2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
12 #include "cache-tree.h"
15 #include "object-store.h"
19 #include "resolve-undo.h"
20 #include "run-command.h"
23 #include "split-index.h"
25 #include "fsmonitor.h"
26 #include "thread-utils.h"
28 #include "sparse-index.h"
30 /* Mask for the name length in ce_flags in the on-disk index */
32 #define CE_NAMEMASK (0x0fff)
36 * The first letter should be 'A'..'Z' for extensions that are not
37 * necessary for a correct operation (i.e. optimization data).
38 * When new extensions are added that _needs_ to be understood in
39 * order to correctly interpret the index file, pick character that
40 * is outside the range, to cause the reader to abort.
43 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
44 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
45 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
46 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
47 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
48 #define CACHE_EXT_FSMONITOR 0x46534D4E /* "FSMN" */
49 #define CACHE_EXT_ENDOFINDEXENTRIES 0x454F4945 /* "EOIE" */
50 #define CACHE_EXT_INDEXENTRYOFFSETTABLE 0x49454F54 /* "IEOT" */
51 #define CACHE_EXT_SPARSE_DIRECTORIES 0x73646972 /* "sdir" */
53 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
54 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
55 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
56 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED | FSMONITOR_CHANGED)
60 * This is an estimate of the pathname length in the index. We use
61 * this for V4 index files to guess the un-deltafied size of the index
62 * in memory because of pathname deltafication. This is not required
63 * for V2/V3 index formats because their pathnames are not compressed.
64 * If the initial amount of memory set aside is not sufficient, the
65 * mem pool will allocate extra memory.
67 #define CACHE_ENTRY_PATH_LENGTH 80
69 static inline struct cache_entry
*mem_pool__ce_alloc(struct mem_pool
*mem_pool
, size_t len
)
71 struct cache_entry
*ce
;
72 ce
= mem_pool_alloc(mem_pool
, cache_entry_size(len
));
73 ce
->mem_pool_allocated
= 1;
77 static inline struct cache_entry
*mem_pool__ce_calloc(struct mem_pool
*mem_pool
, size_t len
)
79 struct cache_entry
* ce
;
80 ce
= mem_pool_calloc(mem_pool
, 1, cache_entry_size(len
));
81 ce
->mem_pool_allocated
= 1;
85 static struct mem_pool
*find_mem_pool(struct index_state
*istate
)
87 struct mem_pool
**pool_ptr
;
89 if (istate
->split_index
&& istate
->split_index
->base
)
90 pool_ptr
= &istate
->split_index
->base
->ce_mem_pool
;
92 pool_ptr
= &istate
->ce_mem_pool
;
95 *pool_ptr
= xmalloc(sizeof(**pool_ptr
));
96 mem_pool_init(*pool_ptr
, 0);
102 static const char *alternate_index_output
;
104 static void set_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
106 if (S_ISSPARSEDIR(ce
->ce_mode
))
107 istate
->sparse_index
= 1;
109 istate
->cache
[nr
] = ce
;
110 add_name_hash(istate
, ce
);
113 static void replace_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
115 struct cache_entry
*old
= istate
->cache
[nr
];
117 replace_index_entry_in_base(istate
, old
, ce
);
118 remove_name_hash(istate
, old
);
119 discard_cache_entry(old
);
120 ce
->ce_flags
&= ~CE_HASHED
;
121 set_index_entry(istate
, nr
, ce
);
122 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
123 mark_fsmonitor_invalid(istate
, ce
);
124 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
127 void rename_index_entry_at(struct index_state
*istate
, int nr
, const char *new_name
)
129 struct cache_entry
*old_entry
= istate
->cache
[nr
], *new_entry
;
130 int namelen
= strlen(new_name
);
132 new_entry
= make_empty_cache_entry(istate
, namelen
);
133 copy_cache_entry(new_entry
, old_entry
);
134 new_entry
->ce_flags
&= ~CE_HASHED
;
135 new_entry
->ce_namelen
= namelen
;
136 new_entry
->index
= 0;
137 memcpy(new_entry
->name
, new_name
, namelen
+ 1);
139 cache_tree_invalidate_path(istate
, old_entry
->name
);
140 untracked_cache_remove_from_index(istate
, old_entry
->name
);
141 remove_index_entry_at(istate
, nr
);
142 add_index_entry(istate
, new_entry
, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
145 void fill_stat_data(struct stat_data
*sd
, struct stat
*st
)
147 sd
->sd_ctime
.sec
= (unsigned int)st
->st_ctime
;
148 sd
->sd_mtime
.sec
= (unsigned int)st
->st_mtime
;
149 sd
->sd_ctime
.nsec
= ST_CTIME_NSEC(*st
);
150 sd
->sd_mtime
.nsec
= ST_MTIME_NSEC(*st
);
151 sd
->sd_dev
= st
->st_dev
;
152 sd
->sd_ino
= st
->st_ino
;
153 sd
->sd_uid
= st
->st_uid
;
154 sd
->sd_gid
= st
->st_gid
;
155 sd
->sd_size
= st
->st_size
;
158 int match_stat_data(const struct stat_data
*sd
, struct stat
*st
)
162 if (sd
->sd_mtime
.sec
!= (unsigned int)st
->st_mtime
)
163 changed
|= MTIME_CHANGED
;
164 if (trust_ctime
&& check_stat
&&
165 sd
->sd_ctime
.sec
!= (unsigned int)st
->st_ctime
)
166 changed
|= CTIME_CHANGED
;
169 if (check_stat
&& sd
->sd_mtime
.nsec
!= ST_MTIME_NSEC(*st
))
170 changed
|= MTIME_CHANGED
;
171 if (trust_ctime
&& check_stat
&&
172 sd
->sd_ctime
.nsec
!= ST_CTIME_NSEC(*st
))
173 changed
|= CTIME_CHANGED
;
177 if (sd
->sd_uid
!= (unsigned int) st
->st_uid
||
178 sd
->sd_gid
!= (unsigned int) st
->st_gid
)
179 changed
|= OWNER_CHANGED
;
180 if (sd
->sd_ino
!= (unsigned int) st
->st_ino
)
181 changed
|= INODE_CHANGED
;
186 * st_dev breaks on network filesystems where different
187 * clients will have different views of what "device"
188 * the filesystem is on
190 if (check_stat
&& sd
->sd_dev
!= (unsigned int) st
->st_dev
)
191 changed
|= INODE_CHANGED
;
194 if (sd
->sd_size
!= (unsigned int) st
->st_size
)
195 changed
|= DATA_CHANGED
;
201 * This only updates the "non-critical" parts of the directory
202 * cache, ie the parts that aren't tracked by GIT, and only used
203 * to validate the cache.
205 void fill_stat_cache_info(struct index_state
*istate
, struct cache_entry
*ce
, struct stat
*st
)
207 fill_stat_data(&ce
->ce_stat_data
, st
);
209 if (assume_unchanged
)
210 ce
->ce_flags
|= CE_VALID
;
212 if (S_ISREG(st
->st_mode
)) {
213 ce_mark_uptodate(ce
);
214 mark_fsmonitor_valid(istate
, ce
);
218 static int ce_compare_data(struct index_state
*istate
,
219 const struct cache_entry
*ce
,
223 int fd
= git_open_cloexec(ce
->name
, O_RDONLY
);
226 struct object_id oid
;
227 if (!index_fd(istate
, &oid
, fd
, st
, OBJ_BLOB
, ce
->name
, 0))
228 match
= !oideq(&oid
, &ce
->oid
);
229 /* index_fd() closed the file descriptor already */
234 static int ce_compare_link(const struct cache_entry
*ce
, size_t expected_size
)
239 enum object_type type
;
240 struct strbuf sb
= STRBUF_INIT
;
242 if (strbuf_readlink(&sb
, ce
->name
, expected_size
))
245 buffer
= read_object_file(&ce
->oid
, &type
, &size
);
248 match
= memcmp(buffer
, sb
.buf
, size
);
255 static int ce_compare_gitlink(const struct cache_entry
*ce
)
257 struct object_id oid
;
260 * We don't actually require that the .git directory
261 * under GITLINK directory be a valid git directory. It
262 * might even be missing (in case nobody populated that
265 * If so, we consider it always to match.
267 if (resolve_gitlink_ref(ce
->name
, "HEAD", &oid
) < 0)
269 return !oideq(&oid
, &ce
->oid
);
272 static int ce_modified_check_fs(struct index_state
*istate
,
273 const struct cache_entry
*ce
,
276 switch (st
->st_mode
& S_IFMT
) {
278 if (ce_compare_data(istate
, ce
, st
))
282 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
286 if (S_ISGITLINK(ce
->ce_mode
))
287 return ce_compare_gitlink(ce
) ? DATA_CHANGED
: 0;
288 /* else fallthrough */
295 static int ce_match_stat_basic(const struct cache_entry
*ce
, struct stat
*st
)
297 unsigned int changed
= 0;
299 if (ce
->ce_flags
& CE_REMOVE
)
300 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
302 switch (ce
->ce_mode
& S_IFMT
) {
304 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
305 /* We consider only the owner x bit to be relevant for
308 if (trust_executable_bit
&&
309 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
310 changed
|= MODE_CHANGED
;
313 if (!S_ISLNK(st
->st_mode
) &&
314 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
315 changed
|= TYPE_CHANGED
;
318 /* We ignore most of the st_xxx fields for gitlinks */
319 if (!S_ISDIR(st
->st_mode
))
320 changed
|= TYPE_CHANGED
;
321 else if (ce_compare_gitlink(ce
))
322 changed
|= DATA_CHANGED
;
325 BUG("unsupported ce_mode: %o", ce
->ce_mode
);
328 changed
|= match_stat_data(&ce
->ce_stat_data
, st
);
330 /* Racily smudged entry? */
331 if (!ce
->ce_stat_data
.sd_size
) {
332 if (!is_empty_blob_sha1(ce
->oid
.hash
))
333 changed
|= DATA_CHANGED
;
339 static int is_racy_stat(const struct index_state
*istate
,
340 const struct stat_data
*sd
)
342 return (istate
->timestamp
.sec
&&
344 /* nanosecond timestamped files can also be racy! */
345 (istate
->timestamp
.sec
< sd
->sd_mtime
.sec
||
346 (istate
->timestamp
.sec
== sd
->sd_mtime
.sec
&&
347 istate
->timestamp
.nsec
<= sd
->sd_mtime
.nsec
))
349 istate
->timestamp
.sec
<= sd
->sd_mtime
.sec
354 int is_racy_timestamp(const struct index_state
*istate
,
355 const struct cache_entry
*ce
)
357 return (!S_ISGITLINK(ce
->ce_mode
) &&
358 is_racy_stat(istate
, &ce
->ce_stat_data
));
361 int match_stat_data_racy(const struct index_state
*istate
,
362 const struct stat_data
*sd
, struct stat
*st
)
364 if (is_racy_stat(istate
, sd
))
365 return MTIME_CHANGED
;
366 return match_stat_data(sd
, st
);
369 int ie_match_stat(struct index_state
*istate
,
370 const struct cache_entry
*ce
, struct stat
*st
,
371 unsigned int options
)
373 unsigned int changed
;
374 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
375 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
376 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
377 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
379 if (!ignore_fsmonitor
)
380 refresh_fsmonitor(istate
);
382 * If it's marked as always valid in the index, it's
383 * valid whatever the checked-out copy says.
385 * skip-worktree has the same effect with higher precedence
387 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
))
389 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
391 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
))
395 * Intent-to-add entries have not been added, so the index entry
396 * by definition never matches what is in the work tree until it
397 * actually gets added.
399 if (ce_intent_to_add(ce
))
400 return DATA_CHANGED
| TYPE_CHANGED
| MODE_CHANGED
;
402 changed
= ce_match_stat_basic(ce
, st
);
405 * Within 1 second of this sequence:
406 * echo xyzzy >file && git-update-index --add file
407 * running this command:
409 * would give a falsely clean cache entry. The mtime and
410 * length match the cache, and other stat fields do not change.
412 * We could detect this at update-index time (the cache entry
413 * being registered/updated records the same time as "now")
414 * and delay the return from git-update-index, but that would
415 * effectively mean we can make at most one commit per second,
416 * which is not acceptable. Instead, we check cache entries
417 * whose mtime are the same as the index file timestamp more
418 * carefully than others.
420 if (!changed
&& is_racy_timestamp(istate
, ce
)) {
421 if (assume_racy_is_modified
)
422 changed
|= DATA_CHANGED
;
424 changed
|= ce_modified_check_fs(istate
, ce
, st
);
430 int ie_modified(struct index_state
*istate
,
431 const struct cache_entry
*ce
,
432 struct stat
*st
, unsigned int options
)
434 int changed
, changed_fs
;
436 changed
= ie_match_stat(istate
, ce
, st
, options
);
440 * If the mode or type has changed, there's no point in trying
441 * to refresh the entry - it's not going to match
443 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
447 * Immediately after read-tree or update-index --cacheinfo,
448 * the length field is zero, as we have never even read the
449 * lstat(2) information once, and we cannot trust DATA_CHANGED
450 * returned by ie_match_stat() which in turn was returned by
451 * ce_match_stat_basic() to signal that the filesize of the
452 * blob changed. We have to actually go to the filesystem to
453 * see if the contents match, and if so, should answer "unchanged".
455 * The logic does not apply to gitlinks, as ce_match_stat_basic()
456 * already has checked the actual HEAD from the filesystem in the
457 * subproject. If ie_match_stat() already said it is different,
458 * then we know it is.
460 if ((changed
& DATA_CHANGED
) &&
461 (S_ISGITLINK(ce
->ce_mode
) || ce
->ce_stat_data
.sd_size
!= 0))
464 changed_fs
= ce_modified_check_fs(istate
, ce
, st
);
466 return changed
| changed_fs
;
470 int base_name_compare(const char *name1
, int len1
, int mode1
,
471 const char *name2
, int len2
, int mode2
)
473 unsigned char c1
, c2
;
474 int len
= len1
< len2
? len1
: len2
;
477 cmp
= memcmp(name1
, name2
, len
);
482 if (!c1
&& S_ISDIR(mode1
))
484 if (!c2
&& S_ISDIR(mode2
))
486 return (c1
< c2
) ? -1 : (c1
> c2
) ? 1 : 0;
490 * df_name_compare() is identical to base_name_compare(), except it
491 * compares conflicting directory/file entries as equal. Note that
492 * while a directory name compares as equal to a regular file, they
493 * then individually compare _differently_ to a filename that has
494 * a dot after the basename (because '\0' < '.' < '/').
496 * This is used by routines that want to traverse the git namespace
497 * but then handle conflicting entries together when possible.
499 int df_name_compare(const char *name1
, int len1
, int mode1
,
500 const char *name2
, int len2
, int mode2
)
502 int len
= len1
< len2
? len1
: len2
, cmp
;
503 unsigned char c1
, c2
;
505 cmp
= memcmp(name1
, name2
, len
);
508 /* Directories and files compare equal (same length, same name) */
512 if (!c1
&& S_ISDIR(mode1
))
515 if (!c2
&& S_ISDIR(mode2
))
517 if (c1
== '/' && !c2
)
519 if (c2
== '/' && !c1
)
524 int name_compare(const char *name1
, size_t len1
, const char *name2
, size_t len2
)
526 size_t min_len
= (len1
< len2
) ? len1
: len2
;
527 int cmp
= memcmp(name1
, name2
, min_len
);
537 int cache_name_stage_compare(const char *name1
, int len1
, int stage1
, const char *name2
, int len2
, int stage2
)
541 cmp
= name_compare(name1
, len1
, name2
, len2
);
552 static int index_name_stage_pos(struct index_state
*istate
, const char *name
, int namelen
, int stage
)
557 last
= istate
->cache_nr
;
558 while (last
> first
) {
559 int next
= first
+ ((last
- first
) >> 1);
560 struct cache_entry
*ce
= istate
->cache
[next
];
561 int cmp
= cache_name_stage_compare(name
, namelen
, stage
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
571 if (istate
->sparse_index
&&
573 /* Note: first <= istate->cache_nr */
574 struct cache_entry
*ce
= istate
->cache
[first
- 1];
577 * If we are in a sparse-index _and_ the entry before the
578 * insertion position is a sparse-directory entry that is
579 * an ancestor of 'name', then we need to expand the index
580 * and search again. This will only trigger once, because
581 * thereafter the index is fully expanded.
583 if (S_ISSPARSEDIR(ce
->ce_mode
) &&
584 ce_namelen(ce
) < namelen
&&
585 !strncmp(name
, ce
->name
, ce_namelen(ce
))) {
586 ensure_full_index(istate
);
587 return index_name_stage_pos(istate
, name
, namelen
, stage
);
594 int index_name_pos(struct index_state
*istate
, const char *name
, int namelen
)
596 return index_name_stage_pos(istate
, name
, namelen
, 0);
599 int remove_index_entry_at(struct index_state
*istate
, int pos
)
601 struct cache_entry
*ce
= istate
->cache
[pos
];
603 record_resolve_undo(istate
, ce
);
604 remove_name_hash(istate
, ce
);
605 save_or_free_index_entry(istate
, ce
);
606 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
608 if (pos
>= istate
->cache_nr
)
610 MOVE_ARRAY(istate
->cache
+ pos
, istate
->cache
+ pos
+ 1,
611 istate
->cache_nr
- pos
);
616 * Remove all cache entries marked for removal, that is where
617 * CE_REMOVE is set in ce_flags. This is much more effective than
618 * calling remove_index_entry_at() for each entry to be removed.
620 void remove_marked_cache_entries(struct index_state
*istate
, int invalidate
)
622 struct cache_entry
**ce_array
= istate
->cache
;
625 for (i
= j
= 0; i
< istate
->cache_nr
; i
++) {
626 if (ce_array
[i
]->ce_flags
& CE_REMOVE
) {
628 cache_tree_invalidate_path(istate
,
630 untracked_cache_remove_from_index(istate
,
633 remove_name_hash(istate
, ce_array
[i
]);
634 save_or_free_index_entry(istate
, ce_array
[i
]);
637 ce_array
[j
++] = ce_array
[i
];
639 if (j
== istate
->cache_nr
)
641 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
642 istate
->cache_nr
= j
;
645 int remove_file_from_index(struct index_state
*istate
, const char *path
)
647 int pos
= index_name_pos(istate
, path
, strlen(path
));
650 cache_tree_invalidate_path(istate
, path
);
651 untracked_cache_remove_from_index(istate
, path
);
652 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
653 remove_index_entry_at(istate
, pos
);
657 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
659 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
662 static int index_name_pos_also_unmerged(struct index_state
*istate
,
663 const char *path
, int namelen
)
665 int pos
= index_name_pos(istate
, path
, namelen
);
666 struct cache_entry
*ce
;
671 /* maybe unmerged? */
673 if (pos
>= istate
->cache_nr
||
674 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
677 /* order of preference: stage 2, 1, 3 */
678 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
679 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
680 !compare_name(ce
, path
, namelen
))
685 static int different_name(struct cache_entry
*ce
, struct cache_entry
*alias
)
687 int len
= ce_namelen(ce
);
688 return ce_namelen(alias
) != len
|| memcmp(ce
->name
, alias
->name
, len
);
692 * If we add a filename that aliases in the cache, we will use the
693 * name that we already have - but we don't want to update the same
694 * alias twice, because that implies that there were actually two
695 * different files with aliasing names!
697 * So we use the CE_ADDED flag to verify that the alias was an old
698 * one before we accept it as
700 static struct cache_entry
*create_alias_ce(struct index_state
*istate
,
701 struct cache_entry
*ce
,
702 struct cache_entry
*alias
)
705 struct cache_entry
*new_entry
;
707 if (alias
->ce_flags
& CE_ADDED
)
708 die(_("will not add file alias '%s' ('%s' already exists in index)"),
709 ce
->name
, alias
->name
);
711 /* Ok, create the new entry using the name of the existing alias */
712 len
= ce_namelen(alias
);
713 new_entry
= make_empty_cache_entry(istate
, len
);
714 memcpy(new_entry
->name
, alias
->name
, len
);
715 copy_cache_entry(new_entry
, ce
);
716 save_or_free_index_entry(istate
, ce
);
720 void set_object_name_for_intent_to_add_entry(struct cache_entry
*ce
)
722 struct object_id oid
;
723 if (write_object_file("", 0, blob_type
, &oid
))
724 die(_("cannot create an empty blob in the object database"));
725 oidcpy(&ce
->oid
, &oid
);
728 int add_to_index(struct index_state
*istate
, const char *path
, struct stat
*st
, int flags
)
730 int namelen
, was_same
;
731 mode_t st_mode
= st
->st_mode
;
732 struct cache_entry
*ce
, *alias
= NULL
;
733 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_IGNORE_SKIP_WORKTREE
|CE_MATCH_RACY_IS_DIRTY
;
734 int verbose
= flags
& (ADD_CACHE_VERBOSE
| ADD_CACHE_PRETEND
);
735 int pretend
= flags
& ADD_CACHE_PRETEND
;
736 int intent_only
= flags
& ADD_CACHE_INTENT
;
737 int add_option
= (ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
|
738 (intent_only
? ADD_CACHE_NEW_ONLY
: 0));
739 int hash_flags
= HASH_WRITE_OBJECT
;
740 struct object_id oid
;
742 if (flags
& ADD_CACHE_RENORMALIZE
)
743 hash_flags
|= HASH_RENORMALIZE
;
745 if (!S_ISREG(st_mode
) && !S_ISLNK(st_mode
) && !S_ISDIR(st_mode
))
746 return error(_("%s: can only add regular files, symbolic links or git-directories"), path
);
748 namelen
= strlen(path
);
749 if (S_ISDIR(st_mode
)) {
750 if (resolve_gitlink_ref(path
, "HEAD", &oid
) < 0)
751 return error(_("'%s' does not have a commit checked out"), path
);
752 while (namelen
&& path
[namelen
-1] == '/')
755 ce
= make_empty_cache_entry(istate
, namelen
);
756 memcpy(ce
->name
, path
, namelen
);
757 ce
->ce_namelen
= namelen
;
759 fill_stat_cache_info(istate
, ce
, st
);
761 ce
->ce_flags
|= CE_INTENT_TO_ADD
;
764 if (trust_executable_bit
&& has_symlinks
) {
765 ce
->ce_mode
= create_ce_mode(st_mode
);
767 /* If there is an existing entry, pick the mode bits and type
768 * from it, otherwise assume unexecutable regular file.
770 struct cache_entry
*ent
;
771 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
773 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
774 ce
->ce_mode
= ce_mode_from_stat(ent
, st_mode
);
777 /* When core.ignorecase=true, determine if a directory of the same name but differing
778 * case already exists within the Git repository. If it does, ensure the directory
779 * case of the file being added to the repository matches (is folded into) the existing
780 * entry's directory case.
783 adjust_dirname_case(istate
, ce
->name
);
785 if (!(flags
& ADD_CACHE_RENORMALIZE
)) {
786 alias
= index_file_exists(istate
, ce
->name
,
787 ce_namelen(ce
), ignore_case
);
790 !ie_match_stat(istate
, alias
, st
, ce_option
)) {
791 /* Nothing changed, really */
792 if (!S_ISGITLINK(alias
->ce_mode
))
793 ce_mark_uptodate(alias
);
794 alias
->ce_flags
|= CE_ADDED
;
796 discard_cache_entry(ce
);
801 if (index_path(istate
, &ce
->oid
, path
, st
, hash_flags
)) {
802 discard_cache_entry(ce
);
803 return error(_("unable to index file '%s'"), path
);
806 set_object_name_for_intent_to_add_entry(ce
);
808 if (ignore_case
&& alias
&& different_name(ce
, alias
))
809 ce
= create_alias_ce(istate
, ce
, alias
);
810 ce
->ce_flags
|= CE_ADDED
;
812 /* It was suspected to be racily clean, but it turns out to be Ok */
815 oideq(&alias
->oid
, &ce
->oid
) &&
816 ce
->ce_mode
== alias
->ce_mode
);
819 discard_cache_entry(ce
);
820 else if (add_index_entry(istate
, ce
, add_option
)) {
821 discard_cache_entry(ce
);
822 return error(_("unable to add '%s' to index"), path
);
824 if (verbose
&& !was_same
)
825 printf("add '%s'\n", path
);
829 int add_file_to_index(struct index_state
*istate
, const char *path
, int flags
)
832 if (lstat(path
, &st
))
833 die_errno(_("unable to stat '%s'"), path
);
834 return add_to_index(istate
, path
, &st
, flags
);
837 struct cache_entry
*make_empty_cache_entry(struct index_state
*istate
, size_t len
)
839 return mem_pool__ce_calloc(find_mem_pool(istate
), len
);
842 struct cache_entry
*make_empty_transient_cache_entry(size_t len
)
844 return xcalloc(1, cache_entry_size(len
));
847 enum verify_path_result
{
853 static enum verify_path_result
verify_path_internal(const char *, unsigned);
855 int verify_path(const char *path
, unsigned mode
)
857 return verify_path_internal(path
, mode
) != PATH_INVALID
;
860 struct cache_entry
*make_cache_entry(struct index_state
*istate
,
862 const struct object_id
*oid
,
865 unsigned int refresh_options
)
867 struct cache_entry
*ce
, *ret
;
870 if (!verify_path(path
, mode
)) {
871 error(_("invalid path '%s'"), path
);
876 ce
= make_empty_cache_entry(istate
, len
);
878 oidcpy(&ce
->oid
, oid
);
879 memcpy(ce
->name
, path
, len
);
880 ce
->ce_flags
= create_ce_flags(stage
);
881 ce
->ce_namelen
= len
;
882 ce
->ce_mode
= create_ce_mode(mode
);
884 ret
= refresh_cache_entry(istate
, ce
, refresh_options
);
886 discard_cache_entry(ce
);
890 struct cache_entry
*make_transient_cache_entry(unsigned int mode
, const struct object_id
*oid
,
891 const char *path
, int stage
)
893 struct cache_entry
*ce
;
896 if (!verify_path(path
, mode
)) {
897 error(_("invalid path '%s'"), path
);
902 ce
= make_empty_transient_cache_entry(len
);
904 oidcpy(&ce
->oid
, oid
);
905 memcpy(ce
->name
, path
, len
);
906 ce
->ce_flags
= create_ce_flags(stage
);
907 ce
->ce_namelen
= len
;
908 ce
->ce_mode
= create_ce_mode(mode
);
914 * Chmod an index entry with either +x or -x.
916 * Returns -1 if the chmod for the particular cache entry failed (if it's
917 * not a regular file), -2 if an invalid flip argument is passed in, 0
920 int chmod_index_entry(struct index_state
*istate
, struct cache_entry
*ce
,
923 if (!S_ISREG(ce
->ce_mode
))
930 ce
->ce_mode
&= ~0111;
935 cache_tree_invalidate_path(istate
, ce
->name
);
936 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
937 mark_fsmonitor_invalid(istate
, ce
);
938 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
943 int ce_same_name(const struct cache_entry
*a
, const struct cache_entry
*b
)
945 int len
= ce_namelen(a
);
946 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
950 * We fundamentally don't like some paths: we don't want
951 * dot or dot-dot anywhere, and for obvious reasons don't
952 * want to recurse into ".git" either.
954 * Also, we don't want double slashes or slashes at the
955 * end that can make pathnames ambiguous.
957 static int verify_dotfile(const char *rest
, unsigned mode
)
960 * The first character was '.', but that
961 * has already been discarded, we now test
965 /* "." is not allowed */
966 if (*rest
== '\0' || is_dir_sep(*rest
))
971 * ".git" followed by NUL or slash is bad. Note that we match
972 * case-insensitively here, even if ignore_case is not set.
973 * This outlaws ".GIT" everywhere out of an abundance of caution,
974 * since there's really no good reason to allow it.
976 * Once we've seen ".git", we can also find ".gitmodules", etc (also
977 * case-insensitively).
981 if (rest
[1] != 'i' && rest
[1] != 'I')
983 if (rest
[2] != 't' && rest
[2] != 'T')
985 if (rest
[3] == '\0' || is_dir_sep(rest
[3]))
989 if (skip_iprefix(rest
, "modules", &rest
) &&
990 (*rest
== '\0' || is_dir_sep(*rest
)))
995 if (rest
[1] == '\0' || is_dir_sep(rest
[1]))
1001 static enum verify_path_result
verify_path_internal(const char *path
,
1006 if (has_dos_drive_prefix(path
))
1007 return PATH_INVALID
;
1009 if (!is_valid_path(path
))
1010 return PATH_INVALID
;
1016 if (is_dir_sep(c
)) {
1020 if (is_hfs_dotgit(path
))
1021 return PATH_INVALID
;
1022 if (S_ISLNK(mode
)) {
1023 if (is_hfs_dotgitmodules(path
))
1024 return PATH_INVALID
;
1028 #ifdef GIT_WINDOWS_NATIVE
1030 return PATH_INVALID
;
1032 if (is_ntfs_dotgit(path
))
1033 return PATH_INVALID
;
1034 if (S_ISLNK(mode
)) {
1035 if (is_ntfs_dotgitmodules(path
))
1036 return PATH_INVALID
;
1041 if ((c
== '.' && !verify_dotfile(path
, mode
)) ||
1043 return PATH_INVALID
;
1045 * allow terminating directory separators for
1046 * sparse directory entries.
1049 return S_ISDIR(mode
) ? PATH_DIR_WITH_SEP
:
1051 } else if (c
== '\\' && protect_ntfs
) {
1052 if (is_ntfs_dotgit(path
))
1053 return PATH_INVALID
;
1054 if (S_ISLNK(mode
)) {
1055 if (is_ntfs_dotgitmodules(path
))
1056 return PATH_INVALID
;
1065 * Do we have another file that has the beginning components being a
1066 * proper superset of the name we're trying to add?
1068 static int has_file_name(struct index_state
*istate
,
1069 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
1072 int len
= ce_namelen(ce
);
1073 int stage
= ce_stage(ce
);
1074 const char *name
= ce
->name
;
1076 while (pos
< istate
->cache_nr
) {
1077 struct cache_entry
*p
= istate
->cache
[pos
++];
1079 if (len
>= ce_namelen(p
))
1081 if (memcmp(name
, p
->name
, len
))
1083 if (ce_stage(p
) != stage
)
1085 if (p
->name
[len
] != '/')
1087 if (p
->ce_flags
& CE_REMOVE
)
1092 remove_index_entry_at(istate
, --pos
);
1099 * Like strcmp(), but also return the offset of the first change.
1100 * If strings are equal, return the length.
1102 int strcmp_offset(const char *s1
, const char *s2
, size_t *first_change
)
1107 return strcmp(s1
, s2
);
1109 for (k
= 0; s1
[k
] == s2
[k
]; k
++)
1114 return (unsigned char)s1
[k
] - (unsigned char)s2
[k
];
1118 * Do we have another file with a pathname that is a proper
1119 * subset of the name we're trying to add?
1121 * That is, is there another file in the index with a path
1122 * that matches a sub-directory in the given entry?
1124 static int has_dir_name(struct index_state
*istate
,
1125 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
1128 int stage
= ce_stage(ce
);
1129 const char *name
= ce
->name
;
1130 const char *slash
= name
+ ce_namelen(ce
);
1135 * We are frequently called during an iteration on a sorted
1136 * list of pathnames and while building a new index. Therefore,
1137 * there is a high probability that this entry will eventually
1138 * be appended to the index, rather than inserted in the middle.
1139 * If we can confirm that, we can avoid binary searches on the
1140 * components of the pathname.
1142 * Compare the entry's full path with the last path in the index.
1144 if (istate
->cache_nr
> 0) {
1145 cmp_last
= strcmp_offset(name
,
1146 istate
->cache
[istate
->cache_nr
- 1]->name
,
1149 if (len_eq_last
== 0) {
1151 * The entry sorts AFTER the last one in the
1152 * index and their paths have no common prefix,
1153 * so there cannot be a F/D conflict.
1158 * The entry sorts AFTER the last one in the
1159 * index, but has a common prefix. Fall through
1160 * to the loop below to disect the entry's path
1161 * and see where the difference is.
1164 } else if (cmp_last
== 0) {
1166 * The entry exactly matches the last one in the
1167 * index, but because of multiple stage and CE_REMOVE
1168 * items, we fall through and let the regular search
1178 if (*--slash
== '/')
1180 if (slash
<= ce
->name
)
1187 * (len + 1) is a directory boundary (including
1188 * the trailing slash). And since the loop is
1189 * decrementing "slash", the first iteration is
1190 * the longest directory prefix; subsequent
1191 * iterations consider parent directories.
1194 if (len
+ 1 <= len_eq_last
) {
1196 * The directory prefix (including the trailing
1197 * slash) also appears as a prefix in the last
1198 * entry, so the remainder cannot collide (because
1199 * strcmp said the whole path was greater).
1204 * LT: last: xxx/file_A
1210 if (len
> len_eq_last
) {
1212 * This part of the directory prefix (excluding
1213 * the trailing slash) is longer than the known
1214 * equal portions, so this sub-directory cannot
1215 * collide with a file.
1224 * This is a possible collision. Fall through and
1225 * let the regular search code handle it.
1232 pos
= index_name_stage_pos(istate
, name
, len
, stage
);
1235 * Found one, but not so fast. This could
1236 * be a marker that says "I was here, but
1237 * I am being removed". Such an entry is
1238 * not a part of the resulting tree, and
1239 * it is Ok to have a directory at the same
1242 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
1246 remove_index_entry_at(istate
, pos
);
1254 * Trivial optimization: if we find an entry that
1255 * already matches the sub-directory, then we know
1256 * we're ok, and we can exit.
1258 while (pos
< istate
->cache_nr
) {
1259 struct cache_entry
*p
= istate
->cache
[pos
];
1260 if ((ce_namelen(p
) <= len
) ||
1261 (p
->name
[len
] != '/') ||
1262 memcmp(p
->name
, name
, len
))
1263 break; /* not our subdirectory */
1264 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
1266 * p is at the same stage as our entry, and
1267 * is a subdirectory of what we are looking
1268 * at, so we cannot have conflicts at our
1269 * level or anything shorter.
1278 /* We may be in a situation where we already have path/file and path
1279 * is being added, or we already have path and path/file is being
1280 * added. Either one would result in a nonsense tree that has path
1281 * twice when git-write-tree tries to write it out. Prevent it.
1283 * If ok-to-replace is specified, we remove the conflicting entries
1284 * from the cache so the caller should recompute the insert position.
1285 * When this happens, we return non-zero.
1287 static int check_file_directory_conflict(struct index_state
*istate
,
1288 const struct cache_entry
*ce
,
1289 int pos
, int ok_to_replace
)
1294 * When ce is an "I am going away" entry, we allow it to be added
1296 if (ce
->ce_flags
& CE_REMOVE
)
1300 * We check if the path is a sub-path of a subsequent pathname
1301 * first, since removing those will not change the position
1304 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
1307 * Then check if the path might have a clashing sub-directory
1310 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
1313 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1316 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
1317 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
1318 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
1319 int new_only
= option
& ADD_CACHE_NEW_ONLY
;
1321 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1322 cache_tree_invalidate_path(istate
, ce
->name
);
1325 * If this entry's path sorts after the last entry in the index,
1326 * we can avoid searching for it.
1328 if (istate
->cache_nr
> 0 &&
1329 strcmp(ce
->name
, istate
->cache
[istate
->cache_nr
- 1]->name
) > 0)
1330 pos
= index_pos_to_insert_pos(istate
->cache_nr
);
1332 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1334 /* existing match? Just replace it. */
1337 replace_index_entry(istate
, pos
, ce
);
1342 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1343 untracked_cache_add_to_index(istate
, ce
->name
);
1346 * Inserting a merged entry ("stage 0") into the index
1347 * will always replace all non-merged entries..
1349 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
1350 while (ce_same_name(istate
->cache
[pos
], ce
)) {
1352 if (!remove_index_entry_at(istate
, pos
))
1359 if (!verify_path(ce
->name
, ce
->ce_mode
))
1360 return error(_("invalid path '%s'"), ce
->name
);
1362 if (!skip_df_check
&&
1363 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
1365 return error(_("'%s' appears as both a file and as a directory"),
1367 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1373 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1377 if (option
& ADD_CACHE_JUST_APPEND
)
1378 pos
= istate
->cache_nr
;
1381 ret
= add_index_entry_with_check(istate
, ce
, option
);
1387 /* Make sure the array is big enough .. */
1388 ALLOC_GROW(istate
->cache
, istate
->cache_nr
+ 1, istate
->cache_alloc
);
1392 if (istate
->cache_nr
> pos
+ 1)
1393 MOVE_ARRAY(istate
->cache
+ pos
+ 1, istate
->cache
+ pos
,
1394 istate
->cache_nr
- pos
- 1);
1395 set_index_entry(istate
, pos
, ce
);
1396 istate
->cache_changed
|= CE_ENTRY_ADDED
;
1401 * "refresh" does not calculate a new sha1 file or bring the
1402 * cache up-to-date for mode/content changes. But what it
1403 * _does_ do is to "re-match" the stat information of a file
1404 * with the cache, so that you can refresh the cache for a
1405 * file that hasn't been changed but where the stat entry is
1408 * For example, you'd want to do this after doing a "git-read-tree",
1409 * to link up the stat cache details with the proper files.
1411 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
1412 struct cache_entry
*ce
,
1413 unsigned int options
, int *err
,
1419 struct cache_entry
*updated
;
1421 int refresh
= options
& CE_MATCH_REFRESH
;
1422 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
1423 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
1424 int ignore_missing
= options
& CE_MATCH_IGNORE_MISSING
;
1425 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
1427 if (!refresh
|| ce_uptodate(ce
))
1430 if (!ignore_fsmonitor
)
1431 refresh_fsmonitor(istate
);
1433 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1434 * that the change to the work tree does not matter and told
1437 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
)) {
1438 ce_mark_uptodate(ce
);
1441 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
)) {
1442 ce_mark_uptodate(ce
);
1445 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
)) {
1446 ce_mark_uptodate(ce
);
1450 if (has_symlink_leading_path(ce
->name
, ce_namelen(ce
))) {
1460 if (lstat(ce
->name
, &st
) < 0) {
1461 if (ignore_missing
&& errno
== ENOENT
)
1468 changed
= ie_match_stat(istate
, ce
, &st
, options
);
1470 *changed_ret
= changed
;
1473 * The path is unchanged. If we were told to ignore
1474 * valid bit, then we did the actual stat check and
1475 * found that the entry is unmodified. If the entry
1476 * is not marked VALID, this is the place to mark it
1477 * valid again, under "assume unchanged" mode.
1479 if (ignore_valid
&& assume_unchanged
&&
1480 !(ce
->ce_flags
& CE_VALID
))
1481 ; /* mark this one VALID again */
1484 * We do not mark the index itself "modified"
1485 * because CE_UPTODATE flag is in-core only;
1486 * we are not going to write this change out.
1488 if (!S_ISGITLINK(ce
->ce_mode
)) {
1489 ce_mark_uptodate(ce
);
1490 mark_fsmonitor_valid(istate
, ce
);
1498 if (ie_modified(istate
, ce
, &st
, options
)) {
1504 updated
= make_empty_cache_entry(istate
, ce_namelen(ce
));
1505 copy_cache_entry(updated
, ce
);
1506 memcpy(updated
->name
, ce
->name
, ce
->ce_namelen
+ 1);
1507 fill_stat_cache_info(istate
, updated
, &st
);
1509 * If ignore_valid is not set, we should leave CE_VALID bit
1510 * alone. Otherwise, paths marked with --no-assume-unchanged
1511 * (i.e. things to be edited) will reacquire CE_VALID bit
1512 * automatically, which is not really what we want.
1514 if (!ignore_valid
&& assume_unchanged
&&
1515 !(ce
->ce_flags
& CE_VALID
))
1516 updated
->ce_flags
&= ~CE_VALID
;
1518 /* istate->cache_changed is updated in the caller */
1522 static void show_file(const char * fmt
, const char * name
, int in_porcelain
,
1523 int * first
, const char *header_msg
)
1525 if (in_porcelain
&& *first
&& header_msg
) {
1526 printf("%s\n", header_msg
);
1532 int repo_refresh_and_write_index(struct repository
*repo
,
1533 unsigned int refresh_flags
,
1534 unsigned int write_flags
,
1536 const struct pathspec
*pathspec
,
1537 char *seen
, const char *header_msg
)
1539 struct lock_file lock_file
= LOCK_INIT
;
1542 fd
= repo_hold_locked_index(repo
, &lock_file
, 0);
1543 if (!gentle
&& fd
< 0)
1545 if (refresh_index(repo
->index
, refresh_flags
, pathspec
, seen
, header_msg
))
1547 if (0 <= fd
&& write_locked_index(repo
->index
, &lock_file
, COMMIT_LOCK
| write_flags
))
1553 int refresh_index(struct index_state
*istate
, unsigned int flags
,
1554 const struct pathspec
*pathspec
,
1555 char *seen
, const char *header_msg
)
1559 int really
= (flags
& REFRESH_REALLY
) != 0;
1560 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
1561 int quiet
= (flags
& REFRESH_QUIET
) != 0;
1562 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
1563 int ignore_submodules
= (flags
& REFRESH_IGNORE_SUBMODULES
) != 0;
1565 int in_porcelain
= (flags
& REFRESH_IN_PORCELAIN
);
1566 unsigned int options
= (CE_MATCH_REFRESH
|
1567 (really
? CE_MATCH_IGNORE_VALID
: 0) |
1568 (not_new
? CE_MATCH_IGNORE_MISSING
: 0));
1569 const char *modified_fmt
;
1570 const char *deleted_fmt
;
1571 const char *typechange_fmt
;
1572 const char *added_fmt
;
1573 const char *unmerged_fmt
;
1574 struct progress
*progress
= NULL
;
1575 int t2_sum_lstat
= 0;
1576 int t2_sum_scan
= 0;
1578 if (flags
& REFRESH_PROGRESS
&& isatty(2))
1579 progress
= start_delayed_progress(_("Refresh index"),
1582 trace_performance_enter();
1583 modified_fmt
= in_porcelain
? "M\t%s\n" : "%s: needs update\n";
1584 deleted_fmt
= in_porcelain
? "D\t%s\n" : "%s: needs update\n";
1585 typechange_fmt
= in_porcelain
? "T\t%s\n" : "%s: needs update\n";
1586 added_fmt
= in_porcelain
? "A\t%s\n" : "%s: needs update\n";
1587 unmerged_fmt
= in_porcelain
? "U\t%s\n" : "%s: needs merge\n";
1589 * Use the multi-threaded preload_index() to refresh most of the
1590 * cache entries quickly then in the single threaded loop below,
1591 * we only have to do the special cases that are left.
1593 preload_index(istate
, pathspec
, 0);
1594 trace2_region_enter("index", "refresh", NULL
);
1595 /* TODO: audit for interaction with sparse-index. */
1596 ensure_full_index(istate
);
1597 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1598 struct cache_entry
*ce
, *new_entry
;
1599 int cache_errno
= 0;
1602 int t2_did_lstat
= 0;
1603 int t2_did_scan
= 0;
1605 ce
= istate
->cache
[i
];
1606 if (ignore_submodules
&& S_ISGITLINK(ce
->ce_mode
))
1609 if (pathspec
&& !ce_path_match(istate
, ce
, pathspec
, seen
))
1613 while ((i
< istate
->cache_nr
) &&
1614 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
1620 show_file(unmerged_fmt
, ce
->name
, in_porcelain
,
1621 &first
, header_msg
);
1629 new_entry
= refresh_cache_ent(istate
, ce
, options
,
1630 &cache_errno
, &changed
,
1631 &t2_did_lstat
, &t2_did_scan
);
1632 t2_sum_lstat
+= t2_did_lstat
;
1633 t2_sum_scan
+= t2_did_scan
;
1634 if (new_entry
== ce
)
1637 display_progress(progress
, i
);
1641 if (really
&& cache_errno
== EINVAL
) {
1642 /* If we are doing --really-refresh that
1643 * means the index is not valid anymore.
1645 ce
->ce_flags
&= ~CE_VALID
;
1646 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
1647 mark_fsmonitor_invalid(istate
, ce
);
1648 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
1653 if (cache_errno
== ENOENT
)
1655 else if (ce_intent_to_add(ce
))
1656 fmt
= added_fmt
; /* must be before other checks */
1657 else if (changed
& TYPE_CHANGED
)
1658 fmt
= typechange_fmt
;
1662 ce
->name
, in_porcelain
, &first
, header_msg
);
1667 replace_index_entry(istate
, i
, new_entry
);
1669 trace2_data_intmax("index", NULL
, "refresh/sum_lstat", t2_sum_lstat
);
1670 trace2_data_intmax("index", NULL
, "refresh/sum_scan", t2_sum_scan
);
1671 trace2_region_leave("index", "refresh", NULL
);
1673 display_progress(progress
, istate
->cache_nr
);
1674 stop_progress(&progress
);
1676 trace_performance_leave("refresh index");
1680 struct cache_entry
*refresh_cache_entry(struct index_state
*istate
,
1681 struct cache_entry
*ce
,
1682 unsigned int options
)
1684 return refresh_cache_ent(istate
, ce
, options
, NULL
, NULL
, NULL
, NULL
);
1688 /*****************************************************************
1690 *****************************************************************/
1692 #define INDEX_FORMAT_DEFAULT 3
1694 static unsigned int get_index_format_default(struct repository
*r
)
1696 char *envversion
= getenv("GIT_INDEX_VERSION");
1698 unsigned int version
= INDEX_FORMAT_DEFAULT
;
1701 prepare_repo_settings(r
);
1703 if (r
->settings
.index_version
>= 0)
1704 version
= r
->settings
.index_version
;
1705 if (version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1706 warning(_("index.version set, but the value is invalid.\n"
1707 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1708 return INDEX_FORMAT_DEFAULT
;
1713 version
= strtoul(envversion
, &endp
, 10);
1715 version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1716 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1717 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1718 version
= INDEX_FORMAT_DEFAULT
;
1724 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1725 * Again - this is just a (very strong in practice) heuristic that
1726 * the inode hasn't changed.
1728 * We save the fields in big-endian order to allow using the
1729 * index file over NFS transparently.
1731 struct ondisk_cache_entry
{
1732 struct cache_time ctime
;
1733 struct cache_time mtime
;
1741 * unsigned char hash[hashsz];
1743 * if (flags & CE_EXTENDED)
1746 unsigned char data
[GIT_MAX_RAWSZ
+ 2 * sizeof(uint16_t)];
1747 char name
[FLEX_ARRAY
];
1750 /* These are only used for v3 or lower */
1751 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1752 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,data) + (len) + 8) & ~7)
1753 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1754 #define ondisk_data_size(flags, len) (the_hash_algo->rawsz + \
1755 ((flags & CE_EXTENDED) ? 2 : 1) * sizeof(uint16_t) + len)
1756 #define ondisk_data_size_max(len) (ondisk_data_size(CE_EXTENDED, len))
1757 #define ondisk_ce_size(ce) (ondisk_cache_entry_size(ondisk_data_size((ce)->ce_flags, ce_namelen(ce))))
1759 /* Allow fsck to force verification of the index checksum. */
1760 int verify_index_checksum
;
1762 /* Allow fsck to force verification of the cache entry order. */
1763 int verify_ce_order
;
1765 static int verify_hdr(const struct cache_header
*hdr
, unsigned long size
)
1768 unsigned char hash
[GIT_MAX_RAWSZ
];
1771 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
1772 return error(_("bad signature 0x%08x"), hdr
->hdr_signature
);
1773 hdr_version
= ntohl(hdr
->hdr_version
);
1774 if (hdr_version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< hdr_version
)
1775 return error(_("bad index version %d"), hdr_version
);
1777 if (!verify_index_checksum
)
1780 the_hash_algo
->init_fn(&c
);
1781 the_hash_algo
->update_fn(&c
, hdr
, size
- the_hash_algo
->rawsz
);
1782 the_hash_algo
->final_fn(hash
, &c
);
1783 if (!hasheq(hash
, (unsigned char *)hdr
+ size
- the_hash_algo
->rawsz
))
1784 return error(_("bad index file sha1 signature"));
1788 static int read_index_extension(struct index_state
*istate
,
1789 const char *ext
, const char *data
, unsigned long sz
)
1791 switch (CACHE_EXT(ext
)) {
1792 case CACHE_EXT_TREE
:
1793 istate
->cache_tree
= cache_tree_read(data
, sz
);
1795 case CACHE_EXT_RESOLVE_UNDO
:
1796 istate
->resolve_undo
= resolve_undo_read(data
, sz
);
1798 case CACHE_EXT_LINK
:
1799 if (read_link_extension(istate
, data
, sz
))
1802 case CACHE_EXT_UNTRACKED
:
1803 istate
->untracked
= read_untracked_extension(data
, sz
);
1805 case CACHE_EXT_FSMONITOR
:
1806 read_fsmonitor_extension(istate
, data
, sz
);
1808 case CACHE_EXT_ENDOFINDEXENTRIES
:
1809 case CACHE_EXT_INDEXENTRYOFFSETTABLE
:
1810 /* already handled in do_read_index() */
1812 case CACHE_EXT_SPARSE_DIRECTORIES
:
1813 /* no content, only an indicator */
1814 istate
->sparse_index
= 1;
1817 if (*ext
< 'A' || 'Z' < *ext
)
1818 return error(_("index uses %.4s extension, which we do not understand"),
1820 fprintf_ln(stderr
, _("ignoring %.4s extension"), ext
);
1826 static struct cache_entry
*create_from_disk(struct mem_pool
*ce_mem_pool
,
1827 unsigned int version
,
1828 struct ondisk_cache_entry
*ondisk
,
1829 unsigned long *ent_size
,
1830 const struct cache_entry
*previous_ce
)
1832 struct cache_entry
*ce
;
1835 const unsigned hashsz
= the_hash_algo
->rawsz
;
1836 const uint16_t *flagsp
= (const uint16_t *)(ondisk
->data
+ hashsz
);
1838 size_t copy_len
= 0;
1840 * Adjacent cache entries tend to share the leading paths, so it makes
1841 * sense to only store the differences in later entries. In the v4
1842 * on-disk format of the index, each on-disk cache entry stores the
1843 * number of bytes to be stripped from the end of the previous name,
1844 * and the bytes to append to the result, to come up with its name.
1846 int expand_name_field
= version
== 4;
1848 /* On-disk flags are just 16 bits */
1849 flags
= get_be16(flagsp
);
1850 len
= flags
& CE_NAMEMASK
;
1852 if (flags
& CE_EXTENDED
) {
1854 extended_flags
= get_be16(flagsp
+ 1) << 16;
1855 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1856 if (extended_flags
& ~CE_EXTENDED_FLAGS
)
1857 die(_("unknown index entry format 0x%08x"), extended_flags
);
1858 flags
|= extended_flags
;
1859 name
= (const char *)(flagsp
+ 2);
1862 name
= (const char *)(flagsp
+ 1);
1864 if (expand_name_field
) {
1865 const unsigned char *cp
= (const unsigned char *)name
;
1866 size_t strip_len
, previous_len
;
1868 /* If we're at the beginning of a block, ignore the previous name */
1869 strip_len
= decode_varint(&cp
);
1871 previous_len
= previous_ce
->ce_namelen
;
1872 if (previous_len
< strip_len
)
1873 die(_("malformed name field in the index, near path '%s'"),
1875 copy_len
= previous_len
- strip_len
;
1877 name
= (const char *)cp
;
1880 if (len
== CE_NAMEMASK
) {
1882 if (expand_name_field
)
1886 ce
= mem_pool__ce_alloc(ce_mem_pool
, len
);
1888 ce
->ce_stat_data
.sd_ctime
.sec
= get_be32(&ondisk
->ctime
.sec
);
1889 ce
->ce_stat_data
.sd_mtime
.sec
= get_be32(&ondisk
->mtime
.sec
);
1890 ce
->ce_stat_data
.sd_ctime
.nsec
= get_be32(&ondisk
->ctime
.nsec
);
1891 ce
->ce_stat_data
.sd_mtime
.nsec
= get_be32(&ondisk
->mtime
.nsec
);
1892 ce
->ce_stat_data
.sd_dev
= get_be32(&ondisk
->dev
);
1893 ce
->ce_stat_data
.sd_ino
= get_be32(&ondisk
->ino
);
1894 ce
->ce_mode
= get_be32(&ondisk
->mode
);
1895 ce
->ce_stat_data
.sd_uid
= get_be32(&ondisk
->uid
);
1896 ce
->ce_stat_data
.sd_gid
= get_be32(&ondisk
->gid
);
1897 ce
->ce_stat_data
.sd_size
= get_be32(&ondisk
->size
);
1898 ce
->ce_flags
= flags
& ~CE_NAMEMASK
;
1899 ce
->ce_namelen
= len
;
1901 hashcpy(ce
->oid
.hash
, ondisk
->data
);
1902 memcpy(ce
->name
, name
, len
);
1903 ce
->name
[len
] = '\0';
1905 if (expand_name_field
) {
1907 memcpy(ce
->name
, previous_ce
->name
, copy_len
);
1908 memcpy(ce
->name
+ copy_len
, name
, len
+ 1 - copy_len
);
1909 *ent_size
= (name
- ((char *)ondisk
)) + len
+ 1 - copy_len
;
1911 memcpy(ce
->name
, name
, len
+ 1);
1912 *ent_size
= ondisk_ce_size(ce
);
1917 static void check_ce_order(struct index_state
*istate
)
1921 if (!verify_ce_order
)
1924 for (i
= 1; i
< istate
->cache_nr
; i
++) {
1925 struct cache_entry
*ce
= istate
->cache
[i
- 1];
1926 struct cache_entry
*next_ce
= istate
->cache
[i
];
1927 int name_compare
= strcmp(ce
->name
, next_ce
->name
);
1929 if (0 < name_compare
)
1930 die(_("unordered stage entries in index"));
1931 if (!name_compare
) {
1933 die(_("multiple stage entries for merged file '%s'"),
1935 if (ce_stage(ce
) > ce_stage(next_ce
))
1936 die(_("unordered stage entries for '%s'"),
1942 static void tweak_untracked_cache(struct index_state
*istate
)
1944 struct repository
*r
= the_repository
;
1946 prepare_repo_settings(r
);
1948 if (r
->settings
.core_untracked_cache
== UNTRACKED_CACHE_REMOVE
) {
1949 remove_untracked_cache(istate
);
1953 if (r
->settings
.core_untracked_cache
== UNTRACKED_CACHE_WRITE
)
1954 add_untracked_cache(istate
);
1957 static void tweak_split_index(struct index_state
*istate
)
1959 switch (git_config_get_split_index()) {
1960 case -1: /* unset: do nothing */
1963 remove_split_index(istate
);
1966 add_split_index(istate
);
1968 default: /* unknown value: do nothing */
1973 static void post_read_index_from(struct index_state
*istate
)
1975 check_ce_order(istate
);
1976 tweak_untracked_cache(istate
);
1977 tweak_split_index(istate
);
1978 tweak_fsmonitor(istate
);
1981 static size_t estimate_cache_size_from_compressed(unsigned int entries
)
1983 return entries
* (sizeof(struct cache_entry
) + CACHE_ENTRY_PATH_LENGTH
);
1986 static size_t estimate_cache_size(size_t ondisk_size
, unsigned int entries
)
1988 long per_entry
= sizeof(struct cache_entry
) - sizeof(struct ondisk_cache_entry
);
1991 * Account for potential alignment differences.
1993 per_entry
+= align_padding_size(per_entry
, 0);
1994 return ondisk_size
+ entries
* per_entry
;
1997 struct index_entry_offset
1999 /* starting byte offset into index file, count of index entries in this block */
2003 struct index_entry_offset_table
2006 struct index_entry_offset entries
[FLEX_ARRAY
];
2009 static struct index_entry_offset_table
*read_ieot_extension(const char *mmap
, size_t mmap_size
, size_t offset
);
2010 static void write_ieot_extension(struct strbuf
*sb
, struct index_entry_offset_table
*ieot
);
2012 static size_t read_eoie_extension(const char *mmap
, size_t mmap_size
);
2013 static void write_eoie_extension(struct strbuf
*sb
, git_hash_ctx
*eoie_context
, size_t offset
);
2015 struct load_index_extensions
2018 struct index_state
*istate
;
2021 unsigned long src_offset
;
2024 static void *load_index_extensions(void *_data
)
2026 struct load_index_extensions
*p
= _data
;
2027 unsigned long src_offset
= p
->src_offset
;
2029 while (src_offset
<= p
->mmap_size
- the_hash_algo
->rawsz
- 8) {
2030 /* After an array of active_nr index entries,
2031 * there can be arbitrary number of extended
2032 * sections, each of which is prefixed with
2033 * extension name (4-byte) and section length
2034 * in 4-byte network byte order.
2036 uint32_t extsize
= get_be32(p
->mmap
+ src_offset
+ 4);
2037 if (read_index_extension(p
->istate
,
2038 p
->mmap
+ src_offset
,
2039 p
->mmap
+ src_offset
+ 8,
2041 munmap((void *)p
->mmap
, p
->mmap_size
);
2042 die(_("index file corrupt"));
2045 src_offset
+= extsize
;
2052 * A helper function that will load the specified range of cache entries
2053 * from the memory mapped file and add them to the given index.
2055 static unsigned long load_cache_entry_block(struct index_state
*istate
,
2056 struct mem_pool
*ce_mem_pool
, int offset
, int nr
, const char *mmap
,
2057 unsigned long start_offset
, const struct cache_entry
*previous_ce
)
2060 unsigned long src_offset
= start_offset
;
2062 for (i
= offset
; i
< offset
+ nr
; i
++) {
2063 struct ondisk_cache_entry
*disk_ce
;
2064 struct cache_entry
*ce
;
2065 unsigned long consumed
;
2067 disk_ce
= (struct ondisk_cache_entry
*)(mmap
+ src_offset
);
2068 ce
= create_from_disk(ce_mem_pool
, istate
->version
, disk_ce
, &consumed
, previous_ce
);
2069 set_index_entry(istate
, i
, ce
);
2071 src_offset
+= consumed
;
2074 return src_offset
- start_offset
;
2077 static unsigned long load_all_cache_entries(struct index_state
*istate
,
2078 const char *mmap
, size_t mmap_size
, unsigned long src_offset
)
2080 unsigned long consumed
;
2082 istate
->ce_mem_pool
= xmalloc(sizeof(*istate
->ce_mem_pool
));
2083 if (istate
->version
== 4) {
2084 mem_pool_init(istate
->ce_mem_pool
,
2085 estimate_cache_size_from_compressed(istate
->cache_nr
));
2087 mem_pool_init(istate
->ce_mem_pool
,
2088 estimate_cache_size(mmap_size
, istate
->cache_nr
));
2091 consumed
= load_cache_entry_block(istate
, istate
->ce_mem_pool
,
2092 0, istate
->cache_nr
, mmap
, src_offset
, NULL
);
2097 * Mostly randomly chosen maximum thread counts: we
2098 * cap the parallelism to online_cpus() threads, and we want
2099 * to have at least 10000 cache entries per thread for it to
2100 * be worth starting a thread.
2103 #define THREAD_COST (10000)
2105 struct load_cache_entries_thread_data
2108 struct index_state
*istate
;
2109 struct mem_pool
*ce_mem_pool
;
2112 struct index_entry_offset_table
*ieot
;
2113 int ieot_start
; /* starting index into the ieot array */
2114 int ieot_blocks
; /* count of ieot entries to process */
2115 unsigned long consumed
; /* return # of bytes in index file processed */
2119 * A thread proc to run the load_cache_entries() computation
2120 * across multiple background threads.
2122 static void *load_cache_entries_thread(void *_data
)
2124 struct load_cache_entries_thread_data
*p
= _data
;
2127 /* iterate across all ieot blocks assigned to this thread */
2128 for (i
= p
->ieot_start
; i
< p
->ieot_start
+ p
->ieot_blocks
; i
++) {
2129 p
->consumed
+= load_cache_entry_block(p
->istate
, p
->ce_mem_pool
,
2130 p
->offset
, p
->ieot
->entries
[i
].nr
, p
->mmap
, p
->ieot
->entries
[i
].offset
, NULL
);
2131 p
->offset
+= p
->ieot
->entries
[i
].nr
;
2136 static unsigned long load_cache_entries_threaded(struct index_state
*istate
, const char *mmap
, size_t mmap_size
,
2137 int nr_threads
, struct index_entry_offset_table
*ieot
)
2139 int i
, offset
, ieot_blocks
, ieot_start
, err
;
2140 struct load_cache_entries_thread_data
*data
;
2141 unsigned long consumed
= 0;
2143 /* a little sanity checking */
2144 if (istate
->name_hash_initialized
)
2145 BUG("the name hash isn't thread safe");
2147 istate
->ce_mem_pool
= xmalloc(sizeof(*istate
->ce_mem_pool
));
2148 mem_pool_init(istate
->ce_mem_pool
, 0);
2150 /* ensure we have no more threads than we have blocks to process */
2151 if (nr_threads
> ieot
->nr
)
2152 nr_threads
= ieot
->nr
;
2153 CALLOC_ARRAY(data
, nr_threads
);
2155 offset
= ieot_start
= 0;
2156 ieot_blocks
= DIV_ROUND_UP(ieot
->nr
, nr_threads
);
2157 for (i
= 0; i
< nr_threads
; i
++) {
2158 struct load_cache_entries_thread_data
*p
= &data
[i
];
2161 if (ieot_start
+ ieot_blocks
> ieot
->nr
)
2162 ieot_blocks
= ieot
->nr
- ieot_start
;
2168 p
->ieot_start
= ieot_start
;
2169 p
->ieot_blocks
= ieot_blocks
;
2171 /* create a mem_pool for each thread */
2173 for (j
= p
->ieot_start
; j
< p
->ieot_start
+ p
->ieot_blocks
; j
++)
2174 nr
+= p
->ieot
->entries
[j
].nr
;
2175 p
->ce_mem_pool
= xmalloc(sizeof(*istate
->ce_mem_pool
));
2176 if (istate
->version
== 4) {
2177 mem_pool_init(p
->ce_mem_pool
,
2178 estimate_cache_size_from_compressed(nr
));
2180 mem_pool_init(p
->ce_mem_pool
,
2181 estimate_cache_size(mmap_size
, nr
));
2184 err
= pthread_create(&p
->pthread
, NULL
, load_cache_entries_thread
, p
);
2186 die(_("unable to create load_cache_entries thread: %s"), strerror(err
));
2188 /* increment by the number of cache entries in the ieot block being processed */
2189 for (j
= 0; j
< ieot_blocks
; j
++)
2190 offset
+= ieot
->entries
[ieot_start
+ j
].nr
;
2191 ieot_start
+= ieot_blocks
;
2194 for (i
= 0; i
< nr_threads
; i
++) {
2195 struct load_cache_entries_thread_data
*p
= &data
[i
];
2197 err
= pthread_join(p
->pthread
, NULL
);
2199 die(_("unable to join load_cache_entries thread: %s"), strerror(err
));
2200 mem_pool_combine(istate
->ce_mem_pool
, p
->ce_mem_pool
);
2201 consumed
+= p
->consumed
;
2209 /* remember to discard_cache() before reading a different cache! */
2210 int do_read_index(struct index_state
*istate
, const char *path
, int must_exist
)
2214 unsigned long src_offset
;
2215 const struct cache_header
*hdr
;
2218 struct load_index_extensions p
;
2219 size_t extension_offset
= 0;
2220 int nr_threads
, cpus
;
2221 struct index_entry_offset_table
*ieot
= NULL
;
2223 if (istate
->initialized
)
2224 return istate
->cache_nr
;
2226 istate
->timestamp
.sec
= 0;
2227 istate
->timestamp
.nsec
= 0;
2228 fd
= open(path
, O_RDONLY
);
2230 if (!must_exist
&& errno
== ENOENT
)
2232 die_errno(_("%s: index file open failed"), path
);
2236 die_errno(_("%s: cannot stat the open index"), path
);
2238 mmap_size
= xsize_t(st
.st_size
);
2239 if (mmap_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2240 die(_("%s: index file smaller than expected"), path
);
2242 mmap
= xmmap_gently(NULL
, mmap_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2243 if (mmap
== MAP_FAILED
)
2244 die_errno(_("%s: unable to map index file"), path
);
2247 hdr
= (const struct cache_header
*)mmap
;
2248 if (verify_hdr(hdr
, mmap_size
) < 0)
2251 hashcpy(istate
->oid
.hash
, (const unsigned char *)hdr
+ mmap_size
- the_hash_algo
->rawsz
);
2252 istate
->version
= ntohl(hdr
->hdr_version
);
2253 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
2254 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
2255 CALLOC_ARRAY(istate
->cache
, istate
->cache_alloc
);
2256 istate
->initialized
= 1;
2260 p
.mmap_size
= mmap_size
;
2262 src_offset
= sizeof(*hdr
);
2264 if (git_config_get_index_threads(&nr_threads
))
2267 /* TODO: does creating more threads than cores help? */
2269 nr_threads
= istate
->cache_nr
/ THREAD_COST
;
2270 cpus
= online_cpus();
2271 if (nr_threads
> cpus
)
2278 if (nr_threads
> 1) {
2279 extension_offset
= read_eoie_extension(mmap
, mmap_size
);
2280 if (extension_offset
) {
2283 p
.src_offset
= extension_offset
;
2284 err
= pthread_create(&p
.pthread
, NULL
, load_index_extensions
, &p
);
2286 die(_("unable to create load_index_extensions thread: %s"), strerror(err
));
2293 * Locate and read the index entry offset table so that we can use it
2294 * to multi-thread the reading of the cache entries.
2296 if (extension_offset
&& nr_threads
> 1)
2297 ieot
= read_ieot_extension(mmap
, mmap_size
, extension_offset
);
2300 src_offset
+= load_cache_entries_threaded(istate
, mmap
, mmap_size
, nr_threads
, ieot
);
2303 src_offset
+= load_all_cache_entries(istate
, mmap
, mmap_size
, src_offset
);
2306 istate
->timestamp
.sec
= st
.st_mtime
;
2307 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
2309 /* if we created a thread, join it otherwise load the extensions on the primary thread */
2310 if (extension_offset
) {
2311 int ret
= pthread_join(p
.pthread
, NULL
);
2313 die(_("unable to join load_index_extensions thread: %s"), strerror(ret
));
2315 p
.src_offset
= src_offset
;
2316 load_index_extensions(&p
);
2318 munmap((void *)mmap
, mmap_size
);
2321 * TODO trace2: replace "the_repository" with the actual repo instance
2322 * that is associated with the given "istate".
2324 trace2_data_intmax("index", the_repository
, "read/version",
2326 trace2_data_intmax("index", the_repository
, "read/cache_nr",
2330 istate
->repo
= the_repository
;
2331 prepare_repo_settings(istate
->repo
);
2332 if (istate
->repo
->settings
.command_requires_full_index
)
2333 ensure_full_index(istate
);
2335 return istate
->cache_nr
;
2338 munmap((void *)mmap
, mmap_size
);
2339 die(_("index file corrupt"));
2343 * Signal that the shared index is used by updating its mtime.
2345 * This way, shared index can be removed if they have not been used
2348 static void freshen_shared_index(const char *shared_index
, int warn
)
2350 if (!check_and_freshen_file(shared_index
, 1) && warn
)
2351 warning(_("could not freshen shared index '%s'"), shared_index
);
2354 int read_index_from(struct index_state
*istate
, const char *path
,
2357 struct split_index
*split_index
;
2362 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2363 if (istate
->initialized
)
2364 return istate
->cache_nr
;
2367 * TODO trace2: replace "the_repository" with the actual repo instance
2368 * that is associated with the given "istate".
2370 trace2_region_enter_printf("index", "do_read_index", the_repository
,
2372 trace_performance_enter();
2373 ret
= do_read_index(istate
, path
, 0);
2374 trace_performance_leave("read cache %s", path
);
2375 trace2_region_leave_printf("index", "do_read_index", the_repository
,
2378 split_index
= istate
->split_index
;
2379 if (!split_index
|| is_null_oid(&split_index
->base_oid
)) {
2380 post_read_index_from(istate
);
2384 trace_performance_enter();
2385 if (split_index
->base
)
2386 discard_index(split_index
->base
);
2388 CALLOC_ARRAY(split_index
->base
, 1);
2390 base_oid_hex
= oid_to_hex(&split_index
->base_oid
);
2391 base_path
= xstrfmt("%s/sharedindex.%s", gitdir
, base_oid_hex
);
2392 trace2_region_enter_printf("index", "shared/do_read_index",
2393 the_repository
, "%s", base_path
);
2394 ret
= do_read_index(split_index
->base
, base_path
, 1);
2395 trace2_region_leave_printf("index", "shared/do_read_index",
2396 the_repository
, "%s", base_path
);
2397 if (!oideq(&split_index
->base_oid
, &split_index
->base
->oid
))
2398 die(_("broken index, expect %s in %s, got %s"),
2399 base_oid_hex
, base_path
,
2400 oid_to_hex(&split_index
->base
->oid
));
2402 freshen_shared_index(base_path
, 0);
2403 merge_base_index(istate
);
2404 post_read_index_from(istate
);
2405 trace_performance_leave("read cache %s", base_path
);
2410 int is_index_unborn(struct index_state
*istate
)
2412 return (!istate
->cache_nr
&& !istate
->timestamp
.sec
);
2415 int discard_index(struct index_state
*istate
)
2418 * Cache entries in istate->cache[] should have been allocated
2419 * from the memory pool associated with this index, or from an
2420 * associated split_index. There is no need to free individual
2421 * cache entries. validate_cache_entries can detect when this
2422 * assertion does not hold.
2424 validate_cache_entries(istate
);
2426 resolve_undo_clear_index(istate
);
2427 istate
->cache_nr
= 0;
2428 istate
->cache_changed
= 0;
2429 istate
->timestamp
.sec
= 0;
2430 istate
->timestamp
.nsec
= 0;
2431 free_name_hash(istate
);
2432 cache_tree_free(&(istate
->cache_tree
));
2433 istate
->initialized
= 0;
2434 istate
->fsmonitor_has_run_once
= 0;
2435 FREE_AND_NULL(istate
->fsmonitor_last_update
);
2436 FREE_AND_NULL(istate
->cache
);
2437 istate
->cache_alloc
= 0;
2438 discard_split_index(istate
);
2439 free_untracked_cache(istate
->untracked
);
2440 istate
->untracked
= NULL
;
2442 if (istate
->ce_mem_pool
) {
2443 mem_pool_discard(istate
->ce_mem_pool
, should_validate_cache_entries());
2444 FREE_AND_NULL(istate
->ce_mem_pool
);
2451 * Validate the cache entries of this index.
2452 * All cache entries associated with this index
2453 * should have been allocated by the memory pool
2454 * associated with this index, or by a referenced
2457 void validate_cache_entries(const struct index_state
*istate
)
2461 if (!should_validate_cache_entries() ||!istate
|| !istate
->initialized
)
2464 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2466 BUG("cache entry is not allocated from expected memory pool");
2467 } else if (!istate
->ce_mem_pool
||
2468 !mem_pool_contains(istate
->ce_mem_pool
, istate
->cache
[i
])) {
2469 if (!istate
->split_index
||
2470 !istate
->split_index
->base
||
2471 !istate
->split_index
->base
->ce_mem_pool
||
2472 !mem_pool_contains(istate
->split_index
->base
->ce_mem_pool
, istate
->cache
[i
])) {
2473 BUG("cache entry is not allocated from expected memory pool");
2478 if (istate
->split_index
)
2479 validate_cache_entries(istate
->split_index
->base
);
2482 int unmerged_index(const struct index_state
*istate
)
2485 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2486 if (ce_stage(istate
->cache
[i
]))
2492 int repo_index_has_changes(struct repository
*repo
,
2496 struct index_state
*istate
= repo
->index
;
2497 struct object_id cmp
;
2501 cmp
= tree
->object
.oid
;
2502 if (tree
|| !get_oid_tree("HEAD", &cmp
)) {
2503 struct diff_options opt
;
2505 repo_diff_setup(repo
, &opt
);
2506 opt
.flags
.exit_with_status
= 1;
2508 opt
.flags
.quick
= 1;
2509 do_diff_cache(&cmp
, &opt
);
2511 for (i
= 0; sb
&& i
< diff_queued_diff
.nr
; i
++) {
2513 strbuf_addch(sb
, ' ');
2514 strbuf_addstr(sb
, diff_queued_diff
.queue
[i
]->two
->path
);
2517 return opt
.flags
.has_changes
!= 0;
2519 /* TODO: audit for interaction with sparse-index. */
2520 ensure_full_index(istate
);
2521 for (i
= 0; sb
&& i
< istate
->cache_nr
; i
++) {
2523 strbuf_addch(sb
, ' ');
2524 strbuf_addstr(sb
, istate
->cache
[i
]->name
);
2526 return !!istate
->cache_nr
;
2530 #define WRITE_BUFFER_SIZE (128 * 1024)
2531 static unsigned char write_buffer
[WRITE_BUFFER_SIZE
];
2532 static unsigned long write_buffer_len
;
2534 static int ce_write_flush(git_hash_ctx
*context
, int fd
)
2536 unsigned int buffered
= write_buffer_len
;
2538 the_hash_algo
->update_fn(context
, write_buffer
, buffered
);
2539 if (write_in_full(fd
, write_buffer
, buffered
) < 0)
2541 write_buffer_len
= 0;
2546 static int ce_write(git_hash_ctx
*context
, int fd
, void *data
, unsigned int len
)
2549 unsigned int buffered
= write_buffer_len
;
2550 unsigned int partial
= WRITE_BUFFER_SIZE
- buffered
;
2553 memcpy(write_buffer
+ buffered
, data
, partial
);
2554 buffered
+= partial
;
2555 if (buffered
== WRITE_BUFFER_SIZE
) {
2556 write_buffer_len
= buffered
;
2557 if (ce_write_flush(context
, fd
))
2561 write_buffer_len
= buffered
;
2563 data
= (char *) data
+ partial
;
2568 static int write_index_ext_header(git_hash_ctx
*context
, git_hash_ctx
*eoie_context
,
2569 int fd
, unsigned int ext
, unsigned int sz
)
2574 the_hash_algo
->update_fn(eoie_context
, &ext
, 4);
2575 the_hash_algo
->update_fn(eoie_context
, &sz
, 4);
2577 return ((ce_write(context
, fd
, &ext
, 4) < 0) ||
2578 (ce_write(context
, fd
, &sz
, 4) < 0)) ? -1 : 0;
2581 static int ce_flush(git_hash_ctx
*context
, int fd
, unsigned char *hash
)
2583 unsigned int left
= write_buffer_len
;
2586 write_buffer_len
= 0;
2587 the_hash_algo
->update_fn(context
, write_buffer
, left
);
2590 /* Flush first if not enough space for hash signature */
2591 if (left
+ the_hash_algo
->rawsz
> WRITE_BUFFER_SIZE
) {
2592 if (write_in_full(fd
, write_buffer
, left
) < 0)
2597 /* Append the hash signature at the end */
2598 the_hash_algo
->final_fn(write_buffer
+ left
, context
);
2599 hashcpy(hash
, write_buffer
+ left
);
2600 left
+= the_hash_algo
->rawsz
;
2601 return (write_in_full(fd
, write_buffer
, left
) < 0) ? -1 : 0;
2604 static void ce_smudge_racily_clean_entry(struct index_state
*istate
,
2605 struct cache_entry
*ce
)
2608 * The only thing we care about in this function is to smudge the
2609 * falsely clean entry due to touch-update-touch race, so we leave
2610 * everything else as they are. We are called for entries whose
2611 * ce_stat_data.sd_mtime match the index file mtime.
2613 * Note that this actually does not do much for gitlinks, for
2614 * which ce_match_stat_basic() always goes to the actual
2615 * contents. The caller checks with is_racy_timestamp() which
2616 * always says "no" for gitlinks, so we are not called for them ;-)
2620 if (lstat(ce
->name
, &st
) < 0)
2622 if (ce_match_stat_basic(ce
, &st
))
2624 if (ce_modified_check_fs(istate
, ce
, &st
)) {
2625 /* This is "racily clean"; smudge it. Note that this
2626 * is a tricky code. At first glance, it may appear
2627 * that it can break with this sequence:
2629 * $ echo xyzzy >frotz
2630 * $ git-update-index --add frotz
2633 * $ echo filfre >nitfol
2634 * $ git-update-index --add nitfol
2636 * but it does not. When the second update-index runs,
2637 * it notices that the entry "frotz" has the same timestamp
2638 * as index, and if we were to smudge it by resetting its
2639 * size to zero here, then the object name recorded
2640 * in index is the 6-byte file but the cached stat information
2641 * becomes zero --- which would then match what we would
2642 * obtain from the filesystem next time we stat("frotz").
2644 * However, the second update-index, before calling
2645 * this function, notices that the cached size is 6
2646 * bytes and what is on the filesystem is an empty
2647 * file, and never calls us, so the cached size information
2648 * for "frotz" stays 6 which does not match the filesystem.
2650 ce
->ce_stat_data
.sd_size
= 0;
2654 /* Copy miscellaneous fields but not the name */
2655 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry
*ondisk
,
2656 struct cache_entry
*ce
)
2659 const unsigned hashsz
= the_hash_algo
->rawsz
;
2660 uint16_t *flagsp
= (uint16_t *)(ondisk
->data
+ hashsz
);
2662 ondisk
->ctime
.sec
= htonl(ce
->ce_stat_data
.sd_ctime
.sec
);
2663 ondisk
->mtime
.sec
= htonl(ce
->ce_stat_data
.sd_mtime
.sec
);
2664 ondisk
->ctime
.nsec
= htonl(ce
->ce_stat_data
.sd_ctime
.nsec
);
2665 ondisk
->mtime
.nsec
= htonl(ce
->ce_stat_data
.sd_mtime
.nsec
);
2666 ondisk
->dev
= htonl(ce
->ce_stat_data
.sd_dev
);
2667 ondisk
->ino
= htonl(ce
->ce_stat_data
.sd_ino
);
2668 ondisk
->mode
= htonl(ce
->ce_mode
);
2669 ondisk
->uid
= htonl(ce
->ce_stat_data
.sd_uid
);
2670 ondisk
->gid
= htonl(ce
->ce_stat_data
.sd_gid
);
2671 ondisk
->size
= htonl(ce
->ce_stat_data
.sd_size
);
2672 hashcpy(ondisk
->data
, ce
->oid
.hash
);
2674 flags
= ce
->ce_flags
& ~CE_NAMEMASK
;
2675 flags
|= (ce_namelen(ce
) >= CE_NAMEMASK
? CE_NAMEMASK
: ce_namelen(ce
));
2676 flagsp
[0] = htons(flags
);
2677 if (ce
->ce_flags
& CE_EXTENDED
) {
2678 flagsp
[1] = htons((ce
->ce_flags
& CE_EXTENDED_FLAGS
) >> 16);
2682 static int ce_write_entry(git_hash_ctx
*c
, int fd
, struct cache_entry
*ce
,
2683 struct strbuf
*previous_name
, struct ondisk_cache_entry
*ondisk
)
2687 unsigned int saved_namelen
;
2688 int stripped_name
= 0;
2689 static unsigned char padding
[8] = { 0x00 };
2691 if (ce
->ce_flags
& CE_STRIP_NAME
) {
2692 saved_namelen
= ce_namelen(ce
);
2697 size
= offsetof(struct ondisk_cache_entry
,data
) + ondisk_data_size(ce
->ce_flags
, 0);
2699 if (!previous_name
) {
2700 int len
= ce_namelen(ce
);
2701 copy_cache_entry_to_ondisk(ondisk
, ce
);
2702 result
= ce_write(c
, fd
, ondisk
, size
);
2704 result
= ce_write(c
, fd
, ce
->name
, len
);
2706 result
= ce_write(c
, fd
, padding
, align_padding_size(size
, len
));
2708 int common
, to_remove
, prefix_size
;
2709 unsigned char to_remove_vi
[16];
2711 (ce
->name
[common
] &&
2712 common
< previous_name
->len
&&
2713 ce
->name
[common
] == previous_name
->buf
[common
]);
2715 ; /* still matching */
2716 to_remove
= previous_name
->len
- common
;
2717 prefix_size
= encode_varint(to_remove
, to_remove_vi
);
2719 copy_cache_entry_to_ondisk(ondisk
, ce
);
2720 result
= ce_write(c
, fd
, ondisk
, size
);
2722 result
= ce_write(c
, fd
, to_remove_vi
, prefix_size
);
2724 result
= ce_write(c
, fd
, ce
->name
+ common
, ce_namelen(ce
) - common
);
2726 result
= ce_write(c
, fd
, padding
, 1);
2728 strbuf_splice(previous_name
, common
, to_remove
,
2729 ce
->name
+ common
, ce_namelen(ce
) - common
);
2731 if (stripped_name
) {
2732 ce
->ce_namelen
= saved_namelen
;
2733 ce
->ce_flags
&= ~CE_STRIP_NAME
;
2740 * This function verifies if index_state has the correct sha1 of the
2741 * index file. Don't die if we have any other failure, just return 0.
2743 static int verify_index_from(const struct index_state
*istate
, const char *path
)
2748 unsigned char hash
[GIT_MAX_RAWSZ
];
2750 if (!istate
->initialized
)
2753 fd
= open(path
, O_RDONLY
);
2760 if (st
.st_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2763 n
= pread_in_full(fd
, hash
, the_hash_algo
->rawsz
, st
.st_size
- the_hash_algo
->rawsz
);
2764 if (n
!= the_hash_algo
->rawsz
)
2767 if (!hasheq(istate
->oid
.hash
, hash
))
2778 static int repo_verify_index(struct repository
*repo
)
2780 return verify_index_from(repo
->index
, repo
->index_file
);
2783 static int has_racy_timestamp(struct index_state
*istate
)
2785 int entries
= istate
->cache_nr
;
2788 for (i
= 0; i
< entries
; i
++) {
2789 struct cache_entry
*ce
= istate
->cache
[i
];
2790 if (is_racy_timestamp(istate
, ce
))
2796 void repo_update_index_if_able(struct repository
*repo
,
2797 struct lock_file
*lockfile
)
2799 if ((repo
->index
->cache_changed
||
2800 has_racy_timestamp(repo
->index
)) &&
2801 repo_verify_index(repo
))
2802 write_locked_index(repo
->index
, lockfile
, COMMIT_LOCK
);
2804 rollback_lock_file(lockfile
);
2807 static int record_eoie(void)
2811 if (!git_config_get_bool("index.recordendofindexentries", &val
))
2815 * As a convenience, the end of index entries extension
2816 * used for threading is written by default if the user
2817 * explicitly requested threaded index reads.
2819 return !git_config_get_index_threads(&val
) && val
!= 1;
2822 static int record_ieot(void)
2826 if (!git_config_get_bool("index.recordoffsettable", &val
))
2830 * As a convenience, the offset table used for threading is
2831 * written by default if the user explicitly requested
2832 * threaded index reads.
2834 return !git_config_get_index_threads(&val
) && val
!= 1;
2838 * On success, `tempfile` is closed. If it is the temporary file
2839 * of a `struct lock_file`, we will therefore effectively perform
2840 * a 'close_lock_file_gently()`. Since that is an implementation
2841 * detail of lockfiles, callers of `do_write_index()` should not
2844 static int do_write_index(struct index_state
*istate
, struct tempfile
*tempfile
,
2845 int strip_extensions
)
2847 uint64_t start
= getnanotime();
2848 int newfd
= tempfile
->fd
;
2849 git_hash_ctx c
, eoie_c
;
2850 struct cache_header hdr
;
2851 int i
, err
= 0, removed
, extended
, hdr_version
;
2852 struct cache_entry
**cache
= istate
->cache
;
2853 int entries
= istate
->cache_nr
;
2855 struct ondisk_cache_entry ondisk
;
2856 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
2857 int drop_cache_tree
= istate
->drop_cache_tree
;
2859 int ieot_entries
= 1;
2860 struct index_entry_offset_table
*ieot
= NULL
;
2863 for (i
= removed
= extended
= 0; i
< entries
; i
++) {
2864 if (cache
[i
]->ce_flags
& CE_REMOVE
)
2867 /* reduce extended entries if possible */
2868 cache
[i
]->ce_flags
&= ~CE_EXTENDED
;
2869 if (cache
[i
]->ce_flags
& CE_EXTENDED_FLAGS
) {
2871 cache
[i
]->ce_flags
|= CE_EXTENDED
;
2875 if (!istate
->version
) {
2876 istate
->version
= get_index_format_default(the_repository
);
2877 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0))
2878 init_split_index(istate
);
2881 /* demote version 3 to version 2 when the latter suffices */
2882 if (istate
->version
== 3 || istate
->version
== 2)
2883 istate
->version
= extended
? 3 : 2;
2885 hdr_version
= istate
->version
;
2887 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
2888 hdr
.hdr_version
= htonl(hdr_version
);
2889 hdr
.hdr_entries
= htonl(entries
- removed
);
2891 the_hash_algo
->init_fn(&c
);
2892 if (ce_write(&c
, newfd
, &hdr
, sizeof(hdr
)) < 0)
2895 if (!HAVE_THREADS
|| git_config_get_index_threads(&nr_threads
))
2898 if (nr_threads
!= 1 && record_ieot()) {
2899 int ieot_blocks
, cpus
;
2902 * ensure default number of ieot blocks maps evenly to the
2903 * default number of threads that will process them leaving
2904 * room for the thread to load the index extensions.
2907 ieot_blocks
= istate
->cache_nr
/ THREAD_COST
;
2908 cpus
= online_cpus();
2909 if (ieot_blocks
> cpus
- 1)
2910 ieot_blocks
= cpus
- 1;
2912 ieot_blocks
= nr_threads
;
2913 if (ieot_blocks
> istate
->cache_nr
)
2914 ieot_blocks
= istate
->cache_nr
;
2918 * no reason to write out the IEOT extension if we don't
2919 * have enough blocks to utilize multi-threading
2921 if (ieot_blocks
> 1) {
2922 ieot
= xcalloc(1, sizeof(struct index_entry_offset_table
)
2923 + (ieot_blocks
* sizeof(struct index_entry_offset
)));
2924 ieot_entries
= DIV_ROUND_UP(entries
, ieot_blocks
);
2928 offset
= lseek(newfd
, 0, SEEK_CUR
);
2933 offset
+= write_buffer_len
;
2935 previous_name
= (hdr_version
== 4) ? &previous_name_buf
: NULL
;
2937 for (i
= 0; i
< entries
; i
++) {
2938 struct cache_entry
*ce
= cache
[i
];
2939 if (ce
->ce_flags
& CE_REMOVE
)
2941 if (!ce_uptodate(ce
) && is_racy_timestamp(istate
, ce
))
2942 ce_smudge_racily_clean_entry(istate
, ce
);
2943 if (is_null_oid(&ce
->oid
)) {
2944 static const char msg
[] = "cache entry has null sha1: %s";
2945 static int allow
= -1;
2948 allow
= git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2950 warning(msg
, ce
->name
);
2952 err
= error(msg
, ce
->name
);
2954 drop_cache_tree
= 1;
2956 if (ieot
&& i
&& (i
% ieot_entries
== 0)) {
2957 ieot
->entries
[ieot
->nr
].nr
= nr
;
2958 ieot
->entries
[ieot
->nr
].offset
= offset
;
2961 * If we have a V4 index, set the first byte to an invalid
2962 * character to ensure there is nothing common with the previous
2966 previous_name
->buf
[0] = 0;
2968 offset
= lseek(newfd
, 0, SEEK_CUR
);
2973 offset
+= write_buffer_len
;
2975 if (ce_write_entry(&c
, newfd
, ce
, previous_name
, (struct ondisk_cache_entry
*)&ondisk
) < 0)
2983 ieot
->entries
[ieot
->nr
].nr
= nr
;
2984 ieot
->entries
[ieot
->nr
].offset
= offset
;
2987 strbuf_release(&previous_name_buf
);
2994 /* Write extension data here */
2995 offset
= lseek(newfd
, 0, SEEK_CUR
);
3000 offset
+= write_buffer_len
;
3001 the_hash_algo
->init_fn(&eoie_c
);
3004 * Lets write out CACHE_EXT_INDEXENTRYOFFSETTABLE first so that we
3005 * can minimize the number of extensions we have to scan through to
3006 * find it during load. Write it out regardless of the
3007 * strip_extensions parameter as we need it when loading the shared
3011 struct strbuf sb
= STRBUF_INIT
;
3013 write_ieot_extension(&sb
, ieot
);
3014 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_INDEXENTRYOFFSETTABLE
, sb
.len
) < 0
3015 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
3016 strbuf_release(&sb
);
3022 if (!strip_extensions
&& istate
->split_index
&&
3023 !is_null_oid(&istate
->split_index
->base_oid
)) {
3024 struct strbuf sb
= STRBUF_INIT
;
3026 err
= write_link_extension(&sb
, istate
) < 0 ||
3027 write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_LINK
,
3029 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
3030 strbuf_release(&sb
);
3034 if (!strip_extensions
&& !drop_cache_tree
&& istate
->cache_tree
) {
3035 struct strbuf sb
= STRBUF_INIT
;
3037 cache_tree_write(&sb
, istate
->cache_tree
);
3038 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_TREE
, sb
.len
) < 0
3039 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
3040 strbuf_release(&sb
);
3044 if (!strip_extensions
&& istate
->resolve_undo
) {
3045 struct strbuf sb
= STRBUF_INIT
;
3047 resolve_undo_write(&sb
, istate
->resolve_undo
);
3048 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_RESOLVE_UNDO
,
3050 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
3051 strbuf_release(&sb
);
3055 if (!strip_extensions
&& istate
->untracked
) {
3056 struct strbuf sb
= STRBUF_INIT
;
3058 write_untracked_extension(&sb
, istate
->untracked
);
3059 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_UNTRACKED
,
3061 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
3062 strbuf_release(&sb
);
3066 if (!strip_extensions
&& istate
->fsmonitor_last_update
) {
3067 struct strbuf sb
= STRBUF_INIT
;
3069 write_fsmonitor_extension(&sb
, istate
);
3070 err
= write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_FSMONITOR
, sb
.len
) < 0
3071 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
3072 strbuf_release(&sb
);
3076 if (istate
->sparse_index
) {
3077 if (write_index_ext_header(&c
, &eoie_c
, newfd
, CACHE_EXT_SPARSE_DIRECTORIES
, 0) < 0)
3082 * CACHE_EXT_ENDOFINDEXENTRIES must be written as the last entry before the SHA1
3083 * so that it can be found and processed before all the index entries are
3084 * read. Write it out regardless of the strip_extensions parameter as we need it
3085 * when loading the shared index.
3087 if (offset
&& record_eoie()) {
3088 struct strbuf sb
= STRBUF_INIT
;
3090 write_eoie_extension(&sb
, &eoie_c
, offset
);
3091 err
= write_index_ext_header(&c
, NULL
, newfd
, CACHE_EXT_ENDOFINDEXENTRIES
, sb
.len
) < 0
3092 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
3093 strbuf_release(&sb
);
3098 if (ce_flush(&c
, newfd
, istate
->oid
.hash
))
3100 if (close_tempfile_gently(tempfile
)) {
3101 error(_("could not close '%s'"), get_tempfile_path(tempfile
));
3104 if (stat(get_tempfile_path(tempfile
), &st
))
3106 istate
->timestamp
.sec
= (unsigned int)st
.st_mtime
;
3107 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
3108 trace_performance_since(start
, "write index, changed mask = %x", istate
->cache_changed
);
3111 * TODO trace2: replace "the_repository" with the actual repo instance
3112 * that is associated with the given "istate".
3114 trace2_data_intmax("index", the_repository
, "write/version",
3116 trace2_data_intmax("index", the_repository
, "write/cache_nr",
3122 void set_alternate_index_output(const char *name
)
3124 alternate_index_output
= name
;
3127 static int commit_locked_index(struct lock_file
*lk
)
3129 if (alternate_index_output
)
3130 return commit_lock_file_to(lk
, alternate_index_output
);
3132 return commit_lock_file(lk
);
3135 static int do_write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
3139 int was_full
= !istate
->sparse_index
;
3141 ret
= convert_to_sparse(istate
);
3144 warning(_("failed to convert to a sparse-index"));
3149 * TODO trace2: replace "the_repository" with the actual repo instance
3150 * that is associated with the given "istate".
3152 trace2_region_enter_printf("index", "do_write_index", the_repository
,
3153 "%s", get_lock_file_path(lock
));
3154 ret
= do_write_index(istate
, lock
->tempfile
, 0);
3155 trace2_region_leave_printf("index", "do_write_index", the_repository
,
3156 "%s", get_lock_file_path(lock
));
3159 ensure_full_index(istate
);
3163 if (flags
& COMMIT_LOCK
)
3164 ret
= commit_locked_index(lock
);
3166 ret
= close_lock_file_gently(lock
);
3168 run_hook_le(NULL
, "post-index-change",
3169 istate
->updated_workdir
? "1" : "0",
3170 istate
->updated_skipworktree
? "1" : "0", NULL
);
3171 istate
->updated_workdir
= 0;
3172 istate
->updated_skipworktree
= 0;
3177 static int write_split_index(struct index_state
*istate
,
3178 struct lock_file
*lock
,
3182 prepare_to_write_split_index(istate
);
3183 ret
= do_write_locked_index(istate
, lock
, flags
);
3184 finish_writing_split_index(istate
);
3188 static const char *shared_index_expire
= "2.weeks.ago";
3190 static unsigned long get_shared_index_expire_date(void)
3192 static unsigned long shared_index_expire_date
;
3193 static int shared_index_expire_date_prepared
;
3195 if (!shared_index_expire_date_prepared
) {
3196 git_config_get_expiry("splitindex.sharedindexexpire",
3197 &shared_index_expire
);
3198 shared_index_expire_date
= approxidate(shared_index_expire
);
3199 shared_index_expire_date_prepared
= 1;
3202 return shared_index_expire_date
;
3205 static int should_delete_shared_index(const char *shared_index_path
)
3208 unsigned long expiration
;
3210 /* Check timestamp */
3211 expiration
= get_shared_index_expire_date();
3214 if (stat(shared_index_path
, &st
))
3215 return error_errno(_("could not stat '%s'"), shared_index_path
);
3216 if (st
.st_mtime
> expiration
)
3222 static int clean_shared_index_files(const char *current_hex
)
3225 DIR *dir
= opendir(get_git_dir());
3228 return error_errno(_("unable to open git dir: %s"), get_git_dir());
3230 while ((de
= readdir(dir
)) != NULL
) {
3231 const char *sha1_hex
;
3232 const char *shared_index_path
;
3233 if (!skip_prefix(de
->d_name
, "sharedindex.", &sha1_hex
))
3235 if (!strcmp(sha1_hex
, current_hex
))
3237 shared_index_path
= git_path("%s", de
->d_name
);
3238 if (should_delete_shared_index(shared_index_path
) > 0 &&
3239 unlink(shared_index_path
))
3240 warning_errno(_("unable to unlink: %s"), shared_index_path
);
3247 static int write_shared_index(struct index_state
*istate
,
3248 struct tempfile
**temp
)
3250 struct split_index
*si
= istate
->split_index
;
3251 int ret
, was_full
= !istate
->sparse_index
;
3253 move_cache_to_base_index(istate
);
3254 convert_to_sparse(istate
);
3256 trace2_region_enter_printf("index", "shared/do_write_index",
3257 the_repository
, "%s", get_tempfile_path(*temp
));
3258 ret
= do_write_index(si
->base
, *temp
, 1);
3259 trace2_region_leave_printf("index", "shared/do_write_index",
3260 the_repository
, "%s", get_tempfile_path(*temp
));
3263 ensure_full_index(istate
);
3267 ret
= adjust_shared_perm(get_tempfile_path(*temp
));
3269 error(_("cannot fix permission bits on '%s'"), get_tempfile_path(*temp
));
3272 ret
= rename_tempfile(temp
,
3273 git_path("sharedindex.%s", oid_to_hex(&si
->base
->oid
)));
3275 oidcpy(&si
->base_oid
, &si
->base
->oid
);
3276 clean_shared_index_files(oid_to_hex(&si
->base
->oid
));
3282 static const int default_max_percent_split_change
= 20;
3284 static int too_many_not_shared_entries(struct index_state
*istate
)
3286 int i
, not_shared
= 0;
3287 int max_split
= git_config_get_max_percent_split_change();
3289 switch (max_split
) {
3291 /* not or badly configured: use the default value */
3292 max_split
= default_max_percent_split_change
;
3295 return 1; /* 0% means always write a new shared index */
3297 return 0; /* 100% means never write a new shared index */
3299 break; /* just use the configured value */
3302 /* Count not shared entries */
3303 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3304 struct cache_entry
*ce
= istate
->cache
[i
];
3309 return (int64_t)istate
->cache_nr
* max_split
< (int64_t)not_shared
* 100;
3312 int write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
3315 int new_shared_index
, ret
;
3316 struct split_index
*si
= istate
->split_index
;
3318 if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0))
3319 cache_tree_verify(the_repository
, istate
);
3321 if ((flags
& SKIP_IF_UNCHANGED
) && !istate
->cache_changed
) {
3322 if (flags
& COMMIT_LOCK
)
3323 rollback_lock_file(lock
);
3327 if (istate
->fsmonitor_last_update
)
3328 fill_fsmonitor_bitmap(istate
);
3330 if (!si
|| alternate_index_output
||
3331 (istate
->cache_changed
& ~EXTMASK
)) {
3333 oidclr(&si
->base_oid
);
3334 ret
= do_write_locked_index(istate
, lock
, flags
);
3338 if (git_env_bool("GIT_TEST_SPLIT_INDEX", 0)) {
3339 int v
= si
->base_oid
.hash
[0];
3341 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3343 if (too_many_not_shared_entries(istate
))
3344 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3346 new_shared_index
= istate
->cache_changed
& SPLIT_INDEX_ORDERED
;
3348 if (new_shared_index
) {
3349 struct tempfile
*temp
;
3352 /* Same initial permissions as the main .git/index file */
3353 temp
= mks_tempfile_sm(git_path("sharedindex_XXXXXX"), 0, 0666);
3355 oidclr(&si
->base_oid
);
3356 ret
= do_write_locked_index(istate
, lock
, flags
);
3359 ret
= write_shared_index(istate
, &temp
);
3361 saved_errno
= errno
;
3362 if (is_tempfile_active(temp
))
3363 delete_tempfile(&temp
);
3364 errno
= saved_errno
;
3370 ret
= write_split_index(istate
, lock
, flags
);
3372 /* Freshen the shared index only if the split-index was written */
3373 if (!ret
&& !new_shared_index
&& !is_null_oid(&si
->base_oid
)) {
3374 const char *shared_index
= git_path("sharedindex.%s",
3375 oid_to_hex(&si
->base_oid
));
3376 freshen_shared_index(shared_index
, 1);
3380 if (flags
& COMMIT_LOCK
)
3381 rollback_lock_file(lock
);
3386 * Read the index file that is potentially unmerged into given
3387 * index_state, dropping any unmerged entries to stage #0 (potentially
3388 * resulting in a path appearing as both a file and a directory in the
3389 * index; the caller is responsible to clear out the extra entries
3390 * before writing the index to a tree). Returns true if the index is
3391 * unmerged. Callers who want to refuse to work from an unmerged
3392 * state can call this and check its return value, instead of calling
3395 int repo_read_index_unmerged(struct repository
*repo
)
3397 struct index_state
*istate
;
3401 repo_read_index(repo
);
3402 istate
= repo
->index
;
3403 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3404 struct cache_entry
*ce
= istate
->cache
[i
];
3405 struct cache_entry
*new_ce
;
3411 len
= ce_namelen(ce
);
3412 new_ce
= make_empty_cache_entry(istate
, len
);
3413 memcpy(new_ce
->name
, ce
->name
, len
);
3414 new_ce
->ce_flags
= create_ce_flags(0) | CE_CONFLICTED
;
3415 new_ce
->ce_namelen
= len
;
3416 new_ce
->ce_mode
= ce
->ce_mode
;
3417 if (add_index_entry(istate
, new_ce
, ADD_CACHE_SKIP_DFCHECK
))
3418 return error(_("%s: cannot drop to stage #0"),
3425 * Returns 1 if the path is an "other" path with respect to
3426 * the index; that is, the path is not mentioned in the index at all,
3427 * either as a file, a directory with some files in the index,
3428 * or as an unmerged entry.
3430 * We helpfully remove a trailing "/" from directories so that
3431 * the output of read_directory can be used as-is.
3433 int index_name_is_other(struct index_state
*istate
, const char *name
,
3437 if (namelen
&& name
[namelen
- 1] == '/')
3439 pos
= index_name_pos(istate
, name
, namelen
);
3441 return 0; /* exact match */
3443 if (pos
< istate
->cache_nr
) {
3444 struct cache_entry
*ce
= istate
->cache
[pos
];
3445 if (ce_namelen(ce
) == namelen
&&
3446 !memcmp(ce
->name
, name
, namelen
))
3447 return 0; /* Yup, this one exists unmerged */
3452 void *read_blob_data_from_index(struct index_state
*istate
,
3453 const char *path
, unsigned long *size
)
3457 enum object_type type
;
3461 pos
= index_name_pos(istate
, path
, len
);
3464 * We might be in the middle of a merge, in which
3465 * case we would read stage #2 (ours).
3469 (pos
< 0 && i
< istate
->cache_nr
&&
3470 !strcmp(istate
->cache
[i
]->name
, path
));
3472 if (ce_stage(istate
->cache
[i
]) == 2)
3477 data
= read_object_file(&istate
->cache
[pos
]->oid
, &type
, &sz
);
3478 if (!data
|| type
!= OBJ_BLOB
) {
3487 void stat_validity_clear(struct stat_validity
*sv
)
3489 FREE_AND_NULL(sv
->sd
);
3492 int stat_validity_check(struct stat_validity
*sv
, const char *path
)
3496 if (stat(path
, &st
) < 0)
3497 return sv
->sd
== NULL
;
3500 return S_ISREG(st
.st_mode
) && !match_stat_data(sv
->sd
, &st
);
3503 void stat_validity_update(struct stat_validity
*sv
, int fd
)
3507 if (fstat(fd
, &st
) < 0 || !S_ISREG(st
.st_mode
))
3508 stat_validity_clear(sv
);
3511 CALLOC_ARRAY(sv
->sd
, 1);
3512 fill_stat_data(sv
->sd
, &st
);
3516 void move_index_extensions(struct index_state
*dst
, struct index_state
*src
)
3518 dst
->untracked
= src
->untracked
;
3519 src
->untracked
= NULL
;
3520 dst
->cache_tree
= src
->cache_tree
;
3521 src
->cache_tree
= NULL
;
3524 struct cache_entry
*dup_cache_entry(const struct cache_entry
*ce
,
3525 struct index_state
*istate
)
3527 unsigned int size
= ce_size(ce
);
3528 int mem_pool_allocated
;
3529 struct cache_entry
*new_entry
= make_empty_cache_entry(istate
, ce_namelen(ce
));
3530 mem_pool_allocated
= new_entry
->mem_pool_allocated
;
3532 memcpy(new_entry
, ce
, size
);
3533 new_entry
->mem_pool_allocated
= mem_pool_allocated
;
3537 void discard_cache_entry(struct cache_entry
*ce
)
3539 if (ce
&& should_validate_cache_entries())
3540 memset(ce
, 0xCD, cache_entry_size(ce
->ce_namelen
));
3542 if (ce
&& ce
->mem_pool_allocated
)
3548 int should_validate_cache_entries(void)
3550 static int validate_index_cache_entries
= -1;
3552 if (validate_index_cache_entries
< 0) {
3553 if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))
3554 validate_index_cache_entries
= 1;
3556 validate_index_cache_entries
= 0;
3559 return validate_index_cache_entries
;
3562 #define EOIE_SIZE (4 + GIT_SHA1_RAWSZ) /* <4-byte offset> + <20-byte hash> */
3563 #define EOIE_SIZE_WITH_HEADER (4 + 4 + EOIE_SIZE) /* <4-byte signature> + <4-byte length> + EOIE_SIZE */
3565 static size_t read_eoie_extension(const char *mmap
, size_t mmap_size
)
3568 * The end of index entries (EOIE) extension is guaranteed to be last
3569 * so that it can be found by scanning backwards from the EOF.
3576 const char *index
, *eoie
;
3578 size_t offset
, src_offset
;
3579 unsigned char hash
[GIT_MAX_RAWSZ
];
3582 /* ensure we have an index big enough to contain an EOIE extension */
3583 if (mmap_size
< sizeof(struct cache_header
) + EOIE_SIZE_WITH_HEADER
+ the_hash_algo
->rawsz
)
3586 /* validate the extension signature */
3587 index
= eoie
= mmap
+ mmap_size
- EOIE_SIZE_WITH_HEADER
- the_hash_algo
->rawsz
;
3588 if (CACHE_EXT(index
) != CACHE_EXT_ENDOFINDEXENTRIES
)
3590 index
+= sizeof(uint32_t);
3592 /* validate the extension size */
3593 extsize
= get_be32(index
);
3594 if (extsize
!= EOIE_SIZE
)
3596 index
+= sizeof(uint32_t);
3599 * Validate the offset we're going to look for the first extension
3600 * signature is after the index header and before the eoie extension.
3602 offset
= get_be32(index
);
3603 if (mmap
+ offset
< mmap
+ sizeof(struct cache_header
))
3605 if (mmap
+ offset
>= eoie
)
3607 index
+= sizeof(uint32_t);
3610 * The hash is computed over extension types and their sizes (but not
3611 * their contents). E.g. if we have "TREE" extension that is N-bytes
3612 * long, "REUC" extension that is M-bytes long, followed by "EOIE",
3613 * then the hash would be:
3615 * SHA-1("TREE" + <binary representation of N> +
3616 * "REUC" + <binary representation of M>)
3618 src_offset
= offset
;
3619 the_hash_algo
->init_fn(&c
);
3620 while (src_offset
< mmap_size
- the_hash_algo
->rawsz
- EOIE_SIZE_WITH_HEADER
) {
3621 /* After an array of active_nr index entries,
3622 * there can be arbitrary number of extended
3623 * sections, each of which is prefixed with
3624 * extension name (4-byte) and section length
3625 * in 4-byte network byte order.
3628 memcpy(&extsize
, mmap
+ src_offset
+ 4, 4);
3629 extsize
= ntohl(extsize
);
3631 /* verify the extension size isn't so large it will wrap around */
3632 if (src_offset
+ 8 + extsize
< src_offset
)
3635 the_hash_algo
->update_fn(&c
, mmap
+ src_offset
, 8);
3638 src_offset
+= extsize
;
3640 the_hash_algo
->final_fn(hash
, &c
);
3641 if (!hasheq(hash
, (const unsigned char *)index
))
3644 /* Validate that the extension offsets returned us back to the eoie extension. */
3645 if (src_offset
!= mmap_size
- the_hash_algo
->rawsz
- EOIE_SIZE_WITH_HEADER
)
3651 static void write_eoie_extension(struct strbuf
*sb
, git_hash_ctx
*eoie_context
, size_t offset
)
3654 unsigned char hash
[GIT_MAX_RAWSZ
];
3657 put_be32(&buffer
, offset
);
3658 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3661 the_hash_algo
->final_fn(hash
, eoie_context
);
3662 strbuf_add(sb
, hash
, the_hash_algo
->rawsz
);
3665 #define IEOT_VERSION (1)
3667 static struct index_entry_offset_table
*read_ieot_extension(const char *mmap
, size_t mmap_size
, size_t offset
)
3669 const char *index
= NULL
;
3670 uint32_t extsize
, ext_version
;
3671 struct index_entry_offset_table
*ieot
;
3674 /* find the IEOT extension */
3677 while (offset
<= mmap_size
- the_hash_algo
->rawsz
- 8) {
3678 extsize
= get_be32(mmap
+ offset
+ 4);
3679 if (CACHE_EXT((mmap
+ offset
)) == CACHE_EXT_INDEXENTRYOFFSETTABLE
) {
3680 index
= mmap
+ offset
+ 4 + 4;
3689 /* validate the version is IEOT_VERSION */
3690 ext_version
= get_be32(index
);
3691 if (ext_version
!= IEOT_VERSION
) {
3692 error("invalid IEOT version %d", ext_version
);
3695 index
+= sizeof(uint32_t);
3697 /* extension size - version bytes / bytes per entry */
3698 nr
= (extsize
- sizeof(uint32_t)) / (sizeof(uint32_t) + sizeof(uint32_t));
3700 error("invalid number of IEOT entries %d", nr
);
3703 ieot
= xmalloc(sizeof(struct index_entry_offset_table
)
3704 + (nr
* sizeof(struct index_entry_offset
)));
3706 for (i
= 0; i
< nr
; i
++) {
3707 ieot
->entries
[i
].offset
= get_be32(index
);
3708 index
+= sizeof(uint32_t);
3709 ieot
->entries
[i
].nr
= get_be32(index
);
3710 index
+= sizeof(uint32_t);
3716 static void write_ieot_extension(struct strbuf
*sb
, struct index_entry_offset_table
*ieot
)
3722 put_be32(&buffer
, IEOT_VERSION
);
3723 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3726 for (i
= 0; i
< ieot
->nr
; i
++) {
3729 put_be32(&buffer
, ieot
->entries
[i
].offset
);
3730 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3733 put_be32(&buffer
, ieot
->entries
[i
].nr
);
3734 strbuf_add(sb
, &buffer
, sizeof(uint32_t));