, name
, len
+ 1);
1816 *ent_size
= ondisk_ce_size(ce
);
1821 static void check_ce_order(struct index_state
*istate
)
1825 if (!verify_ce_order
)
1828 for (i
= 1; i
< istate
->cache_nr
; i
++) {
1829 struct cache_entry
*ce
= istate
->cache
[i
- 1];
1830 struct cache_entry
*next_ce
= istate
->cache
[i
];
1831 int name_compare
= strcmp(ce
->name
, next_ce
->name
);
1833 if (0 < name_compare
)
1834 die(_("unordered stage entries in index"));
1835 if (!name_compare
) {
1837 die(_("multiple stage entries for merged file '%s'"),
1839 if (ce_stage(ce
) > ce_stage(next_ce
))
1840 die(_("unordered stage entries for '%s'"),
1846 static void tweak_untracked_cache(struct index_state
*istate
)
1848 switch (git_config_get_untracked_cache()) {
1849 case -1: /* keep: do nothing */
1852 remove_untracked_cache(istate
);
1855 add_untracked_cache(istate
);
1857 default: /* unknown value: do nothing */
1862 static void tweak_split_index(struct index_state
*istate
)
1864 switch (git_config_get_split_index()) {
1865 case -1: /* unset: do nothing */
1868 remove_split_index(istate
);
1871 add_split_index(istate
);
1873 default: /* unknown value: do nothing */
1878 static void post_read_index_from(struct index_state
*istate
)
1880 check_ce_order(istate
);
1881 tweak_untracked_cache(istate
);
1882 tweak_split_index(istate
);
1883 tweak_fsmonitor(istate
);
1886 static size_t estimate_cache_size_from_compressed(unsigned int entries
)
1888 return entries
* (sizeof(struct cache_entry
) + CACHE_ENTRY_PATH_LENGTH
);
1891 static size_t estimate_cache_size(size_t ondisk_size
, unsigned int entries
)
1893 long per_entry
= sizeof(struct cache_entry
) - sizeof(struct ondisk_cache_entry
);
1896 * Account for potential alignment differences.
1898 per_entry
+= align_padding_size(sizeof(struct cache_entry
), -sizeof(struct ondisk_cache_entry
));
1899 return ondisk_size
+ entries
* per_entry
;
1902 struct index_entry_offset
1904 /* starting byte offset into index file, count of index entries in this block */
1908 struct index_entry_offset_table
1911 struct index_entry_offset entries
[FLEX_ARRAY
];
1914 static struct index_entry_offset_table
*read_ieot_extension(const char *mmap
, size_t mmap_size
, size_t offset
);
1915 static void write_ieot_extension(struct strbuf
*sb
, struct index_entry_offset_table
*ieot
);
1917 static size_t read_eoie_extension(const char *mmap
, size_t mmap_size
);
1918 static void write_eoie_extension(struct strbuf
*sb
, git_hash_ctx
*eoie_context
, size_t offset
);
1920 struct load_index_extensions
1923 struct index_state
*istate
;
1926 unsigned long src_offset
;
1929 static void *load_index_extensions(void *_data
)
1931 struct load_index_extensions
*p
= _data
;
1932 unsigned long src_offset
= p
->src_offset
;
1934 while (src_offset
<= p
->mmap_size
- the_hash_algo
->rawsz
- 8) {
1935 /* After an array of active_nr index entries,
1936 * there can be arbitrary number of extended
1937 * sections, each of which is prefixed with
1938 * extension name (4-byte) and section length
1939 * in 4-byte network byte order.
1941 uint32_t extsize
= get_be32(p
->mmap
+ src_offset
+ 4);
1942 if (read_index_extension(p
->istate
,
1943 p
->mmap
+ src_offset
,
1944 p
->mmap
+ src_offset
+ 8,
1946 munmap((void *)p
->mmap
, p
->mmap_size
);
1947 die(_("index file corrupt"));
1950 src_offset
+= extsize
;
1957 * A helper function that will load the specified range of cache entries
1958 * from the memory mapped file and add them to the given index.
1960 static unsigned long load_cache_entry_block(struct index_state
*istate
,
1961 struct mem_pool
*ce_mem_pool
, int offset
, int nr
, const char *mmap
,
1962 unsigned long start_offset
, const struct cache_entry
*previous_ce
)
1965 unsigned long src_offset
= start_offset
;
1967 for (i
= offset
; i
< offset
+ nr
; i
++) {
1968 struct ondisk_cache_entry
*disk_ce
;
1969 struct cache_entry
*ce
;
1970 unsigned long consumed
;
1972 disk_ce
= (struct ondisk_cache_entry
*)(mmap
+ src_offset
);
1973 ce
= create_from_disk(ce_mem_pool
, istate
->version
, disk_ce
, &consumed
, previous_ce
);
1974 set_index_entry(istate
, i
, ce
);
1976 src_offset
+= consumed
;
1979 return src_offset
- start_offset
;
1982 static unsigned long load_all_cache_entries(struct index_state
*istate
,
1983 const char *mmap
, size_t mmap_size
, unsigned long src_offset
)
1985 unsigned long consumed
;
1987 if (istate
->version
== 4) {
1988 mem_pool_init(&istate
->ce_mem_pool
,
1989 estimate_cache_size_from_compressed(istate
->cache_nr
));
1991 mem_pool_init(&istate
->ce_mem_pool
,
1992 estimate_cache_size(mmap_size
, istate
->cache_nr
));
1995 consumed
= load_cache_entry_block(istate
, istate
->ce_mem_pool
,
1996 0, istate
->cache_nr
, mmap
, src_offset
, NULL
);
2001 * Mostly randomly chosen maximum thread counts: we
2002 * cap the parallelism to online_cpus() threads, and we want
2003 * to have at least 10000 cache entries per thread for it to
2004 * be worth starting a thread.
2007 #define THREAD_COST (10000)
2009 struct load_cache_entries_thread_data
2012 struct index_state
*istate
;
2013 struct mem_pool
*ce_mem_pool
;
2016 struct index_entry_offset_table
*ieot
;
2017 int ieot_start
; /* starting index into the ieot array */
2018 int ieot_blocks
; /* count of ieot entries to process */
2019 unsigned long consumed
; /* return # of bytes in index file processed */
2023 * A thread proc to run the load_cache_entries() computation
2024 * across multiple background threads.
2026 static void *load_cache_entries_thread(void *_data
)
2028 struct load_cache_entries_thread_data
*p
= _data
;
2031 /* iterate across all ieot blocks assigned to this thread */
2032 for (i
= p
->ieot_start
; i
< p
->ieot_start
+ p
->ieot_blocks
; i
++) {
2033 p
->consumed
+= load_cache_entry_block(p
->istate
, p
->ce_mem_pool
,
2034 p
->offset
, p
->ieot
->entries
[i
].nr
, p
->mmap
, p
->ieot
->entries
[i
].offset
, NULL
);
2035 p
->offset
+= p
->ieot
->entries
[i
].nr
;
2040 static unsigned long load_cache_entries_threaded(struct index_state
*istate
, const char *mmap
, size_t mmap_size
,
2041 int nr_threads
, struct index_entry_offset_table
*ieot
)
2043 int i
, offset
, ieot_blocks
, ieot_start
, err
;
2044 struct load_cache_entries_thread_data
*data
;
2045 unsigned long consumed
= 0;
2047 /* a little sanity checking */
2048 if (istate
->name_hash_initialized
)
2049 BUG("the name hash isn't thread safe");
2051 mem_pool_init(&istate
->ce_mem_pool
, 0);
2053 /* ensure we have no more threads than we have blocks to process */
2054 if (nr_threads
> ieot
->nr
)
2055 nr_threads
= ieot
->nr
;
2056 data
= xcalloc(nr_threads
, sizeof(*data
));
2058 offset
= ieot_start
= 0;
2059 ieot_blocks
= DIV_ROUND_UP(ieot
->nr
, nr_threads
);
2060 for (i
= 0; i
< nr_threads
; i
++) {
2061 struct load_cache_entries_thread_data
*p
= &data
[i
];
2064 if (ieot_start
+ ieot_blocks
> ieot
->nr
)
2065 ieot_blocks
= ieot
->nr
- ieot_start
;
2071 p
->ieot_start
= ieot_start
;
2072 p
->ieot_blocks
= ieot_blocks
;
2074 /* create a mem_pool for each thread */
2076 for (j
= p
->ieot_start
; j
< p
->ieot_start
+ p
->ieot_blocks
; j
++)
2077 nr
+= p
->ieot
->entries
[j
].nr
;
2078 if (istate
->version
== 4) {
2079 mem_pool_init(&p
->ce_mem_pool
,
2080 estimate_cache_size_from_compressed(nr
));
2082 mem_pool_init(&p
->ce_mem_pool
,
2083 estimate_cache_size(mmap_size
, nr
));
2086 err
= pthread_create(&p
->pthread
, NULL
, load_cache_entries_thread
, p
);
2088 die(_("unable to create load_cache_entries thread: %s"), strerror(err
));
2090 /* increment by the number of cache entries in the ieot block being processed */
2091 for (j
= 0; j
< ieot_blocks
; j
++)
2092 offset
+= ieot
->entries
[ieot_start
+ j
].nr
;
2093 ieot_start
+= ieot_blocks
;
2096 for (i
= 0; i
< nr_threads
; i
++) {
2097 struct load_cache_entries_thread_data
*p
= &data
[i
];
2099 err
= pthread_join(p
->pthread
, NULL
);
2101 die(_("unable to join load_cache_entries thread: %s"), strerror(err
));
2102 mem_pool_combine(istate
->ce_mem_pool
, p
->ce_mem_pool
);
2103 consumed
+= p
->consumed
;
2111 /* remember to discard_cache() before reading a different cache! */
2112 int do_read_index(struct index_state
*istate
, const char *path
, int must_exist
)
2116 unsigned long src_offset
;
2117 const struct cache_header
*hdr
;
2120 struct load_index_extensions p
;
2121 size_t extension_offset
= 0;
2122 int nr_threads
, cpus
;
2123 struct index_entry_offset_table
*ieot
= NULL
;
2125 if (istate
->initialized
)
2126 return istate
->cache_nr
;
2128 istate
->timestamp
.sec
= 0;
2129 istate
->timestamp
.nsec
= 0;
2130 fd
= open(path
, O_RDONLY
);
2132 if (!must_exist
&& errno
== ENOENT
)
2134 die_errno(_("%s: index file open failed"), path
);
2138 die_errno(_("%s: cannot stat the open index"), path
);
2140 mmap_size
= xsize_t(st
.st_size
);
2141 if (mmap_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2142 die(_("%s: index file smaller than expected"), path
);
2144 mmap
= xmmap(NULL
, mmap_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2145 if (mmap
== MAP_FAILED
)
2146 die_errno(_("%s: unable to map index file"), path
);
2149 hdr
= (const struct cache_header
*)mmap
;
2150 if (verify_hdr(hdr
, mmap_size
) < 0)
2153 hashcpy(istate
->oid
.hash
, (const unsigned char *)hdr
+ mmap_size
- the_hash_algo
->rawsz
);
2154 istate
->version
= ntohl(hdr
->hdr_version
);
2155 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
2156 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
2157 istate
->cache
= xcalloc(istate
->cache_alloc
, sizeof(*istate
->cache
));
2158 istate
->initialized
= 1;
2162 p
.mmap_size
= mmap_size
;
2164 src_offset
= sizeof(*hdr
);
2166 if (git_config_get_index_threads(&nr_threads
))
2169 /* TODO: does creating more threads than cores help? */
2171 nr_threads
= istate
->cache_nr
/ THREAD_COST
;
2172 cpus
= online_cpus();
2173 if (nr_threads
> cpus
)
2180 if (nr_threads
> 1) {
2181 extension_offset
= read_eoie_extension(mmap
, mmap_size
);
2182 if (extension_offset
) {
2185 p
.src_offset
= extension_offset
;
2186 err
= pthread_create(&p
.pthread
, NULL
, load_index_extensions
, &p
);
2188 die(_("unable to create load_index_extensions thread: %s"), strerror(err
));
2195 * Locate and read the index entry offset table so that we can use it
2196 * to multi-thread the reading of the cache entries.
2198 if (extension_offset
&& nr_threads
> 1)
2199 ieot
= read_ieot_extension(mmap
, mmap_size
, extension_offset
);
2202 src_offset
+= load_cache_entries_threaded(istate
, mmap
, mmap_size
, nr_threads
, ieot
);
2205 src_offset
+= load_all_cache_entries(istate
, mmap
, mmap_size
, src_offset
);
2208 istate
->timestamp
.sec
= st
.st_mtime
;
2209 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
2211 /* if we created a thread, join it otherwise load the extensions on the primary thread */
2212 if (extension_offset
) {
2213 int ret
= pthread_join(p
.pthread
, NULL
);
2215 die(_("unable to join load_index_extensions thread: %s"), strerror(ret
));
2217 p
.src_offset
= src_offset
;
2218 load_index_extensions(&p
);
2220 munmap((void *)mmap
, mmap_size
);
2223 * TODO trace2: replace "the_repository" with the actual repo instance
2224 * that is associated with the given "istate".
2226 trace2_data_intmax("index", the_repository
, "read/version",
2228 trace2_data_intmax("index", the_repository
, "read/cache_nr",
2231 return istate
->cache_nr
;
2234 munmap((void *)mmap
, mmap_size
);
2235 die(_("index file corrupt"));
2239 * Signal that the shared index is used by updating its mtime.
2241 * This way, shared index can be removed if they have not been used
2244 static void freshen_shared_index(const char *shared_index
, int warn
)
2246 if (!check_and_freshen_file(shared_index
, 1) && warn
)
2247 warning(_("could not freshen shared index '%s'"), shared_index
);
2250 int read_index_from(struct index_state
*istate
, const char *path
,
2253 struct split_index
*split_index
;
2258 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2259 if (istate
->initialized
)
2260 return istate
->cache_nr
;
2263 * TODO trace2: replace "the_repository" with the actual repo instance
2264 * that is associated with the given "istate".
2266 trace2_region_enter_printf("index", "do_read_index", the_repository
,
2268 trace_performance_enter();
2269 ret
= do_read_index(istate
, path
, 0);
2270 trace_performance_leave("read cache %s", path
);
2271 trace2_region_leave_printf("index", "do_read_index", the_repository
,
2274 split_index
= istate
->split_index
;
2275 if (!split_index
|| is_null_oid(&split_index
->base_oid
)) {
2276 post_read_index_from(istate
);
2280 trace_performance_enter();
2281 if (split_index
->base
)
2282 discard_index(split_index
->base
);
2284 split_index
->base
= xcalloc(1, sizeof(*split_index
->base
));
2286 base_oid_hex
= oid_to_hex(&split_index
->base_oid
);
2287 base_path
= xstrfmt("%s/sharedindex.%s", gitdir
, base_oid_hex
);
2288 trace2_region_enter_printf("index", "shared/do_read_index",
2289 the_repository
, "%s", base_path
);
2290 ret
= do_read_index(split_index
->base
, base_path
, 1);
2291 trace2_region_leave_printf("index", "shared/do_read_index",
2292 the_repository
, "%s", base_path
);
2293 if (!oideq(&split_index
->base_oid
, &split_index
->base
->oid
))
2294 die(_("broken index, expect %s in %s, got %s"),
2295 base_oid_hex
, base_path
,
2296 oid_to_hex(&split_index
->base
->oid
));
2298 freshen_shared_index(base_path
, 0);
2299 merge_base_index(istate
);
2300 post_read_index_from(istate
);
2301 trace_performance_leave("read cache %s", base_path
);
2306 int is_index_unborn(struct index_state
*istate
)
2308 return (!istate
->cache_nr
&& !istate
->timestamp
.sec
);
2311 int discard_index(struct index_state
*istate
)
2314 * Cache entries in istate->cache[] should have been allocated
2315 * from the memory pool associated with this index, or from an
2316 * associated split_index. There is no need to free individual
2317 * cache entries. validate_cache_entries can detect when this
2318 * assertion does not hold.
2320 validate_cache_entries(istate
);
2322 resolve_undo_clear_index(istate
);
2323 istate
->cache_nr
= 0;
2324 istate
->cache_changed
= 0;
2325 istate
->timestamp
.sec
= 0;
2326 istate
->timestamp
.nsec
= 0;
2327 free_name_hash(istate
);
2328 cache_tree_free(&(istate
->cache_tree
));
2329 istate
->initialized
= 0;
2330 istate
->fsmonitor_has_run_once
= 0;
2331 FREE_AND_NULL(istate
->cache
);
2332 istate
->cache_alloc
= 0;
2333 discard_split_index(istate
);
2334 free_untracked_cache(istate
->untracked
);
2335 istate
->untracked
= NULL
;
2337 if (istate
->ce_mem_pool
) {
2338 mem_pool_discard(istate
->ce_mem_pool
, should_validate_cache_entries());
2339 istate
->ce_mem_pool
= NULL
;
2346 * Validate the cache entries of this index.
2347 * All cache entries associated with this index
2348 * should have been allocated by the memory pool
2349 * associated with this index, or by a referenced
2352 void validate_cache_entries(const struct index_state
*istate
)
2356 if (!should_validate_cache_entries() ||!istate
|| !istate
->initialized
)
2359 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2361 BUG("cache entry is not allocated from expected memory pool");
2362 } else if (!istate
->ce_mem_pool
||
2363 !mem_pool_contains(istate
->ce_mem_pool
, istate
->cache
[i
])) {
2364 if (!istate
->split_index
||
2365 !istate
->split_index
->base
||
2366 !istate
->split_index
->base
->ce_mem_pool
||
2367 !mem_pool_contains(istate
->split_index
->base
->ce_mem_pool
, istate
->cache
[i
])) {
2368 BUG("cache entry is not allocated from expected memory pool");
2373 if (istate
->split_index
)
2374 validate_cache_entries(istate
->split_index
->base
);
2377 int unmerged_index(const struct index_state
*istate
)
2380 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2381 if (ce_stage(istate
->cache
[i
]))
2387 int repo_index_has_changes(struct repository
*repo
,
2391 struct index_state
*istate
= repo
->index
;
2392 struct object_id cmp
;
2396 cmp
= tree
->object
.oid
;
2397 if (tree
|| !get_oid_tree("HEAD", &cmp
)) {
2398 struct diff_options opt
;
2400 repo_diff_setup(repo
, &opt
);
2401 opt
.flags
.exit_with_status
= 1;
2403 opt
.flags
.quick
= 1;
2404 do_diff_cache(&cmp
, &opt
);
2406 for (i
= 0; sb
&& i
< diff_queued_diff
.nr
; i
++) {
2408 strbuf_addch(sb
, ' ');
2409 strbuf_addstr(sb
, diff_queued_diff
.queue
[i
]->two
->path
);
2412 return opt
.flags
.has_changes
!= 0;
2414 for (i
= 0; sb
&& i
< istate
->cache_nr
; i
++) {
2416 strbuf_addch(sb
, ' ');
2417 strbuf_addstr(sb
, istate
->cache
[i
]->name
);
2419 return !!istate
->cache_nr
;
2423 #define WRITE_BUFFER_SIZE 8192
2424 static unsigned char write_buffer
[WRITE_BUFFER_SIZE
];
2425 static unsigned long write_buffer_len
;
2427 static int ce_write_flush(git_hash_ctx
*context
, int fd
)
2429 unsigned int buffered
= write_buffer_len
;
2431 the_hash_algo
->update_fn(context
, write_buffer
, buffered
);
2432 if (write_in_full(fd
, write_buffer
, buffered
) < 0)
2434 write_buffer_len
= 0;
2439 static int ce_write(git_hash_ctx
*context
, int fd
, void *data
, unsigned int len
)
2442 unsigned int buffered
= write_buffer_len
;
2443 unsigned int partial
= WRITE_BUFFER_SIZE
- buffered
;
2446 memcpy(write_buffer
+ buffered
, data
, partial
);
2447 buffered
+= partial
;
2448 if (buffered
== WRITE_BUFFER_SIZE
) {
2449 write_buffer_len
= buffered
;
2450 if (ce_write_flush(context
, fd
))
2454 write_buffer_len
= buffered
;
2456 data
= (char *) data
+ partial
;
2461 static int write_index_ext_header(git_hash_ctx
*context
, git_hash_ctx
*eoie_context
,
2462 int fd
, unsigned int ext
, unsigned int sz
)
2467 the_hash_algo
->update_fn(eoie_context
, &ext
, 4);
2468 the_hash_algo
->update_fn(eoie_context
, &sz
, 4);
2470 return ((ce_write(context
, fd
, &ext
, 4) < 0) ||
2471 (ce_write(context
, fd
, &sz
, 4) < 0)) ? -1 : 0;
2474 static int ce_flush(git_hash_ctx
*context
, int fd
, unsigned char *hash
)
2476 unsigned int left
= write_buffer_len
;
2479 write_buffer_len
= 0;
2480 the_hash_algo
->update_fn(context
, write_buffer
, left
);
2483 /* Flush first if not enough space for hash signature */
2484 if (left
+ the_hash_algo
->rawsz
> WRITE_BUFFER_SIZE
) {
2485 if (write_in_full(fd
, write_buffer
, left
) < 0)
2490 /* Append the hash signature at the end */
2491 the_hash_algo
->final_fn(write_buffer
+ left
, context
);
2492 hashcpy(hash
, write_buffer
+ left
);
2493 left
+= the_hash_algo
->rawsz
;
2494 return (write_in_full(fd
, write_buffer
, left
) < 0) ? -1 : 0;
2497 static void ce_smudge_racily_clean_entry(struct index_state
*istate
,
2498 struct cache_entry
*ce
)
2501 * The only thing we care about in this function is to smudge the
2502 * falsely clean entry due to touch-update-touch race, so we leave
2503 * everything else as they are. We are called for entries whose
2504 * ce_stat_data.sd_mtime match the index file mtime.
2506 * Note that this actually does not do much for gitlinks, for
2507 * which ce_match_stat_basic() always goes to the actual
2508 * contents. The caller checks with is_racy_timestamp() which
2509 * always says "no" for gitlinks, so we are not called for them ;-)
2513 if (lstat(ce
->name
, &st
) < 0)
2515 if (ce_match_stat_basic(ce
, &st
))
2517 if (ce_modified_check_fs(istate
, ce
, &st
)) {
2518 /* This is "racily clean"; smudge it. Note that this
2519 * is a tricky code. At first glance, it may appear
2520 * that it can break with this sequence:
2522 * $ echo xyzzy >frotz
2523 * $ git-update-index --add frotz
2526 * $ echo filfre >nitfol
2527 * $ git-update-index --add nitfol
2529 * but it does not. When the second update-index runs,
2530 * it notices that the entry "frotz" has the same timestamp
2531 * as index, and if we were to smudge it by resetting its
2532 * size to zero here, then the object name recorded
2533 * in index is the 6-byte file but the cached stat information
2534 * becomes zero --- which would then match what we would
2535 * obtain from the filesystem next time we stat("frotz").
2537 * However, the second update-index, before calling
2538 * this function, notices that the cached size is 6
2539 * bytes and what is on the filesystem is an empty
2540 * file, and never calls us, so the cached size information
2541 * for "frotz" stays 6 which does not match the filesystem.
2543 ce
->ce_stat_data
.sd_size
= 0;
2547 /* Copy miscellaneous fields but not the name */
2548 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry
*ondisk
,
2549 struct cache_entry
*ce
)
2552 const unsigned hashsz
= the_hash_algo
->rawsz
;
2553 uint16_t *flagsp
= (uint16_t *)(ondisk
->data
+ hashsz
);
2555 ondisk
->ctime
.sec
= htonl(ce
->ce_stat_data
.sd_ctime
.sec
);
2556 ondisk
->mtime
.sec
= htonl(ce
->ce_stat_data
.sd_mtime
.sec
);
2557 ondisk
->ctime
.nsec
= htonl(ce
->ce_stat_data
.sd_ctime
.nsec
);
2558 ondisk
->mtime
.nsec
= htonl(ce
->ce_stat_data
.sd_mtime
.nsec
);
2559 ondisk
->dev
= htonl(ce
->ce_stat_data
.sd_dev
);
2560 ondisk
->ino
= htonl(ce
->ce_stat_data
.sd_ino
);
2561 ondisk
->mode
= htonl(ce
->ce_mode
);
2562 ondisk
->uid
= htonl(ce
->ce_stat_data
.sd_uid
);
2563 ondisk
->gid
= htonl(ce
->ce_stat_data
.sd_gid
);
2564 ondisk
->size
= htonl(ce
->ce_stat_data
.sd_size
);
2565 hashcpy(ondisk
->data
, ce
->oid
.hash
);
2567 flags
= ce
->ce_flags
& ~CE_NAMEMASK
;
2568 flags
|= (ce_namelen(ce
) >= CE_NAMEMASK
? CE_NAMEMASK
: ce_namelen(ce
));
2569 flagsp
[0] = htons(flags
);
2570 if (ce
->ce_flags
& CE_EXTENDED
) {
2571 flagsp
[1] = htons((ce
->ce_flags
& CE_EXTENDED_FLAGS
) >> 16);
2575 static int ce_write_entry(git_hash_ctx
*c
, int fd
, struct cache_entry
*ce
,
2576 struct strbuf
*previous_name
, struct ondisk_cache_entry
*ondisk
)
2580 unsigned int saved_namelen
;
2581 int stripped_name
= 0;
2582 static unsigned char padding
[8] = { 0x00 };
2584 if (ce
->ce_flags
& CE_STRIP_NAME
) {
2585 saved_namelen
= ce_namelen(ce
);
2590 size
= offsetof(struct ondisk_cache_entry
,data
) + ondisk_data_size(ce
->ce_flags
, 0);
2592 if (!previous_name
) {
2593 int len
= ce_namelen(ce
);
2594 copy_cache_entry_to_ondisk(ondisk
, ce
);
2595 result
= ce_write(c
, fd
, ondisk
, size
);
2597 result
= ce_write(c
, fd
, ce
->name
, len
);
2599 result
= ce_write(c
, fd
, padding
, align_padding_size(size
, len
));
2601 int common
, to_remove
, prefix_size
;
2602 unsigned char to_remove_vi
[16];
2604 (ce
->name
[common
] &&
2605 common
< previous_name
->len
&&
2606 ce
->name
[common
] == previous_name
->buf
[common
]);
2608 ; /* still matching */
2609 to_remove
= previous_name
->len
- common
;
2610 prefix_size
= encode_varint(to_remove
, to_remove_vi
);
2612 copy_cache_entry_to_ondisk(ondisk
, ce
);
2613 result
= ce_write(c
, fd
, ondisk
, size
);
2615 result
= ce_write(c
, fd
, to_remove_vi
, prefix_size
);
2617 result
= ce_write(c
, fd
, ce
->name
+ common
, ce_namelen(ce
) - common
);
2619 result
= ce_write(c
, fd
, padding
, 1);
2621 strbuf_splice(previous_name
, common
, to_remove
,
2622 ce
->name
+ common
, ce_namelen(ce
) - common
);
2624 if (stripped_name
) {
2625 ce
->ce_namelen
= saved_namelen
;
2626 ce
->ce_flags
&= ~CE_STRIP_NAME
;
2633 * This function verifies if index_state has the correct sha1 of the
2634 * index file. Don't die if we have any other failure, just return 0.
2636 static int verify_index_from(const struct index_state
*istate
, const char *path
)
2641 unsigned char hash
[GIT_MAX_RAWSZ
];
2643 if (!istate
->initialized
)
2646 fd
= open(path
, O_RDONLY
);
2653 if (st
.st_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2656 n
= pread_in_full(fd
, hash
, the_hash_algo
->rawsz
, st
.st_size
- the_hash_algo
->rawsz
);
2657 if (n
!= the_hash_algo
->rawsz
)
2660 if (!hasheq(istate
->oid
.hash
, hash
))
2671 static int repo_verify_index(struct repository
*repo
)
2673 return verify_index_from(repo
->index
, repo
->index_file
);
2676 static int has_racy_timestamp(struct index_state
*istate
)
2678 int entries
= istate
->cache_nr
;
2681 for (i
= 0; i
< entries
; i
++) {
2682 struct cache_entry
*ce
= istate
->cache
[i
];
2683 if (is_racy_timestamp(istate
, ce
))
2689 void repo_update_index_if_able(struct repository
*repo
,
2690 struct lock_file
*lockfile
)
2692 if ((repo
->index
->cache_changed
||
2693 has_racy_timestamp(repo
->index
)) &&
2694 repo_verify_index(repo
))
2695 write_locked_index(repo
->index
, lockfile
, COMMIT_LOCK
);
2697 rollback_lock_file(lockfile
);
2700 static int record_eoie(void)
2704 if (!git_config_get_bool("index.recordendofindexentries", &val
))
2708 * As a convenience, the end of index entries extension
2709 * used for threading is written by default if the user
2710 * explicitly requested threaded index reads.
2712 return !git_config_get_index_threads(&val
) && val
!= 1;
2715 static int record_ieot(void)
2719 if (!git_config_get_bool("index.recordoffsettable", &val
))
2723 * As a convenience, the offset table used for threading is
2724 * written by default if the user explicitly requested
2725 * threaded index reads.
2727 return !git_config_get_index_threads(&val
) && val
!= 1;
2731 * On success, `tempfile` is closed. If it is the temporary file
2732 * of a `struct lock_file`, we will therefore effectively perform
2733 * a 'close_lock_file_gently()`. Since that is an implementation
2734 * detail of lockfiles, callers of `do_write_index()` should not
2737 static int do_write_index(struct index_state
*istate
, struct tempfile
*tempfile
,
2738 int strip_extensions
)
2740 uint64_t start
= getnanotime();
2741 int newfd
= tempfile
->fd
;
2742 git_hash_ctx c
, eoie_c
;
2743 struct cache_header hdr
;
2744 int i
, err
= 0, removed
, extended
, hdr_version
;
2745 struct cache_entry
**cache
= istate
->cache
;
2746 int entries
= istate
->cache_nr
;
2748 struct ondisk_cache_entry ondisk
;
2749 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
2750 int drop_cache_tree
= istate
->drop_cache_tree
;
2752 int ieot_entries
= 1;
2753 struct index_entry_offset_table
*ieot
= NULL
;
2756 for (i
= removed
= extended
= 0; i
< entries
; i
++) {
2757 if (cache
[i
]->ce_flags
& CE_REMOVE
)
2760 /* reduce extended entries if possible */
2761 cache
[i
]->ce_flags
&= ~CE_EXTENDED
;
2762 if (cache
[i
]->ce_flags
& CE_EXTENDED_FLAGS
) {
2764 cache
[i
]->ce_flags
|= CE_EXTENDED
;
2768 if (!istate
->version
) {
2769 istate
->version
= get_index_format_default(the_repository
);
2770 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0))
2771 init_split_index(istate
);
2774 /* demote version 3 to version 2 when the latter suffices */
2775 if (istate
->version
== 3 || istate
->version
== 2)
2776 istate
->version
= extended
? 3 : 2;
2778 hdr_version
= istate
->version
;
2780 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
2781 hdr
.hdr_version
= htonl(hdr_version
);
2782 hdr
.hdr_entries
= htonl(entries
- removed
);
2784 the_hash_algo
->init_fn(&c
);
2785 if (ce_write(&c
, newfd
, &hdr
, sizeof(hdr
)) < 0)
2788 if (!HAVE_THREADS
|| git_config_get_index_threads(&nr_threads
))
2791 if (nr_threads
!= 1 && record_ieot()) {
2792 int ieot_blocks
, cpus
;
2795 * ensure default number of ieot blocks maps evenly to the
2796 * default number of threads that will process them leaving
2797 * room for the thread to load the index extensions.
2800 ieot_blocks
= istate
->cache_nr
/ THREAD_COST
;
2801 cpus
= online_cpus();
2802 if (ieot_blocks
> cpus
- 1)
2803 ieot_blocks
= cpus
- 1;
2805 ieot_blocks
= nr_threads
;
2806 if (ieot_blocks
> istate
->cache_nr
)
2807 ieot_blocks
= istate
->cache_nr
;
2811 * no reason to write out the IEOT extension if we don't
2812 * have enough blocks to utilize multi-threading
2814 if (ieot_blocks
> 1) {
2815 ieot
= xcalloc(1, sizeof(struct index_entry_offset_table
)
2816 + (ieot_blocks
* sizeof(struct index_entry_offset
)));
2817 ieot_entries
= DIV_ROUND_UP(entries
, ieot_blocks
);
2821 offset
= lseek(newfd
, 0, SEEK_CUR
);
2826 offset
+= write_buffer_len
;
2828 previous_name
= (hdr_version
== 4) ? &previous_name_buf
: NULL
;
2830 for (i
= 0; i
< entries
; i
++) {
2831 struct cache_entry
*ce
= cache
[i
];
2832 if (ce
->ce_flags
& CE_REMOVE
)
2834 if (!ce_uptodate(ce
) && is_racy_timestamp(istate
, ce
))
2835 ce_smudge_racily_clean_entry(istate
, ce
);
2836 if (is_null_oid(&ce
->oid
)) {
2837 static const char msg
[] = "cache entry has null sha1: %s";
2838 static int allow
= -1;
2841 allow
= git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2843 warning(msg
, ce
->name
);
2845 err
= error(msg
, ce
->name
);
2847 drop_cache_tree
= 1;
2849 if (ieot
&& i
&& (i
% ieot_entries
== 0)) {
2850 ieot
->entries
[ieot
->nr
].nr
= nr
;
2851 ieot
->entries
[ieot
->nr
].offset
= offset
;
2854 * If we have a V4 index, set the first byte to an invalid
2855 * character to ensure there is nothing common with the previous
2859 previous_name
->buf
[0] = 0;
2861 offset
= lseek(newfd
, 0, SEEK_CUR
);
2866 offset
+= write_buffer_len
;
2868 if (ce_write_entry(&c
, newfd
, ce
, previous_name
, (struct ondisk_cache_entry
*)&ondisk
) < 0)
2876 ieot
->entries
[ieot
->nr
].nr
= nr
;
2877 ieot
->entries
[ieot
->nr
].offset
= offset
;
2880 strbuf_release(&previous_name_buf
);
2887 /* Write extension data here */
2888 offset
= lseek(newfd
, 0, SEEK_CUR
);
2893 offset
+= write_buffer_len
;
2894 the_hash_algo
->init_fn(&eoie_c
);
2897 * Lets write out CACHE_EXT_INDEXENTRYOFFSETTABLE first so that we
2898 * can minimize the number of extensions we have to scan through to
2899 * find it during load. Write it out regardless of the
2900 * strip_extensions parameter as we need it when loading the shared
2904 struct strbuf sb
= STRBUF_INIT
;
2906 write_ieot_extension(&sb
, ieot
);
2907 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_INDEXENTRYOFFSETTABLE
, sb
.len
) < 0
2908 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2909 strbuf_release(&sb
);
2915 if (!strip_extensions
&& istate
->split_index
&&
2916 !is_null_oid(&istate
->split_index
->base_oid
)) {
2917 struct strbuf sb
= STRBUF_INIT
;
2919 err
= write_link_extension(&sb
, istate
) < 0 ||
2920 write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_LINK
,
2922 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2923 strbuf_release(&sb
);
2927 if (!strip_extensions
&& !drop_cache_tree
&& istate
->cache_tree
) {
2928 struct strbuf sb
= STRBUF_INIT
;
2930 cache_tree_write(&sb
, istate
->cache_tree
);
2931 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_TREE
, sb
.len
) < 0
2932 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2933 strbuf_release(&sb
);
2937 if (!strip_extensions
&& istate
->resolve_undo
) {
2938 struct strbuf sb
= STRBUF_INIT
;
2940 resolve_undo_write(&sb
, istate
->resolve_undo
);
2941 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_RESOLVE_UNDO
,
2943 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2944 strbuf_release(&sb
);
2948 if (!strip_extensions
&& istate
->untracked
) {
2949 struct strbuf sb
= STRBUF_INIT
;
2951 write_untracked_extension(&sb
, istate
->untracked
);
2952 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_UNTRACKED
,
2954 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2955 strbuf_release(&sb
);
2959 if (!strip_extensions
&& istate
->fsmonitor_last_update
) {
2960 struct strbuf sb
= STRBUF_INIT
;
2962 write_fsmonitor_extension(&sb
, istate
);
2963 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_FSMONITOR
, sb
.len
) < 0
2964 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2965 strbuf_release(&sb
);
2971 * CACHE_EXT_ENDOFINDEXENTRIES must be written as the last entry before the SHA1
2972 * so that it can be found and processed before all the index entries are
2973 * read. Write it out regardless of the strip_extensions parameter as we need it
2974 * when loading the shared index.
2976 if (offset
&& record_eoie()) {
2977 struct strbuf sb
= STRBUF_INIT
;
2979 write_eoie_extension(&sb
, &eoie_c
, offset
);
2980 err
= write_index_ext_header(&c
, NULL
, newfd
, CACHE_EXT_ENDOFINDEXENTRIES
, sb
.len
) < 0
2981 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2982 strbuf_release(&sb
);
2987 if (ce_flush(&c
, newfd
, istate
->oid
.hash
))
2989 if (close_tempfile_gently(tempfile
)) {
2990 error(_("could not close '%s'"), tempfile
->filename
.buf
);
2993 if (stat(tempfile
->filename
.buf
, &st
))
2995 istate
->timestamp
.sec
= (unsigned int)st
.st_mtime
;
2996 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
2997 trace_performance_since(start
, "write index, changed mask = %x", istate
->cache_changed
);
3000 * TODO trace2: replace "the_repository" with the actual repo instance
3001 * that is associated with the given "istate".
3003 trace2_data_intmax("index", the_repository
, "write/version",
3005 trace2_data_intmax("index", the_repository
, "write/cache_nr",
3011 void set_alternate_index_output(const char *name
)
3013 alternate_index_output
= name
;
3016 static int commit_locked_index(struct lock_file
*lk
)
3018 if (alternate_index_output
)
3019 return commit_lock_file_to(lk
, alternate_index_output
);
3021 return commit_lock_file(lk
);
3024 static int do_write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
3030 * TODO trace2: replace "the_repository" with the actual repo instance
3031 * that is associated with the given "istate".
3033 trace2_region_enter_printf("index", "do_write_index", the_repository
,
3034 "%s", lock
->tempfile
->filename
.buf
);
3035 ret
= do_write_index(istate
, lock
->tempfile
, 0);
3036 trace2_region_leave_printf("index", "do_write_index", the_repository
,
3037 "%s", lock
->tempfile
->filename
.buf
);
3041 if (flags
& COMMIT_LOCK
)
3042 ret
= commit_locked_index(lock
);
3044 ret
= close_lock_file_gently(lock
);
3046 run_hook_le(NULL
, "post-index-change",
3047 istate
->updated_workdir
? "1" : "0",
3048 istate
->updated_skipworktree
? "1" : "0", NULL
);
3049 istate
->updated_workdir
= 0;
3050 istate
->updated_skipworktree
= 0;
3055 static int write_split_index(struct index_state
*istate
,
3056 struct lock_file
*lock
,
3060 prepare_to_write_split_index(istate
);
3061 ret
= do_write_locked_index(istate
, lock
, flags
);
3062 finish_writing_split_index(istate
);
3066 static const char *shared_index_expire
= "2.weeks.ago";
3068 static unsigned long get_shared_index_expire_date(void)
3070 static unsigned long shared_index_expire_date
;
3071 static int shared_index_expire_date_prepared
;
3073 if (!shared_index_expire_date_prepared
) {
3074 git_config_get_expiry("splitindex.sharedindexexpire",
3075 &shared_index_expire
);
3076 shared_index_expire_date
= approxidate(shared_index_expire
);
3077 shared_index_expire_date_prepared
= 1;
3080 return shared_index_expire_date
;
3083 static int should_delete_shared_index(const char *shared_index_path
)
3086 unsigned long expiration
;
3088 /* Check timestamp */
3089 expiration
= get_shared_index_expire_date();
3092 if (stat(shared_index_path
, &st
))
3093 return error_errno(_("could not stat '%s'"), shared_index_path
);
3094 if (st
.st_mtime
> expiration
)
3100 static int clean_shared_index_files(const char *current_hex
)
3103 DIR *dir
= opendir(get_git_dir());
3106 return error_errno(_("unable to open git dir: %s"), get_git_dir());
3108 while ((de
= readdir(dir
)) != NULL
) {
3109 const char *sha1_hex
;
3110 const char *shared_index_path
;
3111 if (!skip_prefix(de
->d_name
, "sharedindex.", &sha1_hex
))
3113 if (!strcmp(sha1_hex
, current_hex
))
3115 shared_index_path
= git_path("%s", de
->d_name
);
3116 if (should_delete_shared_index(shared_index_path
) > 0 &&
3117 unlink(shared_index_path
))
3118 warning_errno(_("unable to unlink: %s"), shared_index_path
);
3125 static int write_shared_index(struct index_state
*istate
,
3126 struct tempfile
**temp
)
3128 struct split_index
*si
= istate
->split_index
;
3131 move_cache_to_base_index(istate
);
3133 trace2_region_enter_printf("index", "shared/do_write_index",
3134 the_repository
, "%s", (*temp
)->filename
.buf
);
3135 ret
= do_write_index(si
->base
, *temp
, 1);
3136 trace2_region_leave_printf("index", "shared/do_write_index",
3137 the_repository
, "%s", (*temp
)->filename
.buf
);
3141 ret
= adjust_shared_perm(get_tempfile_path(*temp
));
3143 error(_("cannot fix permission bits on '%s'"), get_tempfile_path(*temp
));
3146 ret
= rename_tempfile(temp
,
3147 git_path("sharedindex.%s", oid_to_hex(&si
->base
->oid
)));
3149 oidcpy(&si
->base_oid
, &si
->base
->oid
);
3150 clean_shared_index_files(oid_to_hex(&si
->base
->oid
));
3156 static const int default_max_percent_split_change
= 20;
3158 static int too_many_not_shared_entries(struct index_state
*istate
)
3160 int i
, not_shared
= 0;
3161 int max_split
= git_config_get_max_percent_split_change();
3163 switch (max_split
) {
3165 /* not or badly configured: use the default value */
3166 max_split
= default_max_percent_split_change
;
3169 return 1; /* 0% means always write a new shared index */
3171 return 0; /* 100% means never write a new shared index */
3173 break; /* just use the configured value */
3176 /* Count not shared entries */
3177 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3178 struct cache_entry
*ce
= istate
->cache
[i
];
3183 return (int64_t)istate
->cache_nr
* max_split
< (int64_t)not_shared
* 100;
3186 int write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
3189 int new_shared_index
, ret
;
3190 struct split_index
*si
= istate
->split_index
;
3192 if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0))
3193 cache_tree_verify(the_repository
, istate
);
3195 if ((flags
& SKIP_IF_UNCHANGED
) && !istate
->cache_changed
) {
3196 if (flags
& COMMIT_LOCK
)
3197 rollback_lock_file(lock
);
3201 if (istate
->fsmonitor_last_update
)
3202 fill_fsmonitor_bitmap(istate
);
3204 if (!si
|| alternate_index_output
||
3205 (istate
->cache_changed
& ~EXTMASK
)) {
3207 oidclr(&si
->base_oid
);
3208 ret
= do_write_locked_index(istate
, lock
, flags
);
3212 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0)) {
3213 int v
= si
->base_oid
.hash
[0];
3215 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3217 if (too_many_not_shared_entries(istate
))
3218 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3220 new_shared_index
= istate
->cache_changed
& SPLIT_INDEX_ORDERED
;
3222 if (new_shared_index
) {
3223 struct tempfile
*temp
;
3226 /* Same initial permissions as the main .git/index file */
3227 temp
= mks_tempfile_sm(git_path("sharedindex_XXXXXX"), 0, 0666);
3229 oidclr(&si
->base_oid
);
3230 ret
= do_write_locked_index(istate
, lock
, flags
);
3233 ret
= write_shared_index(istate
, &temp
);
3235 saved_errno
= errno
;
3236 if (is_tempfile_active(temp
))
3237 delete_tempfile(&temp
);
3238 errno
= saved_errno
;
3244 ret
= write_split_index(istate
, lock
, flags
);
3246 /* Freshen the shared index only if the split-index was written */
3247 if (!ret
&& !new_shared_index
&& !is_null_oid(&si
->base_oid
)) {
3248 const char *shared_index
= git_path("sharedindex.%s",
3249 oid_to_hex(&si
->base_oid
));
3250 freshen_shared_index(shared_index
, 1);
3254 if (flags
& COMMIT_LOCK
)
3255 rollback_lock_file(lock
);
3260 * Read the index file that is potentially unmerged into given
3261 * index_state, dropping any unmerged entries to stage #0 (potentially
3262 * resulting in a path appearing as both a file and a directory in the
3263 * index; the caller is responsible to clear out the extra entries
3264 * before writing the index to a tree). Returns true if the index is
3265 * unmerged. Callers who want to refuse to work from an unmerged
3266 * state can call this and check its return value, instead of calling
3269 int repo_read_index_unmerged(struct repository
*repo
)
3271 struct index_state
*istate
;
3275 repo_read_index(repo
);
3276 istate
= repo
->index
;
3277 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3278 struct cache_entry
*ce
= istate
->cache
[i
];
3279 struct cache_entry
*new_ce
;
3285 len
= ce_namelen(ce
);
3286 new_ce
= make_empty_cache_entry(istate
, len
);
3287 memcpy(new_ce
->name
, ce
->name
, len
);
3288 new_ce
->ce_flags
= create_ce_flags(0) | CE_CONFLICTED
;
3289 new_ce
->ce_namelen
= len
;
3290 new_ce
->ce_mode
= ce
->ce_mode
;
3291 if (add_index_entry(istate
, new_ce
, ADD_CACHE_SKIP_DFCHECK
))
3292 return error(_("%s: cannot drop to stage #0"),
3299 * Returns 1 if the path is an "other" path with respect to
3300 * the index; that is, the path is not mentioned in the index at all,
3301 * either as a file, a directory with some files in the index,
3302 * or as an unmerged entry.
3304 * We helpfully remove a trailing "/" from directories so that
3305 * the output of read_directory can be used as-is.
3307 int index_name_is_other(const struct index_state
*istate
, const char *name
,
3311 if (namelen
&& name
[namelen
- 1] == '/')
3313 pos
= index_name_pos(istate
, name
, namelen
);
3315 return 0; /* exact match */
3317 if (pos
< istate
->cache_nr
) {
3318 struct cache_entry
*ce
= istate
->cache
[pos
];
3319 if (ce_namelen(ce
) == namelen
&&
3320 !memcmp(ce
->name
, name
, namelen
))
3321 return 0; /* Yup, this one exists unmerged */
3326 void *read_blob_data_from_index(const struct index_state
*istate
,
3327 const char *path
, unsigned long *size
)
3331 enum object_type type
;
3335 pos
= index_name_pos(istate
, path
, len
);
3338 * We might be in the middle of a merge, in which
3339 * case we would read stage #2 (ours).
3343 (pos
< 0 && i
< istate
->cache_nr
&&
3344 !strcmp(istate
->cache
[i
]->name
, path
));
3346 if (ce_stage(istate
->cache
[i
]) == 2)
3351 data
= read_object_file(&istate
->cache
[pos
]->oid
, &type
, &sz
);
3352 if (!data
|| type
!= OBJ_BLOB
) {
3361 void stat_validity_clear(struct stat_validity
*sv
)
3363 FREE_AND_NULL(sv
->sd
);
3366 int stat_validity_check(struct stat_validity
*sv
, const char *path
)
3370 if (stat(path
, &st
) < 0)
3371 return sv
->sd
== NULL
;
3374 return S_ISREG(st
.st_mode
) && !match_stat_data(sv
->sd
, &st
);
3377 void stat_validity_update(struct stat_validity
*sv
, int fd
)
3381 if (fstat(fd
, &st
) < 0 || !S_ISREG(st
.st_mode
))
3382 stat_validity_clear(sv
);
3385 sv
->sd
= xcalloc(1, sizeof(struct stat_data
));
3386 fill_stat_data(sv
->sd
, &st
);
3390 void move_index_extensions(struct index_state
*dst
, struct index_state
*src
)
3392 dst
->untracked
= src
->untracked
;
3393 src
->untracked
= NULL
;
3394 dst
->cache_tree
= src
->cache_tree
;
3395 src
->cache_tree
= NULL
;
3398 struct cache_entry
*dup_cache_entry(const struct cache_entry
*ce
,
3399 struct index_state
*istate
)
3401 unsigned int size
= ce_size(ce
);
3402 int mem_pool_allocated
;
3403 struct cache_entry
*new_entry
= make_empty_cache_entry(istate
, ce_namelen(ce
));
3404 mem_pool_allocated
= new_entry
->mem_pool_allocated
;
3406 memcpy(new_entry
, ce
, size
);
3407 new_entry
->mem_pool_allocated
= mem_pool_allocated
;
3411 void discard_cache_entry(struct cache_entry
*ce
)
3413 if (ce
&& should_validate_cache_entries())
3414 memset(ce
, 0xCD, cache_entry_size(ce
->ce_namelen
));
3416 if (ce
&& ce
->mem_pool_allocated
)
3422 int should_validate_cache_entries(void)
3424 static int validate_index_cache_entries
= -1;
3426 if (validate_index_cache_entries
< 0) {
3427 if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))
3428 validate_index_cache_entries
= 1;
3430 validate_index_cache_entries
= 0;
3433 return validate_index_cache_entries
;
3436 #define EOIE_SIZE (4 + GIT_SHA1_RAWSZ) /* <4-byte offset> + <20-byte hash> */
3437 #define EOIE_SIZE_WITH_HEADER (4 + 4 + EOIE_SIZE) /* <4-byte signature> + <4-byte length> + EOIE_SIZE */
3439 static size_t read_eoie_extension(const char *mmap
, size_t mmap_size
)
3442 * The end of index entries (EOIE) extension is guaranteed to be last
3443 * so that it can be found by scanning backwards from the EOF.
3450 const char *index
, *eoie
;
3452 size_t offset
, src_offset
;
3453 unsigned char hash
[GIT_MAX_RAWSZ
];
3456 /* ensure we have an index big enough to contain an EOIE extension */
3457 if (mmap_size
< sizeof(struct cache_header
) + EOIE_SIZE_WITH_HEADER
+ the_hash_algo
->rawsz
)
3460 /* validate the extension signature */
3461 index
= eoie
= mmap
+ mmap_size
- EOIE_SIZE_WITH_HEADER
- the_hash_algo
->rawsz
;
3462 if (CACHE_EXT(index
) != CACHE_EXT_ENDOFINDEXENTRIES
)
3464 index
+= sizeof(uint32_t);
3466 /* validate the extension size */
3467 extsize
= get_be32(index
);
3468 if (extsize
!= EOIE_SIZE
)
3470 index
+= sizeof(uint32_t);
3473 * Validate the offset we're going to look for the first extension
3474 * signature is after the index header and before the eoie extension.
3476 offset
= get_be32(index
);
3477 if (mmap
+ offset
< mmap
+ sizeof(struct cache_header
))
3479 if (mmap
+ offset
>= eoie
)
3481 index
+= sizeof(uint32_t);
3484 * The hash is computed over extension types and their sizes (but not
3485 * their contents). E.g. if we have "TREE" extension that is N-bytes
3486 * long, "REUC" extension that is M-bytes long, followed by "EOIE",
3487 * then the hash would be:
3489 * SHA-1("TREE" + <binary representation of N> +
3490 * "REUC" + <binary representation of M>)
3492 src_offset
= offset
;
3493 the_hash_algo
->init_fn(&c
);
3494 while (src_offset
< mmap_size
- the_hash_algo
->rawsz
- EOIE_SIZE_WITH_HEADER
) {
3495 /* After an array of active_nr index entries,
3496 * there can be arbitrary number of extended
3497 * sections, each of which is prefixed with
3498 * extension name (4-byte) and section length
3499 * in 4-byte network byte order.
3502 memcpy(&extsize
, mmap
+ src_offset
+ 4, 4);
3503 extsize
= ntohl(extsize
);
3505 /* verify the extension size isn't so large it will wrap around */
3506 if (src_offset
+ 8 + extsize
< src_offset
)
3509 the_hash_algo
->update_fn(&c
, mmap
+ src_offset
, 8);
3512 src_offset
+= extsize
;
3514 the_hash_algo
->final_fn(hash
, &c
);
3515 if (!hasheq(hash
, (const unsigned char *)index
))
3518 /* Validate that the extension offsets returned us back to the eoie extension. */
3519 if (src_offset
!= mmap_size
- the_hash_algo
->rawsz
- EOIE_SIZE_WITH_HEADER
)
3525 static void write_eoie_extension(struct strbuf
*sb
, git_hash_ctx
*eoie_context
, size_t offset
)
3528 unsigned char hash
[GIT_MAX_RAWSZ
];
3531 put_be32(&buffer
, offset
);
3532 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3535 the_hash_algo
->final_fn(hash
, eoie_context
);
3536 strbuf_add(sb
, hash
, the_hash_algo
->rawsz
);
3539 #define IEOT_VERSION (1)
3541 static struct index_entry_offset_table
*read_ieot_extension(const char *mmap
, size_t mmap_size
, size_t offset
)
3543 const char *index
= NULL
;
3544 uint32_t extsize
, ext_version
;
3545 struct index_entry_offset_table
*ieot
;
3548 /* find the IEOT extension */
3551 while (offset
<= mmap_size
- the_hash_algo
->rawsz
- 8) {
3552 extsize
= get_be32(mmap
+ offset
+ 4);
3553 if (CACHE_EXT((mmap
+ offset
)) == CACHE_EXT_INDEXENTRYOFFSETTABLE
) {
3554 index
= mmap
+ offset
+ 4 + 4;
3563 /* validate the version is IEOT_VERSION */
3564 ext_version
= get_be32(index
);
3565 if (ext_version
!= IEOT_VERSION
) {
3566 error("invalid IEOT version %d", ext_version
);
3569 index
+= sizeof(uint32_t);
3571 /* extension size - version bytes / bytes per entry */
3572 nr
= (extsize
- sizeof(uint32_t)) / (sizeof(uint32_t) + sizeof(uint32_t));
3574 error("invalid number of IEOT entries %d", nr
);
3577 ieot
= xmalloc(sizeof(struct index_entry_offset_table
)
3578 + (nr
* sizeof(struct index_entry_offset
)));
3580 for (i
= 0; i
< nr
; i
++) {
3581 ieot
->entries
[i
].offset
= get_be32(index
);
3582 index
+= sizeof(uint32_t);
3583 ieot
->entries
[i
].nr
= get_be32(index
);
3584 index
+= sizeof(uint32_t);
3590 static void write_ieot_extension(struct strbuf
*sb
, struct index_entry_offset_table
*ieot
)
3596 put_be32(&buffer
, IEOT_VERSION
);
3597 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3600 for (i
= 0; i
< ieot
->nr
; i
++) {
3603 put_be32(&buffer
, ieot
->entries
[i
].offset
);
3604 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3607 put_be32(&buffer
, ieot
->entries
[i
].nr
);
3608 strbuf_add(sb
, &buffer
, sizeof(uint32_t));