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 *pool_ptr
= xmalloc(sizeof(**pool_ptr
));
94 mem_pool_init(*pool_ptr
, 0);
100 static const char *alternate_index_output
;
102 static void set_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
104 istate
->cache
[nr
] = ce
;
105 add_name_hash(istate
, ce
);
108 static void replace_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
110 struct cache_entry
*old
= istate
->cache
[nr
];
112 replace_index_entry_in_base(istate
, old
, ce
);
113 remove_name_hash(istate
, old
);
114 discard_cache_entry(old
);
115 ce
->ce_flags
&= ~CE_HASHED
;
116 set_index_entry(istate
, nr
, ce
);
117 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
118 mark_fsmonitor_invalid(istate
, ce
);
119 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
122 void rename_index_entry_at(struct index_state
*istate
, int nr
, const char *new_name
)
124 struct cache_entry
*old_entry
= istate
->cache
[nr
], *new_entry
;
125 int namelen
= strlen(new_name
);
127 new_entry
= make_empty_cache_entry(istate
, namelen
);
128 copy_cache_entry(new_entry
, old_entry
);
129 new_entry
->ce_flags
&= ~CE_HASHED
;
130 new_entry
->ce_namelen
= namelen
;
131 new_entry
->index
= 0;
132 memcpy(new_entry
->name
, new_name
, namelen
+ 1);
134 cache_tree_invalidate_path(istate
, old_entry
->name
);
135 untracked_cache_remove_from_index(istate
, old_entry
->name
);
136 remove_index_entry_at(istate
, nr
);
137 add_index_entry(istate
, new_entry
, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
140 void fill_stat_data(struct stat_data
*sd
, struct stat
*st
)
142 sd
->sd_ctime
.sec
= (unsigned int)st
->st_ctime
;
143 sd
->sd_mtime
.sec
= (unsigned int)st
->st_mtime
;
144 sd
->sd_ctime
.nsec
= ST_CTIME_NSEC(*st
);
145 sd
->sd_mtime
.nsec
= ST_MTIME_NSEC(*st
);
146 sd
->sd_dev
= st
->st_dev
;
147 sd
->sd_ino
= st
->st_ino
;
148 sd
->sd_uid
= st
->st_uid
;
149 sd
->sd_gid
= st
->st_gid
;
150 sd
->sd_size
= st
->st_size
;
153 int match_stat_data(const struct stat_data
*sd
, struct stat
*st
)
157 if (sd
->sd_mtime
.sec
!= (unsigned int)st
->st_mtime
)
158 changed
|= MTIME_CHANGED
;
159 if (trust_ctime
&& check_stat
&&
160 sd
->sd_ctime
.sec
!= (unsigned int)st
->st_ctime
)
161 changed
|= CTIME_CHANGED
;
164 if (check_stat
&& sd
->sd_mtime
.nsec
!= ST_MTIME_NSEC(*st
))
165 changed
|= MTIME_CHANGED
;
166 if (trust_ctime
&& check_stat
&&
167 sd
->sd_ctime
.nsec
!= ST_CTIME_NSEC(*st
))
168 changed
|= CTIME_CHANGED
;
172 if (sd
->sd_uid
!= (unsigned int) st
->st_uid
||
173 sd
->sd_gid
!= (unsigned int) st
->st_gid
)
174 changed
|= OWNER_CHANGED
;
175 if (sd
->sd_ino
!= (unsigned int) st
->st_ino
)
176 changed
|= INODE_CHANGED
;
181 * st_dev breaks on network filesystems where different
182 * clients will have different views of what "device"
183 * the filesystem is on
185 if (check_stat
&& sd
->sd_dev
!= (unsigned int) st
->st_dev
)
186 changed
|= INODE_CHANGED
;
189 if (sd
->sd_size
!= (unsigned int) st
->st_size
)
190 changed
|= DATA_CHANGED
;
196 * This only updates the "non-critical" parts of the directory
197 * cache, ie the parts that aren't tracked by GIT, and only used
198 * to validate the cache.
200 void fill_stat_cache_info(struct index_state
*istate
, struct cache_entry
*ce
, struct stat
*st
)
202 fill_stat_data(&ce
->ce_stat_data
, st
);
204 if (assume_unchanged
)
205 ce
->ce_flags
|= CE_VALID
;
207 if (S_ISREG(st
->st_mode
)) {
208 ce_mark_uptodate(ce
);
209 mark_fsmonitor_valid(istate
, ce
);
213 static int ce_compare_data(struct index_state
*istate
,
214 const struct cache_entry
*ce
,
218 int fd
= git_open_cloexec(ce
->name
, O_RDONLY
);
221 struct object_id oid
;
222 if (!index_fd(istate
, &oid
, fd
, st
, OBJ_BLOB
, ce
->name
, 0))
223 match
= !oideq(&oid
, &ce
->oid
);
224 /* index_fd() closed the file descriptor already */
229 static int ce_compare_link(const struct cache_entry
*ce
, size_t expected_size
)
234 enum object_type type
;
235 struct strbuf sb
= STRBUF_INIT
;
237 if (strbuf_readlink(&sb
, ce
->name
, expected_size
))
240 buffer
= read_object_file(&ce
->oid
, &type
, &size
);
243 match
= memcmp(buffer
, sb
.buf
, size
);
250 static int ce_compare_gitlink(const struct cache_entry
*ce
)
252 struct object_id oid
;
255 * We don't actually require that the .git directory
256 * under GITLINK directory be a valid git directory. It
257 * might even be missing (in case nobody populated that
260 * If so, we consider it always to match.
262 if (resolve_gitlink_ref(ce
->name
, "HEAD", &oid
) < 0)
264 return !oideq(&oid
, &ce
->oid
);
267 static int ce_modified_check_fs(struct index_state
*istate
,
268 const struct cache_entry
*ce
,
271 switch (st
->st_mode
& S_IFMT
) {
273 if (ce_compare_data(istate
, ce
, st
))
277 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
281 if (S_ISGITLINK(ce
->ce_mode
))
282 return ce_compare_gitlink(ce
) ? DATA_CHANGED
: 0;
283 /* else fallthrough */
290 static int ce_match_stat_basic(const struct cache_entry
*ce
, struct stat
*st
)
292 unsigned int changed
= 0;
294 if (ce
->ce_flags
& CE_REMOVE
)
295 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
297 switch (ce
->ce_mode
& S_IFMT
) {
299 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
300 /* We consider only the owner x bit to be relevant for
303 if (trust_executable_bit
&&
304 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
305 changed
|= MODE_CHANGED
;
308 if (!S_ISLNK(st
->st_mode
) &&
309 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
310 changed
|= TYPE_CHANGED
;
313 /* We ignore most of the st_xxx fields for gitlinks */
314 if (!S_ISDIR(st
->st_mode
))
315 changed
|= TYPE_CHANGED
;
316 else if (ce_compare_gitlink(ce
))
317 changed
|= DATA_CHANGED
;
320 BUG("unsupported ce_mode: %o", ce
->ce_mode
);
323 changed
|= match_stat_data(&ce
->ce_stat_data
, st
);
325 /* Racily smudged entry? */
326 if (!ce
->ce_stat_data
.sd_size
) {
327 if (!is_empty_blob_sha1(ce
->oid
.hash
))
328 changed
|= DATA_CHANGED
;
334 static int is_racy_stat(const struct index_state
*istate
,
335 const struct stat_data
*sd
)
337 return (istate
->timestamp
.sec
&&
339 /* nanosecond timestamped files can also be racy! */
340 (istate
->timestamp
.sec
< sd
->sd_mtime
.sec
||
341 (istate
->timestamp
.sec
== sd
->sd_mtime
.sec
&&
342 istate
->timestamp
.nsec
<= sd
->sd_mtime
.nsec
))
344 istate
->timestamp
.sec
<= sd
->sd_mtime
.sec
349 int is_racy_timestamp(const struct index_state
*istate
,
350 const struct cache_entry
*ce
)
352 return (!S_ISGITLINK(ce
->ce_mode
) &&
353 is_racy_stat(istate
, &ce
->ce_stat_data
));
356 int match_stat_data_racy(const struct index_state
*istate
,
357 const struct stat_data
*sd
, struct stat
*st
)
359 if (is_racy_stat(istate
, sd
))
360 return MTIME_CHANGED
;
361 return match_stat_data(sd
, st
);
364 int ie_match_stat(struct index_state
*istate
,
365 const struct cache_entry
*ce
, struct stat
*st
,
366 unsigned int options
)
368 unsigned int changed
;
369 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
370 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
371 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
372 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
374 if (!ignore_fsmonitor
)
375 refresh_fsmonitor(istate
);
377 * If it's marked as always valid in the index, it's
378 * valid whatever the checked-out copy says.
380 * skip-worktree has the same effect with higher precedence
382 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
))
384 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
386 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
))
390 * Intent-to-add entries have not been added, so the index entry
391 * by definition never matches what is in the work tree until it
392 * actually gets added.
394 if (ce_intent_to_add(ce
))
395 return DATA_CHANGED
| TYPE_CHANGED
| MODE_CHANGED
;
397 changed
= ce_match_stat_basic(ce
, st
);
400 * Within 1 second of this sequence:
401 * echo xyzzy >file && git-update-index --add file
402 * running this command:
404 * would give a falsely clean cache entry. The mtime and
405 * length match the cache, and other stat fields do not change.
407 * We could detect this at update-index time (the cache entry
408 * being registered/updated records the same time as "now")
409 * and delay the return from git-update-index, but that would
410 * effectively mean we can make at most one commit per second,
411 * which is not acceptable. Instead, we check cache entries
412 * whose mtime are the same as the index file timestamp more
413 * carefully than others.
415 if (!changed
&& is_racy_timestamp(istate
, ce
)) {
416 if (assume_racy_is_modified
)
417 changed
|= DATA_CHANGED
;
419 changed
|= ce_modified_check_fs(istate
, ce
, st
);
425 int ie_modified(struct index_state
*istate
,
426 const struct cache_entry
*ce
,
427 struct stat
*st
, unsigned int options
)
429 int changed
, changed_fs
;
431 changed
= ie_match_stat(istate
, ce
, st
, options
);
435 * If the mode or type has changed, there's no point in trying
436 * to refresh the entry - it's not going to match
438 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
442 * Immediately after read-tree or update-index --cacheinfo,
443 * the length field is zero, as we have never even read the
444 * lstat(2) information once, and we cannot trust DATA_CHANGED
445 * returned by ie_match_stat() which in turn was returned by
446 * ce_match_stat_basic() to signal that the filesize of the
447 * blob changed. We have to actually go to the filesystem to
448 * see if the contents match, and if so, should answer "unchanged".
450 * The logic does not apply to gitlinks, as ce_match_stat_basic()
451 * already has checked the actual HEAD from the filesystem in the
452 * subproject. If ie_match_stat() already said it is different,
453 * then we know it is.
455 if ((changed
& DATA_CHANGED
) &&
456 (S_ISGITLINK(ce
->ce_mode
) || ce
->ce_stat_data
.sd_size
!= 0))
459 changed_fs
= ce_modified_check_fs(istate
, ce
, st
);
461 return changed
| changed_fs
;
465 int base_name_compare(const char *name1
, int len1
, int mode1
,
466 const char *name2
, int len2
, int mode2
)
468 unsigned char c1
, c2
;
469 int len
= len1
< len2
? len1
: len2
;
472 cmp
= memcmp(name1
, name2
, len
);
477 if (!c1
&& S_ISDIR(mode1
))
479 if (!c2
&& S_ISDIR(mode2
))
481 return (c1
< c2
) ? -1 : (c1
> c2
) ? 1 : 0;
485 * df_name_compare() is identical to base_name_compare(), except it
486 * compares conflicting directory/file entries as equal. Note that
487 * while a directory name compares as equal to a regular file, they
488 * then individually compare _differently_ to a filename that has
489 * a dot after the basename (because '\0' < '.' < '/').
491 * This is used by routines that want to traverse the git namespace
492 * but then handle conflicting entries together when possible.
494 int df_name_compare(const char *name1
, int len1
, int mode1
,
495 const char *name2
, int len2
, int mode2
)
497 int len
= len1
< len2
? len1
: len2
, cmp
;
498 unsigned char c1
, c2
;
500 cmp
= memcmp(name1
, name2
, len
);
503 /* Directories and files compare equal (same length, same name) */
507 if (!c1
&& S_ISDIR(mode1
))
510 if (!c2
&& S_ISDIR(mode2
))
512 if (c1
== '/' && !c2
)
514 if (c2
== '/' && !c1
)
519 int name_compare(const char *name1
, size_t len1
, const char *name2
, size_t len2
)
521 size_t min_len
= (len1
< len2
) ? len1
: len2
;
522 int cmp
= memcmp(name1
, name2
, min_len
);
532 int cache_name_stage_compare(const char *name1
, int len1
, int stage1
, const char *name2
, int len2
, int stage2
)
536 cmp
= name_compare(name1
, len1
, name2
, len2
);
547 static int index_name_stage_pos(const struct index_state
*istate
, const char *name
, int namelen
, int stage
)
552 last
= istate
->cache_nr
;
553 while (last
> first
) {
554 int next
= first
+ ((last
- first
) >> 1);
555 struct cache_entry
*ce
= istate
->cache
[next
];
556 int cmp
= cache_name_stage_compare(name
, namelen
, stage
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
568 int index_name_pos(const struct index_state
*istate
, const char *name
, int namelen
)
570 return index_name_stage_pos(istate
, name
, namelen
, 0);
573 int remove_index_entry_at(struct index_state
*istate
, int pos
)
575 struct cache_entry
*ce
= istate
->cache
[pos
];
577 record_resolve_undo(istate
, ce
);
578 remove_name_hash(istate
, ce
);
579 save_or_free_index_entry(istate
, ce
);
580 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
582 if (pos
>= istate
->cache_nr
)
584 MOVE_ARRAY(istate
->cache
+ pos
, istate
->cache
+ pos
+ 1,
585 istate
->cache_nr
- pos
);
590 * Remove all cache entries marked for removal, that is where
591 * CE_REMOVE is set in ce_flags. This is much more effective than
592 * calling remove_index_entry_at() for each entry to be removed.
594 void remove_marked_cache_entries(struct index_state
*istate
, int invalidate
)
596 struct cache_entry
**ce_array
= istate
->cache
;
599 for (i
= j
= 0; i
< istate
->cache_nr
; i
++) {
600 if (ce_array
[i
]->ce_flags
& CE_REMOVE
) {
602 cache_tree_invalidate_path(istate
,
604 untracked_cache_remove_from_index(istate
,
607 remove_name_hash(istate
, ce_array
[i
]);
608 save_or_free_index_entry(istate
, ce_array
[i
]);
611 ce_array
[j
++] = ce_array
[i
];
613 if (j
== istate
->cache_nr
)
615 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
616 istate
->cache_nr
= j
;
619 int remove_file_from_index(struct index_state
*istate
, const char *path
)
621 int pos
= index_name_pos(istate
, path
, strlen(path
));
624 cache_tree_invalidate_path(istate
, path
);
625 untracked_cache_remove_from_index(istate
, path
);
626 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
627 remove_index_entry_at(istate
, pos
);
631 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
633 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
636 static int index_name_pos_also_unmerged(struct index_state
*istate
,
637 const char *path
, int namelen
)
639 int pos
= index_name_pos(istate
, path
, namelen
);
640 struct cache_entry
*ce
;
645 /* maybe unmerged? */
647 if (pos
>= istate
->cache_nr
||
648 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
651 /* order of preference: stage 2, 1, 3 */
652 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
653 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
654 !compare_name(ce
, path
, namelen
))
659 static int different_name(struct cache_entry
*ce
, struct cache_entry
*alias
)
661 int len
= ce_namelen(ce
);
662 return ce_namelen(alias
) != len
|| memcmp(ce
->name
, alias
->name
, len
);
666 * If we add a filename that aliases in the cache, we will use the
667 * name that we already have - but we don't want to update the same
668 * alias twice, because that implies that there were actually two
669 * different files with aliasing names!
671 * So we use the CE_ADDED flag to verify that the alias was an old
672 * one before we accept it as
674 static struct cache_entry
*create_alias_ce(struct index_state
*istate
,
675 struct cache_entry
*ce
,
676 struct cache_entry
*alias
)
679 struct cache_entry
*new_entry
;
681 if (alias
->ce_flags
& CE_ADDED
)
682 die(_("will not add file alias '%s' ('%s' already exists in index)"),
683 ce
->name
, alias
->name
);
685 /* Ok, create the new entry using the name of the existing alias */
686 len
= ce_namelen(alias
);
687 new_entry
= make_empty_cache_entry(istate
, len
);
688 memcpy(new_entry
->name
, alias
->name
, len
);
689 copy_cache_entry(new_entry
, ce
);
690 save_or_free_index_entry(istate
, ce
);
694 void set_object_name_for_intent_to_add_entry(struct cache_entry
*ce
)
696 struct object_id oid
;
697 if (write_object_file("", 0, blob_type
, &oid
))
698 die(_("cannot create an empty blob in the object database"));
699 oidcpy(&ce
->oid
, &oid
);
702 int add_to_index(struct index_state
*istate
, const char *path
, struct stat
*st
, int flags
)
704 int namelen
, was_same
;
705 mode_t st_mode
= st
->st_mode
;
706 struct cache_entry
*ce
, *alias
= NULL
;
707 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_IGNORE_SKIP_WORKTREE
|CE_MATCH_RACY_IS_DIRTY
;
708 int verbose
= flags
& (ADD_CACHE_VERBOSE
| ADD_CACHE_PRETEND
);
709 int pretend
= flags
& ADD_CACHE_PRETEND
;
710 int intent_only
= flags
& ADD_CACHE_INTENT
;
711 int add_option
= (ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
|
712 (intent_only
? ADD_CACHE_NEW_ONLY
: 0));
713 int hash_flags
= HASH_WRITE_OBJECT
;
714 struct object_id oid
;
716 if (flags
& ADD_CACHE_RENORMALIZE
)
717 hash_flags
|= HASH_RENORMALIZE
;
719 if (!S_ISREG(st_mode
) && !S_ISLNK(st_mode
) && !S_ISDIR(st_mode
))
720 return error(_("%s: can only add regular files, symbolic links or git-directories"), path
);
722 namelen
= strlen(path
);
723 if (S_ISDIR(st_mode
)) {
724 if (resolve_gitlink_ref(path
, "HEAD", &oid
) < 0)
725 return error(_("'%s' does not have a commit checked out"), path
);
726 while (namelen
&& path
[namelen
-1] == '/')
729 ce
= make_empty_cache_entry(istate
, namelen
);
730 memcpy(ce
->name
, path
, namelen
);
731 ce
->ce_namelen
= namelen
;
733 fill_stat_cache_info(istate
, ce
, st
);
735 ce
->ce_flags
|= CE_INTENT_TO_ADD
;
738 if (trust_executable_bit
&& has_symlinks
) {
739 ce
->ce_mode
= create_ce_mode(st_mode
);
741 /* If there is an existing entry, pick the mode bits and type
742 * from it, otherwise assume unexecutable regular file.
744 struct cache_entry
*ent
;
745 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
747 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
748 ce
->ce_mode
= ce_mode_from_stat(ent
, st_mode
);
751 /* When core.ignorecase=true, determine if a directory of the same name but differing
752 * case already exists within the Git repository. If it does, ensure the directory
753 * case of the file being added to the repository matches (is folded into) the existing
754 * entry's directory case.
757 adjust_dirname_case(istate
, ce
->name
);
759 if (!(flags
& ADD_CACHE_RENORMALIZE
)) {
760 alias
= index_file_exists(istate
, ce
->name
,
761 ce_namelen(ce
), ignore_case
);
764 !ie_match_stat(istate
, alias
, st
, ce_option
)) {
765 /* Nothing changed, really */
766 if (!S_ISGITLINK(alias
->ce_mode
))
767 ce_mark_uptodate(alias
);
768 alias
->ce_flags
|= CE_ADDED
;
770 discard_cache_entry(ce
);
775 if (index_path(istate
, &ce
->oid
, path
, st
, hash_flags
)) {
776 discard_cache_entry(ce
);
777 return error(_("unable to index file '%s'"), path
);
780 set_object_name_for_intent_to_add_entry(ce
);
782 if (ignore_case
&& alias
&& different_name(ce
, alias
))
783 ce
= create_alias_ce(istate
, ce
, alias
);
784 ce
->ce_flags
|= CE_ADDED
;
786 /* It was suspected to be racily clean, but it turns out to be Ok */
789 oideq(&alias
->oid
, &ce
->oid
) &&
790 ce
->ce_mode
== alias
->ce_mode
);
793 discard_cache_entry(ce
);
794 else if (add_index_entry(istate
, ce
, add_option
)) {
795 discard_cache_entry(ce
);
796 return error(_("unable to add '%s' to index"), path
);
798 if (verbose
&& !was_same
)
799 printf("add '%s'\n", path
);
803 int add_file_to_index(struct index_state
*istate
, const char *path
, int flags
)
806 if (lstat(path
, &st
))
807 die_errno(_("unable to stat '%s'"), path
);
808 return add_to_index(istate
, path
, &st
, flags
);
811 struct cache_entry
*make_empty_cache_entry(struct index_state
*istate
, size_t len
)
813 return mem_pool__ce_calloc(find_mem_pool(istate
), len
);
816 struct cache_entry
*make_empty_transient_cache_entry(size_t len
)
818 return xcalloc(1, cache_entry_size(len
));
821 struct cache_entry
*make_cache_entry(struct index_state
*istate
,
823 const struct object_id
*oid
,
826 unsigned int refresh_options
)
828 struct cache_entry
*ce
, *ret
;
831 if (!verify_path(path
, mode
)) {
832 error(_("invalid path '%s'"), path
);
837 ce
= make_empty_cache_entry(istate
, len
);
839 oidcpy(&ce
->oid
, oid
);
840 memcpy(ce
->name
, path
, len
);
841 ce
->ce_flags
= create_ce_flags(stage
);
842 ce
->ce_namelen
= len
;
843 ce
->ce_mode
= create_ce_mode(mode
);
845 ret
= refresh_cache_entry(istate
, ce
, refresh_options
);
847 discard_cache_entry(ce
);
851 struct cache_entry
*make_transient_cache_entry(unsigned int mode
, const struct object_id
*oid
,
852 const char *path
, int stage
)
854 struct cache_entry
*ce
;
857 if (!verify_path(path
, mode
)) {
858 error(_("invalid path '%s'"), path
);
863 ce
= make_empty_transient_cache_entry(len
);
865 oidcpy(&ce
->oid
, oid
);
866 memcpy(ce
->name
, path
, len
);
867 ce
->ce_flags
= create_ce_flags(stage
);
868 ce
->ce_namelen
= len
;
869 ce
->ce_mode
= create_ce_mode(mode
);
875 * Chmod an index entry with either +x or -x.
877 * Returns -1 if the chmod for the particular cache entry failed (if it's
878 * not a regular file), -2 if an invalid flip argument is passed in, 0
881 int chmod_index_entry(struct index_state
*istate
, struct cache_entry
*ce
,
884 if (!S_ISREG(ce
->ce_mode
))
891 ce
->ce_mode
&= ~0111;
896 cache_tree_invalidate_path(istate
, ce
->name
);
897 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
898 mark_fsmonitor_invalid(istate
, ce
);
899 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
904 int ce_same_name(const struct cache_entry
*a
, const struct cache_entry
*b
)
906 int len
= ce_namelen(a
);
907 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
911 * We fundamentally don't like some paths: we don't want
912 * dot or dot-dot anywhere, and for obvious reasons don't
913 * want to recurse into ".git" either.
915 * Also, we don't want double slashes or slashes at the
916 * end that can make pathnames ambiguous.
918 static int verify_dotfile(const char *rest
, unsigned mode
)
921 * The first character was '.', but that
922 * has already been discarded, we now test
926 /* "." is not allowed */
927 if (*rest
== '\0' || is_dir_sep(*rest
))
932 * ".git" followed by NUL or slash is bad. Note that we match
933 * case-insensitively here, even if ignore_case is not set.
934 * This outlaws ".GIT" everywhere out of an abundance of caution,
935 * since there's really no good reason to allow it.
937 * Once we've seen ".git", we can also find ".gitmodules", etc (also
938 * case-insensitively).
942 if (rest
[1] != 'i' && rest
[1] != 'I')
944 if (rest
[2] != 't' && rest
[2] != 'T')
946 if (rest
[3] == '\0' || is_dir_sep(rest
[3]))
950 if (skip_iprefix(rest
, "modules", &rest
) &&
951 (*rest
== '\0' || is_dir_sep(*rest
)))
956 if (rest
[1] == '\0' || is_dir_sep(rest
[1]))
962 int verify_path(const char *path
, unsigned mode
)
966 if (has_dos_drive_prefix(path
))
969 if (!is_valid_path(path
))
980 if (is_hfs_dotgit(path
))
983 if (is_hfs_dotgitmodules(path
))
988 #ifdef GIT_WINDOWS_NATIVE
992 if (is_ntfs_dotgit(path
))
995 if (is_ntfs_dotgitmodules(path
))
1001 if ((c
== '.' && !verify_dotfile(path
, mode
)) ||
1002 is_dir_sep(c
) || c
== '\0')
1004 } else if (c
== '\\' && protect_ntfs
) {
1005 if (is_ntfs_dotgit(path
))
1007 if (S_ISLNK(mode
)) {
1008 if (is_ntfs_dotgitmodules(path
))
1018 * Do we have another file that has the beginning components being a
1019 * proper superset of the name we're trying to add?
1021 static int has_file_name(struct index_state
*istate
,
1022 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
1025 int len
= ce_namelen(ce
);
1026 int stage
= ce_stage(ce
);
1027 const char *name
= ce
->name
;
1029 while (pos
< istate
->cache_nr
) {
1030 struct cache_entry
*p
= istate
->cache
[pos
++];
1032 if (len
>= ce_namelen(p
))
1034 if (memcmp(name
, p
->name
, len
))
1036 if (ce_stage(p
) != stage
)
1038 if (p
->name
[len
] != '/')
1040 if (p
->ce_flags
& CE_REMOVE
)
1045 remove_index_entry_at(istate
, --pos
);
1052 * Like strcmp(), but also return the offset of the first change.
1053 * If strings are equal, return the length.
1055 int strcmp_offset(const char *s1
, const char *s2
, size_t *first_change
)
1060 return strcmp(s1
, s2
);
1062 for (k
= 0; s1
[k
] == s2
[k
]; k
++)
1067 return (unsigned char)s1
[k
] - (unsigned char)s2
[k
];
1071 * Do we have another file with a pathname that is a proper
1072 * subset of the name we're trying to add?
1074 * That is, is there another file in the index with a path
1075 * that matches a sub-directory in the given entry?
1077 static int has_dir_name(struct index_state
*istate
,
1078 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
1081 int stage
= ce_stage(ce
);
1082 const char *name
= ce
->name
;
1083 const char *slash
= name
+ ce_namelen(ce
);
1088 * We are frequently called during an iteration on a sorted
1089 * list of pathnames and while building a new index. Therefore,
1090 * there is a high probability that this entry will eventually
1091 * be appended to the index, rather than inserted in the middle.
1092 * If we can confirm that, we can avoid binary searches on the
1093 * components of the pathname.
1095 * Compare the entry's full path with the last path in the index.
1097 if (istate
->cache_nr
> 0) {
1098 cmp_last
= strcmp_offset(name
,
1099 istate
->cache
[istate
->cache_nr
- 1]->name
,
1102 if (len_eq_last
== 0) {
1104 * The entry sorts AFTER the last one in the
1105 * index and their paths have no common prefix,
1106 * so there cannot be a F/D conflict.
1111 * The entry sorts AFTER the last one in the
1112 * index, but has a common prefix. Fall through
1113 * to the loop below to disect the entry's path
1114 * and see where the difference is.
1117 } else if (cmp_last
== 0) {
1119 * The entry exactly matches the last one in the
1120 * index, but because of multiple stage and CE_REMOVE
1121 * items, we fall through and let the regular search
1131 if (*--slash
== '/')
1133 if (slash
<= ce
->name
)
1140 * (len + 1) is a directory boundary (including
1141 * the trailing slash). And since the loop is
1142 * decrementing "slash", the first iteration is
1143 * the longest directory prefix; subsequent
1144 * iterations consider parent directories.
1147 if (len
+ 1 <= len_eq_last
) {
1149 * The directory prefix (including the trailing
1150 * slash) also appears as a prefix in the last
1151 * entry, so the remainder cannot collide (because
1152 * strcmp said the whole path was greater).
1157 * LT: last: xxx/file_A
1163 if (len
> len_eq_last
) {
1165 * This part of the directory prefix (excluding
1166 * the trailing slash) is longer than the known
1167 * equal portions, so this sub-directory cannot
1168 * collide with a file.
1177 * This is a possible collision. Fall through and
1178 * let the regular search code handle it.
1185 pos
= index_name_stage_pos(istate
, name
, len
, stage
);
1188 * Found one, but not so fast. This could
1189 * be a marker that says "I was here, but
1190 * I am being removed". Such an entry is
1191 * not a part of the resulting tree, and
1192 * it is Ok to have a directory at the same
1195 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
1199 remove_index_entry_at(istate
, pos
);
1207 * Trivial optimization: if we find an entry that
1208 * already matches the sub-directory, then we know
1209 * we're ok, and we can exit.
1211 while (pos
< istate
->cache_nr
) {
1212 struct cache_entry
*p
= istate
->cache
[pos
];
1213 if ((ce_namelen(p
) <= len
) ||
1214 (p
->name
[len
] != '/') ||
1215 memcmp(p
->name
, name
, len
))
1216 break; /* not our subdirectory */
1217 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
1219 * p is at the same stage as our entry, and
1220 * is a subdirectory of what we are looking
1221 * at, so we cannot have conflicts at our
1222 * level or anything shorter.
1231 /* We may be in a situation where we already have path/file and path
1232 * is being added, or we already have path and path/file is being
1233 * added. Either one would result in a nonsense tree that has path
1234 * twice when git-write-tree tries to write it out. Prevent it.
1236 * If ok-to-replace is specified, we remove the conflicting entries
1237 * from the cache so the caller should recompute the insert position.
1238 * When this happens, we return non-zero.
1240 static int check_file_directory_conflict(struct index_state
*istate
,
1241 const struct cache_entry
*ce
,
1242 int pos
, int ok_to_replace
)
1247 * When ce is an "I am going away" entry, we allow it to be added
1249 if (ce
->ce_flags
& CE_REMOVE
)
1253 * We check if the path is a sub-path of a subsequent pathname
1254 * first, since removing those will not change the position
1257 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
1260 * Then check if the path might have a clashing sub-directory
1263 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
1266 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1269 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
1270 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
1271 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
1272 int new_only
= option
& ADD_CACHE_NEW_ONLY
;
1274 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1275 cache_tree_invalidate_path(istate
, ce
->name
);
1278 * If this entry's path sorts after the last entry in the index,
1279 * we can avoid searching for it.
1281 if (istate
->cache_nr
> 0 &&
1282 strcmp(ce
->name
, istate
->cache
[istate
->cache_nr
- 1]->name
) > 0)
1283 pos
= index_pos_to_insert_pos(istate
->cache_nr
);
1285 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1287 /* existing match? Just replace it. */
1290 replace_index_entry(istate
, pos
, ce
);
1295 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1296 untracked_cache_add_to_index(istate
, ce
->name
);
1299 * Inserting a merged entry ("stage 0") into the index
1300 * will always replace all non-merged entries..
1302 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
1303 while (ce_same_name(istate
->cache
[pos
], ce
)) {
1305 if (!remove_index_entry_at(istate
, pos
))
1312 if (!verify_path(ce
->name
, ce
->ce_mode
))
1313 return error(_("invalid path '%s'"), ce
->name
);
1315 if (!skip_df_check
&&
1316 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
1318 return error(_("'%s' appears as both a file and as a directory"),
1320 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1326 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1330 if (option
& ADD_CACHE_JUST_APPEND
)
1331 pos
= istate
->cache_nr
;
1334 ret
= add_index_entry_with_check(istate
, ce
, option
);
1340 /* Make sure the array is big enough .. */
1341 ALLOC_GROW(istate
->cache
, istate
->cache_nr
+ 1, istate
->cache_alloc
);
1345 if (istate
->cache_nr
> pos
+ 1)
1346 MOVE_ARRAY(istate
->cache
+ pos
+ 1, istate
->cache
+ pos
,
1347 istate
->cache_nr
- pos
- 1);
1348 set_index_entry(istate
, pos
, ce
);
1349 istate
->cache_changed
|= CE_ENTRY_ADDED
;
1354 * "refresh" does not calculate a new sha1 file or bring the
1355 * cache up-to-date for mode/content changes. But what it
1356 * _does_ do is to "re-match" the stat information of a file
1357 * with the cache, so that you can refresh the cache for a
1358 * file that hasn't been changed but where the stat entry is
1361 * For example, you'd want to do this after doing a "git-read-tree",
1362 * to link up the stat cache details with the proper files.
1364 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
1365 struct cache_entry
*ce
,
1366 unsigned int options
, int *err
,
1372 struct cache_entry
*updated
;
1374 int refresh
= options
& CE_MATCH_REFRESH
;
1375 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
1376 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
1377 int ignore_missing
= options
& CE_MATCH_IGNORE_MISSING
;
1378 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
1380 if (!refresh
|| ce_uptodate(ce
))
1383 if (!ignore_fsmonitor
)
1384 refresh_fsmonitor(istate
);
1386 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1387 * that the change to the work tree does not matter and told
1390 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
)) {
1391 ce_mark_uptodate(ce
);
1394 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
)) {
1395 ce_mark_uptodate(ce
);
1398 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
)) {
1399 ce_mark_uptodate(ce
);
1403 if (has_symlink_leading_path(ce
->name
, ce_namelen(ce
))) {
1413 if (lstat(ce
->name
, &st
) < 0) {
1414 if (ignore_missing
&& errno
== ENOENT
)
1421 changed
= ie_match_stat(istate
, ce
, &st
, options
);
1423 *changed_ret
= changed
;
1426 * The path is unchanged. If we were told to ignore
1427 * valid bit, then we did the actual stat check and
1428 * found that the entry is unmodified. If the entry
1429 * is not marked VALID, this is the place to mark it
1430 * valid again, under "assume unchanged" mode.
1432 if (ignore_valid
&& assume_unchanged
&&
1433 !(ce
->ce_flags
& CE_VALID
))
1434 ; /* mark this one VALID again */
1437 * We do not mark the index itself "modified"
1438 * because CE_UPTODATE flag is in-core only;
1439 * we are not going to write this change out.
1441 if (!S_ISGITLINK(ce
->ce_mode
)) {
1442 ce_mark_uptodate(ce
);
1443 mark_fsmonitor_valid(istate
, ce
);
1451 if (ie_modified(istate
, ce
, &st
, options
)) {
1457 updated
= make_empty_cache_entry(istate
, ce_namelen(ce
));
1458 copy_cache_entry(updated
, ce
);
1459 memcpy(updated
->name
, ce
->name
, ce
->ce_namelen
+ 1);
1460 fill_stat_cache_info(istate
, updated
, &st
);
1462 * If ignore_valid is not set, we should leave CE_VALID bit
1463 * alone. Otherwise, paths marked with --no-assume-unchanged
1464 * (i.e. things to be edited) will reacquire CE_VALID bit
1465 * automatically, which is not really what we want.
1467 if (!ignore_valid
&& assume_unchanged
&&
1468 !(ce
->ce_flags
& CE_VALID
))
1469 updated
->ce_flags
&= ~CE_VALID
;
1471 /* istate->cache_changed is updated in the caller */
1475 static void show_file(const char * fmt
, const char * name
, int in_porcelain
,
1476 int * first
, const char *header_msg
)
1478 if (in_porcelain
&& *first
&& header_msg
) {
1479 printf("%s\n", header_msg
);
1485 int repo_refresh_and_write_index(struct repository
*repo
,
1486 unsigned int refresh_flags
,
1487 unsigned int write_flags
,
1489 const struct pathspec
*pathspec
,
1490 char *seen
, const char *header_msg
)
1492 struct lock_file lock_file
= LOCK_INIT
;
1495 fd
= repo_hold_locked_index(repo
, &lock_file
, 0);
1496 if (!gentle
&& fd
< 0)
1498 if (refresh_index(repo
->index
, refresh_flags
, pathspec
, seen
, header_msg
))
1500 if (0 <= fd
&& write_locked_index(repo
->index
, &lock_file
, COMMIT_LOCK
| write_flags
))
1506 int refresh_index(struct index_state
*istate
, unsigned int flags
,
1507 const struct pathspec
*pathspec
,
1508 char *seen
, const char *header_msg
)
1512 int really
= (flags
& REFRESH_REALLY
) != 0;
1513 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
1514 int quiet
= (flags
& REFRESH_QUIET
) != 0;
1515 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
1516 int ignore_submodules
= (flags
& REFRESH_IGNORE_SUBMODULES
) != 0;
1518 int in_porcelain
= (flags
& REFRESH_IN_PORCELAIN
);
1519 unsigned int options
= (CE_MATCH_REFRESH
|
1520 (really
? CE_MATCH_IGNORE_VALID
: 0) |
1521 (not_new
? CE_MATCH_IGNORE_MISSING
: 0));
1522 const char *modified_fmt
;
1523 const char *deleted_fmt
;
1524 const char *typechange_fmt
;
1525 const char *added_fmt
;
1526 const char *unmerged_fmt
;
1527 struct progress
*progress
= NULL
;
1528 int t2_sum_lstat
= 0;
1529 int t2_sum_scan
= 0;
1531 if (flags
& REFRESH_PROGRESS
&& isatty(2))
1532 progress
= start_delayed_progress(_("Refresh index"),
1535 trace_performance_enter();
1536 modified_fmt
= in_porcelain
? "M\t%s\n" : "%s: needs update\n";
1537 deleted_fmt
= in_porcelain
? "D\t%s\n" : "%s: needs update\n";
1538 typechange_fmt
= in_porcelain
? "T\t%s\n" : "%s: needs update\n";
1539 added_fmt
= in_porcelain
? "A\t%s\n" : "%s: needs update\n";
1540 unmerged_fmt
= in_porcelain
? "U\t%s\n" : "%s: needs merge\n";
1542 * Use the multi-threaded preload_index() to refresh most of the
1543 * cache entries quickly then in the single threaded loop below,
1544 * we only have to do the special cases that are left.
1546 preload_index(istate
, pathspec
, 0);
1547 trace2_region_enter("index", "refresh", NULL
);
1548 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1549 struct cache_entry
*ce
, *new_entry
;
1550 int cache_errno
= 0;
1553 int t2_did_lstat
= 0;
1554 int t2_did_scan
= 0;
1556 ce
= istate
->cache
[i
];
1557 if (ignore_submodules
&& S_ISGITLINK(ce
->ce_mode
))
1560 if (pathspec
&& !ce_path_match(istate
, ce
, pathspec
, seen
))
1564 while ((i
< istate
->cache_nr
) &&
1565 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
1571 show_file(unmerged_fmt
, ce
->name
, in_porcelain
,
1572 &first
, header_msg
);
1580 new_entry
= refresh_cache_ent(istate
, ce
, options
,
1581 &cache_errno
, &changed
,
1582 &t2_did_lstat
, &t2_did_scan
);
1583 t2_sum_lstat
+= t2_did_lstat
;
1584 t2_sum_scan
+= t2_did_scan
;
1585 if (new_entry
== ce
)
1588 display_progress(progress
, i
);
1592 if (really
&& cache_errno
== EINVAL
) {
1593 /* If we are doing --really-refresh that
1594 * means the index is not valid anymore.
1596 ce
->ce_flags
&= ~CE_VALID
;
1597 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
1598 mark_fsmonitor_invalid(istate
, ce
);
1599 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
1604 if (cache_errno
== ENOENT
)
1606 else if (ce_intent_to_add(ce
))
1607 fmt
= added_fmt
; /* must be before other checks */
1608 else if (changed
& TYPE_CHANGED
)
1609 fmt
= typechange_fmt
;
1613 ce
->name
, in_porcelain
, &first
, header_msg
);
1618 replace_index_entry(istate
, i
, new_entry
);
1620 trace2_data_intmax("index", NULL
, "refresh/sum_lstat", t2_sum_lstat
);
1621 trace2_data_intmax("index", NULL
, "refresh/sum_scan", t2_sum_scan
);
1622 trace2_region_leave("index", "refresh", NULL
);
1624 display_progress(progress
, istate
->cache_nr
);
1625 stop_progress(&progress
);
1627 trace_performance_leave("refresh index");
1631 struct cache_entry
*refresh_cache_entry(struct index_state
*istate
,
1632 struct cache_entry
*ce
,
1633 unsigned int options
)
1635 return refresh_cache_ent(istate
, ce
, options
, NULL
, NULL
, NULL
, NULL
);
1639 /*****************************************************************
1641 *****************************************************************/
1643 #define INDEX_FORMAT_DEFAULT 3
1645 static unsigned int get_index_format_default(struct repository
*r
)
1647 char *envversion
= getenv("GIT_INDEX_VERSION");
1649 unsigned int version
= INDEX_FORMAT_DEFAULT
;
1652 prepare_repo_settings(r
);
1654 if (r
->settings
.index_version
>= 0)
1655 version
= r
->settings
.index_version
;
1656 if (version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1657 warning(_("index.version set, but the value is invalid.\n"
1658 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1659 return INDEX_FORMAT_DEFAULT
;
1664 version
= strtoul(envversion
, &endp
, 10);
1666 version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1667 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1668 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1669 version
= INDEX_FORMAT_DEFAULT
;
1675 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1676 * Again - this is just a (very strong in practice) heuristic that
1677 * the inode hasn't changed.
1679 * We save the fields in big-endian order to allow using the
1680 * index file over NFS transparently.
1682 struct ondisk_cache_entry
{
1683 struct cache_time ctime
;
1684 struct cache_time mtime
;
1692 * unsigned char hash[hashsz];
1694 * if (flags & CE_EXTENDED)
1697 unsigned char data
[GIT_MAX_RAWSZ
+ 2 * sizeof(uint16_t)];
1698 char name
[FLEX_ARRAY
];
1701 /* These are only used for v3 or lower */
1702 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1703 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,data) + (len) + 8) & ~7)
1704 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1705 #define ondisk_data_size(flags, len) (the_hash_algo->rawsz + \
1706 ((flags & CE_EXTENDED) ? 2 : 1) * sizeof(uint16_t) + len)
1707 #define ondisk_data_size_max(len) (ondisk_data_size(CE_EXTENDED, len))
1708 #define ondisk_ce_size(ce) (ondisk_cache_entry_size(ondisk_data_size((ce)->ce_flags, ce_namelen(ce))))
1710 /* Allow fsck to force verification of the index checksum. */
1711 int verify_index_checksum
;
1713 /* Allow fsck to force verification of the cache entry order. */
1714 int verify_ce_order
;
1716 static int verify_hdr(const struct cache_header
*hdr
, unsigned long size
)
1719 unsigned char hash
[GIT_MAX_RAWSZ
];
1722 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
1723 return error(_("bad signature 0x%08x"), hdr
->hdr_signature
);
1724 hdr_version
= ntohl(hdr
->hdr_version
);
1725 if (hdr_version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< hdr_version
)
1726 return error(_("bad index version %d"), hdr_version
);
1728 if (!verify_index_checksum
)
1731 the_hash_algo
->init_fn(&c
);
1732 the_hash_algo
->update_fn(&c
, hdr
, size
- the_hash_algo
->rawsz
);
1733 the_hash_algo
->final_fn(hash
, &c
);
1734 if (!hasheq(hash
, (unsigned char *)hdr
+ size
- the_hash_algo
->rawsz
))
1735 return error(_("bad index file sha1 signature"));
1739 static int read_index_extension(struct index_state
*istate
,
1740 const char *ext
, const char *data
, unsigned long sz
)
1742 switch (CACHE_EXT(ext
)) {
1743 case CACHE_EXT_TREE
:
1744 istate
->cache_tree
= cache_tree_read(data
, sz
);
1746 case CACHE_EXT_RESOLVE_UNDO
:
1747 istate
->resolve_undo
= resolve_undo_read(data
, sz
);
1749 case CACHE_EXT_LINK
:
1750 if (read_link_extension(istate
, data
, sz
))
1753 case CACHE_EXT_UNTRACKED
:
1754 istate
->untracked
= read_untracked_extension(data
, sz
);
1756 case CACHE_EXT_FSMONITOR
:
1757 read_fsmonitor_extension(istate
, data
, sz
);
1759 case CACHE_EXT_ENDOFINDEXENTRIES
:
1760 case CACHE_EXT_INDEXENTRYOFFSETTABLE
:
1761 /* already handled in do_read_index() */
1764 if (*ext
< 'A' || 'Z' < *ext
)
1765 return error(_("index uses %.4s extension, which we do not understand"),
1767 fprintf_ln(stderr
, _("ignoring %.4s extension"), ext
);
1773 static struct cache_entry
*create_from_disk(struct mem_pool
*ce_mem_pool
,
1774 unsigned int version
,
1775 struct ondisk_cache_entry
*ondisk
,
1776 unsigned long *ent_size
,
1777 const struct cache_entry
*previous_ce
)
1779 struct cache_entry
*ce
;
1782 const unsigned hashsz
= the_hash_algo
->rawsz
;
1783 const uint16_t *flagsp
= (const uint16_t *)(ondisk
->data
+ hashsz
);
1785 size_t copy_len
= 0;
1787 * Adjacent cache entries tend to share the leading paths, so it makes
1788 * sense to only store the differences in later entries. In the v4
1789 * on-disk format of the index, each on-disk cache entry stores the
1790 * number of bytes to be stripped from the end of the previous name,
1791 * and the bytes to append to the result, to come up with its name.
1793 int expand_name_field
= version
== 4;
1795 /* On-disk flags are just 16 bits */
1796 flags
= get_be16(flagsp
);
1797 len
= flags
& CE_NAMEMASK
;
1799 if (flags
& CE_EXTENDED
) {
1801 extended_flags
= get_be16(flagsp
+ 1) << 16;
1802 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1803 if (extended_flags
& ~CE_EXTENDED_FLAGS
)
1804 die(_("unknown index entry format 0x%08x"), extended_flags
);
1805 flags
|= extended_flags
;
1806 name
= (const char *)(flagsp
+ 2);
1809 name
= (const char *)(flagsp
+ 1);
1811 if (expand_name_field
) {
1812 const unsigned char *cp
= (const unsigned char *)name
;
1813 size_t strip_len
, previous_len
;
1815 /* If we're at the beginning of a block, ignore the previous name */
1816 strip_len
= decode_varint(&cp
);
1818 previous_len
= previous_ce
->ce_namelen
;
1819 if (previous_len
< strip_len
)
1820 die(_("malformed name field in the index, near path '%s'"),
1822 copy_len
= previous_len
- strip_len
;
1824 name
= (const char *)cp
;
1827 if (len
== CE_NAMEMASK
) {
1829 if (expand_name_field
)
1833 ce
= mem_pool__ce_alloc(ce_mem_pool
, len
);
1835 ce
->ce_stat_data
.sd_ctime
.sec
= get_be32(&ondisk
->ctime
.sec
);
1836 ce
->ce_stat_data
.sd_mtime
.sec
= get_be32(&ondisk
->mtime
.sec
);
1837 ce
->ce_stat_data
.sd_ctime
.nsec
= get_be32(&ondisk
->ctime
.nsec
);
1838 ce
->ce_stat_data
.sd_mtime
.nsec
= get_be32(&ondisk
->mtime
.nsec
);
1839 ce
->ce_stat_data
.sd_dev
= get_be32(&ondisk
->dev
);
1840 ce
->ce_stat_data
.sd_ino
= get_be32(&ondisk
->ino
);
1841 ce
->ce_mode
= get_be32(&ondisk
->mode
);
1842 ce
->ce_stat_data
.sd_uid
= get_be32(&ondisk
->uid
);
1843 ce
->ce_stat_data
.sd_gid
= get_be32(&ondisk
->gid
);
1844 ce
->ce_stat_data
.sd_size
= get_be32(&ondisk
->size
);
1845 ce
->ce_flags
= flags
& ~CE_NAMEMASK
;
1846 ce
->ce_namelen
= len
;
1848 hashcpy(ce
->oid
.hash
, ondisk
->data
);
1849 memcpy(ce
->name
, name
, len
);
1850 ce
->name
[len
] = '\0';
1852 if (expand_name_field
) {
1854 memcpy(ce
->name
, previous_ce
->name
, copy_len
);
1855 memcpy(ce
->name
+ copy_len
, name
, len
+ 1 - copy_len
);
1856 *ent_size
= (name
- ((char *)ondisk
)) + len
+ 1 - copy_len
;
1858 memcpy(ce
->name
, name
, len
+ 1);
1859 *ent_size
= ondisk_ce_size(ce
);
1864 static void check_ce_order(struct index_state
*istate
)
1868 if (!verify_ce_order
)
1871 for (i
= 1; i
< istate
->cache_nr
; i
++) {
1872 struct cache_entry
*ce
= istate
->cache
[i
- 1];
1873 struct cache_entry
*next_ce
= istate
->cache
[i
];
1874 int name_compare
= strcmp(ce
->name
, next_ce
->name
);
1876 if (0 < name_compare
)
1877 die(_("unordered stage entries in index"));
1878 if (!name_compare
) {
1880 die(_("multiple stage entries for merged file '%s'"),
1882 if (ce_stage(ce
) > ce_stage(next_ce
))
1883 die(_("unordered stage entries for '%s'"),
1889 static void tweak_untracked_cache(struct index_state
*istate
)
1891 struct repository
*r
= the_repository
;
1893 prepare_repo_settings(r
);
1895 if (r
->settings
.core_untracked_cache
== UNTRACKED_CACHE_REMOVE
) {
1896 remove_untracked_cache(istate
);
1900 if (r
->settings
.core_untracked_cache
== UNTRACKED_CACHE_WRITE
)
1901 add_untracked_cache(istate
);
1904 static void tweak_split_index(struct index_state
*istate
)
1906 switch (git_config_get_split_index()) {
1907 case -1: /* unset: do nothing */
1910 remove_split_index(istate
);
1913 add_split_index(istate
);
1915 default: /* unknown value: do nothing */
1920 static void post_read_index_from(struct index_state
*istate
)
1922 check_ce_order(istate
);
1923 tweak_untracked_cache(istate
);
1924 tweak_split_index(istate
);
1925 tweak_fsmonitor(istate
);
1928 static size_t estimate_cache_size_from_compressed(unsigned int entries
)
1930 return entries
* (sizeof(struct cache_entry
) + CACHE_ENTRY_PATH_LENGTH
);
1933 static size_t estimate_cache_size(size_t ondisk_size
, unsigned int entries
)
1935 long per_entry
= sizeof(struct cache_entry
) - sizeof(struct ondisk_cache_entry
);
1938 * Account for potential alignment differences.
1940 per_entry
+= align_padding_size(per_entry
, 0);
1941 return ondisk_size
+ entries
* per_entry
;
1944 struct index_entry_offset
1946 /* starting byte offset into index file, count of index entries in this block */
1950 struct index_entry_offset_table
1953 struct index_entry_offset entries
[FLEX_ARRAY
];
1956 static struct index_entry_offset_table
*read_ieot_extension(const char *mmap
, size_t mmap_size
, size_t offset
);
1957 static void write_ieot_extension(struct strbuf
*sb
, struct index_entry_offset_table
*ieot
);
1959 static size_t read_eoie_extension(const char *mmap
, size_t mmap_size
);
1960 static void write_eoie_extension(struct strbuf
*sb
, git_hash_ctx
*eoie_context
, size_t offset
);
1962 struct load_index_extensions
1965 struct index_state
*istate
;
1968 unsigned long src_offset
;
1971 static void *load_index_extensions(void *_data
)
1973 struct load_index_extensions
*p
= _data
;
1974 unsigned long src_offset
= p
->src_offset
;
1976 while (src_offset
<= p
->mmap_size
- the_hash_algo
->rawsz
- 8) {
1977 /* After an array of active_nr index entries,
1978 * there can be arbitrary number of extended
1979 * sections, each of which is prefixed with
1980 * extension name (4-byte) and section length
1981 * in 4-byte network byte order.
1983 uint32_t extsize
= get_be32(p
->mmap
+ src_offset
+ 4);
1984 if (read_index_extension(p
->istate
,
1985 p
->mmap
+ src_offset
,
1986 p
->mmap
+ src_offset
+ 8,
1988 munmap((void *)p
->mmap
, p
->mmap_size
);
1989 die(_("index file corrupt"));
1992 src_offset
+= extsize
;
1999 * A helper function that will load the specified range of cache entries
2000 * from the memory mapped file and add them to the given index.
2002 static unsigned long load_cache_entry_block(struct index_state
*istate
,
2003 struct mem_pool
*ce_mem_pool
, int offset
, int nr
, const char *mmap
,
2004 unsigned long start_offset
, const struct cache_entry
*previous_ce
)
2007 unsigned long src_offset
= start_offset
;
2009 for (i
= offset
; i
< offset
+ nr
; i
++) {
2010 struct ondisk_cache_entry
*disk_ce
;
2011 struct cache_entry
*ce
;
2012 unsigned long consumed
;
2014 disk_ce
= (struct ondisk_cache_entry
*)(mmap
+ src_offset
);
2015 ce
= create_from_disk(ce_mem_pool
, istate
->version
, disk_ce
, &consumed
, previous_ce
);
2016 set_index_entry(istate
, i
, ce
);
2018 src_offset
+= consumed
;
2021 return src_offset
- start_offset
;
2024 static unsigned long load_all_cache_entries(struct index_state
*istate
,
2025 const char *mmap
, size_t mmap_size
, unsigned long src_offset
)
2027 unsigned long consumed
;
2029 istate
->ce_mem_pool
= xmalloc(sizeof(*istate
->ce_mem_pool
));
2030 if (istate
->version
== 4) {
2031 mem_pool_init(istate
->ce_mem_pool
,
2032 estimate_cache_size_from_compressed(istate
->cache_nr
));
2034 mem_pool_init(istate
->ce_mem_pool
,
2035 estimate_cache_size(mmap_size
, istate
->cache_nr
));
2038 consumed
= load_cache_entry_block(istate
, istate
->ce_mem_pool
,
2039 0, istate
->cache_nr
, mmap
, src_offset
, NULL
);
2044 * Mostly randomly chosen maximum thread counts: we
2045 * cap the parallelism to online_cpus() threads, and we want
2046 * to have at least 10000 cache entries per thread for it to
2047 * be worth starting a thread.
2050 #define THREAD_COST (10000)
2052 struct load_cache_entries_thread_data
2055 struct index_state
*istate
;
2056 struct mem_pool
*ce_mem_pool
;
2059 struct index_entry_offset_table
*ieot
;
2060 int ieot_start
; /* starting index into the ieot array */
2061 int ieot_blocks
; /* count of ieot entries to process */
2062 unsigned long consumed
; /* return # of bytes in index file processed */
2066 * A thread proc to run the load_cache_entries() computation
2067 * across multiple background threads.
2069 static void *load_cache_entries_thread(void *_data
)
2071 struct load_cache_entries_thread_data
*p
= _data
;
2074 /* iterate across all ieot blocks assigned to this thread */
2075 for (i
= p
->ieot_start
; i
< p
->ieot_start
+ p
->ieot_blocks
; i
++) {
2076 p
->consumed
+= load_cache_entry_block(p
->istate
, p
->ce_mem_pool
,
2077 p
->offset
, p
->ieot
->entries
[i
].nr
, p
->mmap
, p
->ieot
->entries
[i
].offset
, NULL
);
2078 p
->offset
+= p
->ieot
->entries
[i
].nr
;
2083 static unsigned long load_cache_entries_threaded(struct index_state
*istate
, const char *mmap
, size_t mmap_size
,
2084 int nr_threads
, struct index_entry_offset_table
*ieot
)
2086 int i
, offset
, ieot_blocks
, ieot_start
, err
;
2087 struct load_cache_entries_thread_data
*data
;
2088 unsigned long consumed
= 0;
2090 /* a little sanity checking */
2091 if (istate
->name_hash_initialized
)
2092 BUG("the name hash isn't thread safe");
2094 istate
->ce_mem_pool
= xmalloc(sizeof(*istate
->ce_mem_pool
));
2095 mem_pool_init(istate
->ce_mem_pool
, 0);
2097 /* ensure we have no more threads than we have blocks to process */
2098 if (nr_threads
> ieot
->nr
)
2099 nr_threads
= ieot
->nr
;
2100 data
= xcalloc(nr_threads
, sizeof(*data
));
2102 offset
= ieot_start
= 0;
2103 ieot_blocks
= DIV_ROUND_UP(ieot
->nr
, nr_threads
);
2104 for (i
= 0; i
< nr_threads
; i
++) {
2105 struct load_cache_entries_thread_data
*p
= &data
[i
];
2108 if (ieot_start
+ ieot_blocks
> ieot
->nr
)
2109 ieot_blocks
= ieot
->nr
- ieot_start
;
2115 p
->ieot_start
= ieot_start
;
2116 p
->ieot_blocks
= ieot_blocks
;
2118 /* create a mem_pool for each thread */
2120 for (j
= p
->ieot_start
; j
< p
->ieot_start
+ p
->ieot_blocks
; j
++)
2121 nr
+= p
->ieot
->entries
[j
].nr
;
2122 p
->ce_mem_pool
= xmalloc(sizeof(*istate
->ce_mem_pool
));
2123 if (istate
->version
== 4) {
2124 mem_pool_init(p
->ce_mem_pool
,
2125 estimate_cache_size_from_compressed(nr
));
2127 mem_pool_init(p
->ce_mem_pool
,
2128 estimate_cache_size(mmap_size
, nr
));
2131 err
= pthread_create(&p
->pthread
, NULL
, load_cache_entries_thread
, p
);
2133 die(_("unable to create load_cache_entries thread: %s"), strerror(err
));
2135 /* increment by the number of cache entries in the ieot block being processed */
2136 for (j
= 0; j
< ieot_blocks
; j
++)
2137 offset
+= ieot
->entries
[ieot_start
+ j
].nr
;
2138 ieot_start
+= ieot_blocks
;
2141 for (i
= 0; i
< nr_threads
; i
++) {
2142 struct load_cache_entries_thread_data
*p
= &data
[i
];
2144 err
= pthread_join(p
->pthread
, NULL
);
2146 die(_("unable to join load_cache_entries thread: %s"), strerror(err
));
2147 mem_pool_combine(istate
->ce_mem_pool
, p
->ce_mem_pool
);
2148 consumed
+= p
->consumed
;
2156 /* remember to discard_cache() before reading a different cache! */
2157 int do_read_index(struct index_state
*istate
, const char *path
, int must_exist
)
2161 unsigned long src_offset
;
2162 const struct cache_header
*hdr
;
2165 struct load_index_extensions p
;
2166 size_t extension_offset
= 0;
2167 int nr_threads
, cpus
;
2168 struct index_entry_offset_table
*ieot
= NULL
;
2170 if (istate
->initialized
)
2171 return istate
->cache_nr
;
2173 istate
->timestamp
.sec
= 0;
2174 istate
->timestamp
.nsec
= 0;
2175 fd
= open(path
, O_RDONLY
);
2177 if (!must_exist
&& errno
== ENOENT
)
2179 die_errno(_("%s: index file open failed"), path
);
2183 die_errno(_("%s: cannot stat the open index"), path
);
2185 mmap_size
= xsize_t(st
.st_size
);
2186 if (mmap_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2187 die(_("%s: index file smaller than expected"), path
);
2189 mmap
= xmmap_gently(NULL
, mmap_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2190 if (mmap
== MAP_FAILED
)
2191 die_errno(_("%s: unable to map index file"), path
);
2194 hdr
= (const struct cache_header
*)mmap
;
2195 if (verify_hdr(hdr
, mmap_size
) < 0)
2198 hashcpy(istate
->oid
.hash
, (const unsigned char *)hdr
+ mmap_size
- the_hash_algo
->rawsz
);
2199 istate
->version
= ntohl(hdr
->hdr_version
);
2200 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
2201 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
2202 istate
->cache
= xcalloc(istate
->cache_alloc
, sizeof(*istate
->cache
));
2203 istate
->initialized
= 1;
2207 p
.mmap_size
= mmap_size
;
2209 src_offset
= sizeof(*hdr
);
2211 if (git_config_get_index_threads(&nr_threads
))
2214 /* TODO: does creating more threads than cores help? */
2216 nr_threads
= istate
->cache_nr
/ THREAD_COST
;
2217 cpus
= online_cpus();
2218 if (nr_threads
> cpus
)
2225 if (nr_threads
> 1) {
2226 extension_offset
= read_eoie_extension(mmap
, mmap_size
);
2227 if (extension_offset
) {
2230 p
.src_offset
= extension_offset
;
2231 err
= pthread_create(&p
.pthread
, NULL
, load_index_extensions
, &p
);
2233 die(_("unable to create load_index_extensions thread: %s"), strerror(err
));
2240 * Locate and read the index entry offset table so that we can use it
2241 * to multi-thread the reading of the cache entries.
2243 if (extension_offset
&& nr_threads
> 1)
2244 ieot
= read_ieot_extension(mmap
, mmap_size
, extension_offset
);
2247 src_offset
+= load_cache_entries_threaded(istate
, mmap
, mmap_size
, nr_threads
, ieot
);
2250 src_offset
+= load_all_cache_entries(istate
, mmap
, mmap_size
, src_offset
);
2253 istate
->timestamp
.sec
= st
.st_mtime
;
2254 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
2256 /* if we created a thread, join it otherwise load the extensions on the primary thread */
2257 if (extension_offset
) {
2258 int ret
= pthread_join(p
.pthread
, NULL
);
2260 die(_("unable to join load_index_extensions thread: %s"), strerror(ret
));
2262 p
.src_offset
= src_offset
;
2263 load_index_extensions(&p
);
2265 munmap((void *)mmap
, mmap_size
);
2268 * TODO trace2: replace "the_repository" with the actual repo instance
2269 * that is associated with the given "istate".
2271 trace2_data_intmax("index", the_repository
, "read/version",
2273 trace2_data_intmax("index", the_repository
, "read/cache_nr",
2276 return istate
->cache_nr
;
2279 munmap((void *)mmap
, mmap_size
);
2280 die(_("index file corrupt"));
2284 * Signal that the shared index is used by updating its mtime.
2286 * This way, shared index can be removed if they have not been used
2289 static void freshen_shared_index(const char *shared_index
, int warn
)
2291 if (!check_and_freshen_file(shared_index
, 1) && warn
)
2292 warning(_("could not freshen shared index '%s'"), shared_index
);
2295 int read_index_from(struct index_state
*istate
, const char *path
,
2298 struct split_index
*split_index
;
2303 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2304 if (istate
->initialized
)
2305 return istate
->cache_nr
;
2308 * TODO trace2: replace "the_repository" with the actual repo instance
2309 * that is associated with the given "istate".
2311 trace2_region_enter_printf("index", "do_read_index", the_repository
,
2313 trace_performance_enter();
2314 ret
= do_read_index(istate
, path
, 0);
2315 trace_performance_leave("read cache %s", path
);
2316 trace2_region_leave_printf("index", "do_read_index", the_repository
,
2319 split_index
= istate
->split_index
;
2320 if (!split_index
|| is_null_oid(&split_index
->base_oid
)) {
2321 post_read_index_from(istate
);
2325 trace_performance_enter();
2326 if (split_index
->base
)
2327 discard_index(split_index
->base
);
2329 split_index
->base
= xcalloc(1, sizeof(*split_index
->base
));
2331 base_oid_hex
= oid_to_hex(&split_index
->base_oid
);
2332 base_path
= xstrfmt("%s/sharedindex.%s", gitdir
, base_oid_hex
);
2333 trace2_region_enter_printf("index", "shared/do_read_index",
2334 the_repository
, "%s", base_path
);
2335 ret
= do_read_index(split_index
->base
, base_path
, 1);
2336 trace2_region_leave_printf("index", "shared/do_read_index",
2337 the_repository
, "%s", base_path
);
2338 if (!oideq(&split_index
->base_oid
, &split_index
->base
->oid
))
2339 die(_("broken index, expect %s in %s, got %s"),
2340 base_oid_hex
, base_path
,
2341 oid_to_hex(&split_index
->base
->oid
));
2343 freshen_shared_index(base_path
, 0);
2344 merge_base_index(istate
);
2345 post_read_index_from(istate
);
2346 trace_performance_leave("read cache %s", base_path
);
2351 int is_index_unborn(struct index_state
*istate
)
2353 return (!istate
->cache_nr
&& !istate
->timestamp
.sec
);
2356 int discard_index(struct index_state
*istate
)
2359 * Cache entries in istate->cache[] should have been allocated
2360 * from the memory pool associated with this index, or from an
2361 * associated split_index. There is no need to free individual
2362 * cache entries. validate_cache_entries can detect when this
2363 * assertion does not hold.
2365 validate_cache_entries(istate
);
2367 resolve_undo_clear_index(istate
);
2368 istate
->cache_nr
= 0;
2369 istate
->cache_changed
= 0;
2370 istate
->timestamp
.sec
= 0;
2371 istate
->timestamp
.nsec
= 0;
2372 free_name_hash(istate
);
2373 cache_tree_free(&(istate
->cache_tree
));
2374 istate
->initialized
= 0;
2375 istate
->fsmonitor_has_run_once
= 0;
2376 FREE_AND_NULL(istate
->cache
);
2377 istate
->cache_alloc
= 0;
2378 discard_split_index(istate
);
2379 free_untracked_cache(istate
->untracked
);
2380 istate
->untracked
= NULL
;
2382 if (istate
->ce_mem_pool
) {
2383 mem_pool_discard(istate
->ce_mem_pool
, should_validate_cache_entries());
2384 FREE_AND_NULL(istate
->ce_mem_pool
);
2391 * Validate the cache entries of this index.
2392 * All cache entries associated with this index
2393 * should have been allocated by the memory pool
2394 * associated with this index, or by a referenced
2397 void validate_cache_entries(const struct index_state
*istate
)
2401 if (!should_validate_cache_entries() ||!istate
|| !istate
->initialized
)
2404 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2406 BUG("cache entry is not allocated from expected memory pool");
2407 } else if (!istate
->ce_mem_pool
||
2408 !mem_pool_contains(istate
->ce_mem_pool
, istate
->cache
[i
])) {
2409 if (!istate
->split_index
||
2410 !istate
->split_index
->base
||
2411 !istate
->split_index
->base
->ce_mem_pool
||
2412 !mem_pool_contains(istate
->split_index
->base
->ce_mem_pool
, istate
->cache
[i
])) {
2413 BUG("cache entry is not allocated from expected memory pool");
2418 if (istate
->split_index
)
2419 validate_cache_entries(istate
->split_index
->base
);
2422 int unmerged_index(const struct index_state
*istate
)
2425 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2426 if (ce_stage(istate
->cache
[i
]))
2432 int repo_index_has_changes(struct repository
*repo
,
2436 struct index_state
*istate
= repo
->index
;
2437 struct object_id cmp
;
2441 cmp
= tree
->object
.oid
;
2442 if (tree
|| !get_oid_tree("HEAD", &cmp
)) {
2443 struct diff_options opt
;
2445 repo_diff_setup(repo
, &opt
);
2446 opt
.flags
.exit_with_status
= 1;
2448 opt
.flags
.quick
= 1;
2449 do_diff_cache(&cmp
, &opt
);
2451 for (i
= 0; sb
&& i
< diff_queued_diff
.nr
; i
++) {
2453 strbuf_addch(sb
, ' ');
2454 strbuf_addstr(sb
, diff_queued_diff
.queue
[i
]->two
->path
);
2457 return opt
.flags
.has_changes
!= 0;
2459 for (i
= 0; sb
&& i
< istate
->cache_nr
; i
++) {
2461 strbuf_addch(sb
, ' ');
2462 strbuf_addstr(sb
, istate
->cache
[i
]->name
);
2464 return !!istate
->cache_nr
;
2468 #define WRITE_BUFFER_SIZE (128 * 1024)
2469 static unsigned char write_buffer
[WRITE_BUFFER_SIZE
];
2470 static unsigned long write_buffer_len
;
2472 static int ce_write_flush(git_hash_ctx
*context
, int fd
)
2474 unsigned int buffered
= write_buffer_len
;
2476 the_hash_algo
->update_fn(context
, write_buffer
, buffered
);
2477 if (write_in_full(fd
, write_buffer
, buffered
) < 0)
2479 write_buffer_len
= 0;
2484 static int ce_write(git_hash_ctx
*context
, int fd
, void *data
, unsigned int len
)
2487 unsigned int buffered
= write_buffer_len
;
2488 unsigned int partial
= WRITE_BUFFER_SIZE
- buffered
;
2491 memcpy(write_buffer
+ buffered
, data
, partial
);
2492 buffered
+= partial
;
2493 if (buffered
== WRITE_BUFFER_SIZE
) {
2494 write_buffer_len
= buffered
;
2495 if (ce_write_flush(context
, fd
))
2499 write_buffer_len
= buffered
;
2501 data
= (char *) data
+ partial
;
2506 static int write_index_ext_header(git_hash_ctx
*context
, git_hash_ctx
*eoie_context
,
2507 int fd
, unsigned int ext
, unsigned int sz
)
2512 the_hash_algo
->update_fn(eoie_context
, &ext
, 4);
2513 the_hash_algo
->update_fn(eoie_context
, &sz
, 4);
2515 return ((ce_write(context
, fd
, &ext
, 4) < 0) ||
2516 (ce_write(context
, fd
, &sz
, 4) < 0)) ? -1 : 0;
2519 static int ce_flush(git_hash_ctx
*context
, int fd
, unsigned char *hash
)
2521 unsigned int left
= write_buffer_len
;
2524 write_buffer_len
= 0;
2525 the_hash_algo
->update_fn(context
, write_buffer
, left
);
2528 /* Flush first if not enough space for hash signature */
2529 if (left
+ the_hash_algo
->rawsz
> WRITE_BUFFER_SIZE
) {
2530 if (write_in_full(fd
, write_buffer
, left
) < 0)
2535 /* Append the hash signature at the end */
2536 the_hash_algo
->final_fn(write_buffer
+ left
, context
);
2537 hashcpy(hash
, write_buffer
+ left
);
2538 left
+= the_hash_algo
->rawsz
;
2539 return (write_in_full(fd
, write_buffer
, left
) < 0) ? -1 : 0;
2542 static void ce_smudge_racily_clean_entry(struct index_state
*istate
,
2543 struct cache_entry
*ce
)
2546 * The only thing we care about in this function is to smudge the
2547 * falsely clean entry due to touch-update-touch race, so we leave
2548 * everything else as they are. We are called for entries whose
2549 * ce_stat_data.sd_mtime match the index file mtime.
2551 * Note that this actually does not do much for gitlinks, for
2552 * which ce_match_stat_basic() always goes to the actual
2553 * contents. The caller checks with is_racy_timestamp() which
2554 * always says "no" for gitlinks, so we are not called for them ;-)
2558 if (lstat(ce
->name
, &st
) < 0)
2560 if (ce_match_stat_basic(ce
, &st
))
2562 if (ce_modified_check_fs(istate
, ce
, &st
)) {
2563 /* This is "racily clean"; smudge it. Note that this
2564 * is a tricky code. At first glance, it may appear
2565 * that it can break with this sequence:
2567 * $ echo xyzzy >frotz
2568 * $ git-update-index --add frotz
2571 * $ echo filfre >nitfol
2572 * $ git-update-index --add nitfol
2574 * but it does not. When the second update-index runs,
2575 * it notices that the entry "frotz" has the same timestamp
2576 * as index, and if we were to smudge it by resetting its
2577 * size to zero here, then the object name recorded
2578 * in index is the 6-byte file but the cached stat information
2579 * becomes zero --- which would then match what we would
2580 * obtain from the filesystem next time we stat("frotz").
2582 * However, the second update-index, before calling
2583 * this function, notices that the cached size is 6
2584 * bytes and what is on the filesystem is an empty
2585 * file, and never calls us, so the cached size information
2586 * for "frotz" stays 6 which does not match the filesystem.
2588 ce
->ce_stat_data
.sd_size
= 0;
2592 /* Copy miscellaneous fields but not the name */
2593 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry
*ondisk
,
2594 struct cache_entry
*ce
)
2597 const unsigned hashsz
= the_hash_algo
->rawsz
;
2598 uint16_t *flagsp
= (uint16_t *)(ondisk
->data
+ hashsz
);
2600 ondisk
->ctime
.sec
= htonl(ce
->ce_stat_data
.sd_ctime
.sec
);
2601 ondisk
->mtime
.sec
= htonl(ce
->ce_stat_data
.sd_mtime
.sec
);
2602 ondisk
->ctime
.nsec
= htonl(ce
->ce_stat_data
.sd_ctime
.nsec
);
2603 ondisk
->mtime
.nsec
= htonl(ce
->ce_stat_data
.sd_mtime
.nsec
);
2604 ondisk
->dev
= htonl(ce
->ce_stat_data
.sd_dev
);
2605 ondisk
->ino
= htonl(ce
->ce_stat_data
.sd_ino
);
2606 ondisk
->mode
= htonl(ce
->ce_mode
);
2607 ondisk
->uid
= htonl(ce
->ce_stat_data
.sd_uid
);
2608 ondisk
->gid
= htonl(ce
->ce_stat_data
.sd_gid
);
2609 ondisk
->size
= htonl(ce
->ce_stat_data
.sd_size
);
2610 hashcpy(ondisk
->data
, ce
->oid
.hash
);
2612 flags
= ce
->ce_flags
& ~CE_NAMEMASK
;
2613 flags
|= (ce_namelen(ce
) >= CE_NAMEMASK
? CE_NAMEMASK
: ce_namelen(ce
));
2614 flagsp
[0] = htons(flags
);
2615 if (ce
->ce_flags
& CE_EXTENDED
) {
2616 flagsp
[1] = htons((ce
->ce_flags
& CE_EXTENDED_FLAGS
) >> 16);
2620 static int ce_write_entry(git_hash_ctx
*c
, int fd
, struct cache_entry
*ce
,
2621 struct strbuf
*previous_name
, struct ondisk_cache_entry
*ondisk
)
2625 unsigned int saved_namelen
;
2626 int stripped_name
= 0;
2627 static unsigned char padding
[8] = { 0x00 };
2629 if (ce
->ce_flags
& CE_STRIP_NAME
) {
2630 saved_namelen
= ce_namelen(ce
);
2635 size
= offsetof(struct ondisk_cache_entry
,data
) + ondisk_data_size(ce
->ce_flags
, 0);
2637 if (!previous_name
) {
2638 int len
= ce_namelen(ce
);
2639 copy_cache_entry_to_ondisk(ondisk
, ce
);
2640 result
= ce_write(c
, fd
, ondisk
, size
);
2642 result
= ce_write(c
, fd
, ce
->name
, len
);
2644 result
= ce_write(c
, fd
, padding
, align_padding_size(size
, len
));
2646 int common
, to_remove
, prefix_size
;
2647 unsigned char to_remove_vi
[16];
2649 (ce
->name
[common
] &&
2650 common
< previous_name
->len
&&
2651 ce
->name
[common
] == previous_name
->buf
[common
]);
2653 ; /* still matching */
2654 to_remove
= previous_name
->len
- common
;
2655 prefix_size
= encode_varint(to_remove
, to_remove_vi
);
2657 copy_cache_entry_to_ondisk(ondisk
, ce
);
2658 result
= ce_write(c
, fd
, ondisk
, size
);
2660 result
= ce_write(c
, fd
, to_remove_vi
, prefix_size
);
2662 result
= ce_write(c
, fd
, ce
->name
+ common
, ce_namelen(ce
) - common
);
2664 result
= ce_write(c
, fd
, padding
, 1);
2666 strbuf_splice(previous_name
, common
, to_remove
,
2667 ce
->name
+ common
, ce_namelen(ce
) - common
);
2669 if (stripped_name
) {
2670 ce
->ce_namelen
= saved_namelen
;
2671 ce
->ce_flags
&= ~CE_STRIP_NAME
;
2678 * This function verifies if index_state has the correct sha1 of the
2679 * index file. Don't die if we have any other failure, just return 0.
2681 static int verify_index_from(const struct index_state
*istate
, const char *path
)
2686 unsigned char hash
[GIT_MAX_RAWSZ
];
2688 if (!istate
->initialized
)
2691 fd
= open(path
, O_RDONLY
);
2698 if (st
.st_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2701 n
= pread_in_full(fd
, hash
, the_hash_algo
->rawsz
, st
.st_size
- the_hash_algo
->rawsz
);
2702 if (n
!= the_hash_algo
->rawsz
)
2705 if (!hasheq(istate
->oid
.hash
, hash
))
2716 static int repo_verify_index(struct repository
*repo
)
2718 return verify_index_from(repo
->index
, repo
->index_file
);
2721 static int has_racy_timestamp(struct index_state
*istate
)
2723 int entries
= istate
->cache_nr
;
2726 for (i
= 0; i
< entries
; i
++) {
2727 struct cache_entry
*ce
= istate
->cache
[i
];
2728 if (is_racy_timestamp(istate
, ce
))
2734 void repo_update_index_if_able(struct repository
*repo
,
2735 struct lock_file
*lockfile
)
2737 if ((repo
->index
->cache_changed
||
2738 has_racy_timestamp(repo
->index
)) &&
2739 repo_verify_index(repo
))
2740 write_locked_index(repo
->index
, lockfile
, COMMIT_LOCK
);
2742 rollback_lock_file(lockfile
);
2745 static int record_eoie(void)
2749 if (!git_config_get_bool("index.recordendofindexentries", &val
))
2753 * As a convenience, the end of index entries extension
2754 * used for threading is written by default if the user
2755 * explicitly requested threaded index reads.
2757 return !git_config_get_index_threads(&val
) && val
!= 1;
2760 static int record_ieot(void)
2764 if (!git_config_get_bool("index.recordoffsettable", &val
))
2768 * As a convenience, the offset table used for threading is
2769 * written by default if the user explicitly requested
2770 * threaded index reads.
2772 return !git_config_get_index_threads(&val
) && val
!= 1;
2776 * On success, `tempfile` is closed. If it is the temporary file
2777 * of a `struct lock_file`, we will therefore effectively perform
2778 * a 'close_lock_file_gently()`. Since that is an implementation
2779 * detail of lockfiles, callers of `do_write_index()` should not
2782 static int do_write_index(struct index_state
*istate
, struct tempfile
*tempfile
,
2783 int strip_extensions
)
2785 uint64_t start
= getnanotime();
2786 int newfd
= tempfile
->fd
;
2787 git_hash_ctx c
, eoie_c
;
2788 struct cache_header hdr
;
2789 int i
, err
= 0, removed
, extended
, hdr_version
;
2790 struct cache_entry
**cache
= istate
->cache
;
2791 int entries
= istate
->cache_nr
;
2793 struct ondisk_cache_entry ondisk
;
2794 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
2795 int drop_cache_tree
= istate
->drop_cache_tree
;
2797 int ieot_entries
= 1;
2798 struct index_entry_offset_table
*ieot
= NULL
;
2801 for (i
= removed
= extended
= 0; i
< entries
; i
++) {
2802 if (cache
[i
]->ce_flags
& CE_REMOVE
)
2805 /* reduce extended entries if possible */
2806 cache
[i
]->ce_flags
&= ~CE_EXTENDED
;
2807 if (cache
[i
]->ce_flags
& CE_EXTENDED_FLAGS
) {
2809 cache
[i
]->ce_flags
|= CE_EXTENDED
;
2813 if (!istate
->version
) {
2814 istate
->version
= get_index_format_default(the_repository
);
2815 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0))
2816 init_split_index(istate
);
2819 /* demote version 3 to version 2 when the latter suffices */
2820 if (istate
->version
== 3 || istate
->version
== 2)
2821 istate
->version
= extended
? 3 : 2;
2823 hdr_version
= istate
->version
;
2825 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
2826 hdr
.hdr_version
= htonl(hdr_version
);
2827 hdr
.hdr_entries
= htonl(entries
- removed
);
2829 the_hash_algo
->init_fn(&c
);
2830 if (ce_write(&c
, newfd
, &hdr
, sizeof(hdr
)) < 0)
2833 if (!HAVE_THREADS
|| git_config_get_index_threads(&nr_threads
))
2836 if (nr_threads
!= 1 && record_ieot()) {
2837 int ieot_blocks
, cpus
;
2840 * ensure default number of ieot blocks maps evenly to the
2841 * default number of threads that will process them leaving
2842 * room for the thread to load the index extensions.
2845 ieot_blocks
= istate
->cache_nr
/ THREAD_COST
;
2846 cpus
= online_cpus();
2847 if (ieot_blocks
> cpus
- 1)
2848 ieot_blocks
= cpus
- 1;
2850 ieot_blocks
= nr_threads
;
2851 if (ieot_blocks
> istate
->cache_nr
)
2852 ieot_blocks
= istate
->cache_nr
;
2856 * no reason to write out the IEOT extension if we don't
2857 * have enough blocks to utilize multi-threading
2859 if (ieot_blocks
> 1) {
2860 ieot
= xcalloc(1, sizeof(struct index_entry_offset_table
)
2861 + (ieot_blocks
* sizeof(struct index_entry_offset
)));
2862 ieot_entries
= DIV_ROUND_UP(entries
, ieot_blocks
);
2866 offset
= lseek(newfd
, 0, SEEK_CUR
);
2871 offset
+= write_buffer_len
;
2873 previous_name
= (hdr_version
== 4) ? &previous_name_buf
: NULL
;
2875 for (i
= 0; i
< entries
; i
++) {
2876 struct cache_entry
*ce
= cache
[i
];
2877 if (ce
->ce_flags
& CE_REMOVE
)
2879 if (!ce_uptodate(ce
) && is_racy_timestamp(istate
, ce
))
2880 ce_smudge_racily_clean_entry(istate
, ce
);
2881 if (is_null_oid(&ce
->oid
)) {
2882 static const char msg
[] = "cache entry has null sha1: %s";
2883 static int allow
= -1;
2886 allow
= git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2888 warning(msg
, ce
->name
);
2890 err
= error(msg
, ce
->name
);
2892 drop_cache_tree
= 1;
2894 if (ieot
&& i
&& (i
% ieot_entries
== 0)) {
2895 ieot
->entries
[ieot
->nr
].nr
= nr
;
2896 ieot
->entries
[ieot
->nr
].offset
= offset
;
2899 * If we have a V4 index, set the first byte to an invalid
2900 * character to ensure there is nothing common with the previous
2904 previous_name
->buf
[0] = 0;
2906 offset
= lseek(newfd
, 0, SEEK_CUR
);
2911 offset
+= write_buffer_len
;
2913 if (ce_write_entry(&c
, newfd
, ce
, previous_name
, (struct ondisk_cache_entry
*)&ondisk
) < 0)
2921 ieot
->entries
[ieot
->nr
].nr
= nr
;
2922 ieot
->entries
[ieot
->nr
].offset
= offset
;
2925 strbuf_release(&previous_name_buf
);
2932 /* Write extension data here */
2933 offset
= lseek(newfd
, 0, SEEK_CUR
);
2938 offset
+= write_buffer_len
;
2939 the_hash_algo
->init_fn(&eoie_c
);
2942 * Lets write out CACHE_EXT_INDEXENTRYOFFSETTABLE first so that we
2943 * can minimize the number of extensions we have to scan through to
2944 * find it during load. Write it out regardless of the
2945 * strip_extensions parameter as we need it when loading the shared
2949 struct strbuf sb
= STRBUF_INIT
;
2951 write_ieot_extension(&sb
, ieot
);
2952 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_INDEXENTRYOFFSETTABLE
, sb
.len
) < 0
2953 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2954 strbuf_release(&sb
);
2960 if (!strip_extensions
&& istate
->split_index
&&
2961 !is_null_oid(&istate
->split_index
->base_oid
)) {
2962 struct strbuf sb
= STRBUF_INIT
;
2964 err
= write_link_extension(&sb
, istate
) < 0 ||
2965 write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_LINK
,
2967 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2968 strbuf_release(&sb
);
2972 if (!strip_extensions
&& !drop_cache_tree
&& istate
->cache_tree
) {
2973 struct strbuf sb
= STRBUF_INIT
;
2975 cache_tree_write(&sb
, istate
->cache_tree
);
2976 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_TREE
, sb
.len
) < 0
2977 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2978 strbuf_release(&sb
);
2982 if (!strip_extensions
&& istate
->resolve_undo
) {
2983 struct strbuf sb
= STRBUF_INIT
;
2985 resolve_undo_write(&sb
, istate
->resolve_undo
);
2986 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_RESOLVE_UNDO
,
2988 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2989 strbuf_release(&sb
);
2993 if (!strip_extensions
&& istate
->untracked
) {
2994 struct strbuf sb
= STRBUF_INIT
;
2996 write_untracked_extension(&sb
, istate
->untracked
);
2997 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_UNTRACKED
,
2999 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
3000 strbuf_release(&sb
);
3004 if (!strip_extensions
&& istate
->fsmonitor_last_update
) {
3005 struct strbuf sb
= STRBUF_INIT
;
3007 write_fsmonitor_extension(&sb
, istate
);
3008 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_FSMONITOR
, sb
.len
) < 0
3009 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
3010 strbuf_release(&sb
);
3016 * CACHE_EXT_ENDOFINDEXENTRIES must be written as the last entry before the SHA1
3017 * so that it can be found and processed before all the index entries are
3018 * read. Write it out regardless of the strip_extensions parameter as we need it
3019 * when loading the shared index.
3021 if (offset
&& record_eoie()) {
3022 struct strbuf sb
= STRBUF_INIT
;
3024 write_eoie_extension(&sb
, &eoie_c
, offset
);
3025 err
= write_index_ext_header(&c
, NULL
, newfd
, CACHE_EXT_ENDOFINDEXENTRIES
, sb
.len
) < 0
3026 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
3027 strbuf_release(&sb
);
3032 if (ce_flush(&c
, newfd
, istate
->oid
.hash
))
3034 if (close_tempfile_gently(tempfile
)) {
3035 error(_("could not close '%s'"), get_tempfile_path(tempfile
));
3038 if (stat(get_tempfile_path(tempfile
), &st
))
3040 istate
->timestamp
.sec
= (unsigned int)st
.st_mtime
;
3041 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
3042 trace_performance_since(start
, "write index, changed mask = %x", istate
->cache_changed
);
3045 * TODO trace2: replace "the_repository" with the actual repo instance
3046 * that is associated with the given "istate".
3048 trace2_data_intmax("index", the_repository
, "write/version",
3050 trace2_data_intmax("index", the_repository
, "write/cache_nr",
3056 void set_alternate_index_output(const char *name
)
3058 alternate_index_output
= name
;
3061 static int commit_locked_index(struct lock_file
*lk
)
3063 if (alternate_index_output
)
3064 return commit_lock_file_to(lk
, alternate_index_output
);
3066 return commit_lock_file(lk
);
3069 static int do_write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
3075 * TODO trace2: replace "the_repository" with the actual repo instance
3076 * that is associated with the given "istate".
3078 trace2_region_enter_printf("index", "do_write_index", the_repository
,
3079 "%s", get_lock_file_path(lock
));
3080 ret
= do_write_index(istate
, lock
->tempfile
, 0);
3081 trace2_region_leave_printf("index", "do_write_index", the_repository
,
3082 "%s", get_lock_file_path(lock
));
3086 if (flags
& COMMIT_LOCK
)
3087 ret
= commit_locked_index(lock
);
3089 ret
= close_lock_file_gently(lock
);
3091 run_hook_le(NULL
, "post-index-change",
3092 istate
->updated_workdir
? "1" : "0",
3093 istate
->updated_skipworktree
? "1" : "0", NULL
);
3094 istate
->updated_workdir
= 0;
3095 istate
->updated_skipworktree
= 0;
3100 static int write_split_index(struct index_state
*istate
,
3101 struct lock_file
*lock
,
3105 prepare_to_write_split_index(istate
);
3106 ret
= do_write_locked_index(istate
, lock
, flags
);
3107 finish_writing_split_index(istate
);
3111 static const char *shared_index_expire
= "2.weeks.ago";
3113 static unsigned long get_shared_index_expire_date(void)
3115 static unsigned long shared_index_expire_date
;
3116 static int shared_index_expire_date_prepared
;
3118 if (!shared_index_expire_date_prepared
) {
3119 git_config_get_expiry("splitindex.sharedindexexpire",
3120 &shared_index_expire
);
3121 shared_index_expire_date
= approxidate(shared_index_expire
);
3122 shared_index_expire_date_prepared
= 1;
3125 return shared_index_expire_date
;
3128 static int should_delete_shared_index(const char *shared_index_path
)
3131 unsigned long expiration
;
3133 /* Check timestamp */
3134 expiration
= get_shared_index_expire_date();
3137 if (stat(shared_index_path
, &st
))
3138 return error_errno(_("could not stat '%s'"), shared_index_path
);
3139 if (st
.st_mtime
> expiration
)
3145 static int clean_shared_index_files(const char *current_hex
)
3148 DIR *dir
= opendir(get_git_dir());
3151 return error_errno(_("unable to open git dir: %s"), get_git_dir());
3153 while ((de
= readdir(dir
)) != NULL
) {
3154 const char *sha1_hex
;
3155 const char *shared_index_path
;
3156 if (!skip_prefix(de
->d_name
, "sharedindex.", &sha1_hex
))
3158 if (!strcmp(sha1_hex
, current_hex
))
3160 shared_index_path
= git_path("%s", de
->d_name
);
3161 if (should_delete_shared_index(shared_index_path
) > 0 &&
3162 unlink(shared_index_path
))
3163 warning_errno(_("unable to unlink: %s"), shared_index_path
);
3170 static int write_shared_index(struct index_state
*istate
,
3171 struct tempfile
**temp
)
3173 struct split_index
*si
= istate
->split_index
;
3176 move_cache_to_base_index(istate
);
3178 trace2_region_enter_printf("index", "shared/do_write_index",
3179 the_repository
, "%s", get_tempfile_path(*temp
));
3180 ret
= do_write_index(si
->base
, *temp
, 1);
3181 trace2_region_leave_printf("index", "shared/do_write_index",
3182 the_repository
, "%s", get_tempfile_path(*temp
));
3186 ret
= adjust_shared_perm(get_tempfile_path(*temp
));
3188 error(_("cannot fix permission bits on '%s'"), get_tempfile_path(*temp
));
3191 ret
= rename_tempfile(temp
,
3192 git_path("sharedindex.%s", oid_to_hex(&si
->base
->oid
)));
3194 oidcpy(&si
->base_oid
, &si
->base
->oid
);
3195 clean_shared_index_files(oid_to_hex(&si
->base
->oid
));
3201 static const int default_max_percent_split_change
= 20;
3203 static int too_many_not_shared_entries(struct index_state
*istate
)
3205 int i
, not_shared
= 0;
3206 int max_split
= git_config_get_max_percent_split_change();
3208 switch (max_split
) {
3210 /* not or badly configured: use the default value */
3211 max_split
= default_max_percent_split_change
;
3214 return 1; /* 0% means always write a new shared index */
3216 return 0; /* 100% means never write a new shared index */
3218 break; /* just use the configured value */
3221 /* Count not shared entries */
3222 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3223 struct cache_entry
*ce
= istate
->cache
[i
];
3228 return (int64_t)istate
->cache_nr
* max_split
< (int64_t)not_shared
* 100;
3231 int write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
3234 int new_shared_index
, ret
;
3235 struct split_index
*si
= istate
->split_index
;
3237 if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0))
3238 cache_tree_verify(the_repository
, istate
);
3240 if ((flags
& SKIP_IF_UNCHANGED
) && !istate
->cache_changed
) {
3241 if (flags
& COMMIT_LOCK
)
3242 rollback_lock_file(lock
);
3246 if (istate
->fsmonitor_last_update
)
3247 fill_fsmonitor_bitmap(istate
);
3249 if (!si
|| alternate_index_output
||
3250 (istate
->cache_changed
& ~EXTMASK
)) {
3252 oidclr(&si
->base_oid
);
3253 ret
= do_write_locked_index(istate
, lock
, flags
);
3257 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0)) {
3258 int v
= si
->base_oid
.hash
[0];
3260 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3262 if (too_many_not_shared_entries(istate
))
3263 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3265 new_shared_index
= istate
->cache_changed
& SPLIT_INDEX_ORDERED
;
3267 if (new_shared_index
) {
3268 struct tempfile
*temp
;
3271 /* Same initial permissions as the main .git/index file */
3272 temp
= mks_tempfile_sm(git_path("sharedindex_XXXXXX"), 0, 0666);
3274 oidclr(&si
->base_oid
);
3275 ret
= do_write_locked_index(istate
, lock
, flags
);
3278 ret
= write_shared_index(istate
, &temp
);
3280 saved_errno
= errno
;
3281 if (is_tempfile_active(temp
))
3282 delete_tempfile(&temp
);
3283 errno
= saved_errno
;
3289 ret
= write_split_index(istate
, lock
, flags
);
3291 /* Freshen the shared index only if the split-index was written */
3292 if (!ret
&& !new_shared_index
&& !is_null_oid(&si
->base_oid
)) {
3293 const char *shared_index
= git_path("sharedindex.%s",
3294 oid_to_hex(&si
->base_oid
));
3295 freshen_shared_index(shared_index
, 1);
3299 if (flags
& COMMIT_LOCK
)
3300 rollback_lock_file(lock
);
3305 * Read the index file that is potentially unmerged into given
3306 * index_state, dropping any unmerged entries to stage #0 (potentially
3307 * resulting in a path appearing as both a file and a directory in the
3308 * index; the caller is responsible to clear out the extra entries
3309 * before writing the index to a tree). Returns true if the index is
3310 * unmerged. Callers who want to refuse to work from an unmerged
3311 * state can call this and check its return value, instead of calling
3314 int repo_read_index_unmerged(struct repository
*repo
)
3316 struct index_state
*istate
;
3320 repo_read_index(repo
);
3321 istate
= repo
->index
;
3322 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3323 struct cache_entry
*ce
= istate
->cache
[i
];
3324 struct cache_entry
*new_ce
;
3330 len
= ce_namelen(ce
);
3331 new_ce
= make_empty_cache_entry(istate
, len
);
3332 memcpy(new_ce
->name
, ce
->name
, len
);
3333 new_ce
->ce_flags
= create_ce_flags(0) | CE_CONFLICTED
;
3334 new_ce
->ce_namelen
= len
;
3335 new_ce
->ce_mode
= ce
->ce_mode
;
3336 if (add_index_entry(istate
, new_ce
, ADD_CACHE_SKIP_DFCHECK
))
3337 return error(_("%s: cannot drop to stage #0"),
3344 * Returns 1 if the path is an "other" path with respect to
3345 * the index; that is, the path is not mentioned in the index at all,
3346 * either as a file, a directory with some files in the index,
3347 * or as an unmerged entry.
3349 * We helpfully remove a trailing "/" from directories so that
3350 * the output of read_directory can be used as-is.
3352 int index_name_is_other(const struct index_state
*istate
, const char *name
,
3356 if (namelen
&& name
[namelen
- 1] == '/')
3358 pos
= index_name_pos(istate
, name
, namelen
);
3360 return 0; /* exact match */
3362 if (pos
< istate
->cache_nr
) {
3363 struct cache_entry
*ce
= istate
->cache
[pos
];
3364 if (ce_namelen(ce
) == namelen
&&
3365 !memcmp(ce
->name
, name
, namelen
))
3366 return 0; /* Yup, this one exists unmerged */
3371 void *read_blob_data_from_index(const struct index_state
*istate
,
3372 const char *path
, unsigned long *size
)
3376 enum object_type type
;
3380 pos
= index_name_pos(istate
, path
, len
);
3383 * We might be in the middle of a merge, in which
3384 * case we would read stage #2 (ours).
3388 (pos
< 0 && i
< istate
->cache_nr
&&
3389 !strcmp(istate
->cache
[i
]->name
, path
));
3391 if (ce_stage(istate
->cache
[i
]) == 2)
3396 data
= read_object_file(&istate
->cache
[pos
]->oid
, &type
, &sz
);
3397 if (!data
|| type
!= OBJ_BLOB
) {
3406 void stat_validity_clear(struct stat_validity
*sv
)
3408 FREE_AND_NULL(sv
->sd
);
3411 int stat_validity_check(struct stat_validity
*sv
, const char *path
)
3415 if (stat(path
, &st
) < 0)
3416 return sv
->sd
== NULL
;
3419 return S_ISREG(st
.st_mode
) && !match_stat_data(sv
->sd
, &st
);
3422 void stat_validity_update(struct stat_validity
*sv
, int fd
)
3426 if (fstat(fd
, &st
) < 0 || !S_ISREG(st
.st_mode
))
3427 stat_validity_clear(sv
);
3430 sv
->sd
= xcalloc(1, sizeof(struct stat_data
));
3431 fill_stat_data(sv
->sd
, &st
);
3435 void move_index_extensions(struct index_state
*dst
, struct index_state
*src
)
3437 dst
->untracked
= src
->untracked
;
3438 src
->untracked
= NULL
;
3439 dst
->cache_tree
= src
->cache_tree
;
3440 src
->cache_tree
= NULL
;
3443 struct cache_entry
*dup_cache_entry(const struct cache_entry
*ce
,
3444 struct index_state
*istate
)
3446 unsigned int size
= ce_size(ce
);
3447 int mem_pool_allocated
;
3448 struct cache_entry
*new_entry
= make_empty_cache_entry(istate
, ce_namelen(ce
));
3449 mem_pool_allocated
= new_entry
->mem_pool_allocated
;
3451 memcpy(new_entry
, ce
, size
);
3452 new_entry
->mem_pool_allocated
= mem_pool_allocated
;
3456 void discard_cache_entry(struct cache_entry
*ce
)
3458 if (ce
&& should_validate_cache_entries())
3459 memset(ce
, 0xCD, cache_entry_size(ce
->ce_namelen
));
3461 if (ce
&& ce
->mem_pool_allocated
)
3467 int should_validate_cache_entries(void)
3469 static int validate_index_cache_entries
= -1;
3471 if (validate_index_cache_entries
< 0) {
3472 if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))
3473 validate_index_cache_entries
= 1;
3475 validate_index_cache_entries
= 0;
3478 return validate_index_cache_entries
;
3481 #define EOIE_SIZE (4 + GIT_SHA1_RAWSZ) /* <4-byte offset> + <20-byte hash> */
3482 #define EOIE_SIZE_WITH_HEADER (4 + 4 + EOIE_SIZE) /* <4-byte signature> + <4-byte length> + EOIE_SIZE */
3484 static size_t read_eoie_extension(const char *mmap
, size_t mmap_size
)
3487 * The end of index entries (EOIE) extension is guaranteed to be last
3488 * so that it can be found by scanning backwards from the EOF.
3495 const char *index
, *eoie
;
3497 size_t offset
, src_offset
;
3498 unsigned char hash
[GIT_MAX_RAWSZ
];
3501 /* ensure we have an index big enough to contain an EOIE extension */
3502 if (mmap_size
< sizeof(struct cache_header
) + EOIE_SIZE_WITH_HEADER
+ the_hash_algo
->rawsz
)
3505 /* validate the extension signature */
3506 index
= eoie
= mmap
+ mmap_size
- EOIE_SIZE_WITH_HEADER
- the_hash_algo
->rawsz
;
3507 if (CACHE_EXT(index
) != CACHE_EXT_ENDOFINDEXENTRIES
)
3509 index
+= sizeof(uint32_t);
3511 /* validate the extension size */
3512 extsize
= get_be32(index
);
3513 if (extsize
!= EOIE_SIZE
)
3515 index
+= sizeof(uint32_t);
3518 * Validate the offset we're going to look for the first extension
3519 * signature is after the index header and before the eoie extension.
3521 offset
= get_be32(index
);
3522 if (mmap
+ offset
< mmap
+ sizeof(struct cache_header
))
3524 if (mmap
+ offset
>= eoie
)
3526 index
+= sizeof(uint32_t);
3529 * The hash is computed over extension types and their sizes (but not
3530 * their contents). E.g. if we have "TREE" extension that is N-bytes
3531 * long, "REUC" extension that is M-bytes long, followed by "EOIE",
3532 * then the hash would be:
3534 * SHA-1("TREE" + <binary representation of N> +
3535 * "REUC" + <binary representation of M>)
3537 src_offset
= offset
;
3538 the_hash_algo
->init_fn(&c
);
3539 while (src_offset
< mmap_size
- the_hash_algo
->rawsz
- EOIE_SIZE_WITH_HEADER
) {
3540 /* After an array of active_nr index entries,
3541 * there can be arbitrary number of extended
3542 * sections, each of which is prefixed with
3543 * extension name (4-byte) and section length
3544 * in 4-byte network byte order.
3547 memcpy(&extsize
, mmap
+ src_offset
+ 4, 4);
3548 extsize
= ntohl(extsize
);
3550 /* verify the extension size isn't so large it will wrap around */
3551 if (src_offset
+ 8 + extsize
< src_offset
)
3554 the_hash_algo
->update_fn(&c
, mmap
+ src_offset
, 8);
3557 src_offset
+= extsize
;
3559 the_hash_algo
->final_fn(hash
, &c
);
3560 if (!hasheq(hash
, (const unsigned char *)index
))
3563 /* Validate that the extension offsets returned us back to the eoie extension. */
3564 if (src_offset
!= mmap_size
- the_hash_algo
->rawsz
- EOIE_SIZE_WITH_HEADER
)
3570 static void write_eoie_extension(struct strbuf
*sb
, git_hash_ctx
*eoie_context
, size_t offset
)
3573 unsigned char hash
[GIT_MAX_RAWSZ
];
3576 put_be32(&buffer
, offset
);
3577 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3580 the_hash_algo
->final_fn(hash
, eoie_context
);
3581 strbuf_add(sb
, hash
, the_hash_algo
->rawsz
);
3584 #define IEOT_VERSION (1)
3586 static struct index_entry_offset_table
*read_ieot_extension(const char *mmap
, size_t mmap_size
, size_t offset
)
3588 const char *index
= NULL
;
3589 uint32_t extsize
, ext_version
;
3590 struct index_entry_offset_table
*ieot
;
3593 /* find the IEOT extension */
3596 while (offset
<= mmap_size
- the_hash_algo
->rawsz
- 8) {
3597 extsize
= get_be32(mmap
+ offset
+ 4);
3598 if (CACHE_EXT((mmap
+ offset
)) == CACHE_EXT_INDEXENTRYOFFSETTABLE
) {
3599 index
= mmap
+ offset
+ 4 + 4;
3608 /* validate the version is IEOT_VERSION */
3609 ext_version
= get_be32(index
);
3610 if (ext_version
!= IEOT_VERSION
) {
3611 error("invalid IEOT version %d", ext_version
);
3614 index
+= sizeof(uint32_t);
3616 /* extension size - version bytes / bytes per entry */
3617 nr
= (extsize
- sizeof(uint32_t)) / (sizeof(uint32_t) + sizeof(uint32_t));
3619 error("invalid number of IEOT entries %d", nr
);
3622 ieot
= xmalloc(sizeof(struct index_entry_offset_table
)
3623 + (nr
* sizeof(struct index_entry_offset
)));
3625 for (i
= 0; i
< nr
; i
++) {
3626 ieot
->entries
[i
].offset
= get_be32(index
);
3627 index
+= sizeof(uint32_t);
3628 ieot
->entries
[i
].nr
= get_be32(index
);
3629 index
+= sizeof(uint32_t);
3635 static void write_ieot_extension(struct strbuf
*sb
, struct index_entry_offset_table
*ieot
)
3641 put_be32(&buffer
, IEOT_VERSION
);
3642 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3645 for (i
= 0; i
< ieot
->nr
; i
++) {
3648 put_be32(&buffer
, ieot
->entries
[i
].offset
);
3649 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3652 put_be32(&buffer
, ieot
->entries
[i
].nr
);
3653 strbuf_add(sb
, &buffer
, sizeof(uint32_t));